USF Libraries
USF Digital Collections

Analysis of two strategies for structuring medicare reimbursement to maximize profitability in acute care general hospitals

MISSING IMAGE

Material Information

Title:
Analysis of two strategies for structuring medicare reimbursement to maximize profitability in acute care general hospitals
Physical Description:
Book
Language:
English
Creator:
Barrington, James
Publisher:
University of South Florida
Place of Publication:
Tampa, Fla
Publication Date:

Subjects

Subjects / Keywords:
DRG
Investor owned hospital
Discretionary procedure
Marginal effects
Reciprocal of the debt ratio
Gaming
Dissertations, Academic -- Health Policy & Management -- Masters -- USF   ( lcsh )
Genre:
non-fiction   ( marcgt )

Notes

Abstract:
ABSTRACT: ABSTRACT The healthcare literature sometimes cites Medicare as a negative determinant of hospital profitability. However, a review of Florida acute care short-term general hospital data revealed a subset of profitable hospitals with high percentages of their revenue structure comprised of Medicare reimbursements. Some investigators might contend that these hospitals are just better managed; that hospital profitability is not related to patient mix or payer source. Although good management enhances financial health, there are perhaps other reasons why certain hospitals can become profitable with Medicare as their primary revenue source. Research findings indicate there is wide geographic variability shown for per-capita volumes of discretionary procedures reimbursed by Medicare, and broad variations in Medicare spending per enrollee for general acute care short-term hospitalizations. It was also found that many of the hospitals performing higher rates of discretionary procedures and showing the ability to make a profit with Medicare are investor owned. The focus of this study, covering years 2000-2005, was to examine two strategies using discretionary procedures under Medicare that Florida investor owned hospitals may employ to increase profitability and maintain long-term financial health. Part 1 of the study examined the association between long-term financial viability, measured by the total assets ÷ total liabilities (TATL) ratio (the reciprocal of the debt ratio) and percentages per hospital of two discretionary cardiac and orthopedic procedure variables, reimbursed by conventional Medicare. A positive association was found between the TATL ratio and these variables, as well as significant marginal effects in the association between the TATL ratio and interaction terms for hospital ownership (where investor owned = 1 and not-for-profit = 0) and the discretionary cardiac procedure variable and ownership and the discretionary orthopedic procedure variable. Part 2 used total charges as the dependent variable for patient discharges reimbursed by Medicare HMO. It was found that investor owned hospitals generally assess significantly higher charges than not-for-profits for discretionary CABG and valve replacement procedures for patients with equivalent levels of medical services and hospitalization. It was also found that charges significantly increase for both investor-owned and not-for-profit hospitals located in the southern region of Florida.
Thesis:
Dissertation (PHD)--University of South Florida, 2010.
Bibliography:
Includes bibliographical references.
System Details:
Mode of access: World Wide Web.
System Details:
System requirements: World Wide Web browser and PDF reader.
Statement of Responsibility:
by James Barrington.
General Note:
Title from PDF of title page.
General Note:
Document formatted into pages; contains X pages.

Record Information

Source Institution:
University of South Florida Library
Holding Location:
University of South Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
usfldc doi - E14-SFE0004617
usfldc handle - e14.4617
System ID:
SFS0027932:00001


This item is only available as the following downloads:


Full Text
xml version 1.0 encoding UTF-8 standalone no
record xmlns http:www.loc.govMARC21slim xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.loc.govstandardsmarcxmlschemaMARC21slim.xsd
leader nam 22 Ka 4500
controlfield tag 007 cr-bnu---uuuuu
008 s2010 flu s 000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0004617
035
(OCoLC)
040
FHM
c FHM
049
FHMM
090
XX9999 (Online)
1 100
Barrington, James.
0 245
Analysis of two strategies for structuring medicare reimbursement to maximize profitability in acute care general hospitals
h [electronic resource] /
by James Barrington.
260
[Tampa, Fla] :
b University of South Florida,
2010.
500
Title from PDF of title page.
Document formatted into pages; contains X pages.
502
Dissertation (PHD)--University of South Florida, 2010.
504
Includes bibliographical references.
516
Text (Electronic thesis) in PDF format.
538
Mode of access: World Wide Web.
System requirements: World Wide Web browser and PDF reader.
3 520
ABSTRACT: ABSTRACT The healthcare literature sometimes cites Medicare as a negative determinant of hospital profitability. However, a review of Florida acute care short-term general hospital data revealed a subset of profitable hospitals with high percentages of their revenue structure comprised of Medicare reimbursements. Some investigators might contend that these hospitals are just better managed; that hospital profitability is not related to patient mix or payer source. Although good management enhances financial health, there are perhaps other reasons why certain hospitals can become profitable with Medicare as their primary revenue source. Research findings indicate there is wide geographic variability shown for per-capita volumes of discretionary procedures reimbursed by Medicare, and broad variations in Medicare spending per enrollee for general acute care short-term hospitalizations. It was also found that many of the hospitals performing higher rates of discretionary procedures and showing the ability to make a profit with Medicare are investor owned. The focus of this study, covering years 2000-2005, was to examine two strategies using discretionary procedures under Medicare that Florida investor owned hospitals may employ to increase profitability and maintain long-term financial health. Part 1 of the study examined the association between long-term financial viability, measured by the total assets total liabilities (TATL) ratio (the reciprocal of the debt ratio) and percentages per hospital of two discretionary cardiac and orthopedic procedure variables, reimbursed by conventional Medicare. A positive association was found between the TATL ratio and these variables, as well as significant marginal effects in the association between the TATL ratio and interaction terms for hospital ownership (where investor owned = 1 and not-for-profit = 0) and the discretionary cardiac procedure variable and ownership and the discretionary orthopedic procedure variable. Part 2 used total charges as the dependent variable for patient discharges reimbursed by Medicare HMO. It was found that investor owned hospitals generally assess significantly higher charges than not-for-profits for discretionary CABG and valve replacement procedures for patients with equivalent levels of medical services and hospitalization. It was also found that charges significantly increase for both investor-owned and not-for-profit hospitals located in the southern region of Florida.
590
Advisor: Alan M. Sear, Ph.D.
653
DRG
Investor owned hospital
Discretionary procedure
Marginal effects
Reciprocal of the debt ratio
Gaming
690
Dissertations, Academic
z USF
x Health Policy & Management
Masters.
773
t USF Electronic Theses and Dissertations.
4 856
u http://digital.lib.usf.edu/?e14.4617



PAGE 1

Analysis of Two Strategies for Structuring Medicare Reimbursement to Maximize Profitability in Acute Care General Hospitals by James D. Barrington A dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy Department of Health Policy & Management College of Public Health University of South Florida Major Professor: Alan M. Sear, Ph.D. Barbara Langland Orban, Ph.D. Etienne E. Pracht, Ph.D. Yiliang Zhu, Ph.D. Date of Approval : April 1 6 2010 Keywords: DRG, investor owned hospital discretionary procedure marginal effects reciprocal of the debt ratio gaming Copyright 2010 James D. Barrington

PAGE 2

Dedication It is a wonderful opportunity and great honor for me to dedicate my dissertation to the memory of my father, James Durrel Barrington (1920 2003). During what turned out to be our last lucid conversation I was able to inform my dad that I had b een accepted in to the PhD program. He was elated My father had grown up in the depression in abject poverty. Before he was ten years old he was working as a farm hand to help his mother support their household. Perhaps influenced by his own harsh yout h, h is primary goal that he worked so hard all his life to create was a better life for his children. He set an example of what patience, tenacity, and a strong work ethic could accomplish. He was stern yet exercised wisdom and a sense of fairness and ju stice, which often are qualities nurtured in people who have experienced hard work and hard times. tour of duty as a first lieutenant. He incurred a back injury during combat, the effects of which would cause him pain the rest of his life. He then worked hard and diligently for the next several decades, giving of himself to his children and grandchildren, making sure none of them would ever experience what he had as a boy, that of poverty, hunger, and a demand to grow up immediately He died in January 2003. He would have been 83 the following month. Though h is body was greatly weakened, he still left the earth with the same qualities he had lived by; that of a quiet and bold acceptance of the facts and an admirable sense of dignity. It is often said that the WWII generation was the greatest generation. Having the privilege to have known my father I am convinced this is true.

PAGE 3

Acknowledgments Sincere appreciation is expressed to my doctoral committee members, Major Advisor Dr. Alan M. Sear, Dr. Barbara Langland Orban, Dr. Etienne E. Pracht, and Dr. Yiliang Zhu. Dr. Sear is a stern, no nonsense type of professor that is brutally honest. His old school approach and unbridled candor were not always well received, but in retrospect I realize these qualities helped intensify my focus and bring the dissertation full c ircle. Dr. Sear also has an immense background in health outcomes research of which was a n invaluable and replete source of information and direction. Dr. Orban examined the dissertation in great detail and made numerous salient recommendations which resulted in substantial improvement of the document. Dr. Orban also provided invaluable guidance on the determinants of hospital profitability as well as the importance of the case mix index and Medicare mix when examining financial pe rformance in the context of Medicare. Dr. Pracht also helped to improve the document by specifically requir ing the inclusion of the technical appendix (Appendix 9) which featured the addition of three important variables The appendix helped to assure that parsimony in model development had not occurred at the expense of excluding valuable information. Dr. Zhu provided valuable help in understanding and using interacting variables in model development as well as the calculation of marginal effects and standard errors of marginal effects I also wish to thank Dissertation Defense Chair Dr. Thomas Mason who managed the session at an efficient pace without overlooking any relevant detail.

PAGE 4

i Table of Contents List of Tables v i List of Figures ix Abstract x Chapter 1: Introduction/Statement of the Problem 1 Increased Legal & Regulatory Provisions 2 Ubiquitous and Recurrent Nature of Gaming 3 Focus of the Investigation 4 Chapter 2: Review of the Literature 9 Introduction 9 Brief Summary of the Literature Review 9 Indicators of Financial Performance 12 Multiple Indicators of Financial Performance 13 Medicare is Not a Profitable Revenue Base? 16 Good Management is the Strongest Determinant of Profitability 19 Profitability after Regulatory Changes 21 Managing Assets to Raise Capital & Assure Future Profitability 22 Management & Strategic Direction in Initial Location Decisions 23 Hospital Ownership 24 Securing Ongoing Capital 25 Not for x Exemption 25 The Inurement Principle & Pressures from Governing Boards 27 Does Good Management Fully Explain Inordinately High Profitability 29 Investor 30 Prevalence of Health Care Fraud 33 35 Discretionary Versus Nondiscretionary Procedures 37 Nondiscretionary Procedures 38 Discretionary Procedures & Efficacy of Alternative Treatments 40 Life Expectancy for CABG Patients vs. Those Who Do Not Elect CABG 40 Mastectomy vs. Lumpectomy 40 Benign Prostatic Hyperplasia (BPH) & Radical Prostatectomy 41 Joint Replacement & Revision 4 1 Back Surgery 42

PAGE 5

ii Geog raphic Variability of Discretionary Procedures 42 Medicare Rates for Certain Short Term Inpatient Hospital S tays in Florida 2005 43 Further Literature Findings of Geographic Variation of Discretionary Procedures 47 49 The Effects of Geographic Disparities in Medicare Spending 51 Higher & Lower Levels of Medicare Spending Between States 53 Geographic Differences in Per Capita Medicare Spending in Florida 56 Me dicare Reimbursements for Short Term Inpatient Hospital Stays 56 Total Medicare Spending in Florida 58 Prevalence of Procedures and Levels of Medicare Expenditures 60 Surgical Signature in Several Florida Cities 61 Geographic location 6 3 Research Summary 63 Chapter 3: Research Methods Data Structure & Identification of Variables 6 5 Research Goals & Objectives 6 5 Goal for Part 1 6 5 Objectives for Part 1 66 Goal for Part 2 6 6 Objectives for Part 2 6 6 Data Sources 6 7 Structure of the Datasets 6 7 Selection of Hospitals 6 8 Data Structure for Part 1 6 8 Data Structure for Part 2 6 9 Deriving an Operational Definition of Discretionary Procedures 70 Criteria for Identifying Discretionary Procedures 71 Using a Nondiscretionary Procedure as a Benchmark 71 Using the Coefficient of Variation as an Indicator of Geographic Variability 72 Selecting Discretionary Procedures for Further Investigation 7 4 Dependent & Indepe ndent Variables for Part 1 7 9 Selection of Variables for Part 1 7 9 Measurng 1Z in Model Development 83 Selection Critiera for DRG Codes to be used in Model Development 83 Selection of Variables for Part 2 84 Chapter 4: Research Methods Plan of Analysis for Parts 1 & 2 8 8 Rules for Interaction Models, Model 1 B in Part 1 & Model 2 B in Part 2 8 9 Constitutive Terms 8 9 Managing Multicollinearity in Interaction Terms 8 9

PAGE 6

iii Chapter 5: Research Methods Model Develop m ent for Part 1 9 2 Descriptive Statistics for Continuous Variables Used in Part 1 92 Preliminary Analysis for Part 1 93 Examinations of Potential Control Variables for Part 1 94 Pretesting for Interaction Using the Breslow Day Test 95 Correlation Matrices 97 Comparison of Discretionary Procedure Mean Percentages 9 8 Statements of Hypotheses for Part 1 100 Chapter 6 : Research Methods Model Development for Part 2 10 2 Preliminary Analysis for Part 2 10 2 Descriptive Statistics for Continuou s Variables Used in Part 2 103 Selection of DRG Codes for Part 2 104 Analysis of Mean Charges for DRG Codes Reimbursed by Medicare HMO 10 4 Selection of CABG & Valve Replacements as Population Parameters 106 Review & Selection of Potential Covariates 107 Controlling for Patient Disease Severity 1 07 Charges per Hospital Day 10 9 Geographic Region (Florida Inspection Area) 1 10 Contingency Tables 11 3 Homogeneity of the Odds Ratios 11 4 Statements of Hypotheses for Part 2 11 5 Chapter 7 : Model Results for Part 1: Testing the Association between the TATL Ratio & Discretionary Procedures ( Reimbursed by Conventional Medicare ) 117 Organization of Model Findings 117 Model 1 A: Main Effects Model 117 Regression Equation for Model 1 A 117 Model 1 B : Interaction Model 120 Regression Equation for Model 1 B 120 Significance of the Interaction Model Using the F Test 121 Parameter Estimates 12 2 Interpreting Marginal Effects of Interaction Terms in Model 1 B 12 3 Marginal Effects of the Interaction Term Orthopedic Group Procedures x Ownership 123 Marginal Effect of Orthoped ic Group Procedures When Ownership = 1 12 5 Standard Error & t Value for the Marginal Effect 1 2 6 Marginal Effects of the Interaction Term Cardiac Group Procedures x Ownership 1 27 Marginal Effect of Cardiac Group Procedures When Ownership = 1 128 Discussion of Findings for Part 1 130

PAGE 7

iv Testing for Unnecessary & Marginally Necessary Procedure s 130 Gaming or Good Management 1 31 Part 1 Regression Diagnostics 1 32 Outliers Press Statistic 1 33 1 3 4 Influence Statistics 1 3 4 Specification Errors 1 34 Heterogeneous Variances 1 35 Weaknesses of Part 1 Models 1 35 Randomly Splitting the Data to Assure Model Validity 1 3 6 Grouping DRG Codes 1 36 Procedure Volumes 1 36 County Median Income as a Proxy for Geographic Location 1 37 Rigid Adherence to Parsimony 13 8 Chapter 8: Model Results for Part 2: Testing the Association between Total Charges & Ownership for CABG & Valve Replacement Procedures (Reimbursed by Medicare HMO) 139 Model 2 A: Main Effects Model 139 Regression Equation for Model 2 A 140 Model 2 B : Interaction Model 1 41 Regression Equation for Model 2 B, Intera ction Model 142 Interpreting Marginal Effects of Interaction Terms in Model 2 B 144 Discussion of Findings for Part 2 145 Total Charges vs. Reimbursed Charges 146 Part 2 Regression Diagnostics 1 4 8 Outliers 14 8 Studentized 149 Influence Sta tistics 149 Specification Errors 150 Heterogeneous Variables 150 Multicollinearity 150 Weaknesses of Models in Part 2 1 51 Assessing Validity by Randomly Splitting the Data 151 Total Patient Charges vs. Reimbursements 151 Ch apter 9: Conclusion 152 Policy Recommendations 15 5 A Few Ca veats on Interpretation 155 Recommendat ions to Federal Agencies 156 A Brief Example of Ongoing Monitoring of Rates of Discretionary Procedures Reimbursed by Medicare 157 Recommendations t o Hospital Administrators 159 State Average 160

PAGE 8

v List of Reference s 1 63 Bibliography 173 Appendices 18 3 Appendix 1: Rates of Surgical Procedures p er 1,000 Medicare Enrollees, All Florida Hospital Referral Regions, 2003 1 84 Appendix 2: Medicare Reimbursements for Inpatient Short Stays per Enrollee, Florida HRRs, 2000 2005 187 Appendix 3: Correlation Matrices, TATL Ratio & Selected Discretionary Procedures 1 91 Appendix 4: HUD Prescribed Median Income Levels by Quartiles for Total ( 67 ) Florida Counties 19 3 Appendix 5: Total Medicare HMO Patient Charges Valve Replacement & CABG Procedure s Mean Charges by Quartile, 2000 200 5 19 4 Appendix 6: AHCA 11 Inspection Regions by Florida Geographic Secto r 1 95 Appendix 7: Variance Covariance Matrix Model 1 B 1 96 Appendix 8: P art 1, Model 1 B Collinearity Diagnostics ( I ntercept A djusted) 1 98 Appendix 9: Technical Appendix : Circumventing a Possible Specification Error : Supplement to Part 1, Models 1 A & 1 B 199 Appendix 10 : Variance Covariance Matrix Model 2 B 23 7 Appendix 1 1 : Part 2, Model 2 C Collinearity Diagnostics (Intercept Adjusted) 23 8 About the Author End Page

PAGE 9

vi List of Tables Table 1 Distribution of Fraudulent Activities in Healthcare 1993 3 3 Table 2 Range of Rates for Discretionary Procedures Per 1,000 Medicare Enrollees, Highest & Lowest State Rates in US and Highest & Lowest City Rates in Florida, 1999 43 Table 3 Medicare Spending by City, State & Hospital Referral Region, Higher & Lower Levels of Medicare Reimbursements per Medicare Enrollee, 2003 5 3 Table 4 Medicare Reimbursements for Short Term Inpatient Hospital Stays Pe r 1,000 Medicare Enrollees, 2005 5 7 Table 5 Inflation Adjusted Total Medicare Spending Per Enrollee by Hospital Referral Region (HRR) in Florida, 1992 & 2006, Actual & Percentage Growth 59 Table 6 Procedures Defined as Discretionary Based on Clinical & Geographic Variability Criteria 74 Table 7 Cardiac Group Procedures Variable, Florida Short Term Acute Care General Hospitals, 2000 2005 7 6 Table 8 Orthopedic Group Procedures Variable, Florida Short Term Acute Care General Hospitals, 2000 2005 7 6 Table 9 Raw Data Breakdown of Cardiac & Orthopedic Group Procedures by Ownership 2000 2005 7 7 Table 10 Descriptive Statistics for Continuous Variables used in Part 1, Models 1 A & 1 B 92 Table 11 2x2 Contingency Tables for TATL Ratio & Discretionary Procedures Reimbursed by Conventional Medicare, 2000 2005 94

PAGE 10

vii Table 12 2x2 Contingency Tables for TATL Ratio & Hospital Ownership, Geographic Location, 2000 2005 95 Table 13 Breslow Day Tests to Assess Interaction using 2x2 Contingency Tables & Stratified Analysis, 2000 2005 97 Table 14 Pearson & Spearman Correlation Coefficients for the Association between the TATL Ratio and Discretionary Procedures, Stratified by Ownership, 2000 2005 98 Table 15 Mean Percentage Differences between Investor Owned Hospitals with High TATL Ratios and Al l Other Hospitals, Cardiac & Orthopedic Group Procedures, Reimbursed by Conventional Medicare, 2000 2005 99 Table 16 Descriptive Statistics for Continuous Variables used in Part 2 10 4 Table 17 Analysis of Mean Total Charges for Selected Medicare HMO Procedures by Ownership, 2000 2005 10 5 Table 18 Pearson & Spearman Correlation Coefficients, Associations between Total Charges & Potential Variables for Model Development, 2000 2005 10 8 Table 19 Charges Per Patient Day Valve Replacement & CABG Procedures, Investor Owned vs. Not for Profit Hospitals, Medicare HMO, 2000 2005 10 9 Table 2 0 2x2 Contingency Table Analysis for Total Charges and Ownership & Geographic Region, Medicare HMO Financed Valve Replacement & CABG Procedures, 2000 2005 1 1 4 Table 2 1 Breslow Day Test to Assess Interaction using a 2x2 Contingency Stratified Table, CABG & Valve Replacements Financed by Medicare HMO, 2000 2005 1 1 5 Table 22 Model 1 A: Main Effects Model 11 8 Table 23 Model 1 A: Main Effects Model Parameter Estimates 11 9 Table 24 Model 1 B: Interaction Model 1 20 Table 25 Model 1 B: Interaction Model Parameter Estimates 1 2 2

PAGE 11

viii Table 26 Marginal Effects for the Interaction Term Orthopedic Group Procedures x Ownership, Four Conditions 1 2 4 Table 27 Marginal Effects for the Interaction Term Cardiac Group Procedures x Ownership, Four Conditions 1 2 7 Table 2 8 Sum of Residuals, Sum of Squared Residuals & PRESS for Model 1 B 1 3 3 Table 2 9 Model 2 A: Main Effects Model 1 40 Table 30 Model 2 A: Main Effects Model Parameter Estimates 1 4 1 Table 31 Model 2 B: Interaction Model 1 4 2 Table 32 Model 2 B: Interaction Model Parameter Estimates 1 4 3 Table 33 Percentage of Net Revenue for Inpatient Services, Investor Owned vs. Not for Profit Hospitals in the South Region vs. Combined North & Central Regions, Medicare HMO Inpatient Services, 2000 2005 147 Table 3 4 Sum of Residuals, Sum of Squared Residuals & PRESS for Model 2 B 1 4 8

PAGE 12

ix List of Figures Figure 1 PCI Discharges per 1,000 Medicare Enrollees, All Florida H ospital Referral Regions 2005 4 4 Figure 2 Discharges for Knee Replacement per 1,000 Medicare Enrollees, All Florida Hospital Referral Regions 2005 4 5 Figure 3 Discharges for Back Surgery per 1,000 Medicare Enrollees, All Florida Hospital Referral Regions 2005 4 6 Figure 4 Mean Charges & Volumes of Medicare HMO Financed Valve Replacements & CABG Procedures by Florida Geographic Sector, 2000 2005 1 1 1 Figure 5 Mean Charges & Volumes of Medicare HMO Financed Valve Replacements by Geographic Sector, 2000 2005 1 1 2 Figure 6 Mean Charges & Volumes of Medicare HMO Financed CABG Procedures by Geographic Sector, 2000 2005 1 1 2 Figure 7 Comparison of Total Patient Medicare HMO Charges by Ownership & Region, 2000 2005 1 1 3 Figure 8 CABG Rates per 1,000 Medicare Enrollees by Florida Hospital Referral Region 2005 15 8 Figure 9 CABG Discharges per 1,000 Medicare Enrollees 1 6 1

PAGE 13

x Analysis of Two Strategies for Structuring Medicare Reimbursement to Maximize Profitability in Acute Care General Hospitals James Barrington A BSTRACT The healthcare literature sometimes cites Medicare as a negative determinant of hospital profitability. However, a review of Florida acute care short term general hospital data revealed a subset of profitable hospitals with high percentages of their revenue structure comprised of Medicare re imbursements. Some investigators might contend that these hospitals are just better managed; that hospital profitability is not related to patient mix or payer source Although good management enhances financial health there are perhaps other reasons wh y certain hospitals can become profitable with Medicare as their primary revenue source. R esearch findings indicate there is wide geographic variability shown for per capita volumes of discretionary procedures reimbursed by Medicare, and broad variations in Medicare spending per enrollee for general acute care short term hospitalizations. It was also found that many of the hospitals performing higher rates of discretionary procedures and showing the ability to make a profit with Medicare are investor own ed. The focus of this study covering years 2000 2005, was to examine two strategies using discretionary procedures under Medicare that Florida investor owned hospitals may employ to increase profitability and maintain long term financial health.

PAGE 14

xi Part 1 of the study examined the association between long term financial viability measured by the total assets total liabilities (TATL) ratio (the reciprocal of the debt ratio) and percentages per hospital of two discretionary cardiac and orthopedic pr ocedure variables reimbursed by conventional Medicare. A positive association was found between the TATL ratio and these variables, as well as significant marginal effects in the association between the TATL ratio and interaction terms for hospital owner ship (where investor owned = 1 and not for profit = 0) and the discretionary cardiac procedure variable and ownership and the discretionary orthopedic procedure variable. Part 2 used total charges as the dependent variable for patient discharges reimbursed by Medicare HMO. It was found that investor owned hospitals generally assess significantly higher charges than not for profits for discretionary CABG and valve replacement procedures for patients with equivalent levels of medical services and hospitaliza tion. It was also found that charges significantly increase for both investor owned and not for profit hospitals located in the southern region of Florida.

PAGE 15

1 Chapter 1 Introduction / Statement of the Problem Of the $ 2 0 trillion spent on health services in the United States in 2005 16.0% of the Gross Domestic Product (GDP) $ 611 billion was spent on hospital services (Catlin, Cowan, Heffler et a l 2007). Of the amount spent on hospital services, Medicare paid close to $200 billion ( CMS 2005). S ince the Medicare Prospective Payment System (PPS) for hospitals was introduced in October 1983, there have been reports in the media (and sometimes found in little known government publications, trade publications, and newsletters published by professional associations, citizen groups, and law firms) documenting hospital Medicare fraud, e.g., DRG upcoding ( ACFE 2005 06, Psaty, Boineau, Kuller et al 1999, Rosenberg, Fryback & Katz 2000), misuse of the PPS DRG outlier methodology (CMS 2003, Darmiento 2002, Becker & Kessler 2005), and the performance of unnecessary c oronary a rtery b ypass g rafts (CABG), valve replacements, and percutaneous coronary interven tions (PCI) ( Darmiento 2002 Kesselheim & Brennan 2005). upcoding medical procedures which they performed. Upcoding involves billing for a higher level of service than was actually ren dered via code manipulation; altering a diagnostic code to reflect greater patient severity and more intensive levels of medical services performed

PAGE 16

2 to, in turn, obtain higher reimbursement (ACFE 2005 06). Although this method was effective in generating g reater revenue in the short term, Medicare eventually began to investigate and enforce sanctions against hospitals using the upcoding strategy, causing these hospitals to be more cautious in gaming DRG codes (Middleton 200 2 practice, upcoding has run the gamut from questionable to outright fraud. Ferreting out 2 ). Medicare outlier payment syst em as a source of additional revenue. CMS (200 3 ) cited a increasing their charges compared to costs, costing taxpayers over $2 billion in 2002 and $1 billion to $2 billion a year in inappropriate overpayments for each of the last four Despite these sporadic reports there have been no scientific studies documenting the annual amounts of Medicare funds lost to hospital Medicare fraud. In 2002 it was discovered that for several years Redding Medical Center, a subsidiary of Tenet Healthcare Inc. had been submitting false billings to Medicare, abusing the Medicare outlier program, and even performing and charging for invasive surgeries that were medically unnecessary ( Darmiento 2002 HHS OIG, December 11, 2003, DOJ May 17, 2004 ) Increased Legal & Regulatory Provisions Increased strength of the Federal False Claims Act [31 USC § 3729, 3730 (1)] to use against any organization engaging in fraudulent activities with federal funds (Zack 2001). The 1996 HIPAA legislation also gave f ederal investigators increased funding and a mandate

PAGE 17

3 ed some hospitals to down 2 ) In 2002 the Sarbanes Oxley (SOX) act was passed into law in response to the corporate scandals of 2001 and 2002 (Bisk 2009) In the context of healthc are, this act was passed in a political environment where rooting out fraud and abuse was paramount. This law significantly changed the laws of corporate governance and the rules and regulations under which accounting firms and corporations must operate (ACFE 2005 06). Additionally, certain provisions substantially increased the responsibilities of management and the accounting profession regarding financial reporting, design and operation of the internal control structure, and prevention and detection of fraud (ACFE 2005 06) SOX applied only to publically traded companies, but many of the provisions of SOX have been implemented as best practices by not for profit hospitals (Bisk 2009) Ubiquitous and Recurrent Nature of Gaming It seems that no matt er what laws are enacted or how strictly regulatory actions are enforced, there will always be sectors of the healthcare industry involved in gaming. When one game becomes too risky, there is always another game that laws do not yet cover, or that regulato rs are not focusing upon. its inception, doctors discovered that by increasing their fees they would receive higher 2 ). Historically, a

PAGE 18

4 Gaming in this study is defined as a recurrent strategic maneuvering across legal and/or ethical t hresholds of conduct for the purpose of acquiring additional unwarranted profit The recurrent nature of g aming in this context implies scienter which means the gamer (i.e., offending party) has knowledge and willful intent of e ngag ing in wrongful behavi or (Convisor Duffy 1996 ) Listed below is a brief typology of gaming behaviors that are historically characteristic of the health care industry (ACFE 2005 06 Cleverly & Cameron 2002 ) Kickbacks o Payment for referrals of patients o Payment for insurance con tracts o Payment for vendor contracts Inflated billings o Alterations of claim forms, amounts charged, and financial ledgers o Adding services never rendered to actual services received by a patient o Upcoding ( defined above) o Experimentation with the Medicare outlier methodology Unbundling Charges/Fragmentation o Submitting charges of a comprehensive code as well as one or more component codes o Billing for m utually exclusive procedures o Global service period violations billing for a major procedure (e.g., sur gery) as well as related procedures, when the fee for the major procedure already includes fee s for related procedures within the global service period (pre defined period of which all related services are covered following a major procedure ) (ACFE 2005 06 ) Focus of the Investigation This study will investigate other type s of gaming, one s that are not usually listed in a typology of gaming behaviors. These include hospitals systematically increasing the number s of certain types of discretionary procedure s performed under conventional Medicare and inflating charges under the Medicare HMO program. The example cited above of fraud at Redding Medical Center initially inspired this research.

PAGE 19

5 The study period for this project covers years 2000 2005 which begins soon after the passage of the Balanced Budget Act (BBA) of 1997 This act was an omnibus legislative package designed to balance the federal budget by 2002 with expectations of savings from Medicare of $112 billion between 1998 through 2002 (Guterman 2000) It introduced major Medicare reforms which included reduced payments to providers (Doherty Date Unknown ) though payments were raised in subsequent legislation under the Balanced Budget Reform Act (BBRA) of 1999, and certain reimburseme nts were reinstated in 2000 (Doherty Date Unknown, Guterman 2000 ) On the surface t his would appear to be an era when Medicare may not be a profitable revenue source for hospitals. D espite this surface appearance there were some Florida short term gener al hospitals that maintained unusually high profitability while at the same time deriving high percentages of their annual total revenue from Medicare during the years of this study, 2000 2005 High profitability defined in this study ( and attributed to t he group of hospitals just mentioned) is based on operating margin as a short term assets total liabilities (TATL) ratio (the reciprocal of the debt ratio) as a long term (> one year) indicator derived from In the dataset that wil l be used for part of this analysis (described on page 67 ) observations are provided for 891 Florida hospitals between 2000 and 2005 The dataset shows amounts of net revenue derived from Medicare for each hospital as well as t wo important financial indic ators, the operating margin and the TATL ratio, just mentioned Preliminary analyses showed that certain short term general hospitals with high had

PAGE 20

6 0 15 .0 The magnitude s of operating margin s and TATL ratio s in these hospitals can be better understood by a brief discussion of the scales of these financial indicators in the dataset. The dataset showed that of the 891 hos pitals t he mean operating margin was 2.25 % and the median is 3.3 0% The third quartile is 9.5 % while the 95 th percentile is 19.5 % The mean average for the TATL ratio is 3.92 with a median of 1.94. The third quartile for the TATL ratio is 4.33 and the 9 0 th percentile is 1 2 00 Considering the scales of operating margin and the TATL ratio among total hospitals it is clear that the Medicare, are very profitable. The research question is what is different about certain hospitals, that they could realize inordinately high profitability when their largest r evenue source, Medicare, is consistently identified in the healthcare industry and in the literature as unprofitable (Williams & Hadley 1992, Younis & Forgione 2005) One explanation for higher profitability and better financial health in these hospitals m ay be due to better management, which would entail prudent cost cutting and highly efficient service delivery. Also because of sound strategic direction these hospitals may have structured their markets to offer a very profitable patient mix of DRG proced ures and are extremely competent in delivering these services to maximize both efficiency and quality. Still another possible explanation, because the ir operating margins and TATL ratios are so high, would be that these hospitals (the physicians on staff t here) are

PAGE 21

7 performing higher volumes of certain procedures, some of which may be questionable or perhaps marginally necessary. Additionally, these hospitals may be assessing monetary charges for certain procedures that exceed the extent of medical services performed. Because hospital administrators investigations for upcoding and excessive use of the outlier methodology, certain hospitals may be using a strategy of boosting profitability by overusing or overcharging of certain DRG codes. It was found that hospitals that are profitable and also have high percentages of their revenue base comprised of Medicare payments carry out high rates of surgical procedures described as The definition of discretionary procedure will be examined in detail in Chapter 3, Research Methods Data Structure & Identification of Variables but in short it refers to a procedure for which there is a lack of full scientific agreement regarding its effectiveness. D iscretionary would also mean that the procedure is performed in variable per capita rates per geographic area or region. It is distinguished from a nondiscretionary procedure for which clinical agreement has been established and medical practice is sta tistically uniform regardless of the hospital or geographic area. If financial gaming occurs in connection with discretionary procedures financial incentives and medical practice would differ depending on the type of Medicare payer. Conventional Medicare reimburses procedures on a fixed rate per DRG code while Medicare HMO compensates hospitals on a charge basis. The incentive for hospitals to make a profit with Medicare HMO would be to assess charges higher than are warranted by the extent of medical se rvices provided Even if Medicare agrees to pay only a

PAGE 22

8 fraction of patient charges the hospital may still financially benefit if reimbursement turns out being higher than it would have been had charges been at realistic levels. The incentive under convent ional Medicare would be to increase volumes of certain types of procedures for which the costs of care would be, on the average, consistently less than the DRG reimbursement. Overcharging for medical services or overusing certain DRG codes may be two of th e strategies used by certain hospitals that may allow them to be very profitable while depending so heavily on Medicare as their primary revenue base. As will be shown in the Research Methods s ection s (chapters 3 6) this study will be presented in two p arts, Part 1 and Part 2. Part 1 will test the association between financial performance and discretionary cardiac and orthopedic procedures carried out in investor owned hospitals. The population in Part 1 will be confined to general acute care Florida h ospitals, 2000 2005. Explanatory variables will include percentages of certain cardiac and orthopedic procedures performed per hospital, reimbursed by conventional Medicare, 2000 2005. The dependent (outcome) variable will be the TATL ratio the recipro cal of the debt ratio Part 2 will test the association between total patient charges and discretionary cardiac procedures carried out in investor owned hospitals. The population will be confined to valve replacement and CABG procedures reimbursed by Medi care HMO and the outcome variable will be total charges. The explanatory variable will be hospital ownership.

PAGE 23

9 Chapter 2 Review of the Literature Introduction The Review of the Literature is intended to present rationale for carrying out this inves tigation and to provide a theoretical basis for setting research objectives deriving hypotheses, and selecting variables Brief Summary of the Literature Review The review begin s with an overview of hospital financial performance indicators and a discuss ion of factors which may favorably or adversely affect these indicators. This will help determine which indicators might be useful as variables in the Research Methods section (Chapter 3) and will establish a foundation for the selection of certain contro l variables. One factor often adversely associated with financial performance indicators is a (Younis & Forgione 2005 ) A review of previous research will show that Med icare is often negatively associated with profitability, and as explained in the p roblem s tatement, it is counter intuitive that a group of Florida hospitals with high percentages of Medicare comprising their revenue structure s to simultaneously have high operating margins and other favorable financial performance indicators.

PAGE 24

10 The lite rature re view then cite s investigators that have concluded percentage of Medicare revenue is irrelevant, that good management regardless of payment sources or patient mix is the key to profitability (Clarke 1991, Cleverly 2004) These investigators also conclude that, though there are inherent differences between investor owned and not for profit hospitals, ownership per se does not account for strong financia l performance. I t is the characteristics of many investor owned hospitals that make them more profitable such as reduced labor costs, better long term capital planning, etc. (Galloro 2004, Smith 2002). These characteristics are often induced by boards o f directors in investor owned hospitals that tend to pay closer attention to the bottom line (Smith 2002 HFMA 2004 ) Although profitability is positively associated with good management practices, there are also other mechanisms that may improve financial performance. For example, certain Florida hospitals could be practicing gaming behaviors especially considering the excessively high operating margins that are characteristic of certain Florida hospitals with Medicare as their greatest revenue source. G aming has already been defined in the problem statement as recurrent maneuvering across legal and/or ethical thresholds of conduct for the purpose of ostensibly improving financial performance. Gaming behaviors in previous research have focused on upcodi ng and gaming the DRG outlier methodology, but no research has addressed performing inordinately high volumes of discretionary procedures, some of which may be unnecessary, as a gaming strategy. Also, very few previous studies have examined charges asses sed by profitable hospitals to determine if these charges are significantly higher than other hospitals that

PAGE 25

11 carry out the same medical treatments and perform the same extent of medical services. A good example of why such a study might be important is Re dding Hospital, a subsidiary of Tenet Healthcare, which was found performing unnecessary heart surgeries submitting false billings and overcharging for services (Darmiento 2002). The l iterature r eview lays the theoretical groundwork for identifying poten tial DRG codes which may b e subject to gaming behaviors. As the literature review ensues it will be shown that DRG codes vulnerable to gaming tend to represent discretionary procedures. The criteria to determine a procedure as discretionary in this study are based on lack of clinical consensus of effectiveness and variability in the performance of these procedures across geographic areas that are not linked to rates of disease (Wennberg 2005, Dartmouth 1999). The literature review will then discuss an ass ociation between the rates of discretionary ( and possibly gamed ) procedures and geographic variations of medical practice patterns. Numerous investigators are cited who provide evidence that rates of certain procedures and per capita Medicare expenditures var y widely depending on location. Wide fluctuatio ns are observed between states and selected cities in the nation and between cities and hospital referral regions (HRRs) in Florida. (2005) surgical signature theory will be introduced in the literature review and referenced several times in this study. The surgical signature is identified by w ide geographic variability in f ederal spending and in the performance of discretionary procedures reimbursed by Medicare

PAGE 26

12 Indicators of Financial Per formance Many studies use operating margin (operating income total operating revenues), the proportion of profit remaining after subtracting total operating expenditures from operating revenues (Cleverly & Cameron 2002) as a dependent variable because i t is an excellent measure of financial performance. Operating margin is a summary measure of a income statement reflecting profitability typically over a one year period. It reflects ongoing revenue and expenses over a short term (one y ear) period of time The TATL ratio (total assets total liabilities) is a reliable summary indicator of the balance sheet reflecting long term solvency, viability, and the risk of bankruptcy (Gardiner 1995). In the composition of the TATL ratio, total l iabilities include current and long term liabilities and current and fixed assets as well as other assets such as goodwill if relevant. The TATL ratio is the reciprocal of the debt ratio. The debt ratio may be expressed as ( total liabilities total ass ets ) or ( total debt total assets ) It indicates the ( i.e., n et assets or fund balance in a not for profit s balance sheet ) (Williams, Haka, Bettner et al 2008) The r equity and debt. It can help diagnose the overall financial risk confronting an organization, as generally the greater amount of debt held by an organization the greater the financ ial risk of bankruptcy (Williams, Haka, Bettner et al 2008)

PAGE 27

13 As the inverse to the debt ratio, the TATL ratio also is an indicator of liquidity vs. leverage and is a measure of both cash flow and balance sheet solvency. It reflects a snapshot of long te rm financial condition at a given point in time (Plank & Plank 2000). Other long term measures instead of the TATL ratio were considered, such as the debt to equity (fund balance) ratio and the total liabilities to net equity (fund balance) ratio. But d ata generated for the TATL ratio was cleaner, had fewer missing values, and was determined to be more reliable for the analysis. One problem for the l iterature r eview was that, though there are many research findings using operating margin as well as total margin that can be cited, there are virtually none that use a TATL ratio. For that matter there are very few studies in the healthcare literature that have used other financial ratios as dependent or independent variables Because so many studies use s hort term measures of financial strength, such as operating margin or total margin as outcome variables, and this study uses the long term TATL ratio, it seemed necessary to present findings from the literature regarding ot her useful (long term) financial indicators, which are provided in the following subsection. This information may establish further rationale for using a long term financial indicator such as the TATL ratio. Multiple Indicators of Financial Performance Despite the consistent use of opera ting margin or total margin as outcome variables in studies in the healthcare literature, these margins represent just one dimension of financial health and are just two among many indicators of financial performance. Cleverly (200 4 ) and Lee (1984) stress the importance of examining financial data representing multiple dimensions of the organization This means that the

PAGE 28

14 assessment of financial performance necessitates a simultaneous understanding of a comprehensive set of indicators including liquidity, p rofitability, activity, and debt, as well as determining the i nterrelationships of these indicators (Lee 1984) Several financial indicators representing long term dimensions are presented in the next few paragraphs most of which a re derived from items o n the balance sheet and can be en capsulated with in the TATL ratio. term viability debt leverage, These indicators of long term risk ar e derived from the balance sheet as is the TATL ratio Operating margin or total margin derived from the income statement are not indicator s of long term risk, but only of short Schulman (2008) has explained that hospital financial performance must be examined via separate analyses using operating margin, liquidity ratios, debt leverage ratios, LOS/patient di scharges, and average payment period ratios. Each tool reflects a different aspect of the organization These separate assessments should be reviewed with a comprehension of how they interrelate, over one and five year periods to be assured of gaining b oth short and long term perspectives of financial health. Cleverly (200 4 ) states that, though total margin is an excellent indicator of a [(gain on investment cost of investment) cost of investment] is the most reliable overall measure of long term profitability because financial performance is measured in terms of invested capital.

PAGE 29

15 line is also critical, 4 ). Capital planning is dependent upon addressing questions of how capital requirements affect required margins by s pecialty area, and the level of capital that will be needed for future investment Healthcare organizations that practice systematic capital planning are better able to judiciously invest in long term assets to modernize plant operations while skillfully negotiating the most favorable terms to finance improvements ( HFMA 2004, Cleverly 2004 Shattuck Hammond 1997 ) This practice generally results in a high TATL ratio that is sustained for many years. Financial ratios and margins are the best indicators (on paper) of financial performance, but it must be remembered that even they do not always provide a completely clear representation of financial health because they are indicators that must be extracted from financial statements For example, off balance s heet (risks) liabilities are items not reported in the body of financial statements as a liability, but represent that the organization may be required to incur severe future financial implications. Off balance sheet liabilities often include litigation, guarantees of future performance, and renegotiations of claims under a government contract (Sieg e l & Shim 1995 Plank & Plank 2000 ). T here is no way to reflect off balance sheet liabilities in ratios or margins and no way to account for them in this stu dy. Fortunately, substantial off balance sheet liabilities are not frequent and their presence in connection with certain hospitals should not affect overall findings.

PAGE 30

16 M edicare is N ot a Profitable Revenue Base ? Factors most often cited in connection with poor hospital profitability are aging of plant facilities, hospital location in economically disadvantaged areas (B rown 2001 Younis & Forgione 2005 ) high debt (Vogel, Langland Orban & Gapenski 1993) too few or too many beds (Kim Glover & Stoskopf et a l 200 2 ) low patient occupancy (Langland Orban & Gapenski 1996 Younis & Forgione 2005 ) high LOS (Younis 2004) and high percentages of uncompensated care and delinquent accounts (Brown 2001) In addition to the factors mentioned above a number of stud ies will be cited in the next few paragraphs that indicate that high volumes of Medicare (and Medicaid) patients are a negative determinant of hospital profitability, hospital administrators and physicians pective payment system (and the Balanced Budget Amendment of 1997). Younis & Forgione (2005) sampling not for profit short term community hospitals and using total profit margin (i.e., margins including non operating revenue and expenses net of taxes and extraordinary items) as the measure of profitability, found that profitability is adversely affected by low occupancy, hospital location in poor neighborhoods, and high volumes of Medicare and Medicaid patients. Williams Hadley & Pettingill (1992) found that hospital profitability was adversely related to rural location as well as a high percentage of Medicare patients, while it was positively associated with larger hospital size, larger community population, and high patient case mix. Carpenter & McCue (200 1 ) noted that investors lack confidence in hospitals with greater dependencies on either Medicare or Medicaid reimbursement because these

PAGE 31

17 hospitals are perceived as being vulnerable to political uncertainty and budgetary retrenchment. Investors also p erceive high percentages of Medicaid patients to be concomitant with serving a disproportionate share of uncompensated care patients. Jantzen & Loubeau (2003), using a sample of hospitals in the United States that issued fixed rate bonds in 1995 and 1996, found that LOS was one of the factors of to keep LOS in check to make a profit with Medicare. hos performance hospitals were also identified as having significantly higher operating margins tha n low performance hospitals, as well as better liquidity and newer plants and equipment. control which may contribute to poor financial performance. Clarke found that low per formance hospitals had higher Medicare and Medicaid caseloads and tended to be Conversely, Silverman Skinner & Fisher (1999) shed lig ht on Medicare as a potentially profitabl e revenue source. They studied hospital conversions between 1989 and 1995 and found that not for profits recently converted to investor owned status tend to be the most aggressive in seeking profit and cutting costs. The study entailed an examin ation of total per capita Medicare spending in areas served by for profit and not for profit hospitals, as well as for profit hospitals that had recently been converted from

PAGE 32

18 not for profit. Medicare expenditures were examined in 208 areas where all hospitals (in t hese areas) remained under for profit ownership (i.e., all beds in the area were in for profit hospitals), 2,860 areas where all hospitals remained under not for profit ownership (i.e., all beds in the area were in not for profit hospitals), and 33 areas w here all hospitals had been converted from not for profit to for profit ownership. Federal spending was defined as Medicare expenditures per area and included hospital services, physician services, home healthcare, and services at other facilities. Expen ditures for each health service category were examined separately. It was found that per capita Medicare spending was greater among all health services categories in areas containing for profit hospitals vs. areas containing not for profits. It was also found that per capita Medicare spending grew more rapidly in the 33 areas containing recently converted hospitals relative to the 2,860 areas containing not for profits. This is an important finding because it shows that recently converted hospitals had pe rhaps identified effective strategies to make a profit with Medicare R ecently converted hospitals would likely need to identify profitable revenue sources as o wnership conversion requires restructuring of the organization to satisfy new shareholders and to begin meeting a new liability, the tax burden. It may not be of Medicare revenue to total revenue) per se that contributes adversely to financial performance. Instead it may be the ty pes of DRG codes that are emphasized (utilized with greatest frequency) by hospitals under Medicare that contribute to high or low financial performance G apenski, Vogel & Langland Orban (1993) found that patient mix variables are very useful predictors of profitability. They also found that higher

PAGE 33

19 hospital profitability is associated to a higher case mix index (CMI ) ( i.e., average DRG and Medicare mix (Langland Orban, Gapenski & Vogel 1996). Good Management is the Strongest Determinant of Profitability Some researchers contend that pay er mix is irrelevant, as it is good management and sound strategic direction that results in profitability and long term financial health for hosp itals These researchers explain that the determinants of hospital profitability include market factors, operational efficiency, patient demographics, patient occupancy, cost control, structure of the labor mix, LOS hospital physician relationships, and recent changes in f ederal and state legislation. Determinants of financial health associated with the physical structure of the hospital include geographic location aging of plant and equipment, and econom ies of scale (numbers of beds). In the contex t of plant and equipment, a primary determinant affecting a hospital ability to generate capital to invest in new technologies and upgrade aging facilities. The ability to simultaneously m anag e and strategically negotiat e t hese determinants of hospital profitability ( and long term viability ) listed above reflect what most researchers identify as good management practices. as just mentioned in the discussion on Medicare. Although Clarke conceded that high volumes of Medicare and Medicaid patients and tight regulatory requirements hampered there were some hospitals under these constraints that sti ll performed well. Clarke found that high performance hospitals used 15 percent fewer

PAGE 34

20 labor hours per case (adjusted for case mix and outpatient volume) than low performance related expenses represent more than 60 percent of al l operating expenses for hospitals, labor usage is an important factor in financial stating that effective hospital management is the key to successful financial performance regardless of ownership, patient mix financing sources, etc. He buttresses this position by citing that payment mechanisms do not differ materially between profitable and un profitable hospitals but financial performance is enhance efficiency and control costs. Explaining how hospitals with high numbers of Medicare patients remain financially healthy Cleverly (2004) cited findings from a sample of 3,076 hospitals, including 971 teaching hospitals and 2,105 non teaching ho spitals. Cleverly reports. than not for profit s not due to ownership status per se, but because of more efficient operations and better strategic direction. Investor owned hospitals are generally more aggressive in optimizing their labor force, eliminating unprofitable services, and tai loring their mix of services with the intent of realizing highest reimbursements at lowest treatment costs (Galloro 2004). Sear (1991) reported findings, consistent with Cleverly in examin ing a sample of 50 investor owned and 60 not for profit hospital s in Florida between 1982 and 1988

PAGE 35

21 Sear found that wages per adjusted patient day and LOS were the most important fac t ors in explaining why investor owned hospitals are more profitable. Garrison & Wayne (2008) found that ongoing data examination by mana gement of LOS, especially severity adjusted LOS, and disciplined efforts to reduce LOS are among several characteristics associated with successful hospitals. Schulman (2008) found that hospitals are more likely to have higher operating margins if they ha ve shorter LOS, lower personnel cost percentages, and higher outpatient gross revenue percentages. Schulman (2008) also found that investor owned hospitals financially out per formed not for profits. Between 2003 & 2007 operating margins ranged from 5.45 to 8.87 for investor owned hospitals while not for profit margins ranged from 1.86 to 0.54. The difference in financial performance was attributed in part to investor owned h ospitals incurring 20 percent less in personnel costs than not for profits. Investor owned hospitals also had lower accounts receivable days and wrote off delinquent debt more rapidly. Average LOS was also found to be slightly less in investor owned vs. not for profit hospitals. Profitability after Regulatory Changes Sear (2004) demonstrated that good management can often transcend legal and regulatory changes that affect revenue such as the BBA s effects on Medicare and Medicaid Using a sample of 25 a cute care hospitals, Sear examined operating margins before and after passage of the BBA, between 1990 and 1997 and between 1998 and 2001. Certain hospitals with negative margins, ostensibly caused by BBA, actually had negative margins for the entire 12 y ear period, before and after BBA.

PAGE 36

22 Other hospitals experienced cyclical margins moving higher and lower and not concomitant with BBA. Still other hospitals maintained strong operating margins over e hospitals established management strategies to adapt to changes in the reimbursement and expense the percentage of operating revenue derived from Medicare does not account for a sig nificant amount of the variance in average operating margins in the post Sear showed that poor operating margins could not be attributed to BBA, but to poor management before and after its passage. S ound management, cost control, and strategic direction may be the most important determinants of profitability and may assure adaptation to changes in f ederal reimbursement and market structures Managing Assets to Raise Capital & Assure Future Profitability As already mentioned, the o utcome variable for Part 1 of this study is the TATL Ratio. A high TATL ratio often reflect s sound strategic direction for long term investments and covering of long term liabilities. For long term viability hospitals should define and estimate their fut ure available capital capacity, prioritize capital spending, and their capital capacity analysis should dovetail with strategic financing issues. This should result in a systematic strategy for deriving capital for long term investments such as upgrading plant and equipment and restructuring debt Gapenski Vogel, Langland Orban (1993), for example, found that aging of plant facilities as well as high debt were both negative determinants of profitability. These negative determinants can be many times mit igated or reversed by methodical long term capital planning. But for many hospitals long term planning is not systematic. Shattuck

PAGE 37

23 Hammond (1997) reported that large sectors of the health care industry plan only for the annual budget, and long term invest ments are based on management or board intuition. As discussed in more detail later in the l iterature r eview boards of directors in investor owned hospitals may engender an environment that is more conducive to long term systematic corporate planning th an not for profits. It would not be surprising for a n organization guided by decisions that have a factual basis and long tem perspective to be financially stronger than one that is guided by ambiguity and visceral perception Management & Strategic Dire ction in Initial Location Decisions Geographic location is described in the Research Methods s ection s (chapters 3 6) as an important control variable. Previous studies have found an association between profitability and geographic location and have found that many hospitals located in more affluent areas are investor owned. Investor owned hospitals tend to comprise about 15 percent of the hospitals in most areas, and they tend to concentrate in the most desirable and potentially lucrative locations, resu lting in the exclusion of economically disadvantaged patients (Brown 2001) These areas often include elevated densities of highly reimbursed physician specialties. However, this does not imply there are differences in objectives among ownership types. N ot for profits do not necessarily locate in economically disadvantaged areas because it is part of their organizational mission. In fact, not for profits would more often locate in higher income areas if they could. Investor owned hospitals are generally more likely to possess the ability to shift capital to higher income, new growth communities with greater celerity than not for profit hospitals.

PAGE 38

24 Investor owned hospitals are not always able to avoid poor neighborhoods, as location is driven by market d ynamics. The ability to shift capital and locate hospitals in upper income areas diminishes in the presence of increased spatial competition. When investor owned hospitals are late entering the market they may find rigid densities of competition already ensconced in wealthier areas, circumventing their ability to shift capital for geographic preference (Brown 2001). In Nashville, Tennessee, for example, because about half the hospitals are investor owned spatial competition forces certain investor owned hospitals to locate in poor areas. In fact, though most maternity patients covered under Medicaid and them seek care from investor owned hospitals than not for profit s ( Brown 2001). Hospital Ownership As will be discussed in Chapter 3 hospital ownership is one of the most important variable s in model development of this study. Ownership is discussed in the context of profitability in several other places in the l iter ature r eview. But because ownership is an important variable in this study, a separate discussion is devoted to it in this subsection. In general, investor owned hospitals have higher operating margins and TATL ratios than their not for profit counterpart s and tend to be represented across the continuum by more favorable financial indicators. Investor owned hospitals also tend to be more capable in raising essential long term capital which allow s them to more effectively modernize facilities, invest in th e most current medical technology, and restructure debt. These long term capital investments, in turn, help investor owned hospitals to maintain favorable financial performance indicators.

PAGE 39

25 Securing Ongoing Capital L ong term investments which assures lo ng term sustained profitability can be financed by investor owned hospitals via contributed capital, often through public stock issuances (Smith 2002). F or example, Triad Hospitals, Inc. accumulated $ 85.4 million shares which yielded $189 million in net p roceeds in connection with a stock offering in March 2005. The net proceeds were used for capital expenditures for expansion of working capital, and repayment of in debtedness (Triad Hospitals, Inc., Supplement to Prospectus March 31, 2005). Not for N ot for profit hospitals cannot issue stock to raise capital but they do have the benefit of a tax exemption that may at least partly bala nce their disadvantage of not being able to make stock offerings (Smith 2002) It warrants mentioning that just like investor owned hospitals not for profits can raise large amounts of capital via taxable and tax exempt bond issuances, bank loans, philant hropy, joint ventures, etc. It should also be pointed out that the tax exemption available to not for profit hospitals include s f ederal income taxes, Florida corporation income taxes, state sales taxes, local property taxes, etc. The f ederal, state, and local tax exemptions aggregated together represent a very large monetary amount. Some researchers argue that the benefits accruing to not for profits from a tax exemption are depleted because they must absorb higher levels of uncompensated care and Medicai d patients and they offer needed community services that are not induced by a profit motive Younis (2004) found that not for profit hospitals had longer LOS for

PAGE 40

26 Medicare patients, but concluded that this enhanced the quality of care for these patients w hich was consistent with a more benevolent organizational mission of not for profit hospitals. Herzlinger & Krasker (1987) reached a different conclusion regarding community service and charitable care objectives exercised by not for profits which presumab ly serve s to drain them of their tax exemption advantage. Herzlinger & Krasker explained that, not for profits, though they have the advantage of a tax exemption, do not return greater investments in community service. Therefore, not for inabili ty to financially perform as well as investor owned hospitals could not be explained by their expenditure of resources to benefit the community. Herzlinger & Krasker expressed that a serious question arises when not for profits behave like investor owned hospitals in terms of abandoning a community purpose or systematically reducing services to less profitable patients, such as Medicaid and uninsured patients. If a not for profit behaves like a n investor owned hospital in this context, why should the not for profit continue to enjoy the advantages of tax exemption? Wolfson & Hopes in South Central Florida reveals that almost all institutions benefited financially from their tax exempt s If these conclusions are accurate that (some) not for profit s generally do not provide public service benefits in exchange for their tax exemption then it could be concluded that not for profit s have a financial advantage over investor owned hospitals. The tax exemption may indeed help to balance certain disadvantageous incurred by not for profits relative to investor owned hospitals, such as the inability to issue stock.

PAGE 41

27 Perhaps certain factor s such as the ability to issue stock does not fully account for why investor owned hospitals tend to outperform not for profits. Cleverly (2004) has expressed that it is not ownership status but characteristics that are often associated with investor owne d hospitals, greater efficiency and sound strategic direction, that make them more profitable. Streamlined decision making allows investor owned hospitals to quickly shift capital when necessary, subsequently incurring less costs of capital for investmen ts. This gives them the ability to raise sufficient capital to invest in plant and equipment modernizations, restructure debt, organize the case mix to offer procedures that maximize profitability, and to market the most current treatments and procedures m ade possible through advances in medical research. The Inurement Principle & Pressures from Governing Boards A f undamental reason investor owned hospitals generally are more profitable than their not for profit counterparts is because of inherent differenc es in how these organizations relate to their governing boards. Not for profits, though they are allowed to make a profit, are bound by the doctrine which prohibits them from distributing net earnings for the benefit of private shareholders [(I RC Section 501 (c) (3)]. They are governed by voluntary boards with diverse constituencies, which generally results in protracted decision making and less incentive to aggressively optimize the labor force, eliminate unprofitable services, and tailor thei r mix of services with the intent of realizing highest reimbursements at lowest treatment costs (Galloro 2004 Smith 2002). Investor owned hospitals distribute a portion of their profits back to their shareholders which stimulates focus on the bottom line and pressure for management to

PAGE 42

28 continually adapt to market forces and streamline operations (Smith 2002). The motivations of corporate boards can sometimes place exacting and even unrealistic profit goals upon management which results in very efficient o perations but also can lead to financial statement misrepresentation and gaming payment systems (ACFE 2005 06) aggressively pursuing revenue goals and gaming Medicare. But not for profit hospitals sometime also feel the pressu re from governing boards to improve financial and operational performance and they often respond by emulating behaviors of investor owned hospitals. Galloro (2004) reported that behavioral diffe rences between not for profit and investor owned hospitals diminish when tax exempt hospitals are confronted with increasing competition. Not for profits in the investor owned hospital competition began mimicking investor ow ned behaviors with respect to cost control, labor utilization, service mix, and even the likelihood of upcoding to maximize reimbursement through the use of diagnostic coding. Findings by Silverman Skinner & Fisher (1999), discussed in a previous section, showed that hospitals recently converted from not for profit to investor owned status were much more aggressive in seeking profit than not for profits or even other long term investor owned hospitals. Needleman Lamphere & Choilet (1999) found that hos pitals, formerly public or not for profit, recently converted to investor owned status tend to be the most aggressive in reducing uncompensated care, even more aggressive than hospitals that have been investor owned for a long time.

PAGE 43

29 This shows that inheren t behavioral differences between investor owned and not for profit hospitals do not originate with ownership status per se but with the objectives of governing boards rising competition, and how hospital administration is expected to respond to changes in hospital structure, healthcare markets, and the bottom line. Does Good Management Fully Explain Inordinately High Profitability There is no doubt that findings reported by researchers in the previous section s of this chapter regarding the association bet ween profitability and good management are accurate. Organizations that can be quickly and decisively aligned to changes in market and regulatory environment s will be more profitable than hospitals without these characteristics and more favorable measure s of financial performance for well managed hospitals should be expected. It should be anticipated that the ability to restructure debt and finance facility and equipment modernizations should result in long term financial health and viability. The ques tion is does good management alone account for financial indicators such as operating margins and TATL ratios being as inordinately high for certain investor owned hospitals as those described in the p roblem s tatement At what point does efficiency begin to encounter diminishing returns in its contribution to high profitability ? Even if a hospital financial condition is made stronger by reductions in p ersonnel or LOS, for example, there is a floor whereby certain resources cannot be reduced any further. There may also be a ceiling of which financial indicators cannot climb beyond due only to sound management practices and fiscal discipline Operating margins and TATL ratios that are higher than what would be explainable by good management may be due to gaming payment system s either by

PAGE 44

30 artificially increasing codes or charges and/ or by driving up v olumes of profitable procedures. But how prevalent is gaming in healthcare a nd how surpris ing should it be if gaming strategies are partly responsible for fina ncial health in some hospitals ? T he following two section s address these questions by providing a historical overview of healthcare fraud from academic and trade literature and real world prosecutions. Investor Owned Hospitals Propensity for Gaming Medi care Investor owned hospitals have reportedly carried out gaming behaviors with greater frequency and at more egregious levels than not for profit s. I nvestor owned hospitals may be more aggressive or proficient in gaming payment s ystem s than not for profi t s in the hospital industry and historically they ha ve been particularly bold in gaming Medicare. This section provides an overview from studies done by Silverman & Skinner (2001) and Dafny (20 03) on gaming procedures and payment systems in the context of investor owned hospitals. Silverman & Skinner found that investor owned hospitals are more profitable than not for profit s, but the greatest difference between the two was that investor owned hospitals were generally more involved in the shifting of pati that yield higher reimbursements from the Medicare system, i.e., upcoding. Silverman & Skinner also cite a history of lawsuits in connection with upcoding against hospitals and hospital chains, explaining that these gaming behavio rs provide a profit and non profit hospitals make tradeoffs between pecuniary benefits and reputational or penalty costs Skinner 2001).

PAGE 45

31 Silverman study focused on hospital admissions invo lving pneumonia and respiratory infections. These two diagnostic categories are often difficult to distinguish from each other, but a diagnosis for respiratory infection at that time paid about $2,000 more to the hospital. Between 1989 and 1996, the incid ence of the most expensive DRG codes for pneumonia and respiratory infection increased by ten percent among stable non profit hospitals. Conversely, this same diagnosis increased by 23 % among stable investor owned hospitals and rose by 37 % for hospitals, f ormerly not for profit that had been converted to investor owned status There was also evidence that no t for profit hospitals operating in heavily investor owned markets were almost as likely to upcode as were investor owned hospitals. Silverman & Skinne r conclude their discussion by stating that after 1996, the upcoding index ha d dropped significantly in response to adverse publicity and lawsuits. Nevertheless, other lucrative gaming strategies would likely evolve. Dafny (2003) examined a 20% sample of all hospitalizations of Medicare enrollees from the 1985 1991 Medicare Provider Analysis and Review data to assess responses by hospitals to relative price changes under Medicare PPS and other public and private insurers. nable to isolate price change responses because policy change that generated large price c hanges for 43 percent of all Medicare eliminate age as a criterion as part of certain diagnostic categories.

PAGE 46

32 des were paired for each diagnosis. For example, cardiac arrhythmia would have one code for patients aged 70 and above, or with complications, and a second code was used for patients under 70 patients without complications were similar, regardless of age. Dafny then recalibrated DRG prices to eliminate age as a criterion for data beyond 1988. This resulted in an average 11 percent increase in DRG prices for the top codes in each pair (those with complications) and an average six percent decrease in DRG prices for the bottom codes. Dafny found that hospitals responded to these price changes by upcoding patients to diagnostic codes associated with larger reimbursements, resulting in an estimat ed additional reimbursement of $330 $425 million annually. This response to price changes, as reported by Dafny, was sophisticated, with more upcoding in DRG codes where the spread between top and bottom codes had increased the greatest. Dafny also empha sized that responses to price changes were particularly strong among investor owned hospitals. With the exception of elective diagnoses, Dafny found increases, wh ere intensity is measured by total costs, length of stay, number of surgical procedures, and number of intensive care volume of patients admitted to more remunerative diagnoses, notwithstanding the strong a p riori expectation that such a response should prevail in fixed price settings. Dafny concluded that hospitals, particularly investor owned do not alter their treatment or admissions policies based on diagnos t i c specific prices. Instead they employ

PAGE 47

33 sophis ticated coding strategies in order to maximize total reimbursement. It was concluded from this study that changes in DRG levels were unrelated to changes in costs. Prevalence of Health Care Fraud Uncovering gaming behavior in healthcare is not isolated t o a few academic studies such as Silverman historically these behaviors cover the gamut of the health care industry. According to a 1993 survey by the Health Insurance Association of America (CMS, September 2004) fraudulent act ivities in health care break down as follows in Table 1 Table 1 Distribution of Fraudulent Activities in Healthcare 1993 Fraudulent Activity Percentage Fraudulent Diagnosis 43% Billing for Services Not Rendered 34% False Reporting of Patient Deduct ibles and Co Payments 21% Other 2% Source: CMS, 2004 Specifically for Medicare, CMS (September 2004) and Benson (2005) identified the most common forms of fraud which include: Billing for services not furnished (or not furnished as billed upcodin g ) Misrepresenting the diagnosis to justify payment Soliciting, offering, or receiving a kickback Carrying out plans of treatment and falsifying medical records to justify payment Fraudulent c ost r eports Kickbacks and s elf r eferrals Grant or r esearch f raud The following paragraphs present only a few cases of the numerous prosecutions each year show ing the magnitude of tax dollars lost for healthcare organizations caught gaming Medicare.

PAGE 48

34 T he U.S. Department of Health & Human Servic Office of the Inspector General (OIG) reported on February 11, 2004 that under a settlement agreement Greenville, SC based St. Francis Hospital, Inc., agreed to pay almost $9.5 million to resolve Medicare improprieties from 1997 1999 in its home health hospice, and durable medical equipment programs. This represented one of the largest OIG settlements obtained under the Self Disclosure Protocol (Compliance Monitor, February 18, 2004) The OIG and United States Attorney for the District of Maryland ( H HS OIG February 14, 2003 resolve charges of fraudulent Medicare billings. The settlement resolve d a f ederal investigation under the False Claims Act (31 USC § 3729 3733) arising from alleg ed false and improper billings for services of teaching physicians to Medicare beneficiaries covering calendar year 1994 (HHS OIG February 14, 2003) In December 2000, HCA (formerly known as Columbia HCA), the largest investor owned hospital chain in th e United States, pled guilty to criminal conduct and agreed to pay more than $840 million in criminal fines, civil penalties and damages for unlawful billing practices. Of this amount, $731,400,000 was recovered under the False Claims Act. Under the settl ement agreement, HCA's payment was to resolve several allegations regarding the manner in which it billed the f ederal government and certain states for health care costs ( DOJ, December 14, 2000). The settlement for HCA was designed to resolve allegations f or billing for lab tests that were medically un necessary and not ordered by physicians u pcoding patients medical problems in orde r to get higher reimbursements b illing Medicare for advertising

PAGE 49

35 under the guise of "community education and b illing the f ederal government for non reimbursable costs incurred in the purchase of home health agencies around the country ( DOJ December 14, 2000). Other allegations for HCA were not resolved by the settlement which i nclude d HCA unlawfully charg ing the f ederal go vernment costs for operating its hospitals and paying kickbacks to physicians to get Medicare and Medicaid business ( DOJ, December 14, 2000). Ten e t Billing and Medical Malfeasance Cases of b illing fraud, financial statement misrepresentation, defalcation kickbacks, corruption, etc. have been shown to be abundant in the healthcare industry. But far more sinister types of fraud have also occurred which have had negative effects billing payment systems (HHS OIG Decemb er 11, 2006, DOJ May 17, 2004) On Oct. 31, 2002 Tenet Healthcare Corp., the n the investor owned hospital chain, disclosed that f ederal prosecutors were investigating allegations that two doctors at Tenet subsidiary Redding Medical Center in Redding, CA, conducted unnecessary heart procedures and billed falsely for them. Later it was found that Tenet had been charged with medical malfeasance affecting at least 750 former patients, primarily recipients of CABG surgeries, valve repla cements, and catheterizations who subsequently pursued litigation for medical malpractice and fraud. healthcare entities and encountering legal damages that could amount to settlements in excess of $50 Million ( DOJ May 17, 2004 ). Tenet stock also went down by $20 (from

PAGE 50

36 $49.31 down to $28.75 per share ) on the trading day the information was released, representing a decline in market value greater than $10 billion (Darmiento 2002). It was also discovered in connection with allegations of unnecessary surgery that Tenet competitors. It was estimated that outlier payments comprised 16.7 % or $418 million of It was also found that Tenet inflated its prices to be able to claim large r outlier payments from Medicare, similar large "stop loss" payments from managed care companies, and extra payments from Workers Compensation DSH and M dramatic financial recovery from a low at the end of 1999, to a market darling in October 2002, was largely due to a massive increase in these outlier, stop loss and related payments. Outlie r and stop loss rates were 23 % 26% in Tenet hospitals compared with an average of 3 % The behavior of the managers and physicians at Redding may have been an example of possible inducements to carry out unnecessary invasive surgeries in order to adhere to exacting and unrealistic revenue projections which are sometimes characteristic of investor owned hospitals Aggressive earnings goals motivate streamlined efficiency and sound financial policy which strengthen th e hospital financial health without endangering patients. But goals that are so high they are unrealistic can induce management to cross the line from an emphasis on productivity to behaviors that are unethical even illegal and dangerous to patients General sued Tenet claiming it had repeatedly gamed and overcharged Medicare, inflated

PAGE 51

37 its hospitals prices, and seriously abused the Medicare outlier program It was rep orted (RICO) Act (Davis 2005). Discretionary Versus Nondiscretionary Procedures Why did Tenet use CABG valve replacements, and catheterizations as the most frequent proced ure s for gam ing ? Why not use another type of procedure to game such as organ transplantation, tracheotomies, craniotomies, or treatments for burn p atients ? The rationale for gaming patient volumes and payment systems using CABG valve replacements, and catheterizations w as their discretionary nature Perhaps gaming activities that include the performance of unnecessary medical procedures are carried out more easily when these procedures are discretionary. A discretionary procedu re is subject to physician judgment and is not uniformly is about the usefulness of a procedure, the more likely it is that there will be substantial professional disagreement about its use. S ome physicians will be strongly in favor of using the procedure and others will be far less enthusiastic in their endorsement of 1999 ). R ate s of performance for nondiscretionary proced ures tend to be fairly uniform per 1,000 Medicare enrollees across geographic boundaries, mirroring rates of disease rather than medical practice style But the rates of performance for discretionary procedures show wide geographic variability with less o r sometimes little association to disease rates. S ometimes whether or not a patient has a particular procedure may depend on where treatment is sought for a given set of symptoms. A male diagnosed with early

PAGE 52

38 stage prostate cancer in one part of Florida c ould likely wind up having a radical prostatectomy, but if this same patient had sought treatment in another part of Florida likelihood of radical surgery (Dartmouth 1999 Luther & Studnicki 2003 ). Perhaps it is unfair and unfounded to conclude that discretionary procedures are unnecessary because they occur in geographic areas where rates o f these procedures are high. But it does invoke questions for example when per ca pita rates for certain discretionary procedures are much higher in certain Florida cities counties or hospital referral regions (HRR) relative to the state average ( Weinstein, Bronner, Morgan et al 2004, Weinstein, Lurie, Olson, et al 2006, Dartmouth 200 5 ). A discussion of discretionary versus nondiscretionary procedures follows in the next few paragraphs. Distinctions between these two types of procedures are very important as they lay down the theoretical basis to determine the DRG codes that are su sceptible to gaming and will be chosen as potential variables in the Research Methods section (Chapter 3) Nondiscretionary Procedures If discretionary procedures are subject to physician discretion and carried out with per capita geographic variability (Dartmouth 1999 ), then it is reasonable to assume that a nondiscretionary procedure would be less subject to physician discretion Less physician discretion should then be reflected in per capita rates with much more geographic uniformity (Dartmouth 1999 ) For example, a patient with colon cancer in Phoenix, AR is likely to receive the same surgical approach as a patient with the same condition in Tampa, FL given the type and level of progression of the disease is the same.

PAGE 53

39 Later in the Review of the Lite rature a number of medical procedures are discussed that are cited often in the literature as discretionary. Listed below are just a few procedures that are never defined as discretionary in the literature These procedures were selected by the author ba sed on their nondiscretionary nature to demonstrate a contrast between discretionary and non discretionary procedures. A brief discussion of each of these nondiscretionary procedures may give insights as to why they may be less profitable and why they wou ld be less subject to gaming. Heart transplantation (DRG 103) along with many other types of transplantation (e.g., lung, pancreas, bone marrow, etc.), usually follows a long adverse medical history accompanied by the patient waiting for a donor Although heart and other transplantation s are highly reimbursed Medicare procedure s (CMS 2004) the y are often associated with factors such as costly post operative complications that make them less profitable procedure of which there is absolute scientific agreement on its application, and the procedure is used for patients uniformly who have suffered burns of similar severity. But h ospitals do not routinely profit from burn procedures, as patient care and ge neral operation of a burn facility are very cost intensive. Mouth & Neck Diagnosis (DRG 483) is a highly reimbursed DRG, but this procedure is carried out with little margin for m iscalculation or medical opinion, is done usually in connection with end of life care, and is associated with high treatment costs.

PAGE 54

40 Discretionary Procedures & Efficacy of Alternative Treatments Several discretionary procedures are described in the next few paragraphs with discussions regarding perceptions by some physicians that alter n ative treatments, often less invasive, may sometimes be preferable. Life Expectancy for CABG Patients vs. Those Who Do Not Elect CABG A fter more than 20 years of clinical trials and health outcomes research, it has been found that there is no difference in life expectancy for patients electing as opposed to not electing CABG surgery. CABG does not provide greater life expectancy though that may be the perception of many CABG patients as well as the public. For many patients CABG is recommended to improve angina symptoms, better breathing capacity, and generally better quality of life, and the decision for surgery is presumably based on physician opinion and patient pre ference. But the variation in rates of CABG across geographic regions suggests that physicians may have different symptom thresholds for recommending surgery and may not interpret patient preferences in a uniform manner (Luther & Studnicky 2003) Mastect omy vs. Lumpectomy As of early 2000 despite evidence that the survival rate is identical for breast sparing surgery (lumpectomy) and for mastectomy, there continued to be wide variation in surgical rates indicating that perhaps surgery was based more on pr actice patterns and physician preference instead of disease incidence and preferences of patients. In some cases, patients may choose traditional mas te ctomy over breast conserving therapy (BCT) due to fears of radiation therapy, especially when radiation o ncology counseling is not offered in connection with treatment options. It has also been found

PAGE 55

41 that BCT rates have been highest in hospitals with radiation oncology facilities on the hospital grounds ( Elward, Penberthy, Bear et al 1998) The use of BCT decreases when the patient must be sent to an outside facility for radiation treatment. In other words, certain patients may have been advised to forego BCT for more radical surgery because of medic al logistics as opposed to disease stat us. Benign Prostatic Hyperplasia ( BPH ) & Radical Prostatectomy Although prostatectomy has been shown to improve urinary symptoms, slightly more effective than pharmaceutical regimens, the surgery is accompanied by significant risk of side effects including retrograde ejaculation and a small risk of incontinence. There are striking geographic variations in rates for this procedure which indicate differing medical opinions among physicians concerning the necessity and benefits of surgery. Widespread access ibility of prostate cancer screening has disclosed many early stage cancer s especially among older males. Efficacy of different treatment regimens has not been established, but side effects, incontinence and impotenc e have been well established for surgery and radiation. Many patients, when informed of the alternatives, may elect to avoid the harmful side effects and forego the possibility that active treatment may be more effective. Joint Replacement & Revision The prevalence of joint replacement and revi sion (i.e., surgical replacement of an artificial worn out joint) for degeneration of the knee joint has shown high degrees of variation even among neighboring regions. Elderly people experience joint stiffness and pain due to osteoarthritis (i.e., chroni c deterioration of joint surfaces). Most physicians

PAGE 56

42 function, though clinical trial findings are still limited to support this, and the risks and side effects associa ted with this type of surgery have been well established and include post operative infection and mortality, prosthesis related complications, and long periods of rehabilitation and recovery. Also, joint revisions are more complex than replacements with l ess predictable results. assessments of the risks and potential benefits of this type of surgery. These assessments are likely driven by the level of pain and difficulty in functioning communicated to the physician by the patient. Back Surgery Many patients experience back pain due to spinal stenosis or herniated discs, though the precise cause for some patients is never fully established. There is evidence that back surger y results in immediate improvement, but there is much disagreement among researchers about the long term benefits of back surgery, chiefly because there is little research on the natural history of these conditions treated without surgery. As with knee re placement, back surgery may be driven by physician preference as well as Geographic Variability of Discretionary Procedures Dartmouth (2003) identified wide geographic variability in connec tion with the discretionary procedures just discussed. Table 2 below shows minimum and maximum rates for each procedure between states in the nation and cities in Florida.

PAGE 57

43 Table 2 Range of Rates of Discretionary Procedures Per 1,000 Medicare Enrollees, Highest & Lowest State Rates in US and Highest & Lowest City Rates in Florida, 1999 Procedures Minimum & Maximum Rates of Procedure per 1,000 Medicare Enrollees States in US Cities in Florida Coronary Artery Bypass Grafting 3.1 11.3 5.4 9.5 Mas tectomy for Breast Cancer 0.8 3.3 0.9 2.7 Surgery for Benign Prostatic Hyperplasia 2.6 14.6 2.9 10.1 Joint Replacement for Degeneration of the Knee Joint 1.6 10.0 3.7 7.1 Surgery for Back Pain (including spinal fusion) 1.3 9.0 2.7 7.1 Radical Prostatectomy for Prostate Cancer 0.5 5.0 0.8 3.9 Source: Dartmouth, www.dartmouthatlas.org Table 2 shows CABG, BPH, knee joint replacement, and back pain surgery each with a large range between the lowest and highest rates, both for Florida and the United States. BPH shows a very wide range of seven points (10.1 2.9) for Florida with a range of 12 points (14.6 2.6) for the nation. CABG and back pain surgery for Florida show ranges > 4 points respectively 9.5 5.4 and 7.1 2.7 and > 3, respectively, 7.1 3.7 & 3.9 0.8, for joint replacement and radical prostatectomy. The national ranges for these procedures are even much broader. Medicare Rates for Certain Short Term Inpatient Hospital Stays in Florida 2005 Table 2 provide d examples of geographic variability for procedures carried out per 1,000 Medicare enrollees in Florida and the United States in 1999. Figure s 1 3 provide 2005 data per 1,000 Medicare enrollee s for percutaneous coronary interv ention (PCI), knee replacement, and back surgery. Each of these figure s depict s the broad variation between HRRs in Florida in carrying out these procedures.

PAGE 58

44 Figure 1 showing PCI procedures per 1,000 Medicare enrollees per HRR is shown below. Figure 1 Source: 2005 Dartmouth Surgical Discharges www.dartmouthatlas.org The horizontal dashed line on the chart represents the mean average rate of 12.67 of PCI procedures for all 18 Florida HRRs The figure shows much variation around the mean with a standard deviation of 2.75 and range of 8.44 (17.94 9.50). According to Figure 1 Miami has the lowest number of PCI procedures per 1,000 Medicare enrollees. This low rate in Miami is not exclu sive to PCI, but is also shown for back surgery and knee replacements shown in figure s 2 & 3. Close to Miami in low rates of PCI, also shown in Figure 1, are Ormond Beach and Tallahassee with rates at 9.6 0 and 9.57 per 1,000 Medicare enrollees.

PAGE 59

45 As will b e seen in Table 4 shown in a subsequent section, Miami has the highest dollar figure in Florida for Medicare reimbursements per Medicare enrollee for short term inpatient discharges as well as for total Medicare spending Yet it has the lowest number of o verall surgical discharges per 1,000 Medicare enrollees in Florida, 2005 (Dartmouth Surgical Discharges, 2005 ) Ocala and Clearwater show the highest numbers of PCI in Figure 1 with more than 17 per 1,000 Medicare enrollee s Gainesville and Panama City show over 15 Figure 2 shows discharges for knee replacements by HRR per 1,000 Medicare enrollee s, 2005. Figure 2 Source: 2005 Dartmouth Surgical Discharges, www.dartmouthatlas.org

PAGE 60

46 The horizontal dashe d line shown in Figure 2 represents mean average knee procedures of 8.14 for all 18 Florida HRRs per 1,000 Medicare enrollees. With a mean of 8.14 there is a broad range of 6.26 (11.14 4.88). As depicted in Figure 2, Miami again shows the lowest rate at 4.88 for knee replacements with the highest rate at 11.14 in Fort Myers. The assorted heights of the histogram show the broad variation around the mean in the performance of knee replacements between HRRs in Florida. A similar scenario is shown in Figu re 3 which shows back surgery per 1,000 Medicare enrollees for each Florida HRR, 2005. Again there is a broad range of rates reflecting dissimilar practice patterns for back surgery between HRRs Figure 3 Source: 2005 Dartmouth Surgical Discharges www.dartmouthatlas.org

PAGE 61

47 As with figure s 1&2, the horizontal dashed line on Figure 3 represents the mean average of 4.78 for back surgery discharges for all Florida HRRs per 1,000 Medicare enrollees. There is also a broad range for back surgery discharges of 4.57 (6.99 2.42). Miami shows the lowest rate of back surgeries per 1,000 Medicare enrollees with 2.42. Lakeland is slightly higher with 2.71. At the other end of the spectrum are Fort Myers, Bradenton, Oca la, and Sarasota respectively with 6.99, 6.70, 6.65, and 6.36. Broad differences in practice patterns are observed in figure s 1 3 for PCI, knee replacement, and back surgery. The final section in the l iterature r (Chapter 2) discussion on geographi c variation of discretionary procedures features (2005) surgical signature theory The signature theory is introduced as a potential explanation of geographic disparities in medical practice patterns in the performance of certain types of discr etionary procedures. Further Literature Findings of Geographic Variation of Discretionary Procedures In 1999 a Dartmouth report ( Dartmouth Atlas Quick Report for Florida Discretionary Surgery, 1999 ) identified several procedures as discretionary based o n criteria of scientific uncertainty and geographic variability. These procedures include CABG, mastectomy, BPH joint replacement for degeneration of the knee joint, surgery for back pain, endarterectomy for carotid artery disease, lower extremety bypass surgery, and radical prostatectomy for prostate cancer. These procedures and a few others are shown in Appendix 1 and they will be used in subsequent phases of the analysis. A subsequent Dartmouth report ( Studies of Surgical Variation: Cardiac Surgery Report 2005) defined certain cardiac procedures as discretionary, again based on grounds

PAGE 62

48 of clinical uncertainty and geographic disparity. These procedures included CABG, aortic and mitral valve replacements, and PCI Wennberg, Brownlee, Fisher et al (200 prostate, enlarged prostate, early stage cancer of the breast, osteoarthritis of the knee, osteoarthritis of the hip, osteoarth ritis of the spine, chest pain due to coronary artery disease, stroke threat from carotid artery disease, ischemia due to peripheral artery disease, and gall stones. Birkmeyer, Sharp, Finlayson, et al (1998 in Florida using 1995 hospital patient discharge data for Medicare patients, aged 65 99 years, to test varia bility of surgical rates across national HRRs It was found that rates of variability, only 1.9 2.9 fold across HRRs CABG, transurethral prostatectomy, mastectomy, an 4.7 10.1 fol d across HRRs The American College of Physicians (2001) also found low regional varia tion with treatments for hip fractures and colectomies In contrast, regional rates of CABG, back surgery, radical prostatectomy, and peripheral angioplasty showed high variability. Fisher, Goodman, Skinner et al (2009) reported that Medicare spending in 2006 varied more than threefold across national H R variation is due to differences in the prices paid for similar services, a nd some is due to

PAGE 63

49 differences in illness; but even after accounting for these factors, twofold differences remain. In other words, the differences in spending are almost entirely explained by differences in the volume of health care services received by s for gam ing is consistent with the types of discretionary surgeries described earlier by Wennberg (2008) and Dartmouth (2005) which have included CABG, valve replaceme nts and catheterizations Catheterizations include coronary angiography, coronary arteriography, and PCI (American Heart Association 2009). Weinstein, Lurie, Olson et al (2006) identified hip replacement, lumbar discectomy/laminectomy and lumbar fusion as discretionary based on a lack of scientific surgery are among the most variable of all surgeries. The underlying causes of international and regional variations found i n rates of spine surgery include lack of scientific evidence, financial incentives and disincentives to surgical intervention, and Burns, Moskowitz, Ash et al (1997) studied geographic variations in the performance of hip procedures in Houston, Pittsburg, and Minneapolis. They found hip replacement is used to treat non displaced fractures much more frequently i n Houston Among heart procedures discussed in this study pacemaker implants and implantable cardioverter defibrillators (ICD) appear to have the greatest medical

PAGE 64

50 consensus regarding efficacy. But geographic variations were still found with these procedures the determinants of which were economic and demographic factors ( Ovsyshcher & Furman 2003). Essentially, pacemaker implants were less likely to occur among patients residing in economically disadvantaged communities Carlisle, Valdez Shapiro et al (1995) examined geographic variations in Los Angeles, CA for CABG, carotid endarterectomy, hysterectom y, artery angioplasty, permanent pacemaker insertion, mastectomy, and transurethral prostate resection (TURP). Large geographic variation was found for all cardiac procedures with the exception of pacemaker implants. In fact, out of all procedures studie d pacemaker implants and hysterectomies had the smallest geographic variation, though even these procedures showed variation with respect to race and ethnicity and socioeconomic status. Spencer, Fung, Wang et al (2004) investigated a sample of 16,352 case s from with stage I or II prostate cancer. They found that patients (veterans) in the western part of the U.S. had a higher likelihood of undergoing surgery than radiation compared to the northeast, south, or M idwest. They also found that African American men with lower grade and higher stage tumors as well as unmarried men of all races were more likely to undergo radiation rather than curative treatment or surgery. The d iscussion in the Research Methods section (Chapter 3) will show that many of the discretionary procedures discussed above are a good theoretical starting point for selecting DRG codes as variables that contribute to hospital profitability and may be subjec t to gaming.

PAGE 65

51 The Effects of Geographic Disparities in Medicare Spending It is plausible that Medicare spending would be roughly similar per capita between regions accounting for differences with respect to health status age etc But Medicare spendin g is as geographically variable as volumes of discretionary procedures. After adjusting for clinical, demographic, hospital, and regional characteristics Cowper, DeLong, Peterson et al (1997) found wide variability among states in patient level cost and LO S for CABG surgery. T hey suggest there is no association between higher medical costs and lower rates of readmission or mortality. Fisher, Wennberg, Stukel et al (2003) have noted that Medicare spending increases in areas with high rates of discretionary procedures. But they conclude. spending regions receive more care than those in lower s Dartmouth (1999) concluded that d ifferences in Medicare spending cannot be adjustm ent for these factors has almost no effect on the range of variation in Medicare Higher Medicare spending in certain areas, and more specifically for certain hospitals, induced by profitable DRG codes result s in higher total rev enue for a hospital. If these DRG codes represent highly reimbursed discretionary procedures then the direct effect is inordinate Medicare spending when perhaps more affordable alternative s are available, as well as subjecting patients to procedures oft en more invasive, that may not be in their best interest. But the effect for the hospital, whether intentional or inadvertent,

PAGE 66

5 2 is greater operating revenue, which will, in turn, increase profitability all other things being equal. Baicker & Chandra (2004 ) found that geographic disparities for Medicare spending are correlated with the number of practicing physicians in different areas of the country. A similar scenario may exist between hospitals with high operating margins and those with modest to poor m argins. Operating margins may be affected by the mix of physician specialties found in certain hospitals. A higher concentration of specialists may result in the performance of higher Medicare reimbursement procedures which would increase spending in cert ain geographic areas while also improving the bottom line in the hospitals where an excessive number of highly reimbursed procedures are performed. Areas populated with a substantial number of general practitioners experience lower Medicare spending levels than areas with concentrations of certain types of specialists. According to Baicker & Chandra this explains wide disparities between Medicare reimbursements per capita between geographic areas. Baiker & Chandra concluded that additional spending for Me dicare does not result in more effective care for the geographic areas with the highest spending rates. that has not been & Chandra 2004).

PAGE 67

53 Higher & Lower Levels of Medicare Spending Between States Table 3 shows Medicare reimbursements per enrollee for short term inpatient hospital stays for several different HRRs across the nation. Regions shown at the top of Table 3 represent the top ten highest Me dicare per enrollee reimbursements for short term stays The bottom of the table represents the top ten lowest levels of per e nrollee Medicare reimbursement for short term stays. Table 3 Medicare Spending by Cit y State & Hospital Referral Region, Higher & Lower Levels of Medicare Reimbursements Per Medicare Enrollee, 2003 City Name State Hospital Referral Region Per Enrolle e Reimbursements Top 10 Cities with Highest Levels of Medicare Reimbursements Per Medicare Enrollee for Short Term Inpatient Stays, 2003 Baltimore MD 223 3,776 Miami FL 127 3,792 Philadelphia PA 356 3,871 Los Angeles CA 56 3,991 Chalmette LA 190 14 4,072 East Long Island NY 301 4,176 McAllen TX 402 4,211 Covington LA 19018 4,361 Manhattan NY 303 5,386 Bronx NY 297 6,395 Top 10 Cities with Lowest Levels of Medicare Reimbursements Per Medicare Enrollee for Short Term Inpatient Stays, 2003 Mason City IA 195 1,603 Appleton WI 446 1,607 Neenah WI 452 1,685 South Bend IN 187 1,738 Sioux City IA 196 1,746 Cedar Rapids IA 190 1,783 La Crosse WI 448 1,792 Dubuque IA 193 1,795 Fort Collins CO 104 1,807 Salt Lake City UT 423 1,842 Reimbu rsements are not adjusted for inflation Reimbursements are sorted in ascending order both top 10 lowest & highest level Medicare reimbursement groups Source: Dartmouth Atlas of Healthcare, dartmouthatlas.org

PAGE 68

54 The Bronx and Manhattan, NY, represent the hig hest per Medicare enrollee reimbursements in the nation for short term hospitals stays at $6,395 and $5,386 respectively. Three other cities in the northeast in the top 10 highest per enrollee Medicare spending group are also in the northeast and include Baltimore, Philadelphia, and East Long Island. Two cities from Louisiana, Covington & Chalmette, have reimbursement levels $4,361 & $4,072 McAllen, Texas, has the forth highest Medicare reimbursement level at $4,211. Miami is shown to have the ninth hig hest Medicare reimbursement level in the nation with $3,792. Subsequent review will show that Miami has the highest levels of Medicare spending in Florida. At the other end of the continuum listed in Table 3 are ten top cities with the lowest levels of pe r enrollee Medicare reimbursements for short term hospitals stays. Four cities in Iowa are listed in this group including Mason City, Sioux City, Cedar Rapids, and Dubuque which show per enrollee spending of $1,603, $1,746 $1,783 and $1,795. Three cities in Wisconsin, Appleton, Neenah, and La Crosse, are also shown at $1,607, 1,685, and $1,792. What is the difference between short term inpatient treatments in Mason City, Iowa per Medicare enrollee and spending in T he Bronx or Manhattan, New York ? Why a re spending levels higher in the Northeast and certain cities in Louisiana and Texas, but so much lower in Iowa and Wisconsin? The most probable reasons for these differences in spending are unionized labor in the northeast and high concentration s of inves to r owned hospitals in southern states such as Louisiana and Texas. High ratios of specialist physicians per 1,000 Medicare enrollees may also help explain higher Medicare spending in certain areas

PAGE 69

55 W ith a few exceptions it was found that mean Medicare r e imbursements for inpatient hospital care were generally lower in western and mountain states and higher in southern and eastern states. It was also not uncommon for substantially different levels of Medicare per capita spending even among contiguous state s It has been suggested that geographic differences in volumes of procedures and Medicare spending can be partially explained by culture upbringing and psycho social factors which would result in varying preferences with regard to the number of visits to physicians, the perception of personal medical need, and the willingness to undergo surgery. patient preferences vary as sharply according to boundaries of health care markets as does su Although patients are consumers they do not have the same objectivity and latitude in their decision making as they would in the purchase of other commodities. Patients tend to follow the recommendations of their physicians becau se they trust the expertise of the medical profession and tend to question much less the competency or motivation of their physician as they would the providers of other types of services Fisher, Bynum & Skinner (2009) found that between 1992 and 2006 pe r capita inflation adjusted spending in Miami F lorida grew at an annual rate of 5.0%, relative to 2.3% in Salem, Oregon, and 2.4% in San Francisco. P er capita growth in Medicare expenditures between 1992 and 2006 in Miami was $8,085 per Medicare enrollee It was also found that 26 HRRs across the nation had more rapid spending than Miami and 18 regions had slower growth in Medicare expenditures than Salem.

PAGE 70

56 As already mentioned, w hat may account for differences in surgical volumes and Medicare spending is what Wennberg (2005) has labeled the s urgical s in this chapter beginning on page 60, after the per Capita Medicare Spending in Wennberg argues that g eographic variability in Medicare spending between national regions as well as cities in Florida can be explained by the surgical signature. G eographic Differences in Per Capita Medicare Spending in Florida Florida is a very important state for the Medicar e program In 2001 Florida had 1 ,803,422 Medicare enrollees which exceed ed the number of enrollees in New York and was second only to California which ha d the highest number of Medicare enrollees in the nation By 2005 the number of Medicare enrollees in Florida had grown to 2,115,397, an increase of 17.3% (Dartmouth Surgical Discharges 2005) Medicare spending in Florida occurs with similar variability between cities as was shown between states in the previous section. re enrollment, if variability indicates that more spending occurs in c ertain areas than is warranted, then the re is a sizeable misplacement of f ederal tax dollars Medicare Reimbursements for Short Term Inpatient Hospital Stays Table 4 shows 200 5 Medicare reimbursements for short term inpatient hospital stays per Medicare enrollee for 18 different HRRs in Florida. The second column of this table converts reimbursements to z scores to ascertain how many standard deviations a reimbursement level may be awa y from the mean.

PAGE 71

57 Table 4 Medicare Reimbursements for Short Term Inpatient Hospital Stays P er 1,000 Medicare Enrollee s 2005 City Name HRR 2005 Reimbursements Z Score 2005 Reimbursements Bradenton 115 2,110 (1.19) Clearwater 116 2,824 0.37 Fort Lauderdale 118 2,687 0.07 Fort Myers 119 2,328 (0.71) Gainesville 120 2,989 0.73 Hudson 122 2,431 (0.49) Jacksonville 123 2,956 0.66 Lakeland 124 2,758 0.23 Miami 127 3,905 2.74 Ocala 129 2,411 (0.53) Orlando 130 2,900 0.54 Ormond Beach 131 2,428 (0.49) Panama City 133 3,007 0.77 Pensacola 134 2,491 (0.35) Sarasota 137 1,720 (2.04) St. Petersburg 139 2,707 0.12 Tallahassee 140 2,351 (0.66) Tampa 141 2,756 0.22 National Average 2,955 Adjusted State Average 2,743 Unadjusted State Mean Aver age 2,653 Standard Deviation 458 Coefficient of Variation 0.17 HRR: Hospital Referral Region Z = (Observation Mean) / Standard Deviation = (Observation 2, 653 ) / 458 Source: The Dartmouth Atlas of Healthcare, http://www.dartmouthatlas.org Table 4 above shows that Miami is an outlier compared to the other cities in the table. Z scores were calculated for each of the cities to determine how many standard deviations each city is from the unadjusted mean of $ 2, 653 for Florida. Miami is almost three s tandard deviations above the mean for the state and is significantly above the national average of $ 2, 955 shown also in Table 4 K eep in mind that in figure s 1 3 Miami showed the lowest rates of PCI, back surgery, and knee replacement per 1,000 Medicare e nrollees.

PAGE 72

58 Panama City and Gainesville are each about of one standard deviation above the mean at 0.77 & 0.73. Jacksonville is not far behind at 0.66. The question arises as to why there is so much regional fluctuation for Medicare spending. Why is Miam term inpatient stays so much higher than other HRRs ? What are the differences between areas with high reimbursements per enrollee and those with lower levels? As already mentioned one explanation is the surgical signat ure theory which is discussed in the next section Appendix 2 provides a more thorough review of Medicare reimbursements for short term inpatient stays per Medicare enrollee. The appendix also presents comparisons between 2000 & 2005 including growth in spending in the five year period. Total Medicare Spending in Florida Medicare reimbursements shown in Table 4 referred exclusively to short term inpatient hospital stays. Table 5 shows i nflation a djusted t otal M edicare s pending p er e nrollee by Florida H RR for 1992 & 2006. S ome but not all HRRs follow similar patterns in total Medicare spending across Florida HRRs as they did in spending for short term inpatient care. The point is that w hat might be expected would be comparable levels of spending per Me dicare enrollee, but instead the table shows geographic variation between HRRs

PAGE 73

59 Table 5 Inflation Adjusted Total Medicare Spending Per Enrollee by Hospital Referral Region (HRR) in Florida, 1992 & 2006, Actual & Percentage Growth HRR Place #HRR Medica re Spending Per Enrollee, 1992 Medicare Spending Per Enrollee, 2006 Growth in spending 1992 2006 Annual growth rate 1992 2006 Bradenton 115 4,442 7,640 3,197 3.95% Clearwater 116 5,432 8,697 3,266 3.42% Fort Lauderdale 118 6,321 9,816 3,495 3.19% F ort Myers 119 5,401 8,243 2,842 3.07% Gainesville 120 5,235 8,357 3,122 3.40% Hudson 122 5,356 9,058 3,701 3.82% Jacksonville 123 5,610 8,733 3,123 3.21% Lakeland 124 4,487 8,799 4,313 4.93% Miami 127 8,266 16,351 8,085 4.99% Ocala 129 4,614 8,097 3, 483 4.10% Orlando 130 5,408 8,588 3,179 3.36% Ormond Beach 131 5,010 8,298 3,288 3.67% Panama City 133 5,413 9,678 4,265 4.24% Pensacola 134 5,477 7,946 2,468 2.69% Sarasota 137 4,983 7,466 2,484 2.93% St. Petersburg 139 5,801 9,103 3,301 3.27% Tall ahassee 140 5,140 7,257 2,117 2.49% Tampa 141 5,368 8,991 3,624 3.75% Florida Mean (all 18 HRRs ) 5,431 8,951 3,520 3.58% Florida Std Dev (all 18 HRRs ) 842 1,973 1,269 0.68% Florida CV (all 18 HRRs ) 0.155 0.220 0.361 19.02% *Inflation adjusted Medic are spending Source: The Dartmouth Atlas of Healthcare, http://www.dartmouthatlas.org The most noticeable HRR in Table 5 is HRR 127, Miami which shows $16,351 of 2006 inflation adjusted total Medicare spending per enrollee Miami also showed the highest level of short term inpatient spending in Table 4. As already mentioned under Table 4 and in figure s 1 3, Miami had shown the lowest rate of PCI, back surgery, and knee replacements of all Florida HRRs Miami also shows the lowest overall surgical rate for all HRRs in Florida. Fort Lauderdale, Panama City, St Petersburg, and Hudson show 2006 reimbursements per enrollee of over $9,000. In fact, Fort Lauderdale shows over $9,800

PAGE 74

60 and Panama City close to $9,700. Bradenton, Pensacola, and Sarasota each sho w reimbursements per enrollee under $8,000. The 2006 rate of growth, inflation adjusted between 1992 & 2006 is $3,520 with a large standard deviation of $1,269 with a comparably large coefficient of variation of 0.361. The re is also a high coefficient of variation at 19% for the annual percentage growth rate. The Surgical Signature as a Theor y to Account for Differences in Prevalence of Procedures and Levels of Medicare Expenditures Broad variations in certain types of surgical rates per 1,000 Medicare e nrollees and in Medicare per capita spending may be explained by what Wennberg (2005) refers theory. Wennberg describes the surgical s ignature in the ecialize in a find candidates that meet clinical appropriateness criteria. S urgical specialists tend to become expert in a subset of the procedures that their specialt y performs and to orient t Wennberg tested this theory by reviewing rates of several orthopedic procedures among cities in Florida. Particularly in cases of hip and knee replacemen ts decisions regarding surgery tend to be made by small groups of surgeons. This evolves practice patterns that result in steady increases in these types of surgeries until available surgeons can no longer accommodate further increases. Numbers of surgeri es for the community then stabilize at a high level and remain there for long periods of time.

PAGE 75

61 Wennberg reported that the number of per capita orthopedic surgeons show variations of four to seven fold among regions with no correlation between the numbers of surgeons and the performance of surgeries. W ennberg explains the surgical signature by showing comparisons between Manhattan, New York with four cities in Florida including Miami, Fort Lauderdale, Fort Myers, and Sarasota discussed below Wennberg c oncludes that the surgical signature accounts for differences in surgical rates, discharges, and medical expenditures more than any other factor. Surgical Signature in Several Florida Cities Between 2000 and 2001, Medicare per capita rates for knee surge ry in Fort Myers were three times higher than the rate in Manhattan. The rate in Sarasota was 2.5 times higher, and the rate in Fort Lauderdale was 1.8 times higher. The rate of hip replacement in these three cities was twice the rate for Manhattan. Back surgeries were more than three times higher for Fort Myers and Sarasota compared to Manhattan, and Fort Lauderdale showed a back surgery rate that was five times higher. Oddly, hip replacements were lower in Miami than Manhattan. Within a decade surgeons in Fort Myers performed 7,250 more back operations, 7,000 more knee replacements, and 2,600 more hip replacements than would have been done had the Manhattan rate per 1,000 Medicare enrollees prevailed in those communities. Miami surgeons carried out 870 more back surgeries, 1,400 more knee surgeries, but 56 fewer hip replacements if the Manhattan rate had been applied to Miami over the same ten year period. In 2005, surgical discharges for back surgery in Manhattan were 1.94 per 1,000 Medicare enrollees. Fort Myers and Sarasota, on the other hand, had, respectively, 6.99

PAGE 76

62 and 6.36 back surgery discharges per 1,000 Medicare enrollees. Recall from Figure 3 that Miami had the lowest per capita rate of back surgeries with 2.42 per 1,000 Medicare enrollees. But even Miami, with the lowest per capita back surgery rate among all HRRs in Florida showed a higher rate than Manhattan which was 1.94 (Dartmouth 2005). This is also found for knee replacements in 2005 Although the Miami HRR has the lowest per capita knee replacement rate in Florida with a rate of 4.88 per 1,000 Medicare enrollees (see Figure 2), Manhattan has a per capita rate of 3.72. Fort Myers, Sarasota, and Bradenton have per capita rates of 11.14, 9.16, and 9.42 (see Figure 2) (Dartmouth 2005) P er capita d ischarges for CABG surgeries in 2005 were 3.48 in Manhattan relative to 5.51 in Sarasota. Per capita rates were also 4.27 and 4.28 for Fort Lauderdale and Fort Myers, respectively (Dartmouth 2005). Per capita PCI discharges in Manhattan were 12 .99 in 2005 Contrast this rate to those of Ocala and Clearwater which show ed rates of 17.94 and 17.54 (Dartmouth 2005) Commonly cited reasons for these geographic variations in Medicare expenditures related to back surgeries include the incidence of osteoarthriti s and herniated discs, but considering such wide disparities between communities it is doubtful that these differences in expenditures were completely dictated by medical need. Dartmouth ( Date Unknown ) found a positive correlation between physician visits and density of cardiologists per capita 100,000 residents. For example, patient visits to cardiologists were five times higher in areas with 12 cardiologists per 100,000 residents relative to two per capita 100,000.

PAGE 77

63 Dartmouth also found that rates of proc edures, which are lower in central and Midwestern states, are concomitant with lower density of specialists per 100,000 residents relative to east and west coast states. I n 1995 1996 Dartmouth also found positive correlations between the density of h ospit al beds and discharges for all medical conditions per 1,000 Medicare enrollees. Geographic location Findings from the literature regarding geographic variations in medical practice patterns, particularly those in Florida, provided the impetus to create t wo variable s for geographic location In Part 1 of the study, geographic location is based on median income levels for 67 countie s in Florida and is dichotomized according to high vs. low income level s In Part 2 of the study geographic location is based on north, central, and south geographic regions in Florida. This variable is a dichotomy with the combined northern and central regions as the base level. Research Summary The Literature Review up to this point has examined several financial performance Medicare revenue or amount of Medicare patients, good management and prudent strategic direction, possible gaming of payment systems and overuse or overcharging of certain discretiona ry medical procedures. The Literature Review has also assessed what constitutes a discretionary procedure based on scientific uncertainty and geographic variability and cited several studies that reached similar conclusions on what type of procedures shou ld be considered discretionary.

PAGE 78

64 The investigation will now proceed to first define and operationalize the term discretionary procedure for purposes of this study (Chapter 3) The investigation will then go on to test the association between the TATL ratio and certain types of discretionary heart and orthopedic procedures reimbursed by conventional Medicare ( c hapter s 5 & 7 ) The analysis will also focus on the interaction between these two discretionary procedure variables and hospital ownership. This w ill help to address the question if the performance of higher rates of certain types of discretionary procedures relates to stronger balance sheets (i.e., higher TATL ratios) in investor owned vs. not for profit hospitals. It may also help to shed light o n fluctuating rates in performing these discretionary procedures with respect to hospital owne rship and geographic location The analysis will then switch from an investigation of rates of discretionary procedures under conventional Medicare to an analysis of total patient charges under Medicare HMO (chapters 6 & 8 ) This will address large disparities in patient charges, unrelated to patient disease severity or the extent of medical services provided, with respect to hospital ownership and Florida geograp hic region. This summary is expressed in more detail under Research Goals & Objectives, shown in Chapter 3, Research Methods Data Structure & Identification of Variables, which is next.

PAGE 79

65 Chapter 3 Research Methods Data Structure & Identificat ion of Variables Research Goals & Objectives The intent of this research is to determine which DRG codes are associated with investor owned hospitals with high TATL ratios, if these codes represent discretionary procedures, how these DRG codes differ from codes associated with TATL ratios of less profitable hospitals, if these DRG codes are being used with inordinate frequency considering the medical needs of the patient population, and to determine if charges are commensurate with other hospitals that per form the same procedures. Two broad goals and several objectives for parts 1 & 2 have been formulated to address the research questions. The two goals are listed as follows in connection with parts 1 & 2 of this study Goal for Part 1 Determine if inv estor owned hospitals with high TATL ratios are using conventional Medicare to perform discretionary procedures at significantly higher rates than their not for profit counterparts. It is important to distinguish that the context is a positive association between the TATL ratio and percentages of discretionary procedures not higher volumes. It was shown in the preliminary analyses that not for profits performed higher overall volumes of discretionary procedures during the study period, 2000 2005 (See Tabl e 9) However, preliminary analyses also indicated that percentages per hospital of certain discretionary procedures, reimbursed by conventional Medicare, increased concomitantly with rising TATL ratios in investor owned hospitals.

PAGE 80

66 Objectives for Part 1 1) To operationalize a definition for 2) To identify DRG codes which represent discretionary procedures 3) s increase corresponding to increases in percentages per hospital of certain discretionary procedures reimbursed by conventional Medicare 4) To determine if the association between the TATL ratio and percentages of certain discretionary procedures, reimbursed by conventional Medicare, differ according to hospital ownership 5) To determine if the ass ociation between the TATL ratio and percentages of certain discretionary procedures, reimbursed by conventional Medicare, differ according to geographic location Goal for Part 2 Determine if investor owned hospitals are charging significantly higher amou nts for the same extent of medical services than their not for profit counterparts using Medicare HMO to perform certain discretionary procedures. Objectives for Part 2 1) To operationalize a definition for 2) To identify DRG codes w hich represent discretionary procedures 3) To determine if mean charges for these discretionary procedures reimbursed by Medicare HMO are significantly greater for investor owned hospitals than mean charges for not for profit Florida general short term acute care hospitals. 4) To determine if mean charges for these discretionary procedures reimbursed by Medicare HMO differ according to geographic region The goals differ for parts 1 & 2 because incentives under conventional Medicare and Medicare HMO also differ. As already mentioned, s ince conventional Medicare is based on a fixed rate per DRG code there is no incentive for hospitals to artificially increase charges. Their incentive would be to increase volumes of certain procedures that make them profitable, w hile attempting to minimize other types of procedures. Inflating charges would be an incentive for hospitals reimbursed by Medicare HMO, since it is similar to an insurance charge based payment system.

PAGE 81

67 It should also be observed that references to geograp hic location in Part 1 and geographic region (or sector) surgical signature, as discussed in the literature review in Chapter 2 Data Sources Hypotheses will be addressed using four datasets that prov ide clinical, demographic, operational, and financial data. These datasets include the Florida Agency for Health Care Administration (AHCA) Discharge Data for all patients treated in Florida hospitals over a six year period, 2000 2005, AHCA Financial data (hardcopy) 2000 2005, AHCA Financial data (electronic file) for the same time period, and median income data for each county in Florida for each year, 2000 2005, available from the U.S. Department of Housing & Urban Development (www.hud.gov). AHCA hardco py and electronic financial data are considered two different datasets because though they overlap in some areas they report on different financial items in others. For example the hardcopy dataset provides some income statement data including operating m argins, as well as operating revenues and expenses that comprise these margins for all Florida hospitals, but it does not report balance sheet data. The electronic data does not provide income statement data with the exception of total margins, but it does provide data on hospital balance sheets, including items that comprise total assets and total liabilities, of which the TATL was constructed. Structure of the Datasets Datasets for parts 1 & 2 will be composites of respectively four and three of the datas ets listed above.

PAGE 82

68 Selection of Hospitals The selection of hospitals to be used in the analysis were derived from the AHCA Financial data and included only acute care general investor owned (C corporation) hospitals religious and other not for profit [ 501(c)(3)] hospitals, and governmental hospital district facilities. Excluded from the data set were rehabilitation and psychiatric facilities, certain government owned hospitals (e.g., state and county hospitals, etc.), teaching hospitals, specialty hosp itals, limited liability companies (LLCs), and partnerships. A total of 891 hospital observations were comprised of roughly 150 hospitals from each year, 2000 2005. For example, of the 281 health facilities listed in the 2005 AHCA Financial Data, 151 facilities were used and were comprised of 69 investor owned hospitals, 20 religious not for profit hospitals, 46 other not for profit hospitals, and 16 hospital district facilities. Discharge and financial data were segmented to include only patients dis charged from these general acute care short term hospitals. Data Structure for Part 1 Total observations for Part 1 included the entire 891 observations representing repeated measures o f approximately 150 acute care general hospitals for each year, 2000 2 005. For Part 1 this means that a large AHCA dataset of 4 million conventional Medicare patient observations must be merged with a much smaller AHCA hospital financial data set comprised of 891 observations. The structure of the models must be driven b y the dependent variable which for Part 1 is the TATL ratio. This means that because the TATL ratio is specific to each

PAGE 83

69 hospital, each observation must represent an individual hospital in a given year. This allows each observation of the dependent variab le to be unique. Individual patient data used in the analysis for Part 1 were downloaded from the AHCA Discharge Data and aggregated into data specific to each hospital. For example, conventional Medicare CABG patient observations are aggregated into spec ific hospitals in which they occurred. These aggregated totals then become total CABG procedures performed within each specific hospital in a given year. Procedures are then converted to percentages per hospital by dividing total CABG procedures reimburs ed by conventional Medicare by total discharges performed by each hospital. Part 1 of the study uses geographic location as a covariate. HUD promulgated median household income data is used as a proxy for geographic location to distinguish hospitals locat ed in higher vs. lower income counties. HUD assigns annual median household income levels for each of the 67 counties in Florida. Data Structure for Part 2 Part 2 of the study features total Medicare HMO charges for valve replacements and CABG procedures as the dependent variable. The dependent variable, total charges, is the driver in structuring the model. AHCA Discharge data reports a unique observation for total charges for each patient. This allows population parameters to be defined as patients t reated with CABG or valve replacement procedures (DRG codes 104, 105, 106, 107,109, 547, 548, 549, 550) financed by Medicare HMO, 2000 2005. This resulted in 12,972 patient observations.

PAGE 84

70 The variable hospital ownership was then coded for each patient obse rvation to represent whether the hospital of which the CABG or valve replacement procedure occurred was either investor owned or not for profit. Not all hospitals within the 891 hospital selection perform CABG or valve replacement procedures reimbursed by Medicare HMO Only 360 hospitals (approximately 60 hospitals per year, 2000 2005) perform these types of procedures and are the hospitals used in the analysis for Part 2. Length of stay ( LOS ) and geographic region are covariates in the models for Part 2. With LOS as a covariate the objective was to examine increasing charges corresponding to increases in days of hospitalization. Th e association between charges and the number of days of hospitalization may differ for patients who are discharged due to d eath as opposed to patients discharge d under other conditions Therefore, it was determined that observations representing patients whose discharge status was due to death should be eliminated from the dataset. This resulted in 12,420 total observations u sed for the analysis. Deriving an Operational Definition of Discretionary Procedures Recall that the first two objectives listed under Research Goals & Objectives in this chapter pertain to operationalizing the concept of discretionary procedure in the con text of this investigation. This concept of discretionary procedure is a critical feature in both parts 1 & 2. Part 1 contain s two variables representing discretionary procedures, and Part 2 parameter s are defined in part, by two discretion ary procedures. B efore moving forward with a discussion of the variables used in this analysis, it would be useful to define and operationalize the concept of discretionary procedure.

PAGE 85

71 As already discussed in the l iterature r eview in Chapter 2 a discreti onary procedure is defined as a medical procedure which lacks scientific consensus regarding efficacy and per capita geographic uniformity in practice patterns. Criteria for Identifying Discretionary Procedures Wennberg (2005) and Dartmouth (1999) defined discretionary procedures as those used to treat chronic as opposed to acute illnesses, medical uncertainty regarding efficacy, and geographic variability in how these procedures are prescribed. Of these three criteria the most useful for efficiently iden tifying discretionary procedures is geographic variability. In fact, geographic variability is an indicator for medical uncertainty because it is the lack of scientific consensus that results in regionally non uniform rates of procedures that are inconsis tent with disease patterns. By contrast, because medical consensus exists for nondiscretionary procedures they tend to mirror disease patterns and occur with more predictable uniformity per 1,000 Medicare enrollees across geographic regions (Fisher, Bynum & Skinner 2009) Using a Nondiscretionary Procedure as a Benchmark Wennberg (2005) used geographic variability to identify discretionary procedures by using the geographic uniformity of a nondiscretionary procedure as a benchmark. Hip fracture was used as a benchmark reference because it is acute, painful, and motivates the physicians, irrespective of their specialty or geographic location, agree uniformly on the need for h Wennberg calculated rates for hip fracture and several other types of procedures, for all national HRRs 2002 2003, using the incidence of procedures for prescribed

PAGE 86

72 geograph ic areas as the numerator and the resident population as the denominator. These rates were then expressed as ratios to the national average for each procedure. Wennberg systematic component of variation, a measure that geographic variability (Vitale, Krant, Gelijns et al 1999, Weinstein, Goodman & Wennberg 1998, Keller, Soule, Wennberg et al 1990) Hip fracture showed ratios of 0.9 and 1.3 depending on location, 10% below and 30% above the national average with rates stable over time and consistently the highest in the southwest and Texas regions. The systematic component of variation for hip fracture was 13.8, which was used as the benchmark to make comparisons with other procedures. The hip fracture benchmark was compared with many procedures, three of which were knee surgery, back surgery, and hip replacement. The component of variation for knee surgery was 55.0, more than four times that of the hip fracture benchmark of 13.8. Hip replacement and back surgery had components of variation of 67.2 and 93.6, five and seven times greater, respectively, than that of hip fracture. Using the Coefficient of Variation as an Indicator of Geographic Variability method showed variability by examining differences in mean ratios for certain skewed due to a few very high or very low ratios. This is not as much of a problem in HRRs To derive a measure of variability for procedures in Florida, which has only 18 HRRs a more conservative test,

PAGE 87

73 the coefficient of variation (CV) is needed. The CV, also known as unitized risk, is expressed as the relationship between the standard deviation and the mean for th e data distribution (Garson 2006, Rajaram 2004). Appendix 1 which was referenced in Chapter 1 shows rates per 1,000 Medicare enrollees of surgical procedures for each of the 18 Florida HRRs 2003. Below the tables containing these rates in the appendix are descriptive statistics including the CV. The CV for all Florida surgical discharges is 0.07 which is consistent with two nondiscretionary procedures which are used as benchmarks, hip fracture repair and resection for colon cancer, which have CVs of 0 .09 and 0.08 respectively. The CV indicates wide geographic variation, based on the association of the standard deviation to the mean for each rate, for each of these procedures in comparison to the benchmarks. CABG and coronary angiography show CVs of 0 .17 and 0.18 respectively. Back surgery has a CV of 0.28 and mastectomy 0. 24 Hip and knee replacement s show 0.19 and 0.18, and benign prostatic hyperplasia (BPH) shows a CV of 0.37. Based on the broad distance between the CVs of each procedure and the benchmarks, as shown in Appendix 1, several procedures are determined to qualify under the geographic variability criterion as discretionary and are eligible to be used as variables in the study. Table 6 shows these procedures with their respective DRG c odes.

PAGE 88

74 Table 6 Procedures Defined as Discretionary Based on Clinical & Geographic Variability Criteria Procedure DRG Codes Valve Replacement 104, 105 Coronary Artery Bypass 106, 107, 109, 547, 548, 549, 550 Percutaneous Cardiovascular Intervention ( PCI) (with & without stents) 112, 516, 517, 518, 526, 527, 555, 556, 557, 558 Major Joint & Limb Reattachment 209 Hip & Femur Procedures 210, 211 Back Surgery Spinal Fusion 496, 497, 498, 519, 520, 546 Knee Procedures 501, 502, 503, 544, 5 45 Prostatectomy 306, 307, 336, 337 Mastectomy 257 258 259 260 Discretionary procedures were identified under four major diagnostic categories (MDC). Cardiac procedures are represented under MDC 5, Diseases and Disorders of the Circulatory System, an d orthopedic procedures are under MDC 8, Diseases and Disorders of the Musculoskeletal System and Connective Tissue. Mastectomy is classified under MDC 9, Diseases and Disorders of the Skin, Subcutaneous Tissue and Breast, and BPH and radical prostatectom y fall under MDC 11, Diseases and Disorders of the Kidney and Urinary Tract (Ingenix 2003 Utah DOH 200 8 ). The procedures identified in Table 6 will now be explored for potential statistical usefulness in model development. Selecting Discretionary Procedu res for Further Investigation To begin a selection process, the procedures listed in Table 6 were initially tested using a Pearson and Spearman correlation matrix shown in Appendix 3 Variables used in the matrices included the TATL ratio and percentages of discretionary procedures including CABG, PCI, valve replacement, spinal fusion, hip & femur procedures joint &

PAGE 89

75 limb procedures knee, mastectomy, and prostate procedures. Using CABG as an example of how percentages of procedures are calculated, the pe rcentage of CABG procedures equals the number of CABG procedures performed in a particular hospital and reimbursed by conventional Medicare divided by the total number of procedures from all payer s in that hospital in a given year. In the correlation matr ix in Appendix 3, a ssociations are observed between the TATL and each medical procedure tested in the matrix. With the exception of mastectomies, coefficients between the TATL ratio and each procedure (e.g., CABG, PCI, etc.) are in excess of 0.20 (p < 0.0 001). This means that a significantly positive association exists between the TATL ratio and each procedure. What is also noticed are significant and high correlations between different type s of heart procedure s There is a Pearson correlation of 0.93 b etween valve replacements and CABG procedures. There is also a Pearson correlation of 0.65 between PCI and valve procedures and 0.68 between PCI and CABG. There is a Spearman correlation of 0.87 between valve replacements and PCI and 0.88 between PCI and CABG. Similar associations, though not as high, are also observed between certain orthopedic procedures. There is a Pearson correlation of 0.75 (p < 0.0001) and Spearman correlation of 0.81 (p < 0.0001) between hip & femur procedures and joint & limb pr ocedures. There were also Pearson and Spearman coefficients of 0.50 (p < 0.0001) and 0.62 (p < 0.0001), respectively, between joint & limb procedures and knee procedures. These high correlations between certain types of medical procedures indicate problem s of multicollinearity if each procedure was to be used as an individual variable. After a review of the matrices it was determined that several medical procedures could be

PAGE 90

76 efficiently and coherently packaged into two independent variables. Although prost atectomy showed significant and high associations with the TATL ratio it was dropped from further consideration because the remaining procedures fall into either MDC 5 or MDC 8. It was determined that they could be grouped together to form two variables r epresenting cardiac group procedures and orthopedic group procedures. Grouping these procedures into two variables will reduce the number of terms used in model development which will add statistical power and help assure validity and will also allow seve ral DRG codes to be tested at one time The two newly created groups cardiac group procedures and orthopedic group procedures, and the patient populations that comprise them are shown in tables 7 and 8 Table 7 Cardiac Group Procedures Variable Florid a Short Term Acute Care General Hospitals, 2000 2005 Procedures (Reimbursed by Conventional Medicare) Total Procedure Volumes Valve Replacement 20,551 Coronary Artery Bypass 50,158 Percutaneous Cardiovascular Procedures 114,894 Total 185,603 Table 8 Orthopedic Group Procedures Variable Florida Short Term Acute Care General Hospitals, 2000 2005 Procedures (Reimbursed by Conventional Medicare) Total Procedures Volumes Major Joint & Limb Reattachment 136,121 Back Surgery Spinal Fusion 25,432 Kn ee Procedures 9,583 Hip & Femur Procedures 51,389 Total 222,525

PAGE 91

77 The cardiac and orthopedic group procedure variables will be discussed further under the next subsection. It will explain how these groups of procedures will be structured and used as v ariables. Listed below in Table 9 is a breakdown of volumes of procedures stratified by ownership for the time period 2000 2005 This table reports volume data for the newly created variables, cardiac group procedures and orthopedic group procedures and f or total procedures reimbursed by conventional Medicare and total procedures performed in Florida hospitals. Also listed are percentages of cardiac and orthopedic group procedures relative to total procedures and procedures reimbursed by conventional Med icare. Table 9 Raw Data & Percentage Breakdown of Cardiac & Orthopedic Group Procedures by Ownership 2000 2005 Procedures Investor Owned Not for Profit Total Procedures 3,492,321 6,788,961 Conventional Medicare Procedures 1,475,156 2,507,551 Cardi ac Group Procedures 63,064 122,539 Orthopedic Group Procedures 76,471 146,054 ( Cardiac Group Total Procedures ) x 100 1.81% 1.80% ( Orthopedic Group Total Procedures ) x 100 2.19% 2.15% ( Cardiac Group Convention Medicare ) x 100 4.28% 4.89% ( Orth opedic Group Conventional Medicare ) x 100 5.18% 5.82% Totals for cardiac group procedures and orthopedic group procedures in Table 9 represent procedures reimbursed by Conventional Medicare Total procedures represent procedures reimbursed by conventi onal Medicare and all other payers. The dataset shows, as depicted in Table 9 that investor owned hospitals performed a total of 63,064 cardiac group procedures and 76,471 orthopedic group procedures reimbursed by conventional Medicare between 2000 and 2005. Not for profits performed about twice the volume of investor owned hospitals carrying out

PAGE 92

78 122,539 cardiac group procedures and 146,054 orthopedic procedures during the same time period. Between 2000 and 2005, investor owned hospitals performed 3,4 92,321 total procedures ( total payers) and 1,475,156 procedures reimbursed by conventional Medicare. Not for profits performed 6,788,961 total procedures ( total payers) with 2,507,551 reimbursed by conventional Medicare. Investor owned hospitals do not exceed that of not for profits in terms of volumes or percentages. In fact, percentages of cardiac and orthopedic group procedures relative to total conventional Medicare procedures are slightly higher in not for profit hospitals with 4.89% and 5.82% resp ectively, compared to 4.28% and 5.18% in investor owned hospitals. Percentages of cardiac group procedures and orthopedic group procedures relative to total procedures ( total payers) are almost even between not for profit and investor owned hospitals show ing respectively 1.80% and 1.81% for cardiac group procedures and 2.15% and 2.19% for orthopedic group procedures. As will be discussed more fully in Chapter 5, Research Methods Model Development for Part 1 t he difference between investor owned hospital s and not for profits in this context is that there is a positive association between the TATL ratio and the cardiac group procedure and orthopedic group procedure variables in investor owned hospitals; that the TATL ratio increases corresponding to percen tage increases in the cardiac group procedure and orthopedic group procedure variables. This association was not observed among not for profits.

PAGE 93

79 Dependent & Independent Variables for Part 1 Following is a discussion of dependent and independent variable s use d in Part 1 and Part 2 of this study Specific variables under Part 1 are discussed in the following paragraphs. This will be followed by a discussion of dependent and independent variables used in Part 2. These variables will be used in model deve lopment and will be discussed again in chapters 5 & 7 for Part 1 and chapters 6 & 8 for Part 2. Selection of Variables for Part 1 TATL Ratio : The literature is rife with researchers using operating margin as the outcome variable to operationalize hospital profitability. Operating margin is an income statement item and is a measure of short term (one year) hospital profitability. The focus of this study was a trend of long term (multi year) financial health and solvency which was better reflected by the TATL ratio The TAT L ratio is defined as the ratio of ( current + long term assets ) / ( current + long term liabilities ) The TATL ratio is a comprehensive long t erm indicator of financial solvency and viability and is derived from the balance sheet. Th e T ATL ratio usually comprised of long term debt. It also reflects investments in plant improvements and equipment, as the greatest portion of assets is usually derived from long term assets including property, plant, and equipment. Based on the objectives of this study the TATL ratio as the outcome variable was selected as an appropriate measure of long term financial strength and solvency for Part 1. The TAT L ratio was used in the modeling phase of the study as a continuous variable. But in some of the preliminary, especially nonparametric analyses the TATL

PAGE 94

80 ratio was examined as a categorical variable. Several descriptive statistical techniques were used to summarize the TATL distribution and to determine a breakpoint needed to convert the TATL ratio to a dichotomy. distribution it was determined that the breakpoint for a high vs. low TA T L ratio would be 5/8 q uartile ( 5/ 8 Q) ( half the distance between the median and third quartile ) The 5/8 TATL quartile was selected because it could be used as a demarcation between h ospitals with high levels of financial strength and solvency vs. hospitals with low le vels. The third quartile, as well as the six th and seventh qu i ntiles, were also examined as potential breakpoints and generated results similar to those when the 5/8 quartile was used It was determined that results of nonparametric tests used in this an alysis would be reported using the 5/8 quartile of the TATL distribution, though any of the other quartile breakpoints would have been acceptable. Ownership: Represented as a dichotomous variable where investor owned hospitals equal 1.0 and all other hospi tals equal zero. All other hospitals in this case include religious not for profit hospitals, other not for profit hospitals, an d hospital district hospitals. Geographic location County Median Income : As cited in the literature review hospital profita bility is associated with geographic location Preliminary analyses showed that h ospitals located in higher income areas tend to have higher TATL ratios than hospitals in economically disadvantaged areas. It often reflects sound strategic direction for management, if it ha s the flexibility, to seek out and locate hospitals in more affluent areas. It is also investor owned hospitals that tend to be able to exercise greater flexibility in geographic location The association

PAGE 95

81 between geographic location a s a strategy for sound management, and the TATL ratio makes geographic location a very good control variable The rationale for selecting geographic location as an independent variable is consistent with discussions in the l iterature r eview regarding Wennb signature theory As already discussed, this theory posits that the geographic variability of per capita rates and Medicare spending for discretionary procedures are more related to geographic location than disease status of populati ons. The best indicator for geographic location for which data was available, was a proxy measure of HUD promulgated median income levels for each of 67 Florida c ount ies 2000 2005 ( www. hud.gov) High income geographic location s were defined as hospitals located in Florida counties with high median income levels. This variable worked well as a continuous, but even better as a categorical variable The high vs. low median income breakpoint is based on quartiles using annualized distributions for county m edian income. It was determined that the optimum breakpoint to distinguish a high income vs. low income county was 5/8 Q (half the distance between the third and fourth quartile) A list of breakpoints for each year at 5/8 Q as well as all quartiles is fou nd in Appendix 4 Discretionary cardiac group procedures: This variable was identified previously in Table 7 and was developed in conjunction with the definition and operationalization of discretionary procedure. It was impractical to make a variable ou t of every DRG code or even out of every procedure. Instead, a number of DRG codes were grouped together to comprise a composite variable titled discretionary cardiac group procedures. T his variable is

PAGE 96

82 comprised of procedures reimbursed by conventional M edicare under M DC 5 (heart procedures) and include valve replacements ( DRG codes 104 & 105 ) CABG procedures (DRG codes 106, 107, 109, 547, 548, 549 & 550), and PCI (DRG codes 112, 516, 517, 518 526, 527, 555, 556, 557 & 558 ) Observations for the cardiac group procedures variable were expressed as percentages for each hospital. Percentages of discretionary cardiac procedure s w ere calculated via the frequency of discretionary cardiac group procedures reimbursed by conventional Medicare for a given hospita l divided by total discharges carried out by that hospital. Percentages for each hospital are also calculated each year, 2000 2005. Of the hospitals that perform cardiac group procedures reimbursed by conventional Medicare, 2000 2005, percentages per hosp ital range from 0.0 % to 15.4 08 % The mean percentage per hospital is 1.1 27 % and standard deviation is 2. 208 %. Discretionary orthopedic group procedures: This variable was identified in Table 8 and was developed in connection with the definition and opera tionalization of the concept of discretionary procedure. Just as with discretionary cardiac group procedures, this variable includes several orthopedic group procedures reimbursed by conventional Medicare under M DC 8 (orthopedic procedures) and include ma jor hip and femur procedures ( DRG code 209 ) back surgery (spinal fusion) (DRG codes 496, 497, 519, 520 & 546 ) and knee procedures (DRG codes 501, 502, 503, 544 & 545 ) Frequencies of the above listed DRG codes were downloaded from the ACHA discharge data for each hospital and year. Percentages were then calculated by dividing

PAGE 97

83 the frequencies of these procedures by total procedures for each hospital for each year, 2000 2005. Of the hospitals that perform orthopedic group procedures reimbursed by convention al Medicare 2000 2005, the percentages per hospital range between 0. 0 % and 11.7 5 %. The mean percentage is 2.1 26 % with a standard deviation of 1. 772 %. In the final mode l s of this study, rates of cardiac group procedures and orthopedic group procedures were transformed to z scores. This was done to manage problem s of multicollinearity and is discussed further in the discussion on model development in Chapter 5. Measuring 1Z in Model Development As just mentioned standard deviations for cardiac group proced ures and orthopedic group procedures, respectively, are 2.2 08 % and 1. 772 %. Later in the development of regression models and results ( c hapter s 5 & 7 ) cardiac and orthopedic group procedures will be described as increasing by 1Z. An increase by 1Z for ca rdiac group procedures will mean that the percentage per hospital increases by 2.2 08 %. An increase of 1Z for orthopedic group procedures will mean that the percentage per hospital will increase by 1. 772 %. Selection Criteria for DRG Codes to be u sed in Mod el Development Observations derived from a DRG code that w as discontinued by Medicare during the study period, but was replaced by a code ( s ) representing the same procedure w ere allowed to remain eligible as observations in either the cardiac group procedu res or orthopedic group procedures variable.

PAGE 98

84 For example, a code in the cardiac group procedures variable, DRG 112, was discontinued in October 2001, but was replaced at the same time by DRG codes 516, 517 & 518, and these codes remained active through 200 5 ( Utah DOH 200 8 ). DRG 112 represented PCI (i.e., angioplasty) while code 516 also represents a PCI, but includ es myocardial infarction (MI). DRG code 517 represents PCI without MI but with an artery stent implant DRG code 518 represents PCI without MI and without a stent implant (Utah DOH 2008) DRG codes 516, 517 & 518 were discontinued and replaced in October 2005 with codes 555 (PCI with MCV diagnosis), 556 (PCI with non drug eluting stent without MCV diagnosis), 557 (PCI with non drug eluting ste nt with MCV diagnosis) & 558 (PCI with non drug eluting stent without MCV diagnosis) (Utah DOH 2008) The newer codes are expanded to provide greater detail on the surgical procedure But all of these codes still represent roughly the same procedure for purposes of this investigation Selection of Variables for Part 2 Total Patient Charges: The identification of an outcome variable for Part 2 was not difficult. Since the research question is whether charges are significantly different with respect to ow nership (i.e., do investor owned hospitals charge significantly more for providing the same procedures and extent of medical services), a sound and evident choice for an outcome variable was total patient charges. Total patient charges are defined as the aggregate amount of charges for all procedures performed for each specific patient prior to discounts. Total charges in the AHCA discharge database include each revenue code including room charges, intensive

PAGE 99

85 care, coronary care, cardiology, anesthesia, e mergency room, anesthesia, intensive care, etc. (AHCA 2005). As already mentioned, total charges in Part 2 of this study apply only to Medicare HMO patients, as there is no impetus for hospitals to increase charges under the fixed rate per DRG code conven tional Medicare program. Total charges are used in this study as a continuous variable in model development and as a dichotomous (high vs. low charges) variable in preliminary categorical analyses. As a dichotomy t he breakpoint for high vs. low charges is at 5/8 Q (half the distance between the median and third quartile). A special provision is made for this variable because total charges increase each year. To adjust for these increases the 5/8Q breakpoint had to be annualized, where the 5/8Q breakpoint equals a different A table is provided in Appendix 5 with a breakdown of total charges by year and quartile. Owne rship: Hospital ownership is the only independent variable that will be used in both parts 1 and 2 of the analysis. This variable is critical to both parts since the focus of the study is about investor owned vs. not for profit hospitals. Just as in Part 1, hospital ownership is r epresented as a dichotomous variable where investor owed hospitals equal 1.0 and all other hospitals equal zero. All other hospitals in this case include religious not for profit hospitals, other not for profit hospitals, and hos pital district hospitals. Geographic Region ( Florida Inspection Area ) : Geographic location was used in Part 2 of the analysis, but was not based on median income as it was in Part 1. As a state regulatory agency AHCA oversees 11 inspection regions that are divided into north,

PAGE 100

86 central, and south Florida geographic sectors. These inspection regions are not relevant to general, acute care hospitals. Instead the inspection regions are used to aid the state of Florida in conducting regulatory compliance ins pections of nursing homes that accept Medicare and Medicaid payments. But as it turned out these regions were found to be suitable proxies for geographic location as Medicare HMO total patient charges differed considerably by geographic region A listin g of Florida counties within each inspection region by north, central, and south geographic sector s is shown in Appendix 6 It should also be observed that disproportionately fewer CABG or valve replacement procedures occurred in the north Florida region by investor owned hospitals. One of the reasons for this was that there are fewer investor owned hospitals that perform CABG and valve replacement procedures in the northern sector. As a result, the north and central regions were collapsed into one level and the variable geographic region was reduced from a three level categorical variable to a dichotomy where south = 1 and north & central = 0. Disease Severity : The preferred variable to adjust for disease severity was APR DRG codes, but this was unava ilable in the AHCA dataset. Other variables included LOS and multiple diagnoses. LOS is defined as discharge date (AHCA 2005) Higher LOS indicates patients being treated are very ill and requ ire greater numbers of days for recovery (Chen 1994). Increasing patient days subsequently generates rising treatment costs ( Rivers Tsai & Munchus 2005) which will drive up hospital charges. This makes LOS a sound control variable in its relationship to charges in Part 2.

PAGE 101

87 Multiple diagnoses are derived from nine secondary clinical diagnostic categories, recorded for all Florida patients. Each additional secondary diagnosis may indicate a higher level of severity, and in turn, a higher level of costs an d charges. The number of secondary procedures was added for each patient to form a continuous variable, which ranged from zero, reflecting least severity to nine, reflecting greatest severity. Although multiple diagnoses provided valuable information in preliminary analyses LOS was determined to be a be tter variable in model development. Valve P rocedure: Since patient charges in Part 2 are derived from both CABG and valve replacement procedures and because the data shows that charges for valve replace ments tend to be higher than CABG procedures, a dichotomous variable was needed to differentiate between the two. This variable uses CABG as the base level. Therefore, coding for this variable is valve replacements = 1, CABG procedures = 0.

PAGE 102

88 Cha pter 4 Research Methods Plan of Analysis for Parts 1 & 2 As described in the p roblem s tatement in Chapter 1 the literature repeatedly provides evidence that hospitals cannot make a profit with Medicare. Yet as this research unfolded it became apparen t that many financially strong hospitals had Medicare as a majority portion of their revenue base. It was also noticeable that many of these hospitals were investor owned and performed higher percentages of discretionary procedures covered by Medicare. T he intent of this research is to understand two strategies certain investor owned hospitals may use to remain financially strong with Medicare as an appreciable portion of their revenue base. Part 1 in chapters 5 & 7 will examine utilizations of certai n DRG codes, representing discretionary procedures that are reimbursed by conventional Medicare to investor owned hospitals The purpose of Part 1 is to determine if selected discretionary procedures reimbursed by conventional Medicare are being utilized at significantly higher rates by investor owned hospitals with high TATL ratios. As discussed in previous sections the TATL ratio was selected to be the outcome variable for Part 1. The population includes 891 investor owned, religious not for profit, o ther not for profit and hospital district acute care general hospitals, 2000 2005.

PAGE 103

89 As discussed in model development and model results for Part 2 in chapters 6 & 8, t he outcome variable for Part 2 is total charges, and the population includes all CABG and valve replacement patient discharges reimbursed by Medicare HMO that occurred in acute care general hospitals, 2000 2005. The purpose of Part 2 is to determine if investor owned hospitals are charging significantly higher amounts than not for profit hospi tals under the same DRG codes reimbursed by Medicare HMO. Rules for Interact ion Models, Model 1 B in Part 1 & Model 2 B in Part 2 Part 1 and Part 2 of this study present both main effects and interaction models. A few ground rules must be established fo r the interaction models entailing the use of constitutive terms and methods of managing multicollinearity due to the inclusion of interacting variables. Constitutive Terms It must be expressed that interaction models in this study include all constitutive (Brambor, Clark & Golder 200 6 ). This means that in a model 0 1 X 1 2 X 2 + 3 X 1 X 2 if interaction is found for 3 X 1 X 2 then both 1 X 1 and 2 X 2 must remain in the model even if these constitutive terms are not statistically significant (Cortina 1993, Sincich 2004 Jaccard & Turrisi 2003 Brambor, Clark & Go lder 2006 ). Managing Multicollinearity in Interaction Models Numerous studies have been done where multicollinearity was incorrectly managed by excluding certain constitutive terms in interaction models. Brambor Clark & Golder (200 6 ) ha ve admonished re searchers that constitutive terms must always be included in regression models except in very rare circumstances.

PAGE 104

90 As will be shown under m odel r esults for Part 1 (Chapter 7) multicollinearity, t hough not a problem in the main effects models in this study, had to be managed in the interaction model in Part 1, Model 1 B. Multicollinearity often accompanies interacting variables even when it i product term (X 1 X 2 ) is an exact nonlinear function of the constituent variables (X 1 and X 2 (Cortina 1993). As just mentioned it is unsound to manage multicollinearity by eliminating constitutive terms or by avoiding interaction altogether (Brambor, Clark & Golder 200 6 ). (Brambor, Clark & Go lder 2006). In a multiplicative interaction model what is of direct interest is not the significance of the model parameters per se, but the marginal effect of X on Y. So the standard error of interest corresponds to the marginal effect also, not the sta ndard error of the parameter estimate (Brambor, Clark & Golder 2006, Jaccard & Turrisi 2003). For interaction models affected by multicollinearity, Brambor, Clark & Golder ( 2006 ) ha ve recommended certain data transformations that may be helpful in model in terpretation. For the models in this study several transformations were used, such as natural logs and centering variables, but z scores were found to be the most useful. The use of z scores helped to deflate variance inflation and provided better compre hension of interacting variables and marginal effects. The use of z score transformations for certain continuous variables in Part 1 are discussed in chapters 5 & 7.

PAGE 105

91 Variance inflation factors (VIF) are an effective and convenient method for assessing mul ticollinearity and is reported with each model in this study Other methods were also used in evaluating multicollinearity and are presented in parts 1 & 2 in the discussion on regression diagnostics in chapters 7 & 8

PAGE 106

92 Chapter 5 Research Methods Mod el Development for Part 1 Descriptive Statistics for Continuous Variables Used in Part 1 The preliminary analysis begins with a review of descriptive statistics of the continuous variables to be used in model development for Part 1 Listed in Table 10 a re the dependent variable, the TATL ratio, and independent variables, orthopedic and cardiac group procedure percentages per hospital Table 10 Descriptive Statistics for Continuous Variables used in Part 1, Models 1 A & 1 B Variable Mean Standard Deviat ion Minimum Maximum TATL Ratio 4.000 4. 530 0.263 31.569 Cardiac Group Procedures 1.127% 2.208% 0.00% 15.408% Orthopedic Group Procedures 2.126% 1.772% 0.00% 11.750% According to Table 10 the TATL ratio has a mean of 4.000 with a standard deviation o f 4.530. The TATL ratio has a minimum value of 0.263 and a maximum value of 31.569. The percentage of cardiac group procedure s shows a mean average of 1.127% with a standard deviation of 2.208%. The percentag e s of cardiac group procedures performed in Florida hospitals range from zero percent (no cardiac group procedures performed in the hospital ) to 15.408% O rthopedic group procedures show a

PAGE 107

93 mean of 2.126% and standard deviation of 1.772% and range from zero percent (no orthopedic group procedures p erformed in the hospital ) to 11.750%. Preliminary Analysis for Part 1 A major criticism of many studies levied by Bagley White & Golomb (2001) was the lack of preliminary analyses preceding model development such as nonparametric and bivariate tests to d etermine the directions and magnitudes of variables. The problem with insufficient preliminary analyses is that variables may be employed in models based on the visceral intuition of the researcher rather than a factual comprehension of the data. In an ef fort to avoid this type of criticism, several preliminary tests were used as precursors to the development of a regression model. Th ese tests were used to establish a secure foundation for selecting independent variables, assuring model validity, and gene rating more confidence in the findings from the models. The preliminary analysis began with a review of contingency tables. Before contingency tables could be created certain continuous variables had to be converted to dichotomies. This required the assi gnment of breakpoints to identify high vs. low levels. As already discussed, the breakpoint for the TATL ratio was set at 5/8 Q. Breakpoints for both cardiac and orthopedic group procedures were determined to be at 0.75Z (three quarters of one standard d eviation above the mean of z = 0) Recall that earlier it was discussed that these variables had been converted to percentages per hospital which had then been transformed to z scores. Table 11 shows results of contingency tables testing the association b etween the TATL ratio and the two chief explanatory variables, cardiac group procedures and orthopedic group procedures.

PAGE 108

94 Table 11 2x2 Contingency Table s for TATL Ratio & Discretionary Procedures Reimbursed by Conventional Medicare, 2000 2005 Variables Odds Ratio & CI Chi Square (p value) Cardiac Procedures Group (0.75Z) x TATL ( 5/8 Q) 15.09 (0.0001) Orthopedic Procedures Group (0.75Z) x TATL ( 5/8 Q) 5.72 (0.017) TATL ( 5/8Q 5/8 Q for the TATL di stribution for all hospitals. Orthopedic Procedures Group (0.75Z): Orthopedic Procedures Gro up Z Cardiac Procedures Group (0.75Z): Cardiac Procedures Gro up Z The association between the TATL ratio and cardiac group procedures is indicated via an odds ratio of 2.03 (p < 0.0001). The odds ratio for orthopedic group procedures is 1.51 (p < 0.0017) with the lower end of the confidence limit down to 1.08, indicating fairly weak association. Although t he magnitude of association between the TATL ratio and orthopedic group procedures is shown to be low in Table 11 association was later shown to be higher between the TATL ratio and the interacting variable of ownership and orthopedic group procedures. Ex aminations of Potential Control Variables for Part 1 O ther factors aside from cardiac & orthopedic group procedures may help explain rises or declines in the TATL ratio The l iterature r eview cites several authors who argue that profitability and long ter m financial strength represented by high TATL ratios should be credited to good management more than any other factor. Two factors that are associated with both good management and long term financial health are hospital ownership and geographic location 5/8 Q median county income in the distribution of 67 Florida counties).

PAGE 109

95 C ontingency table analysis was used to test each control variable and its association to the TATL ratio. Results of the analyses are shown in Table 1 2 Table 1 2 2x2 Cont ingency Table s for TATL Ratio & Hospital Ownership, Geographic L ocation 2000 2005 Variables Odds Ratio & CI Chi Square (p value) Ownership x TATL ( 5/8 Q) 165.96 (0.0001) Geographic location (county 5/8 Q ) x TATL (5/8Q ) 5.55 (0.0194) TATL ( 5/8 5/8 Q for the TATL distribution for all hospitals. Ownership: Investor Owned Hospital = 1; else = 0 Geographic location ( 5/8 5/8 Q for the median incom e 5/8 Q = 1; else = 0. Ownership shows the greatest association with the TATL ratio demonstrated by an odds ratio = 7.50 (p < 0.0001) The odds ratio for geographic location represented by 5/8 Q) vs. low Florida county median income level s shows a low er odds ratio of 1. 42 with the lower bounds of the CI at 1.06 Geographic location is not as strong of a variable as ownership, but l ater the analysis will show it to be more useful as an interacting variable. Pretesting for Interaction Using the Breslow Day Test In this study potential interaction was assessed in preliminary analyses and then incorporated into hypotheses preceding model development. This practice adhered partly to crit icisms imposed by Bagley White & Golomb (2001) and warnings by Louis (2008) Bagley White & Golomb ha ve been critical of researchers who do not test for interaction in their models, even when such testing is warranted. Louis warned against the cavalier addition of interaction terms to final models and strongly recommends that

PAGE 110

96 interacting variables should be represented in hypotheses after being detected in preliminary analyses. d on careful preliminary examination is because interacting variables have less power than main effects variables, and as already mentioned often result in problems with multicollinearity. Interactions are also vulnerable to outlier effects, and the highe r order the interaction the more distorting the outlier can be. To circumvent possible criticisms that could be levied against the methods employed in this investigation, the potential presence of interaction was explored thoroughly before incorporating interacting variables in hypotheses and placing them in regression models. Breslow Day tests were examined for s t ratified tables to test the potential presence of interaction between variables in their association with the TATL ratio. The Breslow Day tes t is used to assess homogeneity between the odds ratios in stratified contingency tests (Agresti 1999, Stokes Davis & Kock 2003) I nteracting variables identified by the Breslow Day test may be used to generate hypotheses for interaction occurring in mod el development (Schwartz 2004). Results of the Breslow Day tests and Cochran Mantell Henzell ( CMH ) odds ratios are shown in Table 1 3 Bear in mind that the most useful information on the table is the results of the Breslow Day tests. The CMH odds ratio i s not so useful because it is usually recommended to calculate odds ratios for separate tables when the Breslow Day test indicates that interaction is present.

PAGE 111

97 Table 1 3 Breslow Day Tests to Assess Interaction using 2x2 Contingency Table s & Stratified Ana lysis, 2000 2005 Variables CMH & CI Chi Square (p value) Breslow Day (p value) Ownership x TATL (2.5Q) controlling for Geographic location ( 5/8 Q) 165.54 (0.0001) 13.81 (0.0002) Orthopedic Group Procedures x TATL ( 5/8 Q) controlling for ownership 1.12 1.68 2.49 6.01 (0.0142) 21.92 (0.0001) Cardiac Group Procedures x TATL ( 5/8 Q) controlling for Ownership 1.42 2.18 3.32 12.32 (0.0004) 8.46 (0.0036) Ownership: Investor Owned Hospital = 1; else = 0 Orthopedic Group Procedures (0.75Z): Orthopedic Group Procedures Z Cardiac Group Procedures (0.75Z): Cardiac Group Procedures Z else = 0. Geographic location ( 5/8 5/8 Q for the median income distribution for 5/8 Q = 1; else = 0. Results i n Table 1 3 indicate a low p value (0.0002) for the Breslow Day test for ownership and TATL ratio controlling for geographic location This may indicate two way interaction between ownership and geographic location Findings also suggest there is interaction between orthopedic group procedures and ownership in their association with the TATL ratio (Breslow Day = 21.92, p < 0.0001) as well as cardiac group procedures and ownership (Breslow Day = 8.46, p < 0.0036 ) These findings present a sound basis to incorporate these interacting terms in model development. Correla tion Matrices A follow up method is used to buttress findings from contingency tables regarding the variable ownership and orthopedic and cardiac group procedure variables. A correlation matrix to determine magnitude s of association between the TATL rati o and each variable, stratified by ownership is shown in Table 1 4

PAGE 112

98 Table 1 4 Pearson & Spearman Correlation Coefficients for the Association between the TATL Ratio and Discretionary Procedures Stratified by Ownership 2000 2005 Pearson Coefficients As sociation with TATL Spearman Coefficients Association with TATL TATL Investor Owned Not for Profit Investor Owned Not for Profit Cardiac Group Procedures 0.33 1 <.0001 0.00298 0.9474 0. 44 2 0. 0001 0.04714 0.2957 Orthopedic Group Procedures 0.466 <.000 1 0.02875 0.5238 0.49 7 0. 0001 0.00122 0.9783 For each variable listed, whether the association is assessed by Pe a rson linear or Spearman ordinal ( via rankings ) correlations the coe fficients show a significant association under the investor owned stra tum A Pearson coefficient of 0.331 (p < 0.0001) and Spearman coefficient of 0.442 (p < 0.0001) are shown for the association between the TATL ratio and cardiac group procedures. Pearson and Spearman show, respectively, coefficients of 0 .46 6 (p < 0.0001) and 0.497 (p < 0.0001) for the association between the TATL ratio and orthopedic group procedures. No significant association is found using the Pearson or Spearman correlation matri ces between the TATL ratio and either of the cardiac or orthopedic group procedure variables in the not for profit category Comparison of Discretionary Procedure Mean Percentages C ardiac and orthopedic group procedures, reimbursed by conventional Medicare represent small percentages per hospital. Keep in mind that the denomi nator in the percentage calculation is total procedures per hospital under all payment sources. This is very large compared to the numerator of cardiac or orthopedic group procedures reimbursed only by conventional Medicare.

PAGE 113

99 But even though percentage s of cardiac and orthopedic procedures are small, they are still significantly different with respect to hospital ownership and TATL ratio. An additional test was performed to compare the mean percentages of cardiac and orthopedic procedures based on ownership and TATL ratio. The ownership variable and TATL variable were combined to form a special variable to represent investor owned hospitals with TATL ratios 5/8Q. This variable was formed so t test s could be carried out to compare the means of cardiac group procedure percentages and orthopedic group procedure percentages between investor owned hospitals with TATL ratios 5/8Q vs. all other hospitals (i.e., all not for profit hospitals and investor owned hospitals with TATL ratios < 5/8Q). The results of the tests are shown in Table 1 5 Table 1 5 Mean Percentage Differences between Investor Owned Hospitals with High TATL Ratios and All Other Hospitals Card iac & Orthopedic Group Procedures, Reimbursed by Conventional Medicare, 2000 2005 Discretionary Procedures Investor Owned, TATL ratios 5/8Q All Other Hospitals t test (p value) Cardiac Group Procedures 1.6% 0.9% 3.48 (0.0006) Orthopedic Group Procedures 2.67 % 1.95 % 5.23 (0.0001) Means for cardiac group procedures for investor owned hospitals with TATL ratios 5/8Q vs. all other hos pitals is 1.6% and 0.9%, respectively. A Swaitherwhite t test was performed (i.e., the folded F test indicated unequal variances, p < 0.0001 so a pooled t test was not used ) and yielded t = 3.48 (p = 0.0006). This indicates that mean percentages of car diac group procedures are significantly higher in investor owned hospitals with high TATL ratios vs. all other hospitals (not for

PAGE 114

100 profit hospitals with high and low TATL ratios and investor owned hospitals with TATL< 5/8Q). Mean percentages of orthopedic g roup procedures for investor owned hospitals with TATL ratios 5/8Q vs. all other hospitals were 2.67% and 1.95%. A pooled t test was performed (i.e., the folded F test did not indicate unequal variances, p < 0.8858) and yielded t = 5.23 (p < 0.0001). T his indicates that mean percentages of orthopedic group procedures are significantly higher in investor owned hospitals with high TATL ratios vs. all other hospitals ( total not for profit hospitals with high and low TATL ratios and investor owned hospitals with TATL< 5/8Q) Statements of Hypotheses for Part 1 Now t hat the data structure, parameters, variables, DRG selection criteria and preliminary analyses have been described, hypotheses can be delineated. Several hypotheses were formulated in connectio n with the research objectives and preliminary analyses presented in Part 1. Hypotheses for Part 1 addressing the association between the TATL ratio and cardiac and orthopedic group procedures financed by conventional Medicare are written as follows: 1) Ther e is a statistically significant positive association between the continuous dependent variable TATL ratio and ownership, controlling for other variables which may also be associated with the TATL ratio. 2) There is a statistically significant positive asso ciation between the continuous dependent variable TATL ratio and the two way interaction term of ownership and cardiac group procedures, controlling for other variables which may also be associated with the TATL ratio. 3) There is a statistically significan t positive association between the continuous dependent variable TATL ratio and the two way interaction term of ownership and orthopedic group procedures, controlling for other variables which may also be associated with the TATL ratio.

PAGE 115

101 4) There is a stat istically significant positive association between the continuous dependent variable TATL ratio and the two way interaction term of ownership and geographic location controlling for other variables which may also be associated with the TATL ratio. Thes e hypotheses include interacting variables, so it is assumed that all pertinent main effects and constitutive terms will remain in the model. Now that h ypotheses for Part 1 have been delinea ted they will be tested via regression models which are reported in a discussion on model results for Part 1 in Chapter 7 Preceding the testing of hypotheses and model results for Part 1 is a discussion on model development for Part 2 which is provided in Chapter 6.

PAGE 116

102 Chapter 6 Research Methods Model Development for Part 2 Preliminary Analysis for Part 2 The discussion of research methods now shifts from an examination of volumes and percentages of discretionary procedures reimbursed by conventional Medicare to an investigation of total charges for certain discre tionary procedures reimbursed by Medicare HMO. The population parameter for Part 1 was investor owned and not for profit Florida acute care general hospitals reimbursed by conventional Medicare, 2000 2005. One of two population parameters for Part 2 is patient discharge cases in acute care general hospitals reimbursed by Medicare HMO, 2000 2005. A second population parameter in Part 2 is defined by two discretionary procedures, CABG and valve replacements, of which are discussed more fully later in this chapter Part 2 of the analysis will test the association between total charges, as a continuous dependent variable and the dichotomous independent variable hospital ownership. The analysis will address the question as to whether investor owned hospita ls charge significantly greater amounts for the same DRG codes and extent of medical services than their not for profit counterparts.

PAGE 117

103 was discussed in the literature review and refers to the geographic variability not o nly of the prevalence of Medicare reimbursed discretionary procedures, but also of Medicare spending levels that cannot be explained by the disease status of the population or the extent of medical services provided. One of the objectives of Part 2 is to d etermine if investor owned hospitals, as well as not for profits, assess significantly higher charges under the Medicare HMO program if investor owned hospitals char ge more for the same services, and if these charges increase disproportionately by geographic area. The analysis of total charges in Part 2 began, just as it had in Part 1, with a review of discretionary procedures represented by DRG codes from Appendix 1. Part 2 imposed the same criteria of selecting DRG codes that could be defined as discretionary based on clinical and geographic (variability) grounds. As it turned out the same DRG codes that were determined to be important for the study under convention al Medicare in Part 1 were also found to be useful in the study of total charges under Medicare HMO in Part 2. Just as it was in Part 1, model development in Part 2 is preceded by a review of descriptive statistics and extensive bivariate and nonparametric analyses to develop strong groundwork for the selection of variables and structuring of models. Descriptive Statistics for Continuous Variables Used in Part 2 The preliminary analysis for Part 2 begins with a review of descriptive statistics of the cont inuous variables that will be used in model development for Part 2 Listed in Table 1 6 is the dependent variable, patient charges, and independent variable LOS.

PAGE 118

104 Table 1 6 Descriptive Statistics for Continuous Variables used in Part 2 Variable Mean Stand ard Deviation Minimum Maximum Patient Charges $102,488 63,696 3,744 810,610 Length of Stay (LOS) 10.327 6.360 1 .0 92.0 The table shows that mean patient charges are $102,488 with a standard deviation of $63,696. This variable has a very broad range with a minimum of $3 744 and maximum of $810,610. Mean days of hospitalization is 10.327 days with a standard deviation of 6.360 days Minimum LOS is one day with a maximum of 92.0 days. It was surprising that the minimum value for LOS was one day, as it was thought that the minimum value may be higher. Upon closer examination it was found that 33 observations had LOS equal to one day. Of these observations, three were discharged to home care under the supervision of a home healthcare organization whi le the others were discharged to home without further supervision As the preliminary analys i s ensues more information will be provided about these two variables Additional information forthcoming will be even more meaningful because it will show patien t charges and LOS stratified by hospital ownership and procedures (i.e., CABG vs. valve replacements). Selection of DRG Codes for Part 2 Analysis of Mean Charges for DRG Codes Reimbursed by Medicare HMO The selection of DRG codes entailed a review of diffe rences of mean total charges between investor owned and not for profit facilities. An examination of these differences in mean charges is shown in Table 1 7 using percentage changes and t tests. Two of the t tests shown in the table are pooled t tests, bu t all others are Satterthwaite t

PAGE 119

105 tests. The Satterthwaite t test was used for tests where the significant folded F test indicated unequal variances. Before reviewing the table, it should be pointed out that any reader familiar with hospital reimbursemen t amounts under Medicare (conventional or HMO) will find these amounts excessive. What is actually reimbursed under Medicare HMO tends to be only a fraction of total patient charges. Reimbursement levels are also important because they, not charges, repr esent actual amounts of federal spending. Therefore, levels of reimbursement under Medicare HMO will be addressed in model development for Part 2 in Chapter 8. charges, and the inconsistencies in charging patterns based on hospital ownership and geography. Table 1 7 is now presented below. Table 1 7 Analysis of Mean Total Charges for Selected Medicare HMO Procedures by Ownership, 2000 2005 Procedure Investor Owned Not for Profit Differences Actual (%) t test (P values) Valve Replacement 168,685 114,004 54,681 (48.0%) 17.89 (0.0001) CABG 126,402 81,734 44,668 (54.7%) 33.01 (0.0001) PCI 57,233 42,980 14,253 (33.2%) 32.27 (0.0001) Joint & Limb 47,134 34,899 12,235 (35.1%) 35.8 5 (0.0001) Hip & Femur 39,324 29,216 10,108 (34.6%) 20.63 (0.0001) Spinal Fusion 62,567 58,861 3,706 (6.3%) 1.77 (0.0764) Knee Procedures 53,734 43,727 10,007 (22.9%) 5.86* (0.0001) Prostatectomy 20,317 14,935 5,382 (36.0%) 10.73 (0.0001) Mastectomy 21,408 15,439 5,969 (38.7%) 10.50* (0.0001) Pooled t tests where variances are equal. All other tests were Swaitherwaite t tests, as Folded F tests showed that variances of total charges between investor owned and not for profit hospitals were found to be unequal

PAGE 120

106 With the exception of spinal fusion (t = 1.77, p = 0.0764), results of t tests indicate that investor owned facilities assess total charges for each procedure significantly higher than their not for profit counterparts. Percentage difference s for each of the means of CABG and valve procedures respectively are approximately 48.0% & 54.7% greater. For PCI, investor owned hospitals charge about 33.2% more than not for profits. Investor owned hospitals charge between 34% & 39% more for joint & limb procedures, hip & femur procedures, prostitectomies, and mastectomies, and 23% more for knee procedures. Large as well as statistically significant differences for mean charges between investor owned and not for profit facilities provide a strong bas is for using these procedures in model development for total charges. Selection of CABG & Valve Replacements as Population Parameters It has already been stated that population parameters for models in Part 2 are patients treated with CABG and valve replac ement procedures under Medicare HMO, 2000 2005. It should be addressed why CABG and valve procedures were selected as population parameters and not the other discretionary procedures listed back on Table 1 7 These procedures were chosen because they best adhered to research criteria regarding DRG code selection. They show the highest dollar amounts for total patient charges, and the highest percentage differences between investor owned and not for profit hospitals. Consequently, CABG (DRG codes 106, 107, 109, 547, 548, 549, 550) and valve replacement procedures (DRG codes 104, 105), reimbursed by Medicare HMO, 2000

PAGE 121

107 2005, represent population parameters, and procedures performed under these DRG codes will populate the dataset for analyses in Part 2. Revi ew & Selection of Potential Covariates Model development will feature ownership as the explanatory variable. Just as in Part 1, models in Part 2 need to control for factors that in addition to ownership are associated with the dependent variable, total c harges. After a review of preliminary analyses, the selection of covariates included LOS and geographic sector. LOS is a continuous variable while geographic sector is a dichotomy representing the south region vs. the combined north & central regions, w ith north and central regions as the base level. Controlling for Patient Disease Severity LOS and multiple diagnoses were both considered as covariates because they represent measures of patient severity. Total patient charges will increase corresponding to additional days of hospitalization or in connection with greater numbers of secondary diagnoses. There was a problem using both variables in the same model because they are each associated with charges and with each other. Because these variables are related and redundant, a determination had to be made to select the variable that had the strongest correlation with total charges. To do this, a correlation matrix was created using total charges, LOS, and multiple diagnoses. The results are shown belo w in Table 1 8

PAGE 122

108 Table 1 8 Pearson & Spearman Correlation Coefficients, Associations between Total Charges & Potential Variables for Model Development, 2000 2005 Total Patient Charges Pearson (p value) Spearman (p value) LOS Pearson (p value) Spearman (p value) Multiple Diagnoses Pearson (p value) Spearman (p value) Total Patient Charges Pearson (p value) Spearman (p value) 1.00 1.00 0.648 (0.0001) 0.550 (0.0001) 0.331 (0.0001) 0.390 (0.0001) LOS Pearson (p value) Spearman (p value) 0.648 (0.0001 ) 0.550 (0.0001) 1.00 1.00 0.376 (0.0001) 0.482 (0.0001) Multiple Diagnoses Pearson (p value) Spearman (p value) 0.331 (0.0001) 0.390 (0.0001) 0.376 (0.0001) 0.482 (0.0001) 1.00 1.00 Not surprisingly, the Pearson correlation coefficient for LOS is very high at 0.648 (p < 0.0001) for the linear association between charges and LOS. The association between ordinal rankings using the Spearman correlation coefficient is 0.550 (p < 0.0001). The linear association between charges and multiple diagnoses is also fairly high at 0.331 with ordinal (rankings) association even higher at 0.390(p < 0.0001). Although the coefficients for multiple diagnoses are high, they are still not nearly as high as LOS. What is also noticeable, and as previously discussed, is the association between LOS and multiple diagnoses, Pearson = 0.376 (p < 0.0001) and Spearman = 0.482 (p < 0.0001) indicating high association between these two covariates. Ultimately, LOS was selected as the covariate in model development because it de monstrated much higher correlation with total patient charges.

PAGE 123

109 Charges per Hospital Day The mean average LOS for CABG procedures is 9.907 days with a standard deviation of 5.816 days. There is a minimum of two days and a maximum of 92 days. Valve proced ures show a mean average of 11.404 days with a standard deviation of 7.47 days, a minimum of one day and a maximum of 71 days. Investor owned hospitals show a higher mean LOS for CABG procedures of 10.113 days, while not for profits show a mean of 9.806 da ys. Investor owned hospitals also show slightly higher mean average LOS for valve replacement procedures. Not for profits show a mean of 11.308 days, and investor owned hospitals show a mean of 11.644 days. What is different between investor owned hospit als and not for profits with respect to LOS is charges per day of hospitalization. LOS and total charges were used to calculate total charges per patient day for valve replacements and CABG procedures, stratified by ownership (i.e., total charges total days of hospitalization), which is presented in Table 1 9 Table 1 9 Charges Per Patient Day Valve Replacement & CABG Procedures, Investor Owned vs. Not for Profit Hospitals, Medicare HMO, 2000 2005 Valve Replacements CABG Charges/LOS Investor O wned Not for Profit Investor Owned Not for Profit Mean Charges/LOS $18,143 $12,515 $13,526 $9,246 Standard Deviation $17,730 $11,529 $5,820 $6,129 Coefficient of Variation 0.98 0.92 0.43 0.66 Mean charges per patient day for valve replacements in in vestor owned hospitals is $18,143 compared to $12,515 in not for profits. Average charges per patient day in

PAGE 124

110 investor owned hospitals are $13,526 for CABG procedures compared to $9,246 in not for profit hospitals. The preliminary findings in Table 1 9 indi cate that higher mean total patient charges by investor owned hospitals are not due to higher average LOS. A similar table (not shown) was constructed for mean total patient charges by numbers of secondary diagnoses. It was found that mean total patient charges were also much higher for each diagnostic level for investor owned hospitals. This indicates that mean higher charges are not due to investor owned hospitals caring for sicker patients (i.e., those with higher numbers of secondary diagnoses). Geog raphic Region (Florida Inspection Area) As already mentioned, the geographic region variable was collapsed from a three level to a dichotomous variable where combined north and central regions represent the base level. However, it is still useful to revie w the disparities in charges between each region. Geographic region turned out to be a very important variable because it was found that total charges for valve replacement and CABG procedures differ significantly with respect to geographic sector. Figu re 4 is a Pareto chart showing mean charges on the left y axis and volumes on the right y axis.

PAGE 125

111 Figure 4 Source: AHCA Patient Discharge Dataset, 2000 2005 Mean charges, covering years 2000 2005, for valve replacement and CABG procedures run $72,834 in North Florida while they are $100,021 in central Florida and over 120,000 in south Florida. Volumes also differ with respect to geographic sector, as there are around 2,000 valve replacement and CABG procedures performed in north Florida, over 5,000 in central, and a little over 6,000 in south Florida. Figure s 5 and 6 are also Pareto charts showing mean charges and volumes by sector for each individual procedure. These two charts report the same information as Figure 4, but charges are broken down se parately for CABG and valve procedures.

PAGE 126

112 Figure 5 Figure 6 Source: AHCA Patient Discharge Dataset, 2000 2005 Source: AHCA Patient Discharge Dataset, 2000 2005 Differences in m ean charges as w ell as volumes by geographic sector are consistent with Wennberg surgical signature theory Recall that a certain aspect of this theory, discussed in the literature review is that certain types of discretionary procedures and related federal spending s ignificantly differ by geographic region with no link to disease status of the population. It is possible that differences in volumes and patient charges, shown in figures 5 and 6 are consistent with the surgical signature theory. It has been demonstrat ed so far that patient charges are broadly variable with respect to hospital ownership and geography. Figure 7 now shows a comparison of Medicare HMO total patient CABG and valve replacement charges by geographic region, stratified by ownership. Mean Charges & Volumes of Medicare HMO Financed Valve Replacements by Geographic Sector 2000 2005 20,000 40,000 60,000 80,000 100,000 120,000 140,000 160,000 North Central South 200 400 600 800 1,000 1,200 1,400 1,600 1,800 2,000 Volumes Mean Charges Vollumes Mean Charges Mean Charges & Volumes of Medic are HMO Financed CABG Procedures by Geographic Sector 2000 2005 20,000 40,000 60,000 80,000 100,000 120,000 North Central South Mean Charges 500 1,000 1,500 2,000 2,500 3,000 3,500 4,000 4,500 Volumes Mean Charges Volumes

PAGE 127

113 Figur e 7 Source: AHCA Discharge Dataset, 2000 2005 It appears from this chart that investor owned hospitals are not the only ones increasing Medicare HMO charges in certain regions of Florida. Though not as high as investor owned hospitals, southern Florida n ot for profits also assess higher CABG and valve replacement charges relative to not for profits in the north and central regions. Total patient charges increase by approximately $14,000 (10.8%) between the south geographic region and other regions for inv estor owned hospitals. But not for profits increase charges in general by almost $30,000 (37.5%) for procedures occurring in hospitals located in the southern regions vs. those in the central and northern regions. Contingency Tables Similar to methods us ed in Part 1, covariates were also tested using contingency tables in Part 2. For total charges to be used in the contingency table it had to be & Dependent Variables

PAGE 128

114 patient charges is 5/8Q, half the distance between the second and third quartile in the distribution of total charges. Results of the first contingency table analyses are shown in Table 20 The se tables are used to test the association between total charges and ownership and total charges and south geographic region. Table 20 2x2 Contingency Table Analysis for Total Charges and Ownership & Geographic Region, Medicare HMO Financed Valve Replacem ent & CABG Procedures, 2000 2005 Variables Odds Ratio & CI Chi Square (p value) Ownership x Total Patient Charges 913.94 (< 0.0001) South Geographic Region x Total Patient Charges 201.53 (< 0.0001) Ownership: Inv estor owned hospital = 1; Not for profit= 0. South Geographic Sector: South = 1; Central or North = 0 The association between ownership and total charges shows an odds ratio o f 3.67 greater in investor owned hospitals relative to not for profits. Another odds ratio, 1.83 (p < 0.0001) is then observed between total patient charges and south geographic region. These odds ratios demonstrate a positive association between total char ges and ownership and total charges and south geographic region. This indicates that ownership and geographic region will be sound variables in model development. Homogeneity of the Odds Ratios As in Part 1, c ontingency tables are again used to pre test the presence of interaction for the model development stage. This is done via stratifying the analysis and then testing for the homogeneity of the odds ratio via a Breslow Day test. Table 2 1

PAGE 129

115 presents results of stratified contingency table analysis. As stated in Part 1, it is the Breslow Day test that is the most meaningful information on the table, as it is often advisable to calculate a separate odds ratio for each table when interaction is present. Table 2 1 Breslow Day Test to Assess Interaction using a 2x2 Contingency Stratified Table, CABG & Valve Replacements Financed by Medicare HMO, 2000 2005 Variables CMH & CI Chi Square (p value) Breslow Day (p value) Ownership x Total Charges controlling for South Geographic Sector 872.46 (< 0.0001) 36.98 ( 0.0001) The table shows results of the contingency table testing the association between total patient charges and south geographic sector controlling for ownership. A Breslow Day test value of 36.98 (p = 0. 0001) indicates the presence of interaction between ownership and south geographic region. It provides sound rationale to test the interaction of these two variables in model development. Statements Hypotheses for Part 2 Now that the preliminary analyses have been completed, hypotheses will be stated for Part 2 which addresses charges for CABG and valve replacement procedures financed by Medicare HMO. These hypotheses are written as follows. 1) There is a statistically significant positive relationship bet ween the continuous dependent variable, total patient charges and LOS for CABG and valve replacement procedures reimbursed by Medicare HMO, 2000 2005, controlling for other variables which may also be associated with total patient charges. 2) There is a stati stically significant positive relationship between the continuous dependent variable, total patient charges and ownership for CABG and valve replacement procedures reimbursed by Medicare HMO, 2000 2005, controlling for other variables which may also be ass ociated with total patient charges. 3) There is a statistically significant positive relationship between the continuous dependent variable, total patient charges and geographic sector for CABG and valve replacement procedures reimbursed by Medicare HMO, 2000 2005,

PAGE 130

116 controlling for other variables which may also be associated with total patient charges. 4) There is a statistically significant positive relationship between the continuous dependent variable, total patient charges and the two way interaction term for ownership and geographic sector for CABG and valve replacement procedures reimbursed by Medicare HMO, 2000 2005, controlling for other variables which may also be associated with total patient charges. Now that hypotheses for Part 2 have been delineated they will be tested via regression models of which will be presented in Chapter 8. Prior to the discussion on model results for Part 2 is a presentation on model results for Part 1 which is provided in Chapter 7.

PAGE 131

117 Chapter 7 Model Results for Part 1 Testing the Association between the TATL Ratio and Discretionary Procedures ( Reimbursed by Conventional Medicare ) Organization of Model Findings Model results for Part 1 will feature M odel 1 A a main effects model and Model 1 B an interaction model The presentation of model findings is followed by a discussion on regression diagnostics for Model 1 B Model 1 A : Main Effects Model Crude estimates indicated that investor owned hospitals with high TATL ratios carry out certain discretionary cardiac and orthopedic group procedures reimbursed by conventional Medicare at higher rates (not volumes) than other hospitals. This will now be further tested in the development of Model 1 A which contains the covariates discussed in Chapter 3 Model 1 A, a main e ffects MLS model is expressed as follows: Regression Equation for Model 1 A TATL = 0 1 X 2 P 3 C 4 R + Where: X = Ownership P = Geographic location C = Z Cardiac Group Procedures R = Z Orthopedic Group Procedures Model 1 A was employed and an examination of initial findings is shown below in Table 2 2

PAGE 132

118 Tab le 2 2 Model 1 A: Main Effects Model Source DF Sum of Squares Mean Square F Value Pr>F Model 4 5, 647 1,411.8 91.28 <0.0001 Error 8 86 13, 703 15. 46 Corrected Total 8 90 19, 351 Root MSE 3.9 3 Dependent Mean 3.9 3 R 2 0. 2918 CV 100.12 Adjusted R 2 0. 2886 The main effects model shows a global F test of 91.28 (p < 0.0001) indicating that the parameter estimate of at least one variable in the model is significant. Findings also show a dependent mean, 3.9 3 equaling the mean average of the TATL ra tio for all observations of the dependent variable. The (textbook) ideal for the CV in the regression output is 10 .0 (Sincich 2004). Regressions using real world data rarely derive a CV of 10.0 but Model 1 100.12 is high even for real world da ta. There is an R 2 of 0. 2918 and adjusted R 2 of 0. 2886 w hich is fairly low regarding linearity and model fit. It should be pointed out that the R 2 is very close to the adjusted R 2 This may be important in model results because a large distance between t he R 2 and the adjusted R 2 sometimes indicates there are too many variables in the model, and perhaps one or more of them should be excluded (Sincich 2005). Table 2 3 presents parameter estimates, p values, and VIF for Model 1 A At the bottom of this tabl e there is documentation provided on the coding of variables in Model 1 A.

PAGE 133

119 Table 2 3 Model 1 A: Main Effects Model Parameter Estimates Variables Parameter Estimates Standard Errors t Values Pr > |t| VIF Intercept 1.55623 0.23721 6.56 <.0001 0 Ownershi p 4.03304 0.26578 15.17 <.0001 1.00531 Geographic location 0.99093 0.27618 3.59 0.0004 1.07026 Z Cardiac Procedures Group 0.80545 0.13447 5.99 <.0001 1.04047 Z Orthopedic Procedures Group 0.85122 0.13738 6.2 0 <.0001 1.08599 Coding for Variables: Owner ship: Investor owned hospital = 1; Not for Profit = 0 Cardio Group Procedures : Z score transformed p ercent of patients per hospital treated under a DRG code identified as a cardiac group procedure in this study. Orthopedic Group Procedures : Z score trans formed p ercent of patients per hospital treated under a DRG code identified as an orthopedic procedure in this study. Geographic location 5/8 Q = 1; Else = 0 VIF, was very low in Model 1 A, ranging roughly between 1.0 and 1. 08 for all in dependent variables, indicating no problem with multicollinearity. P arameter estimates for ownership and geographic location shown in Table 2 3 are significant and very high. Ownership shows a parameter estimate of 4.03 indicating that investor owned h ospitals on the average have TATL ratios that are four points above the average TATL ratio of not for profits Hospitals located in higher income counties distribution for all Florida counties), on the average, have TATL ratios that are almost one (0.99) point higher than facilities located in lower income counties. C a rdiac and orthopedic group procedures have significant but low er betas of 0. 81 & 0. 7 5 respectively. Bear in mind that cardiac and orthopedic group procedures are coded as z scores. So in the interpretation of the betas, h ospitals that perform these procedures at a rate of 1Z higher than the mean (z = 0) are show n to have TATL ratios

PAGE 134

120 respectively 4/5 of a point and 3/4 of a point higher than hospitals that perform these procedures at a rate of z = 0. Model 1 B : Interaction Model The main effects model was very useful in evaluating variables and gaining further in sights into the workings of the data, but beyond this it became less relevant in light of the findings in the subsequent interaction model, 1 B. Thus, Model 1 B is the model used to test the hypotheses delineated for Part 1 of this investigation. The MLS regression equation for Model 1 B is shown as follows. Regression Equation for Model 1 B TATL = 0 1 X 2 P 3 C 4 R 5 XP 6 XC 7 X R Where: X = Ownership P = Geographic location C = Z Cardiac Group Procedures R = Z Orthopedic Group Procedures Partial F tests were very helpful in evaluating the inclusion and usefulnes s of certain interactive variables. The use of partial residuals helped evaluate and determine unnecessary the inclu sion of a quadratic term ( s ) in the model Initial findings of Model 1 B are shown below. Table 2 4 Model 1 B : Interaction Model Source DF Sum of Squares Mean Square F Value Pr>F Model 7 7,867.30 1,123.90 86 42 <0.0001 Error 8 83 11,483 1 3 00 Corrected Total 890 19 351 Root MSE 3. 61 Dependent Mean 3.93 R 2 0. 4066 Coef ficient of Var iation (CV) 91 8 0 Adjusted R 2 0. 4019

PAGE 135

121 Tabl e 2 4 shows that the interaction model 1 B has seven degrees of freedom (i.e., seven variables) due to interaction terms added in the model compared to the main effects model which had four The dependent (TATL ratio) mean in the interaction model is of co urse identical to that of the main effects model at 3.93 The interaction model is a better fitting model than the main effects model. The root MSE and CV are slightly smaller in the interaction model at 3. 61 and 91 8 0 compared to 3. 93 and 100 12 in th e main effects model. Likewise the adjusted R 2 is 0. 4019 in the interaction model compared to 0. 2886 in the main effects model, reflecting better goodness of fit and a higher percentage of explanation of the variance in the data by the model Significanc e of the Interaction Model Using the F Test Before proceeding with a review of parameter estimates, i t should first be determined if the interaction model is significant relative to the main effects model. A method to test the significance of the interacti on model is the general F test Th e formula for the F test is as follows: Source s : Aichen & West 1991, Young 2006 Jaccard & Turrisi 2003 The test was calculated yielding a very high F value of 55. 76 (p < 0.0001) with R 1 2 = 0.2886 and R 2 2 = 0.4019; df 1 = k 2 k 1 (7 4 ) = 3 and df 2 = N k 2 1 (891 7 1 ) = 883. The F test indicates that M odel 1 B is significant relative to Model 1 A.

PAGE 136

122 Parameter Estimates Now that the significance of the interaction model 1 B has been established the parameter estimates for this model are shown in Table 2 5 Table 2 5 Model 1 B : Interaction Model Parameter Estimates Variable Parameter Estimate Standard Error t Value P<| t | VIF 0 Intercept 2.03691 0.25435 8.01 <.0001 0 1 X Owner ship 2.76 6 0.38406 7.2 <.0001 2.49652 2 P Geographic location 0.095 7 0.34185 0.28 0.7796 1.95004 3 C Z Cardiac Group Procedures 0.0004 0.20616 0 0.9984 2.90873 4 R Z Orthopedic Group Procedures 0.0779 0.16577 0.47 0.6385 1.88050 5 XP Ownership x Geo graphic location 2.232 6 0.50934 4.38 <.0001 3.45155 6 XC Ownership x Z Cardiac Group Procedures 1.1402 0.25742 4.43 <.0001 2.90506 6 X R Ownership x Z Orthopedic Procedures Group 2.3385 0.25543 9.16 <.0001 1.85193 Coding for Variables: Ownership: Inves tor owned hospital = 1; Not for Profit = 0 Cardio Group Procedures : Z score transformed percent of patients per hospital treated under a DRG code identified as a cardiac group procedure in this study. Orthopedic Group Procedures : Z score transformed perce nt of patients per hospital treated under a DRG code identified as an orthopedic procedure in this study. Geographic location Table 2 5 shows that VIF is very low, less than the (conservative) rule of thumb of 5.0, indicating that the conversion to z scores for cardiac and orthopedic group procedures was effective in managing multicollineari ty. Table 2 5 shows a parameter estimate for ownership of 2.7 7 Recall that the beta for ownership in the main effects model, 1 A was 4.06. The smaller beta for ownership in the interaction model is due to a more thorough explanation of the relationship b etween the TATL ratio and ownership by the inclusion of variables that interact with ownership.

PAGE 137

123 Parameter estimates for z orthopedic group procedures and z cardiac group procedures are very low and not significant. But they still must remain in the model because they are constitutive terms in connection with interaction variables ownership x z orthopedic group procedures and ownership x z cardiac group procedures. The same is true for the variable geographic location Marginal effects for interacting va riables will be discussed in the following section. Interpreting Marginal Effects of Interaction Terms in Model 1 B Marginal Effects of the Interaction Term Orthopedic Group Procedures x Ownership Brambor, Clark & Golder ( 2006 ) have pointed out that many researchers establish interaction models but then do not go on to calculate the marginal effects of the interaction terms. To avoid this type of shortcoming in the investigation, marginal effects were calculated for the interaction terms in Model 1 B. I n examining the interaction terms, the models can be presented according to four conditions in connection with the interaction of orthopedic group procedures and ownership: R = 1 Z X = 1; R = 0 Z X = 0 ; R = 1 Z X = 0; R = 0 Z X = 1 (where R = orthopedic group procedures and X = ownership)

PAGE 138

124 Table 2 6 shows marginal effects of each condition. Table 2 6 Marginal Effects for the Interaction Term Orthopedic Group Procedures x Ownership Four Conditions Variable R =1 Z X=1 R =0 Z X=0 R =1 Z X=0 R =0 Z X=1 0 Intercept 2.037 2.037 2.037 2.037 1 X Ownership 2.766 2.766 2 P Geographic location 0.096 0.096 0.096 0.096 3 C Z Cardiac Group Procedures 0.000 0.000 0.000 0.000 4 R Z Orthopedic Group Procedures 0.07 79 0.078 5 XP Ownership x Geographic loc ation 2.233 2.233 6 XC Ownership x Z Cardiac Group Procedures 1.140 1.140 7 X R Ownership x Z Orthopedic Procedures Group 2.339 Totals 10.533 2.133 2.055 8.27 2 Where R = Z Orthopedic Group Procedures and X = Ownership The mar ginal effect of each condition is as follows: R = 1 Z X = 1 = 10.533; R = 0 Z X = 0 = 2.133; R = 1 Z X = 0 = 2.055; and R = 0 Z X = 1 = 8.272. An important difference is found between models where X = 1, R = 1 and X = 0, R = 0 which is shown below 0 1 2 P 3 C 4 5 P 6 C 7 0 2 P 3 C = 10.53 3 2.133 = 8.40 The true difference between the models where z orthopedic group procedures = 1Z and ownership = 1 vs. z orthopedic group procedures = 0Z and ownership = 0 amounts to 8.40 TATL ratio points. This is a very large difference between TATL ratios which is due chiefly to ownership.

PAGE 139

125 Marginal Effect of Orthopedic Group Procedures When Ownership = 1 To determine the contribution of orthopedic group procedures two other models are co mpared. This includes z orthopedic group procedures = 1Z and ownership = 1 vs. z orthopedic group procedures = 0Z and ownership = 1. The difference between these two models is: 0 1 2 P 3 C 4 5 P 6 C 7 0 + 1 + 2 P + 3 C + 5 P + 6 C = 10.53 8.272 = 2.261 The marginal effect of the interaction term for orthopedic group procedures and ownership is 2.26 1 TATL ratio points. The interaction term can also be wri tten 0 + 1 X + 4 R + 7 R X (where R = z orthopedic group procedures and X = ownership ) of w hich the marginal effect may be interpreted R = 4 7 X (i.e., 4 at X ) (Jaccard & Turrisi 2003 Aichen & West 1991 ). T he marginal effect may then be calcu lated 0.0779 + 2.33 9 (X) = 2.26 1 It should be understood that the focal variable in this case is orthopedic group procedures and the modifying variable is ownership. In other words, the marginal effect is determined by the association between the TAT L ratio and orthopedic group procedures at 1Z (one standard deviation above the mean of z = 0) when ownership is present (i.e., ownership = 1) (Jaccard & Turrisi 2003 ) So in general investor owned hospitals will have TAT L ratios that are 2.26 points high er if they perform conventional Medicare reimbursed orthopedic group procedures at 1Z above the mean.

PAGE 140

126 Standard Error & t Value for the Marginal Effect Recall from Table 25 that the parameter for interaction term X R is significant (p < 0.0001). But this d oes not necessarily mean that the marginal effect of the interaction is significant (Brambor Clark & Golder 2007 Jaccard & Turrisi 2003 ) To determine the significance of the marginal effect of orthopedic group procedures on the TATL ratio when ownership is present the following formula is employed. This formula shows the standard error for the association between the TATL ratio and orthopedic group procedures when ownership = 1 (i.e., 4 at X) (Aichen & West 1991) R = [ v 4 ) + X 2 v ar ( 7 ) + 2 X c ov ( 4 7 )] 1/2 The standard error formula contains variances for 4 and 7 and a covariance for ( 4 7 ) (Jaccard & Turrisi 2003) These variances and covariance are shown in Appendix 7 which provides a variance covariance matrix for Model 1 B. The appendix also shows the calculations in connection with the standard error formula. From this formula it is determined that the standard error for orthopedic group procedures when ownership = 1 is 0.1943. It had already been determined that the marginal effect ( 4 at X ) is 2.261. To determine the significance of the marginal effect a t value is computed t = ( 4 at X ) / SE ( 4 at X) = 2.261 / 0.1943 = 11.633. The t value is distributed at N k 1 degrees of freedom, where k is the number of predictor terms in the interactive model (in this case k = 7 so df = 891 7 1 = 883 ).

PAGE 141

127 The marginal effect of the interaction term is determined to be significant with t = 11.633 (p value < 0.0001) (Aiken West 1991, Jaccard & Turissi 2003 ). This indicates that invest or owned hospitals that perform orthopedic group procedures at 1Z above the mean have, on the average, TATL ratios that are 2.261 points higher than investor owned hospitals that perform orthopedic group procedures at the mean (z = 0). Marginal Effect s of the Interaction Term Cardiac Group Procedures x Ownership A second interaction term is now tested featuring z cardiac group procedures x ownership. Table 2 7 is constructed to show the marginal effects under four conditions, C = 1 Z X = 1, C = 0 Z X = 0, C = 1 Z X = 0, C = 0 Z X = 1 (where C = z cardiac group procedures and X = ownership). Table 2 7 Marginal Effects for the Interaction Term Cardiac Group Procedures x Ownership Four Conditions Variable C=1, X=1 Z C=0 Z X=0 C=1 Z X=0 C=0 Z X=1 0 Intercept 2.037 2.037 2.037 2.037 1 X Ownership 2.766 2.766 2 P Geographic location 0.096 0.096 0.096 0.096 3 C Z Cardiac Group Procedures 0.0004 0.000 4 4 R Z Orthopedic Group Procedures 0.078 0.078 0.078 0.078 5 XP Ownership x Geographic location 2.233 2.233 6 XC Ownership x Z Cardiac Group Procedures 1.140 7 X R Ownership x Z Orthopedic Procedures Group 2.339 2.339 Totals 10.53 3 4 2.05 50 2.055 4 9.39 3 Where C = Z Cardiac Group Procedures and X = Ownership Marginal effects for each condition include: C = 1Z, X = 1 = 10.53 3 ; C = 0 Z X = 0 = 2.05 50 ; C = 1Z, X = 0 = 2.055 4 ; and C = 0 Z X = 1 = 9.39 3

PAGE 142

128 The true difference between models where X = 1, C = 1 Z and X = 0, C = 0 Z is shown as follows. 0 1 2 P 3 + 4 R + 5 P 6 7 R 0 2 4 R = 10.53 3 4 2.05 50 = 8.4 78 0 The difference between the models, where z cardiac group procedures = 1Z and ownership = 1 vs. z cardiac group procedures = 0Z and ownership = 0 amounts to 8.478 0 TATL ratio points. Marginal Effect of C ardiac Group Procedures When Ownership = 1 To determine the contribution of cardiac group procedures a comparison is made between the models where cardiac group procedures = 1Z and ownership = 1 vs. cardia c group procedures = 0Z and ownership = 1. The dif ference between these two models is: 0 1 2 P 3 4 R 5 P 6 7 R 0 1 2 P 4 5 P 7 R = 10.53 3 4 9.39 3 0 = 1.140 4 In this case, the focal variable is cardiac group procedures, so the marginal effect will be determined by the association between the TATL ratio and cardiac group procedures at 1Z (one standard deviation above the mean, z = 0) when ownership is present (i.e., ownership = 1). The interaction term is then written 0 3 C 1 6 CX (where C = cardiac group procedures and X = ownership) of wh ich the marginal effect is interpreted C = 3 6 X. The marginal effect may then be calculated .0004 + 1.40(X) = 1. 1 4 04 This indicates that on the average investor owned hospitals that

PAGE 143

129 perform cardiac group procedures at 1Z have TATL ratios that are 1. 1 4 04 points higher than other investor owned hospitals that perform cardiac group procedures at the mean (Z = 0) For the marginal effect for cardiac group procedures when ownership = 1 ( 3 at X ) to be meaningful a test of significance must be perfor med, just as it was previously for orthopedic group procedures when ownership = 1. This first requires the calculation of a standard error for the marginal effect of which the formula is written below. C = [ v 3 ) + X 2 v ar ( 6 ) + 2Xc ov( 3 6 )] 1/2 As a lready mentioned Appendix 7 provides the variance covariance matrix for Model 1 B and shows marginal effects and standard error calculations for From this formula it is demonstrated that the standard error for cardiac group procedures when ownersh ip = 1 is 0.1 541 The marginal effect for the interaction between cardiac group procedures and ownership is 1.140 4 To determine the significance of the marginal effect ( 3 at X = 1) a t value distributed with N k 1 (883) degrees of freedom is comput ed t = ( 3 at X ) / SE ( 3 at X) = 1.140 4 / 0.1 541 = 7.4003 (p < 0.0001) The marginal effect of the interaction term for cardiac group procedures and ownership is found to be significant with t = 7.4003 (p < 0.0001). This indicates that investor owned h ospitals that perform cardiac group procedures at 1Z above the mean have, on the average, TATL ratios that are 1.140 4 points higher than investor owned hospitals that perform cardiac group procedures at the mean level (z = 0).

PAGE 144

130 Discussion of Findings for Pa rt 1 Regression is a technique used for observatory studies so causation cannot be established. But positive association was demonstrated between TATL ratios and rates of cardiac group and orthopedic group procedures. A positive association was also fo und for the interaction between cardiac group procedures and ownership as well as the interaction for orthopedic group procedures and ownership. The model showed that the TATL r atio was, on the average, 1.1404 points higher when cardiac group procedures w ere administered at 1Z above the mean (Z = 0) in investor owned hospitals. The model also showed that the TATL ratio was on the average 2.26 1 points higher when orthopedic group procedures were administered at a rate of 1Z higher than the mean (Z = 0) amo ng investor owned hospitals. Testing for Unnecessary & Marginally Necessary Procedures unnecessary heart procedures. One of the objectives of this investigation was to determine if a portion of discretionary procedures performed by profitable investor owned hospitals are unnecessary or marginally necessary However evidence was not uncovered to indicate that procedures were being performed that were medically unwarranted. Invest or owned hospitals with strong balance sheets (i.e., high TATL ratios) perform cardiac group and orthopedic group procedures at higher rates per hospital (not higher volumes) than not for profits or unprofitable investor owned hospitals. But i t was found in the preliminary analysis that less profitable not for profit hospitals carry out by far the highest volumes of these procedures.

PAGE 145

131 To make inferences about the medical necessity of these procedures it would have to be demonstrated that profitable investor owned hospitals perform volumes of cardiac group and orthopedic group procedures at levels that exceed the medical needs of the population. This inference cannot occur if volumes of procedures performed by investor owned hospitals do not exceed those of other types of hospitals. Gaming or Good Management Orthopedic and cardiac group procedures generally represent small percentages per hospital relative to the number of total procedures performed. Higher TATL ratios may be associated to higher percentage s of cardiac and orthopedic procedures, but it is also likely that other factors contribute to these high ratios. But a significantly positive association has also been demonstrated between TATL ratios and the interacting variables of ownership and orthop edic group procedures and ownership and cardiac group procedures. It is then likely that the performance of these procedures contribute to long term profitability. Certain examples in the literature review may point to h igher percentages of certain types of discretionary procedures being an example of gaming Medicare, emphasizing certain profitable D R G codes in the patient mix while avoiding others. Keep in mind that investor owned hospitals tend to have fewer beds than not for profits resulting in fewer total discharges for inpatient care. This explains why certain investor owned hospitals that emphasize these types of discretionary procedures have higher rates of discretionary procedures without generating higher volumes. But does this mean that many investor owned hospitals attempt to fill their beds with patients

PAGE 146

132 being treated with procedures represented by lucrative DRG codes, while less profitable patients are shifted to not for profits ? The literature review also cited many researchers who conten d that good management explains why in general investor owned hospitals are financially stronger than not for profits. An emphasis on certain types of discretionary DRG codes may just reflect another example of good management practice. What can be render ed from this analysis is that investor owned hospitals with high TATL ratios tend to offer packages of discretionary cardiac and orthopedic services that are attractive to specialists and their patients. Part 1 Regression Diagnostics Bagley White & Golom b (2001) have issued criticisms o n numerous (peer reviewed) studies in the literature, some of which included no regression diagnostics or validation analysis, no testing for multicollinearity, and no testing for the needed inclusion of higher order terms Multicollinearity and specification errors ( testing to include higher order terms) have already been discussed in connection with Model 1 B In addition to these diagnostics other potential regression v iolations were also tested and included outliers a nd influential observations other aspects of specification errors, and h eteroscedasticity In model development multicollinearity was assessed and resulted in the transformation of observations to z scores for cardiac group procedures and orthopedic gr oup procedures. Problems with multicollinearity have been partly addressed via the inclusion of VIF with each model. In addition both adjusted and non adjusted intercept

PAGE 147

133 collinearity diagnostics were carried out. The results of collinearity diagnostics in connection with Model 1 B using an intercept adjustment are shown in Appendix 8 Outliers Press Statistic A good summary measure available in testing outliers and influential observations is the predicted residual sum of squares (PRESS). The PRESS s um of squares is the sum of squares of residuals using models obtained by estimating the equation with all other observations, i.e., ( y i i 1 ) (Freund & Little 2000) Table 2 8 provides information regarding the PRESS statistic. T able 2 8 Sum of Residuals Sum of Squared Residuals & PRESS for Model 1 B Sum of Residuals 0 .0 Sum of Squared Residuals 11,483 Predicted Residual SS (PRESS) 11,703 In diagnosing residuals using PRESS, t he sum of residuals should equal zero which it does in Table 2 8 The sum of squared residuals (SSR) should be the same as the Error SS in the regression output for Model 1 B Recall that the Error SS for Model 1 B is 11,483 which is equal to the su m of squared residuals in Table 2 8 The PRESS statistic should also be compa red to the SSR When the PRESS statistic is appreciably larger than the SSR there is rationale to suspect that some influential observations and outliers exist. For Model 1 B the PRESS statistic is only slightly larger than the SS R so it can be conclud ed that this test does not indicate there are observations that unduly influen ce model results Univariate statistics including

PAGE 148

134 normal probability plots box plots, and stem leaf plots were also reviewed for potential anomalies and none were observed. Stu Additional t esting for outliers beg a n with a review of studentized residuals and d statistic. Careful scrutiny was applied to observations that exceed residual test thresholds. These include studentized re siduals 2 .5 .5 [ or 5 asterisks which equals | 2 .5 | 4/891) ] Overall there were only around 25 observations with residuals that exceeded either of these thresholds. There were also a few studenized residuals in close to pr oximity to 2 .0 or 2 0. Influence Statistics Influence statistics are helpful because o utliers are not always readily detectable by an examination of residuals alone In some cases the least squares estimation procedure may pull the estimated regression response toward observations that have extreme values in either x or y dimensions (Freund, Littell, 2000). I nfluence statistics were carried out including hat diag, covariance ratio, DFFITS, and DFBETAS Only a very small number of observations exceeded prescribed thresholds e.g. for DFFITS 2 [(m + 1)/n] 1/2. Specification Errors A specification error occurs when a specified model does not contain all of the essential parameters. This might occur when only linear terms are specifi ed in the model, but actual associations are curvilinear. The analysis of residuals and particularly partial residual s indicated no need to include one or mo re squared terms in the model or any additional interaction terms.

PAGE 149

135 There are, however, some conce rns regarding another type of specification error in a model that does not includ e all relevant main effects variables. One of the research objectives in this project was to form parsimonious models. Although this is a worthwhile objective rigid adherenc e to parsimony should not occur at the expense of excluding important information. To be assured that a specification error did not occur in Part 1 a technical appendix is added, Appendix 9 where regression models are set up to include the variables use d in Part 1 plus three additional variables which include hospital occupancy, labor costs represented by manhours per adjusted patient day, and market concentration measured by the Herfindahl Hirshman Index (HHI). In the concluding remarks of A ppendix 9 it is explain ed that model results, parameter estimates and marginal effects via interaction terms, are very similar in the expanded models shown in the technical appendix relative to the parsimonious models presented in the body of this document for Part 1. This enhances con fidence in model validity and in the findings derived from Part 1 Heterogeneous Variances An underlying regression assumption is that all random errors have the same variances. A review of residuals indicated no systematic residua l patterns, such as nonrandom residual distributions, recognizable patterns in the magnitudes of the variances, or increases or decreases in variation of larger values of the outcome variable (Freund, Littell, 2000). Weaknesses of Part 1 Models All resea rch projects have inherent weaknesses, and this study is no different. These weaknesses are described in the following paragraphs.

PAGE 150

136 Randomly Splitting the Data to Assure Model Validity Many researchers will enhance confidence in the validity of their find ings by first randomly splitting a certain portion of the data, then building the regression model on one split data set and then applying the prediction equation to the other (Sincich, 2005). Splitting the data was not performed in this analysis chiefly because the size of the project with two parts 1 and 2 was getting out of hand. Additional activities would expand the project even further. An examination using split data might be beneficial in a follow up replication study. Other methods such as a thor ough review of residual analysis, particular ly the use of studentized ( jackknife ) residuals, and a strong knowledge of the data using extensive preliminary analyses were found to also be good methods for evaluating model validity. Grouping DRG Codes The mo dels used in Part 1 featured two variables, cardiac and orthopedic group procedures that were comprised of grouped DRG codes. This was a strong point of the investigation in that it allowed several DRG codes to be tested at one time in the models. It also avoided further problems with multicollinearity since many medical procedures, such as CABG and valve replacements, were shown to be highly correlated. But this was also a weakness because findings would have been more precise had only one DRG code or at least one surgical procedure been tested at a time Procedure Volumes There was also a problem of different sized volumes of certain procedures within cardiac group and orthopedic group procedures Back in the preliminary analysis for Part 1 tables 7 & 8 provided numbers of observations for each procedure comprising

PAGE 151

137 cardiac group and orthopedic group procedures. Of the 185,603 discharges for cardiac group procedures, v alve replacements and CABG procedures showed 20,551 and 50,158 cases while PCI showed 114 894 Findings using the cardiac group procedures variable would be weighted more heavily by PCI. There was a similar problem with orthopedic group procedures of which the total number of 222 525 observations was comprised in part by 136,121 major join t and limb reattachments (DRG 209). A mitigating factor is that the larger PCI and major joint and limb reattachment volumes are more diversely distributed in their performance across types of hospital ownership. This means that bias generated by their in clusion and preponderance in cardiac and orthopedic group procedure variables would be toward the null hypothesis. County Median Income as a Proxy for Geographic location The proxy for geographic location in this study was Florida county median income leve ls. Although this proxy served well in the models, it may have been advantageous to have used a more refined and detailed proxy (if one had been available). Median income levels by county planning areas (i.e., contiguous census tracts within a county agg regated by natural geographic boundaries or community characteristics ) may have further elucidated their association with the TATL ratio. Hospital locations within affluent planning areas may h ave generated more precise observations than hospital locations in higher income counties.

PAGE 152

138 Rigid Adherence to Parsimony In the discussion on specification error under the subsection, Regression Diagnostics, there is a reference to Appendix 9 a technical appendix which presents models using the same variables in Part 1, with three additional variables to reflect patient occupancy, labor costs, and market concentration. It is possible that a rigid adherence to parsimon y in Part 1 may have led to creating models that ignored variables that were both relevant to the dependent variable and valuable to the analysis. If a specification error has occurred the technical appendix represents an attempt to correct it. The reader is cordially invited to review par simonious models in Part 1 in the body of the document as well as the expanded models in Appendix 9

PAGE 153

139 Chapter 8 Model Results for Part 2 Testing the Association between Total Charges & Ownership for CABG & Valve R eplacement Procedures (Reimbursed by Medicare HMO ) Model 2 A : Main Effects Model Now that hypotheses have been stated based on findings in the preliminary analyses which helped identify and test outcome explanatory and control variables in Chapter 6 the investigation can proceed to the first main effects model in Part 2 The dependent variable in this model is total patient charges which is a continuous variable, so a n MLS regression equation is used (Mendenhall & Sincich 2003). The MLS equation for the main effects model in 2 A i ncludes four terms three of which includ e ownership, LOS, and s outh geographic region An additional term include s valve procedure which is a term used to adjust for differences in charges between valve replacements and CABG procedures It is c oded val ve procedure = 1 and CABG ( as the base level ) = 0

PAGE 154

140 The MLS equation for the main effects mod el, 2 A, is written as follows: R egression Equation for Model 2 A y Charges = 0 1 X 2 S 3 L + 4 V Where: X = Ownership S = South g eographic sector L = LOS V = Valve procedure The regression procedure was undertaken using the prediction equation for Model 2 A. F indings for the main effects model are shown in Table 2 9 Table 2 9 Model 2 A: Main Effects Model Source DF Sum of Squares Mean Square F Value Pr>F Model 4.0 2.87E + 13 7.17E + 12 4,101.02 <.0001 Error 12,415 2.17E + 13 1.748E + 09 Corrected Total 12,419 5.04E + 13 Root MSE 41,813 Dependent Mea n 102,488 R 2 0.5692 CV 40.80 Adjusted R 2 0.5691 As shown in Table 2 9 mean total charges (i.e., dependent mean) for all cases in the dataset is $10 2 488 The model shows an adjusted R 2 of 0. 5691 reflecting goodness of fit where close to 57 % of the v ariance is explained by the model. The model shows a CV of 40 80 As discussed previously the ideal for the CV is 10.0 (Sincich 2005). The CV of 40.80 in connection with Model 2 A is greater than the ideal of 10.0, but is still acceptable for real world data.

PAGE 155

141 P arameter estimates for Model 2 A are shown in Table 30 Table 30 Model 2 A: Main Effects Model Parameter Estimates Parameters Parameter Estimate Standard Error t Value P > | t | VIF Intercept 9,585 835.24 11.48 <0.0001 0 Ownership 43,351 810.63 53.48 <0.0001 1.00639 South 19,714 754.94 26.11 <0.0001 1.00629 LOS 6,219 59.3 7 104.75 <0.0001 1.01274 Valve 21,174 840.75 25.18 <0.0001 1.01399 Coding for Variables: Ownership: Investor owned hospital = 1, Not for Profit = 0; LOS: Continuous variable (days from admission to discharge); Geographic Region: South = 1, North & Central = 0; Valve Procedure: Valve replacement = 1, CABG procedure = 0 Model 2 A shows the parameter estimate for ownership at $ 4 3 351 indicating that on the average, investor owned hospitals assess total charges $ 4 3 351 higher than not for profits. Total charges for each additional one day in the hospital are $6, 2 19 as shown by the parameter estimate for LOS. Geographic reg ion shows a beta of 19 714 for the south Florida geographic sector indicating that in general charges in the south sector are $ 19 714 greater than the base level north ern & central Florida. The dummy variable valve procedure yields a beta of 21, 174 repre senting that charges for valve replacements in general are $21, 174 more than the base level CABG procedures. M odel 2 B : Interaction Model The main effects model demonstrated that total patient charges are significantly higher for investor owned hospitals relative to not for profits and for hospitals located in the southern region of Florida relative to the central and northern regions. Model 2 B is an interaction model that contains only one interaction term ownership x south geographic region. The nece ssity for including a quadratic term in

PAGE 156

142 connection with LOS, the only continuous independent variable, was explored and was found unwarranted. The regression equation for Model 2 B is written below Regression Equation for Model 2 B Interaction Model y Charges 0 1 X 2 S 3 L 4 XS 5 Where: X = Ownership S = South geographic region L = LOS XS = Ownership x south geographic region V = Valve procedure Findings were derived based on the interaction model 2 B and are presented below in Tabl e 3 1 Table 3 1 Model 2 B : Interaction Model Source DF Sum of Squares Mean Square F Value Pr>F Model 5 2.87E + 13 5.75E + 12 3,293.43 <0.0001 Error 12,414 2.17E + 13 1.745E + 09 Corrected Total 12,419 5.04E + 13 Root MSE 41,768 Dependent Mean 102,488 R 2 0.5702 CV 40.75 Adjusted R 2 0.5700 Model 2 B shows an adjusted R 2 of 0. 5700 indicating that roughly 57 % of the variance in the data can be explained by the model. The R 2 and CV ( 40.75 ) are close to the R 2 and CV in the main effects mode l 2 A Before proceeding with an examination of parameter estimates an F test is carried out to compare the interaction model 2 B with the main effects model 2 A The test

PAGE 157

143 yielded F = 17.42 (p < 0.0001) for R 1 2 = 0.5 691 and R 2 2 = 0.5700 with df 1 = k 2 k 1 = (5 4) = 1 and df 2 = N k 2 1 ( 12,420 5 1 ) = 12,41 4 The F test is sufficiently large enough to warrant using the interaction model, but it should be mentioned that the distance between the R 2 in these two models is very close, and a signific ant F value was yielded only because of the large df 2 which was due to a large n umber of observations (N = 12,420) used in models 2 A and 2 B. Even though the R 2 in each model is close, t he interaction model still generates very useful results. It generat es findings that help to address certain research questions and issues that had been initiated in the p roblem s tatement and literature r eview Therefore, reporting the interaction model is consistent with the objectives of this study. Findings of the inte raction model, 2 B are shown in Table 3 2 Table 3 2 Model 2 B : Interaction Model Parameter Estimates Parameters Parameter Estimate Standard Error t Value P > | t | VIF Intercept 8,478 860 9.85 <.0001 0 Ownership 47,505 1,130 42.03 <.0001 1.9611 South Sector 22,429 913 24.56 <.0001 1.4762 LOS 6,214 59 104.76 <.0001 1.0130 Ownership x South Sector 8,529 1,619 5.27 <.0001 2.5127 Valve 21,046 840 25.05 <.0001 1.0148 As shown in Model 2 B, t he re is a parameter estimate for ownership of $ 47 505 Rec A was slightly less, $43,351. What accounts for the larger beta in Model 2 B is the inclusion of the interaction term, ownership x south geographic sector. What is observed by the interaction term is that the distanc e in mean patient charges narrows between investor owned hospitals and not for profits in the south region This indicates though investor owned hospitals generally assess higher charges than not

PAGE 158

144 for profits in all regions, not for profits in the south re gion charge higher than not for profits in the north and central regions This finding is also consistent with the preliminary analysis shown in Figure 7 which showed mean charges stratified by ownership and location. This chart showed a narrowing of the distance of charges between investor owned and not for profits in the south region. Days of hospitalization are almost identical in the interaction model as the main effects model. This indicates that differences in charges with respect to hospital owner ship and geographic region are not due to disease severity, as operationalized by LOS. Interpreting Marginal Effects of Interaction Terms in Model 2 B The interaction model for ownership x south geographic region is expressed as y Charges 0 1 X 2 S 4 XS In this case, the focal variable is ownership and the modifying variable is south region. The 1 at S = 1 + 4 S ) = 47,505 + ( 8,529) S = 38,976 This indicates that in general investor owned hospitals, locate d in the south region, assess charges at $38,976 higher than not for profits located in the south region (Jaccard & Turisi 2003) A standard error and t value must be computed to determine if the marginal effect of ownership and south region is significa nt. The standard error calculation was performed using the formula shown below. The formula (Jaccard & Turisi 2003) incorporates variances for 1 and 4 and the covariance for ( 1 4 ) which is derived from the variance covariance matrix shown in Appendix 10 The appendix also shows the calculations using this formula.

PAGE 159

145 1 ) + S 2 4 1 4 )] 1/2 a standard error of 1,160. To determine the significance of the marginal effect a t value is computed as follows: t = ( 1 at S) / SE ( 1 at S ) = 38,976. / 1,160 = 33.60 (p < 0.0001) The t value distributed at N k 1 (12,420 5 1 = 12,414) degr ees of freedom, is very high and the marginal effect is significant at p < 0.0001. From this model it can be concluded that investor owned hospitals assess significantly higher charges under the Medicare HMO program for CABG and valve replacement proced ures. The interaction term ownership x south region indicates that although investor owned hospitals assess higher charges in the south region, the distance is narrowed in amounts between investor own ed and not for profit hospitals. Not for profits are s hown to accelerate charges at a higher rate (not monetary value) than investor owned hospitals in the south region Discussion of Findings for Part 2 It is clear from models 2 A and 2 B that investor owned hospitals assess significantly higher charges tha n not for profits in the performance of CABG and valve replacement procedures reimbursed by Medicare HMO. It is also unambiguous that investor owned and not for profit hospitals, located in the southern Florida region, assess higher charges for the same s ervices relative to their hospital counterparts in central and north Florida. In fact, though investor owned hospitals assess higher charges in all geographic regions, the distance between amounts of charges narrows between investor owned and not for prof it hospitals in the south region.

PAGE 160

146 It has also been established that higher charges in connection with ownership or geographic region are not due to increased days of hospitalization or the administration of more intensive medical services. Total Charges vs Reimbursed Charges A comment was made previously in connection with Table 17 that average total charges for valve replacement and CABG procedure s are excessively higher than what is actually reimbursed by Medicare HMO. The AHCA financial data (electron ic file) provides financial data for hospitals according to each payment source. Amounts are listed for inpatient and outpatient services in terms of total revenue, revenue deductions, and net revenue. Percentages were calculated for Medicare HMO inpatie nt net revenue [ Medicare HMO net inpatient revenue percentage = ( Medicare HMO gross inpatient revenue deductions from Medicare HMO gross inpatient revenue) / Medicare HMO gross inpatient revenue] I t was found that in general not for profit hospitals rea lize only around 30% of their Medicare HMO gross revenue while i nvestor owned hospitals realize, on the average, only 24%. These percentages of net revenue were calculated not for all hospitals, only the investor owned and not for profit hospitals represe nted in the study population for Part 2. In other words, t hese included only hospitals that performed valve replacement or CABG procedures reimbursed by Medicare HMO Because total patient charges are much higher in investor owned hospitals it would be ex pected that percentages of reimbursements under Medicare HMO would be

PAGE 161

147 lower than that of not for profits. As already mentioned, reimbursements are, in fact, lower for investor owned hospitals. Likewise, b ecause charges increase for both investor owned and not for profit hospitals in the southern region of Florida, it would be anticipated that net revenue percentages should decrease relative to those in the north and central regions. But this does not occur, as shown in Table 3 3 Table 3 3 Percentage of Net Revenue for Inpatient Services Investor Owned vs. Not for Profit Hospitals in the South Region vs Combined North & Central Region s Medicare HMO Inpatient Services 2000 2005 Region Investor Owned Not for Profit South 23.88% 29.86% North & Central 24.81% 30.84% Source: AHCA (2000 2005) Financial Analysis Annual Electronic Reports The table shows that net revenue, on the average, for investor owned hospitals, located in the southern region, is only about one percentage point (23.88% vs. 24.81%) lower than net revenue in the north and central regions. Similarly there is only about one percentage point (29.86% vs. 30.84%) difference for the south vs. north & central regions for not for profit hospitals. One important question arises regarding the differential between total patient charges and reimbursements. Data to calculate p ercentages of net revenue are provided for Medicare HMO inpatient services, not specifically for valve replacements and CABG procedures. So it is not possible to determine if the average 30% and 24% levels of Medicare HMO reimbursement for not for profits and investor owned hospitals, respectively, apply specifically to valve replacements and CABG procedures.

PAGE 162

148 However, if these percentages are accurate it would mean that r eimbursements are higher in the south region irrespective of disease severity or the extent of medical services provided. It would also mean that reimbursements under the Medicare HMO program occur with wide variability with little attention to anomalies in f ederal spending. It has been stated that one of the goals of Part 2 (Chapter 6) was to test As discussed this theory entails geographic disparities in per capita rates and f ederal expenditures in connect ion with discretionary procedures. The findings in Part 2 add credence to Part 2 Regression Diagnostics The same set of tools used for regression diagnostics in Part 1 ( Chapter 7 ) are employed again in Part 2 Diag nostic results provided in this section are in connection with Model 2 B Outliers Part 2 begins its presentation on diagnostics with a discussion on potential outliers. The PRESS statistic is used again as a summary measure. As stated in Part 1, PRESS is the sum of squares of residuals using models obtained by estimating the equation with all other observations; that is ( y i yhat i 1 ) (Freund Littell 2000). Information on PRESS is shown in Table 3 4 below. Table 3 4 Sum of Residuals, Sum of Squared Res iduals & PRESS for Model 2 B Sum of Residuals 0.0 Sum of Squared Residuals 2 1, 657 ,2 8 0,000,000 Predicted Residual SS (PRESS) 21 692 210 ,000,000

PAGE 163

149 The sum of residuals in Table 3 4 should equal zero which it does The sum of squared residuals should be t he same as the Error SS in the regression output for Model 2 B The Error SS under Model 2 B was 21,657,280,000,000 ( 2.17 E+13 ) which equals the sum of square d residuals shown in the t able. The PRESS statistic is now compared with the sum of squared res iduals as i nfluential observations and outliers would be indic a ted if PRESS is appreciably larger But as shown in the t able PRESS is only slightly larger than the sum of squared residuals with a miniscule difference possibly even due to rounding. It can be concluded that information derived from Table 3 4 does not indicate the presence of influential observations or outliers. Studentized residuals and Cook s d were also used to perform additional tests for o utliers. O bservations in connection with Model 2 B were reviewed that exceeded prescribed thresholds | 2.0 | N (i.e., 4/12, 414 ) (Freund, Littell 2000). Influence Statistics I nfluence statistics including hat diag, covariance ratio, DFFITS and DFBETAS were used to identify and examine observations tha t exceeded prescribed thresholds. A small number of observations were identified as potentially influential, but overall no major problems regarding influential observations or outliers occurred.

PAGE 164

150 Specification Errors As mentioned in the regression dia gnostic discussion under Part 1, a specification error occurs when a model lacks all essential parameters which may occur when the true association is curvilinear, and the model includes only linear terms. Residual analysis and in particular partial resi duals were very helpful in determin ing if models contained all necessary parameters. The incorporation of quadratic terms in model s for Part 2 w as tested and their inclusion in the final model was determined to be unwarranted. Heterogeneous Variables An assessment was done to determine if random errors have equal variances. Charts were analyzed to detect non random residual patterns, obvious patterns in the magnitudes of the variances, and increases or decreases of larger values of the outcome variable ( Freund, Littell 2000). Overall, no major violations of heterogeneity were observed. Multicollinearity Appendix 1 1 presents an annotated assessment of multicollinearity in connection with Model 2 B Problems with multicollinearty have been discussed, a nd VIF has been presented with each model. In addition, b oth adjusted and non adjusted intercept collinearity diagnostics were performed for each model. The collinearity diagnostics presented in this appendix is intercept adjusted in connection with Mode l 2 B As discussed under Part 1 of the study, the rule of thumb in reducing problems of VIF was less than 3.0 in connection with all parameter estimates for models 2 A & 2 B. Overa ll, n o

PAGE 165

151 major problems with multicollinearity were detected for the main effects model 2 A or interaction model 2 B. Weaknesses of Models in Part 2 As with Part 1, Part 2 also had certain weaknesses in analysis and model development. These are described in the following paragraphs. Assessing Validity by Randomly Splitting the Data Just as it was discussed in Part 1 (Chapter 7) a good strategy to evaluate model validity is to randomly split the data, develop the model for one portion of the data and then apply the prediction equation to the other portion. As it was mentioned under Part 1, t his was not done because of the magnitude of this project. Had this project been only one part instead of two then random splitting of the data would have been done. O ther methods used to evaluate the validity of the model included a thorough review of regression diagnostics and residual analysis as well as a substantial review of the data preceding model development via extensive preliminary bivariate and nonparametric analyses. Total Patient Charges vs. Reimbursements It has been discussed that actual amounts reimbursed for each patient was not reported in the AHCA Discharge dataset only total charges. Reimbursements are only known more generally in connection with Medicare HMO inpatient services shown in the AHCA Financial (electronic) data Meaningful findings were still derived in Part 2 especially considering wide disparities in levels of total patient charges with respect to ownership and geographic sector. But more specific information about reimbursements at the patient level would have been helpful.

PAGE 166

152 Chapter 9 Conclu s ion It should be asseverated that this study does not take the position that there is anything wrong with solvency or profitability. Fina ncial strength in hospitals saves lives and aids in offering better quality of care to patients. There are many previous studies with compelling findings regarding the association between healthcare quality and hospital profi tability as well as the associ ation with poor quality and low profitability. T he Agency for Healthcare Research and Quality (2005) examined medical errors in the files of approximately one million patients who had undergone surgery in 176 Florida hospitals between 1996 and 2000. Th e study concluded that a decrease in a before, during, and after surgery. It can be concluded that financially healthy hospitals make valuable contributions to medi making a profit with Medicare, as long as it i s done within legal and ethical parameters. W hat is perplexing is that a s discussed in the l iterature r eview in Chapter 2 there is a consensus a mong researchers that Medicare is not a profitable revenue source for hospitals. Yet a number of profitable short term general hospitals in Florida show Medicare to comprise a large, even majority, percentage of their revenue base.

PAGE 167

15 3 This study has attemp ted to shed light on two strategies certain hospitals use to make a profit with Medicare. It was found in Part 1 of this study that certain investor owned hospitals perform significantly higher rates per hospital of certain types of discretionary cardiac and orthopedic group procedures, and these higher rates are positively associated with TATL ratios representing long term financial health Perhaps these investor owned hospitals are gaming conventiona l Medicare by emphasizing DRG codes that represent ce rtain discretionary procedures. But there are some researchers who contend that the types of procedures performed in hospitals are irrelevant to solvency and profitability. C opious citations in the literature review point to the association between fina ncial strength and good management practices. The most notable of these researchers, Cleverly assets that better financial health in investor owned hospitals is a product exclusively of good management, and that it is entirely irrespective of patient mix or payer sources. But Part 1 of this investigation demonstrated a positive association between the TATL ratio and discretionary cardiac and orthopedic procedures. So perhaps in addition to good management, financial health is also due to systematic tar geting of the most profitable DRG codes. It could also be argued that an emphasis on the most lucrative medical procedures is in fact good management. Perhaps certain hospitals merely demonstrate the acumen to structure their patient mix to maximize volu mes of DRG codes that are the most profitable under conventional Medicare. In Part 2 it was found that investor owned hospitals assess significantly higher total patient charges than not for profits for CABG and valve replacement procedures

PAGE 168

154 under the Med icare HMO program. Higher charges could not be explained by patient disease severity or the provision of more extensive medical services. On the flipside it was also found that the investor owned hospitals that assess significantly higher charges also ha ve appreciably higher percentages of deductions from total revenue. Although the deductions from revenue attenuate the argument that investor owned hospitals are overcharging, it does not address the question of why these charges are assessed so high in t he first place. It also does not address why total patient charges accelerate for both investor owned and not for profit hospitals in the southern region of Florida. Evidence was not uncovered in Part 1 which might indicate that investor owned hospitals are performing procedures unnecessarily or at rates that exceed the medical needs of the Florida population. Generating research findings that indicate unnecessary or marginally necessary surgical procedures are being performed in investor owned hospitals is an ambitious objective, one that would likely require many investigations But the crux of this investigation was less ambitious with the aim of identifying a couple of strate gies hospitals, chiefly investor owned, employ to make a profit with Medicare. Certain types of discretionary procedures under conventional Medicare increase per capita with respect to hospital ownership Likewise, investor owned hospitals and all hospit als located in the southern region of Florida assess excessive charges for medical services under the Medicare HMO program irrespective of patient disease severity and days of hospitalization

PAGE 169

155 (2005) surgical signature theory was described in the literature review and referenced several times in this document. In the context of discretionary procedures, this theory refers to fluctuating levels of per capita rates and f ederal s pending that are not clearly linked to the disease status of the po pulation The findings in this investigation Policy Recommendations This study would not be complete without some recommendations that may help to improve health policy. This final section discusses s ome preemptive methods that could be undertaken to uncover potential gaming activities by routine monitoring and subsequent identification of inordinate rates of certain discretionary procedures. A Few Caveats on Interpretation It should be taken into acc ount that this project is an observational study using secondary data. Associations can be drawn, but causation cannot be established. It should also be considered that this study is a pioneering effort, meaning there are very few prior studies dealing w ith gaming in this context increasing rates of procedures and inflating charges. Most studies in the literature address gaming in the context of up coding. It should also be recalled that no inferences could be demonstrated in this study to indicate tha t any unnecessary cardiac or orthopedic procedures occurred. What was demonstrated was that a positive association exists between the TATL ratio and cardiac and orthopedic group procedures exclusively in investor owned hospitals. It is possible that this association is indicative of gaming, but it is also feasible that the association has occurred due to very good management in certain investor owned hospitals.

PAGE 170

156 Administrators in these hospitals have perhaps structured their patient mix to maximize DRG co des that tend to be profitable, while minimizing those that are not. This reflects managerial acumen, and there is no legal or ethical violation unless it could be demonstrated that surgeries are being performed that are not clinically warranted. There was strong evidence that investor owned hospitals assessed higher charges than not for profits under the Medicare HMO program, and that both investor owned hospitals and not for profits assessed higher patient charges in the southern sector of Florida than the other sectors. These findings raised very good questions, but it could not be shown definitively if higher charges were resulting in higher and unwarranted reimbursements. Recommendations to Federal Agencies Now that these caveats, discussed in the p revious section of this chapter have been taken into account, there are some practical recommendations resulting from this research. Recommendations should not only be offered to hospital administrators and s Office of the Inspector General (OIG) and personnel and officials involved in regulatory and oversight functions of conventional Medicare and Medicare HMO programs. Recommendations to Medicare would include routine monitoring of rates of certain discret ionary procedures, the objective of which would be to identify anomalous fluctuations. Gaming, if it occurs, may be elucidated, especially on a large scale, by examining per capita rates in certain geographic areas that unexplainably exceed those of other areas, or that show sudden inordinate increases. Hospitals or hospital referral

PAGE 171

157 regions (HHRs) demonstrating appreciably higher rates than other hospitals or regions would be the focus of further investigation. Anomalies may be found not only in rates, but also in patient charges for discretionary procedures. In Part 2 of the narrative, figures 5 7 provide patient charges for the south, central, and north geographic sectors of Florida. These charts in themselves shed light on a geographic area of inte rest the south sector of Florida. Medicare should begin examining mean patient charges and reimbursement levels by geographic area to determine if reimbursement amounts are warranted. A Brief Example of Ongoing Monitoring of Rates of Discretionary Pro cedures Reimbursed by Medicare A brief hypothetical demonstration of how Medicare could monitor rates of discretionary procedures using real world data is shown in Figure 8 This figure provides CABG rates per 1,000 Medicare enrollees by Florida HRR, 200 5. Each bar on the chart is accompanied by a vertical error bar which ranges from one standard deviation below the mean to one standard deviation above.

PAGE 172

158 Figure 8 Source: Dartmouth Atlas of Healthcare, 2005 The CABG rate for the Panama City HRR at 5. 99, is the highest on the chart and also exceeds the upper limit (one standard deviation) of the error bar. Perhaps further review of Panama City is warranted, but it would not be the most judicious use of investigative resources because Panama City only has 27,000 Medicare enrollees (Dartmouth 2005). Therefore, priority would be given to another HRR with high CABG rates and with high numbers of enrollees. Figure 8 shows that the Orlando HRR has a CABG rate of 5.72, and the Jacksonville HRR shows a rate o f 5. 6 4. As it turns out, these two HRRs have Medicare enrollments, respectively, of over 400,000 and almost 150,000 (Dartmouth 2005). Because the Orlando and Jacksonville HRRs have both high numbers of Medicare enrollees and high CABG rates they would ha review.

PAGE 173

159 At this point auditors should focus on the Orlando and Jacksonville HRRs, and determine which hospitals comprising these regions contribute most heavily to the high rate of CABG surgery. Recommendations to Hospital Administrators Recommendations extending to hospital administrators would also include a review of rate fluctuations of discretionary procedures. They should be aware if rates of discretionary procedures at their hospital are appreciably higher t han rates of these procedures in comparable hospitals in the region. If rates of certain procedures are not excessive they would likely not be targeted by Medicare for further inquiry. If rates of certain discretionary procedures are disproportionately h igher than other hospitals, regions, or state average, administrators should understand that they may be on discretionary procedures, there is no reason to reduce the performance of these procedures if they are financially worthwhile to the hospital. However, administrators must perform their due diligence to assure that higher rates never represent unwarranted rates. This would compel them to perform the same regional review as was just prescribed for Medicare to determine where their hospital falls in terms of numbers of enrollees and rates of certain discretionary procedures. If they are located in a region with inordinately high CABG rates they should determine how much their hospital contributes to that rate. If the contribution is substantial then both a financial and clinical dimension should be included in their

PAGE 174

160 internal performance audit procedures of surgical operations. This would also entail convincing their board of directors that such audit standards are necessary. In the literature review there was a lengthy discussion on Redding Medical Center, a subsidiary of T enet Healthcare, Inc. To conclude the discussion on the importance of monitoring rates of discretionary procedures by geographic area by both federal officials and hospital administrators, there is no better example to cite than Tenet. Monitoring rates o f discretionary procedures, as it was shown in the hypothetical example using Figure 8 may have helped to circumvent unnecessary surgeries and other disreputable acts by Tenet The performance of unnecessary and marginally necessary procedures at Tenet be gan to come to light when two men were diagnosed with severe life threatening artery blockage by a Redding physician who recommended immediate bypass surgery. The two men sought second medical opinions, and it was found that both were in good health, had clean arteries, and were not in need of bypass surgery (Chan 2002). The two men reported their suspicious diagnoses to the Federal Bureau of Investigation (FBI). Results of the operations performed at R edding w ere unnecessary. The two men then prevailed in a qui tam lawsuit filed under the False Claims Act (Chan 2002). alleging they had undergone surgical procedures that were de termined to be medically unnecessary, specifically CABG, valve replacements, and catheterizations (Darmiento, 2002).

PAGE 175

161 Although malpractice and malfeasances occurred at Redding where physicians and hospital administrators were directly culpable, Medicare ind irectly deserves at least some of the blame for negligence; for not responsibly monitoring the variability of certain procedures by geographic region, a sin of omission. Figure 9 compares CABG discharges per 1,000 Medicare enrollees for the Redding HRR, t he HRR of which Redding Hospital was located, and the California state average, 1992 2003. Figure 9 Source: Dartmouth Atlas of Healthcare: Studies of Surgical Variation, Cardiac Surgery Report 2005 The Redding HRR shows substantially higher rates of CABG surgery relative to the state average, and this disproportionate rate was largely due to activities occurring at Redding Medical Center. It is reasonable to assume that Medicare should have been alerted of the disparities in rates as early as 1992. rates were double that of state averages. The rate then remained at least four points higher in the Redding HRR relative to the mean rate of CABG procedures occurring in

PAGE 176

162 California. The Redding HRR peaked in 2002 when it exceeded more than 12 CABG surgeries per 1,000 Medicare enrollees. Very similar excessive rates of aortic and mitral valve replacements occurred at Redding in the same time period (Dartmouth, 2005). The question beckons as to how Medicare missed these an omalies? A subsequent question is how did Redding administrators not know there were malfeasances occurring? In addition to unwarranted surgeries, other violations uncovered at Redding Medical Center included false billings, inflated charges, and gross misuse of the Medicare outlier methodology (Darmiento 2002). issued by a UBS Warburg (investment banking and securities) analyst. The report was the result of rifling throug Medicare outlier data) and discovering much higher levels than those of its competitors. It was estimated that Medicare outlier payments compri reimbursements for the year ended September 30, 2002 (Darmiento 2002). If an program, it is very likely upper l behavior. Medicare to nefarious behaviors that entailed wholesale fraud and the endangerment of patients. A corporat e culture that refused to tolerate abuses of Medicare and skillful in and the horrible events that led up to it.

PAGE 177

163 List of Refer ences Agency for Health Care Administr ation (AHCA) (2000 2005), Financial Analysis Annual Hardcopy Reports Agency for Health Care Administration (AHCA) (2000 2005), Financial Analysis Annual Electronic Reports Agency for Health Care Administration (AHCA) (2000 2005), Patient Discharge Data set A gency for Health Care Administration, State Center for Health Statistics ( January 2005), Florida Healthcare Expenditures Agency for Healthcare Research and Quality ( November 2005), Patient Safety and Quality, Patient Safety Problems Increase when Ho spital Profit Margins Decline Over Time Agresti, A. (1999), An Introduction to Categorical Data Analysis, Canada, John Wiley & Sons Aiken, L. S., & West, S. G. (1991). Multiple R egression: Testing and Interpreting Interactions. Newbury Park, London, Sage A merican College of Physicians American Society of Internal Medicine (2001), Primer on Geographic Variation in Health Care, Retrieved January 3, 2009 from: http://www.acponline.org/clinical_information/journals_publications/ecp/sepoct01/prime rgeo.pdf A ssociation of Certified Fraud Examiners (ACFE) (200 5 0 6) Fraud Examiners Manual, 2006 Edition, ACFE World Headquarters, Austin Texas Association of Certified Fraud Examiners (ACFE) (2006) Report to the Nation ACFE World Headquarters, Austin Texas Bagley Literature: Standards for Use and Reporting, with Particular Attention to One Medical Journal of Clinical Epidemiology, 54, 10, 979 985

PAGE 178

164 Baicker, K, Chandra, A. ( April 200 4), Medicare Spending, the Physician Workforce, and Higher Spending and Less Use of High Quality, Effective Care, Health Affairs, Retrieved March 21, 2005 from Data Wa tch Data Base Becker, D. & Kessler, D. (2005), Detecting Medicare Abuse, Journal of Health Economics, 24, 1, 199 210 Birkmeyer, J.D., Sharp, S.M., Finlayson, S.R., Fisher, E.S. & Wennberg, J.E., ( November 1998) Variation Profiles of Common Surgical Procedu res, Surgery 124, 5, 917 23 Bisk Publishing ( 2009 ) Benson P. (2005) Warren Benson Law Group, Medicare Fraud & Other Healthcare Fraud Bowman S.M., Zimmerman F.J., Christakis D.A., Sharar S.R., Martin D.P. (2005) Ho spital characteristics associated with the management of pediatric splenic injuries. JAMA. 294: 2611 2617 Brambor T., Clark W., Golder, M. (2006), Understanding Interaction Models: Improving Empirical Analysis, Political Analysis, 14, 1, 63 82 Brown, H Public, Nonprofit, and For Journal of Health Care Finance, 27, 4, (Retrieved October 8, 2005 from High Beam Research, Inc. database) Burns RB Moskowitz MA Ash A Kane RL Finch M McCarth y E.P ., (March 1997) Variations in the Performance of Hip Fracture Procedures, Medical Care, Vol. 35 No. 3, 196 203 Carlisle DM, Valdez RB, Shapiro MF & Brook RH Health Services Research, Vol. 3 0 No. 1, 27 42 Market, Operational, and Financial Factors with Multi Journal of Health Care Finance, 28, 2 Retrieved October 10, 2005 from Hi gh Beam Research, Inc. database Catlin Health Affairs, 26, 1 Retrieved June 4, 2010 from http://content.healthaffairs.org/cgi/content/full/26/1/142/T1 Retrieved from the int ernet on May 1, 2010 from http://www.cbsnews.com/stories/2002/11/21/ eveningnews/main

PAGE 179

165 Clarke R.L ( December 1991) Management is the Key to Financial Performance Hospital Manageme nt, Healthcare Financial Managemen t, Retrieved February 2, 2005 from Healthcare Financial Management Data Base Cleverly W.O. (December 2004 ) Looking Ahead to Medicare 2005: Successful Hospitals Know How to Manage Both Revenue and Costs to Survive Under a Tightened Medicare Reimbursement Structure, Healthcare Financial Management, Retrieved on February 2, 2005 from Healthcare Financial Management Data Base Cleverly W.O. & Cameron, A.E. (2002) Essentials of Health Care Finance, Aspen Publishers, Inc., G aithersburg, Maryland CMS (June 5, 2003), CMS Issues Final Rule for Outlier Payments to Hospitals, Retrieved March 10, 2005 from http://www.cms.hhs.gov/media/press/release.asp CMS ( September 2004 ). Listing of Medicare Fraud Alerts, Retrieved February 3, 2005 from http://www.cms.hhs.gov/providers/fraud/UMFA2.ASP CMS (2004) Medpar, Retrieved June 4 2003 from http://www.cms.hhs.gov/Med icare CMS (2005) National Health Expenditures Aggregate, Per Capita Amounts, Percent Distribution, and Average Annual Percent Growth by Source of Funds: Selected Calendar Years 1960 2008, Tables 1 13 CMS (no date listed), Medicare Definition of Fraud, R etrieved February 3, 2005 from http://www.cms.hhs.gov/providers/fraud/DEFINI2.ASP Hospital to Pay $9.5 Million in OIG Settlement Retrieved May 30, 2010 from http://www.hcpro.com Convisor Duffy (1996), Law, Harcourt Brace Chicago Cortina, J.M. (Winter 1993) Interaction, Nonlinearity and Multicollinearity: Implications for Multiple Regression, Journal of Management, Vol 19 No. 4 Cowper PA DeLong ER Peterson ED Lipscomb J Muhlbaier LH Jollis JG Pryor DB Mark DB source Use for Coronary Artery Medical Care Vol. 35, No. 4, 320 33 Dafny, L. (September 2003). How Do Hospitals Respond to Price Changes? NBER Working Paper 9972 Darmiento, L. (January 4, 200 Los Angeles Business Journa l Retrieved October 18, 2005 from High Beam Research, Inc. database

PAGE 180

166 Dartmouth Medical School Center for the Evaluative Clinical Sciences (200 5 ) Dartm outh Atlas of Healthcare: Studies of Surgical Variation, Cardiac Surgery Report Retrieved June 30, 2008 from http://www.dartmouthatlas.org/ Dartmouth Medical School (Date Unknown) Center for the Evaluative Clinical Sciences, A Dartmouth Atlas Quick Repo rt The Association Between the Supply of Cardiologists and Rates of Visits to Cardiologists, Retrieved February 2, 2005 from http://www.dartmouthatlas.org/ Dartmouth Medical School: The Center for the Evaluative Clinical Sciences (1999). The Dartmouth A tlas Quick Report: Discretionary Surgery Retrieved February 22, 2005 from http://www.dartmouthatlas.org/ Dartmouth Atlas of Healthcare (2005) Selected Surgical Discharge Rates, Retrieved April 19, 2010 from htt p:// www. dartmouthatlas .org/data/region/ surgical discharges Davis, M. (March 2, 2005) Florida Sues Tenet, Real Money Silver, Retrieved July 30, 2008 from http://www.thestreet.com/ De Lew, N. (2000), Medicare: 35 Years of Service, Health Care Financing Review, 22, 1, 75 103 Doherty, E. ( Date Unknown), "The Balanced Budget Act Colorado Gerontological Society, (Retrieved June 1, 2010 fro m http://www.senioranswers.org Elward, K.S., Penberthy, L.T., Bear, H., Schwartz, D.M., Boudreau, R.M. & Cook, S.S. ( April June 1998) Variation in the Use of Breast Conserving Therapy for Medicare Beneficiaries in Virginia: Clinical, Geographic, and Hospital Characteristics, Clin Perform Q ual Health Care, 6, 2, 63 6 9 : New England Journal of Medicine Fisher, E.S., Bynum, J., Skinner, J. (February 27, 2009), The Policy Implications of Variations in Medicare Spending Growt h, The Dartmouth Institute for Health Policy & Clinical Practice, A Dartmouth Atlas Project Brief Report Fisher, E., Goodman, D., Skinner J. & Bronner, K. (February 27, 2009) Health Care Dartmouth Atlas Project Topic Brief Fisher ES, Wennberg DE, Stukel TA, Gottlieb DJ, Lucas FL, Pinder E.L. ( February 2003) The Implications of Regional Variations in Medicare Spending. Part 2: Health Outcomes and Satisfaction with Care. Annals of Internal Medicine, 18,138 (4):288 98 Fisher ES, Wennberg DE, Stukel TA, Gottlieb DJ. ( October 2004) Variations in the Longitudinal Efficiency of Academic Medical Centers, Health Affairs Web Exclusive

PAGE 181

167 Freund R.J. & Littell R.C. (2000), SAS System for Regression, NC, The SAS Ins titute. study the reverberations of a for profit's entry into a hospital market. The findings are Modern Healthcare Retrieved Oct ober 3, 2005 from High Beam Research, Inc. database Gepenski, L.C., Vogel, W.B. & Langland Hospital & Health Services Administration, 38, 1, 63 80 Gardiner, M.A. (September 1995) Finan cial Ratio Definitions Reviewed, Management Accounting (British ) Garrison G.L. & Wayne D. (June 2008) The Right Moves for Success with Medicare: The Medicare Payment Landscape is Changing, and Hospitals Should be Taking Steps to Ensure they Receive Approp riate Payment. Here's How, Healthcare Financial Management. G arson D.G. (2006 ) Stat Notes, Retrieved June 20 06 from http: //faculty chass. ncsu.edu/garson/ PA765/statnote.htm Healthcare Financial Management Association (HFMA) (2004), Financing the Future Report 2: How Are Hospitals Financing the Future? Capital Spending in Health Care Today, HFMO in Cooperation with GE Healthcare Financial Services, www.hfma.org Profits Harvard Business Review, 93 105. Ingenix (2003), DRG Expert: A Comprehensive Reference to the DRG Classification System, 19 th Edition, St. Anthony Publishing/Medicode Internal Revenue Code (IRC): 16 CFR 501, 512, 513 Jaccard J. & Turrisi (20 03) Interaction Effects in Multiple Regression, London, Sage International Advances in Economic Research Retrieved October 17, 2005 from High Beam Research, Inc D atabase Kesselheim, A.S. & Brennan, T.A. (2005), Overbilling vs. Downcoding The Battle Between Physicians and Insurers, New England Journal of Medicine, 352, 9, 855 857

PAGE 182

168 Keller, R.B., Soule, D.N., Wennberg, J.E. & Hanley, D.F. (1990), Dealing with Geographic Variations in the Use of Hospitals. The Experience of the Maine Medical Assessment Foundation Orthopedic Study Group, The Journal of Bone & Joint Surgery, 72: 1286 1293 onship Journal of Health Care Finance, 29, 2 Retrieved October 14, 2005 from High Beam Research, Inc. D atabase Langabeer, J. (March 2008) Hospital Turnaround Strategies. Hospital Topics. R etrieved January 3, 2009 from HighBeam Research Inc. Database Lee C.F. (198 4 ) Financial Analysis and Planning: Theory and Application, Mass. Addison Wesley Louis, W.R. (2008) Moderated Multiple Regression: Work Notes and Syntax, School of Psychology, Uni versity of Queensland Luther, S. & Studnicki, J. (2003) Lectures in Managed Care Department of Health Policy & Management College of Public Health, University of South Florida Mendenhall, W., Sincich, T.L. (2003) A Second Course in Statistics: Regression Analysis, Prentice Hall, New Jersey Middleton, C. ( December 2002), Hospitals Are Just Playing the Medicare Game, Health Policy Prescriptions, 1, 12, 112 117 Retrieved June 1, 2010 from www.mwhcmm.com Needleman, J., Choilet, D., & Lamphere, J. (March Apri Health Affairs, 16, 2, 187 195 Needleman, J., Lamphere, J & Choilet, D. (July Hospital Conversions in Florida: What does it mean for C ommunities when Hospitals Health Affairs 18, 4, 125 133 Ovsyshcher I.E & Furm an S Pacing Clinical Electrophysiology, 26, 1, 2, 474 47 8 Plank, T.M. & Plank, L.R. (2000) Accounting Desk Book: The Acc Instant Answer Book, Prentice Hall 11 th Edition.

PAGE 183

169 Psaty, BM, Boineau, R, Kuller, LH, & L uepker, R.V., (1999), The Potential Costs of Upcoding for Heart Failure in the United States, American Journal of Cardiology 84, 1, 108 9 Rajarman L (Spring 200 4 ) Lectures i n SAS Depar tment of Epidemiology & Biostatistics, University of South Florida Journal of Heal th Care Finance, 31, 3, Retrieved October 10, 2005 from High Beam Research, Inc. database Rosenberg, M.A., Fryback, D.G., Katz, D.A. (2000), A Statistical Model to Detect DRG Upcoding, Health Services and Outcomes Research Methodology, 1, 3 4, 233 252 SAS Cary, North Carolina Schwartz, S. (Fall 2004) Lectures in Advanced Research Methods in Epidemiology, Department of Epidemiology & Biostatistics, College of Public Health, Univer sity of South Florida Owned Multi Hospital Systems with Not for Health Care Management Review, 16, 2, 31 37 Sear, A.M. (2004), Effects of Medicare BBA Spending Reductions on the Profitability of General Acute Care Hospitals, Journal of Healthcare Management, 49, 2, 119 133 Capital How Do You Ensure You Have Enough 18, 2005 form Shattuck Hammond Partners, Inc. data base Schu l man, T.M. (July 2008) Hospital Financial Performance: Trends to Watch: Financial Indicators Derived from Medicare Cost Report Data are Reliable Tools for Assessing the Effectiveness of a Hospita l's Operations, Healthcare Financial Management 2008. Retrieved January 3, 2009 from HighBeam Research Inc. Database Silverman, E., Skinner, J. (2001) Are For Profit Hospital s Really Different? Medicare Upcoding and Market Structure, NBER Working Papers 8133 / National Bureau of Economic Research, Inc., Retrieved March 15, 2005 from http://search.hp.netscape.com/ hp/boomframe.jsp?query=medicare+drg+upcoding&page=1&offset=1&result_url=redir between For Profit Hospital Ownership and Increased The New England Journal of Medicine, 341, 6, 420 426

PAGE 184

170 Silversmith, J. (December 2000 Budget Act Minneapolis, MN, 55413 Sincich, T. (Fall 2004) Lectures in Applied Linear Statistical Models, Department of Decision Science & Information Systems, College of Busines s, University of South Florida Sincich, T. (Spring 2005) Lectures in Multivariate Statistical Models, Department of Decision Science & Information Systems, College of Business, University of South Florida Ultimate Healthcare Issue: Capital How Do You Ensure You Have Enough form Shattuck Hammond Partners, Inc. D atabase Profit and Tax Exempt Nonpr Virginia Polytechnic Institute and State University in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy in Business with a Major in Accounting Spencer, B.A., Fung, C.H., Wang, M., Rubenstein, L.V. & Litwin, M.S. (December of Early Stage Prostate Cancer, Journal of Urology, 6 1, 2362 5 Stokes, M., Davis, C.S. & Kock, G.G. (2003), Categorical Data Analysis Using the SAS System, North Carolina, SAS Institute, Inc. Triad Hospitals, Inc., Prospectus Supplement (Prospectus dated March 31, 2005), Form 424B5, Filing Date June 30, 2005, Filed Pursuant To Rule 424(b)(5), Registration No. 3 33 123294 United States Code (USC) Title 26 (Internal Revenue Code) United States Code (USC) Title 31 3 7 2 9 3 733 (Federal False Claim s Act) U.S. Department of Health and Human Services (HHS), Office of the Inspector General (OIG) (February 14, 2003) OIG News: Johns Hopkins University Will Pay $800,000 to Resolve Medicare Fraud Case U.S. Department of Justice (DOJ) ( November 2003), Civil Fraud Recoveries Total $2.1 Billion for FY 2003. False Claim Act recoveries exceed $12 billion since 1988, Retrieved March 21, 2005 from www.usdoj.gov/opa/pr/2003/November/03_civ_ 613.htm

PAGE 185

171 U.S. Department of Health & Human Services (HHS) Office of the Inspector General (OIG) (December 11, 2003), OIG and Tenet Healthcare Corporation Reach Divestiture Agreement to Address Exclusion of Redding Medical Center, Office of the Inspector General, 330 Independence Ave., SW, Washington, DC, 20201, 202 619 1343 U.S. Department of Housing & Urban Development (HUD), Median Income Data for all Counties in the United States, 2000 2005, Section 8 Program, Retrieved June 1, 2008 from hud.gov U.S. Department of Justice (DOJ) ( May 17, 2004), Tenet Healthcare to Pay U.S. $22.5 M illion for Improperly Billing Medicare, Retrieved March 24, 2005 from www.usdoj.gov/opa/pr/2004/March/04_civ _183.htm U.S. Department of Justice (DOJ) (December 14, 2000), HCA The Health Care Company & Subsidiaries to Pay $840 Million in Criminal Fines and Civil Damages and Penalties: Largest Government Fraud Settlement in U.S. History Retrieved July 30, 20 05 from http://www.justice.gov/opa/pr/2000/December/696civcrm.htm Utah Department of Health (DOH) (July 2008), Diagnostic Related Groups Code List, Retrieved October 1, 2009 from http://health.utah.gov Vitale, M.G., Krant, J.J., Gelijns, A.C., Heitjan D.F., Arons, R. R., Bigliani, L.U., Flatow, E.L. (1999), Geographic Variations in the Rates of Operative Procedures Involving the Shoulder, Including Total Shoulder Replacement, Humeral Head Replacement, and Rotator Cuff Repair, The Journal of Bond & Joi nt Surgery, 81: 763 772 Vogel, W.B., Langland Health Care Management Review, 18, 2, 15 26 Wennberg J.E. ( January 24, 2005) Variation in Use of Medicare Services Among Regions and Selected Academic Medical Centers: Is More Better? Presentation by John E. Wennberg, MD, MPH, Duncan W. Clark Lecture, New York Academy of Medicine, New York, New York Weinstein, J.N., Bronner K.K., Morgan, T.S. & Wennberg, J.E. (October 2004), And Geographic Variations In Major Surgery For Dege nerative Diseases of the Hip, Knee, and Spine: Is there a roadmap for change? Health Affairs, 7 Wennberg, J.E., Brownlee, S., Fisher E.S., Skinner, J.S., Weinstein, J.N. (December 2008), An Agenda for Change: Improving Quality and Curbing Health Car e Spending: Opportunities for Congress and the Obama Administration, A Dartmough Atlas White Paper, The Dartmouth Institute for Health Policy & Clinical Practice The Ort hopedic The Journal of Bone & Joint Surgery, 80: 327 330

PAGE 186

172 Weinstein, J.N., Lurie, J.D., Olson, P.R., Bronner, K.K. & Fi sher, E.S. (2006), United 2003, Spine, 31, 23, t he P4P Ag enda, Part 1: How Medicare Can Improve Patient Decision Making a nd Reduce Unnecessary Care, Health Affairs 26, 6, 1564 1574 ospital Medical Care, 30, 2, 174 187. Williams, J.R., Haka, S.F., Bettner M.S. & Carcello, J.V. (2008), Financial & Managerial Accounting, McGraw Hill Irwin. Exemp t Hospitals Special? (Differences between Not for Profit and For Healthcare Financial Management Retrieved October 31, 2005 from High Beam Research, Inc. D atabase Young, G. (January 2006), A Practical Guide to Multiplicative Interaction Variables in Policy Research, GW Institute of Public Policy Prospective Payment Journal of Health Care Finance, 31, 1 Retrieved October 9, 20 05 from High Beam Research, Inc. D atabase Journal of Healthcare Finance, 31, 3 Retrieve d October 9, 2005 from High Beam Research, Inc. D atabase Zack, G. (2001), Lecture on Managing Non Profit Organizations: Uncovering Hidden Liabilities Non profit Research Center, Inc. Maryland

PAGE 187

173 Bibliography Alam, P., Elshafie, E & Jarjoura, D. (Sept ember 2008), The Effect of Ownership Structure on Performance of Hospitals, Academy of Accounting and Financial Studies Journal. Retrieved January 3, 2009 from HighBeam Research: http://www.high beam.com/doc/1P3 1560861581.html Allison, P.D. (2003) Logistic Regression Using the SAS System: Theory and Application, North Carolina, SAS Institute, Inc. Altman, E.I. (1968) Financial Ratios, Discriminant Analysis, and the Prediction of Corporate Bankr Journal of Finance 23, 589 609 American Hospital Association (AHA) (2005) Annual Survey of Hospitals, 1998 2004 Anderson, S., Wolfe, J. (June 2002). A Perspective on Audit Malpractice Claims, Journal of Accountancy Online, June 2002, Retrieved f rom AICPA February 3, 2005 from http://www.aicpa.org/pubs/jofa/sep2002/anderson.htm Bierman, A.S. (2004), Coexisting Illness and Heart Disease among Elderly Medicare Managed Care Enrollees, Health Care Financing Review, 25, 4, 105 117 Bettner, M.& Collin s, F. (1987), Physicians and Administrators: Inducing Collaboration, Hospital and Health Services Administration 32, 2, 151 60 Medical Care, 28, 9, 853 8 64 Journal of Health Politics, Policy, and Law, 9, 87 118 Undergoing Coronary CMAJ, 172, 9, 1183 1186 For Profit Health Care Strategic Management, Retrieved October 10, 2005 from High Beam Research, Inc. dat abase

PAGE 188

174 Burrill, D.F. (2007) Modeling and Interpreting Interactions in Multiple Regression, Minitab Inc., Minitab.com Business Wire, "Fitch Rates All Children's Hospital (Florida) $90MM Series 2007 Bonds 'AA' Underlying" (2007) Retrieved January 3, 2009 from High Beam Research, Inc. database Carpenter, C.E. (1991) "The Marginal Effect of Bond Insurance on Hospital, Tax Exempt Bond Yields," Inquiry, 28, 1, 67 73 Carpenter, C.E. (1992) "Determination of Hospital Tax Exempt Debt Yields: Corrections for Selection and Simultaneous Equation Bias," Health Services Research, 27, 5, 695 713 Carpenter, C.E. (1994), The Impact of Intra DRG Severity of illness on Hospital Profitability: Implications for Payment Reform, Journal of Health Politics, Policy & Law, 19, 4, 729 51 Casella G. & Lehmann, E.L. (1999) Theory of Point Estimation, Springer. Chirikos, T. (1989) Lectures in Health Economics I & II, Department of Health Policy & Management, College of Public Health, University of South Florida Clarke, Richard L. (1999) Profitability is no measure of viability, Healthcare Financial Management, Retrieved January 3, 2009 from HighBeam Research Database Cleverly, W.O. (1986), Essentials of Health Care Finance. Aspen, Rockville Coddington, D. & Moore, K. (1987), Market Driven Strategies in Health Care. San Francisco, Jossey Bass Cowan, C., Catlin, A., Smith, C. & Sensenig, A. (2004), National Health Expenditures, 2002, Health Care Financing Review, 25, 4, 143 166 Daily, J. (Date Undefined), The Develop ment, Implementation, and Results of Tenet Jennifer Daily, Senior Vice President, Clinical Quality, Chief Medical Officer, Tenet Healthcare Corporation, Dallas, TX, 469 893 2988 Davis, M. (March 3, 2005) Ten et Suit Comes Out of Left Field, Real Money Silver, Retrieved July 30, 2008 from http://www.thestreet.com/ Dalton, K., Norton & E., Kilpatrick, K., (February 2001), A Longitudinal Study of the Effects of Graduate Medical Education on Hospital Operating C osts, Health Services Research Retrieved March 15, 2005 from http://www.hret.org/hret /public ations/hsr.html

PAGE 189

175 Dartmouth Medical School: The Center for the Evaluative Clinical Sciences (1999). The Dartmouth Atlas of Health Care 1999: The Quality of Medical Care in the United States: A Report on the Medicare Program, Health Forum, Inc. (an American Hospital Association Company). Dartmouth Medical School Ce nter for the Evaluative Clinical Sciences, A Dartmouth Atlas Quick Report Hospitalization for Medical Conditions, Retrieved February 2, 2005 from http://www.dartmouthatlas.org/ Dartmouth Medical School Center for the Evaluative Clinical Sciences, A Da rtmouth Atlas Quick Report Percent of Medicare Deaths Associated with Admission to ICU, Retrieved February 2, 2005 from http://www.dartmouthatlas.org/ Dartmouth Medical School Center for the Evaluative Clinical Sciences, A Dartmouth Atlas Quick Report The Association Between Hospital Beds and Hospitalization for Congestive Heart Failure and Hip Fracture, Retrieved February 2, 2005 from http://www.dartmouthatlas.org/ Dawson, J. Interpreting Interaction Effects. Retrieved July 9, 2008 from http://www. jeremydawson.co.uk/slopes.htm Dawson, J. F., & Richter, A. W. (2006). Probing Three Way Interactions in Moderated Multiple Regression: Development and Application of a Slope Difference Test. Journal of Applied Psychology, 91, 917 926 Declerq K. The Odds Ra tio, Retrieved December 1, 2008 from http://www sas.com Defriese, G.H. (December 1996) Non Profit Conversions: A Challenge and an Opportunity for Health Services Research with a Clear Public Benefit, Health Services Research Editorial, Retrieved from th e Gail Group (2004) (originally published by the American College of Healthcare Executives 1996) Retrieved March 24, 2005 from http://findarticles.com/p/articles/mi_m4149/is_n5_v 31/ai_19018155 Ellis, B.H., Shannon, E.D., Cox, J.K. & Aiken, L., et. al., (2004), Chronic Conditions: Results of the Medicare Health Outcomes Survey, 1998 2000, Health Care Financing Review, 25, 4, 75 91 Der, G. & Everitt, B.S. A Handbook of Statistical Analysis Using SAS, Chapman & Hall, CRC Derrig, R.A. (2002, September). Insurance Fraud, The Journal of Risk and Insurance, 69, 3, 271

PAGE 190

176 Devereaux, P.J. Heels Ansdell, D., Lacchetti, C., Haines, T, & Burns, K.E.A, et. al., (2004) Payments for Care at Pr ivate For Profit and Private Not for Profit Hospitals: A Systematic Review and Meta Analysis, Canadian Medical Association Journal, 170, 12, 1817 1824 Dupuy, A.; Benchikhi, H.; Roujeau, J. C.; Bernard, P.; Vaillant, L.; Chosidow, O.; Sassolas, B.; Guillau me, J. C.; Grob, J. J.; and Bastuji Garin, S. (1999). Risk Factors for Erysipelas of the Leg (Cellulitis): Case Control Study. British Medical Journal 318:1591 1594. Operating Margin Down, Modern Healthcare, Retrieved March 15, 2005 from http://www.modernhealthcare.com Federal Register (October 26, 1999), 45 CFR Part 61, Health Care Fraud and Abuse Data Collection Program: Reporting of Final Adverse Actions; Final Rule Feld man, R. (2001, June). An Economic Explanation for Fraud and Abuse in Public Medical Care Programs, The Journal of Legal Studies, 569 577 Fuchs, V.R. (January February 1999), Health Care for the Elderly: How Much? Who Will Pay For It? Health Affairs, 11 21 Florida Agency for Health Care Administration FUHRS Financial Data, FY98 04 Gail Group (2003), Hospital Operating Margins Still Declining, Retrieved from the Gail Group Data Bas e (Reprinted from Healthcare Financial Management, January 2000) on February 2, 2005 Gregory, D. (1992), Strategic Alliances Between Physicians and Hospitals in Multi Hospital Systems, Hospital and Health Services Administration 37, 247 58 Goes, J.B. (Oc tober 1995), The Effects of Hospital Physician Integration Strategies on Hospital Financial Performance, Health Services Research Retrieved on March 17, 2005 from http://www.hret.org/hret/publi cations/hsr.html Haffer, S.C., Bowen, S.E. (2004), Measuring and Improving Health Outcomes in Medicare: The Medicare HOS Program, Health Care Financing Review, 25, 4, 1 3 Hair, J.F, Anderson, R.E., Tatham, R.L. & Black, W.C. (1998) Multivariate Data Ana lysis, Prentice Hall, New Jersey Journal of Health Care Finance, 31, 3, Retrieved October 10, 2005 from High Beam Research, Inc. database

PAGE 191

177 Hasan M.M. (1996), Let's End the Nonp rofit Charade, New England Journal of Medicine, 334, 1055 1057 Healthcare Financial Management, Retrieved October 12, 2005 from High Beam Research, Inc. database Hawryluk, M. (Mar ch 3, 2003), Medicare Fraud: Back to Stricter Scrutiny? Amednews.com Retrieved March 27, 2005 from http://www.ama assn.org/amednews/2003/03/03/gvsa0303.htm Henderson, D.A. Chase, B.W., Woodson, B.M. (January 2002), Performance Measures for Non Profit Orga nizations, Journal of Accountancy Online, Retrieved from AICPA March 5, 2005 from http://www.aicpa.org/pubs/jofa/jan2002/hender.html Henderson, M., Health Care Fraud and Abuse: Diagnosis Related Group (DRG) Creep, Price Waterhouse Coopers (Mr. Henderson i s a partner with Price Waterhouse Coopers in Health Care Fraud and Abuse Practice). http://ahcaxnet.fdhc.state.fl.us/nhcguide/static/state_inspection.html http://v8doc.sas.com/sashtml/stat/chap55/sect14.htm http://www.ats.ucla.edu/stat/sas/faq/sas9_stat_plots.htm http://www.stattutorials.com/SAS/TUTORIAL PROC CORR.htm Hutt, W.H. (1975), A Rehabilitation of Say's Law, Athens, Ohio, Ohio University Press. Hutt, W.H., (1979). The Keynesian Episode, Indianapolis, Ind. Li berty Press Internal Revenue Service (IRS) Ruling 98 15, 1998 12 IR Bulletin No. 6 Jacob, J.A. & Jackson, C. (June 19, 2000). Oxford Pressing Physicians for E&M Upcoding Refunds, Retrieved March 23, 2005 from http:/www.ama.assn.org.amednews /2000/06/19/b il10619.html Jekel, J.F., Katz, D.L. & Elmore, J.G. (2001) Epidemiology, Biostatistics and Preventive Medicine, W.B. Saunders, U.S. Johnson, D.E. (1998) Applied Multivariate Methods for Data Analysis, Brooks/Cole, U.S. Kahn, C.N. & Kuttner, H. (1999, Ja nuary February), Budget Bills and Medicare Policy: The Politics of the BBA, Health Affairs, 37 47

PAGE 192

178 Kahn, H.A., & Sempos, (1989), C.T. Statistical Methods in Epidemiology, New York, Oxford University Press Karp, A.H., Getting Started with Proc Logistic, S tatistics, Data Analysis, and Data Mining (Paper 248 26): Sierra Information Services, Inc., 19229 Sonoma Highway, PMB 264, Sonoma, California 95476 USA, 707 996 7380, Retrieved February 12, 2008 from www.Sier raInformation.com Klemm, J.D. (2000), Medicaid Spending: A Brief History, Health Care Financing Review, 22, 106 112. Kongstvedt, P.R. (2003). Essentials of Managed Health Care, Sudbury, MA, Jones & Bartlett Lawrence, M. & Wells, J.T. (2004, Octo ber), Basic Legal Concepts: Beware of Insufficient Knowledge of the Law, Journal of Accountancy Online, Retrieved March 23, 2005 from http://www.aicpa.org/pubs/jofa/oct2004/lawrence.htm. Lieff, Cabraser, Heimann & Bernstein, LLP, Press Articles on Tenet H ospital, Retrieved February 16, 2005 from, http://www.uncessary heart surgery.com/press.html Leone, A. J., Van Horn, R.L. & Gerard, J.W. (2005), Abnormal Returns and the Regulation of Nonprofit Hospital Sales & Conversions, Journal of Health Economics, 24 1, 113 135 Long H.W. (May June 1991), Fiscal Health of Hospitals in a Decade of Medicare, Physician Executive, Retrieved on February 28, 2005 from http://www.acpe.org/ McCall, N, Korb, J., Petersons, A., & Moore, S. (2002), Constraining Medicare Home Health Reimbursement: What Are the Outcomes? Health Care Financing Review, 24, 257, 57 76 Midkiff, K.A., Journal of Accountancy Online, Retrieved March 23, 2005 from http://www.aicpa.org/pubs/ jofa/jan2004 /chklist.htm Miller, C. R., Epstein, J., Woolhandler, S. & Himmelstein, D. U. (1997), Costs at For Profit and Not for Profit Hospitals, New England Journal of Medicine, 337, 1779 1780 Mougeot, M. & Naegelen, F. (2005), Hospital Price Regulation and Expenditure Cap Policy, Journal of Health Economics, 24, 1, 55 72 Bypass Grafting in Octogenarians: Cl inical and Economic Outcomes at Community 755

PAGE 193

179 NBER Working Papers /National Bureau of Economic Research, Inc., Retr ieved March 15, 2005 from http://search.hp.netscape.com/hp/boomframe.jsp?query =medicare+drg+ upcoding&page=1&offset=1&result_url=redir North Carolina Rural Health Research and Policy Analysis Center Cecil G. Sheps Center for Health Services Research (2003, July), A Primer on Interpreting Hospital Margins, The Universit y of North Carolina at Chapel Hill, Chapel Hill, NC 27599 7590, Retrieved February 28, 2005 from http://www.shepscenter.unc.edu/research_programs/ rural_ program Panzar and Willig (1977) Econ omies of scale and Economies of Scope in Multi Output Production, Quarterly Journal of Economics, 91: 481 493 Patrick, D.L. Grembowski, D., Durham, M., Beresford, S. & Diehr, et. al. (1999), Cost Health Car e Financing Review, 20, 4, 25 43 PR Newswire Association LLC (May 23, 2008), Saint Joseph's Hospital of Atlanta Sued Over Patient Admission and Billing Scheme, According to Page Perry, LLC. PR Newswire Retrieved January 3, 2009 from HighBeam Research, I nc. Database Pope, G.C., Kautter, M.S., Ellis, J., Ash, R.P., et. al. A.S. (2004), Risk Adjustment of Medicare Capitation Payments Using the CMS HCC Model, Healthcare Financing Review, 25, 4, 119 141 Pope, G.C., Kautter, J., Ash, A.S., et al. (1996), R isk Adjusted Medicare Capitation Rates Using Ambulatory and Inpatient Diagnoses, Health Care Financing Review, 17, 3, 77 100 Reinhardt, U. (2000) The Economics of For Profit and Not For Profit Hospitals, Health Affairs, 19, 6, 178 186, Retrieved March 27 2005 from http://www.ncbi.nlm.nih.gov/ entrez/query.fcgi?cmd=Retrieve&db=pubmed&dopt=Abstract&list_uids=11192401&ito ol=iconfft Rizzo J.A. (February 1992) Supply and Demand Factors in the Determination of Medica Health Services Research, Retrieved February 1, 2005 from http://www.hsr.org/hsr/database/search.html Rothman, K.J. & Greenland, S. (1998) Modern Epidemiology, Philadelphia, Lipp encott Williams & Wilkins Saucier, P., Bezanson, L., Booth, M., Bratesman, S., Fralich, J.T. & Gilden, D., et. al., (1998), Linked Data Analysis of Dually Eligible Beneficiaries in New England, Health Care Financing Review, 20, 2, 91 108

PAGE 194

180 Sear, A.M. (1992 Owned Multi Hospital and Health Services Administration, 37, 3, 403 415 Seshamani, M. & Gray, A.M., (2004), Ageing and Health Care Expenditure: the Red Herring Argum ent Revisited, Health Economics, 13, 303 314 Frontiers of Health Services Management 7, 1, 3 32 Solucient, LLC 2005 (June 2001), Hospitals Report Tiny Operating Profit Ma rgin Gain as BBA Reimbursement Levels Exact Ongoing Toll, Reprinted from Health Industry Today, 57 63 Heffler, S., Smith, S., Keehan, S., Borger, C. et. al. (2005), Trends: U.S. Health Spending Projections For 2004 2014, Health Tracking, 23, 74 85. Szk lo, M. & Nieto, F.J. (2004) Epidemiology: Beyond the Basics, Jones & Bartlett, Boston Siegel, S.B. (August, 2003) How to Tame Health Care Costs, Journal of Accountancy Online Retrieved March 23, 2005 from http://www.aicpa.org/pubs/jofa/aug2003/ siegel.htm Silverman, E., Skinner, J., (2004) Medicare Upcoding and Hospital Ownership, Journal of Health Economics, 23, 369 389 Telberg, R. (April 2004) A Joint Effort to Fight Corporate Fraud, Journal of Accountancy Online, Retrieved March 23, 2005 from http://www.aicpa.org/pubs/jofa/apr 2004/telberg.htm The False Claims Act Legal Center, What is the F alse Claims Act & Why is it Important? TAF Education Fund, Retrieved March 19, 2005 from http:// www.taf.org/total2004.htm U.S. Department of Health and Human Services (HHS) (October 1995), Medicare Risk HMO Performance Indicators, OEI 06 91 00734 U.S. De partment of Health and Human Services (HHS), Office of the Inspector General (Date Undefined), Financial Arrangements between Hospitals and Hospital Based Physicians, 0EI 09 89 00330 U.S. Department of Health and Human Services (HHS) (November 2001), Audi t of Medicare Part B Services Billed by Community Urgent Care Medical Group for the Period January 1, 1995 through July 31, 1999, A 09 00 00089

PAGE 195

181 U.S. Department of Health and Human Services (HHS) and the U.S. Department of Justice (DOJ) (September 2003), H ealth Care Fraud and Abuse Control Program: Annual Report for FY 2002, Office of the Inspector General, 330 Independence Ave., SW, Washington, DC, 20201, 202 619 1343 U.S. Department of Health & Human Services (HHS) Office of the Inspector General (OIG) (December 9, 2004), OIG Report Highlights Progress Against Waste, Abuse, and Fraud, Office of the Inspector General, 330 Independence Ave., SW, Washington, DC, 20201, 202 619 1343 Journal of Accountancy Online Retrieved March 15, 2005 from http://www.aicpa.org/pubs/jofa/feb2001/wells.htm Journal of Accountancy Online, Retrieved March 3, 2005 from http://www.aicpa. org/pubs/ jofa/jul2001/ wells.htm W Journal of Accountancy Online, Retrieved March 3, 2005 from http://www.aicpa.org/pubs/jofa/ aug2001/wells.htm Wilensky, G.R. & Newhouse, J.P. (1999, January Health Affairs, 92 106 Willan, A.R., Briggs, A.H. & Hoch, J.S. (2004) Regression Methods for Covariate Adjustment and Subgroup Analysis for Non Censored Cost Effectiveness D ata, Health Economics, 13, 461 475 Woolhandler, S. & Himmelstein, D.U. (1997), Costs of Care and Administration at For Profit and Other Hospitals in the United States, The New England Journal of Medicine, 336, 11, 769 774 Woolhandler S. & Himmelstein D.U (1999), When Money is the Mission the High Costs of Investor Owned Care, New England Journal of Medicine, 341, 6, 444 446 Woolhandler, S., Himmelstein, D.U. (June 2004) The high costs of for profit care, Commentary, 170, 12, Canadian Medical Association Journal, Retrieved March 23, 2005 from doi:10.1503/cmaj.1040779. Wu, Y. (Summer 2004) Lectures in Categorical Data Analysis, Department of Epidemiology & Biostatistics, College of Public Health, Univ ersity of South Florida Hospital Profitability in the Post Journal of Health Care Finance, Retrieved October 9, 2005 from High Beam Research, Inc. Database

PAGE 196

182 Youn Profitability in Florida: a Revisitation.(a Study of the Financial Performance of Several Research in Healthcare Finance, Retrieved October 11, 2005 from Hig h Beam Research, Inc. Database Zhu, Y. (Fall 2003) Lectures in Survey Sampling in the Health Sciences, Department of Epidemiology & Biostatistics, College of Public Health, University of South Florida Zhu, Y. (Spring 2004) Lectures in Survival Analysis, Department of Epidemiology & Biostatistics, College of Public Health, University of South Florida

PAGE 197

183 Appendices

PAGE 198

All Florida Hospital Referral Regions HRR All Surgical Discharges Hip Fracture Repair Resection for Colon Cancer Coronary Angiography CABG Carotid Endarterectomy Cholecystectomy Hip Replacement National Average 102.12 7.53 1.76 22.77 5.18 3.02 4.29 3.18 State Average 108.56 7.46 1.77 25.64 5.42 3.6 4.44 3.27 Bradenton 115 117.94 7.71 2.14 28.13 5.49 4.87 6.6 3.75 Clearwater 116 107.44 7.47 2.09 27.38 5.04 3.54 4.03 3.18 Fort Lauderdale 118 105.83 7.54 1.66 24.6 4.95 2.96 3.67 3.91 Fort Myers 119 109.99 6.74 1.63 21.98 4.94 3.93 4.13 4.2 Gainesville 120 108.00 8.1 1.77 26.72 5.04 2.68 4.58 2.51 Hudson 122 119.24 8.07 1.81 34.2 6.15 4.87 5.78 3.14 Jacksonville 123 107.4 7.81 1.9 32.2 5.98 4.25 3.95 2.68 Lakeland 124 105.35 7.57 1.82 31.1 4.73 2.9 4.18 2.78 Miami 127 101.39 7.21 1.75 22.08 3.36 1.59 4.51 2.08 Ocala 129 110.41 6.9 1.86 27.6 5.6 4.75 4.82 3.17 Orlando 130 111.75 7.16 1.84 26.59 6.47 4.07 4.8 3.27 Ormond Beach 131 97.53 8.02 1.72 22.56 5.73 2.17 3.66 3.49 Panama City 133 126.96 9.73 1.73 36.4 7.97 3.37 5.03 3.07 Pensacola 134 122.13 8.48 1.66 26.76 5.9 3.69 4.88 2.82 Sarasota 137 103.84 7.45 1.75 18.28 5.83 3.26 3.61 4.14 St. Petersburg 139 111.09 8.09 1.97 20.72 4.48 4.9 5.58 3.19 Tallahassee 140 101.69 7.94 1.62 21.96 5.03 3.06 5 2.25 Tampa 141 104.08 7.4 1.7 26.64 5.46 3.98 4.69 2.68 Mean 109.56 7.74 1.8 26.44 5.45 3.6 4.64 3.13 Standard Deviation 7.72 0.67 0.15 4.83 0.95 0.95 0.79 0.6 Coefficient of Variation 0.07 0.09 0.08 0.18 0.17 0.26 0.17 0.19 Source: Dartmouth Atlas of Healthcare, Data Tables, Hospital referral regions (HRRs): Regional markets for Tertiary Medical Care Retrieved on September 24, 2006 from http://www.dartmouthatlas.org/ 184 Appendix 1 Rates of Surgical Procedures per 1,000 Medicare Enrollees, All Florida Hospital Referral Regions, 2003 Appendix 1 presents similar data from the same source as shown in Charts 1 3 in the literature review. Data used for Appendix 1 are from 2003 while data from Charts 1 3 are from 2005. It was determined that 2003 would be appropriate for the presentation in Appendix 1 because the study period is 2000 2005, and 2003 occurs in the middle of the study period.

PAGE 199

All Florida Hospital Referral Regions Knee Replacement Mastectomy for Cancer* PCI Lower Extremity Revascularization Abdominal Aortic Aneurysm Repair TURP for BPH** Back Surgery Valve Replacement National Average 6.88 1.19 11.27 1.41 0.97 5.23 4.02 1.39 State Average 6.71 1.06 12.91 1.36 1.05 4.54 4.52 1.54 Bradenton 7.51 1.12 14.46 1.22 1.11 5.72 6.7 1.28 Clearwater 6.28 1.12 17.93 1.42 1.23 3.95 3.4 2.01 Fort Lauderdale 5.62 0.73 11.51 1.21 0.92 4.42 3.76 1.73 Fort Myers 9.55 0.67 11.8 1.49 1.33 4.87 6.77 1.47 Gainesville 6.24 1.64 14.8 1.1 1.08 5 6.38 0.96 Hudson 7.38 1.11 17.87 1.13 0.93 4.39 3.83 2.06 Jacksonville 5.98 0.9 14.34 1.41 1.36 4.28 3.27 1.51 Lakeland 6.58 1.36 12.86 1.1 0.95 4.99 3.87 1.74 Miami 4.07 1.02 10.34 1.13 0.64 5.09 2.55 1.23 Ocala 6.99 1.52 14.74 1.11 1.04 3.44 5.22 1.74 Orlando 6.83 1.16 13.95 1.33 1.03 3.8 4.12 1.52 Ormond Beach 6.83 0.93 10.17 1.72 0.95 2.65 4.31 1.54 Panama City 6.8 Suppressed*** 13.06 1.33 Suppressed*** 11.51 5.19 1.72 Pensacola 8.76 1.11 14.18 2.58 1.27 4.36 6.5 1.3 Sarasota 8.12 1.21 11.59 1.64 1.04 5.4 6.2 1.72 St. Petersburg 6.59 1.45 9.84 1.8 1.14 6.82 4.44 1.61 Tallahassee 6.74 1.6 9.65 1.26 0.95 5.78 5.15 0.75 Tampa 5.72 1.22 11.32 1.42 1.08 5.28 3.77 1.28 Mean 6.81 1.17 13.02 1.41 1.06 5.1 4.75 1.51 Standard Deviation 1.22 0.28 2.49 0.36 0.18 1.86 1.31 0.34 Coefficient of Variation 0.18 0.24 0.19 0.26 0.17 0.37 0.28 0.22 Per 1,000 Female Medicare Enrollees, ** Per 1,000 Male Medicare Enrollees, ***Suppressed indicates there is not enough data in the HRR to protect patients' identity. Source: Dartmouth Atlas of Healthcare, Data Tables, Hospital referral regions (HRRs): Regional markets for Tertiary Medical Care Retrieved on September 24, 2006 from http://www.dartmouthatlas.org/ 185 Appendix 1 Rates of Surgical Procedures per 1,000 Medicare Enrollees, All Florida Hospital Referral Regions, 2003

PAGE 200

Source: Dartmouth Atlas of Healthcare, Data Tables, Hospital referral regions (HRRs): Regional markets for tertiary medical care. Downloaded from the internet on September 24, 2006 at address: 186 Appendix 1 Rates of Surgical Procedures per 1,000 Medicare Enrollees, All Florida Hospital Referral Regions, 2003 0.17 0.19 0.18 0.24 0.19 0.37 0.28 0.22 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 CABG Hip Replacemen t Knee Replacemen t Mastectomy PCI TURP for BPH Back Surgery Valve Replacemen t Coefficients of Variation for Certain Discretionary Procedures, 2003 The solid horizontal line shows the coefficient of variation for the benchmark hip fracture repair at 0.09. All other procedures have coefficients of variation far above the benchmark figure. The chart shows a graphic presentation of the data presented on the previous two pages. The chart does not include all medical procedures shown in Appendix 1, only those that are discussed further in the investigation.

PAGE 201

City Name HRR Reimbursements 2000 Reimbursements 2005 %Change 2000-2005 Z-Score Reimbursements 2000 Z-Score Reimbursements 2005 Z-Score %Change 2000-2005 Bradenton 115 2,350 2,110 -10.2% (0.41) (1.19) (0.95) Clearwater 116 2,503 2,824 12.8% (0.02) 0.37 0.44 Fort Lauderdale 118 2,710 2,687 -0.8% 0.50 0.07 (0.39) Fort Myers 119 2,261 2,328 3.0% (0.64) (0.71) (0.16) Gainesville 120 2,744 2,989 9.0% 0.59 0.73 0.20 Hudson 122 2,533 2,431 -4.0% 0.05 (0.49) (0.58) Jacksonville 123 2,765 2,956 6.9% 0.64 0.66 0.08 Lakeland 124 2,272 2,758 21.4% (0.61) 0.23 0.95 Miami 127 3,688 3,905 5.9% 3.00 2.74 0.02 Ocala 129 2,220 2,411 8.6% (0.75) (0.53) 0.18 Orlando 130 2,492 2,900 16.4% (0.05) 0.54 0.65 Ormond Beach 131 2,011 2,428 20.7% (1.28) (0.49) 0.91 Panama City 133 3,030 3,007 -0.8% 1.32 0.77 (0.38) Pensacola 134 2,359 2,491 5.6% (0.39) (0.35) 0.00 Sarasota 137 2,128 1,720 -19.2% (0.98) (2.04) (1.49) St. Petersburg 139 2,386 2,707 13.5% (0.32) 0.12 0.47 Tallahassee 140 2,154 2,351 9.2% (0.91) (0.66) 0.21 Tampa 141 2,617 2,756 5.3% 0.27 0.22 (0.02) Adjusted State Average 2,578 2,743 6.4% Unadjusted State Mean Avg 2,512 2,653 5.6% Standard Deviation 392 458 16.6% Coefficient of Variation 0.156 0.172 10.4% Source: http://www.dartmouthatlas.org/ 187 Reimbursement per Enrollee Reimbursement per Enrollee Appendix 2 Medicare Reimbursements for Inpatient Short Stays per Enrollee, Florida HRR's, 2000-2005

PAGE 202

Appendix 2 Medicare Reimbursements for Inpatient Short Stays per Enrollee, Florida HRR's, 2000-2005 Source: http://www.dartmouthatlas.org/ 188 25.0% 20.0% 15.0% 10.0% 5.0% 0.0% 5.0% 10.0% 15.0% 20.0% 25.0% Bradenton Clearwater Fort Lauderdale Fort Myers Gainesville Hudson Jacksonville Lakeland Miami Ocala Orlando Ormond Beach Panama City Pensacola Sarasota St. Petersburg Tallahassee Tampa Medicare Reimbursements for Inpatient Short Stays per Enrollee Percentage Change, 2000 2005 The five year percentage change between 2000 & 2005 for Medicare reimbursements for inpatient short term stays per enrollee show substantial fluctuation between Hospital Referral Regions. Growth is higher than 20% in Lakeland and Ormond Beach. St Pete rsb urg and Clearwater show close to 15%. Sarasota shows a decline of almost 20%. Hudson also shows a decline but by less than 5%. Lakeland & Ormond Beach show percentage growth in excess of 20% between 2000 & 2005

PAGE 203

Appendix 2 Medicare Reimbursements for Inpatient Short Stays per Enrollee, Florida HRR's, 2000-2005 Source: http://www.dartmouthatlas.org/ 189 (1.50) (1.00) (0.50) 0.00 0.50 1.00 1.50 2.00 2.50 3.00 3.50 Bradenton Clearwater Fort Lauderdale Fort Myers Gainesville Hudson Jacksonville Lakeland Miami Ocala Orlando Ormond Beach Panama City Pensacola Sarasota St. Petersburg Tallahassee Tampa Medicare Reimbursements for Inpatient Short Stays per Enrollee Z Scores per HRR, 2000 This chart shows Medicare reimbursements for short term hospital stays for each HRR in Florida converted to z scores, 2000. The chart shows that Miami is 3 standard deviations above the mean with Panama City almost 1.5 standard deviations. Fort Lauderdale, Gainesv ill e, and Jacksonville are each around 0.50 standard deviations above the mean. Ormond Beach is close to 1.5 standard deviations below th e mean with Sarasota at almost 1 standard deviation below. The chart shows much fluctuation between Florida HRR's with respect to short ter m inpatient Medicare reimbursement per enrollee in 2000. Miami is 3 and Panama City is almost 1.5 standard deviations above the mean.

PAGE 204

Appendix 2 Medicare Reimbursements for Inpatient Short Stays per Enrollee, Florida HRR's, 2000-2005 Source: http://www.dartmouthatlas.org/ 190 (3.00) (2.00) (1.00) 0.00 1.00 2.00 3.00 Bradenton Clearwater Fort Lauderdale Fort Myers Gainesville Hudson Jacksonville Lakeland Miami Ocala Orlando Ormond Beach Panama City Pensacola Sarasota St. Petersburg Tallahassee Tampa Medicare Reimbursements for Inpatient Short Stays per Enrollee Z Scores per HRR, 2005 This chart shows Medicare reimbursements for short term hospital stays for each HRR in Florida converted to z scores, 2005. Jus t like the 2000 chart Miami is close to 3 standard deviations above the mean with Panama City almost 1.0 standard deviations ab ove Sarasota is more than 2 standard deviations below the mean. Although there is still wide fluctuation in the 2005 chart, it d oes n't show the extent of variation as was seen in the 2000 chart. Miami is again shown to have the highest standard deviation above the mean. Sarasota shows that it is a little more than 2 standard deviations below the mean. Source: http://www.dartmouthatlas.org

PAGE 205

TATL CABG% PCI% Valve Repl.% Fusion% Hip & Femur% Joint & Limb% Knee Proc% Mastec tomy% Prostate Proc% TATL 1 0.21904 0.22828 0.20818 0.25655 0.20145 0.19437 0.10465 0.16364 0.22529 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 0.0018 <.0001 <.0001 CABG% 0.21904 1 0.68367 0.93267 0.26449 0.04595 0.13811 0.025 0.14222 0.10612 <.0001 <.0001 <.0001 <.0001 0.1706 <.0001 0.4561 <.0001 0.0015 PCI% 0.22828 0.68367 1 0.65751 0.26299 0.0348 0.13333 0.1094 0.07686 0.06435 <.0001 <.0001 <.0001 <.0001 0.2994 <.0001 0.0011 0.0218 0.0548 Valve Repl.% 0.20818 0.93267 0.65751 1 0.25317 0.04819 0.14192 0.03011 0.14089 0.10773 <.0001 <.0001 <.0001 <.0001 0.1506 <.0001 0.3694 <.0001 0.0013 Fusion% 0.25655 0.26449 0.26299 0.25317 1 0.39632 0.50064 0.27941 0.33878 0.29032 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 Hip & Femur% 0.20145 0.04595 0.0348 0.04819 0.39632 1 0.75331 0.28689 0.51225 0.42486 <.0001 0.1706 0.2994 0.1506 <.0001 <.0001 <.0001 <.0001 <.0001 Joint & Limb% 0.19437 0.13811 0.13333 0.14192 0.50064 0.75331 1 0.36702 0.42016 0.41234 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 Knee Proc% 0.10465 0.025 0.1094 0.03011 0.27941 0.28689 0.36702 1 0.08723 0.07706 0.0018 0.4561 0.0011 0.3694 <.0001 <.0001 <.0001 0.0092 0.0214 Mastectomy% 0.16364 0.14222 0.07686 0.14089 0.33878 0.51225 0.42016 0.08723 1 0.46431 <.0001 <.0001 0.0218 <.0001 <.0001 <.0001 <.0001 0.0092 <.0001 Prostate Proc% 0.22529 0.10612 0.06435 0.10773 0.29032 0.42486 0.41234 0.07706 0.46431 1 <.0001 0.0015 0.0548 0.0013 <.0001 <.0001 <.0001 0.0214 <.0001 191 Appendix 3 Correlation Matrices, TATL Ratio & Selected Discretionary Procedures Pearson Correlation Coefficients, N = 891 (Prob > | r | under H0: Rho=0) High Pearson correlation coefficients are seen between valve replacement and CABG at 0.93. A high correlation is also observ ed between valve replacement and PCI at 0.66 and between CABG and PCI at 0.68. High Pearson correlation coefficients are also observed betwe en hip & femur and joint and limb at 0.75 and between joint & limb and fusion at 0.50.

PAGE 206

TATL CABG% PCI% Valve Repl.% Fusion% Hip & Femur% Joint & Limb% Knee Proc% Mastectomy% Prostate Proc% TATL 1 0.22896 0.2186 0.22776 0.25362 0.18534 0.24102 0.21276 0.14491 0.23637 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 CABG% 0.22896 1 0.88233 0.95561 0.44184 -0.01802 0.18476 0.15605 0.13957 0.19681 <.0001 <.0001 <.0001 <.0001 0.591 <.0001 <.0001 <.0001 <.0001 PCI% 0.2186 0.88233 1 0.87223 0.5214 0.05011 0.25493 0.21158 0.17048 0.20759 <.0001 <.0001 <.0001 <.0001 0.135 <.0001 <.0001 <.0001 <.0001 Valve Repl.% 0.22776 0.95561 0.87223 1 0.45688 -0.00474 0.19277 0.15479 0.15283 0.20332 <.0001 <.0001 <.0001 <.0001 0.8877 <.0001 <.0001 <.0001 <.0001 Fusion% 0.25362 0.44184 0.5214 0.45688 1 0.4336 0.62188 0.42865 0.39627 0.42466 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 Hip & Femur% 0.18534 -0.01802 0.05011 -0.00474 0.4336 1 0.80528 0.46811 0.54058 0.5056 <.0001 0.591 0.135 0.8877 <.0001 <.0001 <.0001 <.0001 <.0001 Joint & Limb% 0.24102 0.18476 0.25493 0.19277 0.62188 0.80528 1 0.55669 0.5752 0.5526 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 Knee Proc% 0.21276 0.15605 0.21158 0.15479 0.42865 0.46811 0.55669 1 0.32861 0.34445 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 Mastectomy% 0.14491 0.13957 0.17048 0.15283 0.39627 0.54058 0.5752 0.32861 1 0.56564 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 Prostate Proc% 0.23637 0.19681 0.20759 0.20332 0.42466 0.5056 0.5526 0.34445 0.56564 1 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 192 Correlation Matrices, TATL Ratio & Selected Discretionary Procedures Spearman Correlation Coefficients, N = 891 (Prob > | r | under H0: Rho=0) Appendix 3 High Spearman correlation coefficients are shown between valve replacement and CABG at 0.96. A high Spearman coefficient is als o shown between valve replacement and PCI at 0.87 and between CABG and PCI at 0.88. High Pearson correlation coefficients are al so observed between hip & femur and joint and limb at 0.81 and between joint & limb and fusion at 0.62.

PAGE 207

Quartile 2000 2001 2002 2003 2004 2005 1 $35,200 $36,950 $38,050 $39,100 $41,000 $41,000 2 $42,600 $44,100 $46,300 $46,600 $48,600 $48,650 3 $48,500 $50,550 $52,650 $52,600 $54,300 $55,000 4 $59,100 $65,000 $69,800 $61,400 $63,300 $63,300 5/8Q $45,550 $47,325 $49,475 $49,600 $51,450 $51,825 Mean $42,034 $43,846 $45,716 $45,500 $47,397 $48,011 St Dev $7,672 $8,284 $8,910 $7,548 $7,543 $7,594 193 Source: U.S. Dept. of Housing & Urban Development Appendix 4 HUD Prescribed Median Income Levels by Quartiles for Total (67) Florida Counties Shown above are quartiles for the distribution of median income limits for all counties in Florida for each year, 2000 2005. Quartiles listed include 1st 4th quartiles and also the 5/8 quartile which is half the distance between the 2nd & 3rd quartile. The 5/8 quartile is listed in addition to the other four quartiles because it is used as a breakpoint in the analysis. Geographic Location in Part 1 was operationalized by using median income per Florida County as a proxy. In model development, plant location was used as a categorical variable with the breakpoint defined at 5/8Q. Florida county median income levels, 2000 2005, were obtained from the U.S. Dept. of Housing & Urban Development (www.hud.gov).

PAGE 208

Quartil e 2000 2001 2002 2003 2004 2005 1 48,543 55,095 66,257 72,440 87,090 98,255 2 63,491 72,415 90,931 105,210 124,680 135,127 3 87,247 99,596 124,383 149,477 173,892 180,904 4 356,229 783,196 573,612 648,020 731,168 913,265 5/8Q 75,369 86,006 107,657 127,343 149,286 158,015 Mean 74,143 84,343 104,781 121,816 141,848 153,562 194 Appendix 5 Total Medicare HMO Patient Charges, Valve Replacement & CABG Procedures, Mean Charges by Quartile, 2000-2005 Total charges was used as both a continuous and categorical variable. For categorical analysis, total charges was converted to a dichotomy. The breakpoint to define high vs. low total charges was at 5/8Q. A special provision was used for Total Charges in categorical analysis because charges increase appreciably each year. To adjust for these increases the 5/8Q breakpoint was annualized, where the 5/8Q breakpoint equals $75,369 for year 2000, 86,006 in 2001, 107,657 in 2002, etc.

PAGE 209

Sector Inspection Region North 1 2 3 4 Central 5 6 7 South 8 9 10 11 Source: http://ahcaxnet.fdhc.state.fl.us/nhcguide/static/state_inspection.html 195 Indian River, Martin, Okeechobee, Palm Beach, St. Lucie Broward Dade, Monroe Baker, Clay, Duval, Flagler, Nassau, St. Johns, Volusia Pasco, Pinellas Hardee, Highlands, Hillsborough, Manatee, Polk Brevard, Orange, Osceola, Seminole Charlotte, Collier, De Soto, Glades, Hendry, Lee, Sarasota Escambia, Okaloosa, Santa Rosa, Walton Bay, Calhoun, Franklin, Gadsden, Gulf, Holms, Jackson, Jefferson, Leon, Liberty, Madison, Taylor, Wakulla, Washington Alachua, Bradford, Citrus, Columbia, Dixie, Gilchrist, Hamilton, Hernando, Lafayette, Lake, Levy, Marion, Putnam, Sumter, Suwannee, Union Florida Counties Appendix 6 AHCA 11 Inspection Regions by Florida Geographic Sector The north geographic sector is comprised of inspection regions 1 4 which are comprised of 41 Florida counties. Inspection areas 5 7 comprise the central geographic sector which contains 11 Florida counties. The south sector is comprised of inspection areas 8 11 and include 15 Florida counties.

PAGE 210

0 1 X 2 P 3 C 4 R 5 XP 6 XC 7 XR Variables Intercept Ownership Plant Location Z-Cardiac Grp Z-Orth Grp Ownership x Plant Location Ownership x Z-Cardiac Grp Ownership x Z-Orth Grp Parameter Estimates 2.03691 2.76579 0.09568 0.000404 -0.077899 2.23255 1.14022 2.33852 0 Intercept 0.0647 -0.0647 -0.06683 0.006073 0.007504 0.06683 -0.00607 -0.0075 1 X Ownership -0.0647 0.1475 0.06683 -0.006073 -0.007504 -0.15118 0.00708 0.01687 2 P Plant Location -0.06683 0.06683 0.11686 -0.00629 -0.014402 -0.11686 0.00629 0.0144 3 C Z-Cardiac Group 0.00607 -0.00607 -0.00629 0.042503 -0.004756 0.00629 -0.0425 0.00476 4 R Z-Orth Group 0.0075 -0.0075 -0.0144 -0.004756 0.027478 0.0144 0.00476 -0.02748 5 XP Ownership x Plant Location 0.06683 -0.15118 -0.11686 0.00629 0.014402 0.25943 -0.01098 -0.02836 6 XC Ownership x Z-Cardiac Group -0.00607 0.00708 0.00629 -0.042503 0.004756 -0.01098 0.06626 -0.00953 7 XR Ownership x Z-Orth Group -0.0075 0.01687 0.0144 0.004756 -0.027478 -0.02836 -0.00953 0.06524 Marginal Effects & Standard Errors for Orthopedic Group Procedures x Ownership Orthopedic Group Procedures when Ownership = 1 4 when X = 1) = -0.077899 + X(2.33852) = 2.2606 4 ) + X 2 7 4 7 X)] 1/2 2 (0.06524) + 2X(-0.027478) = 0.1943 t = 2.2606 / 0.1943 = 11.6332 196 Appendix 7 Variance-Covariance Matrix Model 1-B

PAGE 211

Appendix 7 Variance-Covariance Matrix Model 1-B Marginal Effects & Standard Errors for Cardiac Group Procedures x Ownership Cardiac Group Procedures when Ownership = 1 3 when X = 1) = -0.000404 + X(1.14022) = 1.1406 3 ) + X 2 6 3 6 X)] 1/2 2 (0.06626) + 2X(-0.042503) = 0.1541 t = 1.1406 / 0.1541 = 7.4003 Marginal Effects & Standard Errors for Geographic Location x Ownership Plant Location when Ownership = 1 2 when X = 1) = 0.09568 + X(2.2326) = 2.3283 2 ) + X 2 5 2 5 X)] 1/2 2 (0.25943) + 2X(-0.11686) = 0.3776 t = 2.3283 / 0.3776 = 6.166 197

PAGE 212

Proportion of Variation No. Eigen value Condition Index Ownership Plant Location Z-CGP* Z-OGP* Ownership x Plant Location Ownership x Z-CGP* Ownership x Z-OGP* 1 2.27838 1 0.01059 0.02675 0.02887 0.0327 0.02124 0.02737 0.03349 2 1.64416 1.17717 0.07403 0.01788 0.02302 0.01423 0.05671 0.02454 0.015 3 1.44932 1.25381 0.01042 0.01248 0.04609 0.10339 0.00030739 0.04872 0.09102 4 0.91108 1.58138 0.11975 0.34395 0.00053634 0.01353 0.00028198 2.57E-07 0.05398 5 0.37885 2.45232 0.05096 0.00115 0.03311 0.61607 0.03346 0.02533 0.57571 6 0.1876 3.48496 0.01665 0.01478 0.79894 0.13953 0.02604 0.80264 0.13621 7 0.15061 3.8894 0.7176 0.58301 0.06943 0.08056 0.86196 0.07141 0.09458 198 Appendix 8 Part 1, Model 1-B Collinearity Diagnostics (Intercept Adjusted) Eigen values close to zero signify serious problems of multicollinearity. But no Eigen values shown in this matrix appear to be close to zero. As a rule of thumb a condition index > 30 indicates serious problems of multicollinearity. None of the condition index values are even above 3.9. *CGP: Cardiac Group Procedures *OGP: Orthopedic Group Procedures

PAGE 213

199 Appendix 9 Technical Appendix Circumventing a Possible Specification Error Supplement to Part 1, Models 1 A & 1 B Introduction In the Regression Diagnostics section under Part 1 (Chapter 7) specifically under the discussion on Specification Error (page 134) a reference is made to Appendix 9 This is a technical appendix that is added to the analysis to help avoid the possibility of a specification err or in this project. A specification error occurs in a model that does not include a certain variable(s) when its presence is warranted. Specification errors are often discussed in the context of incorporating a higher order term(s) when there is a curvi linear, as opposed to a linear, association. Higher order terms may also refer to interacting variables that should have been placed in the model Another type of spec ification error m ay occur when certain variables are not included but are theoretica l ly grounded and conceptually relevant to the model. Excluding one or more of these variables may result in a specification error as well as validity One of the objectives in this project has been to formulate parsimonious models; to create efficient models using only a few variables that d irectly contribute to the model and address hypotheses. This objective was carried out in the development of models 1 A and 1 B in Part 1 of the narrative (Chapter 7) But the enforceme nt of objective s for parsimony and model efficiency should not occur at the price of foregoing otherwise valuable information especially when an expanded model may include additional

PAGE 214

200 variables of which theory and prior research has established their assoc iation with the dependent variable. If a specification error occurred in models 1 A and 1 B in Part 1 due to the exclusion of important variables, then a n important question arises in connection with these models : can similar results be obtained with additional, relevant variables entered into the model, or will an expanded model provide new information that may alter conclusions reached earlier in Part 1 ? T his question will be addressed via two additional models, en ti tled Model 1 C and Model 1 D in this appendix. If the inclusion of additional variables results in, for example, lower betas or perhaps even insignificant p values for the association between the total assets/total liabilities ( TATL ) ratio and cardiac and orthopedic group procedures, then it may be concluded that the models in Part 1 are less reliable than initially imagined. On the other hand, if the associations between the TATL ratio and cardiac and orthopedic group procedures are similar in the presen ce of these additional variables, then greater confidence can be placed in these associations, evidenced by both parsimonious and expanded models. Adding this technical appendix which tests Part 1 models using t hree additional and very relevant variables is an important endeavor. These three additional variables will reflect labor costs, hospital occupancy, and market conditions. Without this appendix questions may arise as to whether the analysis was incomplete and if model validity and results could be fully trusted.

PAGE 215

201 Recap of Part 1 Variables Before these three additional variables are presented, it is necessary to provide a recap of the variables that were already used in models 1 A and 1 B in Part 1 of the narrative. These original variables include d the dependent variable TATL ratio, and in dependent variables ownership, geographic location, cardiac group procedures, and orthopedic group procedures. A brief summary of these variables are restated in the next few brief paragraphs. The TATL ratio (total assets total liabilities) is the reciprocal of the debt ratio, ( total liabilities total assets ) Total liabilities include current and long term liabilities and total assets include current and fixed assets as well as other assets such as goodwill if relevant. As the inverse to the debt ratio, t he TATL ratio is an indicator of liquidity vs. leverage and is a measure of both cash flow and balance sheet solvency. Essentially, it is a measure of long term financial health and stability Cardiac group procedures represent a set of DRG codes under M ajor Diagnostic Category (M DC ) 5 and include valve replacements, coronary artery bypass grafts (CABG), and percutan i ous coronary interventions (PCI). Orthopedic group procedures represent DRG codes under MDC 8 and include major joint & li mb reattachment, hip & femur procedures, spinal fusions, and knee procedures A full listing of these procedures with corresponding DRG codes can be found o n Table 6 in Chapter 3 of the narrative Both cardiac and orthopedic group procedure variables are ex pressed as percentages which are calculated for each hospital via cardiac procedures reimbursed by conventional Medicare total procedures including all payers. Percentages of orthopedic

PAGE 216

202 group procedures are calculated in the same fashion for each hospital via orthopedic group procedures reimbursed by conventional Medicare total procedures including all payers. Ownership is a dummy variable to contrast investor owned from not for profit hospitals with not for profits as the base level. Geographic location is a dummy variable to contrast hospitals located in higher income vs. lower income counties with hospitals in lower income counties as the base level. The breakpoint to determine high er income counties is the median income quartile for the median income distribution for all Florida counties. Three Additional Variables As mentioned in the Introduction of this appendix t here are three additional dimensions that could be explored in connection with models in Part 1. These include labor costs, market conditions, and facility sizing. These three dimensions will be operationalized by three different continuous variables. Labor costs will be measured by the variable manhours per adjusted patient day f or each hospital Market conditions will be measured by the He r f i ndahl Hirshman Index (HHI) for HMOs for each county and facility sizing will be measured by hospital occupancy percentage for each hospital These three variables will be discussed more fu lly in the following three subsections. Labor Costs Manhours per Adjusted Patient Day There are lengthy discussions regarding labor costs and the efficient use of resources in hospitals in the literature review in Chapter 2 of the narrative These discussions provide much informatio n on the association between a strong balance sheet, reflected in a high TATL ratio, and labor costs.

PAGE 217

203 Several variables were explored which represent labor costs and include hospital average annual LOS, salary per adjusted admission, full time equivalency (FTE), and salary per FTE. However, the best variable that was found to reflect labor costs was manhours per adjusted patient day. This variable is a ratio represented as follows. M anhours per adjusted patient day = Manhours {( T otal of acute and intensive care patient days) [(Total inpatient revenue Inpatient sub acute revenue) Gross revenue]} Manhours are defined as one work year, which is 2080 hours for one full time employee multiplied by the number of fulltime equivalent employees. Adjusted patient days are the sum of acute patient days and intensive care patient days divided by the ratio of inpatient revenues generated from acute, intensive, ambulatory, and ancillary patient services to gross revenue (AHCA 200 5). It is intuitive that measures of hospital profitability would be inversely related to labor costs, and evidence supporting this supposition is replete in the literature. It was also surmised that investor owned hospitals would have fe wer mean average manhours per adjusted day relative to not for profits An examination of the AHCA financial dataset, 2000 2005, revealed that investor owned hospitals have a mean average of 22.90 manhours per adjusted patient day with a standard deviatio n of 10.98. Not for profits, on the other hand, show a mean average of 26.36 manhours per adjusted patient day with a standard deviation of 10.27. These differences between investor owned and not for profit hospitals were supported by Sear (1991) who found that hospital profitability and solvency are associated with reduced manhours per adjusted patient day. Sear examined 50 investor owned and 60 not for profit hospitals in Florida for the period from 1982 through 1988

PAGE 218

204 The results indicated that vestor owned hospitals used significantly fewer FTE staff per bed, had significantly fewer manhours per adjusted patient day, and paid significantly less in wages and had significantly higher operating margins (profit) than did the not for profit instituti Observing mean differences between investor owned and not for profit hospitals with respect to mean average man the model development phase of the analysis it was conjectured that interaction may exist between ownership and manhours per adjusted patient day. Tests showed that the variable, manhours per adjusted patient day was a good control variable in that it had a strong inverse corre lation with the TATL ratio. However, interaction was not found for ownership x adjusted manhours per patient day. Facility Sizing Patient Occupancy It is advantageous for a hospital to be of optimum size, not to contain more beds than are necessary but just enough beds The nontrivial costs of building, staffing, and maintaining unused beds are unnecessary and are ultimately borne by the health care consumer considerable savings are possible by eliminating beds in an over bedded hospital (saving variable costs) and by preventing the c onstruction of unneeded beds (saving both fixed and variable costs) (Hancock, Magerlein, Storer, et al 1978) A good measure for optimum facility size is hospital occupancy. Even if a hospital performs procedures that are very profitable in terms of DR G reimbursement, profitability will still be enervated if patient occupancy is low.

PAGE 219

205 Patient occupancy is a calculation used to show the actual utilization of an inpatient health facility for a given period of time. of stay are common measures of efficiency which provide a rough measurement of how well hospitals are utilizing their primary resource; bed capacity (Plante 2009). Occupancy rate is available i n the AHCA financial data as an annual adjusted mean percentage of occupancy for licensed staffed acute beds during a reporting period (e.g., one year) (AHCA 2005) The formula for occupancy is shown as follows: N umber of total inpatient service days B ed days available Total inpatient service days (i.e., days of care) are the sum of each daily inpatient census for the reporting period B ed days available are equal to the number of staffed beds multiplied by the number of days in the reporting period. For example, a hospital with 50 beds would have an annual 18,2 50 bed days available (50 x 365) (Pennsylvania D OH 2001) In addition to facility sizing there is another reason why occupancy is important in the development of model s in Part 1 and in this appendix Recall that in the narrative it was discu ssed that investor owned hospitals, on the average, have fewer beds than not for profit hospitals. The AHCA financial dataset, 2000 2005, shows that the mean average for beds in investor owned hospitals is 207 with a standard deviation of 120 and a maximu m of 531. In contrast, not for profits, which included religious and other not for profit and hospital district hospitals, showed a mean average of 301 beds with a standard deviation of 230 and a maximum of 955. Since investor owned hospitals have fewer beds, averaging a hundred beds less than not for profits, then b y performing higher rates of certain discretionary procedures,

PAGE 220

206 certain investor owned hospitals not only fill their beds with more profitable DRG codes, but also use fewer beds for less profitable codes. What was surprising in examining the AHCA financial data was that not for profits showed slightly higher occupancy rates than investor owned hospitals. It was expected that the mean occupancy rate in investor owned hospitals would exce ed that of not for profits. The mean occupancy rate for all hospitals in the study, 2000 2005, was 46.19 percent while not for profits showed 47.10 percent, and investor owned hospitals showed only 44.83 percent. The difference in occupancy rate changes w hen the analysis is stratified not only by ownership, but also by TATL ratio levels. Recall that in the narrative the TATL ratio distribution was segmented into quartiles. When occupancy is reviewed to only include investor owned hospitals with TATL rati os equal to or exceeding the TATL ratio quartile then mean occupancy jumps up to 49.27 percent. Occupancy was then examined to include only investor owned hospitals with TATL ratios equal to or exceeding the third TATL ratio quartile and the mean rate jumped again to 50.58 percent. Based on a review of occupancy rates stratified by ownership and TATL ratio it was determined that interaction exist ed between ownership and occupancy in their association with the dependent variable This is discussed furt her in the Preliminary Analyses and Model Development sections in this appendix. Occupancy, as reported in the AHCA financial data, 2000 2005, was determined to be a useful control variable with a positive association between it and the TATL ratio.

PAGE 221

207 M arke t Con centration HMO Herf i ndahl Hirschman Index (HHI) The models presented in this technical appendix attempt to adjust for market conditions by incorporating an additional variable, the HMO HHI. Market conditions in the context of HMO markets refer to the number of health plans competing with each other in a geographic area Fewer available health plans reduce flexibility and negotiating leverage, lead ing to lower reimbursement, while a broad selection of plans gives a hospital a bargaining advantage. As will be discussed subsequent to the Model Development section, t he inclusion of the HMO HHI is only one aspect of market conditions. In future research there should also be consideration in using the h ospital HHI in addition to th e HMO HHI A Brief Discussion of the HHI Before proceeding with the analysis, a brief discussion o f the HHI is presented. The HHI attributed to economists Orris C. Herfindahl and A lbert O. Hirschman has been applied in research on market competition, technology management, and antitrust law ( Szpiro 1987, The Brandes Institute 2004) The H HI is one of the most commonly used indicators to detect anticompetitive behavior in indust ries. In fact, an increase in the value of the index is usually interpreted as an indicator of actions which may lessen co m petition or even create a monopoly (Matsumoto, Merlone, Sxidarovsky) A low HHI indicates that market concentration is low which indicates vigorous competition reflecting that a large number of firms are operating with small market shares. In contrast, a high HHI indicates greater market concentration where economic output is carried out by a small number of firms resulting in weak competition (Rhoades 1993).

PAGE 222

208 The HHI has been used in previous research in measuring horizontal integration of healthcare markets. Horizontal integration often raises antitrust concerns because a newly combined firm will have a larger market share than either firm did before merging (Whinston 2006) From an economic and sometimes legal perspective, a high degree of concentration reflecting a lack of competition may be evidence of concerns In general th e HHI can be a warning signal of possible antitrust problems as well as the emergence of a monopoly or oligopolies within an industry (Zimmerman 2009). For this reason the HHI is used by the federal government to monitor the effects on markets due to merg Transactions that increase the HHI by more than 100 points in concentrated markets presumptively raise antitrust concerns under the Horizontal Merger Guidelines issued by the U.S. Department of Justice and the Federal Trade Commissio (www.justice.gov). Calculating the HHI The HHI is written as follows: N H HI = (M S i ) 2 i = 1 M S is the market share of firm i in the market, and N is the number of firms (Rhodes 1993) An example of the HHI calculation would be if a market had only two firms, and each had 50 percent market share. The HHI would equal 0.50 2 + 0.50 2 = 0.50 (Pogodzinski 2010, Pilsbury & Meaney 2009) Depending on the preference of the researcher, HHI observations may be expressed as a decimal or as a whole number. When whole numbers are used 10,000

PAGE 223

209 represents the maximum value representing the existence of a monopoly in which one firm has 100 percent of the market, that is, the HHI = (100) 2 = 10,000. Conversely, the HHI takes on a very small value theoretically approaching zero if the market is purely competitive where there a re many firms with small market shares. For example, in a market with 100 firms where each have a one percent share of the market the HHI = (1 1 ) + (1 1 ) . (1 1 00 ) = 100 (Rhoades 1993). When expressed as a decimal the HHI ranges from 0 to 1.0 representin g a continuum of market concentration which progresses from a large number of very small firms to a single monopolistic firm. As the HHI approaches zero it is an indication of perfect competition, and as an HHI approaches 1.0 it indicates market dominatio n by an oligopoly or monopoly. Increases in the HHI indicate an increase in market power and a decrease in competition. Conversely, a decrease in the HHI represents decreasing market power and an increase in competition (Pogodzinski 2010). Market conce ntration is often represented by HHI intervals. It is usually considered that an HHI below 0.01 (or 100 when using whole numbers ) represents a highly competitive market. An HHI below 0.1 (1,000) represents a non concentrated index. An HHI between 0.1 an d 0.18 (1,000 to 1,800) indicates moderate concentration, and an HHI above 0.18 ( > 1,800) indicates high market concentration (Pogodzinski 2010, National Grid 2009) and perhaps market domination by a monopoly or oligopoly (Powell 2007 U.S. D OJ & Fed eral Trade Comm ission 1997)

PAGE 224

210 The HHI and Healthcare Markets Th e interpretation of the HHI discussed above, is not totally consistent with hospital and HMO markets. H ealthcare specifically HMO, markets tend to show higher HHIs per city or county. For example, an examination of the HMO HHI data for all counties in Florida, 2000 2004, reveals a mean of 0.29 82 (or 2, 982 ) This is consistent with average HHIs in many other parts of the nation. Table 1 provides a breakdown of HMO/PPO HHIs for 2008 and Hospital Discharge HHIs for 2007 by several cities and states across the country. Table 1 A Delineation of HMO/PPO HHI, 2008 & Hospital System Discharge HHI, 2007 A Survey of Various Cities Across the Na tion City/State HMO/PPO HHI 2008 Hospital System Discharge HHI 2007 Sarasota, FL 1 734 2 436 Honolulu, HI 6 357 1 349 Boise, ID 3 159 2 740 Grand Rapids, MI 4 299 2 623 Knoxville, TN 2 221 1 483 New Port News, VA 1 844 2 302 Fresno, CA 2 260 1 318 San Francisco, CA 2 024 1 659 Fort Wayne, IN 3 112 2 236 Macon, GA 5 507 1 555 Lafayette, LA 4 045 1 014 Beaumont, TX 3 289 2 884 Corpus Christi, TX 2 678 3 226 Hospital Care: High Value for All, National Business Group on Health As shown in Table 1 none of the HMO/PPO HHIs listed show an HHI less than 1,800 a threshold which denotes a monopoly or oligopoly in many other industries. Healthcare markets behave somewhat differently than many other markets

PAGE 225

211 As already mentioned, the HHI is useful as a control variable because it reflects healthcare market concentration the tendency of a market to be dominated (the extent to which price and output are controlled) by a few large firms. In this case the HMO HHI is a useful control variable reflecting market penetration of HMO plans and their tendency of controlling contract terms and reimbursement levels to hospi tals. Therefore, i t was anticipated that there would be a n inverse association between TATL ratio and market concentration. Although an inverse association between the TATL ratio and the HMO HHI was anticipated, a positive albeit weak, assoc iation is what was found This anomaly was investigated and reasons why this unexpected association may have occurred is further discussed following the Model Development section. Da ta Integrity : Three Caveats Before proceeding with model development a few caveats regarding the dataset should be disclosed. First, an HMO HHI was available for each county in Florida, but only for years 2000 2004; data for 2005 was unavailable. But keep in mind that there is not considerable variation by county in the HHI from year to year. S ince the HMO HHI is not The second caveat is that there are three missing values for hospital occupancy. Data entry for this variable was done manually by the author and occupancy could be identified for all but three hospitals. This means that the dataset is reduced from 891 to 88 8 observations. Considering that occupancy is an important variable, and since l ost

PAGE 226

212 data only represents three observations it was determined to keep occupancy in the model. Preliminary Analyses Descriptive Statistics The preliminary analys i s begin s with a presentation of some descriptive statistics shown in Table 2, for the three additional variables manhours per adjusted patient day, occupancy, and HMO HHI Certain types of descriptive statistics have already been provided regarding these three variables, but they are aggregated below so they can be viewed simultan eously in one table. These statistics are presented for total hospitals and hospitals stratified by ownership. Table 2 Descriptive Statistics for Manhours per Adjusted Patient Day, Occupancy & HMO HHI Total, Investor Owned & Not for Profit Hospitals, 2000 2005 Variables All Hospitals Investor Owned Not for Profits Mean Std Dev Mean Std Dev Mean Std Dev Man hours per Adjusted Patient Day 24.84 10.70 22.9 6 10.9 3 26.36 10.27 Occupancy 0. 4619 0. 25 92 0. 44 94 0. 2642 0. 471 9 0. 2549 HMO HHI 0.29 82 0.2 191 0.32 01 0.2 409 0.2805 0.1983 An important element in this table is the standard deviation for each variable. As will be discussed continuous variables in these models are expressed as z scores. Using the association between the TATL ratio and occupancy percentage as an example a change in the TATL ratio would not correspond to a one percent change in occupancy. Instead, correct interpretation would be a change in the TATL ratio corresponding to a change in occupancy by one standard deviation 0.2592

PAGE 227

213 Preliminary Tests A number of preliminary tests many similar to those performed in Part 1 of the na rrative, were completed for the analysis presented in this appendix. One of these tests included the calculation of correlation coefficients for the TATL ratio, HMO HHI, occupancy, and manhours per adjusted day which is shown in tables 3 & 4 These tables provide measures of association using the Pearson correlation test for linear dependence and the Spearman test of ordinal rankings. Since other variables used in this appendix, such as cardiac and orthopedic group procedures, were already tested und er Part 1 of the narrative, they are not presented in tables 3 & 4 Tests of Association Using Pearson & Spearman Correlation Coefficients Table 3 presents a correlation matrix using Pearson coefficients for the TATL ratio, HMO HHI, occupancy, and manhou rs per adjusted patient day. Tabl e 3 Pearson Correlation Coefficients for TATL, HMO HHI, Occupancy & Manhours per Adj. Day All Hospitals, 2000 2005 Variables TATL HMO HHI Occupancy Manhours per Adj. Day TATL 1 0.07042 0.06643 0.14506 ( 0.0356 ) ( 0.0477 ) ( <.0001 ) HMO HHI 0.07 485 1 0.13257 0.07718 ( 0.0 255 ) ( <.0001 ) ( 0.0213 ) Occupancy 0.06643 0.13257 1 0.0624 ( 0.0477 ) ( <.0001 ) ( 0.0631 ) Manhours per Adj. Day 0.14506 0.07718 0.0624 1 ( <.0001 ) ( 0.0213 ) ( 0.0631 ) As shown in Table 3 all correlation coefficients between the TATL ratio and HMO HHI, occupancy, and manhours per adjusted patient day are significant (p < 0.05)

PAGE 228

214 but not very high. The highest negative coefficient, demonstrating an inverse association, is between the TATL ratio and manhours per adjusted day at 0.145 (p < 0.0001). It was anticipated that a negative correlation coefficient reflecting an inverse association, would be found between the TATL ratio and HMO HHI, but the table shows a sign ificantly positive, albeit very low coefficient. There is also a n expected negative coefficient shown between the HMO HHI and occupancy at 0.133 (p < 0.0001). A high negative coefficient indicating an inverse association, was expected to be found betw een occupancy and manhours per adjusted day. The supposition was that higher occupancy would result in greate r efficiency in labor costs. But the correlation coefficient between occupancy and manhours per adjusted patient day was non significant (p = 0.0631) Table 4 shows the same analysis as in Table 3 but using Spearman correlation coefficients. Table 4 Spearman Correlation Coefficients for TATL, HMO HHI, Occupancy & Manhours per Adj. Day Total Hospitals, 2000 2005 Variables TATL HMO HHI Occupancy Manhours per Adj. Day TATL 1 0.03063 0.12737 0.26856 ( 0.361 ) ( 0.0001 ) ( <.0001 ) HMO HHI 0.03 161 1 0.13975 0.0961 ( 0.3 460 ) ( <.0001 ) ( 0.0041 ) Occupancy 0.12737 0.13975 1 0.09831 (< 0.0001 ) ( <.0001 ) ( 0.0034 ) Manhours per Adj. Day 0.26856 0.0961 0.09831 1 ( <.0001 ) ( 0.0041 ) ( 0.0034 ) The Spearman coefficient is higher than the Pearson for the associations between the TATL ratio and occupancy at 0.127 (p < 0.0001). A high er negative coefficient is

PAGE 229

215 found between the TATL ratio and manhours per adjusted patient day at 0.269 (p < 0.0001). An anticipated inverse association between the TATL ratio and the HMO HHI was again left unsubstantiated with a positive non significant coefficient of 0.03 16 (p = 0.3 460 ). A n inverse associa tion was fo u nd again as it was using the Pearson coefficient, between the HMO HHI and occupancy with a coefficient of 0.139 8 (p < 0.0001). Tests of Association Exclusively for In vestor Owned Hospitals Tests of association, stratified by ownership, were also performed. Tables 5 and 6 provide findings using both the Pearson and Spearman correlation coefficients applied exclusively to investor owned hospitals. Table 5 Pearson Correlation Coefficients for TATL, HMO HHI, Occupancy & Manhours per Adj. Day Exclusively for Investor Owned Hospitals 2000 2005 Variables TATL HMO HHI Occupancy Manhours per Adj. Day TATL 1 0.0243 0.1571 0.1147 ( 0.6295 ) ( 0.0017 ) ( 0.0225 ) HMO HHI 0.0 3 4 74 1 0.1267 0.1725 ( 0. 4901 ) ( 0.0116 ) ( 0.0006 ) Occupancy 0.1571 0.1267 1 0.1209 ( 0.0017 ) ( 0.0116 ) ( 0.0162 ) Manhours per Adj. Day 0.1147 0.1725 0.1209 1 ( 0.0225 ) ( 0.0006 ) ( 0.0162 ) Pearson coefficients show a non significant association between the TATL ratio and the HMO HHI when the analysis is stratified to include only investor owned hospitals. T he coefficient between the TATL ratio and occupancy is significantly positive with a coefficient of 0.157 1, and the coefficient between the TATL ratio and manhours per adjusted patient day is significantly negative, 0.1147 (p = 0.0225) reflecting an inverse association.

PAGE 230

216 The matrix also shows an expected negative coefficient of 0.1267 (p = 0.0116) between manhours per adjusted patient day and occupancy. This reflects that as occupancy increases so does efficiency in the use of personnel. There is also a positive coefficient between the HMO HHI and manhours per adjusted patient day This reflects that as market concentration increases labor efficiency, in turn, decreases. Table 6 now shows Spearman rank order correlation coefficients exclusively for investor owned hospitals. Table 6 Spearman Correlation Coefficients for TATL, HMO HHI, Occupancy & Manhours per Adj. Day Exclusively for Investor Owned Hospitals, 2000 2005 Variables TATL HMO HHI Occupancy Manhours per Adj. Day TATL 1 0.0191 0.2254 0.2231 ( 0.7046 ) ( 0.0001 ) ( 0.0001 ) HMO HHI 0.0191 1 0.1518 0.2474 ( 0.7046 ) ( 0.0025 ) ( 0.0001 ) Occupancy 0.2254 0.1518 1 0.2532 ( 0.0001 ) ( 0.0025 ) ( 0.0001 ) Manhours per Adj. Day 0.2231 0.2474 0.2532 1 ( 0.0001 ) ( 0.0001 ) ( 0.0001 ) The findings in Table 6 using Spearman coefficient s are consistent with respect to the association between t he TATL ratio and occupancy, except the Spearman coefficient shows greater magnitude between the two variables with a coefficient of 0.2254 (p < 0.0001). The association between the TATL ratio and m anhours per adjusted patient day also shows greater magnitude using the Spearman coefficient at 0.2231 (p < 0.0001). The coefficient between the TATL ratio and the HMO HHI was found to be negative indicating the anticipated inverse association, but with a high p value (0.7046).

PAGE 231

217 Because of a large magnitude of difference in association between the TATL ratio and occupancy when stratified by ownership interaction was tested and found between these two variables. This is discussed further in the Model Deve lopment section. Assessment of Control Variables Manhours per adjusted patient day performed best in the preliminary analyses in its inverse association with the TATL ratio. This was demonstrated by Pearson and Spearman correlation coefficients shown in t ables 3 6 as well as other preliminary tests carried out in preparation of this appendix. These tests also revealed a positive association between the TATL ratio and occupancy, as well as elucidated the presence of interaction between ownership and occu pancy in their association with the TATL ratio. The HMO HHI variable did not perform well in preliminary analyses. Association between the TATL ratio and the HMO HHI depending on the test that was used, was shown to be either non significant or very small. The positive association between profitability or solvency and the HMO HHI which was found using the Pearson coefficient in Table 3, is also in consistent with theory as there should be an inverse association between the TATL ratio and the HMO HHI Lower hospital TATL ratios should correspond to higher HMO HHI concentration. The inability of preliminary tests to show a strong inverse association between the TATL ratio and the HMO HHI does n o t always mean that an association fails to exist It is just not demonstrated in the tests employed perhaps due to anomalies in the data or other reasons Therefore, i t was determined for the HMO HHI to remain in the models

PAGE 232

218 Further discussion will be provided regarding the unanticipated lack of inverse association between the TATL ratio and the HMO HHI following the present ation of the regression models in the Model Development section. Model Development As mentioned earlier, under Part1 in the narrative (Chapter 7) there w ere a main effects and an interaction model, labeled, respectively, Model 1 A and Model 1 B. Because this appendix is technically a continuation of Part 1 the main effects and interaction model will be en titled Model 1 C and Model 1 D. Keep in mind that mod els 1 C and 1 D are expanded models that include three additional variables representing labor costs per hospital patient occupancy per hospital and market concentration per county for all counties in Florida. As discussed in the narrative, multicollinea rity often accompanies interaction terms. Since Model 1 D is an interaction model steps were taken to manage variance inflation which included the transformation of continuous variables to z scores. Although no problems with multicollinearity were antic ipated with the main effects model, 1 C, z scores were also used for continuous variables in this model, so that comparisons could be made readily between models 1 C and 1 D. Now that additional variables have been identified and tested, they can be used i n a main effects model.

PAGE 233

219 Model 1 C : Expanded Main Effects Model Model 1 C is a n expanded main effects model that includes the TATL ratio as the dependent variable and seven independent variables. The regression equation for Model 1 C is written below. TATL = 0 1 X 2 P 3 C 4 R 5 H 6 O 7 M + Where: X = Ownership P = Geographic Location C = Z Cardiac Group Procedures R = Z Orthopedic Group Procedures O = Z Occupancy H = Z H MO HHI M = Z Manhours per A dj Patient Day The results of the regression model 1 C, are shown in Table 7 Table 7 Model 1 C Main Effects Model Source DF Sum of Squares Mean S quare F Value P r > F Model 7 6 020.18 860.02 6 56.88 <.0001 Error 88 0 13,305 15.11 9 Corrected Total 887 19,325 Root MSE 3.888 R 2 0.31 15 Dependent Mean 3.97 Adjusted R 2 0.30 61 Coefficient of Variation 98.747 Table 7 reports a significant F value in connection with Model 1 C, indicating that at least one variable in the model is significant. A coefficient of variation of 98.747 is also reported which is much higher than the ideal coefficient of 10.0.

PAGE 234

220 The model shows an adjusted R 2 of 0.3061. This is slightly higher tha n the adjusted R 2 that was found for the main effects Model 1 A in the narrative But the higher R 2 could be due simply to a greater number of variables in the model. The parameter estimates are now presented for Model 1 C and are shown in Table 8 below Table 8 Model 1 C Expanded Main Effects Model Parameter Estimates Variable Parameter Estimate Standard Error t Value Pr > |t| VIF 0 Intercept 1.517 0.237 6.39 <.0001 0 1 X Ownership 3.839 0.269 14.26 <.0001 1.052 2 P Geographic Location 1.224 0.283 4.32 <.0001 1.148 3 C Z Cardiac Group Procedures 0.839 0.134 6.26 <.0001 1.057 4 R Z Orthopedic Group Procedures 0.789 0.137 5.76 <.0001 1.100 5 H Z H HI HMO 0.424 0.138 3.06 0.0022 1.120 6 O Z Occupancy 0.331 0.133 2.50 0.0126 1.0278 7 M Z Manhours per Ad j Patient Day 0.462 0.134 3.43 0.0006 1.049 All variables in the model are found to be significant and with the exception of the HMO HHI, are found in directions and magnitudes that were anticipated. The VIF on the table indicates no problems with multicollinearity in the model. The parameter estimate for ownership is 3.84 indicating that, on the average, the TATL ratio is 3.84 points higher in investor owned hospitals relative to not for profits. For geographic location it is found that hospitals located in higher income counties have TATL ratios that are 1.2 points higher than those located in lower income counties. Hospitals located in counties with an HMO HHI at 1Z have TATL ratios that are 4/5 of a point higher than those with an HMO HHI at the mean. As already discussed,

PAGE 235

221 th e directional aberration between the TATL ratio and the HMO HHI will be covered more fully subsequent to the Model Development section. Hospitals with annual average percentages of occupancy at 1Z have TATL ratios that are about one third of a point higher than hospitals with percentages of occupancy at the mean (Z = 0) As expected, hospitals with manhours per adjusted day at 1Z show TAT L ratios at almost a half a point below those with manhours per adjusted patient day at the mean (Z = 0) Betas for cardiac group and orthopedic group procedure percentages show similar results that were shown in Model 1 A under Part 1. In general hosp itals performing cardiac group procedures at 1Z show a TATL ratio of more than 4/5 of a point higher than those that carry out cardiac group procedures at the mean (Z = 0) Likewise, hospitals that perform orthopedic group procedures at 1Z show in general TATL ratios that are more than three quarters of a point higher than hospitals that carry out these procedures at the mean (Z = 0). In addition to the preliminary tests, the main effects model further demonstrated the statistical usefulness of the variables employed. The analysis now proc eeds to the interaction model which uses these same variables. Model 1 D : Expanded Interaction Model The presence of interaction was t ested in Model 1 C for all variables in the model, but was found only for ownership x geographic location, ownership x z c ardiac g roup p rocedures, ownership x z o rthopedic g roup p rocedures, and ownership x z o ccupancy.

PAGE 236

222 T he regression equation is written as follows: Regression Equation for Model 1 D TATL = 0 1 X 2 P 3 C 4 R 5 H 6 O 7 M 8 X P 9 X C 10 X R 11 X O + Where: X = Ownership P = Geographic Location C = Z Cardiac Group Procedures R = Z Orthopedic Group Procedures O = Z Occupancy H = Z HMO HHI M = Z Manhours per A dj P atient D ay The results of the interaction model are shown below in Table 9. Table 9 Model 1 D Interaction Model Source DF Sum of Squares Mean Square F Value Pr > F Model 11 8,265.62 751.42 59.52 <.0001 Error 876 11,059 12.625 Corrected Total 887 19,325 Root MSE 3.553 R 2 0.4277 Dependent Mean 3.938 Adjusted R 2 0.4205 Coefficient of Variation 90.234 Results of Model 1 D show a dependent mean (i.e., the mean of the TATL ratio) of 3.94. The model also shows a high CV at 90.23, which just like the main effects model, is far above the ideal CV of 10.0. Other items listed in Table 9 such as the SSR and SSE will be exam ined later when the PRESS statistic is presented at the conclusion of Model 1 D.

PAGE 237

223 Table 9 shows a higher a djusted R 2 of 0.42 1 which is higher than the a djusted R 2 in the main effects model which was 0. 3 061 Before proceeding to Model 1 significance of the interaction model relative to the main effects model with an F test. The F test was calculated and yield ed a high F value of 46.53 (p < 0.0001) with R 1 2 = 0.3061 and R 2 2 = 0.42 05 ; df 1 = k 2 k 1 (11 7) = 4 and df 2 = N k 2 1 (888 11 1) = 876. This test indicates that the interaction model is significant relative to the main effects model. Now that the F test has indicated that the interaction mod el is significant compared to the main effects model, t he parameter estimates for Model 1 D are shown in Table 10

PAGE 238

224 Table 10 Model 1 D Interaction Model Parameter Estimates Variable Par Est St d. Error t Value Pr > |t| VIF 0 Intercept 1.959 0.254 7.7 <.0001 0 1 X Ownership 2.701 0.382 7.07 <.0001 2.535 2 P Geographic Location 0.370 0.347 1.06 0.2872 2.064 3 C Z Cardiac Group Procedures 0.032 0.205 0.16 0.8746 2.970 4 R Z Orthopedic Group Procedures 0.119 0.164 0.73 0.4673 1.891 5 H Z HMO HHI 0.425 0.127 3.34 0.0009 1.131 6 O Z Occupancy 0.024 0.164 0.15 0.8836 1.887 7 M Z Manhours per Ad j Patient Day 0.255 0.124 2.06 0.04 1.070 8 XP Ownership x Geographic Location 2.124 0.504 4.21 <.0001 3.466 9 XC Ownership x Z Cardiac Group Procedures 1.111 0.256 4.33 <.0001 2.966 10 XR Ownership x Z Orthopedic Group Procedures 2.313 0.253 9.15 <.0001 1.859 11 XO Ownership x Occupancy 0.752 0.241 3.12 0.0019 1.873 VIF is higher in the interaction model relative to the main effects model, but all terms show VIF < 4.0 which is acceptable and far below the general conservative (rule of thumb) threshold for VIF of 5.0 The use of z scores in these models has been very effective in managing problems of multicollinearity. The beta for the HMO HHI is almost unchanged in the interaction model relative to the main effects model. The parameter estimate in the interaction model is 0.425 while it was 0.424 in the main effects model indicating th at hospitals located in counties with HMO HHI at 1Z show TATL ratios that are 0.425 points higher than hospitals located in

PAGE 239

225 counties with HMO HHI at the mean (Z = 0) As it has been discussed already, an inverse association between the TATL ratio and the HMO HHI had been anticipated. Th e unexpected positive association between the TATL ratio and the HMO HHI is more fully discussed following the Model Development section. Although there is almost no change in the HMO HHI, there is considerable change in manhours per adjusted day. The ma in effects model showed a parameter estimate for this variable at 0.462. The interaction model shows a beta of 0.255 Model 1 D: Marginal Effects of Interaction Terms As already mentioned, four interaction terms were found in the model. The marginal effects for each interaction term are shown below. Occupancy was not used as a variable in models 1 A and 1 B. Therefore, it seems fitting to examine the marginal effect of ownership x occupancy fir st and i n greatest detail among the interaction terms in Model 1 D. T he interaction term for ownership x occupancy can be written 0 + 1 X 6 O 11 XO (where O = occupancy and X = ownership) of which the marginal effect may be interpreted O = 6 11 X (i.e., 6 at X ) (Jaccard & Turrisi 2003, Aichen & West 1991). The marginal effect may then be calculated 0.024 + 0.752(X) = 0.776. The focal variable for this interaction term is occupancy, and the modifying variable is ownership. The marginal e ffect is determined by the association between the TATL ratio and occupancy at 1Z (one standard deviation above the mean of z = 0) when ownership (investor owned hospital) is present (i.e., ownership = 1) (Jaccard & Turrisi

PAGE 240

226 2003). Under these conditions it is found that the TATL ratio is 0.7 7 6, indicating that in general investor owned hospitals with occupancy percentages at 1 Z have TATL ratios that are 0.7 7 6 po int s higher than investor owned hospitals with occupancy percentages at the mean (Z = 0). Three other interaction terms were identified in Model 1 D including ownership x geographic location, ownership x cardiac group procedures, and ownership x orthopedic group procedures. Table 11 provides marginal effects for each of th e interaction terms. Table 11 Delineation of Interaction Terms & Marginal Effects in Connection with Model 1 D Interaction Terms Marginal Effects Ownership x Occupancy 1 X 6 O 11 XO 6 11 X 0.024 + 0.752(X) = 0.776 Ownership x Geographic Location 1 X 2 P 8 XP 2 8 X 0.370 + 2.124(X) = 2.494 Ownership x Cardiac Group Procedures 1 X 3 C 9 XC 3 9 X 0.032 + 1.111 (X) = 1.143 Ownership x Orthopedic Group Procedures 1 X 4 R 10 XR 4 10 X 0.119 + 2.313 (X) = 2.194 The interaction between ownership and occupancy has already been discussed. The following paragraphs provide brief discussions on each of the other interacting terms. For each interaction term listed in Table 11 the modifying variable is ownership. The focal variables include geographic location, cardiac group procedures, and orthopedic group procedures

PAGE 241

227 For the interaction term of ownership x geographic location, the focal variable is geographic location. The marginal effect indicates that investor owned hospitals located in higher income counties have TATL ratios that are, on the average, 2.49 points higher than investor owned hospitals in lower income counties. Investor owned hospitals that perform orthopedic group procedures at 1Z have TATL ratios that in general are 2.19 points higher than investor owned hospitals that carry out these procedures at the mean (Z = 0). Likewise, investor owned hospitals that perform cardiac group procedures at 1Z have TATL ratios that in general are 1.14 points hig her than investor owned hospitals that carry out orthopedic group procedures at the mean (Z = 0). Model 1 D: Standard Errors Calculated for Marginal Effects of Interaction Terms Standard errors are now calculated for each marginal effect shown in Table 11 to determine if they are significant. The marginal effect for the interaction term, ownership x occupancy will again be used as an example in applying the standard error formula while results for the remaining interaction terms are shown in Table 12 T he formula to determine the standard error is written belo w. T he variances and covariance used in the formula were extracted from a variance covariance matrix generated in connection with the model.

PAGE 242

228 = [var( 6 ) + X 2 var( 11 ) + 2 Xcov( 6 11 X)] 1/2 =[ 0.0269 + X 2 (0.0580) + 2 X ( 0.0267 )] 1/2 = 0.1778 t = ( 6 at X) / se ( 6 at X) t = 0.7756 / 0. 0.1778 = 4.3619 p < 0.0001 The t value indicates that the interaction term, ownership x z occupancy, is significant. Marginal effects, standard errors, t values and p values are shown in Table 12 Table 12 Marginal Effects & Significance in Connection with Model 1 D Interaction Terms Marginal Effects Standard Errors t Values p Values Ownership x Geographic Location 2.4935 0.3763 6.6269 < 0.0001 Ownership x Z Cardiac Group Procedures 1.1432 0.1533 7.4588 < 0.0001 Ownership x Z Orthopedic Group Procedures 2.1939 0.1930 11.3687 < 0.0001 Ownership x Occupancy 0.7756 0.1778 4.3619 < 0.0001 A high t value and low p value (< 0.0001) are observed for each interaction term used in Model 1 D. This indicates that all interaction terms used in Model 1 D are significant. As will be discussed in the Conclu ding Remarks to this appendix, parameter estimates and marginal effects for original variables, geographic location, cardiac group

PAGE 243

229 procedures, and orthopedic group procedures, showed only minimal changes in models that included the new variables representing labor costs, occupancy, and market concentration. This exercise with the inclusion of additional variables was necessary to assure that all relevant dimensions in association with the dependent variable were represented in the model. Hopefully, as a result of models 1 C and 1 D gre ater confidence can be placed in the validity of Part 1 models ( 1 A 1 D ) and in the findings of this study The Unexpected Positive Association b etween the TATL Ratio and the HMO HHI It has already been explained that an inverse association was anticipated but was not found between the TATL ratio and the HMO HHI variable. Higher concentrations of HMO dominance (i.e., control of price and output) should result in lower profitability for hospitals which should lead to attenuated bargaining levera ge, lower reimbursements, lower profitability, and subs equently a weaker balance sheet reflected by a lower TATL ratio. Conversely, hospitals in areas with lower concentrations of HMO HHI will encounter a much larger selection of health plans. H ospitals that can exercise greater market control will be better able to secure more desirable agreements with HMOs which should result in higher reimbursements. Ultimately, the se hospitals should have stronger balance sheets, represented by a higher TATL ratio. This scenario begs the question, why does the data reveal a positive rather than an inverse association between the TATL ratio and the HMO HHI? This analysis has included only the HMO HHI as a measure of market concentration, but there are more variables than just the HMO HHI that could be applied

PAGE 244

230 to healthcare markets, one of which being the hospital HHI. Therefore, o ne possible explanation why the TATL ratio has a positive association with the HMO HHI is that perhaps in certain counties where the HMO HH I is high the hospital HHI may be high also. If this supposition is accurate, the inclusion of the hospital HHI would be very useful in subsequent research. The re are multiple aspects of market concentration. HMOs have the advantage if health plans are hospital with extensive market share may have greater leverage in negotiating managed care contracts despite a limited selection of plans. Since h ealth care markets tend to be regional if a hospital for example, controls 85% of the acute care capacity in the region, every health plan would be required to include the hospital in its network (Cleverly & Cameron 2002). Even i f only one or a small number of health plans were available in an area, a hospital with a substantial market share may still experience a puissant negotiating advantage. Regression Diagnostics for Model 1 D A full set of regression diagnostic tools used in parts 1 and 2 in the narrative ( c hapter s 7 & 8 ) were a pplied to the main effects and i nteraction models 1 C and 1 D, in this appendix. Outliers Press Statistic As explained in the narrative, a very good summary measure for testing outliers and influential observations is the predicted residual sum of squares (PRESS). The PRESS statistic in connection with Model 1 D is presented in Table 13

PAGE 245

231 Table 13 Sum of Residuals, Sum of Squared Residuals & PRESS for Model 1 D Sum of Residuals 0.0 Sum of Squared Residuals 11, 059 Predicted Residual SS (PRESS) 11, 380 The diagnosis in Table 13 shows that the sum of residuals equals zero which it should. The s um of squared residuals ( SSR ) of 11,059 should also be equal to the Error SS in the regression output for Model 1 D which it is If the PRESS statistic is sub stantially larger than the SSR it would indicate a reason to suspect the existence of certain outliers and influential observations in connection with Model 1 D. H owever, the comparison shows that PRESS is only slightly greater than the SSR. PRESS shows 11,380 while the SSR shows 11,059. This indicates no rationale to be wary of outliers or influential observations (Freund, Littell 2000). In addition to PRESS additional diagnostics were performed to detect potential anomalies which included u nivariate statistics specifically normal probability plots, box plots, and stem leaf plots As a result no potential aberrations were observed. Influence Statistics I nfluence diagnostics were also performed which included hat d iag covariance ratio, DFFITS, and DFBETAS. Just as had been the case with Model 1 B in Part 1 of the

PAGE 246

232 narrative, only a small number of observations were found to exceed prescribed thresholds. Additional testing for outliers began with a revi ew of studentized residuals and 88 )]. There were 47 observations with residuals that exceeded either of these thresholds. All but 13 showed studentized residuals that remained within the Heterogeneous Variances An examination of residuals was carrie d out to assess homoscedasticity. There were no observations of systematic residual patterns, nonrandom residual distributions, recognizable patterns in the magnitudes of the variances, or increases or decreases in variation of larger values of the outcom e variable (Freund, Littell 2000). Multicollinearity Multicollinearity was tested and has already been discussed using VIF for models 1 C and 1 D. Some researchers prescribe a rule of thumb that VIF should not exceed 5.0, while others recommend 10.0 as a VIF threshold VIF corresponding to all parameter estimates in Model 1 D were well below the lower threshold of 5.0 so VIF levels indicated no detection of multicollinearity In addition to VIF both adjusted and non adjusted intercept collinearity diagnostics were performed and there was no evidence of problems with multicollinearity in the models

PAGE 247

233 Specification Errors Additional interaction terms or other higher order terms, such as the inclusion of a quadratic term, were found to be unwarranted in the models. This appendix was added to th e dissertation to help assure that another type of specification error, failure to include other dimensions that may be associated to the TATL ratio, did not occur. This was the rationale for including three additional variables, manhours per adjusted patient day, occupancy, and the HMO HHI. Th is strategy to avoid a specification error in this context is discussed further in Conclu ding Remarks to thi s appendix C oncluding Remarks The pr eceding discussion on regression diagnostics ended with a brief statement regarding testing for specification errors in the models. As stated in the introduction to this appendix specification errors are most often discussed in the context of interaction terms that should have been included or when squared terms should be incorporated in the model because the association is curvilinear as opposed to linear But this appendix has been included to address an additional type of specification error. These types of errors may also occur when certain variables are not included in the model but have a theoretical association and/or ha ve been determined to be associated with the depende nt variable in prior research Models 1 A and 1 B presented in Par t 1 of the narrative (Chapter 7) included the independent variables ownership, geographic location, cardiac group procedures, and orthopedic group procedures. An association was determined between the dependent

PAGE 248

234 variable the TATL ratio and each of the independent variable s Additional variables included in models 1 C and 1 D have also show n an association to the TATL ratio The most important outcome from the models, 1 C and 1 D, in this appendix is that variables were incorporated in the models from several relevant theoretical and prior tested dimensions representing labor costs, occupancy, and market concentration. The inclusion of these variables adds rigor and engenders greater confidence in the findings. W ith the inclusion of these additional variables the betas for cardiac group procedures and orthopedic group procedures changed only slightly. Furthermore, there was little change in the margina l effects of ownership x cardiac group procedures and ownership x orthopedic group procedures. The association between the TATL ratio and cardiac group procedures performed in investor owned hospitals (i.e., when ownership = 1) was almost the same in the presence of additional independent variables HMO HHI, occupancy, and manhours per adjusted day in the model. The same is true for the association between the TATL ratio and orthopedic group procedures when ownership = 1. Findin gs gleaned from expanded models 1 C and 1 D which add three critical variables, buttress the findings in models 1 A and 1 B in the narrative. T his instills greater confidence in the findings from all of the models in connection with Part 1 1 A through 1 D It also enhances credibility in the conclusions drawn from this study.

PAGE 249

235 References to Technical Appendix Agency for Health Care Administration (AHCA) (2005), Financial Analysis Annual Report Bian J. & Morrisey, M.A. (Summer 2006) on, Hospital Competition, Health Care Financing Review Cleverly W.O. & Cameron, A.E. (2002) Essentials of Health Care Finance, Aspen Publishers, Inc., Gaithersburg, Maryland Hancock, W.M., Magerlein, D.B., Storer, R.H. and J.B. Martin (Fall 1978), Health Services Research Huckman, R.S., ( May 2005) Hospital Integration and Vertical Consolidation: an Analysis of Acquisitions in New York State Journal of Health Economics Matsumoto, Merlone, Sxidarovsky (Date Unknown) Some Notes on Applying the Herfindahl Hirschman Index Retrieved May 3, 2010 from http://www2 .tamacc .chuo u.ac.jp /keizaiken/discussno118.pdf Needleman J., Chollet, D.J., Lamphere, J. (March/April 1997) Health Affairs Pennsylvania Department of Health (DOH) (2001) Health Statistics Technical Assistance Tools of the Trade : O ccupancy R ates in H ealth F acilities Plante, C. (January 2009) D ifferentiation between F or P rofit and N onprofit H Research in Healthcare Financial Management Retrieved April 30, 2010 from http://www.allbusiness.com/health care/health care facilities hospitals/13136041 1.html Powell, A.C. (April 2007) , HCMG 903 Presentation Pogodzinski, J.M., Regional Economics, Lecture 3, March 2010. P yenson, B., Kosuke, I., Goldberg, S. et al (March 2010) High Value for All National Business Group on Health Whinston, M. D. ( 2006 ) Lectures on Antitrust Economics The MIT Press, Cambridge, MA Rhoades S. (March 1993) Federal Reserve Bulletin, 188 189

PAGE 250

236 Sear A.M. (Spring 1991) Comparison of E fficiency and P rofitability of I nvestor O wned M ultihospital S ystems with N ot F or P rofit H ospital Health Care Management Review Shen Y C, Wu V. & Melnick, G. (February 2008) Structure: An Analysis of P enetration, Concentration, and Ownership b etween 1994 2005 NBER Working Paper No. 13775 : Bulletin on Aging & Health Szpiro, G.G. (December 1987), Hirschman Versus Herfindahl: Some Topological Properties for the Use of Concentration Indexes, Mathematical Social Sciences, 14, 3, 299 302. The Brandes Institute ( September 2004 ) Concentrated Portfolios: An Examination of Their Characteristics and Effectiveness Brandes Investment Partners, LP. Viscusi, W. K., Harrington Jr. J.E. and Vernon, J.M. ( 2005 ) Economics of Reg u lation and Antitrust The MIT Press, Cambridge, MA U.S. Department of Justice (DOJ) & Federal Trade Commission (1992, rev. 1997), Horizontal Merger Guidelines Retrieved April 30, 2010 from http://www.usdoj. gov/atr /pu blic/ guidelines/hmg.pdf U.S. Department of Justice The Herfindahl Hirshman Index, Rerived May 9, 2010 from www.justice.gov/atr/public/testimony ) Zimmerman, H. and Market Share in Rhode Island: Spectrum Research Services, Inc., Rhode Island Department of Health in Medical Center and Charter Care Health Partners under the Rhode Island Hospital Conversions Act, RIGL Chapter 23 17.14

PAGE 251

0 1 X 2 S 3 L 4 XS 6 V Variables Intercept Ownership South LOS Ownership x South Valve Parameter Estimates 8,478 47,505 22,429 6,214 (8,529) 21,046 0 Intercept 740,272 (351,146) (339,258) (33,724) 340,208 (138,763) 1 X Ownership (351,146) 1,277,769 366,334 (1,741) (1,277,125) 10,030 2 S South (339,258) 366,334 834,299 (1,790) (834,493) (32,045) 3 L LOS (33,724) (1,741) (1,790) 3,518 1,496 (5,239) 4 XS Ownership x South 340,208 (1,277,125) (834,493) 1,496 2,622,028 39,414 6 V Valve (138,763) 10,030 (32,045) (5,239) 39,414 705,934 South = South Geographic Sector Marginal Effects & Standard Errors for Geographic Location x Ownership Ownership when South Geographic Sector = 1 4 when X = 1) = 47,505 + S(-8,529) = 38,976 1 ) + S 2 4 1 4 X)] 1/2 2 (2,622,028) 2X(-1,277,125) = 1,160 t = 38,976 / 1,160 = 33.60 237 Appendix 10 Variance-Covariance Matrix Model 2-B

PAGE 252

Ownership LOS South Ownership x South Valve 1 2.82046 1 0.03122 0.03682 0.03676 0.0271 0.03104 2 0.99034 1.68759 0.05962 0.06172 0.00097473 0.09676 0.33767 3 0.55546 2.25338 0.19821 0.0213 0.40225 0.0004599 0.26779 4 0.45541 2.48863 0.06487 0.53462 0.07253 0.09636 0.36206 5 0.17834 3.97684 0.64609 0.34553 0.48748 0.77932 0.00145 238 Appendix 11 Part 2, Model 2-C Collinearity Diagnostics (Intercept Adjusted) Collinearity Diagnostics Number Eigen value Proportion of Variation Condition Index Eigen values close to zero signify serious problems of multicollinearity. An example could be an Eigen value at 0.09 or below. None of the Eigen values in connection with Model 1 B were close to zero. As a rule of thumb a condition index > 30 indicates serious problems of multicollinearity. None of the condition index values are even as high as 4.0.

PAGE 253

About the Author James Barrington holds Master of Public Administration (MPA) and Master of Public Health (MPH) degrees from the University of South Florida (USF) He also has bachelor degrees in sociology and accounting, as well as a fifth year in accounting. Mr. Barrington is also a C ertified P ublic A ccountant (CPA), C ertified B usiness M anager (CBM), and a C ertified Florida F amily C ourt M ediator. Mr. Barrington has worked for 2 6 years in local government and continued to work fulltime while matriculating thr ough the Ph.D. program in USF s College of Public Health Prior to entering the Ph.D. program he ran side businesses in accounting, tax return preparation, and grant writing. Mr. Barrington has also assi s ted undocumented worker s in obtaining individual tax identification numbers allow ing th em to file tax returns with the IRS and helped several of them establish their own limited liability companies. A serious problem for undocumented workers is that they are often hired to perform physically strenuous jobs, but once the jobs are complete, employers either refuse to pay them or will remunerate to them only a small fraction of the agreed upon amount. Employers are well aware that undocumented workers have little or no legal recourse when they are underpaid or receive no payment at all. Over the years, Mr. Barrington has intervened on behalf of undocumented workers challenging employers to fully pay their workers. This has often resulted in workers receiving a higher fraction of payment, and every now and then has resulted in a worker receiv ing full pay. Mr. Barrington is also married has two children and four grandchildren.