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 Ka
controlfield tag 001 001970413
007 cr mnu|||uuuuu
008 081126s2007 flu sbm 000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0002314
Organizational form of disease management programs :
b a transaction cost analysis
h [electronic resource] /
by Nahush Chandaver.
[Tampa, Fla.] :
University of South Florida,
ABSTRACT: Patient care programs such as wellness, preventive care and specifically disease management programs, which target the chronically ill population, are designed to reduce healthcare costs and improve health, while promoting the efficient use of healthcare resources, and increasing productivity. The organizational form adopted by the health plan for these programs, i.e. in-sourced vs. outsourced is an important factor in the success of these programs and the extent to which the core objectives listed above are fulfilled. Transaction cost economics aims to explain the working arrangement for an organization and to explain why sourcing decisions were made by considering alternate organizational arrangements and comparing the costs of transacting under each. This research aims to understand the nature and sources of transaction costs, how they affect the sourcing decision of disease management and other programs, and its effect on the organization, using current industry data. Predictive models are used to obtain empirical results of the influence of each factor, and also to provide cost estimates for each organizational form available, irrespective of the form currently adopted. The analysis of the primary data obtained by the means of a web-based survey supports and confirms the effect of transaction cost factors on these programs. This implies that in order to reap financial rewards and serve patients better, health plans must aim to minimize transaction costs and select the organizational form that best accomplishes this objective.
Thesis (M.S.I.E.)--University of South Florida, 2007.
Includes bibliographical references.
Text (Electronic thesis) in PDF format.
System requirements: World Wide Web browser and PDF reader.
Mode of access: World Wide Web.
Title from PDF of title page.
Document formatted into pages; contains 174 pages.
Adviser: Kingsley A. Reeves, Jr., Ph.D.
Inverse mills ratio.
x Industrial Engineering
t USF Electronic Theses and Dissertations.
Organizational Form of Disease Management Programs: A Transaction Cost Analysis by Nahush Chandaver A thesis submitted in partial fulfillment of the requirements for the degree of Master of Science in Industrial Engineering Department of Industrial and Management Systems Eng ineering College of Engineering University of South Florida Major Professor: Kingsley A. Reeves, Jr., Ph.D. Michael Weng, Ph.D. Gabriel Picone, Ph.D. Date of Approval: November 14, 2007 Keywords: predictive modelling, logistic regression selection bias, inverse mills ratio, outsourcing, integration Copyright 2008, Nahush Chandaver
i Table of Contents List of Tables..................................................................................................................................iv List of Figures................................................................................................................................vii Abstract.........................................................................................................................................viii Chapter 1. Introduction....................................................................................................................1 1.1 Introduction to the Disease Management Concept.......................................................1 1.2 Outsourcing in the Health Insurance Industry..............................................................4 1.3 Background, Complication and Objectives..................................................................4 1.4 Research Approach and Benefits..................................................................................5 1.5 Structure of the Report..................................................................................................8 Chapter 2. Literature Review...........................................................................................................9 2.1 Introduction...................................................................................................................9 2.2 Transaction Cost Economics (TCE) Theory.................................................................9 2.2.1 Uncertainty..................................................................................................11 2.2.2 Asset Specificity and its Effect on Organizational Form............................15 2.2.3 Similarity and Frequency............................................................................18 2.2.4 Bounded Rationality...................................................................................20 2.2.5 Core Competence and Transaction Costs...................................................21 2.2.6 Empirical Measurement of Transaction Costs............................................24 2.3 Disease Management Literature Review....................................................................25 2.3.1 A Brief History of Disease Management....................................................26 2.3.2 Current State of the Disease Management Industry....................................27 2.3.3 Effect of Transaction Cost Factors on DM Organizational Form...............34 2.3.4 Future of Disease Management..................................................................40 2.4 Selection Bias and the Heckman Two-Step Method..................................................41 Chapter 3. Literature Summary and the Hypotheses.....................................................................44 3.1 Literature Summary and Application to Health Plans................................................44 3.2 Application of TCE Factors to Health Plans...............................................................45 3.3 The Hypotheses...........................................................................................................48 Chapter 4. Methodology................................................................................................................49 4.1 Research Approach.....................................................................................................49 4.2 Research Method........................................................................................................49 4.3 Design and Study Participants....................................................................................52 4.4 Measures, Instruments and Data Sources...................................................................53 4.5 Primary Data Collection.............................................................................................54 4.6 Data Analysis..............................................................................................................57
ii Chapter 5. Numerical Results and Inference.................................................................................59 5.1 Frequencies of Respondents and Corresponding DM Programs................................59 5.2 Data Analysis..............................................................................................................64 5.3 Creating the Training and Validation Sets for the First Stage Select ion Model.........67 5.4 Training Stage.............................................................................................................68 5.4.1 Step 1. Creating the First Stage Selection Model.......................................68 5.4.1 Step 2. Evaluate Results of the Training Stage...........................................73 5.4.1 Step 3. Use the Classification Table to Determine Optimal Cut-Off Point...........................................................................................................77 5.5 Validation Stage..........................................................................................................78 5.5.1 Step 1. Embed the Validation Set Into the Training Set.............................78 5.5.1 Step 2. Results of the First Stage Selection Model With Combined Data Set......................................................................................................82 5.5.2 Classification of the Training Set...............................................................83 5.5.3 Classification of the Validation Set............................................................84 5.5.4 Inference for the First Stage Selection Model............................................84 5.6 First Stage Selection Model Excluding Uncertainty...................................................86 5.6.1 Training Stage.............................................................................................87 5.6.1 Step 1. Creating the First Stage Selection Model Excluding Uncertainty.. 87 5.6.1 Step 2. Evaluate Results of the Training Stage for the Seven Factor Model.........................................................................................................89 5.6.1 Step 3. Use the Classification Table to Determine Optimal Cut-Off Point...........................................................................................................91 5.6.2 Testing Stage...............................................................................................92 5.6.2 Step 1. Embed the Validation Set Into the Training Set.............................92 5.6.2 Step 2. Results of the First Stage Selection Model With Combined Data Set for Seven TCE Factors................................................................95 5.6.3 Classification of the Training Set...............................................................96 5.6.4 Classification of the Validation Set............................................................97 5.7 Comparison of the Model With and Without the TCE Factor Uncertainty................97 5.8 Inference From the Comparison of Full and Seven Factor Model..............................98 5.9 Calculation of Actual In-Sourced and Outsourced Costs...........................................99 5.9.1 Breakup of Actual In-Sourced Costs..........................................................99 5.9.2 Breakup of Actual Outsourced Costs........................................................100 5.9.3 Calculation of the Inverse Mills Ratio and the Help and Control Factor Delta..............................................................................................101 5.10 Frequencies for the In-Sourced Subset...................................................................104 5.10.1 Frequencies for the Organizations in the In-Sourced Subset..................104 5.10.2 Frequencies for the Diseases in the In-Sourced Subset..........................105 5.11 Organization Cost Model for In-Sourced Costs......................................................106 5.11.1 Running the Model.................................................................................107 5.11.2 Results of the In-Sourced Cost Model With Correction for Selection Bias.......................................................................................109 5.11.3 Correcting the Standard Error Terms......................................................111 5.11.4 Inference From the Internal Cost Equation.............................................116 5.12 Comparison of First and Second Stage Results......................................................118 5.13 Comparison of Actual and Predicted In-Sourced Costs..........................................120 5.13.1 Means for the Actual and Predicted In-Sourced Costs...........................120 5.13.2 Comparison of Actual and Predicted Costs for Sub-Sample of Data.......................................................................................................121
iii 5.13.3 Rolling Up the Costs for Each Organization in the In-Sourced Subset....................................................................................................121 5.14 Creating the Log Specification Model for the In-Sourced Costs............................122 5.14.1 Running the Model.................................................................................122 5.14.2 Results of the Log Specification In-Sourced Cost Model.......................124 5.14.3 Correcting the Standard Errors for the Log Spec Model........................125 5.15 Comparison of Actual and Predicted In-Sourced Costs..........................................129 5.15.1 Means for the Actual and Predicted In-Sourced Costs From the Log Specification Model......................................................................130 5.15.2 Comparison of Actual and Predicted Costs for Sub-Sample of Data From the Log Specification Model..............................................131 5.15.3 Combined Actual Cost, Combined Predicted Cost and Combined Error Estimate Using the Log Specification of the Cost Model...........132 5.15.4 Cost and Error Estimates for the Whole In-Sourced Subset Using the Log Specification of the Cost Model..............................................133 5.15.5 Comparison of Coefficients From the Regular and Log Specification of the In-Sourced Cost Model.........................................135 5.15.6 Comparison of Summed Actual and Predicted Costs From the Regular and Log Specification of the In-Sourced Cost Model.............136 Chapter 6. Conclusions and Future Research..............................................................................138 6.1 Conclusions...............................................................................................................138 6.2 Organizational Form and Cost Analysis and Prediction as a Decision Making Tool..........................................................................................................................139 6.3 Future Work..............................................................................................................140 References....................................................................................................................................142 Appendices...................................................................................................................................149 Appendix A. SAS Code..................................................................................................150 Appendix B. Electronic Survey......................................................................................170
iv List of Tables Table 2.1 Disease Management Program Statistics Across the U.S..............................................33 Table 2.2 Method of Disease Management Program Implementation..........................................33 Table 4.1 Probit Model Variable Definitions and Descriptions.....................................................55 Table 5.1 Responding Organizations and Number of Respective Responses................................60 Table 5.2 Frequency of Corresponding DM Programs..................................................................61 Table 5.3 Frequencies of the Organization Form..........................................................................62 Table 5.4 Frequencies of the Eight TCE Factors...........................................................................63 Table 5.5 Means and Statistics for all Modeling Variables...........................................................65 Table 5.6 Correlations for all Modeling Variables........................................................................66 Table 5.7 Frequencies for the Training Set....................................................................................68 Table 5.8 Frequencies for the Validation Set.................................................................................68 Table 5.9 Full Model Response, Fit Statistics and Null Hypothesis for T raining Stage...............70 Table 5.10 Analysis of Parameter Coefficients for the Training St age.........................................76 Table 5.11 Classification Table for Training Set...........................................................................78 Table 5.12 Full Model Response, Fit Statistics and Null Hypothesis for Te sting Phase...............80 Table 5.13 Analysis of Parameter Coefficients for Testing Stage................................ .................82 Table 5.14 Prediction Accuracy for the Training Set....................................................................83 Table 5.15 Prediction Accuracy for the Validation Set.................................................................84 Table 5.16 Seven Factor Model Response, Fit Statistics and Null Hypothesis for Training Stage...........................................................................................................................87 Table 5.17 Analysis of Parameter Coefficients for the Seven Factor Model.................................90
v Table 5.18 Classification Table for the Seven Factor Training Sta ge...........................................92 Table 5.19 Seven Factor Model Response, Fit Statistics and Null Hypothesis for Testing Stage...........................................................................................................................93 Table 5.20 Analysis of Parameter Coefficients for the Testing Sta ge of the Seven Factor Model.........................................................................................................................95 Table 5.21 Prediction Accuracy for the Training Set....................................................................96 Table 5.22 Prediction Accuracy for the Validation Set.................................................................97 Table 5.23 Comparison of the Full and Seven Factor Models.......................................................98 Table 5.24 Means for the Actual In-Sourced Costs.....................................................................100 Table 5.25 Means for Actual Outsourced Costs..........................................................................101 Table 5.26 Results and Statistics for the Inverse Mills Ratio and Control Factor Delta.............103 Table 5.27 Frequencies for Responding Organizations of the In-Sourced Subset .......................105 Table 5.28 Frequencies for the DM Programs of the In-Sourced Subset....................................106 Table 5.29 Analysis of Variance for the In-Sourced Cost Model................................................109 Table 5.30 Parameter Estimates for the Independent Variables in the In-Sourced Cost Model.......................................................................................................................111 Table 5.31 Values of Corrected Variance, Std. Error and Error Correlat ion...............................113 Table 5.32 Results From Heteroskedasticity Correction.............................................................114 Table 5.33 Final Cost Model Results...........................................................................................116 Table 5.34 Comparison of Selection and Substantial Model Coefficients..................................119 Table 5.35 Means and Statistics for Actual and Predicted In-Sourced Costs ..............................120 Table 5.36 Comparison Between Actual and Predicted Costs for Sub-Sample of D ata..............121 Table 5.37 Rolling Up the Costs for Each Organization in the Integrated Subse t.......................122 Table 5.38 Analysis of Variance for Log Specification of In-Sourced Cost Model....................123 Table 5.39 Parameter Estimates for the In-Sourced Log Specification M odel............................125 Table 5.40 Log Specification Corrected Variance, Std. Error and Error Co rrelation..................126
vi Table 5.41 Log Spec Heteroskedasticity Correction Results.......................................................127 Table 5.42 Final Log Spec Cost Model Results...........................................................................129 Table 5.43 Means and Statistics for Actual and Predicted In-Sourced Costs From the Log Specification Model.................................................................................................131 Table 5.44 Comparison of Actual and Predicted Costs for Sub-Sample of Data F rom the Log Specification Model..........................................................................................132 Table 5.45 Summing Up the Actual and Predicted Costs and the Error Estimate for Each Organization.............................................................................................................133 Table 5.46 Cost and Error Estimates for the Full In-Sourced Subset From the Log Specification Model.................................................................................................134 Table 5.47 Coefficient Comparison Between Standard and Log Specification of t he Cost Model.......................................................................................................................135 Table 5.48 Comparison of Costs From the Standard and Log Specification of the Cos t Model.......................................................................................................................136
vii List of Figures Figure 1.1 Disease Management Supply Chain...............................................................................3 Figure 2.1 Flexibility Need v/s Control Need Option Range.........................................................13 Figure 2.2 Competitive Advantage v/s Strategic Vulnerability Matrix .........................................14 Figure 2.3 Percentage of DM Programs Offered by Health Plans in the U.S................................29 Figure 2.4 Organizational Form of DM Programs in Health Plans...............................................35 Figure 2.5 Percentage of Health Plans and Respective Anticipated Areas of E xpansion for DM Programs.............................................................................................................40
viii Organizational Form of Disease Management Programs: A Transaction C ost Analysis Nahush Chandaver ABSTRACT Patient care programs such as wellness, preventive care and specifi cally disease management programs, which target the chronically ill population, are designed to reduc e healthcare costs and improve health, while promoting the efficient use of healthcare resources and increasing productivity. The organizational form adopted by the health plan for these pr ograms, i.e. insourced vs. outsourced is an important factor in the success of these progr ams and the extent to which the core objectives listed above are fulfilled. Transaction cost economics aims to explain the working arrangement for an o rganization and to explain why sourcing decisions were made by considering alternate organizati onal arrangements and comparing the costs of transacting under each. This research aims to unde rstand the nature and sources of transaction costs, how they affect the sourcing decision of disease management and other programs, and its effect on the organization, using current industry da ta. Predictive models are used to obtain empirical results of the influence of each f actor, and also to provide cost estimates for each organizational form available, irrespecti ve of the form currently adopted. The analysis of the primary data obtained by the means of a web-based surve y supports and confirms the effect of transaction cost factors on these programs. Th is implies that in order to reap financial rewards and serve patients better, health plans must ai m to minimize transaction costs and select the organizational form that best accomplishes this ob jective.
1 Chapter 1. Introduction This is a thesis on the concept of outsourcing in the health insurance indust ry with a specific focus on disease management programs operated by the various health plans. It aim s to answer the question if sourcing of disease management programs can be explained base d on transaction cost factors and used to lead to cost savings. 1.1 Introduction to the Disease Management Concept The quality of healthcare and health services has been the subject of publi c scrutiny and much debate, and it has recently heightened due to the rapid growth of costs and li tigation in the form of lawsuits for negligence. There is increasing dissatisfaction of healthcare consumers with their experience due to significant deviations from best care practices, ri se in medical errors and a large addition of unknown or non-value added services in healthcare . A lingering conc ern is the inability of the U.S. healthcare system to deal with the chronically il l population, which has been increasing in recent years . As noted by the Florida Medicaid Disease Management Initiative in 2000 , disease management programs have been proposed in order to improve heal thcare by facilitating and addressing several key issues outlined below. Disease management programs are designed to benefit both the healthcare organization and the patient by following a two-pronged approach. At the patient side, the chronica lly ill and the population at risk for chronic diseases are admitted in these programs. Th e program then takes steps to improve the health outcomes and quality of life for the patient. It does this by fostering self-care/self management of the condition by the patients themselve s, aided by patient education and by raising the awareness of the patient regarding his or her own health c onditions. Doing so also promotes accountability of the patient in the care and treatment deci sions taken. As the awareness of the patient regarding the condition(s) is increased, it leads to a more beneficial and stronger relationship between the physician and the patient. The program staff undertakes patient monitoring and promotes the continuity of care, that is, takes steps to e nsure that the patient
2 completes the entire treatment cycle and also measures patient sati sfaction and treatment effectiveness for each patient on an ongoing basis. At the physician/care provi der side, these programs aid the medical professionals by providing them valuable relevant information and practice/evidence-based guidelines that may prove helpful to them during pat ient treatment and care. By doing so these programs can delay and, in the best cases, even prev ent complications of chronic health conditions. This leads to an improvement in the health outc omes and quality of life for the patient, while at the same time it leads to cost savings f or the patient in terms of healthcare costs, and also for the healthcare provider, and is thus very beneficial to al l parties involved. Disease management programs also promote efficient use of healthcare resources and increase medical productivity by increasing patient awareness levels and helpi ng physicians in their treatment protocol. The supply chain for a disease management program i s as shown in figure 1 below. Disease management programs are particularly applicable and useful to Fl orida as it is the third largest in Medicaid spending in the U.S and ranks 41 st in the nation in per capita expenditures . The state of Florida is also a pioneer in this area as it is the first state to implement these programs in the Medicare and Medicaid fields in 1998 and encourage health plan s to adopt these plans at the same time. The state government has already reduced the annua l budget for Medicare and Medicaid by $ 66 million in anticipation of the savings that were promised by the pr oponents of these programs, and the results of the early studies done to measure the effectiveness and results of these programs . However, subsequent findings have shown that while savings in healthcare costs have occurred for patients, they have been offset t o some extent by rising drug costs. Moreover, the savings for the health management organizations hav e been offset by the cost of implementation of the disease management programs. This has reduce d the actual savings and effectiveness of the programs in terms of efficiency and cost sav ings for the healthcare organizations.
3 No Normal operation Yes No Yes (In-sourced DM program) (Patient Identification/education/monitoring) (Practice based guidelines and relevant info) (Treatment) (Reporting and feedback loop) ** Choice of health plan given to eligible by employers or Medicare/Medica id. ++ Utilization review/ do DM programs need to be implemented? ## Outsourcing decision Figure 1.1 Disease Management Supply Chain Employer/Medicare/Medicaid ** Indemnity (fee-for-service) health plans MCO Managed care health plans HMO PPO HMO health plans POS health plans ++ ## DMO Disease Management Team Data Analysis/ Clinical information systems Case Managers Administrative team Pharmacists Physician Patient PPO health plans
4 1.2 Outsourcing in the Health Insurance Industry In order to remain financially viable and profitable while adhering and pr omoting the disease management principles listed above, a medical insurance organizatio n must develop effective strategies for care provision to the affected population . One method for thi s is the outsourcing of the disease management programs by the Health Management O rganization (HMO) to external disease management organizations (DMOs). To dat e, the decision for outsourcing has been attributed to changes in market costs and not due to inte rnal organization costs. However, internal organization costs have been thought to be just as impor tant to the outsourcing decision as the external market costs, and this was proved empiric ally in the shipbuilding industry . Our objective is to explain the outsourcing or integ ration decision of patient care programs based on transaction cost factors, and to determine to wh at extent that decision is supported, by measuring and comparing costs of the different orga nizational forms. We will study the various transaction cost factors as applied to dis ease management programs, determine the most important ones, that is, the factors which exert the most i nfluence over the outsourcing decision in this industry, and study whether their primary effect is on external market costs or internal organization costs. 1.3 Background, Complication and Objectives The state of Florida is unique in that it was the first state in the countr y to develop and implement disease management programs within the state healthcare plans for eli gible residents, and encourage the implementation of these plans in the states private HMOs and healthcare providers in the late 1990s. The other states in the country are taking an active int erest in the performance of these programs to see if these programs deliver on their promise of reduced healthcare costs, better patient health outcomes and improved efficiency and profitability for the healthcare organizations. While early research has shown improvements in the health outcomes and cost s for patients in the short term, the long-term effects for both the patients and the organizat ions are not clear and need to be studied further [22, 39].
5 Employing this econometric analysis to this industry will allow a study of t he strategies undertaken by the concerned organizations in order to meet these objectives, a nd bring out the effect of transaction costs on these organizations, while highlighting the most important transaction cost factors that apply to this particular industry. Thus it will have an immediate broad impact. The significance of the proposed research is that it will provide a model tha t can be widely disseminated and improved upon to assist in the further research and learning i n the field of transaction cost economics and factors as applied to disease manageme nt programs, medical tasks and the service industry in general. Much will be learned about how intern al organization costs influence the outsourcing decision and what transaction cost factors h ave the greatest influence over the final form of the organization. As various organization costs wil l also be gathered, this research will also yield valuable information on the costs /savings i ncurred by the various forms of organization possible in a specific case. Transaction cost analy sis applied to the outsourcing of disease management programs will contribute to a deeper understanding of the economics followed by the health management/maintenance organizations (HMOs) and gove rnment healthcare entities (Medicare/Medicaid). Objectives of the propo sed thesis research are to: 1) Determine whether transaction cost analysis can be used to validate the ef fectiveness of organizational form in disease management programs. 2) Identify which of the transaction cost factors (such as asset specifi city, uncertainty and complexity) exert greater influence on the outsourcing decisions in diseas e management programs. 3) Determine whether outsourcing leads to fulfillment of disease manageme nt objectives. 4) To isolate the effects of transactions on the cost of in-house care and out sourcing of care. 5) Provide dollar estimates for the costs/savings associated with the s ourcing decision. 1.4 Research Approach and Benefits In economics and other related disciplines, transaction costs are define d as the costs incurred in addition to the price of the intended economic transaction such as a service task or product. A number of kinds of transaction cost have come to be known by particular names [ 38].
6 Search and information costs are costs such as those incurred in determi ning that the required good is available on the market, which has the lowest price, etc. Bargaining costs are the costs required to come to an acceptable agr eement with the other party to the transaction, drawing up an appropriate contract and so on. In game theory this is analyzed for instance in the game of chicken. Policing and enforcement costs are the costs of making sure the other par ty sticks to the terms of the contract, and taking appropriate action (often through the legal system) if this turns out not to be the case. The factors that cause transaction costs to be incurred for organizatio ns can be attributed to various factors that can be explained as follows [75, 78, 82, 51, 36, and 61]: 1) Asset Specificity Williamson [78, 82, and 83] has suggested six main types of asset specificity : Site specificity Physical asset specificity Human asset specificity Brand names Dedicated assets Temporal specificity 2) Uncertainty 3) Similarity/relatedness  4) Frequency These factors have been explained in the next section. Transaction cost ec onomics can be used as a framework for understanding the healthcare organizationÂ’s decision to outso urce or integrate disease management programs based on these factors. Research in this area has encountered significant difficulty due to the difficulty of obtaining the relevant data, and empirical data have not been applied to the disease management programs, and the evidence of their effectiveness is limited [39, 34]. The application of the above analysis to disease managem ent programs is helpful in explaining the outsourcing protocol followed by many medical organizations. It shows which
7 factors are the most influential in the decision to outsource patient c are, and also helps in providing a dollar estimate of the various organizational forms in this sector, which has not been available before. But most importantly, this research helps in i mproving profitability for medical organizations, without compromising the aims of the implemented disease m anagement programs, among which are increasing satisfaction and quality of life and re ducing costs for the patients. The benefits of applying the transaction costs analysis to disease mana gement programs are as follows. 1) It leads to more effective understanding of the organizational structur es of private and government health management/maintenance organizations. In this resea rch, we give explicit attention to the role of internal organization costs in outsourcing deci sions. We use transaction cost analysis as a framework to study these costs. Many pr evious attempts to apply transaction cost economics to various industries have used estimati ons of reduced form relationship between organizational forms and observed characteristics Due to this, it was not possible to decipher whether the resulting organizational form was due t o changes in market transaction costs or from variations in the costs incurred i n organizing the production internally. Using censored regression and the two-stage method outlined below we can overcome the difficulties generally observed in obtaining direct observat ions of data, while at the same time giving explicit attention to the role of internal organi zation costs. Based on this, we can infer whether the effect of a particular variable raises the probability of integration in a particular organization due to increase in the hazards o f market exchange or its effects on the internal organization costs. 2) It increases understanding of the factors and costs that affect outsourcing Application of these methods shows which transaction cost factors exert a stronger influ ence over the outsourcing in the health management organizations, and whether they have a st ronger effect on the costs of internal organization or market exchange costs. 3) Application of censored regression techniques also leads to the isolat ion of the effects of attributes of transactions on the cost of organizing within and between fir ms and provides dollar estimates to these costs. It has been proven that the costs vary systematically with the nature of transaction and that the savings of choosing the right organizati onal arrangement are substantial . Empirically, it has been shown that in the shipbuilding industry mistaken
8 integration of work that is typically outsourced/subcontracted increased i nternal organization costs by 70%, while outsourcing work normally performed internally within the firm led to organizational costs almost three times those incurred if the jobs were done internally . Transaction cost analysis applied to disease management program outsourci ng in the form of censored regression techniques provides a similar estimate of the costs and savings borne by these organizations. 4) It also contributes to the research on transaction costs. This work contributes to the research on transaction cost analysis. Various transaction cost factors have been studied in this research. This method of analysis has been applied to both the manufacturing and the construction industry. The factors for scheduling and engineering intensity hav e been proven to be important in the case of the naval shipbuilding industry . Although the c onditions of bounded rationality and opportunism may be universal, the factors that influ ence them may vary from one industry to another. Hence the effect of the factors considered w ill be different for different industries, and as a result, it is important for case st udies in various other industries be carried out along with more formal empirical analysis. Tra nsaction cost analysis has so far not been applied to the disease management industry and empirical research in this industry using censored regression techniques is yet to be carried out, apa rt from the analysis and results presented herein. The application of the transaction cost analysis framework to this industry enhances our understanding of health plans decisions regarding out sourcing and their organizational behavior. 1.5 Structure of the Report This thesis is organized as follows, spanning six chapters. Chapter 1 introduces the topic area, outlines the reasons for study, and provide s details of the research objectives. Chapter 2 focuses on providing an extract of the lite rature survey prior to forming the hypotheses. Next, chapter 3 summarizes the theoretical concepts and articulates the hypotheses based on the literature review performed on disease management, pat ient care programs and transaction cost economics. Chapter 4 outlines the research met hodology for primary and secondary data collection and analysis that is used for hypothesis testing in our case, and chapter 5 presents the numerical results and inference. Finally, cha pter 6 provides the conclusions and the directions for future research in this area.
9 Chapter 2. Literature Review 2.1 Introduction Outsourcing is a very well researched topic and considerable resear ch has been done in the field of transaction cost economics to explain the cause and effects of outsourci ng in various industries. This chapter aims to provide a history of transaction cost theory and th e previous research on this topic with the help of an extensive literature search involving the study of rel evant theoretical concepts and previous related work. The conclusions reached from this exerci se have been summarized in chapter 3 to form the background to the work done in this research. Another objective is to examine the transaction cost theory and disease m anagement literature to find relevant theories and empirical evidence regarding the in-sourcing vs. outsourcing or build vs. buy decision faced by various organizations. The collection of the availab le results is used in the formulation of the main hypothesis of this research. 2.2 Transaction Cost Economics (TCE) Theory There is an immense body of literature available in the field of tran saction cost economics. A comprehensive review may be found in Shelanski and KleinÂ’s  1995 work. The mai n tenet of transaction cost economics (TCE) suggests that transactions betw een providers and users of goods or services should be organized in a manner such that transaction costs ar e minimized. The theory behind transaction cost analysis was developed by Ronald coase i n his seminal paper, The Nature of the Firm (1937) , which laid the foundation for all further research done in this area, most notably by Oliver E. Williamson. This theory was used by Coase to d evelop a theoretical framework for predicting when certain economic tasks w ould be performed by firms
10 and when they would be performed on the market, as noted by Robert Kissell and Morton Gl antz in Optimal Trading Strategies, AMACOM, 2003 . Subsequently, Oliver E. Wi lliamson coined the term transaction cost and has done extensive research in this area, which is elaborated on below. Organizations and firms usually do not place emphasis on transaction costs. Acc ording to Straub and Ang's (1998)  research, production cost (which is defined as the amount of mone y a customer pays the vendor for its services) is given six times more im portance than transaction costs. McFetridge and Smith (1989)  study outsourcing service contracts i n Canada in their research and find that simple production costs are not sufficient to explain t he pattern of outsourcing, which validates the theory and effects of transaction costs. The theory of transaction cost economics focuses on the costs of transact ions when a good or service is transferred from a provider to a user. When an organization out sources, the transaction costs will include the costs of searching and selecting the supplier(s), dr awing up the contract, performance/results measurement, and dispute resolution (usually involvi ng litigation and/or a third party adjudicator). Conversely, when transactions are internal, t he total costs include managing and monitoring costs in addition to the cost of the capital, inputs and ra w materials required for the transaction. According to Williamson (1989) , the form adopt ed by the organization, (referred to as governance structure, by Williamson) affe cts the transaction costs. Transaction costs occur before and after an economic transaction and a ce ntral proposition of transaction cost economics is that organizations strive for greater efficiency by implementing governance structures that minimize transaction costs. Organizations have many options for organizing these transactions via governance structures which vary from spot/open markets for generic goods and services where the buyers and sellers are immaterial to the transaction, to fully vertically integrat ed organizations, where both buyer and seller can be said to be one and the same and are under joint ownership and contr ol. Between these two extremes of spot markets and vertical integration there ar e various contracting choices available for the organization to complete its transactions, which include shared ownership of assets or joint ventures.
11 Williamson (1979, 1981) [76, 77] states that markets are not the best solution f or transactions involving asset specificity because buyers and sellers can easily wa lk, that is, cancel the transaction without any loss to themselves. Markets are also not ideal when one considers the possibility of opportunistic behavior by the parties, which is explained below. Williamson (1989, 1993) [79, 81] explains opportunistic behavior as follows: the va lue of the transaction-specific assets in question depends on the continued contract be tween the buyer and seller, hence, the party that has not invested in these specific asse ts may be tempted to threaten to walk away from the relationship in order to realize more value from this investment. He also points out that asset specificity plays a major role in the degree o f vertical integration and that vertical integration may be the only solution for costly asset specif ic investments as it is highly difficult for these assets to be transferred or utilized for alternat ive buyers/sellers and used for other tasks and services. The types of transaction cost factors have been broadly defined by Leeman (2006 )  as follows: 1) Uncertainty, 2) Asset specificity and 3) Frequency. Another factor can be said to be the similarity of the tasks and services in question. Asset specificity is generally regarded as the most crucial transaction c ost factor . Others regard Uncertainty to be the most critical factor . 2.2.1 Uncertainty As stated by Leeman , Â“Uncertainty generally refers to how easily per formance can be monitored. Monitoring becomes problematic when the task requirements or outc omes are difficult to predict or when the service purchased requires teamwork, making it difficult to connect the product with an individuals input. The greater the uncertainty, the greater t he transaction costs incurred in developing and executing a contract in a manner such that all part ies are satisfied with the outcomeÂ”.
12 Many researchers have studied the effect of uncertainty on organization al form. Pirrong (1994)  found that in ocean shipping the type of contract used (spot markets, medium and long-term contracts or vertical integration) depends on the uncertainty of providi ng alternative shipping services for the goods at short notice in the event of a problem or holdup. Stigler (1951)  has theorized that due to uncertainty, industries that a re in decline show greater tendency to outsource, whereas the organizations in their growth phas e show industries with a greater tendency to integrate. Casson (1986, 1987) [9, 10] studied the shipping industry and has found that that shipping companies running oil tankers and refrigerate d cargo ships tend to have ownership of the vessels used for transport early in the company deve lopment and are usually leased/contracted in the case of more established companies. This observation supports StiglerÂ’s theory given above. However, a contrasting view to StiglerÂ’s theory is available in the literature and can be seen in HarriganÂ’s (1983)  research, which has analyzed the vertical int egration within 192 firms in 16 different industries in the period between 1960 and 1981. She states that new indust ries which are inexperienced tend to have less integration and more outsourcing i n order to reduce risks. That is, early in an industryÂ’s development, when costs and risks are high, firm s generally operate with less integration. An example of the computer industry is given, which ou tsourced microprocessors and memory chips in its infancy, but began to internalize the production of these components as the industry grew and stabilized. The colloquial is stated as the greater prevalence of outsourcing within industries in decline, in order to meet fluctuating dema nds and market conditions (uncertainty), which can be limited due to high levels of integra tion. Harrigan also states that certain firms with bargaining power over s uppliers, distributors, and customers can reduce prices by reducing supplier profit margins, and can avoi d integration, thus the disadvantages associated with it. These results contrast with StiglerÂ’s (1951)  hypothesis that firms integrate early during industry development in order to achieve competitive advantages.
13 Figure 2.1 Flexibility Need v/s Control Need Option Range Source: Quinn, James Brian, and Frederick G. Hilmer, Strategic Outsourci ng, Sloan Management Review, Summer 1994, pp. 43Â–55.  Quinn and Hilmer (1994)  have studied uncertainty in terms of flexibilit y and control. If the firm is subjected to uncertainty in the form of changing demand for its product s or services then outsourcing gives the firm the flexibility to meet the changing scenarios but causes it to lose some control over the outsourced activity in terms of execution and performance They state that when firms outsource they normally transfer certain risks and investment s that they would have normally incurred by the contracted party. Figure 2 above shows all the choice s of organization form depending on the companyÂ’s control and flexibility needs. In order to minimize the effects of uncertainty, Quinn & Hilmer  in the s ame article also suggest that outsourcing be done by carefully taking the firmÂ’s skills and r esources into account, ShortTerm Contract Call Options Long-Term Contract Retainer Joint Development Partial Full Ownership Ownership Flexibility Need High Control Need Low High Low
14 and also by comparing the potential of gaining a competitive advantage in t he market with all the costs that would be incurred due to contracting. This compromise has been diagrammatically represented by them as per the matrix in Figure 3. If the activity is such that it allows the organization to gain a competitive a dvantage while vulnerability/uncertainty is low, then, it can be outsourced, else it shoul d be integrated. High Low High Low Figure 2.2 Competitive Advantage v/s Strategic Vulnerability Matrix Source: Quinn, James Brian, and Frederick G. Hilmer, Strategic Outsourci ng, Sloan Management Review, Summer 1994, pp. 43Â–55.  Both Badaracco (1991)  and Harrigan and Newman (1990)  state in their re search that there is a potential for knowledge leaks when organizations outsource, which, i f associated with the source of its competitive advantage can lead the organization to suffer a major setback. These uncertainties or risks, which are the loss of critical skil ls or loss of control over a supplier, have to be managed by careful monitoring and management of the outsourcing relationsh ip, which leads to an increase in transaction costs for the organization. High control needed (integrate) Moderate control needed (special venture or contract arrangement) Low control needed (Outsource) Potential for competitive edge Degree of strategic vulnerability
15 The counter argument to this is that transaction costs can still increa se due to uncertainty even if the task or service is integrated. Internal employees and departments m ay fail to perform to their full capacity, and may require policing and monitoring resources to improve perf ormance. In some cases it is more difficult to enforce and measure performance f or internal tasks and services than for external suppliers, which increases uncertainty and thus le ads to more transaction costs for the organization. Thus, as per Blumberg and Blumberg (1994)  an organization m ay lag behind industry best practices if internal departments are not world-class providers, due to an increase in transaction costs. Therefore, if the disease management industry were considered to be in a st ate of growth then previous research evidence would suggest that health plans would tend to ver tically integrate to include disease management operations and not outsource it. The disease ma nagement sector is indeed seen to be in the growth phase (as per our literature review in the nex t section), however, it is seen that organizational form for these programs is likely to be out sourced as a result of the specialized nature of this sector. 2.2.2 Asset Specificity and its Effect on Organizational Form Asset specificity refers to transaction specific investments in human, physical or other forms of capital. Asset specificity also refers to how specifically a partic ular product or service is designed or produced for a specific customer or if the product or service uses a speci fic asset. It is broken up into six main types, as explained below: 1) Site or location specificityÂ— the location of the buyer and seller in orde r to economize on inventories or transportation costs, or transportation and inventory costs specific to the transaction; 2) Physical asset specificityÂ—investments such as specialized equi pment, tools machines or systems designed for a particular customer or applications; 3) Human asset specificityÂ— the skills, experience or knowledge of the people inv olved in the transaction, or one or both of the parties develop skills or knowledge s pecific to the buyer-seller relationship;
16 4) Brand specificityÂ—the evaluation and selection of vendors and suppliers based on their reputations, or when the involved parties must maintain the reputation of a s hared brand name such as a franchise relationship; 5) Dedicated capacityÂ—capacity that is created to serve a particular customer and this capacity is difficult to adapt to use for alternative customers; a nd, 6) Temporal specificityÂ—the level or importance and specificity of the timing of a particular product or service. Joskow (1985)  has studied asset specificity with respect to mines supplying raw materials to electricity generator plants and his study shows that vertical integr ation is positively associated with all forms of asset specificity such as site specificity (when t ransportation costs are high), physical asset specificity, and human capital/know-how specific to the tra nsaction. Stuckey (1983)  and Hennart (1988)  have researched site specificity specif ically and their results support the above results and shows that aluminum refiners generally own their ow n bauxite mines because of high transportation costs (site specificity) whereas t his is not true in the case of tin refiners as the refiners are able to handle different ores Masten (1984)  studied asset specificity in the aerospace industr y and found that integrated components were generally more complex and specialized than Â‘buyÂ’ components. The hi gher transaction costs for these components due to a higher degree of physical and hum an asset specificity were stated as the causes of integration. Masten, Me ehan, and Snyder have extended the above research by studying the organizational form in the U.S. Auto Industry (1989) . They conclude that while physical and site specificity were not the maj or factors that decide vertical integration, engineering intensity is, and the reason for thi s is theorized as the greater human asset specificity required for these components and the difficulty of managing this when they are outside the firm make it more suitable for these components and s ervices to be integrated. Chandler (1961)  has also analyzed the maintenance strategies of the No rth American airline industry after the introduction of jet engines, based on human asset specificity The airlines had always maintained their own piston engines using their internal maintena nce departments. Jet engines were found to require new maintenance skills and facilities but les s frequent maintenance. Due to this, the maintenance of these engines continued to be i ntegrated during the
17 early years of the jet age. However, once maintenance practices and r outines became standard across airlines and engine types, internal maintenance departments we re outsourced to external independent maintenance specialists Fuhr and Thorsten  have studied the Vertical Governance between Airli nes and Airports using transaction cost analysis in 2006. They conclude that temporal speci ficity and uncertainty play a major role in the contracting between airlines and airports of v arious sizes. Anderson and SchmittleinÂ’s (1984)  work on human asset specificity reinforc es the above finding. Their study of the factors that determined the use of company sale s staff as opposed to independent distributors led to findings that an internal sales staff is used to reduce transaction costs when the following is required: 1) specialized training, 2) detaile d or proprietary knowledge of the selling company 3) continuing relationship between salespersons and clie nts 4) detailed knowledge of product or customer, and 5) when output measures of sales staff are unr eliable. Masten, Meehan, and Snyder  have also studied the organizational form and a ssociated costs in their 1991 study of naval shipbuilding industry and found that higher the importance of timely completion/scheduling of the component in construction the higher is the likel ihood of integration. This is because an interruption in any stage of construction d isrupts all subsequent operations by having a cascading effect which causes delays to the whole projec t. This also gives subcontractors incentives to delay in order to gain price concessions, whi ch is a type of opportunistic behavior. Monteverde and Teece (1982a)  found that in General Motors and Ford the probability that a component is produced in-house increased with the engineering effort require d to design it. This has been attributed to human capital specificity due to the engineering knowle dge required in these applications. Monteverde and TeeceÂ’s work (1982b)  on physical spec ificity found that automobile manufacturers in general were more likely to retain title t o the more specialized and expensive tooling used by suppliers. This again supports the theory that great er the asset specificity, the higher the incentive is for organizations to integr ate those tasks/applications. Lehmann and OÂ’Shaughnessy (1974)  have studied Reputation (Brand Specificity) and t hey state that it is very important in the selection and evaluation of vendors and suppliers.
18 They say that this is due to managementÂ’s desire to reduce risks to th eir companies and for themselves. This is achieved by selecting suppliers with a good reputation and high credibility, which can also improve the image of the contracting firm itself in some cas es. Panayides and Cullinane (2002)  have studied the importance of reputati on in ship manager selection. Their research was aimed at finding the most important criter ia for ship manager evaluation and selection. Their sample size consisted of 48 ship management c ompanies and 36 ship owners. They state that the inspection for selection is done mainly on tw o levels; the first level is financial variables, profitability, location and managerial ability. The second and more important level is a measure of the managerÂ’s reputation, image and rel iability, integrity, trustworthiness, and commitment. Thus, they state that brand specificity in the form of the ship managerÂ’s experience, establishment and status is a significant fac tor for the organizational form chosen by shipping companies for ship management. Reputation of the contracting pa rties can be said to reduce the risk of opportunism, which would reduce the monitoring cost s and increase the efficiency and profits to both parties involved. The above literature shows that in general, an increase in asset specif icity in any form is positively related to integration. 2.2.3 Similarity and Frequency Similarity can be said to refer to the nature of the tasks or processes and h ow closely they resemble the ones done on a regular basis by the firm or organization. Lee man (2006)  further states that transaction cost analysis studies the relationship betw een characteristics of transactions and the forms of governance organizations implement to negotiate and execute those transactions. Some examples of organizational structure include long term contracting short-term contracting and internal production. The view shared by shared by most economists is that or ganizations choose specific arrangements by comparing the costs of transacting under each. This insight needed empirical support, which was provided by noting the observable attribut es of transactions by Williamson (1975, 1979) and Klein et al. [75, 76, 60]. However, these efforts have generally concentrated on factors aggrava ting the hazards of market exchange, and the costs of internal organization have been treated only as a ba rrier to be overcome before integration (Masten et al., 1991) .
19 Frequency refers to how often the purchaser transacts in the market. Due to ec onomies of scale, frequency decreases the per transaction cost of asset-specific inves tments. Therefore, greater transaction frequency will enhance the value of asset-specific invest ments, for example, the costs of implementing new care management processes (Leeman, 2006) . Masten et al.  have given specific attention to the role of tran saction cost factors on internal organization costs in organization integration decisions, and have provided t he empirical study of a naval project, which has provided dollar estimates of the costs of vari ous organizational arrangements. The most important result of this research is regarding the contribution of changes in market and internal organization costs to the final arrangement a dopted by the firm. It is well known that internal organization sacrifices the advantages of market e xchange, while preventing problems such as opportunism, scheduling and uncertainty. However, this demands greate r investments in administration and monitoring (Williamson, 1985; 1990) [78, 80]. Economists and theorists have paid little attention to the influenc e that these factors make on the costs of managing and monitoring tasks and services internally, and to what ext ent they weigh on the form finally adopted by the organization. They have concentrated on how these fac tors affect the market prices, while neglecting the former effect. Ronald Coase has been one of the exceptions to this view and states Â“the effect of activities in w hich a firm is already engaged on the cost of undertaking additional activitiesÂ” is essential to expl aining why particular operations are chosen within specific firms (1988:40) . He goes on to say, Â“ The way in which industry is organized isÂ…dependent on the relation between the costs of carrying out transactions on the market and the costs of organizing an activity within that firm which can p erform this task at lowest costs. Furthermore, the costs of organizing an activity within a ny given firm depends on what other activities it is engaged in. A given set of activities wil l facilitate the carrying out of some activities, but hinder the performance of others. It is these relations hips which determine the actual organization of industry.Â” (1972:64) . He also states that inter nal organization costs are likely to be higher for transactions other than those in which the fir m is already engaged in, for which there is a higher degree of uncertainty. Asset specificities of the various types explained above tend to raise organization costs, if integration is carried out, a nd also raises market exchange costs, if outsourcing is favored, however, in the case of this fac tor, integration is usually preferred as it allows greater flexibility for change and modificati ons. Similarly, uncertainty and complexity, while producing a net increase in market as well as interna l organization costs, favors
20 integration to subcontracting as integration gives the organization allow s the organization to adapt to changing situations and circumstances, where outsourcing does not. The s imilarity of transactions, on the other hand, is unlikely to drive down market costs as t he parties engaged in the bargaining are most concerned about the final outcomes and not the manne r in which the goods or services are provided . In order to verify the above statements, empir ical data needs to be collected in order to support or refute them. This has only been done in the shi pbuilding industry and needs to be applied to the disease management field in order to st udy the effects of these factors on disease management program sourcing decision. 2.2.4 Bounded Rationality Managers and organizations have limited managerial time and control, and hence they cannot manage all tasks internally or plan and contract for all possibilities in the future in the case of outsourced tasks or services. This is due to bounded rationality, and thus, bounded rationality influences organizations in their attempts to reduce transaction cost s. The theory of bounded rationality was proposed in 1957 by Herbert Simon [62, 63], and it can be explained as the limitations on decision-making due by time, costs, human abilitie s, availability of information, and technology. He states, Â“Bounded rationality is a central them e in behavioral economics. It is concerned with the ways in which the actual decision-mak ing process influences decisions. Theories of bounded rationality relax one or more assumptions of st andard expected utility theoryÂ”. For most transactions, markets are the preferred governance structure a s markets provide the incentives to cut costs and maximize value net of production costs, while at the same time they allow the parties involved to respond quickly to changes in the market. As sta ted before by Williamson (1981) , markets are not the ideal solution for transact ions involving asset specificity because buyers and sellers can cancel the transaction en tirely. Contracts of differing lengths can offer some protection against the drawbacks of market transac tions; however bounded rationality makes it impossible to draw up contacts that cover all possible circumstances, due to which the involved parties may indulge in opportunistic behavior to make prof its.
21 As a result, complex internal control and monitoring systems may be needed to police the contract, make changes, and settle disputes if needed. Thus, bounded rational ity brings out the negative aspects of market transactions due to the possibility of opport unistic behavior by the parties and increases transaction costs. For other types of transacti ons a more integrated governance structure may be desired. If outsourcing is not possible due to ass et specificity, and bounded rationality, vertical integration can be used in order to maximize profits and reduce transaction costs. Bounded rationality also plays a role in internal org anization as control within the organization may be lacking in certain aspects, due to which opportunism by em ployees and an increase in transaction costs within the firm may be seen. Theref ore bounded rationality affects transaction costs in both governance structures, and organiza tional form should be chosen in order to minimize it. 2.2.5 Core Competence and Transaction Costs Prahalad and Hamel  introduced the concept of core competence in thei r 1990 study which they define as Â“the collective learning in the organization, especiall y in coordinating diverse production skills and integrating multiple streams of technologies.Â” Ex cellence in a few core competencies is what gives the organization a competitive edge in the ma rket. Quinn and Hilmer (1994)  in their article Strategic Outsourcing recomm end outsourcing only non-core activities to minimize transaction costs. This suggestion has bee n made so that firms can concentrate their limited internal resources on a set of core competenci es and tasks where they can achieve pre-eminence and provide unique value for their customers. In order to differentiate and identify these core functions in an organization, they have put forth the g uidelines given below: 1) Core competencies are limited in number. 2) Core competencies are flexible and long-term platforms capable of chang e. 3) Core competencies are skills or knowledge sets not products or functions. The y also cut across traditional functions. Hence, they are activities that are based on knowledge rather than on ownership of assets. 4) Core competencies should be embedded in the organizations systems and not dependent on a few people.
22 5) Core competencies can be used as sources of leverage in the value chain. 6) Core competencies are Core functions/services that are important to t he customer in the long term such as understanding and serving the customer. We can infer from the above that companies must retain only activities that give them the competitive advantage and other tasks and services may be outsourced. Howeve r, Quinn & Hilmer  point out that this is in fact not possible as the Â“supplier ma rkets are not totally reliable and efficientÂ”. According to them most outsourcing will enta il some risks, which have been elaborated above. Harrigan  supports the above recommendation with her 1983 work in which she anal yzes the vertical integration strategies of 192 firms in 16 different industrie s from 1960 to 1981. She finds that generally finds that firms internalize the tasks and service s that they consider to be their core competencies or those that contribute to their competitive advantage i n order to minimize transaction costs. One example cited in this work is how computer firms ma nufactured the logic chips and processors for their product internally but purchased the other c omponents. Another example of this is that pharmaceutical firms used their own trained sa les agents for marketing their medical products in order to protect their patents and increase sa les and also integrated production of certain chemicals and pharmaceuticals during high demand. Thus we see that evidence from the business management literature shows that integration reduces transaction costs associated with market transactions and common ad ministrative functions. However, integration can also increase transaction costs in the form of i nternal coordination, policing and enforcement costs and reduce incentives to maximize perfor mance and efficiency within the organization. Hence it is clear from the above that firms wil l benefit more by firstly outsourcing activities or parts of activities that are less cr itical to its survival. The main benefits of outsourcing can be summarized as stated by Corbett (1995) , and they are: improved business focus, access to world-class capabilities, reduce d cycle times and improved quality, sharing risks and costs in new technology. Other main benefits are reducing
23 operating costs, converting capital investment in non-core functions into operati ng expense and gaining better control integrated tasks. Buzzel (1983)  has studied 1649 manufacturing units from the Profit Impact of Ma rket Strategy (PIMS) database. His research shows that either a very hig h or a very low level of vertical integration yields an above average rate of return while ea rnings are lowest in the middle, and he recommends vertical integration only when a company needs savings as wel l as high control over its tasks and services. A measure of integration is given by the value added to sales ratio. According to him, the advantages of in-sourcing are lower transaction costs, supply assurances, improved coordination, and lower uncertainty. The disadvantages ar e capital investments, unbalanced throughput, reduced flexibility and a loss of special ization. DÂ’Aveni and RavenscraftÂ’s (1994)  work on the benefits of vertical in tegration support BuzzelÂ’s findings by showing that vertical integration can reduce total co sts by avoiding the transaction costs associated with market transactions, combining admi nistrative functions previously performed separately, and providing better information about costs. Lik e Buzzel, they also point out that integration can increase transaction costs in the form of costs for coordination and production. The additional coordination of activities required in integrate d organizations may increase overhead. They state that production costs may increase becaus e of the lack of market pressure to improve the efficiency of internal processes and employees, low er economies of scale, or failure to innovate. Other costs that increase are the costs needed to moni tor market information, manage inventories, and plan and schedule activities. In closing, we see that higher asset specificity in all its forms is generally associated with greater vertical integration due to higher transaction costs of outsourcing the se specific tasks. Other factors that influence the transaction costs and organizational form are bounded rationality, frequency and uncertainty. Thus, in making decisions regarding organizational form, it is important to consider not only the actual cost of the good or service in each case, but also the transaction cost factors that will be most prominent in that scenario, a nd the level of these costs when managing the transaction both internally and externally.
24 2.2.6 Empirical Measurement of Transaction Costs To the authors knowledge based on the literature review, the empirical e stimation of costs incurred due to transaction cost factors has been done only twice before, fi rst by Wallis and North  in 1986, who attempted to measure transaction costs of the economy over a 100 ye ars. However, they faced severe problems in defining and more so in measuring transa ction costs that are detailed in the methodology section. Their analysis concludes that the transaction sector is a significant part of the economy and grew from 25% to 40 % between 1870 and 1970. The second is by Masten Meehan and Snyder in 1991 , who provided empirical evidenc e of both the influence of each transaction cost factor but also provided cost est imates of each organizational form applicable to a shipyard involved in Naval construction pro jects. The problems faced by Wallis and North were mitigated by using switching regre ssion techniques. This was done using probit regression models to compute the effects of each fa ctor on the form actually adopted. A problem of selection bias was encountered for the second stage c ost calculation since the efficient organization structure is chosen, th e other forms are not observed, for which the Heckman two-step procedure and correction factor was used to eli minate the selection bias as outlined below. The structural equations were estimat ed as censored regression models analogous to the way actual and reservation wages are estimated in l abor supply applications. From this technique they obtain actual dollar estimates of transaction costs and can therefore estimate the magnitude of individual coefficients and not just their relative impact. They found that transaction costs account for 14% of the total value of all c omponents analyzed and that costs of the components made internally would rise to three times the ac tual were they to be outsourced. Integration of the contracted components would lead to a 70% increase i n transaction costs. Masten  has applied the above methodology to assess the performance i mplications of governance choices and its effect on business performance. Also reported is the fact that the transaction cost factors mainly af fect the costs if internal organization, rather than market costs as is normally assumed. Thus, the importance of organizational form is substantial.
25 2.3 Disease Management Literature Review Disease management is quickly emerging as one of the most important new are as of medical management as noted by Quilty and Lewis in the article Case Studies in Dis ease Management in. Medical Interface Magazine . The World Health Organization (W HO) estimates that chronic diseases make up 60% of the global disease burden, which is expected to rise t o 80% by the year 2020 for developing countries . For the United States, the Centers for Dis ease Control and Prevention (CDC) estimate the total cost for diabetes as $137.7 billion i n 1995. According to Thompson, Edelsberg, Kinsey, and Oster , nearly half of the American workfor ce is either overweight or obese. Americans with chronic conditions account for 75% of t otal healthcare costs . Chronic illnesses are the major cause of morbidity in the United Sta tes, and due to the increase of senior population in the country, the prevalence of these conditions is bound to increase. At the same time the effectiveness of the U.S. healthcare s ystem in providing care for this segment of the population has been wanting. The medical system has also com e under harsh criticism for rising costs and large deviation from best care pract ices, which is to say the treatment and care which is most suitable for the affected person at t hat particular time. As noted by Wheatley (2002) ,Â” amid rising healthcare expenditures and declining tax revenues state efforts to expand access to health insurance coverage have been put on hold in m any parts of the country. Recently, states have had to take a number of difficult steps to redu ce program expenditures, including restricting eligibility, reducing benefits, an d cutting provider payments. These measures generate cost savings but also restrict access to ca re. Another option, which is now being more widely adopted by states, is to develop disease management (DM) programs that are designed to contain costs by improving health among the chronically ill. Disea se management programs are meant to benefit both the medical insurance organization and t he consumer/patient by containing costs by improving health among the chronically ill. More than 20 sta tes are now engaged in developing and implementing disease management programs for their primary care case management and fee for service populations . The popularity of diseas e management springs from the fact that the proactive management or prevention of chron ic conditions presents the single largest opportunity to improve health and reduce healthcare costs. Disease Management programs focus on patient identification, monitoring a nd early intervention. This shifts healthcare expenses to less invasive and expensive care, thu s, disease management programs are meant to strive to achieve two seemingly conflicting goals : improving health care
26 while achieving cost savings at the same time. These programs work by drawing on the commitment and self-interest of patients, expert coaching, monitoring and trea tment by experienced nurses. The treatment guidelines are grounded in evidence-base d medicine. These resources are deployed to monitor patientsÂ’ conditions and coordinate trea tments with the physicians in various settings and diseases. According to Lewis , in disease management, the word intervention can loos ely be defined as Â“that set of products, services, education, expert resources and data offe red to the patient, patientÂ’s family/caregiver, and/or provider in order to reduce the l ikelihood of acute exacerbations and complications and/or to improve the baseline health status of the member overall.Â” The term intervention is also used in medicine generally, but is done so to defi ne the medical treatments given to a patient. The main difference between a m edical intervention and one through a disease management program is one of duration. A medical interve ntion is usually a treatment, procedure, medical test or therapy, a disease management in tervention consists of patient monitoring, follow-through, support and assistance and outcome reporting. Int erventions done through these programs can have many points of contact and changes as per the condition of the individual, and may last a lifetime . These programs tackle critical factors that have the greatest influence on quality of life, health and associated costs for most of the populations, especially the chronical ly ill segment. Currently many DMOs and health plans have overhauled their programs to manage co-morbid pa tients, i.e. patients with two or more chronic conditions. 2.3.1 A Brief History of Disease Management According to Boston Consulting Group (BCG) ; the earliest known impleme ntation of disease management was the launch of blood glucose monitoring (BGM) units to diabetes pa tients in the 1980s as this required significant education and monitoring of patients along wi th the setup of the required infrastructure, and the mindset of both patients and doctors needed to be modified. This was followed by the first wave of DM programs in the early 1990s, mainl y supported by pharmaceutical companies. Pharmaceutical companies supported these prog rams as they knew that prescription drugs help keep diseases in check and would reduce or minimize hospitalization, particularly in chronic conditions. Many health plans were skeptical, a s they view it as a ploy to
27 sell more drugs. Another limitation they had was that these first-gener ation pharmaceutical company-sponsored programs came with too many formulary constraints . Thus, most of these programs had closed by the end of the decade. A second wave started in the middle of the decade when entrepreneurs began to wor k to serve the large demand for disease management services, which required speciali zed technology, data mining and management. These early DMOs usually focused on a single diseas e at a time, and recently there has been a change towards managing co-morbidities, especi ally in the case of Medicare and Medicaid. The latest and current wave of disease management has been fuelled by the heal th plans as they have widely embraced these programs, and support and provide disease management p rograms either internally or through contracting with external vendors. Today many hea lth plans are working to integrate these programs into other aspects of medical mana gement such as wellness programs. The accreditation by the National Committee for Quality Ass urance (NCQA) has helped in wider acceptance of these programs as well. 2.3.2 Current State of the Disease Management Industry According to the BCG report Realizing the Promise of Disease Manageme nt , published in Feb 2006, today DM enjoys widespread use amongst the majority of U.S. health plans. According to the above report, out of the 120 health plans assessed from the 150 tota l in the U.S, all but 4 offered DM programs, meaning that 96% of the American health pl ans survey offered disease management programs. DM is now viewed as a competitive necessity according to more than 80% of the de cision makers in the health plans studied by BCG in the study noted above. 72 of the 120 health plans surv eyed stated cost savings as their reason for disease management program i mplementation. DM programs are widespread today even though there is marked uncertainty about r esults, savings and outcomes measurement methodology, and DM vendors or disease management organi zations (DMOs) have enjoyed rapid growth over the last decade.
28 The Disease Management Purchasing Consortium  estimates that DMO revenues have increased from $78 million in 1997 to almost $1.2 billion in 2005, which gives a compound annual growth rate (CAGR) of 40%. The revenue is expected to grow to $1.8 billion in 2008, with the growth coming from Medicare and Medicaid. Today many businesses offer DM services, and most DMOs have expanded beyond their focus on a single disease. Huma na is generally acknowledged as the leader in disease management . Among st DMOs, the top five based on market revenues are: Healthways, Health Dialog, CorSolutions, Lif eMasters Supported Healthcare, and Matria Healthcare. DMOs have also diversified int o informatics, where they sell data and analysis tools for employers to allow them to assess their employ ee health and health plan performance (SourceDMPC) . The BCG report Realizing the Promise of Disease Management  finds that health plans are almost equally as likely to develop and run these DM programs internally a s they are to contract with external DMOs to purchase DM services, given the situation within the or ganization and associated transaction costs. Another option available to them is the combinat ion or hybrid approach, where some health plans combine internal and external resource ssuch as in-house nurses and purchased software in order to execute DM programs. Although private U.S health plans are the largest implementers of DM pr ograms today and majority of employers access DM through these health plans, several o ther sectors such as the direct-to-employer segment is a rising trend and this segment is growing ra pidly. Employers are also taking an active interest in managing and coordinating employee health pl ans and disease management programs and frequently request it when contracting with a heal th plan. Large employers are also likely to contract separately with a DMO for dis ease management programs separately from their health plan. These employers usually have multipl e health plans and seek a single DM benefit that they can apply across the organization for thei r employees. For large employers, disease management is growing in importance because they increas ingly see the value of such programs in reducing absenteeism and short-term disability expense, not t o mention employee morale and retention . Federal and state governments are also getting heavily involved in DM usi ng the Centers for Medicare and Medicaid (CMS) pilots to implement these programs. In a ddition, Governments abroad are showing an increased interest in this sector. Given the above, w e see that disease
29 management programs are usually implemented for diabetes, asthma, corona ry artery disease (CAD), congestive heart failure (CHF) and chronic obstructive pulmonary disease (COPD), which are known as the five core chronic diseases. The number of health plan s that offer all five programs represent only 21% of the total, as can be seen from figure 2.3. Also, the number of health plans offering these programs for other chronic diseases such as end-stage renal disease, lower-back pain and cancer are low. Figure 2.3 Percentage of DM Programs Offered by Health Plans in the U.S. Source: BCG Landscape database, Feb 2006.  The highest governing body overseeing all the organizations in the U.S is the Di sease Management Association of America (DMAA) . DMAA is a non-profit assoc iation that represents all stakeholders in the DM community. The association does this through public and private advocacy by targeting the healthcare industry, government agenc ies, employers, and the general public to educate them on the important role DM programs play in improv ing healthcare quality and outcomes for chronically ill patients . The components of disease management as defined by the DMAA  are: 92 68 64 23 21 0 20 40 60 80 100 DiabetesAsthma Cardiac conditions (CAD and CHF) COPD All 5 conditions
30 1) Population identification processes; 2) Evidence-based practice guidelines; 3) Collaborative practice models to include physician and support-service p roviders; 4) Patient self-management education (may include primary prevention, beha vior modification programs, and compliance/surveillance); 5) Process and outcomes measurement, evaluation, and management; 6) Routine reporting/feedback loop (may include communication with patient, physicia n, health plan and ancillary providers, and practice profiling). Full-service disease management programs are those that include all s ix components. Programs consisting of fewer components are known as disease management support servi ces. Traditionally, disease management has focused on the big five chronic disea ses: ischemic heart disease, diabetes, COPD, asthma and heart failure. Disease management pr ograms generally are offered telephonically, involving interaction with a trained nursing profes sional, and require an extended series of interactions, including a strong educational element. Pat ients are expected to play an active role in managing their diseases. Because of the presence of co-morbidities or multiple conditions in most high-risk patients, this approach may become ope rationally difficult to execute, with patients being cared for by more than one program. Over time the industry has moved more toward a whole person model in which all the diseases a patient has are managed by a single disease management program (Source-DMAA) . As stated by the Disease Management Purchasing Consortium (DMPC) , d isease management requires a comprehensive clinical and economic understanding of a di sease state that can only be developed through a team approach. Clinical input is required to desig n the interventions, identify patients, and understand the impact of co-morbidities Information systems input is required to integrate the disparate data bases of medi cal information for a particular disease. Legal and network development assistance is req uired for contracting and to understand how disease costs are impacted by capitation arrangements. And once a program is developed marketing support will be required to develop physician communication materials. According to Managed Care magazine , a typical disease management pr ogram consists of the following teams:
31 Program administrators: These are the individuals who run a health pl an and are the best source of information on organizational structure, goals and expectations, pay and ince ntive programs, and fiscal commitments to the disease management program. The success of s uch programs is dependent on the support it receives from the administration during its developm ent and implementation. Pharmacists: Academic and professional training in pharmacotherapeuti cs and pharmaceutical care empowers pharmacists to play a critical role in disease manage ment. Pharmacists in highly integrated managed care settings participate in formulary decision s, drug treatment protocols and critical pathway design. Pharmacists in disease management programs a lso perform the following activities Â– 1) Patient education concerning drug use, especially in high-risk/high-use c ases. 2) Compliance education and monitoring for selected populations. 3) Disease state monitoring (blood glucose, blood pressure, serum cholesterol etc.). 4) General wellness education. 5) Intervention with physicians to encourage drug protocol adherence. Information managers: Data analysis plays a critical role in designing and operating a DM program. For the implementation of these programs, algorithms, based on specif ic correlates of drug, diagnosis, procedure and specialist codes are needed, to query claims data in order to identify the health planÂ’s members who have the diseases in question. As a r esult of this level of specificity, the entire population with these diseases can be identified. Ba seline measurements are necessary for later comparisons to ascertain whether care has been im proved and costs have been controlled. Information managers help the planning team decide on data formats and definitions. They determine the usefulness of current information systems and also pr omote exchange of appropriate data elements among the partners. Continual improvement of the i nformation systems used in the programs is necessary in order to capture and track data used in outc omes research and the information required for future improvement. Finance managers: The programs finance team is needed to analyze curr ent costs of care, including the costs of failing to achieve intended outcomes and the predicted f inancial consequences of the disease management program. In addition, this team is responsible for
32 negotiating contracts among the disease state management partners, and fo r clarifying arrangements among them with regard to risk sharing and capitation. Florida operates the largest (and one of the oldest) Medicaid diseas e management programs in the country, which was initiated in 1998. Florida has the fourth largest Medicai d population in the nation, with 2.1 million eligibles and $8.8 billion spending in FY 00-01; $9.9 billion appropriations for FY 01-02; $ 11 billion FY 02-03, and $13 billion in FY 04-05 . The Florida disease management program is the most comprehensive disease m anagement program in the nation for Medicaid recipients [68, 1]. The diseases covered by Med icaid DM programs are asthma, HIV/AIDS, CHF, hemophilia, ESRD, diabetes, hypertension, pre-diabetes a nd depression. In May 2001, a Florida legislative audit was released which cri ticized the DM program for not being close to producing the projected savings of $113 million over the period of 1998 to 2001 as was initially expected. It has also been found that while the DM prog rams generally reduced inpatient hospital costs, produced improvements in patient care quality and led to a reduction in spending, these reductions were generally offset by DM progr am costs . Table 2.1 shows the most popular disease management programs in the country, w hile table 2.2 reports the tools used for their implementation.
33 Table 2.1 Disease Management Program Statistics Across the U.S. Source: Managed Healthcare Executive; Apr 2006.  Disease management programs offered by HMOs: Disease state Percentage of HMOs offering programs Diabetes 81.5% Asthma 79.6% Cardiovascular disorders 64.7% High Â– risk pregnancy 31.4% Hypertension 20.0% COPD(chronic obstructive pulmonary disorder) 16.6% Multiple sclerosis 8.4% HIV/AIDS 7.7% Gastrointestinal disorders 4.3% Hormonal therapy 1.9% Other disease management programs offered 45.6% Top three other programs offered: Low Â– back pain 32.1% Smoking cessation 22.6% ESRD (end stage renal disease) 22.2% Table 2.2 Method of Disease Management Program Implementation Source: Tracy Walker, Managed Healthcare Executive; Apr 2006.  Implementation tool Percentage of HMOS offering service Patient education newsletter 71.4% Physician education newsletters 59.6% Information on web site 54.0% Patient education courses 48.4% Reminders at physician visits 38.5% Physician education courses 34.8%
34 The investment required in terms of capital and human resources is thus signi ficant when it comes to the implementation of disease management programs. A healthcare or ganization has the choice of implementing such programs itself, or contracting them to outside di sease management vendors. In order to remain profitable and financially viable while upholding t he principles of disease management and reducing healthcare costs, a medical care provider m ust develop effective strategies, as noted by Einstein . Outsourcing of these progr ams to disease management organizations (DMOs) is one strategy that is widely prac ticed. 2.3.3 Effect of Transaction Cost Factors on DM Organizational Form We can see the effect of TCE factors on health plans in the survey conducte d by BCG in February 2006 . As seen in figure 2.4; health plans are as likely to integrate DM pr ograms as they are to outsource them to a DMO. One way larger health plans have integrated thei r disease management programs is by purchasing the DMO outright. For example, Wellpoint ha s purchased Health Management Corporation and UnitedHealth Group has purchased the D MO Optum. Health plans such as Cigna have contracted with DMOs, while others s uch as Kaiser Permanante have a completely integrated approach. The decision on organizat ional form, according to the February 2006 BCG report, is made at an individual level by eac h health plan. They state Â“itÂ’s not the payerÂ’s size but the perspective of senior ma nagement that largely determines whether the payer develops its own DM programs or turns to the ma rket for external options.Â”
35 Source: Realizing the promise of disease management, BCG, 2006.  The BCG report continues, Â“they (health plans) recognize the capabil ities required to implement the approach and the difficulties involvedÂ”, which can be interpreted as as set specificity in the form of technology and software. Also, they go on to say, Â“[Health plans] view disea se management as a highly specialized set of skills that are difficult to master or replicate at low cost. Some payers may view disease management as so central to their bus iness that they will make every effort to make or bring the approach in-house. Others may feel tha t they cannot afford the fees associated with outsourcing or they can best limit the ir expenditures by relying on an internal or assembled programÂ” . This shows that human and physical asset specificity plays a large part in determining the organizational form for these programs. There is uncertainty regarding the savings for the health plan with thes e programs, and the savings for different programs can be realized at different times. Figure 2.4 Organizational Form of DM Programs in Health Plans 50 17 33 65 33 2 41 50 9 12 88 38 62 47 49 4 0% 20% 40% 60% 80% 100% < 100,000 lives 100,000500,000 lives 500,0001 million lives 1 million5 million lives >5million lives Overall Internal DM Outsourced DM No DM
36 For example, end-stage renal disease savings can begin in as few as 45 day s , whereas a Healthcare Business roundtable consensus showed an average interval befor e savings are realized to be approximately 18 months for other programs . In addition, they BCG report shows that due to lack of a standard methodology in order t o measure effects and outcomes, health plans face uncertainty in terms of measuring results, and by choosing an appropriate organizational form, they try to minimize the trans action costs associated with this uncertainty. The biggest obstacle in the path of disease managem ent is that no standard methodology exists for measuring savings and outcomes. The results reporting done for disease management programs consists of usually three outcome measuresÂ—proces s outcomes, i.e. (Did the compliance rate go up?), health status outcomes, i.e. (Did ER visits decline? D id selfreported health assessment scores improve?), and member satisfaction. How ever, due to no set standards across the industry, the methods used to measure these usually vary from organization to organization. A common mistake is the first is regression to the me an. Any disease management program which starts with last yearÂ’s high usersÂ—a common start ing point in asthma and CHF disease managementÂ—will automatically show improvement simply because few diseases progress linearly. Although various industry groups such as the Disease Management Associati on of America (DMAA) and the DMPC  have issued guidelines, there has not been an agre ement in terms of adopting a particular methodology, which introduces uncertainty and increases the transaction costs of implementing the program in each organizational form as Â“each p ayer will need to examine a variety of issues, such as the magnitude and reliability of its savings measurementsÂ”, and Â“we expect other payers to find disease management so resource intensiv e and difficult to manage effectively that they will turn to DMOs when their serviceddelivery or internal outcomes prove unsatisfactoryÂ” . Uncertainty is also stated as the risk of failure for a disease management program implemented by a health plan, which would cause a setback to t he company. Additionally, health plans are seen to look for Â“common vision and committ ed leadershipÂ” while searching for an appropriate DMO, which is an exampl e of brand specificity. Another view of brand specificity is given in the DMPC report Outsourcing : Lessons Learned as Â“examples of favorable first contracts would be NYLCare-AirLogix, Foundati on-Vivra Specialty Partners, Humana-Ralin, Humana-Paidos, Humana-Baxter, Principal-Accordant, and a large
37 number of health plans fortunate enough to receive programs that were liter ally given away, no strings attached, in order for a vendor to start generating experience a nd outcomes.Â” On the surface, it might seem risky and problematic to contract with a new or inexperienced vendor for disease management programs. However, due to the relatively new nat ure of the industry and the unique requirements it entails, new/inexperienced vendors h ave actually shown better performance as compared to established vendors/DMOs, as noted by the DMPC. As an example, Apria was an established DMO with a vast experience in asthma, and Stua rt Disease Management Services (financed by Zeneca), were handling programs for va rious national health plans, but both pulled out of disease management and left their customers (the hea lth plans with which they were contracted) in the lurch with what are now essentiall y orphan disease management programs . The above shows the pitfalls of stressing on brand name and reputation and its effect on transaction costs to the level that the programs failed. Frequency is reported as the number of interventions as well as the ret ention and penetration among its customers by the DM program. Most health plans screen all policy hold ers for program eligibility using their preferred algorithms, which take into account t he medical history and risk of the individual. If eligible, the individual is enrolled into the program at no expense or for a small monthly fee. The person is free to opt out of the program at any time. D ue to this, the number of people enrolled in a particular program is always in flux. The adheren ce of the patients to the program protocol is also something that needs to be constantly monitored and he nce the frequency of contact within a program can vary significantly based on the cha racteristics of the people enrolled. Hence, this factor also plays an important role in the f inal form adopted by the health plan for these programs. According to the DMPC report Outsourcing: Disease ManagementÂ’s Ma gic Bullet (1999) , Outsourcing is not always the answer for health plans any more than build ing programs internally is always the answer. Many health plan medical directors are given direc tives along the lines of: Â“You have to institute a disease management program, and you have to do it within your existing budget.Â” . Thus, there are many variables that influence internali zing or contracting a particular program in order to maximize benefits and profits.
38 From the above, three main factors can be used to distinguish between those dise ase categories and health plan circumstances which should lead to a buy decision and those which should lead to a build decision : 1) Health plan organization, culture, and budget 2) Severity of disease disease management programs which look like a ty pical health planÂ’s day-to-day operations can be successfully built by most health pla ns, but those which require a set of skills not normally found within a health plan are bett er served through outsourcing to an expert vendor. This shows how similarity may affec t the organization form of these programs in health plans. 3) Availability of tools and expertise The more widespread the expertise and tool s available for patient management in a particular category, the easi er it is to build a program. For instance, health plans often build their own prenatal care progr ams, using readily available scripts to help their call center nurses triage pregnant members to identify those needing the most attention. The experience base in pregnanc y management is built on close to 4,000,000 US pregnancies every year. Rare diseas es by definition lack that experience base, and hence expertise and tools are much ha rder to find. For instance, the nationwide experience base for hemophilia is built on on ly 20,000 patients. In the case of rare diseases, a health plan can spend more time j ust trying to assemble the requisite tools itself (assembling the tools being a sma ll piece of the overall disease management program) than it would spend creating an entire program t hrough an outsource. However, Evaluating and selecting vendors, contracting, and claims a nalysis require some effort and expertise. If integrated, a health plan would need t o purchase its own retrospective claims analysis/predictive utilization softwar e. Such a tool can help identify tomorrowÂ’s high users (the people one wants in a disease management pr ogram) as well as ones from previous periods. Such software, such as CodeReview, is he lpful but not exhaustive. Several vendors have very sophisticated algorithms, supervi sed by medical directors, to find opportunities which software alone can overl ook, and they guarantee significant amounts of savings. The above statements again show how asset specificity in the form of physical a nd human asset specificity affect the form chosen by health plans for these programs.
39 It is also reported by Matheson, et al.  that Â“health plans actually make the build-or-buy decision on a condition-by-condition basis. Harvard Pilgrim exemplifies this appr oach, by having internal programs for some conditions, such as asthma and diabetes, while contra cting with one DMO for a cardiac program and another for rare diseases. Furthermore, some pa yers blend inhouse resources and external services in the same program, for example, us ing in-house nurses in coordination with data analytics purchased from a vendor.Â” This shows t hat the transaction cost factors will affect each program in a health plan differently, leading t o different organizational forms for each as the situation demands. They recommend that DMOs reduce tr ansaction costs for health plans by Â“more effectively targeting and communicating to empl oyer groups and health plans, and differentiating and marketingÂ”. Employers and health plans are alr eady requesting customized reporting on the outcomes of the DM programs, with greater detail i n savings and health improvements which reduce uncertainty at the cost of higher trans action costs. They state that using efficient disease management programs, health plans and employ ers can leverage them strategically in order to build a competitive advantage. The most importa nt element to make this possible is that Â“they should strive for excellence in the management of administrative and information technology costsÂ”, both of which are components of transaction cos ts. According to an article in Disease Management News , Â“Creating a s uccessful disease management will require senior management commitment and dedicated resou rcesÂ”, and that Â“(disease management) programs are difficult because they require a n unprecedented level of coordination, communication, and synthesis of information.Â” Both internal and contra cted programs require time from senior management and commitment of capital and r esources to be successful. In the case of contracted programs, it is seen that there needs to be close communication and information flow between many departments of both the health plan a nd the DMO to build a successful disease management program. The information sy stems department in both firms in particular, needs to have a close bond in order to develop the outcom es tracking and reporting functions. The medical directors of both firms also need to w ork together on the program protocols and integrating the program with the case management f unction . Thus we can see that transaction costs are very prominent in the implementation of these programs and it is imperative that the organization choose a form as to minimize these cos ts. We can conclude that all the major transaction cost factors which are as set specificity, frequency, and uncertainty will play a part in the final form adopted by a particular hea lth plan for these programs.
40 2.3.4 Future of Disease Management Disease management is also expanding worldwide, especially in Europe and Asia, due to its rapid growth in the U.S., particularly in the Medicare and Medicaid sectors Australia has implemented many DM pilots recently, and Singapore has invested significantly in DM. ot her countries implementing DM are Brazil and South Africa, whereas the United Kingdom and the Calgary health region in Canada are developing initiatives in DM . Most of the DMO s and health plans are also looking to apply DM to additional areas such as obesity, cancer, and other cardiac conditions as they seek to achieve additional savings and meet employer demands for these programs, according to Matheson, et al., in 2006 . Also, they are counting on inc reasing the number of people being covered by these programs, mainly by going deeper into the ris k categories for each condition. Most health plan executives and decision make rs view the DM industry to be in its growth phase . Figure 2.5 shows the areas most likel y for expansion and program development in the near future. Figure 2.5 Percentage of Health Plans and Respective Anticipate d Areas of Expansion for DM Programs Source: Market data, BCG, Feb 2006.  50 25 17 13 13 0 10 20 30 40 50 WellnessObesity CancerRenal disease Depression
41 2.4 Selection Bias and the Heckman Two-Step Method There are two forms of the selection bias problem. In the standard case of s election bias, information on the dependent variable for part of the respondents is missi ng. In the other version of the selection bias problem, information on the dependent variable is av ailable for all respondents, but the distribution of respondents over categories of the indep endent variable we are interested in has taken place in a non-random manner. Common to both forms of selection bias is that there is a selection process by which data is divided over two (or more) groups and that non-randomness in this process disturbs the estimation of other relationships which are of substantial interest. Th us, as described by Smits , there are two processes (which can be described with two equations, c alled selection equation and substantial equation) and these processes are related to each o ther. This relationship will be reflected in a non-zero correlation between the error terms of the equations. If such a correlation is present, we cannot estimate the substantial equation wit hout taking the selection process into account. The Heckman two-step procedure can also be used to address both the forms of selection bias, and is taken from the classical papers of Heckma n (1979, 1980). This method was first derived by James Heckman in 1979 . In this paper, the bias that results from the usage of non-randomly selected samples to estimate behavioral re lationships as an ordinary specification error or omitted variable bias is discussed. The specification error framework is assumed to be the same as that specified by Griliches  Breen , and Theil . He states that sample selection bias may arise for two reasons. Fi rst, there may be self selection by individuals or data units being studied. Second, sample selection de cisions by researchers may lead to this bias. Using a computationally tractable te chnique, a simple consistent two stage estimator is considered that enables analysts to utilize simple regression methods to estimate behavioral functions using least squares method. The asymptotic d istribution of the estimator is also derived. In the first step of the Heckman procedure, the selection process which is r esponsible for selection bias problems is studied with the so-called selection model. For this purpose, generally a probit model is estimated (as the error term of this model is normally dist ributed, one of the assumptions underlying the Heckman model).
42 Next, the residuals of the selection equation are used to construct a sel ection bias control factor, which is called lambda. This factor is a summarizing measure which re flects the effects of all unmeasured characteristics which are related to the selection decisi on. Lambda is called the inverse mills ratio and is denoted as: f(z)/F(z), where z is the estim ated value from the probit equation and Â“fÂ” and Â“FÂ” denote the standard normal density and distribution functi ons, respectively. The value of this variable for each of the respondent s is saved and used as an additional variable. In the second step of the Heckman procedure, the main analysis is performed, in thi s case an ordinary least squares (OLS) regression analysis of the effects o f sourcing decision on costs. In this substantial analysis we use the selection bias control factor calc ulated above as an additional independent variable. Because this factor reflects the effect of a ll the unmeasured characteristics which are related to the dependent variable of the initial model, the coe fficients of this factor in the substantial analysis catches the part of the unmeasured characterist ics related to the dependent variable in the secondary equation. Due to the presence of a control factor (la mbda) in the analysis to compensate for the unmeasured characteristics of the depende nt variable, which is also related to the dependent variables in the (initial) selection model the predictors in the equation are freed from this effect and the regression analysis produce s unbiased coefficients. This method was first applied by Hanoch  in labor applications. In this industry wages are observed only for those who actually work. However, one can infer from the decisi on to work and characteristics of the working laborers the reservation wage that m ost likely generated the pattern of observed employment and the observed wages at that time. Heckman has applied his own methodology in his 1980 paper . Here, he presents an empirically tractable model of the life cycle labor supply decisions of married women in an environment of perfect certainty. He integrates two distinct dimensi ons of life Â– time labor supply: annual hours worked and annual participation in the work force using his two Â– s tep approach, and using eight years of panel microdata from the Michigan panel S urvey of Income Dynamics in order to estimate the model. Thus he extends the work done by Hanoch abov e, as that has stated only hours per week and hours per year as the two arbitrary dimensions.
43 He finds that labor supply is inversely related to lifeÂ–time wealth mea sures, children affect lifeÂ– time labor supply decisions, and that future values of variables determine c urrent labor supply decisions. The usage of this methodology in this research has been detaile d in the methodology section. From the above literature review it seems very essential that furth er study of the sourcing decision of these programs be conducted. The proposed project builds on researc h in the study of factors affecting the outsourcing of disease management programs in a me dical insurance organization. It focuses primarily on using transaction cost economics as a framework for better understanding the sourcing decisions and the internal organization costs and the external market costs that lead to this decision.
44 Chapter 3. Literature Summary and the Hypotheses 3.1 Literature Summary and Application to Health Plans As the number of topics related to both transaction cost economics and diseas e management is very large, the researcher acknowledges that the literature revie w is not exhaustive, however the literature reviewed is sufficient to get a grasp on the key iss ues with which the research is concerned. These have been summarized here and used as a basis for the hypothes is detailed in this chapter. We can infer from the literature review that all types of asset spec ificity, uncertainty and frequency affect the levels of transaction costs and hence affec t the organizational form. Bounded rationality also places limits on the organizationÂ’s ability to c omplete all activities internally or outsource completely and foresee and contract for all possibl e contingencies. Firms internalize their most important tasks and personnel to control quality and pr oduction, ensure access to scarce inputs, and have a better understanding of complex production/ service techniques and technology. Based on the particular situation, firms should only integrate transactions that they can perform more effectively in-house than through contracting. This i mplies that if the total cost inclusive of the costs of selection, contract management, performance measurement, and dispute resolution are less than internal costs of providing the same g ood or service, then it must be outsourced, as the associated transaction costs are lower in that case. According to transaction cost theory services formerly performed int ernally will tend to be outsourced if 1) the scale at which the service is performed efficien tly increases relative to demand and 2) if the service becomes more standardized, less customer specif ic or more widely used.
45 Therefore, for an organization considering outsourcing there is not one clear ans wer regarding organizational form. It depends on the type of transaction and the specific c onditions and factors that influence the organization and the industry. 3.2 Application of TCE Factors to Health Plans Transaction cost analysis has been applied to various manufacturing app lications, which deals with continuous processing of a large quantity of material as they move from one processing station to the next, and construction industries, which involves the building of a si ngle or unit at a fixed location, and the finished unit may or may not be made up of a small number of finishe d units. The various transaction cost factors and their effect on the organ izational form in the healthcare sector can be hypothesized as follows: In manufacturing, physical asset specificity is usually higher due to the high volume of production and the portability of the finished goods, compared to construction project s, where the final product is unique or produced in limited quantities, but the assets them selves are multipurpose and mobile. Disease management programs are mainly concerned with the monitoring of the individuals enrolled, which requires advanced software a nd computing power, and the provision of timely information to both the patient and the physician ( which is done through various means of communication), hence, physical asset specificity is likely to be an important factor in the determination of the organizational form of a di sease management program. We state hypothesis 1 such that integration of disease management programs becomes more likely as physical asset specificity increases. Temporal specificity does not play a major role in the organizational form for manufacturing operations as it is of a high volume and continuous nature, whereas in the construct ion field a delay at one stage can reverberate through the entire project, and thus is more important in this application. The same can be said of the disease management, as it requi res the timely dissemination of medical information both to the patient and to the physician. A delay in this regard could potentially lead to serious consequences to the afflicted per son, and to the organization in the form of treatment costs, and hence this factor is lik ely to play an important role in the arrangement of the firm. We state hypothesis 2 such that integr ation of disease management programs becomes more likely as temporal specificity inc reases.
46 The factor site specificity can be explained as the distance between the interacting firms, Or Transportation and inventory costs specific to the transaction. Disea se management programs are mainly concerned with the timely disposal of critical information to patients and physicians and coordination of medical services and tasks between the providers and the pat ients in order to provide best evidence care, and to make the patients active participants in their own care. Thus, these programs are not involved in delivering specific services or compone nts at specific sites or individuals; hence, this factor is hypothesized to exert a very low influe nce on the outsourcing decision, and has thus not been considered in the empirical analysis. We stat e hypothesis 3 such that site specificity does not play an important role in the determination of the organizational form for disease management programs. The factor dedicated assets can be defined as substantial, general-purpose investments specific to the transaction, and that need to be invested in for the proper completion of t he transaction or service, or high-capacity equipment whose capacity is intended to be dedic ated to a particular customer. In this context, dedicated assets may refer to capacity that is created to serve particular/specific customers, so that it would be difficult to find al ternative customers, or an alternative use for the capacity created. Here, the effect of th is factor will depend on both the disease being monitored and the size of the population enrolled. We state hypothesis 4 such that outsourcing/contracting of disease management programs will be more like ly as the dedicated asset specificity rises. The factor human asset specificity is generally not important in the ma nufacturing area due to the generalized and labor-intensive nature of the tasks involved. In the const ruction field, this factor may vary in importance, while generally it mirrors the construction fi eld and the importance of this factor is low, however, there may be some construction applications (s uch as naval shipbuilding) may require specialized knowledge and skills, which increases the influence of this factor over the firm. Similarly, Human asset specificity will most likely exert a big influence over the organization structure as the experience, knowledge and skills needed f or managing and running disease management programs are very specialized and specific. Usua lly, only experienced medical professionals (physicians and nurses) make up any give n disease management team. We state hypothesis 5 such that outsourcing/contracting of d isease management programs will be more likely as human asset specificity ri ses.
47 Uncertainty or Complexity also has a role to play in disease management programs. Disease management programs are generally very complex and require advanced know ledge of medical protocol, treatments and procedures. The symptoms and issues of the enrolled peo ple will differ from person to person and this will lead to a degree of complexity much higher the n that found in either the construction or the manufacturing areas, and is highly likel y to play a major role in the form of the organization. We state hypothesis 6 such that programs with lower uncertainty and complexity will tend to be integrated while those which entail higher unc ertainty and complexity will be contracted. Disease management programs consist of many highÂ–technology, medical knowledg e intensive activities, unlike construction and manufacturing operations, where labor i ntensive, low-tech activities make up the bulk of the work. The tasks involved will also vary significantly on a patient to patient and also on a program-to-program basis. Thus, similarit y in the disease management context is hypothesized to be low (between as well as within progr ams) and will likely play an important factor in determining organizational form. We state hypothesis 7 such that disease management programs similar to the ones already offered by the health plan are likely to be in-sourced, while those dissimilar to current programs will tend to be outsourced. In this context, frequency refers to how often contact is made with the pat ients for interventions relating to their specific conditions. In other industries, it is seen t hat increased frequency leads to a greater probability of outsourcing or contracting to external vendors. T he effect of this factor here is hypothesized to be similar, i.e. programs that require frequent cont act will tend to be outsourced. We state hypothesis 8 such that the higher the frequency, the higher the chances of the disease management program being outsourced or contracted. This study will focus on the factors outlined above and will involve collecti on of data based on the previously stated transaction cost factors as a means of constructi ng a probit regression model to study the effect of these factors on the form adopted by an organization fo r implementing disease management programs and to provide a dollar estimate of the costs borne by the organization. The proposed study will focus on health management organizations who have implem ented disease management programs both internally and through external vendors as a means to gather representative data based on the previously stated transaction cost fa ctors as a means of
48 constructing a regression model to study the effect of these factors on orga nizational form, organization cost, and the role played by the major transaction cost factors in disease management programs through out the country and to provide a dollar estimate of the costs borne by the organization Answers to the above questions can give a better insight into the issues of outsourcing from a health planÂ’s perspective. 3.3 The Hypotheses Based on the understanding and the appreciation of this literature stated above, the researcher formed the following hypotheses to be tested: 1) Transaction cost factors yield influence over the organization form of dis ease management programs in managed care health plans. 2) The transaction cost factors exert their principal effect on the cost s of internal organization, rather than external market costs. The researcherÂ’s primary and secondary data collection is centered on t esting these hypotheses. In order to test these hypotheses the researcher had to answer the four s econdary questions outlined below and explained in the measures, instruments, and data sources section: 1) What is the nature of organization adopted for the disease management prog rams implemented by various health plans in the country? 2) What is the impact of transaction cost factors on integration decision s for disease management programs? 3) What are the implications for designing regression models for predicti on of organization form and costs on the basis of transaction cost factors? 4) To analyze if selective organization leads to savings for the manage d care organization or health insurance organization.
49 Chapter 4. Methodology 4.1 Research Approach The research approach and methodology used in this thesis will be based on quantit ative data analysis collected by surveys and secondary data obtained from organizations in the health insurance industry such as managed care organizations, indemnity health plans Medicare, etc. The primary data and the secondary data will be collected by in-depth survey fr om the internal departments of willing health insurance organizations. 4.2 Research Method Transaction cost analysis of organizational form maintains the hypothesi s that the organization is so arranged as to minimize the cost of governing the transactions. The orga nization of the firm can be expressed as a binary variable, which is make or buy, that is, whether the component or service will be produced in-house or contracted/outsourced to an external ve ndor. There are two methodologies generally used for the measurement of transaction costs: Direct Measurement: the first and most straightforward way of pred icting the organizational form chosen would be by direct measurement and comparison of the costs, for example, if w e denote the form chosen as F*, a model of the choice between the two arrangements can be shown as: F* = Fo, if Co < Cm, and Â“FoÂ” represents the integrated form for the task or se rvice. = Fm, if Co >= Cm, and Â“FmÂ” represents the outsourced form for the task or service. where Â“CoÂ” and Â“CmÂ” represents the costs of internal production and market subcontracting respectively . However, many costs such as inflexibility or nee d of litigation may not be addressed. Also, the most basic and fundamental problem in this approach is tha t organization
50 costs cannot be observed for the organizational forms not chosen. For example if an organization chooses internal production and the associated costs are measured, the costs o f organization for the alternative form i.e. outsourcing cannot be measured as that organization al form does not exist. Thus, direct cost observation is not a feasible method for the appl ication of transaction cost analysis. In order to address this shortcoming, the following methodology has been adopt ed. Reduced form analysis: in this methodology, the transaction costs in each poss ible organization form are related to observable features and then predictions of final adop ted organizational form are made based on these features. Hence, the true costs of organization can be said to be: Co = AX + e, (1) Cm = BZ + u, (2) where X and Z are vectors of attributes (in this case, they are trans action cost factors) influencing the respective organizational costs, A and B are coefficient vect ors and e and u are normally distributed random variables. Thus, the probability of observing organization for m Â‘FoÂ’ becomes: Fo = Pr (Co < Cm) = Pr (e-u < BZ-AX). Thus, the comparison is now based on the signs and magnitudes of the coefficients A and B, and not on the direct costs (Co and Cm) themselves. However, if the variance o f the difference between the random variables (e-u) is not known, the coefficients of the above e quations can only be identified up to a proportionality factor. Additionally, if X and Z share e lements, only the differences between the vectors A and B can be identified . As a result, i t is not possible to deduce where the principal effect of the transaction cost factors lie on internal or market costs. In order to obtain stronger tests of the theory, the method given below will be us ed in this analysis. Two-stage analysis: as the name implies, this method consists of tw o stages. The first stage consists of the construction of a logistic or probit regression model as shown below. The logistic model takes the form:
51 Where Â“ Â” is the constant/intercept obtained from the model, Â“kÂ” represents t he numbers of the independent variables Â“xÂ”, which each can have Â“iÂ” levels as shown above. Â“ Â” is the parameter coefficients obtained for each of the independent variables from the model, and Â“p i Â” is the probability of the task or service being in-sourced. Thus, Â“p i Â” can be calculated as: here, Y = 1 for the in-house case, and Y = 0 for the contracted case. The parameters 1 ..., k are usually estimated by maximum likelihood. The probit model assumes that: where is the cumulative distribution function of the standard normal distri bution, Y is the binary outcome variable, and X the vector of regressors. The paramete rs are typically estimated by maximum likelihood. In this research we estimate a probit model in the fir st stage. In this case, only the costs of organization for the form actually adopted a re calculated. Thus, the model becomes: C = Co = AX + e, if Co < Cm, C = Cm = n.a., If Co >= Cm. In the second stage, switching regression techniques can be used to provide estim ates of the internal organization costs. Estimation of the equations as a censored r egression model will further reduce the need of large quantities of data. First, the inverse mil ls ratio (the ratio of the
52 probability density function over the cumulative distribution function of a d istribution) is calculated as M = f (z)/F (z), where Â“zÂ” is the estimated probit val ues from the model, f = the probability density function and F = cumulative distribution function of a dis tribution. The internal organization cost equation is constructed by regressing each t ransaction cost factor against our measured costs for in-sourcing. The equation for the internal organizational form is thus deduced by using the above equation and the inverse mills ratio. The transaction cost methodology described above was applied to a specific a pplication, which is, the outsourcing of disease management services by health plans and health mana gement organizations (HMOs) to disease management organizations (DMOs). Whe reas previous empirical research has dealt with manufacturing and construction applica tions, the process of disease management is quite different and removed from these, which in tur n influences the circumstances that lead to opportunism and affects the nature of the organ ization and the associated costs. 4.3 Design and Study Participants A linear regression model featuring the decision to integrate as the dependent variable and the various transaction cost factors explained above as independent variables will be constructed. Health management organizations, including Medicare and Medicaid which engag e in disease management plans and its outsourcing will be considered in this study. Initia lly, only those health management organizations situated and serving the population of Florida we re considered. However, in order to obtain sufficient data, the sample size was expanded to i nclude health management organizations from other states in the U.S. as well. After the selection of the health organizations, data was collected for t he construction of this model based on the disease management programs implemented for chronic di seases (see table 2.1). Pertinent data regarding any disease management program that was obtai ned was added to the construction of the model in addition to the basic five diseases. The c ollection of this data provides insight and better understanding of the effect that the considere d factors of transaction cost have on the final organization form and allows estimation of the organiz ation costs incurred with the current form and also under the other allowable alternative. Mo reover, this model shows
53 the importance of internal organization costs in the outsourcing decisio n, which has previously only been applied to the construction and manufacturing fields, and never to the d isease management field. 4.4 Measures, Instruments and Data Sources The inquiry was guided by four research questions: 1) What is the nature of org anization adopted for the disease management programs implemented by various health plans i n the country; 2) what is the impact of transaction cost factors on integration decisions f or disease management programs; 3) what are the implications for designing regression models for prediction of organization form and costs on the basis of transaction cost factors?; and 4) to analyze if selective organization leads to savings for the managed care organization or health i nsurance organization. To determine the nature of organization in the various managed care healt h plans in the country that implement disease management programs (Question 1), an analysis of var ious health management organizations (including Medicare and Medicaid) was conduc ted. The number and types of disease management programs were noted and used to answer this qu estion. The organization of the different plans (vertically integrated or subcont racted) were of particular interest. The purpose is to establish a frame of reference for the ide ntification of the factors to be studied and included in the regression model. To assess the impact of transaction cost factors on integration deci sions for disease management programs (Question 2), all voluntary health management organizations were asked to complete a survey design based on previous surveys done by Monteverde and Teece (1982a) [51 ], Masten (1984) , and Anderson and Schmittlein . These surveys were previously use d to collect data from firms engaged in construction and manufacturing, such as the automoti ve and aerospace industry. The survey covers a sample of tasks and services that can be integrate d or outsourced by a health management organization while implementing a particular disease mana gement program. The original survey has been extensively modified in order to adapt it to ga ther information on the disease management area, and it differs significantly from its original usage in the other industries. The original variables, the definitions, descriptive detail s and layout of the survey along with the modified version will be presented below. It is designed suc h that a team of
54 company officials such as the planning and implementation managers of the specif ic health plans and in some cases the higher management of the company can respond to each item on t he survey based on their judgment. This data enables the construction of the linear r egression model for the estimation of the decision to integrate services within the organizati on and the coefficients of the respective transaction cost factors in order to judge their importance Thirdly, to determine the implications for designing regression models for prediction of organization form and costs on the basis of transaction cost factors (Quest ion 3), a summative analysis of the quantitative data was conducted, along with a comparison of the predictions made by our model with the actual organizational form, in order to determine the e ffectiveness of the constructed model, which also provides a more reliable picture of the performa nce of the model. In addition, data on organization costs were collected in order to estimate the cost of alternate arrangements. Finally, to analyze if selective organization leads to savings for the managed care organization or health insurance organization (Question 4), we obtained the organization cos ts for the disease management programs that have been integrated into the organization, and a lso those that have been contracted to external vendors (DMOs). Thus, we can obtain estimates of org anizational transaction costs for both the cases possible for these programs for eva luation and comparison of the costs incurred. 4.5 Primary Data Collection As noted above, hypotheses regarding the effect of various transaction cost factors on the outsourcing decision for a disease management program have been put forth. To test these hypotheses, data was collected from health management organizations. The independent variables corresponding to the hypotheses stated above are bas ed on the respective transaction cost factors and are scored using a 5-point Liker t scale and are explained in table 4.1 below. The specific questions asked in the survey have been detailed in appendix B at the end of the document.
55 Table 4.1 Probit Model Variable Definitions and Descriptions Question Variable Definition Q Â– 1 Disease class The disease managed by the program. Q Â– 2 Organization form = 1, if the program was in sourced, = 0, if the program was outsourced. Q 3-a, Q 3-b, Q Â– 4, Q Â– 5 Measurement of transaction costs (Co) This can be measured as the time spent in relation to the program X the average hourly management wage. Q 6 Temporal specificity Ranking of the importance of timing of interventions, patient/program effectiveness checks, risk evaluations, etc. Q Â– 7 Physical asset specificity The degree to which the facilities and equipment is specific to the application. Q Â– 8 Human asset specificity The degree to which the knowledge, skills and experience of employees is specific to the application. Q Â– 9 Dedicated asset specificity A ranking of the degree of dedicated assets required for the program. Q Â– 10 Complexity (proxy for uncertainty) A ranking of the complexity of the tasks involved in the program. Q Â– 11 Similarity A variable that ranks a program according to the similarity of tasks with respect to the other programs run by the health plan, and how similar the required care is between the patient classes in the different programs. Q Â– 12 Frequency This variable ranks how often patient and physician contact is made by the program staff. Q Â– 13 Uncertainty (proxy for uncertainty) Provides a ranking of the difficulty in measuring the results, performance evaluation and effectiveness of the program.
56 In addition to these independent variables, data was collected on organizati on costs for the estimation of the structural cost equations as given in (1) and (2). The acquisition of this data has proved difficult and has varied based on the organization. For outside contract ing particularly, this difficulty is exacerbated as outsourcing involves two parties and cost s will be borne by both of them, necessitating the need for data to be collected from two sources Also, contractual failures occur probabilistically over a period of time in the future, wh ich leads to the data being collected being based on the views and expectations of the decision makers i nvolved. By contrast, costs of internal organization (planning, execution) occur in a s ingle organization and in a more routine manner. Thus, these costs are easier to obtain or if actual measurement is not possible, reasonable proxies can be constructed. We thus concentrated on ob taining these costs, that is, the costs of internal organization for the processes and services actually done in house by the firm. Based on this, the costs of organization can be obtained by calcula ting the number of hours consumed by the decision makers for the planning, set up and execution of a service or process times the average hourly wage rate for the manage ment involved, which was found to be $60/hr after investigation into the industry and its associate d wages. In the designed survey, there are four questions pertaining to costs. As trans action costs are not usually measured and recorded, the questions ask for time estimates that are then converted to a monetary value. These questions are the questions 3a, 3b, 4 and 5. Question 3a is conce rned with obtaining the time estimate for the administrative and facility plann ing tasks associated with an in-house program. If the program is to be integrated, there will need to have been substantial time and effort spent in order to fulfill the required administrative and s tartup tasks of starting a program from scratch. These costs will be unique to an in-house program and will not be present in the case of an outsourced program. Question 3b is used to measure the lega l costs incurred while contracting a disease management program. When the decision is made t o outsource, an appropriate contract needs to be drawn up between the two parties in order to def ine and put down the terms and conditions of the partnership. This will involve negotiations and bargaining between the parties involved which leads to an additional cost incurred f or the contracted programs. This cost is unique to outsourced programs and will not occur in the case of programs built internally by the health plans.
57 The next question (question 4) is concerned with the measurement of search and information costs. This question is common to both cases (in-sourced and outsourced), as rele vant information regarding the disease management program will need to be collected reg ardless of the form decided upon. In the integrated case, costs will be incurred in obtaining inf ormation about the tools and facilities required, the expertise needed, and the outcomes and benchmar ks to be set. In the case of outsourced programs, it will involve the selection of a vendor t hat meets all the set requirements from all the choices available in the market. Question 5 is used to record the supervisory costs that are an integral part of transaction costs. This question is again meant for both cases of organizational form, as i n the integrated case, the effectiveness and outcomes of the implemented program will need to be monitor ed and changes will need to be made (if needed) to the internal staff and tools of the he alth plan. In the case of an outsourced program, time will be spent on monitoring the outcomes/results repo rted to the management by the external DMO, and changes or improvements may need to be worke d out as needed based on the decisions of the health plan management. The dollar values f or this question are annualized. Thus, the in-sourced costs are calculated as follows: In-sourced costs = (Q3a + Q4)* working hours*average hourly management wage + Q5*weeks/yr*Average hourly management wage The outsourced costs are calculated as follows: Outsourced costs = Q3b + (Q4)* working hours*average hourly management wage + Q5*weeks/yr*Average hourly management wage 4.6 Data Analysis Upon collection of the data, analysis was conducted upon the gathered data in a two -step procedure. In the first stage, a probit regression model was constructed for the estimation of the selection decision regarding whether the process will be done internally (integr ated) or if the task will be
58 subcontracted (outsourced) to an outside service provider. The building of th is model provided us with the coefficients of the various factors considered and explained in table 4.1, and their effect on the final organization decision and the form chosen. This stage of the analys is provided answers as to the importance of the various factors considered in the analysis and their influence on the outsourcing decision. In the second stage, estimation of the structural equations of the model i s carried out. First, we estimate the internal organization cost equation for each integrated progr am based on the costs obtained for each program from the organization and the values for each of the coefficients obtained from the first stage results. For this, using the sample of integrat ed services, we can estimate the coefficients for the internal organization cost equati on by regressing our measure of internal organization costs against each of the independent variable s. We also obtain the log specifications of the linear internal cost equation calculated, as the log specification will constrain the organization costs in the positive direction and also provide a better f it to the obtained data. Estimates of the transaction costs for the contracted disease manag ement programs are also calculated as detailed in section 4.5 and compared with the in-house costs obt ained. Upon obtaining the cost estimations for both integrated and outsourced programs, t he in-house cost equation is used to estimate the transaction costs for the integra ted organizational form for each program, given its specific attributes. Thus, a comparison of the var ious organizational forms can be made. We obtain the predicted dollar value of the integrated transaction costs to compare with the costs for the organizational form actually adopted. T he costs to the firm that would be incurred if all the tasks/services or programs were to be i ntegrated can also be obtained and compared to find the costs or savings caused to the health plan under each org anizational form.
59 Chapter 5. Numerical Results and Inference This chapter describes the application of transaction cost economics t heory to the area of disease management programs in health plans. Section 5.1 reports the frequencies o f the data set obtained from the electronic survey responses. In Section 5.2 the data analysis is pe rformed. In Section 5.3 the data set is separated in to the training and validation sets. Secti on 5.4 onwards details the modeling, results and inference from the resulting models. TCE analysis is applied to cost prediction for the integrated subset of the obtained data set. These health plans with integrated programs are selected for cost analysis be cause the costs associated with integrated programs are much more accurately measurable as compar ed to the outsourced subset. 5.1 Frequencies of Respondents and Corresponding DM Programs In order to collect relevant data for the construction of the required mode ls for the analysis, an electronic survey was sent to the health plans that agreed to partici pate in the research over a period of two and a half months. The survey was sent to health plans across the nation in order to obtain the largest possible sample size for analysis, and only completed surveys with responses for all questions were included in the analysis. The frequencies of the res ulting aggregated data set are as given below in tables 5.1, 5.2 and 5.3, which list the responding health p lans, the programs implemented and the frequencies of the ranking questions based on the tr ansaction cost factors respectively.
60 Table 5.1 Responding Organizations and Number of Respective Responses Organization Frequency Cumulative Frequency Ault International Medical Management, LLC 1 1 BCBSVT 6 7 Blue Cross Blue Shield of Florida 5 12 BlueCross BlueShield of Tennessee 5 17 CareGuide, Inc. 5 22 Contra Costa Health Plan 1 23 Direct Remedy Inc. 1 24 Florida Health Care Plans 4 28 Great-West healthcare 5 33 Health Alliance Plan 5 38 Health Integrated 1 39 Health Net Inc. 5 44 HealthPartners 5 49 Healthy Futures, Inc 1 50 Humana, Inc. 7 57 IMS Managed Care, Inc. 5 62 Independence Blue Cross 3 65 Interactive Performance Technologies LLC 2 67 Medica 4 71 Memphis Managed Care Corp 1 72 Miller & Huffman Outcome Architects, LLC 2 74 Mountain States Home Care 1 75 Parkland Community Health Plan 1 76 Partners HealthCare 1 77 QualChoice 5 82 Quality First Healthcare, Inc. 1 83 Solucia Inc 5 88 Utah Medicaid 1 89 VillageHealth Disease Management 2 91 WellPoint, Inc. 1 92 William Blair 1 93
61 Table 5.2 Frequency of Corresponding DM Programs Disease Frequency Cumulative Frequency Asthma 17 17 Chronic Obstructive Pulmonary Disease (COPD) 11 28 Congestive Heart Failure (CHF) 17 45 Coronary Artery Disease(CAD) 14 59 Diabetes 17 76 Low Back Pain 1 77 Other: End Stage Renal Disease 2 79 Other: 16 complex chronic conditions (e.g., Crohn's, Parkinson's, Multiple Sclerosis, Sickle Cell, etc.) 1 80 Other: CKD 1 81 Other: Cancer 2 83 Other: Complex Conditions 1 84 Other: High risk pregnancy 1 85 Other: Hypertension 2 87 Other: Integrated program for 5 conditions (asthma, diabetes, congestive heart failure, coronary artery disease, COPD) 1 88 Other: Our Synergy program covers 21 conditions 1 89 Other: Rare Diseases 1 90 Other: all chronic health conditions 1 91 Other: maternal child 1 92 Other: Pressure ulcers 1 93
62 Table 5.3 Frequencies of the Organization Form FORM FORM Frequency Cumulative Frequency Insourced/Integrated 40 40 Outsourced 53 93 DEP DEP Frequency Cumulative Frequency 0 53 53 1 40 93 DEP is the binary variable corresponding to the organization form encountere d for each DM program. Each health plan was requested to fill out one survey for each di sease management program they offered, and the organizational form for each program was ask ed. The answer to this variable was converted to the dependent variable Â“DEPÂ” for the purp ose of modelling. DEP = 1, if the program is in sourced/integrated by the health plan, and DEP = 0, if the program is outsourced. Table 5.4 reports the frequencies of each independent transaction c ost factor.
63 Table 5.4 Frequencies of the Eight TCE Factors UNCERTAINTY Frequency Cum. Freq. PHYSICAL Frequency Cum. Freq. 1 5 5 1 18 18 2 12 17 2 4 22 3 33 50 3 41 63 4 34 84 4 15 78 5 9 93 5 15 93 HUMAN Frequency Cum. Freq. CAPITAL Frequency Cum. Freq. 1 26 26 1 6 6 2 7 33 2 20 26 3 28 61 3 31 57 4 20 81 4 22 79 5 12 93 5 14 93 COMPLEXITY Frequency Cum. Freq. SIMILARITY Frequency Cum. Freq. 1 4 4 1 14 14 2 4 8 2 11 25 3 12 20 3 17 42 4 56 76 4 18 60 5 17 93 5 33 93 FREQUENCY Frequency Cum. Freq. TEMPORAL Frequency Cum. Freq. 2 7 7 2 1 1 3 52 59 3 30 31 4 16 75 4 29 60 5 18 93 5 33 93
64 5.2 Data Analysis In order to ensure the validity of the data, the statistics of the data and the correlations between the eight independent transaction cost variables each and also the correlat ion of each independent factor with the dependent variable needs to be checked. This is done by checki ng the Pearson correlation coefficient between all nine variables involved in the construc tion of the model. The results are as given below. Pearson Correlation Coefficients Â– This statistic measures the st rength and direction of the linear relationship between the two variables. The correlation coefficient can range from -1 to +1, with -1 indicating a perfect negative correlation, +1 indicating a perfect positive correlation, and 0 indicating no correlation at all. (A variable will always have a corre lation coefficient of 1 with itself.) N = 93 This indicates that 93 observations were used in the correlation of e ach pair of variables. Prob > |r| under H0: Rho=0 This is the p-value and indicates the probability of o bserving this correlation coefficient or one that is more extreme under the null hypothes is (Ho) that the correlation (Rho) is 0. The section is constructed in a way so that the t op number is the correlation coefficient and the bottom number is the p-value. The results of the procedure are presented in tables 5.5 and 5.6 below. Table 5.5 shows the means and statistics for all nine variables involved, while table 5.6 reports the results of the correlation procedure carried out where each variable is checked for correlation wi th itself and the eight others included in the analysis.
65 Table 5.5 Means and Statistics for all Modeling Variables The CORR Procedure 9 Variables: DEP TEMPORAL PHYSICAL HUMAN CAPITAL COMPLEXITY SIMILARITY FREQUENCY UNCERTAINTY Variable N Mean Std Dev Sum Minimum Maximum Label DEP 93 0.43011 0.49777 40 0 1 DEP TEMPORAL 93 4.01075 0.85331 373 2 5 TEMPOR AL PHYSICAL 93 3.05376 1.27999 284 1 5 PHYSICA L HUMAN 93 2.83871 1.38541 264 1 5 HUMAN CAPITAL 93 3.19355 1.135 297 1 5 CAPITAL COMPLEXITY 93 3.83871 0.9242 357 1 5 COMPLE XITY SIMILARITY 93 3.48387 1.45672 324 1 5 SIMILARI TY FREQUENCY 93 3.48387 0.89215 324 2 5 FREQUE NCY UNCERTAINTY 93 3.32258 1.00175 309 1 5 UNCERT AINTY
66 Table 5.6 Correlations for all Modeling Variables Pearson Correlation Coefficients, N = 93 Prob > |r| under H0: Rho=0 DEP TEMPORAL PHYSICAL HUMAN CAPITAL 1 0.09135 0.11685 -0.00864 0.37051 DEP 0.3838 0.2647 0.9345 0.0003 0.09135 1 0.23831 0.24054 0.05394 TEMPORAL 0.3838 0.0214 0.0202 0.6076 0.11685 0.23831 1 0.086308 0.09002 PHYSICAL 0.2647 0.0214 <.0001 0.3908 COMPLEXI TY SIMILARIT Y FREQUENCY UNCERTAINTY -0.27286 -0.1852 -0.10659 0.0675 DEP 0.0081 0.0755 0.3092 0.5203 0.15383 -0.4415 0.19298 -0.10583 TEMPORAL 0.141 <.0001 0.0638 0.3127 0.09929 -0.0957 -0.01351 0.09653 PHYSICAL 0.3437 0.3614 0.8977 0.3573 DEP TEMPORAL PHYSICAL HUMAN CAPITAL -0.00864 0.24054 0.086308 1 0.02007 HUMAN 0.9345 0.0202 <.0001 0.8486 0.37051 0.05394 0.09002 0.02007 1 CAPITAL 0.0003 0.6076 0.3908 0.8486 -0.27286 0.15383 0.09929 0.0304 -0.05281 COMPLEXITY 0.0081 0.141 0.3437 0.7724 0.6151 -0.1852 -0.4415 -0.09572 -0.21943 -0.02439 SIMILARITY 0.0755 <.0001 0.3614 0.0346 0.8165 -0.10659 0.19298 -0.01351 0.06383 -0.05056 FREQUENCY 0.3092 0.0638 0.8977 0.5433 0.6303 0.0675 -0.1058 0.09653 -0.0091 0.25041 UNCERTAINT Y 0.5203 0.3127 0.3573 0.931 0.0155
67 Table 5.6 (Continued) COMPLEXIT Y SIMILARITY FREQUENCY UNCERTAINTY 0.0304 -0.2194 0.06383 -0.0091 HUMAN 0.7724 0.0346 0.5433 0.931 -0.05281 -0.0244 -0.05056 0.25041 CAPITAL 0.6151 0.8165 0.6303 0.0155 1 -0.0948 0.20115 0.35032 COMPLEXITY 0.366 0.0532 0.0006 -0.0948 1 -0.35775 -0.01129 SIMILARITY 0.366 0.0004 0.9144 0.20115 -0.3578 1 -0.26168 FREQUENCY 0.0532 0.0004 0.0113 0.35032 -0.0113 -0.26168 1 UNCERTAINT Y 0.0006 0.9144 0.0113 If an independent variable is heavily correlated with another independent v ariable, one of them can be removed as both produce the same effect in the model and it is not necessa ry that both variables be included in the model. If an independent variable is heavily corr elated with the dependent variable of the model, then the effect of the independent factor ca n be explained by the nature of the correlation between the two variables. However, from the a bove results we see that none of the independent variables are correlated strongly with each oth er and neither is any independent variables strongly correlated with the dependent variable (DEP). Hence all 8 of then can be included in the analysis. 5.3 Creating the Training and Validation Sets for the First Stage Selec tion Model The total sample size consists of 93 data points. The responses cover a la rge number of the disease management programs offered by health plans. The responses also show that there was no clear consensus as to which organizational form is better for these pr ograms, as already evidenced from the literature review. For the sample obtained for the purpose of t his research, it is seen that outsourced programs (N = 53) slightly outnumber the integrated cases (N = 40)
68 This data set was randomly split into two groups: the training set on w hich the selection model was built and the inference was deduced, consisting of 80 observations and the vali dation set, which was used to test the model and determine its accuracy, containing 13 observ ations. The frequencies for the two sets are as given below in tables 5.7 and 5.8. Table 5.7 Frequencies for the Training Set The FREQ Procedure DEP DEP Frequency Percent Cumulative Frequency Cumulative Percent 0 47 58.75 47 58.75 1 33 41.25 80 100 Table 5.8 Frequencies for the Validation Set ACTUAL_FORM Frequency Percent Cumulative Frequency Cumulative Percent 0 6 46.15 6 46.15 1 7 53.85 13 100 5.4 Training Stage 5.4.1 Step 1. Creating the First Stage Selection Model The first step is to create the selection model. The results of this model will provide answers to the following questions: 1) Do transaction cost factors affect and influence the sourcing decis ion for disease management programs? 2) If so, which factors play the most important role in determining organizatio nal form and what is their effect?
69 The combined set of responses needs to be split into two for the purposes of mode ling. The first is the training set (with 80 randomly selected observations) on which the mode l is built and the coefficients and significance of the independent factors is noted. The se cond set is the testing set, which has the remaining 13 observations, where the model accuracy will be t ested by comparing the predicted organizational form with the actual, which is known in our cas e. The training set created from the total sample is used to create the first stage probit selection model. To compute the first stage selection model, the command Â“proc probitÂ” can be used i n SAS. This procedure does not allow for the classification table to be obtained, however which is very helpful for checking the model accuracy in this case. As an alternativ e, a the Â“proc logisticÂ” command is used along with the Â“link = probitÂ” command. This command estimates a p robit model based on the given data, while also allowing for the probabilities for ea ch observation and the classification table to be constructed. In this research based on the effect of transaction cost factors on organi zational form of disease management programs, the selection model contains the eight independent tra nsaction cost variables. The dependent variable DEP is an indicator variable with val ue 1 for integrated programs and a value 0 for outsourced programs. The SAS commands are as follows : proc logistic data =TRAINING descending ; model DEP = temporal physical human capital complexity similarity frequency uncertainty/ LINK =PROBIT ctable pprob =( 0.05 to 1 by 0.05 ); output out =prob XBETA = g predicted =phat; TITLE 'FIRST STAGE SELECTION MODEL' ; run ; Where, Â“trainingÂ” is the data set containing the 80 data points, the command Â“ XBETAÂ” gives us the probit scores generated for each of the observations. The command Â“pre dictedÂ” provides us with the probabilities for each of the observations recorded. In the output of this analysis, we find the estimates of the parameters. On the basis of these parameters, for each observation the predicted probit score is also obtained, which is stored in the variable Â“gÂ”.
70 The command Â“predictedÂ” gives us the probability values calculated for ea ch of the observations in the training set. The results are as given below. Table 5.9 gives the model fit statistics and the significance of the probit model for to the data using the statistical parameters given below. Table 5.9 Full Model Response, Fit Statistics and Null Hypothesis f or Training Stage FIRST STAGE SELECTION MODEL The LOGISTIC Procedure Model Information Data Set WORK.TRAINING Response Variable DEP DEP Number of Response Levels 2 Model binary probit Optimization Technique Fisher's scoring Number of Observations Read 80 Number of Observations Used 80 Response Profile Ordered Value DEP Total Frequency 1 1 33 2 0 47 Probability modeled is DEP=1.
71 Table 5.9 (Continued) Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied. Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 110.441 98.614 SC 112.823 120.052 -2 Log L 108.441 80.614 Testing Global Null Hypothesis: BETA=0 Test Chi-Square DF Pr > ChiSq Likelihood Ratio 27.8271 8 0.0005 Score 23.6146 8 0.0027 Wald 18.0993 8 0.0205 An explanation of the terms and results is given below: 1) Data Set The data set used in this procedure. 2) Response Variable The response variable in the logistic regressi on. 3) Number of Response Levels The number of levels our response variable has. Here we have DEP = 0 and DEP = 1. 4) Model The type of regression model that was fit to our data. 5) Optimization Technique This refers to the iterative method of esti mating the regression parameters. In SAS, the default is method is Fisher's scoring method
72 6) Number of Observations Read and Number of Observations Used The number of observations read and the number of observation used in the analysis. The Number of Observations Used may be less than the Number of Observations Read if there are missing values for any variables in the equation. By default, SAS does a list wise deletion of incomplete cases. We see that all 80 observations of the training set have been used for the construction of the model, and none have been deleted. 7) Ordered Value Ordered value refers to how SAS orders/models the levels of the dependent variable. When the descending option is specified in the procedure s tatement, SAS treats the levels of DEP in a descending order (high to low). By default S AS models the 0's (the outsourced cases). The descending option is necessary so that S AS models the 1's, that is, the integrated cases. 8) Total Frequency The frequency distribution of the response variable. O ur response variable has 33 observations with a DEP = 1 and 47 with DEP = 0. 9) Probability modeled is DEP = 1 This is a note informing which level of t he response variable we are modeling. 10) Model Convergence Status this describes whether the maximum-likelihood a lgorithm has converged or not, and what kind of convergence criterion is used to asses convergence. The default criterion is the relative gradient convergenc e criterion (GCONV), and the default precision is 10 -8 11) Criterion Â– this lists various measurements used to assess the model fit which consists of the following: 1) AIC The Akaike Information Criterion. It is calculated as AIC = -2 Log L + 2((k1) + s), where k is the number of levels of the dependent variable and s is the number of predictors in the model. AIC is used for the comparison of models from different samples or non-nested models. The model with the smallest AIC is considered the best. 2) SC This is the Schwarz Criterion. It is defined as 2 Log L + ((k-1) + s)*log ( f i ), where f i 's are the frequency values of the i th observation, and k and s are as defined previously. Like AIC, SC penalizes for the number of predictors in the model a nd the smallest SC is most desirable. 3) Akaike Information Criterion (AIC) and Schwarz Criterion (SC) are devi ants of negative two times the Log-Likelihood (-2 Log L). AIC and SC penalize the l oglikelihood by the number of predictors in the model.
73 4) -2 Log L This is negative two times the log-likelihood. The -2 Log L is used in hypothesis tests for nested models. 12) Intercept Only This column refers to the respective criterion stati stics with no predictors in the model, i.e., just the response variable. 13) Intercept and Covariates This column corresponds to the respective cr iterion statistics for the fitted model. A fitted model includes all independent variables and the intercept. We can compare the values in this column with the criteria corresponding I ntercept Only value to assess model fit/significance. 14) Test These are three asymptotically equivalent Chi-Square tests. They test against the null hypothesis that at least one of the predictors' regression coeffici ent is not equal to zero in the model. 15) Likelihood Ratio The Likelihood Ratio (LR) Chi-Square tests that at l east one of the predictors' regression coefficient is not equal to zero in the model. T he LR Chi-Square statistic can be calculated by -2 Log L(null model) 2 Log L(fitted model) where L(null model) refers to the Intercept Only model and L(fitted model) refers to the Intercept and Covariates model. 16) Score The Score Chi-Square tests that at least one of the predictors' regression coefficients is not equal to zero in the model. 17) Wald The Wald Chi-Square tests that at least one of the predictors' reg ression coefficients is not equal to zero in the model. 18) Chi-Square, DF and Pr > ChiSq The Chi-Square test statistic, Degrees of Fr eedom (DF) and associated p-value (PR>ChiSq) corresponding to the specific test that a ll of the predictors are simultaneously equal to zero. The null hypothesis is that all of the regression coefficients in the model are equal to zero. The DF defines the distribution of the Chi-Square test statistics and is defined by the number of predictors in the model. PR>ChiSq is compared to a specified alpha level (willingness to acce pt a type I error), which is often set at 0.05 or 0.01. 5.4.1 Step 2. Evaluate Results of the Training Stage From the above results, we see that the model is a good fit for the data obta ined via the survey. It is now necessary to obtain the coefficients for each independent transac tion cost factor in order to gauge their effect on the final organizational form chosen by the health p lan for. The statistics and results obtained in this step are explained below:
74 1) Parameter Â– this column lists the predictor variables in the model and t he intercept. 2) DF This column gives the degrees of freedom corresponding to the Paramet er. Each Parameter estimated in the model requires one DF and defines the ChiSquare distribution to test whether the individual regression coefficient is zero, given the other variables in the model. 3) Estimate The binary probit regression estimates for the Paramet ers in the model. 4) Intercept The probit regression estimate when all variables in the mode l are evaluated at zero. 5) Standard Error The standard errors of the individual regression coeff icients. 6) Wald Chi-Square and Pr > ChiSq The test statistics and p-values, respect ively, testing the null hypothesis that an individual predictor's regression coefficient i s zero, given the other predictor variables are in the model. 7) Percent Concordant A pair of observations with different observed response s is said to be concordant if the observation with the lower ordered response value (DEP = 0) has a lower predicted mean score than the observation with the higher ordered res ponse value (DEP = 1). 8) Percent Discordant If an observation with the lower ordered response va lue has a higher predicted mean score than the observation with a higher ordered response val ue, then the pair is discordant. 9) Percent Tied A pair of observations with different responses is neit her concordant nor discordant, and is termed a tied pair. 10) Pairs The total number of distinct pairs with one case having a positive response (DEP = 1) and the other having a negative response (DEP = 0). The total number ways the 93 observations can be paired up (excluding be matched up with themselves) is 93(92)/2 = 4278. 11) Somers' D Somer's D is used to determine the strength and direction of re lation between pairs of variables. Its values range from -1.0 (all pairs disagree) to 1.0 (all pairs agree). It is defined as (n c -n d )/t where n c is the number of pairs that are concordant, n d the number of pairs that are discordant, and t is the number of total number of pairs wi th different responses. 12) Gamma The Goodman-Kruskal Gamma method does not penalize for ties on either variable. Its values range from -1.0 (no association) to 1.0 (full associati on). Because it
75 does not penalize for ties, its value will generally be greater than the v alues for Somer's D. 13) Tau-a Kendall's Tau-a is a modification of Somer's D that takes into th e account the difference between the number of possible paired observations and the number o f paired observations with a different response. It is defined to be the ratio of the difference between the number of concordant pairs and the number of discordant pairs to the number of possible pairs (2(n c -n d )/(N(N-1)). Tau-a is usually smaller than Somer's D since there are many paired observations with the same response. 14) c c ranges from 0.5 to 1, where 0.5 corresponds to the model randomly predicting the response, and a 1 corresponds to the model perfectly predicting the response. Through this step we obtain the parameter estimates for each transaction c ost factor, as presented in table 5.10.
76 Table 5.10 Analysis of Parameter Coefficients for the Training Stage FIRST STAGE SELECTION MODEL The LOGISTIC Procedure Analysis of Maximum Likelihood Estimates Parameter DF Estimate Standard Error Wald ChiSquare Pr > ChiSq Intercept 1 1.6295 1.6646 0.9583 0.3276 TEMPORAL 1 0.0476 0.2179 0.0478 0.827 PHYSICAL 1 0.6327 0.2587 5.9824 0.0144 HUMAN 1 -0.5188 0.2496 4.3196 0.0377 CAPITAL 1 0.5214 0.1813 8.27 0.004 COMPLEXITY 1 -0.6031 0.2401 6.3088 0.012 SIMILARITY 1 -0.2788 0.1416 3.873 0.0491 FREQUENCY 1 -0.1736 0.2283 0.5785 0.4469 UNCERTAINTY 1 -0.067 0.2079 0.104 0.7471 Association of Predicted Probabilities and Observed Responses Percent Concordant 80.1 Somers' D 0.603 Percent Discordant 19.7 Gamma 0.605 Percent Tied 0.2 Tau-a 0.296 Pairs 1551 c 0.802
77 5.4.1 Step 3. Use the Classification Table to Determine Optimal Cut-Off Point In order to maximize the effectiveness of the model in classifying events a nd non events for the validation set, we need to select a cut-off point for the predicted probab ilities, one below which the program will be classified as outsourced, and above which the program wil l be classified as in-sourced. This comparison table was created using the Â“pprobÂ” option in th e modeling syntax available in the SAS software. Each event can be classified as a true positive or a false positive according to the following definitions: 1) Event: if the organizational form of a given DM program is predicted as in-s ourced/integrated then it is termed as an event. 2) NonÂ–event: if the organization form of a given DM program is predicted as outso urced then it is termed as a non-event. 3) False POS (False positive): if an event identified by the model is not a n integrated program it constitutes a false positive. 4) False NEG (False negatives): if the non-event identified by the model a s outsourced program is actually an event (in-sourced organizational form) it is termed as a false negative. The results are as shown in table 5.11.
78 Table 5.11 Classification Table for Training Set Correct Incorrect Percentages Prob Level Event Non Event Event Non Event Correct Sensitivity Specificity False POS False NEG 0.05 33 3 44 0 45 100 6.4 57.1 0 0.1 32 11 36 1 53.8 97 23.4 52.9 8.3 0.15 32 11 36 1 53.8 97 23.4 52.9 8.3 0.2 25 16 31 8 51.3 75.8 34 55.4 33.3 0.25 24 19 28 9 53.8 72.7 40.4 53.8 32.1 0.3 23 21 26 10 55 69.7 44.7 53.1 32.3 0.35 23 25 22 10 60 69.7 53.2 48.9 28.6 0.4 23 30 17 10 66.3 69.7 63.8 42.5 25 0.45 23 38 9 10 76.3 69.7 80.9 28.1 20.8 0.5 20 39 8 13 73.8 60.6 83 28.6 25 0.55 17 44 3 16 76.3 51.5 93.6 15.0 26.7 0.6 15 45 2 18 75 45.5 95.7 11.8 28.6 0.65 14 45 2 19 73.8 42.4 95.7 12.5 29.7 0.7 14 45 2 19 73.8 42.4 95.7 12.5 29.7 0.75 8 45 2 25 66.3 24.2 95.7 20 35.7 0.8 7 45 2 26 65 21.2 95.7 22.2 36.6 0.85 7 46 1 26 66.3 21.2 97.9 12.5 36.1 0.9 7 47 0 26 67.5 21.2 100 0 35.6 0.95 4 47 0 29 63.8 12.1 100 0 38.2 1 0 47 0 33 58.8 0 100 41.3 From the above we see that we have the best prediction and highest value of corre ct classifications (and corresponding lowest number of false positives and n egatives) occurs at the 0.55 probability level, hence that value is selected as the cut-off point to be used for the validation set. 5.5 Validation Stage 5.5.1 Step 1. Embed the Validation Set Into the Training Set In this step we combine the training and validation data sets into one, but we le ave the dependent variable information as unknown for the validation set.
79 When the model is run, the model is built again on the basis of the training set However, the predicted probabilities for the validation set are also calculat ed and displayed. The SAS commands are as follows: proc logistic data =COMBINED descending ; model DEP = temporal physical human capital complexity similarity frequency uncertainty/ LINK =PROBIT ; output out =prob2 XBETA = g2 predicted =phat2; TITLE 'FIRST STAGE SELECTION MODEL FOR COMBINED DATA SET' ; run ; Where, Â“combinedÂ” is the data set containing all the 93 data points. Usi ng the cut-off point described above, one can then classify them as inÂ–sourced or outsourced, and a co mparison with the actual form (known in our case) can be made, if this information is store d in another variable (for this analysis, the actual form is stored in the variable Â“actua l_formÂ”. The initial output for this step again details the number of observations used and the model fit sta tistics with the terms as explained in section 5.4.1 (step 1). The results are detailed in tabl e 5.12, and show that out of the total 93 observations used, only the original 80 are used to construct the model, w hereas the newly added training observations are not used as they have the dependent va riable (DEP) as missing. However, predicted probabilities are still calculated for t he testing set as well as this set contains all the independent variable values for each observation. Thus, a c omparison of the predicted and actual form can be made in the later stages.
80 Table 5.12 Full Model Response, Fit Statistics and Null Hypothesis for Testing Phase FIRST STAGE SELECTION MODEL FOR COMBINED DATA SET The LOGISTIC Procedure Model Information Data Set WORK.COMBINED Response Variable DEP DEP Number of Response Levels 2 Model binary probit Optimization Technique Fisher's scoring Number of Observations Read 93 Number of Observations Used 80 Response Profile Ordered Value DEP Total Frequency 1 1 33 2 0 47 Probability modeled is DEP=1.
81 Table 5.12 (Continued) NOTE: 13 observations were deleted due to missing values for the response or explanatory variables. Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied. Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 110.441 98.614 SC 112.823 120.052 -2 Log L 108.441 80.614 FIRST STAGE SELECTION MODEL FOR COMBINED DATA SET The LOGISTIC Procedure Testing Global Null Hypothesis: BETA=0 Test ChiSquare DF Pr > ChiSq Likelihood Ratio 27.8271 8 0.0005 Score 23.6146 8 0.0027 Wald 18.0993 8 0.0205
82 5.5.1 Step 2. Results of the First Stage Selection Model W ith Combined Data Set As in the earlier case with the training set, in this step we obtain the p arameter estimates for the transaction cost variables included in the modeling. The terms and stat istics are the same as those explained in section 5.4.1 step 2. Table 5.13 reports the parameter coefficient s for the independent factors. Table 5.13 Analysis of Parameter Coefficients for Testing Stage Analysis of Maximum Likelihood Estimates Parameter DF Estimate Standard Error Wald ChiSquare Pr > ChiSq Intercept 1 1.6295 1.6646 0.9583 0.3276 TEMPORAL 1 0.0476 0.2179 0.0478 0.827 PHYSICAL 1 0.6327 0.2587 5.9824 0.0144 HUMAN 1 -0.5188 0.2496 4.3196 0.0377 CAPITAL 1 0.5214 0.1813 8.27 0.004 COMPLEXITY 1 -0.6031 0.2401 6.3088 0.012 SIMILARITY 1 -0.2788 0.1416 3.873 0.0491 FREQUENCY 1 -0.1736 0.2283 0.5785 0.4469 UNCERTAINTY 1 -0.067 0.2079 0.104 0.7471 Association of Predicted Probabilities and Observed Responses Percent Concordant 80.1 Somers' D 0.603 Percent Discordant 19.7 Gamma 0.605 Percent Tied 0.2 Tau-a 0.296 Pairs 1551 c 0.802
83 5.5.2 Classification of the Training Set As we have selected the optimum cut-off point for classifying the observa tions in the model, we can classify the output according to the predicted probabilities for eac h data point in the training set to check the accuracy of the model on the training set. From table 5.11, it is seen that selecting a cut-off point of 55% as the demarcation between integrated and outsource d organizational form gives us the most correct classifications and the least number of fals e positives and false negatives. This cut-off point is now used on the training and validation set Those observations from both sets that have a predicted value of less than 0.55 are classifie d as outsourced and the oneÂ’s that have a predicted value of 0.55 or higher are classified as having a n integrated organizational form. The results for both the training and the validation se t can now be checked for accuracy. The results for the training set are presented in tabl e 5.14. The first column denotes the predicted form, classified on the basis of the cut-off point, whereas t he first row denotes the dependent variable (DEP), which is the actual organizational form for the disease management programs. The diagonal elements represent the frequency of the correct clas sifications, while the non-diagonal elements are the number of observations that have been erroneously c lassified. Table 5.14 Prediction Accuracy for the Training Set The FREQ Procedure Table of PRED_FORM by DEP Frequency 0 1 Total 0 45 13 58 1 2 20 22 Total 47 33 80 Frequency Missing = 13 Hence, prediction accuracy = (45+20)/80 = 81.25 %
84 5.5.3 Classification of the Validation Set We follow the same procedure as in section 5.5.2 in order to check the classi fication of the validation set, which is the true measure of model effectiveness. The ou tcome is reported in table 5.15. Based on the classification process detailed earlier, it is seen tha t 76.92% of the observations in the testing/validation set are correctly classifi ed. Thus, the model is accurate in the prediction of organizational form of disease management programs in he alth plans. Table 5.15 Prediction Accuracy for the Validation Set The FREQ Procedure Table of PRED_FORM by ACTUAL_FORM Frequency 0 1 Total 0 5 2 7 1 1 5 6 Total 6 7 13 Frequency Missing = 80 Hence, prediction accuracy = (5+5)/13 = 76.92 % 5.5.4 Inference for the First Stage Selection Model In table 5.13, results for the probit estimation of the decision to integrate the disease management production using the proxies for the seven transaction cost factors are shown Of these factors, the coefficient for temporal specificity is positive as expected, indica ting that the program is more likely to be integrated the more critical the scheduling of a task or service is to the program.
85 The coefficient for temporal specificity (TEMPORAL) is positive, m eaning that as the importance of scheduling of the various tasks in the program rises, the m ore likely the program is to be integrated and not outsourced. The insignificance of the factor may be due t o the fact that scheduling does not play as vital a role in DM programs as it does in ot her industries such as the automotive and ship building industries, and does not have a reverberating or domino ef fect on the rest of the program if the completion of a particular task or servi ce is not according to schedule. Thus, according to the results, while our hypothesis detailed in section is correct to the extent of the effect of this factor on the organization form, the degre e of importance that was hypothesized was overstated. Thus, hypothesis 2 from section 3.2 is only partiall y satisfied. The coefficient for physical asset specificity (PHYSICAL) is also positive and significant, meaning that as the more specific the tools and assets used in the program ; the more likely the program is to be integrated and not contracted. Results show that hypothesis 1 is completely satisfied and that the effect of physical asset specificity is as s tated in section 3.2. Another factor that demonstrates a similar effect is Â“CAPITALÂ”. T he coefficient for dedicated asset specificity (CAPITAL) is also positive and significant, wh ich supports the hypothesis that integration is more likely for programs that require specific investm ents that are unusable for any other purposes. It is seen that hypothesis 4 is not satisfied and that the effect of dedic ated asset specificity is the converse of what was detailed in section 3.2. ItÂ’s significance is cor rectly stated. The coefficient for Human asset specificity (HUMAN) is negative a nd significant, which supports the hypothesis that contracting/outsourcing is more likely for progr ams needing specific skill sets and experience from the employees. Thus, hypothesis 5 is proved correct as the results from the above model match the effect detailed for this factor in sectio n 3.2. The coefficient for Uncertainty (COMPLEXITY) is negative and signif icant, which again supports the hypothesis detailed in section 3.2 that contracting/outsourcing i s more likely for programs where the outcome reporting and performance measurement may be more difficult for the health plan, which may lead to higher transaction costs if such a pr ogram were integrated. Specifically, increases in complexity make it less likely that t he program will be integrated within the firm. Hypothesis 6 is thus satisfied and results support our claim m ade in the earlier section.
86 The coefficient for similarity (SIMILARITY) is negative and signi ficant, meaning that DM programs which are dissimilar to the ones the health plan may be involved i n have a higher chance of integration than ones which may be similar to the ones already of fered by the organization.The significance of this factor is correctly predi cted, however, the effect on organizational form on DM programs is converse of that noted, leading to a part ial validation of hypothesis 7. The coefficient for frequency (FREQUENCY) is again negative but i nsignificant, meaning that as the frequency of contact required with the patents enrolled in the program rises, the probability of contracting the DM program to an external DMO rises. The insignificance of the factor may be explained by the fact that frequency within a program may vary significant ly based on the individual characteristics of the patient and hence may not be a major factor in the sourcing decision for health plans. The effect agrees with our hypothesis stated in s ection in 3.2. Hypothesis 8 is thus partially satisfied. While the effect is concurren t with the stated hypothesis, the degree of effect exerted by this factor on the sourcing decision is not as high as was postulated. Finally, the coefficient for uncertainty (UNCERTAINTY) is aga in negative but insignificant. This factor has already been covered by the independent variable Â“COMPLE XITYÂ” which is a proxy for uncertainty. We see that the effect is similar to the effe ct of Â“complexityÂ”, i.e. as the uncertainty of effectiveness and outcomes in a DM program rises, health p lans tend to contract rather than build such programs themselves. The insignificance may be due to the fact that the effect has already been covered as stated before. At this stage of the analysis, a reduced form model for the transaction c ost study of integration is constructed. The results are consistent with the hypotheses regarding the potential holdups in the market transactions in the case of human asset specificity and uncerta inty, along with the costs of managing unfamiliar or complex activities within the firm. Thus, from t he above we see that the hypothesis 1 from section 3.3 is satisfied. 5.6 First Stage Selection Model Excluding Uncertainty We create the first stage selection model again, but this time without the factor Â“UNCERTAINTYÂ” in order to note the effect of its exclusion on the model a s a whole.
87 The variable Â“complexityÂ” is generally used as a proxy for measuring unce rtainty in transaction cost analysis. This variable has been included in the first stage mode l as detailed in the previous sections. Due to the specialized nature of disease management programs, anot her question was included in the survey to capture all effects of uncertainty on the final for m chosen by the health plan. Based on the frequencies and the correlations of this variable with t he other independent factors and the dependent variable, it is seen that the effect of this f actor is similar to that of Â“complexityÂ”, but is insignificant in the final model. Thus, it is necessa ry to note the model performance and accuracy with this factor removed. 5.6.1 Training Stage In this stage, we use the same 80 observations of the earlier training set as shown in section 5.4.1 step 1, but without the independent variable uncertainty, in order to study the effe ct of its exclusion on the whole model and its accuracy. The SAS commands and data sets use d are the same as detailed in section 5.4.1 step 1, with the independent factor Â“uncertai ntyÂ” excluded. The results for the model information are as shown below in table 5.16. 5.6.1 Step 1. Creating the First Stage Selection Model Excluding Uncertai nty Table 5.16 Seven Factor Model Response, Fit Statistics and Null Hypoth esis for Training Stage FIRST STAGE SELECTION MODEL WITH NO UNCERTAINTY The LOGISTIC Procedure Model Information Data Set WORK.TRAINING Response Variable DEP DEP Number of Response Levels 2 Model binary probit Optimization Technique Fisher's scoring
88 Table 5.16 (Continued) Number of Observations Read 80 Number of Observations Used 80 Response Profile Ordered Value DEP Total Frequency 1 1 33 2 0 47 Probability modeled is DEP=1. Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied. Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 110.441 96.711 SC 112.823 115.767 -2 Log L 108.441 80.711 Testing Global Null Hypothesis: BETA=0
89 Table 5.16 (Continued) Test ChiSquare DF Pr > ChiSq Likelihood Ratio 27.73 7 0.0002 Score 23.5741 7 0.0014 Wald 18.2744 7 0.0108 5.6.1 Step 2. Evaluate Results of the Training Stage for the Seven Factor Model Once the model is run, we obtain the new coefficients for the independent vari ables involved, which are presented in table 5.17.
90 Table 5.17 Analysis of Parameter Coefficients for the Seven Facto r Model FIRST STAGE SELECTION MODEL WITH NO UNCERTAINTY The LOGISTIC Procedure Analysis of Maximum Likelihood Estimates Parameter DF Standard Estimate Error Wald Chi Â– Square Pr > ChiSq Intercept 1 1.3695 1.4507 0.8912 0.3452 TEMPORAL 1 0.0559 0.2165 0.0667 0.7962 PHYSICAL 1 0.6194 0.2563 5.8406 0.0157 HUMAN 1 -0.5072 0.2469 4.2191 0.04 CAPITAL 1 0.5062 0.1723 8.6348 0.0033 COMPLEXITY 1 -0.6265 0.2302 7.4046 0.0065 SIMILARITY 1 -0.2675 0.1354 3.9058 0.0481 FREQUENCY 1 -0.1452 0.2136 0.462 0.4967
91 Table 5.17 (Continued) Association of Predicted Probabilities and Observed Responses Percent Concordant 80.6 Somers' D 0.622 Percent Discordant 18.4 Gamma 0.628 Percent Tied 1 Tau-a 0.305 Pairs 1551 C 0.811 5.6.1 Step 3. Use the Classification Table to Determine Optimal Cut-Off Point As in section 5.4.1 (step 3), the classification table is constructed once a gain in order to obtain the best possible cut-off point for classification of the data points in the t raining and validation sets. Table 5.18 details the results of this step. It is seen that 55% once agai n serves as the best cut-off point for the classification purposes of the model. The observations from both se ts that have a predicted value of less than 0.55 are classified as outsourced and the oneÂ’s t hat have a predicted value of 0.55 or higher are classified as having an integrated organizationa l form.
92 Table 5.18 Classification Table for the Seven Factor Training Stage Correct Incorrect Percentages Prob Level Event Non Â– Event Event Non Event Correct Sensitivity Specificity False POS False NEG 0.05 33 3 44 0 45 100 6.4 57.1 0 0.1 32 11 36 1 53.8 97 23.4 52.9 8.3 0.15 32 11 36 1 53.8 97 23.4 52.9 8.3 0.2 25 17 30 8 52.5 75.8 36.2 54.5 32 0.25 24 20 27 9 55 72.7 42.6 52.9 31 0.3 23 21 26 10 55 69.7 44.7 53.1 32.3 0.35 23 25 22 10 60 69.7 53.2 48.9 28.6 0.4 23 30 17 10 66.3 69.7 63.8 42.5 25 0.45 23 37 10 10 75 69.7 78.7 30.3 21.3 0.5 20 39 8 13 73.8 60.6 83 28.6 25 0.55 17 44 3 16 76.3 51.5 93.6 15 26.7 0.6 15 45 2 18 75 45.5 95.7 11.8 28.6 0.65 15 45 2 18 75 45.5 95.7 11.8 28.6 0.7 14 45 2 19 73.8 42.4 95.7 12.5 29.7 0.75 9 45 2 24 67.5 27.3 95.7 18.2 34.8 0.8 7 45 2 26 65 21.2 95.7 22.2 36.6 0.85 7 47 0 26 67.5 21.2 100 0 35.6 0.9 7 47 0 26 67.5 21.2 100 0 35.6 0.95 4 47 0 29 63.8 12.1 100 0 38.2 1 0 47 0 33 58.8 0.0 100 41.3 5.6.2 Testing Stage The new model is to be tested again for prediction accuracy, which done as below by running the model again using a combination of both the training and validation sets as input. The SAS commands and data sets are the same as detailed in section 5.5.1 step 1, excluding the factor Â“uncertaintyÂ”. 5.6.2 Step 1. Embed the Validation Set Into the Training Set In the first step, the two sets (training and validation) are combined, and t he model is run again as below. The results are as reported in table 5.19.
93 Table 5.19 Seven Factor Model Response, Fit Statistics and Null Hypoth esis for Testing Stage FIRST STAGE SELECTION MODEL FOR COMBINED DATA SET WITH NO UNCERTAINTY The LOGISTIC Procedure Model Information Data Set WORK.COMBINED_NO_UNCERT Response Variable DEP DEP Number of Response Levels 2 Model binary probit Optimization Technique Fisher's scoring Number of Observations Read 93 Number of Observations Used 80 Response Profile Ordered Value DEP Total Frequency 1 1 33 2 0 47 Probability modeled is DEP=1. NOTE: 13 observations were deleted due to missing values for the response or explanatory variables.
94 Table 5.19 (Continued) Model Convergence Status Convergence criterion (GCONV=1E-8) satisfied. Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 110.441 96.711 SC 112.823 115.767 -2 Log L 108.441 80.711 FIRST STAGE SELECTION MODEL FOR COMBINED DATA SET WITH NO UNCERTAINTY The LOGISTIC Procedure Testing Global Null Hypothesis: BETA=0 Test ChiSquare DF Pr > ChiSq Likelihood Ratio 27.73 7 0.0002 Score 23.5741 7 0.0014 Wald 18.2744 7 0.0108
95 5.6.2 Step 2. Results of the First Stage Selection Model W ith Combined Data Set for Seven TCE Factors The model run provides us with the parameter coefficients again as detai led below in table 5.20. The output parameters are similar to those explained in section 5.4.1 step 2. Table 5.20 Analysis of Parameter Coefficients for the Testing St age of the Seven Factor Model Parameter DF Standard Estimate Error Wald Chi Â– Square Pr > ChiSq Intercept 1 1.3695 1.4507 0.8912 0.3452 TEMPORAL 1 0.0559 0.2165 0.0667 0.7962 PHYSICAL 1 0.6194 0.2563 5.8406 0.0157 HUMAN 1 -0.5072 0.2469 4.2191 0.04 CAPITAL 1 0.5062 0.1723 8.6348 0.0033 COMPLEXITY 1 -0.6265 0.2302 7.4046 0.0065 SIMILARITY 1 -0.2675 0.1354 3.9058 0.0481 FREQUENCY 1 -0.1452 0.2136 0.462 0.4967
96 Table 5.20 (Continued) Association of Predicted Probabilities and Observed Responses Percent Concordant 80.6 Somers' D 0.622 Percent Discordant 18.4 Gamma 0.628 Percent Tied 1 Tau-a 0.305 Pairs 1551 C 0.811 5.6.3 Classification of the Training Set From section 5.6.1 step 3, we have selected 55% as the cut-off point to different iate between integrated and contracted programs (the same as in the case of the ful l model), and we check the accuracy of the new model on the training set first. This is done in orde r to gauge the accuracy of the new model, which may change significantly due to the exclusion of the fact or Â“uncertaintyÂ”. The diagonal elements again represent the correct classifications for both the training and the validation phase. The results for the training set are detailed in table 5.21. Table 5.21 Prediction Accuracy for the Training Set The FREQ Procedure Table of PRED_FORM by DEP Frequency 0 1 Total 0 45 13 58 1 2 20 22 Total 47 33 80 Frequency Missing = 13
97 5.6.4 Classification of the Validation Set We also need to test the new model on the validation set in order to get the tru e accuracy of the model. The results are as shown below in table 5.22. Table 5.22 Prediction Accuracy for the Validation Set The FREQ Procedure Table of PRED_FORM by ACTUAL_FORM Frequency 0 1 Total 0 5 2 7 1 1 5 6 Total 6 7 13 Frequency Missing = 80 From the above, it is seen that the prediction accuracy of the seven fact or model is the same as that of the full model for both the training and validation sets. It can thus b e concluded that exclusion of the factor Â“uncertaintyÂ” does not affect the model accura cy. 5.7 Comparison of the Model W ith and Without the TCE Factor Uncertainty Presented below in table 5.23 is a comparison of the parameter coefficien ts for both the full and seven factor models.
98 Table 5.23 Comparison of the Full and Seven Factor Models Estimate Parameter DF w/ Uncertainty w/o Uncertainty 1.6295 1.3695 Intercept 1 (0.3276) (0.3452) 0.0476 0.0559 TEMPORAL 1 (0.827) (0.7962) 0.6327 0.6194 PHYSICAL 1 (0.0144) (0.0157) -0.5188 -0.5072 HUMAN 1 (0.0377) (0.04) 0.5214 0.5062 CAPITAL 1 (0.004) (0.0033) -0.6031 -0.6265 COMPLEXITY 1 (0.012) (0.0065) -0.2788 -0.2675 SIMILARITY 1 (0.0491) (0.0481) -0.1736 -0.1452 FREQUENCY 1 (0.4469) (0.4967) -0.067 UNCERTAINTY 1 (0.7471) Pr>Chisq statistics in parenthesis 5.8 Inference From the Comparison of Full and Seven Factor Model It is seen that the coefficient and significance of Â“TEMPORALÂ” incr eases slightly, while the factor coefficient of Â“COMPLEXITYÂ” decreases. This effect is cons istent with the observed correlations between the variables for temporal specificity and uncer tainty as Â“TEMPORALÂ” is negatively correlated with the factor Â“UNCERTAINTYÂ” due to which its value increases when the second factor for uncertainty is removed. The coefficients for fact ors Â“SIMILARITYÂ”, Â“FREQUENCYÂ” and Â“HUMANÂ” also show the same effect. On the other hand, Â“COMPLEXITYÂ” is positively correlated with Â“UNCERTAINTYÂ”, and thus the value of this factor decreases upon the removal of Â“UNCERTAINTYÂ”. The same can be said for the values of coefficients for the factors Â“PHYSICALÂ” and Â“CAPITALÂ”. Thus, we se e that the coefficients of the factors change slightly, however the prediction accuracy of the model re mains unchanged.
99 5.9 Calculation of Actual In-Sourced and Outsourced Costs Based on the literature review and industrial inquiry, the average hourly wag e for health plan executives and decision makers responsible for initiating, developing and manag ing DM programs was found to be $60. This value was multiplied to the time estimates that were obtained from our survey in order to obtain an estimate of the actual transaction costs for each DM program, the results of which have been presented below. 5.9.1 Breakup of Actual In-Sourced Costs Using the time estimates from the survey, and the average hourly wage, we obtain the estimates of the transaction costs for both the in-sourced and outsourced cases of the responses. The means and statistics for the in sourced case are reported in table 5.24. For the in-sourced costs, the questions 3a (time taken for administrative/f acility planning tasks), and 4 (search and information time spent) are multiplied with the number of hours in each working day (taken here as 8 hours/day) and then multiplied with the average hourly management wage, which is $60. To this is added the supervisory cost, which is given in te rms of hours per week. To annualize it, the number of weeks in a year is multiplied along wit h the average hourly management wage in order to obtain this cost. The sum of these three ele ments gives us an estimate if the in-sourced costs for a particular integrated disease management program. For the estimation of the outsourced costs, the cost estimates from ques tions 4 and 5 are calculated and added as above; in addition, the value obtained from question 3a (legal/negotiations costs) is directly added to the above as a direc t dollar amount is asked for this question, and no conversion is necessary. The addition of these three values p rovides the outsourced cost estimate for each observation in the outsourced subset. The first row of table 5.24 details the administrative/facility plann ing costs for the integrated cases of disease management programs, while the second and third rows show the search and information costs and supervisory costs respectively. The last row repor ts the means and statistics for the total in-sourced costs, which is the sum of the first three row s.
100 Table 5.24 Means for the Actual In-Sourced Costs The MEANS Procedure Variable N Mean Std Dev Minimum Maximum INSOURCED_ADMIN_COST 40 55812 68014.12 0 288000 SEARCH_INFO_COST 40 27408 47821.8 0 288000 SUPERVISORY_COST 40 52488 68036.08 0 288000 TOTAL_INSOURCED_COST 40 135708 150068.67 0 691200 5.9.2 Breakup of Actual Outsourced Costs As in the earlier case, the means and statistics for the outsourced c ase are presented in table 5.25 below. The first row shows the legal cost, which is in dollars. The second and third rows report the dollar values of the calculated search and information costs and the superv isory costs, while the fourth row reports the means and statistics for the total outsourced c osts, which is the sum of the first three terms explained above.
101 Table 5.25 Means for Actual Outsourced Costs The MEANS Procedure Variable N Mean Std Dev LEGAL_COST_OUTSOURCED_DOLLARS 53 14132.08 24255.99 SEARCH_INFO_COST 53 17750.94 21043.81 SUPERVISORY_COST 53 20513.21 21706.79 TOTAL_OUTSOURCED_COST 53 52396.23 48103.72 Variable Minimum Maximum LEGAL_COST_OUTSOURCED_DOLLARS 0 100000 SEARCH_INFO_COST 0 86400 SUPERVISORY_COST 0 108000 TOTAL_OUTSOURCED_COST 0 197920 We see that outsourced costs are approximately 1/3 of inÂ–sourced costs due to th e difficulties noted by Masten et al., who state that contracting costs are incurred by ea ch party included in the transaction, hence, cost data needs to be collected from two or more sources. In a ddition, the contractual changes and failures that occur in this case occur probabil istically over time, which requires that data be collected on the intangible expectations of the decision m akers. Thus, the collected outsourcing costs have been disregarded and the focus is placed on the inÂ–sourced costs for building of the predictive models in the next stage. 5.9.3 Calculation of the Inverse Mills Ratio and the Help and Control Fact or Delta As the outsourced are disregarded, the dependent variable information is mi ssing for part of our data set. The standard selection bias problem is thus encountered when constru cting the cost equation as detailed in chapter 4. In order to correct the selection bias an additional independent variable is needed to be added along with the transaction cost factors. Thi s variable is the Heckman correction factor lambda, which is the inverse mills ratio. To compute the Heckman correction factor Lambda with a PROBIT selection mod el, the following SAS commands are used:
102 proc logistic data =COMBINED descending ; model DEP = temporal physical human capital complexity similarity frequency uncertainty/ LINK =PROBIT ; output out =prob2 XBETA = g2 predicted =phat2; TITLE 'FIRST STAGE SELECTION MODEL FOR COMBINED DATA SET' ; run ; Where, Â“combinedÂ” is the data set containing all 93 data points, the command Â“ XBETAÂ” gives us the probit scores generated for each of the observations. The command Â“pre dictedÂ” provides us with the probabilities for each of the observations recorded. In the output of this analysis, we find the estimates of the parameters. On t he basis of these parameters, for each observation the predicted probit score is also obta ined, which is stored in the variable Â“g2Â”. These probit scores obtained in the variable Â“g2Â” are us ed to compute the Heckman control factor LAMBDA, using the SAS command as follows: LAMBDA1 = (( 1 /sqrt( 2 3.141592654 ))*(exp(-G2*G2* 0.5 )))/CDF( 'NORMAL' ,G2); Or LAMBDA2 = (PDFG2/CDFG2); Or lambda3 = ( 1 /sqrt( 2 3.141592654 )*exp(1 *g2** 2 / 2 ))/probnorm(g2). For applying the two-step procedure it is important that all rows with mis sing values on variables which are used in the substantial analyses are removed from the active file. This means that all the outsourced cases and the cases where the dependent variable or any of the independent variables are missing are removed, and the following analysis is done on the remaining insourced subset only. The next step is to compute the value of the control fact or: DELTA1 = -LAMBDA1*G2-LAMBDA1*LAMBDA1; DELTA2 = -LAMBDA2*G2-LAMBDA2*LAMBDA2; DELTA3 = -LAMBDA3*G2-LAMBDA3*LAMBDA3;
103 Three values of the control factor are calculated in order to check th e values of all three inverse mills ratios obtained by the different methods. The values of DELTA1, DELTA 2 and DELTA3 should be between -1 and 0. The values of both the inverse mills ratio and the control factor are checked as follows: PROC MEANS DATA = LAMBDA ; VAR LAMBDA1 LAMBDA2 lambda3 DELTA1 DELTA2 DELTA3 H1 H2 H3; TITLE 'RESULTS FOR THE INVERSE MILLS RATIO AND CONTROL FACTOR DEL TA' ; RUN ; The table 5.26 details the values of the inverse mills ratio and the control fa ctor delta. Â“Data = LambdaÂ” denotes the data set lambda, from which all outsourced and missing data poi nts have been excluded. Table 5.26 Results and Statistics for the Inverse Mills Ratio and Cont rol Factor Delta The MEANS Procedure Variable N Mean Std Dev Minimum Maximum LAMBDA1 39 0.6465018 0.4819293 0.0085964 1.7400238 LAMBDA2 39 0.6465018 0.4819293 0.0085964 1.7400238 lambda3 39 0.6465018 0.4819293 0.0085964 1.7400238 DELTA1 39 -0.4992273 0.2347647 -0.8291143 -0.0238978 DELTA2 39 -0.4992273 0.2347647 -0.8291143 -0.0238978 DELTA3 39 -0.4992273 0.2347647 -0.8291143 -0.0238978 h1 39 0.4992273 0.2347647 0.0238978 0.8291143 h2 39 0.4992273 0.2347647 0.0238978 0.8291143 h3 39 0.4992273 0.2347647 0.0238978 0.8291143
104 All three formulas give us the same values for the variable Â“lambdaÂ”. The variables Â“h1Â”, Â“h2Â”, and Â“h3Â” are the inverses of the three control factors calculated. The va lue of the control factor delta (DELTA1 in our case) should be between -1 and 0 which is satisfied for a ll three cases as seen above. Hence, one case of the calculated inverse mills ratio and the control factor delta can be used in the planned analysis. The inverse mills ratio is calculate d and added to the analysis as an additional independent variable as detailed in chapter 4. 5.10 Frequencies for the In-Sourced Subset Before constructing the cost model, the frequencies of the integrated port ion of the data set is presented. 5.10.1 Frequencies for the Organizations in the In-Sourced Subset Table 5.27 reports the frequencies for the health plans that have inÂ–house DM programs, whereas table 5.28 in the next section provides the number of DM program present in this s ubset of the full data set.
105 Table 5.27 Frequencies for Responding Organizations of the In-Sou rced Subset The FREQ Procedure Organization Frequency Cumulative Frequency Ault International Medical Management, LLC 1 1 CareGuide, Inc. 5 6 Contra Costa Health Plan 1 7 Florida Health Care Plans 4 11 Health Alliance Plan 4 15 HealthPartners 2 17 Healthy Futures, Inc 1 18 IMS Managed Care, Inc. 5 23 Memphis Managed Care Corp 1 24 Miller & Huffman Outcome Architects, LLC 2 26 Mountain States Home Care 1 27 Partners HealthCare 1 28 QualChoice 5 33 Quality First Healthcare, Inc. 1 34 Solucia Inc 5 39 WellPoint, Inc. 1 40 5.10.2 Frequencies for the Diseases in the In-Sourced Subset The frequencies for the diseases managed by the programs implemented in-house by the health plans in the in-sourced subset are as shown below in table 5.28.
106 Table 5.28 Frequencies for the DM Programs of the In-Sourced Sub set Disease Frequency Cumulative Frequency Asthma 7 7 Chronic Obstructive Pulmonary Disease (COPD) 4 11 Congestive Heart Failure (CHF) 8 19 Coronary Artery Disease(CAD) 7 26 Diabetes 10 36 Other: Hypertension 2 38 Other: all chronic health conditions 1 39 Pressure ulcers 1 40 5.11 Organization Cost Model for In-Sourced Costs In this analysis, the factor Â“UNCERTAINTYÂ” is removed, as this fa ctor is a proxy for the transaction cost factor uncertainty, which is already accounted for with the variable Â“COMPLEXITYÂ”. Also, as detailed by Heckman, one independent variable from the selection equation must be removed during the substantial analysis. Inclusion of all ei ght independent variables from the first stage into the organization cost equation ca uses the estimated correlation coefficients between the errors in the selection and cost equations to excee d the logical upper bound. The SAS commands used are given below: PROC REG DATA =LAMBDA; MODEL TOTAL_INSOURCED_COST = temporal physical human capital complexity similarity frequency LAMBDA1; output out =INSOURCED_PRED predicted =PRED_IN_COST; TITLE 'ORGANIZATION COST MODEL FOR IN SOURCED COSTS' ; RUN ;
107 In this analysis, the in-sourced cost (total_insourced_cost) is the depe ndent variable, whereas the transaction cost factors (excluding Â“uncertaintyÂ”) and the inver se mills ratio are the independent variables. The data set Â“lambdaÂ” is that part of the combined data set that has no missing independent or dependent variables and contains only the integrated portion of t he total sample. The command Â“predictedÂ” produces the predicted values of the in-source cost s and stores them in the variable called Â“pred_in_costÂ” in the data set Â“insourced_predÂ” (de fined by the Â“output outÂ” command). The results are as detailed in the sections below. 5.11.1 Running the Model In this step, the model is run with the calculated inÂ–house cost as the depend ent variable, and the survey responses to the TCE questions as the independent variables. T he tests and statistics displayed in this part of the results are explained below: 1) Source The source of variance, Model, Residual, and Total. The Total va riance is divided into the variance which can be explained by the independent variable s (Model) and the variance which is not explained by the independent variables (R esidual or Error). 2) DF The degrees of freedom associated with the sources of variance. T he total variance has N-1 degrees of freedom. In this case, N=39, so the DF for total is 38. The mode l degrees of freedom correspond to the number of predictors minus 1 (K-1). The interc ept is automatically included in the model. Including the intercept, there ar e 9 predictors, so the model has 9-1= 8 degrees of freedom. The Residual degree of freedom is the DF total minus the DF model, 38 8 is 30. 3) Sum of Squares The Sum of Squares associated with the three sources of varia nce, Total, Model and Residual. 4) Mean Square The Sum of Squares divided by their respective DF. 5) F Value and Pr > F The F-value is the Mean Square Model divided by the Mean Squa re Residual. The p-value associated with this F value is displayed. The p-value is compared to the alpha level (typically 0.05) and, if smaller, it can be conclude d that the independent variables reliably predict the dependent variable. If the p-value is greater than 0.05, it can be said that the group of independent variables does not show a statistically significant relationship with the dependent varia ble, or that the group of independent variables does not reliably predict the dependent variable
108 6) Root MSE Root MSE is the standard deviation of the error term, and is the sq uare root of the Mean Square Residual (or Error). 7) Dependent Mean The mean of the dependent variable. 8) Coeff Var The coefficient of variation, which is a unit-less measure of variation in the data. It is the root MSE divided by the mean of the dependent variable. 9) R-Square R-Square is the proportion of variance in the dependent variable (in-sourced cost) which can be predicted from the independent variables (the transaction cos t factors and inverse mills ratio). This value indicates that 68.65% of the varia nce in science scores can be predicted from the independent variables. 10) Adj R-Sq As predictors are added to the model, each predictor explains some of the variance in the dependent variable simply due to chance. The adjusted R-square a ttempts to yield a more honest value to estimate the R-squared for the population. The v alue of R-square was 0.6865, while the value of Adjusted R-square was 0.6029. Adjusted R-squared is computed using the formula 1 ((1 Rsq)((N 1) / (N k 1)). Table 5.29 reports the results from the analysis of variance and the R Â– sq uare value for the model constructed.
109 Table 5.29 Analysis of Variance for the In-Sourced Cost Model The REG Procedure Model: MODEL1 Dependent Variable: TOTAL_INSOURCED_COST Number of Observations Read 40 Number of Observations Used 39 Number of Observations with Missing Values 1 Source DF Sum of Squares Mean Square Value F Pr > F Model 8 5.900092E+11 73751145750 8.21 <.0001 Error 30 2.694055E+11 8980184083 Corrected Total 38 8.594147E+11 Root MSE 94764 R-Square 0.6865 Dependent Mean 139188 Adj R-Sq 0.6029 Coeff Var 68.08349 5.11.2 Results of the In-Sourced Cost Model W ith Correction for Selection Bias Table 5.30 below reports the coefficients of the independent factors f or the linear specification of the internal cost equation. The explanation for the terms and statistics found in this result are given below:
110 1) Variable This column shows the predictor variables (the independent tra nsaction cost factor). The first variable represents the intercept. 2) Label This column gives the label for the variable. 3) DF This column give the degrees of freedom associated with each independent variable. 4) Parameter Estimates The values for the regression equation for pr edicting the dependent variable from the independent variable. The regression equation is prese nted in many different ways, for example: Yp = a0 + a1*x1 + a2*x2 + a3*x3 + a4*x4+Â…Â…Â…Â….+aN*xN The column of estimates provides the values for a0, a1, a2, and so on for this e quation. In this case, the regression equation is: TOTAL_INSOURCED_COST = 277864 + (-123102)*TEMPORAL + (-71011)*PHYSICAL + 142950*HUMAN + (-98136)*CAPITAL + 18113*COMPLEXITY + 53406*SIMILARITY + 160158*FREQUENCY + (-335922)*LAMBDA 5) Standard Error The standard errors associated with the coefficients. The standard errors in this case are still biased, which shall be removed using the methodology deta iled in section 5.11.3. 6) t Value and Pr > |t|These columns provide the t-value and 2 tailed p-value us ed in testing the null hypothesis that the coefficient/parameter is 0. Coeff icients having pvalues less than alpha are statistically significant.
111 Table 5.30 Parameter Estimates for the Independent Variables in th e In-Sourced Cost Model Variable Label DF Parameter Estimate Standard Error t Value Pr > |t| Intercept Intercept 1 277864 129379 2.15 0.0399 TEMPORAL TEMPORAL 1 -123102 22140 -5.56 <.0001 PHYSICAL PHYSICAL 1 -71011 67213 -1.06 0.2992 HUMAN HUMAN 1 142950 48164 2.97 0.0058 CAPITAL CAPITAL 1 -98136 34980 -2.81 0.0087 COMPLEXITY COMPLEXITY 1 18113 45284 0.4 0.692 SIMILARITY SIMILARITY 1 53406 22626 2.36 0.025 FREQUENCY FREQUENCY 1 160158 27855 5.75 <.0001 LAMBDA1 1 -335922 139581 -2.41 0.0225 5.11.3 Correcting the Standard Error Terms The above analysis produces unbiased parameter estimates for the independe nt variables. However, the standard estimates of these parameters are biased because of heteroskedasticity. The variance of the error term is not the same for each respondent. To cor rect the standard errors and get the unbiased estimates, the following additional steps have to be t aken.
112 First, a command was added to the substantial regression analysis to save the residuals of the regression model in a new variable (which is called RES), as given be low: PROC REG DATA =LAMBDA; MODEL TOTAL_INSOURCED_COST = temporal physical human capital complexity similarity frequency LAMBDA1; output out =INSOURCED_PRED predicted =PRED_IN_COST residual = RES; RUN ; This variable must be squared: RES2 = RES*RES. Two other help variables must also be computed. The first one is the regress ion coefficient of the variable LAMBDA (the inverse mills ratio) in the OLS analysis which is called LAMB. The second one is the number of cases used in the OLS regression, called N. The results for both are given below. LAMB=-335922. N=39. The variable RES2 and also DELTA, which was computed in the first part of the analysis, have to be summed over all cases. The values of the sum of these two variables ar e as given below. DELTAS1 = -19.4699. RESS = 2.694E11. Where, RESS and DELTAS1 are the sums of the residuals and the control fact or DELTA, respectively. Now the corrected value of the variance (VARC) and the standard error (SE C) of the error term of the substantial equation can be estimated: VARC = RESS/N-LAMB*LAMB*DELTAS/N.
113 SEC = sqrt(VARC). Computation of RHO, the correlation between the error terms of the selec tion and substantial equations: RHO = sqrt(LAMB*LAMB/VARC). If (lamb<0) RHO = 0-RHO. Now the values of VARC, SEC and RHO can be computed, the values of which are note d in table 5.31: Table 5.31 Values of Corrected Variance, Std. Error and Error Correlation The MEANS Procedure Variable N Value VARC 40 63242395246 SEC 40 251480.41 RHO 40 -1.335778 Computation of the standard errors of the separate observations (RHOI) and transformation of the standard errors into weights (WGT): RHOI = sqrt(VARC+LAMB*LAMB*DELTA). WGT = 1/RHOI. Now the corrected standard errors can be computed by running the substantial a nalysis again, but this time as Weighted Least Squares (WLS) regression with WGT as w eight: PROC REG DATA =INSOURCED_PRED; MODEL TOTAL_INSOURCED_COST = temporal physical human capital complexity similarity frequency LAMBDA1; weight WGT;
114 output out =INSOURCED_PRED_NEW predicted =PRED_IN_COST_NEW residual = RES_NEW; title 'CORRECTING HETEROSKEDASTICITY FOR COST MODEL' ; RUN ; The results of the new regression are as reported below in table 5.32. The ter m Â“weightÂ” indicates that the variable Â“WGTÂ” calculated above is used as a weight in this regression. Table 5.32 Results From Heteroskedasticity Correction CORRECTING HETEROSKEDASTICITY FOR COST MODEL The REG Procedure Model: MODEL1 Dependent Variable: TOTAL_INSOURCED_COST Number of Observations Read 40 Number of Observations Used 39 Number of Observations with Missing Values 1 Weight: WGT Analysis of Variance Source DF Sum of Squares Mean Square F Value Pr > F Model 8 4964233 620529 6.58 <.0001 Error 30 2827930 94264 Corrected Total 38 7792163 Root MSE 307.025 RSquare 0.6371 Dependent Mean 141982 Adj RSq 0.5403 Coeff Var 0.21624
115 Table 5.32 (Continued) Parameter Estimates Variable Label DF Parameter Estimate Standard Error t Value Pr > |t| Intercept Intercept 1 196189 127481 1.54 0.1343 TEMPORAL TEMPORAL 1 -105395 24307 -4.34 0.0002 PHYSICAL PHYSICAL 1 -70407 75141 -0.94 0.3562 HUMAN HUMAN 1 129803 55543 2.34 0.0263 CAPITAL CAPITAL 1 -81541 40472 -2.01 0.053 COMPLEXITY COMPLEXITY 1 25955 54356 0.48 0.6365 SIMILARITY SIMILARITY 1 38124 25757 1.48 0.1493 FREQUENCY FREQUENCY 1 149350 30494 4.9 <.0001 LAMBDA1 1 -280786 163067 -1.72 0.0954 By combining the parameter estimates of the substantial analysis with the standard errors of this WLS analysis, the Heckman procedure is completed. To indicate the explain ed variance R 2 of the analysis, the R 2 of the substantial analysis should be taken. Thus, combining the parameter estimates and R 2 from the initial step and the corrected standard errors, we get the final results as shown in table 5.33. The standard errors given below have been corrected for hete roskedasticity and endogeneity of the selection correction index.
116 Table 5.33 Final Cost Model Results Variable Label DF Parameter Estimate Standard Error t Value Pr > |t| Intercept Intercept 1 277864 127481 2.15 0.0399 TEMPORAL TEMPORAL 1 -123102 24307 -5.56 <.0001 PHYSICAL PHYSICAL 1 -71011 75141 -1.06 0.2992 HUMAN HUMAN 1 142950 55543 2.97 0.0058 CAPITAL CAPITAL 1 -98136 40472 -2.81 0.0087 COMPLEXITY COMPLEXITY 1 18113 54356 0.4 0.692 SIMILARITY SIMILARITY 1 53406 25757 2.36 0.025 FREQUENCY FREQUENCY 1 160158 30494 5.75 <.0001 LAMBDA1 1 -335922 163067 -2.41 0.0225 R 2 = 0.6865 5.11.4 Inference From the Internal Cost Equation The second stage results above from the internal organization cost model confirm and strengthen the predictions of the theory and the findings of the first stage selec tion model with regard to the effects of TCE factors on the sourcing decisions for health plans. The effect of temporal specificity (TEMPORAL) on the integration t ransaction costs is negative, as meaning that health plans will tend to reduce their transaction costs if DM programs that require stricter adherence to timing and scheduling are built inÂ–house rathe r than contracted. The significance of the factor indicates that the effect of this fa ctor fosters integration through its effect on internal organization costs rather than by increasing the hazards of market exchange, as
117 noted by Masten et al. The effect of physical asset specificity (PHY SICAL) on the integration transaction costs is also negative, as meaning that health plans will te nd to reduce their transaction costs if DM programs that require more specific tools a nd software are integrated rather than outsourced. However, the coefficient on this factor is not si gnificant, indicating that the principal effect of PHYSICAL on the integration decision derive s from the hazards of market exchange. The cost coefficient of human asset specificity (HUMAN) is positive in the model, meaning that if DM programs requiring specific skills and knowledge from its employee s are in-sourced, transaction costs for the health plan tend to rise. The second-stage est imates indicate that the correlation between the human asset specificity and the likelihood of in tegration is a consequence of the rise in internal organization costs, rather than decrease in cos ts of market exchange as the theory predicts. The coefficient of dedicated asset specificity (CAPITAL) is nega tive, meaning that if DM programs requiring greater investments unique to the program are in-sourc ed, transaction costs for the health plan tend to decrease. The significance of the factor ag ain indicates that the effect of this factor fosters integration through its effect on interna l organization costs rather than by increasing the hazards of market exchange, as with the factors for tempor al and human asset specificity. The effect of uncertainty (COMPLEXITY) is positive, meaning that DM programs for which effectiveness and performance measurement are more difficult should be outsourced by health plans in order to reduce their incurred transaction costs. The coeffici ent on this factor is not significant, indicating that the principal effect of uncertainty on t he integration decision derives from the hazards of market exchange. The effect of similarity (SIMIL ARITY) on the in-sourced transaction cost is also positive, meaning that if DM programs simila r to the ones the health plan may be involved in are integrated, transaction costs for the health plan t end to rise. Unlike complexity, this factor is also significant, indicating that the effe ct of this factor fosters integration through its effect on internal organization costs rather than by increasing the hazards of market exchange.
118 Finally, the effect of frequency (FREQUENCY) in the in-sourced transa ction cost model is positive, meaning that if DM programs health plans risk increasing their i ncurred transaction costs if they integrate DM programs requiring a high frequency of contact wit h the individuals enrolled in the program. The significance of the factor indicates tha t the effect of this factor fosters integration through its effect on internal organization costs, a nd not by increasing the hazards of market exchange. The second stage findings confirm and strengthen the findings of the first sta ge estimation with regards to the transaction cost factors. In addition, we can deduce that the factors similarity, temporal, human, capital and frequency have their primary effect on the inter nal organization costs rather than on market costs as the theory suggests, whereas the fa ctors complexity and physical act principally on the costs of market exchange. Thus, from the above, it is seen that hypothesis 2 stated in section 3.3 is satisfied. 5.12 Comparison of First and Second Stage Results Table 5.34 presents a comparison between the coefficients obtained for the independent variables from the selection and the substantial equation. A side by side comparison of the par ameter coefficients obtained from the first and second stage parameter coeffici ents establishes the fact that the effect of the transaction cost factors is captured both in te rms of effect on organizational form and in terms of costs in the case of health plans.
119 Table 5.34 Comparison of Selection and Substantial Model Coefficients Estimate Parameter DF First stage Second stage cost estimate Intercept 1 1.6295 (0.3276) 277864 (0.0399) TEMPORAL 1 0.0476 (0.8270) 123102 (<.0001) PHYSICAL 1 0.6327 (0.0144) 71011 (0.2992) HUMAN 1 -0.5188 (0.0377) 142950 (0.0058) CAPITAL 1 0.5214 (0.0040) 98136 (0.0087) COMPLEXITY 1 -0.6031 (0.0120) 18113 (0.6920) SIMILARITY 1 -0.2788 (0.0491) 53406 (0.0250) FREQUENCY 1 -0.1736 (0.4469) 160158 (<.0001) UNCERTAINTY 1 -0.067 (0.7471) LAMBDA (inverse mills ratio) 1 335922 (0.0225) Pr>Chisq statistics in parenthesis
120 5.13 Comparison of Actual and Predicted In-Sourced Costs In order to determine the accuracy and effectiveness of the model, we need t o check the means and statistics of the predicted costs with the actual recorded values Table 5.35 presents the means for both the actual and predicted costs, while table 5.36 reports the predi cted value and the error for a sub-sample of the in-sourced set. 5.13.1 Means for the Actual and Predicted In-Sourced Costs From table 5.35, the mean, standard deviation and the minimum and maximum va lues for the actual and predicted in-sourced costs can be inferred. The variable Â“ total_insourced_costÂ” is the actual transaction cost of setting up and maintaining a disease manageme nt program in-house, whereas the variable Â“pred_in_costÂ” is the predicted cost produced by the s econd stage regression model as detailed in the earlier section. Table 5.35 Means and Statistics for Actual and Predicted In-Sourced Costs Variable N Mean Std Dev TOTAL_INSOURCED_COST 40 135708 150068.67 PRED_IN_COST 39 139187.69 124605.61 Variable Minimum Maximum TOTAL_INSOURCED_COST 0 691200 PRED_IN_COST -52659.15 585426.79
121 5.13.2 Comparison of Actual and Predicted Costs for Sub-Sample of Data Presented below in table 5.36 is a sub-sample of the integrated subset from which the prediction accuracy can be determined. Also calculated is the prediction error, whi ch is also reported for the chosen subset. The error for each case is calculated as the (actual c ostpredicted cost)/(actual cost)*100. Table 5.36 Comparison Between Actual and Predicted Costs for Sub-S ample of Data Organization Disease ACTUAL COST ($) PREDICTED COST ($) ERROR (%) Florida Health Care Plans Other: Hypertension 178,560 192,234.7674 -7.658359877 Partners HealthCare Congestive Heart Failure (CHF) 691,200 585,426.7885 15.30283731 Health Alliance Plan Diabetes 376,800 294,539.1615 21.83143271 Ault International Medical Management, LLC Other: all chronic health conditions 288,000 209,266.0364 27.33818179 Memphis Managed Care Corp Diabetes 333,600 216,655.5245 35.05529842 5.13.3 Rolling Up the Costs for Each Organization in the In-Sourced Subset Since most organizations in the integrated subset have multiple DM pr ograms, we can combine the costs for each program to get a total value for each organization. Ta ble 5.37 presents the means of the total costs (actual and predicted) for each organization. T he variable Â“sum_actual_in_costsÂ” represents the sum of the integrated costs for each disease management program offered by a particular health plan. The second variable (sum_pred_in_cos t) is the sum
122 of the predicted in-house costs for each disease management program that wer e obtained from the second stage cost model for each given organization. Table 5.37 Rolling Up the Costs for Each Organization in the Integrated Subset The MEANS Procedure Variable N Mean Std Dev Minimum Maximum SUM_ACTUAL_IN_COST 16 339270 386307.2 0 1507200 SUM_PRED_IN_COST 15 361888 322802.46 -263295.75 981298.71 5.14 Creating the Log Specification Model for the In-Sourced Costs From above we see that the predicted costs are negative for a few data point s. In order to constrain them in the positive direction and also to provide a better fit to the data, the log specification of the model to predict the in-sourced costs of a single DM progr am is taken. The results are as below: 5.14.1 Running the Model For this case, the log value of the recorded transaction costs is calcu lated and used as the dependent variable, whereas the seven independent transaction cost factors and the inverse mills ratio (lambda) are kept unchanged and used as the independent variables for th e following model. The SAS commands for this stage are given below. DATA LAMBDA_LOG; SET LAMBDA; IF TOTAL_INSOURCED_COST ^= 0 THEN TOTAL_INSOURCED_COST_LOG = LOG(TOTAL_INSOURCED_COST); RUN ;
123 The data set Â“lambdaÂ” is the data set that was originally used to constr uct the second stage cost model for the in-source subset. The log value of the actual costs are now use d as the dependent variable, whereas the transaction cost factors and the inverse mil ls ratio are used as the independent factors as before. The Â“predictedÂ” command produces the predicte d value of the costs and stores it in the variable Â“pred_in_cost_logÂ” in the new data s et called Â“insourced_pred_logÂ”. PROC REG DATA =LAMBDA_LOG; MODEL TOTAL_INSOURCED_COST_LOG = temporal physical human capital complexity similarity frequency LAMBDA1; output out =INSOURCED_PRED_LOG predicted =PRED_IN_COST_LOG; RUN ; The results have been reported in table 5.38, and contain the same terms and sta tistics as explained in section 5.11.1. There is one observation that has missing data, whi ch is removed, as in the earlier case and can be seen in the results below. Table 5.38 Analysis of Variance for Log Specification of In-Sourced Cost Model The REG Procedure Model: MODEL1 Dependent Variable: TOTAL_INSOURCED_COST_LOG Number of Observations Read 40 Number of Observations Used 39 Number of Observations with Missing Values 1
124 Table 5.38 (Continued) Analysis of Variance Source DF Sum of Squares Mean Square F Value Pr > F Model 8 45.82037 5.72755 26.71 <.0001 Error 30 6.43346 0.21445 Corrected Total 38 52.25382 Root MSE 0.46309 RSquare 0.8769 Dependent Mean 11.26964 Adj RSq 0.844 Coeff Var 4.10915 5.14.2 Results of the Log Specification In-Sourced Cost Model Table 5.39 reports the coefficients for the log specification of the in ternal cost model. For an explanation of the terms please see section 5.11.2.
125 Table 5.39 Parameter Estimates for the In-Sourced Log Specification Model Variable Label DF Parameter Estimate Standard Error t Value Pr > |t| Intercept Intercept 1 12.64238 0.63224 20 <.0001 TEMPORAL TEMPORAL 1 -1.04877 0.10819 -9.69 <.0001 PHYSICAL PHYSICAL 1 -1.4895 0.32845 -4.53 <.0001 HUMAN HUMAN 1 1.58807 0.23536 6.75 <.0001 CAPITAL CAPITAL 1 -0.50171 0.17094 -2.93 0.0063 COMPLEXITY COMPLEXITY 1 0.61511 0.22129 2.78 0.0093 SIMILARITY SIMILARITY 1 0.54914 0.11057 4.97 <.0001 FREQUENCY FREQUENCY 1 0.99731 0.13612 7.33 <.0001 LAMBDA1 1 -3.33719 0.6821 -4.89 <.0001 5.14.3 Correcting the Standard Errors for the Log Spec Model As before, this analysis produces unbiased parameter estimates for the independent variables. However, the standard estimates of these parameters are again biased be cause of heteroskedasticity, and the variance of the error term is not the same for each respondent. To correct the standard errors and get the unbiased estimates, we follow the s ame steps as outlined in section 5.11.3. The values for the variables RESS and DELTAS1 and the corrected value of the variance (VARC), the standard error (SEC) of the error term of the substantia l equation, and RHO, the correlation between the error terms of the selection and substantial equat ions is as calculated as before and is as shown below, and the values for the corrected variance, st andard error and correlation are noted in table 5.40:
126 DELTAS1 = -19.4699 RESS = 6.433455 Table 5.40 Log Specification Corrected Variance, Std. Error and Error Correlat ion The MEANS Procedure Variable N Value VARC 40 5.7247836 SEC 40 2.392652 RHO 40 -1.3947661 Computation of the standard errors of the separate observations (RHOI) and transformation of the standard errors into weights (WGT): RHOI = sqrt(VARC+LAMB*LAMB*DELTA). WGT = 1/RHOI. Now the substantial analysis is run again, with the computed weights (WGT) as weight. The results of the new regression are as reported below in table 5.41.
127 Table 5.41 Log Spec Heteroskedasticity Correction Results Number of Observations Read 40 Number of Observations Used 39 Number of Observations with Missing Values 1 Weight: WGT Analysis of Variance Source DF Sum of Squares Mean Square F Value Pr > F Model 8 38.31308 4.78914 24.4 <.0001 Error 30 5.88712 0.19624 Corrected Total 38 44.20021 Root MSE 0.44299 R-Square 0.8668 Dependent Mean 11.28769 Adj R-Sq 0.8313 Coeff Var 3.92451
128 Table 5.41 (Continued) Parameter Estimates Variable Label DF Parameter Estimate Standard Error t Value Pr > |t| Intercept Intercept 1 12.48772 0.59526 20.98 <.0001 TEMPORAL TEMPORAL 1 -1.00912 0.11035 -9.14 <.0001 PHYSICAL PHYSICAL 1 -1.38213 0.34778 -3.97 0.0004 HUMAN HUMAN 1 1.53842 0.25731 5.98 <.0001 CAPITAL CAPITAL 1 -0.47016 0.18438 -2.55 0.0161 COMPLEXITY COMPLEXITY 1 0.55885 0.24576 2.27 0.0303 SIMILARITY SIMILARITY 1 0.53643 0.12081 4.44 0.0001 FREQUENCY FREQUENCY 1 0.92685 0.13861 6.69 <.0001 LAMBDA1 1 -3.13836 0.75417 -4.16 0.0002 Now combining the parameter estimates of the substantial analysis wi th the standard errors of this WLS analysis, the Heckman procedure is completed. Thus, combining the parameter estimates and R 2 from the initial step and the standard errors corrected for heteroskeda sticity and endogeneity of the selection correction index, the final results as shown be low in table 5.42.
129 Table 5.42 Final Log Spec Cost Model Results Variable Label DF Parameter Estimate Standard Error t Value Pr > |t| Intercept Intercept 1 12.64238 0.59526 20 <.0001 TEMPORAL TEMPORAL 1 -1.04877 0.11035 -9.69 <.0001 PHYSICAL PHYSICAL 1 -1.4895 0.34778 -4.53 <.0001 HUMAN HUMAN 1 1.58807 0.25731 6.75 <.0001 CAPITAL CAPITAL 1 -0.50171 0.18438 -2.93 0.0063 COMPLEXITY COMPLEXITY 1 0.61511 0.24576 2.78 0.0093 SIMILARITY SIMILARITY 1 0.54914 0.12081 4.97 <.0001 FREQUENCY FREQUENCY 1 0.99731 0.13861 7.33 <.0001 LAMBDA1 1 -3.33719 0.75417 -4.89 <.0001 R 2 = 0.876 5.15 Comparison of Actual and Predicted In-Sourced Costs The predicted costs for the log specification of the cost model are obtaine d by taking the exponential of the predicted values obtained from the results as detai led in section 5.14.2.
130 From the above model, the predicted values of the in-sourced TCE costs are calculated again, along with the prediction error which is given as: ((actual costpredict ed cost)/ actual cost) 100, with the variable name Â“error_actÂ”. The results are as given below. T he SAS commands are as follows: DATA INSOURCED_PRED_LOG; SET INSOURCED_PRED_LOG; PRED_IN_COST_NEW = EXP(PRED_IN_COST_LOG); ERROR1_ACT = ((TOTAL_INSOURCED_COST PRED_IN_COST_NEW)/TOTAL_INSOURCED_COST)* 100 ; RUN ; PROC MEANS DATA = INSOURCED_PRED_LOG; VAR TOTAL_INSOURCED_COST PRED_IN_COST_NEW PRED_IN_COST_LOG ERROR1_ACT ; TITLE 'MEANS FOR WHOLE IN-SOURCED ORGANIZATION SET LOG SPEC' ; RUN ; 5.15.1 Means for the Actual and Predicted In-Sourced Costs From the L og Specification Model The means, standard deviation and minimum and maximum values of the actual and pr edicted costs from the log specification of the model produced by the SAS commands state d above are presented in table 5.43 along with the statistics for the prediction error.
131 Table 5.43 Means and Statistics for Actual and Predicted In-Sourced Costs From the Log Specification Model The MEANS Procedure Variable N Mean TOTAL_INSOURCED_COST 40 135708 PRED_IN_COST_NEW 39 135799.25 PRED_IN_COST_LOG 39 11.2696393 ERROR1_ACT 39 -7.0213513 Variable Std Dev Minimum Maximum TOTAL_INSOURCED_COST 150068.67 0 691200 PRED_IN_COST_NEW 172988.02 9474.78 867799.57 PRED_IN_COST_LOG 1.0980889 9.1563888 13.6737161 ERROR1_ACT 34.8613776 -103.8299626 72.7442827 Looking at the means of the actual and the predicted in-sourced costs, we see t hat the log specification of the model does a better job of cost prediction. The mean er ror is -7.102% which is a small value. We also see that all the predicted costs are constr ained in the positive direction as required. Comparing the predictions and errors of the sub-sample of in-sourced programs given below and the means of the predictions and errors from above, we see that the log specif ication model provides better cost estimates for the data provided. 5.15.2 Comparison of Actual and Predicted Costs for Sub-Sample of Data Fr om the Log Specification Model The increased accuracy and effectiveness of the log specification of the cost model can be seen from table 5.44, where a comparison of the predicted costs and associated error s can be made for a sub-sample of the data.
132 Table 5.44 Comparison of Actual and Predicted Costs for Sub-Sample of Data From the Log Specification Model Organization TOTAL_IN_COST ($) (Actual Measured Cost) PRED COST from log spec ($) PRED COST from normal reg ($) ERROR from log spec (%) ERROR from normal reg (%) QualChoice 84960 84292.11017 118441.2832 0.786123 -39.4083 Memphis Managed Care Corp 333600 205669.8881 216655.5245 38.34835 35.0553 Health Alliance Plan 376800 271550.8577 226399.5965 27.93236 39.91518 IMS Managed Care, Inc. 37440 44010.33494 49318.03133 -17.54897 -31.7255 CareGuide, Inc. 52320 57981.23142 95784.32581 -10.8204 -83.074 5.15.3 Combined Actual Cost, Combined Predicted Cost and Combined Err or Estimate Using the Log Specification of the Cost Model As in the linear specification, we can sum up the integrated costs for the various DM programs of each firm and provide a total estimate for each organization. Table 5.45 presents the means and other statistics for the costs, while table 5.46 presents the costs c alculated for the whole integrated subset. The variable Â“sum_actual_in_cost_logÂ” is the summation of the actu al in-house costs for each responding health plan, while the second variable. Â“sum_pred_in_cost_logÂ” is the total of the predicted in-sourced costs for each of the health plans in the inte grated subset of the data. The variable Â“error_actÂ” is the prediction error at the health plan le vel calculated as detailed in section 5.15.
133 Table 5.45 Summing Up the Actual and Predicted Costs and the Error Es timate for Each Organization The MEANS Procedure Variable N Mean Std Dev Minimum Maximum SUM_ACTUAL_IN_ COST_LOG 16 339,270.00 386,307.20 0 1,507,200.00 SUM_PRED_IN_ COST_LOG 15 353,078.06 382,246.54 47,373.90 1,425,801.08 ERROR_ACT 15 -3.5859624 41.549059 -103.829962 70.2153988 5.15.4 Cost and Error Estimates for the Whole In-Sourced Subset using the Log Specification of the Cost Model The whole integrated subset along with the actual and predicted costs obtai ned from the log specification of the in-sourced cost model and the error, both summed up at the organiz ation level are as reported below in table 5.46.
134 Table 5.46 Cost and Error Estimates for the Full In-Sourced Subse t From the Log Specification Model Organization Number of DM programs SUM of Actual TCE costs ($) SUM of predicted TCE costs ($) ERROR (%) Healthy Futures, Inc 1 86400 176109.0877 -103.8299626 HealthPartners 2 244800 376669.232 -53.86815032 Partners HealthCare 1 691200 867799.5654 -25.54970564 Miller & Huffman Outcome Architects, LLC 2 108000 133761.615 -23.85334724 IMS Managed Care, Inc. 5 187200 220051.6747 -17.54897152 CareGuide, Inc. 5 261600 289906.1571 -10.82039644 Solucia Inc 5 43200 47373.89726 -9.661799223 QualChoice 5 424800 421460.5509 0.786122679 Health Alliance Plan 4 1507200 1425801.083 5.400671268 Mountain States Home Care 1 77760 71412.70046 8.162679445 Contra Costa Health Plan 1 119040 108339.5013 8.988994183 Florida Health Care Plans 4 839520 734717.5781 12.48361229 Memphis Managed Care Corp 1 333600 205669.8881 38.34835489 Ault International Medical Management, LLC 1 288000 152763.6563 46.95706377 Quality First Healthcare, Inc. 1 216000 64334.73854 70.21539882
135 5.15.5 Comparison of Coefficients From the Regular and Log Specification of t he InSourced Cost Model Presented below in table 5.47 is a comparison between the coefficients of the independent transaction cost variables from both the linear and log specification of the models. Table 5.47 Coefficient Comparison Between Standard and Log Specification of the Cost Model Estimate Parameter DF First stage regression Log specification regression 277864 12.64238 Intercept 1 (2.15) (20.00) -123102 -1.04877 TEMPORAL 1 (-5.56) (-9.69) -71011 -1.4895 PHYSICAL 1 (-1.06) (-4.53) 142950 1.58807 HUMAN 1 (2.97) (6.75) -98136 -0.50171 CAPITAL 1 (-2.81) (-2.93) 18113 0.61511 COMPLEXITY 1 (0.4) (2.78) 53406 0.54914 SIMILARITY 1 (2.36) (4.97) 160158 0.99731 FREQUENCY 1 (5.75) (7.33) -335922 -3.33719 LAMBDA (inverse mills ratio) 1 (-2.41) (-4.89) t Â– Statistics in parenthesis
136 It can be seen that the log specification of the model preserves the ef fect of the transaction cost factors on the in-house costs obtained from the linear specification of the i ntegrated cost model, however, as seen from the previous sections, it also produces a better fit f or the data (R 2 = 0.876 as compared to 0.6865) and constrains the predicted costs in the positive dir ection. 5.15.6 Comparison of Summed Actual and Predicted Costs From the Regular and Log Specification of the In-Sourced Cost Model Comparison of the rolled up actual and predicted costs by the normal and the log specification of the model are as shown below in table 5.48. This table provides a side by side represe ntation of the major statistics and means obtained by the normal/linear specificat ion of the cost model with the actual in-house costs recorded. The first row presents the actual costs, while the second row reports the predicted costs obtained via the two separate methods. Table 5.48 Comparison of Costs From the Standard and Log Specification of the Cost Model Variab le N STD DEV ($) MIN ($) MAX ($) MEAN ($) Std Re g Log Spec Std Reg Log Spec Std Reg Log Spec Std Reg Log Spec Std Reg Lo g Spe c SUM OF ACTU AL COST S 16 16 386307 .2 386307.2 0 0 1507 200 15072 00 339270 339270 SUM OF PRED COST S 15 15 322802 .5 382246.5 4 2632 96 4737 3.9 9812 98.7 14258 01.1 361888 353078 .06 Error 9.6 5 % 3.5 8%
137 We see that the log specification model does a good job of TCE cost predictio n for the in-sourced subset. Thus, using both the selection and the log specification of the in-house model in tandem, the most appropriate organization form for a particular DM program and the a ssociated costs for the program if it were to be integrated by the health plan can be accurately determined for the consideration of the management and decision makers.
138 Chapter 6. Conclusions and Future Research This chapter includes the conclusion to this research and presents dir ections for further research. Section 6.1 provides the conclusions drawn from the research. In Section 6.2, the use o f organizational form and cost analysis and prediction using transaction cost e conomics as a decision making tool is presented. Section 6.3 provides the directions for futur e research in this area. 6.1 Conclusions In the preceding research, we have applied predictive modeling and switching r egression techniques to the sourcing decision problem of disease management programs in health plans to determine the factors most heavily influencing this decision and the tra nsaction costs associated with the decision. The results support the hypothesis that transaction c ost factors play a major role in determining the organizational form adopted by a health plan for such programs For the cases that were studied, and from the results of the probit models, one of the princi pal findings is that while we see that integration becomes more likely as the importance of scheduling increases, temporal specificity is not a significant factor in determining orga nizational form, as has been found to be the case in other industries. This effect may be due to the fact t hat while delays in scheduling do have an impact on the transaction costs experienced by the firm, DM pr ograms do not exhibit the phenomena where the delay in one part of the program or task ca n reverberate and cause delays throughout the rest of the project, as is seen in other industrie s such as automotive and shipbuilding industries. We also see that the factors for physical asset specificity, human asset specificity, uncertainty and dedicated asset specificity play a vital role in determining the form adopted by the organization. The results provide evidence that integra tion becomes more probable in the presence of relationship specific physical assets and t ools and for capital investment that is specific to that task, service or program.
139 It is seen that the effect of human asset specificity is the opposite, t hat is, programs that require more specific skills or knowledge are more likely to be contracted rath er than built inÂ–house. In addition, it is seen that organizations tend to outsource programs that are similar to the ones they already have in operation and those that may have a high degree of uncertainty in their performance and effectiveness measurement. The most important findings of this stage are that temporal specificity does not play as major a role as hypothesized in di sease management programs, whereas the transaction cost factors physical asset speci ficity, dedicated asset specificity and uncertainty are major factors in deciding the organi zational form chosen by a health plan for a particular disease management program. Finally, the model to predict the in-sourced costs for DM programs is const ructed. The results indicate that the factors for temporal specificity, human asset speci ficity, dedicated assets and similarity exert their influence on the costs of internal organizati on, whereas the primary effect of physical asset specificity and uncertainty is predominantly on market c osts and not on the costs of internal organization. 6.2 Organizational Form and Cost Analysis and Prediction as a Decision Ma king Tool Transaction costs are unique in the fact that unlike other costs incurre d such as direct and overhead costs, transaction costs are not recorded or measured. However, a s has been shown in this research, the costs are significant, and the efficiency and effe ctiveness of the disease management program can be greatly improved by minimizing the effect of th ese factors, while improving the profitability of the organization. The prediction results from TCE models can be used as a decision making tool by management executives of health plans. The deci sion variables in this approach are the organization form and the predicted costs. Dependi ng on the organization and management of the health plan, the type of disease management program pla nned and the levels of the various TCE factors, the executives can decide the values for the above mentioned variables and feed that into the models, which would provide them with the orga nization form to be used in that case and also the in-sourced costs for that program. Thus, a dec ision can be made whether to integrate or contract the program in order to minimize the tran saction costs incurred. For example, a health plan that has contracted one or more of its DM programs may find that the transaction costs can be reduced if the programs were to be integra ted. Whereas, the transaction costs for a DM program that has been built in-house may be calculated using our model s, and if the associated costs can be reduced by contracting to an external DMO, that option can be
140 exercised by the health plan. Thus, for a highly competitive industry such as hea lth insurance, reducing transaction costs associated with DM programs can give health plan s a competitive edge in the market and improve the management and profitability for this face t of its services considerably. In most cases, the decision makers would want to select the form such that both matches their organizational and management objectives as well as mi nimizes the transaction costs. Concluding, this research presents a decision making approach for planning the setup and management procedures for use by planning and management executives in health pla ns. The importance of internal organization costs distinct from that of market tra nsaction costs suggests that analyses of integration decisions should encompass costs of organizi ng within as well as in between firms, rather than focusing only on market transaction costs, as prev alent economic theory suggests. Based on the levels of the various factors and the possible impact, the planners can make a choice in selecting the form that best fulfils their objective s and minimizes the costs. This is a general approach and can be applied to any health plan and its DM programs for form and cost prediction. 6.3 Future W ork Even though the sample size on which the models are built is 93 observations, it captures most of the facets of the disease management industry. Hence, the methodology used t o study these observations can be extended to much larger sample sizes covering hundreds of he alth plans and DM programs. Increasing the sample size will potentially further streng then the predictive capabilities of the form and cost models. The present study was conducted wi th a small number of observations in a relatively new industry. In addition to the unique features of disease management programs, the tasks and services in this industry are also influenced by government and federal regulations, which can be taken into account by constructing new proxi es for these factors. Also, the independent transaction cost variables are imprecise proxies for the variables of true interest, hence there is a need for the refinement of these varia bles and for new proxies that permit cross firm and most ideally cross industry comparisons of transa ction cost factors and costs. Further, since only data on the internal organization costs were acc urately measurable, the burden of estimating the internal cost model was heavily dependent on the int egrated subset of the total data set.
141 A decision support software can also be developed for a general purpose use by management executives and planners in health plans and government health agencies. The proposed software would have a graphical user interface with input screens to allow users to feed the transaction cost factor levels for any disease management program. Depending on the input the software will run the two models and provide the ideal organization form and the transaction cos ts for the inÂ– sourced form of the program for the consideration of the management.
142 References  A Report on the Florida Medicaid Disease Management Program Â– Historica l Perspective, Start-up, Activities, Current Operations, Future Operati ons and Expectations. 2000. Agency for Health Care Administration  Albertson D. 1997. Is the Disease Management Ship Coming In ? Employee Benefit News, p.29.  Anderson, E. and Schmittlein, D. 1984. Integration of the Sales Force: An Empirical Examination Rand Journal of Economics, Vol.15, pp.385-389.  Badaracco, Joseph L., Jr. 1991. The Knowledge Link: How Firms Compete Through Strategic Alliances. Boston: Harvard Business School Press.  Bagga, Sudhir. 2003. Integrative Medicine-A Practical Application IEEE, Vol.8, No.3, pp.137-142.  Blumberg, Michael R., and Donald F. Blumberg. 1994. Strategic Directions for Optimum Outsourcing The Source Fall/Winter 1994, pp.8-15.  Breen, Richard. 1996. Regression Models: Censored, Sample Selected, or Truncated Data. Sage University Paper No.111. Thousand Oaks: Sage.  Buzzel, R. 1983. Is Vertical Integration Profitable ? Harvard Business Review, Vol.61, pp.92-102.  Casson, Mark. 1986. The Role of Vertical Integration in the Shipping Industry Journal of Transport Economics and Policy, pp 7-29.  Casson, Mark. 1987. The Firm and the Market Cambridge: MIT Press  Chandler, Margaret K. 1961. Competition between the Inside and Outside Labour Force for the Work of the Industrial firm Proceedings of the Industrial Relations Research Association, pp.334-335.
143  Coase, Ronald H. 1937. The Nature of the Firm Economica n.s., Vol.4, pp.386-405.  Coase, Ronald H. 1972. Industrial Organization: A Proposal for Research, Policy Issues and Research Opportunities in Industrial Organization, in V.R. Fuchs (ed.), New York: National Bureau for Economic Research).  Coase, Ronald H. 1988. The Nature of the Firm: Origin, Meaning, Influence, Journal of Law, Economics, and Organization, pp.3-47.  Corbett, Michael F. 1995 Top Ten Reasons Companies Outsource The Outsourcing Institute, 1995 BuyerÂ’s Guide, pp.29-31.  DÂ’Aveni, Richard. A. and David J. Ravenscraft. 1994. Economies of Integration Versus Bureaucracy Costs: Does Vertical Integration Improve Performance ? Academy of Management Journal Vol.5, pp.1167-1206.  Disease Management and Medicaid 2004. The 11 th Princeton Conference: Managing Cost and Quality through the Health Care Delivery System  Disease Management Association of America. http://www.dmaa.org.  Disease Management Purchasing Consortium. http://www.dismgmt.com.  Einstein, Albert B, Jr, Patricia J. Goldsmith, and David W. Maberry. 2006. Contracting with a Medicare HMO Cancer Economics  Fitzner, Karen A., Cynthia Gomez, and Heather Chappell. 2005. DMAA Quality and Research Initiative Disease Management, Vol.8, No.6, pp.392-396.  Fox, Heidi. 2005. Florida Medicaid: Future State of Disease Management Disease Management RFI Workshop, Agency for Health Care Administration.  Fuhr, Johannes, and Thorsten Beckers. 2006. Vertical Governance between Airlines and AirportsA Transaction Cost Analysis Center for Network Industries and Infrastructure  Gillette, Bill. 2006. Disease Management Moments Managed Healthcare Executive, Vol.16, No.3, pp.32-33.  Griliches, Zvt. 1957. Specification Bias in Estimates of Production Function Journal of Farm Economics, Vol.39, pp.8-20.
144  Hanoch, G. 1996. A Multivariate Model of Labor Supply: Methodology and Estimation Princeton, N.J., Princeton University Press.  Harrigan, Kathryn R. 1983. Strategies for Vertical Integration, Lexington, Mass.: Lexington Books.  Harrigan, Kathryn R., and William H. Newman. 1990. Bases of Interorganization Cooperation: Propensity, Power, Persistence Journal of Management Studies Vol.27, No.4, pp.417-434.  Healthcare Business October 1999, p.29.  Heckman, James J. 1979. Sample Selection Bias as a Specification Error Econometrica, Vol.47, No.1, pp.153-161.  Heckman, James J., and Thomas E. Macurdy. 1980. A Life Cycle Model of Female Labour Supply. The Review of Economic Studies Vol.47, No.1, pp.47-74.  Hennart, Jean-Franois. 1988. Upstream Vertical Integration in the Aluminum and Tin Industries Journal of Economic Behavior and Organization Vol.9, pp.281-299.  Hoffman, C., Rice, D., & Sung, H. Y. 1996. Persons with chronic conditions. Their prevalence and costs JAMA 276 pp.1473-1479.  Holtz-Eakin, Douglas. 2004. An Analysis of the Literature on Disease Management Programs Congressional Budget office  Humana DM Program Propelled by Strong CHF Results 1996. Disease Management News p.1.  Joskow, Paul L. 1985. Vertical Integration and Long Term Contracts: The Case of CoalBurning Electric Generation Plants Journal of Law, Economics and Organization Vol.1, pp.33-79.  Kissell ,Robert and Morton Glantz. 2003. Optimal Trading Strategies AMACOM.  Klein, Peter G. 1996. Economic Calculation and the Limits of Organization The Review of Austrian Economics, Vol.9, No.2, pp.3-28.  Leeman, Jennifer, and Barbara Mark. 2006. The Chronic Care Model Versus Disease Management Programs: A Transaction Cost Analysis Approach Health Care Management Review, Vol.31, No.1, pp.18-25.
145  Lehmann, D. R. and OÂ’ Shaughnessy, J. 1974. Difference in attribute importance for different industrial products Journal of Marketing, Vol.38, pp.36-42.  Lewis, A. 1998. Outsourcing: Lessons Learned Disease Management Purchasing Consortium.  Lewis, A. 1999. Outsourcing: Disease ManagementÂ’s Magic Bullet Disease Management Purchasing Consortium.  Managed Care magazine. http://www.managedcaremag.com.  Masten, Scott E. 1993. Transaction Costs, Mistakes, and Performance: Assessing the Importance of Governance Managerial and Decision Economics Vol.27, No.2, pp.119129.  Masten, Scott E. 1984. The Organization of Production: Evidence from the Aerospace Industry Journal of Law and Economics Vol.27, pp.403-417.  Masten, Scott E., James W. Meehan, and Edward A. Snyder. 1989. Vertical Integration in the U.S. Auto Industry: A Note on the Influence of Transaction Specific Assets Journal of Economic Behaviour and Organization Vol.12, pp.265-273.  Masten, Scott E., James W. Meehan, and Edward A. Snyder. 1991. The Costs of Organization Journal of Law, Economics and Organization Vol.7, pp.1-25.  Matheson, M., Anne Wilkins and Daphne Psacharopoulos. 2006. Realizing the promise of Disease ManagementÂ–Payer Trends and Opportunities in the United States The Boston Consulting Group.  McFetridge D.G and Smith D. 1989. The Economics of Vertical Disintegration The Fraser Institute, Vancouver.  Monteverde, Kirk, and David J. Teece. 1982. Supplier Switching Costs and Vertical Integration in the Automobile Industry Bell Journal of Economics Vol.13, Spring 1982a, pp.206-213.  Monteverde, Kirk, and David J. Teece. 1982. Appropriable Rents and Quasi-Vertical Integration Journal of Law and Economics Vol.25, 1982b, pp.321-328.  Photis M. Panayides and K. Cullinane. 2002. The vertical disintegration of ship management: choice criteria for third party selection and evaluation The Hong Kong Polytechnic University Department of Shipping and Transport Logistic s, Maritime Policy & Management, Taylor & Francis Ltd, Vol.29, No.1, pp.45-64.
146  Pirrong, Stephen C. 1994. Contracting Practices in Bulk Shipping Markets: A Transaction Cost Explanation Journal of Law and Economics Vol.36, pp.937-976.  Prahalad, C., and G. Hamel. 1990. The Core Competence of the Corporation Harvard Business Review pp.79-91.  Pruitt, Sheri; Annandale, Steve; Epping-Jordan, JoAnne; Fernndez Daz, Jes s M.; Khan, Mahmud; Kisa, Adnan; Klapow, Joshua; Nuo Solinis, Roberto ; Reddy, Srinath; Wagner, Ed. Innovative care for Chronic Conditions: Building Blocks for Action 2002. WHO Global Report  Quilty M, Lewis A. 1997. Case Studies in Disease Management for Congestive Heart Failure Medical Interface p.97.  Quilty M, Weiner J, Francis M et al 1997. Unleashing the Profit potential of Disease Management: Profile of Leading Strategies and Strategists Spectrum Disease Management: Strategic Approaches and Program Analysis, Decision Reso urces.  Quinn, James Brian, and Frederick G. Hilmer. 1994. Strategic Outsourcing Sloan Management Review pp.43-55.  Renaissance Offers Â‘Sign & SaveÂ’ Program. 1997. Disease Management News p.5.  Shelanski, Howard A., and Peter G. Klein. 1995. Empirical Research in Transaction Cost Economics: A Review and Assessment Journal of Law, Economics & Organization, Vol.11, No.2, pp.335-361.  Silver, M. 1984. Enterprise and the Scope of the Firm Oxford: Martin Robertson.  Simon, H. A. 1957. Models of Man Wiley, New York.  Simon, Herbert. 1957. A Behavioral Model of Rational Choice in Models of Man, Social and Rational: Mathematical Essays on Rational Human Behavior in a Soci al Setting. New York: Wiley.  Smits, Jeroen. 2003. Estimating the Heckman two-step procedure to control for selection bias with SPSS  Stigler, George J. 1951. The Division of Labour Is Limited by the Extent of the Market Journal of Political Economy Vol.59, pp.185-193.
147  Straub Detmar Dr. and Soon Ang. 1998. Production and Transaction Economies and IS Outsourcing: A Study of the U.S. Banking Industry. MIS Quarterly, Vol.22, No.4, pp.535-552.  Stuckey, John. 1983. Vertical Integration and Joint Ventures in the Aluminum Industry Cambridge, Mass.: Harvard University Press.  The Florida Medicaid Disease Management Initiative. 2000. Agency for Hea lth care Administration.  Theil, H. 1957. Specification Errors and the Estimation of Economic Relationships Revue de lÂ’Institut International de Statistique Vol.25, pp.41-51.  Thompson, D., Edelsberg, J., Kinsey, K.L., Oster, G. 1998. Estimated Economic Costs of obesity to U.S Business American Journal of Health Promotion, Vol.13, No.2, pp.120127.  Vendors Lead Voters. 1997. Winners in CSC-DMN Survey Disease Management News, p.3.  Walker, Tracey. 2006. Disease Management Programs Offered by HMOs Managed Healthcare Executive, ABI/INFORM Global, Vol.16, No.4, p.60.  Wallis, J. and North, C. 1986. Measuring the Transaction Sector in the American Economy, 1870-1970. Gallman (ed.), Long Term Factors in American Economic Growth, Chicago, University of Chicago Press, pp.95-148.  Wheatley, Ben. 2002. Disease Management: Findings from Leading State programs Academy Health, Vol.3, No.3.  Williamson, Oliver E. 1975. Markets and Hierarchies: Analysis and Antitrust Implications. Free Press: New York.  Williamson, Oliver E. 1979. Transaction-Cost Economics: The Governance of Contractual Relations Journal of Law and Economics Vol.22, No.2, pp.233-261.  Williamson, Oliver E. 1981. The Economics of Organisation: The Transaction Cost Approach American Journal of Sociology, Vol.87, No.3.  Williamson, Oliver E. 1985. The Economic Institutions of Capitalism, New York: Free Press.
148  Williamson, Oliver E. 1989. Transaction Cost Economics in R. Schmalensee and R. D. Willig (eds.), Handbook of Industrial Organization New York: Elsevier Science Publishers B.V, Vol.1.  Williamson, Oliver E. 1990. Comparative Economic Organization: The Analysis of Discrete Structural Alternatives University of California, Berkeley, School of Business.  Williamson, Oliver E. 1993. Opportunism and Its Critics Managerial and Decision Economics Vol.14, No.2.  Williamson, Oliver E. 1996. The Mechanisms of Governance New York: Oxford University.  Williamson, Oliver E. 2000. The New Institutional Economics: Taking Stock, Looking Ahead Journal of Economic Literature Vol.38, No.3, pp.595-613.
150 Appendix A. SAS Code /******************IMPORTING ****************************************/ /*****************THE DATA SET**********************************/ /***************************************************************/ proc import datafile = C:\Documents and Settings\nchandav\Desktop\SURVEY.xls" out =SURVEY replace ; run ; /******************"SURVEY" IS RAW DATA SET*****************************************/ /***************************************************************************** *****/ data survey1; set survey; if form = 'In-sourced/Integrated' then DEP = 1 ; else if form = 'Outsourced' then DEP = 0 ; run ; data survey2 ; set survey1; if DEP in ( 0 1 ); run ; /**********GETTING RELEVANT FREQUENCIES*** ************************************/ proc freq data = survey2; tables DEP form temporal physical human capital complexity similarity frequency uncertainty ORGANIZATION DISEASE/
151 Appendix A (Continued) norow nocol nopercent ; TITLE 'Responding organizations and number of respective responses' ; run ; /**********CHECKING CORRELATION BETWEEN ALL 9 VARIABLES IN FULL DATA SET*****************/ proc CORR data = survey2; VAR DEP temporal physical human capital complexity similarity frequency uncertainty ; TITLE 'Means and correlations for all variables' ; run ; /* PICK 80 RANDOM DATA POINTS FOR TRAINING SET*/ data SURVEY3 (DROP = address email position email telephone nr date name) ; set survey2; x = ranuni( 4546654 ); run ; proc sort data = SURVEY3; by x; run ; data TRAINING VALIDATION; set SURVEY3; if _n_ <= 80 THEN OUTPUT TRAINING; ELSE OUTPUT VALIDATION; run ; DATA TRAINING; SET TRAINING; TYPE = 'TRAINING' ;
152 Appendix A (Continued) RUN ; PROC FREQ DATA = TRAINING; TABLES DEP; TITLE 'Frequencies for the training set' ; RUN ; DATA VALIDATION; SET VALIDATION; TYPE = 'VALIDATION' ; ACTUAL_FORM = DEP; RUN ; PROC FREQ DATA = VALIDATION; TABLES ACTUAL_FORM; TITLE 'Frequencies for the validation set' ; RUN ; DATA VALIDATION (DROP = DEP); SET VALIDATION; RUN ; /* BUILDING THE MODEL WITH THE TRAINING SET*/ proc logistic data =TRAINING descending ; model DEP = temporal physical human capital complexity similarity frequency uncertainty/ LINK =PROBIT ctable pprob =( 0.05 to 1 by 0.05 ); output out =prob XBETA = g predicted =phat; TITLE 'FIRST STAGE SELECTION MODEL' ; run ;
153 Appendix A (Continued) DATA COMBINED; SET TRAINING VALIDATION; RUN ; /* RUNNING THE MODEL WITH THE COMBINED SET*/ proc logistic data =COMBINED descending ; model DEP = temporal physical human capital complexity similarity frequency uncertainty/ LINK =PROBIT ; output out =prob2 XBETA = g2 predicted =phat2; TITLE 'FIRST STAGE SELECTION MODEL FOR COMBINED DATA SET' ; run ; DATA PROB2; SET PROB2; IF PHAT2 >= 0.55 THEN PRED_FORM = 1 ; ELSE PRED_FORM = 0 ; RUN ; /**********CHECKING CLASSIFICATION OF TRAINING SET***************************************/ PROC FREQ DATA = PROB2; TABLES PRED_FORM*DEP/ norow nocol nopercent ; TITLE 'Classification table for Training set' ; RUN ; /**********CHECKING CLASSIFICATION OF VALIDATION SET***************************************/
154 Appendix A (Continued) PROC FREQ DATA = PROB2; TABLES PRED_FORM*ACTUAL_FORM/ norow nocol nopercent ; TITLE 'Classification table for validation set' ; RUN ; /**********STEP Â– I-b ***************************************/ /**********2 ND CASE FOR MODEL *************************************/ /**********TRAINING VALIDADATION FOR LOGISTIC PROBIT**************/ /**********WITH FACTOR UNCERTAINTY REMOVED***********************/ proc logistic data =TRAINING descending ; model DEP = temporal physical human capital complexity similarity frequency/ LINK =PROBIT ctable pprob =( 0.05 to 1 by 0.05 ); output out =prob_NO_UNCERT XBETA = g_NO_UNCERT predicted =phat_NO_UNCERT; TITLE 'FIRST STAGE SELECTION MODEL WITH NO UNCERTAINTY' ; run ; DATA COMBINED_NO_UNCERT; SET TRAINING VALIDATION; RUN ; proc logistic data =COMBINED_NO_UNCERT descending ; model DEP = temporal physical human capital complexity similarity frequency / LINK =PROBIT ; output out =prob2_NO_UNCERT XBETA = g2_NO_UNCERT predicted =phat2_NO_UNCERT; TITLE 'FIRST STAGE SELECTION MODEL FOR COMBINED DATA SET WITH NO UNCERTAINTY' ;
155 Appendix A (Continued) run ; DATA PROB2_NO_UNCERT; SET PROB2_NO_UNCERT; IF PHAT2_NO_UNCERT >= 0.55 THEN PRED_FORM = 1 ; ELSE PRED_FORM = 0 ; RUN ; /**********CHECKING CLASSIFICATION OF TRAINING SET***************************************/ PROC FREQ DATA = PROB2_NO_UNCERT; TABLES PRED_FORM*DEP/ norow nocol nopercent ; TITLE 'Classification table for training set' ; RUN ; /**********CHECKING CLASSIFICATION OF VALIDATION SET***************************************/ PROC FREQ DATA = PROB2_NO_UNCERT; TABLES PRED_FORM*ACTUAL_FORM/ norow nocol nopercent ; TITLE 'Classification table for validation set' ; RUN ; /**********PART II ************************************************/ /*********CALCULATION OF IN-SOURCED AND ******************************/ /**********OUTSOURCED COSTS **************************************/ /********** ($60 HOURLY RATE) ***************************************/
156 Appendix A (Continued) data COMBINED2; set PROB2; IF ACTUAL_FORM ^= THEN DEP = ACTUAL_FORM; RUN ; data insourced2; set COMBINED2; if (dep = 1 ); INSOURCED_ADMIN_COST = STARTUP_TIME_INSOURCED_DAYS* 8 60 ; if INSOURCED_ADMIN_COST = then INSOURCED_ADMIN_COST = 0 ; SEARCH_INFO_COST = SEARCH_INFO_TIME_DAYS* 8 60 ; if SEARCH_INFO_COST = then SEARCH_INFO_COST = 0 ; SUPERVISORY_COST = SUPERVISORY_POLICING_TIME_HOURS_* 4 12 60 ; if SUPERVISORY_COST= then SUPERVISORY_COST = 0 ; TOTAL_INSOURCED_COST = INSOURCED_ADMIN_COST+SEARCH_INFO_COST+SUPERVISORY_COST; run ; data outsourced2; set COMBINED2; if (dep = 0 ); SEARCH_INFO_COST = SEARCH_INFO_TIME_DAYS* 8 60 ;
157 Appendix A (Continued) if SEARCH_INFO_COST = then SEARCH_INFO_COST = 0 ; SUPERVISORY_COST = SUPERVISORY_POLICING_TIME_HOURS_* 4 12 60 ; if SUPERVISORY_COST= then SUPERVISORY_COST = 0 ; if LEGAL_COST_OUTSOURCED_DOLLARS= then LEGAL_COST_OUTSOURCED_DOLLARS = 0 ; TOTAL_OUTSOURCED_COST = LEGAL_COST_OUTSOURCED_DOLLARS+SEARCH_INFO_COST+SUPERVIS ORY_COS T; run ; /* MEANS FOR IN Â– SOURCED COSTS*/ proc means data = insourced2 ; var INSOURCED_ADMIN_COST SEARCH_INFO_COST SUPERVISORY_COST TOTAL_INSOURCED_COST ; TITLE 'MEANS FOR IN SOURCED COSTS' ; run ; /* MEANS FOR OUTSOURCED COSTS*/ proc means data = outsourced2 ; var LEGAL_COST_OUTSOURCED_DOLLARS SEARCH_INFO_COST SUPERVISORY_COST TOTAL_OUTSOURCED_COST; TITLE 'MEANS FOR OUT SOURCED COSTS' ; run ;
158 Appendix A (Continued) /**********PART III ************************************************/ /*********CALCULATION OF THE ******************************/ /**********THE INVERSE MILLS RATIO ******************************/ /**********FOR THE HECKMAN 2 STAGE ESTIMATION************************/ /**********TO DERIVE THE IN-SOURCED************************/ /*********COST EQUATION ******************************/ DATA LAMBDA; SET INSOURCED2; IF DEP= 1 AND TOTAL_INSOURCED_COST ^= AND TOTAL_INSOURCED_COST ^= 0 THEN DO ; PDFG2 = PDF( 'NORMAL' ,G2); CDFG2 = CDF( 'NORMAL' ,G2); LAMBDA1 = (( 1 /sqrt( 2 3.141592654 ))*(exp(-G2*G2* 0.5 )))/CDF( 'NORMAL' ,G2); LAMBDA2 = (PDFG2/CDFG2); lambda3=( 1 /sqrt( 2 3.141592654 )*exp(1 *g2** 2 / 2 ))/probnorm(g2); DELTA1 = -LAMBDA1*G2-LAMBDA1*LAMBDA1; DELTA2 = -LAMBDA2*G2-LAMBDA2*LAMBDA2; DELTA3 = -LAMBDA3*G2-LAMBDA3*LAMBDA3; h1=lambda1** 2 +lambda1*g2; h2=lambda2** 2 +lambda2*g2; h3=lambda3** 2 +lambda3*g2; END ; RUN ; PROC MEANS DATA = LAMBDA ; VAR LAMBDA1 LAMBDA2 lambda3 DELTA1 DELTA2 DELTA3 H1 H2 H3; TITLE 'RESULTS FOR THE INVERSE MILLS RATIO AND CONTROL FACTOR DEL TA' ; RUN ; /* BUILDING THE IN Â– SOURCED COST MODEL*/
159 Appendix A (Continued) PROC REG DATA =LAMBDA; MODEL TOTAL_INSOURCED_COST = temporal physical human capital complexity similarity frequency LAMBDA1; output out =INSOURCED_PRED predicted =PRED_IN_COST; TITLE 'ORGANIZATION COST MODEL FOR IN SOURCED COSTS' ; RUN ; DATA INSOURCED_PRED; SET INSOURCED_PRED; ERROR = ( (TOTAL_INSOURCED_COST PRED_IN_COST)/TOTAL_INSOURCED_COST)* 100 ; RUN ; /* MEANS FOR PREDICTED AND ACTUAL IN Â– SOURCED COSTS*/ PROC MEANS DATA = INSOURCED_PRED; VAR TOTAL_INSOURCED_COST PRED_IN_COST ; TITLE 'Comparison of Actual and predicted in sourced costs' ; RUN ; /**********CORRECTING HETEROSKEDASTICITY FOR COST MODEL*******************/ /*************************************************************************/ PROC REG DATA =LAMBDA; MODEL TOTAL_INSOURCED_COST = temporal physical human capital complexity similarity frequency LAMBDA1; output out =INSOURCED_PRED predicted =PRED_IN_COST residual = RES; RUN ; DATA INSOURCED_PRED; SET INSOURCED_PRED;
160 Appendix A (Continued) RES2 = RES*RES; LAMB= 335922 ; N= 39 ; RUN ; PROC SQL ; SELECT SUM(DELTA1) AS DELTAS1 FROM INSOURCED_PRED; QUIT ; PROC SQL ; SELECT SUM(RES2) AS RESS FROM INSOURCED_PRED; QUIT ; DATA INSOURCED_PRED; SET INSOURCED_PRED; DELTAS1 = 19.4699 ; RESS = 2.694E11 ; VARC = RESS/N-LAMB*LAMB*DELTAS1/N; SEC = sqrt(VARC); RHO = sqrt(LAMB*LAMB/VARC); If (lamb< 0 ) THEN RHO = 0 -RHO; C = VARC+LAMB*LAMB*DELTA1; RHOI = sqrt(C); WGT = 1 /RHOI; RUN ; PROC MEANS DATA = INSOURCED_PRED; VAR VARC SEC RHO;
161 Appendix A (Continued) RUN ; PROC REG DATA =INSOURCED_PRED; MODEL TOTAL_INSOURCED_COST = temporal physical human capital complexity similarity frequency LAMBDA1; weight WGT; output out =INSOURCED_PRED_NEW predicted =PRED_IN_COST_NEW residual = RES_NEW; title 'CORRECTING HETEROSKEDASTICITY FOR COST MODEL' ; RUN ; /*************************************************************************/ /**********PART IV ************************************************/ /*********CALCULATION OF THE ******************************/ /**********COMBINED COSTS******************************/ /**********FOR EACH HEALTH PLAN IN THE ************************/ /**********IN Â– SOURCED DATA SET************************/ DATA INSOURCED_PRED_C; SET INSOURCED_PRED; IF ORGANIZATION = 'Health Alliance Plan' THEN ORGANIZATION = 'Health Alliance Plan' ; RUN ; PROC SQL ; CREATE TABLE SUM_COSTS AS SELECT *, SUM(TOTAL_INSOURCED_COST) AS SUM_ACTUAL_IN_COST, SUM(PRED_IN_COST) AS SUM_PRED_IN_COST FROM INSOURCED_PRED_C GROUP BY ORGANIZATION; quit ;
162 Appendix A (Continued) PROC SORT DATA = SUM_COSTS; BY ORGANIZATION; RUN ; DATA SUM_COSTS3; SET SUM_COSTS; BY ORGANIZATION; IF FIRST.ORGANIZATION; RUN ; DATA SUM_COSTS3; SET SUM_COSTS3; ERROR_ACT = ((SUM_ACTUAL_IN_COST SUM_PRED_IN_COST)/SUM_ACTUAL_IN_COST)* 100 ; ERROR_ABS = (ABS(SUM_ACTUAL_IN_COST SUM_PRED_IN_COST)/SUM_ACTUAL_IN_COST)* 100 ; RUN ; /* MEANS FOR IN Â– SOURCED COSTS (WHOLE INTEGRATED SUBSET)*/ PROC MEANS DATA = SUM_COSTS3; VAR SUM_ACTUAL_IN_COST SUM_PRED_IN_COST ERROR_ACT ERROR_ABS; TITLE 'MEANS FOR IN SOURCED ORGANIZATION SET ; RUN ; /**************************************************/ /**************************************************/ DATA SUM_COSTS4; SET SUM_COSTS3;
163 Appendix A (Continued) IF TOTAL_INSOURCED_COST^= 0 ; RUN ; PROC MEANS DATA = SUM_COSTS4; VAR SUM_ACTUAL_IN_COST SUM_PRED_IN_COST ERROR_ACT ERROR_ABS; TITLE 'MEANS FOR IN SOURCED ORGANIZATION WITH 1 MISSING REMOVED' ; RUN ; /**************************************************/ /**************************************************/ proc sort data = SUM_COSTS3; BY ERROR_ACT; RUN ; /**********PART V ************************************************/ /**************LOG SPECIFICATION*********************************/ /**************OF THE IN SOURCED************************/ /*********COST MODEL FOR BETTER MODEL FIT****************************/ /*************AND POSITIVE CONSTRAINING******************************/ DATA LAMBDA_LOG; SET LAMBDA; TEMPORAL_LOG = log(TEMPORAL); physical_LOG = log(physical); human_LOG = log(human); capital_LOG = log(capital); complexity_LOG = log(complexity); similarity_LOG = log(similarity); frequency_LOG = log(frequency);
164 Appendix A (Continued) LAMBDA1_LOG = log(LAMBDA1); IF TOTAL_INSOURCED_COST ^= 0 THEN TOTAL_INSOURCED_COST_LOG = LOG(TOTAL_INSOURCED_COST); RUN ; /* LOG SPECIFICATION OF THE COST MODEL*/ PROC REG DATA =LAMBDA_LOG; MODEL TOTAL_INSOURCED_COST_LOG = temporal physical human capital complexity similarity frequency LAMBDA1; output out =INSOURCED_PRED_LOG predicted =PRED_IN_COST_LOG; RUN ; /*************************************************************************/ /***********HETEROSKEDASTICITY CORRECTION FOR LOG SPEC***********************************/ /***************************************************************************** ************/ PROC REG DATA =LAMBDA_LOG; MODEL TOTAL_INSOURCED_COST_LOG = temporal physical human capital complexity similarity frequency LAMBDA1; output out =INSOURCED_PRED_LOG predicted =PRED_IN_COST_LOG residual = RES; RUN ; /**************************************************/ /**************************************************/ DATA INSOURCED_PRED_LOG; SET INSOURCED_PRED_LOG; RES2 = RES*RES; LAMB= 3.33719 ; N= 39 ; RUN ; PROC SQL ;
165 Appendix A (Continued) SELECT SUM(DELTA1) AS DELTAS1 FROM INSOURCED_PRED_LOG; QUIT ; PROC SQL ; SELECT SUM(RES2) AS RESS FROM INSOURCED_PRED_LOG; QUIT ; DATA INSOURCED_PRED_LOG; SET INSOURCED_PRED_LOG; DELTAS1 = 19.4699 ; RESS = 6.433455 ; VARC = RESS/N-LAMB*LAMB*DELTAS1/N; SEC = sqrt(VARC); RHO = sqrt(LAMB*LAMB/VARC); If (lamb< 0 ) THEN RHO = 0 -RHO; C = VARC+LAMB*LAMB*DELTA1; RHOI = sqrt(C); WGT = 1 /RHOI; RUN ; PROC MEANS DATA = INSOURCED_PRED_LOG; VAR VARC SEC RHO; RUN ; PROC REG DATA =INSOURCED_PRED_LOG; MODEL TOTAL_INSOURCED_COST_LOG = temporal physical human capital complexity similarity frequency LAMBDA1; weight WGT;
166 Appendix A (Continued) output out =INSOURCED_PRED_NEW_LOG predicted =PRED_IN_COST_NEW_LOG residual = RES_NEW_LOG; title 'CORRECTING HETEROSKEDASTICITY FOR LOG SPEC OF COST MODEL' ; RUN ; /*****************************************************************************/ /**************************************************/ /**************************************************/ DATA INSOURCED_PRED_LOG; SET INSOURCED_PRED_LOG; PRED_IN_COST_NEW = EXP(PRED_IN_COST_LOG); ERROR1_ACT = ((TOTAL_INSOURCED_COST PRED_IN_COST_NEW)/TOTAL_INSOURCED_COST)* 100 ; ERROR1_ABS = (ABS(TOTAL_INSOURCED_COST PRED_IN_COST_NEW)/TOTAL_INSOURCED_COST)* 100 ; RUN ; PROC MEANS DATA = INSOURCED_PRED_LOG; VAR TOTAL_INSOURCED_COST PRED_IN_COST_NEW PRED_IN_COST_LOG ERROR1_ACT ERROR1_ABS; TITLE 'MEANS FOR WHOLE IN SOURCED ORGANIZATION SET LOG SPEC' ; RUN ; /**************************************************/ /**************************************************/ DATA INSOURCED_PRED_LOG_B; SET INSOURCED_PRED_LOG; IF TOTAL_INSOURCED_COST^= 0 ; RUN ; PROC MEANS DATA = INSOURCED_PRED_LOG_B;
167 Appendix A (Continued) VAR TOTAL_INSOURCED_COST PRED_IN_COST_NEW PRED_IN_COST_LOG ERROR1_ACT ERROR1_ABS; TITLE 'MEANS FOR IN SOURCED ORGANIZATION SET LOG SPEC WITH 1 MISSI NG REMOVED' ; RUN ; /**************************************************/ /**************************************************/ DATA GOOD_ERROR_LOG; SET INSOURCED_PRED_LOG; WHERE ERROR1_ABS <= 100 ; RUN ; PROC MEANS DATA = GOOD_ERROR_LOG; VAR ERROR1_ABS ; RUN ; /**************CREATING TABLE************************/ /***************WITH COMBINED COSTS FOR EACH ORGANIZATION******************/ /******************FOR COMBINED ERROR CALCULATION**************************/ DATA INSOURCED_PRED_LOG2; SET INSOURCED_PRED_LOG; IF ORGANIZATION = 'Health Alliance Plan' THEN ORGANIZATION = 'Health Alliance Plan' ; RUN ; /* FREQUENCIES FOR INTEGRATED SUBSET*/
168 Appendix A (Continued) PROC FREQ DATA = INSOURCED_PRED_LOG2 ; TABLES ORGANIZATION DISEASE ; TITLE 'FREQUENCIES FOR IN SOURCED SUB SET' ; RUN ; /* CALCULATING TOTAL COST PER HEALTH PLAN*/ PROC SQL ; CREATE TABLE SUM_COSTS_LOG AS SELECT *, SUM(TOTAL_INSOURCED_COST) AS SUM_ACTUAL_IN_COST_LOG, SUM(PRED_IN_COST_NEW) AS SUM_PRED_IN_COST_LOG FROM INSOURCED_PRED_LOG2 GROUP BY ORGANIZATION; quit ; DATA SUM_COSTS_LOG2; SET SUM_COSTS_LOG; BY ORGANIZATION; IF FIRST.ORGANIZATION; RUN ; DATA SUM_COSTS_LOG2; SET SUM_COSTS_LOG2; ERROR_ACT = ((SUM_ACTUAL_IN_COST_LOG SUM_PRED_IN_COST_LOG)/SUM_ACTUAL_IN_COST_LOG)* 100 ; ERROR_ABS = (ABS(SUM_ACTUAL_IN_COST_LOG SUM_PRED_IN_COST_LOG)/SUM_ACTUAL_IN_COST_LOG)* 100 ; RUN ; PROC MEANS DATA = SUM_COSTS_LOG2;
169 Appendix A (Continued) VAR SUM_ACTUAL_IN_COST_LOG SUM_PRED_IN_COST_LOG ERROR_ACT ERROR_ABS; TITLE 'MEANS FOR WHOLE ORGANIZATION SET' ; RUN ; DATA SUM_COSTS_LOG3; SET SUM_COSTS_LOG2; IF TOTAL_INSOURCED_COST ^= 0 ; RUN ; PROC MEANS DATA = SUM_COSTS_LOG3; VAR SUM_ACTUAL_IN_COST_LOG SUM_PRED_IN_COST_LOG ERROR_ACT ERROR_ABS; TITLE 'MEANS FOR ORGANIZATION SET WITH MISSING REMOVED' ; RUN ; proc sort data = SUM_COSTS_LOG2; BY ERROR_ACT; RUN ;
170 Appendix B. Electronic Survey Disease Management (DM) Outsourcing Survey Please take the time to answer the following questions; your input is greatly appreciated. Please include your contact information so the results may be sent to you. Preliminary Information: Please enter the name of your organization: Please enter your name: Please enter your position: Please enter your email address: Please enter your telephone number: Please enter your mailing address:
171 Appendix B (Continued) Disease management Questions: Q 1) Please enter the disease for which the program has been imp lemented: Diabetes Asthma Coronary Artery Disease(CAD) Congestive Heart Failure (CHF) Chronic Obstructive Pulmonary Disease (COPD) Other: Q 2) Is this Disease Management program: In-sourced/Integrated (go to Q3-a next) Outsourced (go to Q3-b next) other: Please specify Q3-a) What was the approx. time spent (in days) in administrative, facility planning, and other start-up tasks prior to implementation of this in-sou rced program? (Go to Q 4 next) Q3-b) What was the approx. legal cost (in $) involved in bargaining, negotia ting and drawing up an appropriate contract for this outsourced DM program? Q 4) What was the approx. time spent (in days) to obtain relevant information in preparation for implementing this program?
172 Appendix B (Continued) Q 5) What is the approx. time spent (in hours) per week for supervis ory and managerial tasks for this program? Please rate the following on a scale from 1 to 5: Q 6) Scheduling requirements for a particular task or service (su ch as patient interventions, patient/program effectiveness checks, and risk evaluations) are som etimes critical in a particular program. On the other hand, there is more flexibility r egarding the timely completion of tasks and services in other disease management program s. Using the scale, rate how important, in terms of costs, it is to have tasks in this p rogram done on schedule. 1 "Not Important" 2 3 4 5 "Very Important" Q 7) To what extent are the tools and assets such as the clinical dat abases and feedback systems, predictive models for patient identification, and monitor ing and reporting processes specific to this program? Using the scale below, rate t he specificity of the assets required for this program. 1 "Relatively Standard" 2 3 "Somewhat Specific" 4 5 "Very Specific" Relatively Standard the facilities, assets, etc., used in the program can be easily adapted fo r use by other industries and other disease management programs. Somewhat Specific the facilities, assets, etc., used in the program can be easily adapted fo r use by other disease management programs. Very Specific the facilities, assets, etc., used in the program cannot be easily adapted for use by others, even other disease management programs.
173 Appendix B (Continued) Q 8) To what extent are the skills, knowledge, or experience of the pr ogram employees specific to the tasks/services involved in this particular progr am? 1 "Relatively standard" 2 3 "Somewhat specific" 4 5 "Very specific" Relatively standard Â– the skills, knowledge, and experience of the employees used in the program are comparably valued in applications by other industries and other disease mana gement programs. Somewhat specific Â– the skills, knowledge, and experience of the employees used in the program are comparably valued in applications by other disease management programs. Very specific the skills, knowledge, and experience of the employees used in the program woul d not be comparably valued in applications by others, even other disease management programs. Q 9) Using the scale below, please rate the investment made in t he program in terms of capital, facilities, software and equipment for the setup and monit oring of this specific program, which cannot be used for another program. 1 "Low" 2 3 4 5 "Very High" Q 10) Please rate the complexity of tasks and services involved in this program using the following scale. 1 "Fairly Simple" 2 3 4 5 "Very Complex" Q 11) How similar is this program to the other disease management programs offered by the health plan? 1 "Not Similar" 2 3 4 5 "Very Similar"
174 Appendix B (Continued) Q 12) Using the scale below, please rate the frequency of conta ct with the individuals enrolled in the program. 1 "Very Rare" 2 3 4 5 "Very Frequent" Q 13) Using the scale below, please rate the difficulty in measu ring the outcomes, effectiveness and performance of this program. 1 "Easy" 2 3 4 5 "Very Difficult" S ubmit USF Disease Management Outsourcing Survey