USF Libraries
USF Digital Collections

Racial disparities in breast cancer surgical treatment and radiation therapy use

MISSING IMAGE

Material Information

Title:
Racial disparities in breast cancer surgical treatment and radiation therapy use
Physical Description:
Book
Language:
English
Creator:
Koehlmoos, Tracey
Publisher:
University of South Florida
Place of Publication:
Tampa, Fla.
Publication Date:

Subjects

Subjects / Keywords:
Insurance
Health access
Women's health
Lumpectomy
Rurality
Dissertations, Academic -- Health Policy and Management -- Doctoral -- USF   ( lcsh )
Genre:
government publication (state, provincial, terriorial, dependent)   ( marcgt )
bibliography   ( marcgt )
theses   ( marcgt )
non-fiction   ( marcgt )

Notes

Summary:
ABSTRACT: This study explores the relationship between race and surgical treatment and radiation therapy use for localized breast cancer patients in the state of Florida in 2001. The study will be useful in raising awareness of the relationship between Black race and appropriate breast cancer treatment within the Florida Cancer Data System. The Healthy People 2010 initiatives call to eliminate racial disparities and the high placement of breast cancer on the national research agenda make this study timely and insightful for health policymakers, clinicians and other health researchers.Also, the study evaluates the effect of other health system and patient related factors such as insurance provider and rural versus urban residence, to the appropriate use of cancer therapy in order to present an up-to-date and accurate picture of the quality of breast cancer care for women in the state of Florida.The study used multivariate logistic regression modeling and chi-square distribution to compare models in order to disentangle the effects of age, rural residence, marital status and primary health insurance provider from race and to determine how these factors influenced breast conserving surgery versus mastectomy use.Further, the second research question exclusively focused on the population that received breast conserving surgery in order to examine the impact of race and the other covariates as explanatory measures of appropriate receipt of radiation therapy.The first hypothesis found that there was no statistically significant difference between Black and White women in terms of receipt of breast conserving surgery for treatment of localized breast cancer. The second hypothesis, which focused on appropriate receipt of radiation therapy following breast conserving surgery, found that there was a statistically significant interaction between Black race and Medicaid as primary health insurance provider.The study concludes by examining possible areas of improvement in data collection in the State of Florida.
Thesis:
Thesis (Ph.D.)--University of South Florida, 2005.
Bibliography:
Includes bibliographical references.
System Details:
System requirements: World Wide Web browser and PDF reader.
System Details:
Mode of access: World Wide Web.
Statement of Responsibility:
by Tracey Koehlmoos.
General Note:
Title from PDF of title page.
General Note:
Document formatted into pages; contains 98 pages.

Record Information

Source Institution:
University of South Florida Library
Holding Location:
University of South Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
aleph - 001681164
usfldc doi - E14-SFE0001085
usfldc handle - e14.1085
System ID:
SFS0025406:00001


This item is only available as the following downloads:


Full Text

PAGE 1

Racial Disparities in Breast Cancer Surgical Treatment and Radiation Therapy Use by Tracey Lynn Koehlmoos A dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy Department of Health Policy and Management College of Public Health University of South Florida Major Professor: James Studnicki, Sc.D. Stephen L. Luther, Ph.D. Skai W. Schwartz, Ph.D. Ann L. Abbott, Ph.D. Date of Approval: February 22, 2005 Keywords: Insurance, Health Access Women Health, Lumpectomy, Rurality @ Copyright 2005, Tracey Lynn Koehlmoos

PAGE 2

Dedication To my beloved grandmother, Florence Zitowski Kaufman, 1920-1993, who lost the battle with breast cancer Thank you for encouragement and inspiration to last beyond your lifetime.

PAGE 3

Acknowledgements I would like to recognize the many family, friends and colleagues that encouraged and supported me while I completed this en deavor. To my husband, Randy, thank you for creating a marriage with enough room to pursue both of our dreams. To Robert, Michael, and David, thank you for intrinsica lly understanding and general goodness. To my highly esteemed committee members, Dr. Studnicki, thank you for rigor and relevance. I am Eliza to you. Dr. Luther, thank you for holding my hand through every step of the marathon. Dr. Schwartz, thank you for methods, vision, and friendship. To Dr. Abbott, thank you for bringing balance to th e Force. To Monika, Kabir, Norma, Tom The Copy Man Ross, Kari and Peggy; tha nk you for love and support big and small. Lastly, I would be remiss to not acknowledge the tangible input of Mr. and Mrs. J.W. Weber who recognized a need and became kin to me and the boys. You are great Americans and we love you. Thank you!

PAGE 4

i Table of Contents List of Tables................................................................................................................. ....iii List of Figures......................................................................................................................v List of Abbreviations.........................................................................................................v i Abstract....................................................................................................................... ......vii Introduction................................................................................................................... .......1 Statement of the Problem.....................................................................................................2 Defining Disparities.................................................................................................2 Disparity Reduction.................................................................................................4 Breast Conserving Surgery......................................................................................6 Need for the Study...................................................................................................7 Review of the Literature......................................................................................................8 Use of Breast Conserving Surgery...........................................................................8 Use of Radiation Therapy......................................................................................11 Study Contribution to the Literature......................................................................13 Limitations.............................................................................................................14 Hypotheses.............................................................................................................15 Hypothesis One..........................................................................................15 Hypothesis Two.........................................................................................15 Research Design.................................................................................................................16 Data Source............................................................................................................16 Confidentiality.......................................................................................................17 Approach................................................................................................................17 Inclusion and Exclusion Criteria............................................................................18 Gender........................................................................................................18 Stage of Disease.........................................................................................19 Missing Values.......................................................................................................20 Dependent (Outcome) Variables............................................................................20 Surgery Type..............................................................................................20 Radiation Therapy......................................................................................21 Predictor Variable of Interest.................................................................................22 Race............................................................................................................22 Covariates..............................................................................................................22 Age.............................................................................................................22

PAGE 5

ii Payer..........................................................................................................24 Rurality .....................................................................................................25 Marital Status.............................................................................................29 Results........................................................................................................................ ........31 Overview................................................................................................................31 Data Reduction.......................................................................................................31 Hypothesis One, Breast Conserving Surgery Use.................................................36 Descriptive Statistics, Univariate...............................................................36 Descriptive Statistics, Bivariate.................................................................38 Modeling....................................................................................................42 Hypothesis Two, Radi ation Therapy Use..............................................................47 Data Reduction...........................................................................................47 Descriptive Statistics, Univariate...............................................................47 Descriptive Statistics, Bivariate.................................................................50 Modeling....................................................................................................53 Discussion and Conclusion................................................................................................60 Breast Conserving Surgery Hypothesis.................................................................60 Radiation Therapy Hypothesis...............................................................................61 Implications and Recommendations......................................................................63 References..........................................................................................................................66 Bibliography................................................................................................................... ...72 Appendices.........................................................................................................................73 Appendix A: Summary of Logis tic Regression Models for Breast Conserving Sergery...........................................................................................74 Appendix B: Comparison of Models for Breast Conserving Sergery..................76 Appendix C: Results of Interacti on Term Modeling for Hypothesis One............77 Appendix D: Summary of Logistic Regression for Receipt of Radiation Therapy..............................................................................................................79 Appendix E: Comparison of Models for Receipt of Radiation Therapy..............83 Appendix F: Results of Interaction Term Modeling for Breast Conserving Surgery Population............................................................................................84 About the Author...................................................................................................End Page

PAGE 6

iii List of Tables Table 1 SEER Summary Stage...............................................................................19 Table 2 FCDS Radiation Therapy...........................................................................21 Table 3 Age Categories and Descriptions...............................................................23 Table 4 Primary Payer Categories and Descriptions..............................................25 Table 5 USDA Rural-Urban Continuum Codes.....................................................27 Table 6 Florida Counties by USDA Rural-Urban Continuum Codes.....................28 Table 7 Rural-Urban Cate gories and Descriptions Used in Analysis.....................29 Table 8 Marital Status Codes and Descriptions......................................................30 Table 9 Descriptive Statistics for Lo cal Stage Breast Cancer Population..............37 Table 10 Bivariate Descriptive Statis tics, Categorical Covariates by Race.............40 Table 11 Mean, SD, and Range of Age by Race......................................................41 Table 12 Crude Model for Surgical Treatment.........................................................44 Table 13 Full Model for Surgical Treatment............................................................44 Table 14 Full Model with Interact ion Terms for Surgical Treatment.......................45 Table 15 Candidate for Best BCS Explanatory Model with Black, C8....................46 Table 16 Candidate for Best BCS Expl anatory Model, without Black, C9..............46 Table 17 Descriptive Sta tistics of Breast Conser ving Surgery Population...............48 Table 18 Descriptive Statistics, Cate gorical Covariates by Race for BCS...............51 Table 19 Mean, SD, and Range of Age by Race for BCS Population......................52 Table 20 Crude Model for RT...................................................................................55

PAGE 7

iv Table 21 Full Model for RT......................................................................................55 Table 22 Full Model with Interaction Terms for RT................................................56 Table 23 Candidate for Best Explanatory Model, J15..............................................58 Table 24 Candidate for Best Explanatory Model, J16..............................................58 Table 25 Interaction E ffect of Black and Medicaid in Model J16............................59

PAGE 8

v List of Figures Figure 1 Conceptual Framework: Potent ial Barriers to Cancer Treatment................4 Figure 2 Breast Cancer Policy Timeline....................................................................5 Figure 3 Data Reduction Strategy............................................................................34 Figure 4 Histograms of Bivariate Analysis of Breast Conserving Surgery by Race and Category.....................................................................................41 Figure 5 Data Reduction for Ra diation Therapy Analysis.......................................47 Figure 6 Histogram of Bivariate Analysis by Race and Payer for RT.....................57

PAGE 9

vi List of Abbreviations ACoS American College of Surgeons ACS American Cancer Society BCS Breast Conserving Surgery BCT Breast Conserving Therapy CDC Centers for Disease Control and Prevention DHHS Department of Health and Human Services FCDS Florida Cancer Data System IOM Institute of Medicine MQSA Mammography Quality Standards Act NBCCEDP National Breast and Cervical Cancer Early Detection Program NCI National Cancer Institute NIH National Institutes of Health NOS Not Otherwise Specified RT Radiation Therapy SAS Statistical Applications Software SEER Surveillance, Epidemiology, and End Results USDA United States Department of Agriculture

PAGE 10

vii Racial Disparities in Breast Cancer Surgical Treatment and Radiation Therapy Use Tracey Lynn Koehlmoos ABSTRACT This study explores the relationship between race and surgical treatment and radiation therapy use for localized breast can cer patients in the stat e of Florida in 2001. The study will be useful in raising awareness of the relationship between Black race and appropriate breast cancer treatment within th e Florida Cancer Data System. The Healthy People 2010 initiatives call to eliminate r acial disparities and the high placement of breast cancer on the national res earch agenda make this study timely and insightful for health policymakers, clinicians and other health researchers. Also, the study evaluates the effect of other health system and patient re lated factors such as insurance provider and rural versus urban residence, to the appropria te use of cancer therapy in order to present an up-to-date and accurate picture of the qual ity of breast cancer care for women in the state of Florida. The study used multivariate logistic regression modeling and chi-square distribution to compare models in order to dise ntangle the effects of age, rural residence, marital status and primary health insurance provider from race and to determine how these factors influenced breast conserving surg ery versus mastectomy use. Further, the second research question exclusively focuse d on the population that received breast conserving surgery in order to examine the impact of race and the other covariates as explanatory measures of appropria te receipt of ra diation therapy.

PAGE 11

viii The first hypothesis found that there wa s no statistically significant difference between Black and White women in terms of receipt of breast conserving surgery for treatment of localized br east cancer. The second hypothesis, which focused on appropriate receipt of radi ation therapy following breast conserving surgery, found that there was a statistically significant intera ction between Black race and Medicaid as primary health insurance provider. The study concludes by examining possibl e areas of improvement in data collection in the State of Fl orida. Also, the study contains recommendations as to previously unexplored facets of breast cance r research and breast cancer health policy that could be beneficial in the reduction of health and heal thcare disparities in other geographic areas and in other diseases.

PAGE 12

1 Introduction The disparity between Black and White women in terms of incidence and mortality from breast cancer is well documented. Over the la st twenty years the racial disparity in mortality has continued to increase. The di fference between the two groups was first recorded in 1981 when medical advances in breast cancer treatment dramatically increased breast cancer survivability dispr oportionately for White women (Brawley 471, Ries 1998). The federal Healthy People 2010 initiative established the goal of eliminating racial and ethnic disparities in six major areas, incl uding cancer. (U.S. Department of Health and Human Services 2000a). Recent declines in overall breast cancer mortality and increases in screening us e and survivability over the last ten years masks the plight of the medically unders erved and minority populations (Shingawa 2000). This study examined an important component of the Black/White racial disparities in breast cancer treatment ba sed on the appropriate receipt of breast conserving surgery (BCS) with radiation therapy (RT).

PAGE 13

2 Statement of the Problem Defining Disparities In 2003 the Institute of Medicine (IOM) published a comprehensive study which demonstrated that "racial and ethnic minorities received a lower quality of healthcare than non-minorities, even when access-related f actors, such as patients' insurance status and income, are controlled" (Institute of Medi cine 2003). The essential message of this report was that the health of the individual ca nnot be separated from the health of the larger community and from the health of the state and the nation. The term disparities can be defined in a number of ways. The IOM study defined disparities as racial or ethnic differences in healthcare that were not due to access-related factors or clinical needs, preferences, and appropriateness of in tervention (IOM 2003 p.32). The Centers for Disease Control and Prevention (CDC) applied a different definition of how it measures disparities for the purpose of monitoring the Healt hy People 2010 objectives. The principal measure of disparity was a relative measure, the percent difference. The difference was measured from the "best" or most favorable group rate, which was not always the rate for White, Non-Hispanics. The CDC expressed difference measured for indicators in terms of adve rse outcomes not favorable outcomes. For example, the percent of people without health insurance rather than the pe rcent with health insurance (Keppel 20 February 2004). Last, the Department of H ealth and Human Services (DHHS) used a broader definition of disparities: In the absence of consensus on the defin ition of disparities, this report [The National Healthcare Disparities Report] will focus on presenting the facts. Where

PAGE 14

3 we find variation among populations, this va riation will simply be described as a difference. By allowing the data to speak for themselves, there is no implication that these differences resu lt in adverse health outcomes or imply moral error or prejudice in any way (DHHS 2003 p.11-12). Racial disparities in healthcare can be seen on a myriad of levels including ethical, public health, justice, economic and is indicative of fl aws in the over all quality of healthcare. Policymakers and healthcare provide rs alike must recognize that disparities exist and hamper efforts to improve social jus tice and national quality of life. The United States is poised to become populated by close to 50% of what is currently considered minority group members by 2050 (U.S. Bureau of the Census 2000). In 2001, African-Americans had a 33% highe r risk of dying from all types of cancer than non-Hispanic, Whites. Shavers a nd Brown (Figure 1) developed a framework in order to conceptualize the potential barriers to the receipt of optimal cancer treatment. The framework was constructed around structur al, provider and patient related factors that can negatively influence an individuals treatment options for all types of cancer.

PAGE 15

STRUCTURAL BARRIERS PHYSICIAN CLINICAL/FACTORS PATIENT FACTORS Health insurance status Physician Recommendation Socioeconomic status Type of health insurance --Clinical Stage Patient preferences/decision making Type of institution where care is received --Other clinical prognostic indicators Cost/Copayment Geographic region where care is received --Co-morbidity Transportation --Pain assessment Time required for treatment --Physician perceptions/biases Family/Other support Note. Shavers 2002 Figure 1. Conceptual Framework: Potential Barriers to Cancer Treatment Disparity Reduction During the 1990's numerous programs were developed to attempt to reduce overall breast cancer mortality as well as reduce the racial/ethnic and socioeconomic disparities in breast cancer mortality. Governmental policy changes and additional funding for programs like the Breast and Cervical Cancer Mortality Prevention Act of 1990 (Public Law 101-354) covered screening but not treatment of breast cancer through the Centers for Disease Control and Prevention's (CDC) National Breast and Cervical Cancer Early Detection Program (NBCCEDP). In order to bridge the screening/treatment gap, Congress approved the Breast and Cervical Cancer Treatment and Prevention Act of 2000, which appropriated $900 million over 10 years to provide treatment for women diagnosed with cancer via the NBCCEDP. As a social equalizer, NBCCEDP has the ability to qualify indigent and otherwise uninsured women for breast cancer treatment via state administered Medicaid programs. As of 2002 was available in every state and five 4

PAGE 16

tribal areas except Oklahoma (CDC 2003). Another program, the Mammography Quality Standards Act (MQSA) of 1992 sought to regulate the quality of mammography equipment, but failed to assure the accuracy of interpretation or the effectiveness of post-screening follow-up. (Bickell 2002). However, these programs do nothing to guarantee or measure the effectiveness or appropriateness of breast cancer treatment. Figure 2 presents a timeline of major health policy initiatives that sought to increase access to and monitor quality of care in breast cancer screening and treatment. 1990 1992 1994 1996 1998 2000 2002 2004 2006 N BCCEDP active in 49 states, 8 territories, DC and all Native American tribal areas Mammography Quality Standards Act ( M Q SA ) Breast and Cervical Cancer Treatment and Prevention Act (NBCCEDP) Breast and Cervical Cancer Mortality Prevention Act Figure 2. Breast Cancer Policy Timeline In the new millennium breast cancer remains the second leading cause of cancer death for Black women in the United States, and the mortality rate for Black women from breast cancer is higher than it is for White women despite the lower incidence rate. Because of the later stage at which breast cancer is typically diagnosed in Black women, 5

PAGE 17

6 the five-year survival rate for Black women remains significantly lower than for White women (75% vs. 88%) (Ameri can Cancer Society 2003). Breast Conserving Surgery Numerous prospective, randomized clinical trials in the Unit ed States and in Europe during the 1980's demonstrated that surv ival after BCS was equal to survival after mastectomy (Fisher 1989, Blichart-Toft 1988, Veronesi 1981, Sarrazin 1989, Licter 1992, Van Dongen 1992). This strong evidence led a 1990 National Institutes of Health consensus panel to determine that breast cons ervation treatment is an appropriate method of primary therapy for the majority of women with localized breast cancer (NIH 1992). Breast Conserving Treatment (BCT) is defined as the excision of the primary tumor and adjacent breast tissue followed by radiation therapy. Breast Conserving Treatment is not an option for every case of breast carcinoma. However, the National Cancer Data Base issued a report in 1994 that concluded that up to 75% of diagnosed cases of breast cancer can be considered localized, thus eligible for BCT (Osteen 1994). Based on the guidelines released in 1992 by an interdisciplinary group whic h consisted of represented the American College of Surgeons (ACoS) American College of Radiology (ACR), the College of American Pathologists (CAP), and the Society of Surgical Oncology, there are only four absolute exclusion factors agains t the use of Breast Conserving Therapy. The four contraindications are listed below: Patient in the first or second trimester pregnancy History of therapeutic irra diation to the breast region Multiple primary tumors in se parate quadrants of the breast Extensive indeterminate or malignant-appearing calcifications throughout the breast (Winchester 1992).

PAGE 18

7 Need for the Study Considering that Black women have a si gnificantly worse prognosis from breast carcinoma compared to White women, even when stage at diagnosis is equivalent, it is worthwhile then to analyze racial differences in the treatment of women with early stage breast cancer as a measure of quality of care (Joslyn 2002, Ro etzheim 2000). That there was no recorded difference in breast cancer survivability before 1981 led to the hypothesis in many research articl es that advances in breast cancer treatment were only benefiting one segment of the population (the primarily White, upper and middle classes) leaving less adequate care for poor and mi nority women (Diehr 1989, Boyer-Chammard 1999, Breen 1999). Health care for medica lly underserved populations remains a high priority for the U.S. government. Particular in terest has been paid to cancer care because it is both common and expensive, and require s a complex interaction between patients, providers and the health care system (Hewitt 1999). Breast cancer is firmly entrenched on the national cancer research agenda because of its enormous impact on society (Jones 2002, NCI 2001). This study describes the impact of race on appropriate receipt of su rgical treatment and radiation therapy for localized breast cancer. Results of this study provide policy makers, health researchers and health agencies with an assessment of the degree to which Florida has progressed toward adoption of breast conserving surgery and radiation therapy for the treatment of Early Stage Breast Cancer. Also, the study demons trates the effect of other health system and patient related factors such as insurance provider and rural versus urban residence, to the appropriate use of cancer therapy in or der to present an up-to-date and accurate picture of the quality of breast cancer care for women in the state of Florida.

PAGE 19

8 Review of the Literature In the 13 years since the NIH report, ther e continues to be an underused of BCT as a less radical alternative to mastectomy for treatment of local stage breast cancer despite increases in the detection. The aver age BCS rate for the United States remains very low with only 40-50% of eligible pati ents undergoing the less invasive procedure (Morrow 2001, Lazovich 1999). Mastectomy rather than BCS remains the primary surgical approach to treating local breast cancer in many geogra phic regions. Patient advocacy groups, researchers, clinicians and policymakers consistently argue that BCS plus radiation is superior for most women b ecause it is less invasive, preserves the breast, and post-surgical psychological adjustment is less difficult (Lantz 2002). Considering that Black women have a significantly worse prognosis from breast carcinoma compared to White women, even when stage at diagnosis is equivalent, it is worthwhile then to analyze racial differences in the treatment of women with early stag e breast cancer as a measure of quality of care (Joslyn 2002, Roetzheim 2000). Use of Breast Conserving Surgery Overall studies of breast cancer surgery using race/ethnicity as a predictive variable have produced conflic ting results. Several studies from the 1990's reported that breast cancer treatment differed between African-American women and White women but there is wide variation among researchers as to the perceived cause of the difference. National data from the early 1990's show that Black women were less likely to undergo BCS than white women (Muss 1992). Natti nger and associates and Johantgen and

PAGE 20

9 associates found that surgical treatment varied even after adjusting for breast cancer stage at diagnosis (Nattinger 1992, Johantgen 1995). More recently articles by Newman and associates and Velanovich and associates demonstrate that Black women continue to be less likely than White women to undergo BCS. However, the latter two studies focused on a limited population within the same urban area (New man 1999, Velanovich 1999). Also, the Carolina Breast Can cer Study from 1995 through 2000 showed that fewer Black women received BCS than White women (Dunmore 2000). On the other side of the spectrum, in 2002 Joslyn released a study that analyzed SEER program data from 1988-1998 which found that Black women were slightly more likely the receive BCS than White women, but the differences were not clinically significant; however, greater th an one-third of the Black women were premenopausal compared with fewer than one-fourth of White women (Joslyn 2002). Other studies showed that after adjusting for sociodem ographic information that race was not a significant indicator of type of surgery used to treat breast cancer. Using data from the Surveillance, Epidemiology, and End Results (SEER) national tumor registry program, Gilligan and associates show that from 1983-1996 the use of BCS increased consistently across all racial groups and that when resear chers adjust for socioeconomic covariates such as county level of education and count y income, race proved not to be a significant indicator of the use of BCS (Gillligan 2002). In addition, a study of California, New Mexico and Detroit, Michigan breast cancer cases from 1992-1995, Richardson and associates found that once they had adjusted for stage of disease at diagnosis, there was no significant difference in surgical treatment (Richardson 2001). Bradley and asso ciates in a different study of the Detroit,

PAGE 21

10 Michigan area that linked SEER program and Census data found that when controlling for age, race, marital status, cancer stage, Medicaid status, and census tract poverty level, Black women were statistically significantly as likely as White women to receive BCS. However, within the Medicaid population, Bl ack women were more likely to receive BCS than White women, were when not adju sting for health insurance, Black women were overall less likely to receive any form of breast cancer surg ery than White women (Bradley 2002). Bradley cautions that although poverty presents itse lf as the strongest indicator of poor cancer stage of diagnoses, treatment and d eath, the fact remains that Black women are more likely to have lower incomes and live at or below the poverty level than White women. Using data from 1997 and 1998, Luther and Studnicki related physician volume to the use of mastectomy versus BCS. In contrast to the aforementioned Bradley study, Luther found that Medicaid insured women were significantly less likely to receive BCS. Although BCS had become the most common form of breast cancer su rgical treatment, they found that in Florida non-whites were more likely to be treated by low volume surgeons who were more likely to perform mastectomies; whereas White women and privately insured women were more likely to receive BCS (Luther 2001). However, this study grouped all non-white case s into one category which can bias study results toward the null in that non-white Hispanic, Native Am erican, and Asian raci al and ethnic groups tend to vary widely from Black women in their cancer outcomes and sociodemographic factors (Bradley 2002). In another Florida-based breast cance r study using data from 1994, Roetzheim and associates found there was no racial diffe rences in the use of BCS although insurance

PAGE 22

11 type varied greatly by ra ce/ethnicity. Black women constituted 6.9% of the study population yet had a higher risk of being covered by Medicaid ( 6.6% for Black v. 1.8% for White women) or to be uninsured ( 10.0% for Non-White women v. 3.6% for White women) (Roetzheim 2000). Use of Radiation Therapy Similar to the divergent findings for the use of breast conserving surgery, research conducted on the use of radiation therapy ha s produced equivocal results. Radiation therapy (RT) is the factor that makes Br east Conserving Therapy a definitive primary treatment for women undergoing breast cons erving surgery. In addition to race, advanced age (75+ or 80+ years of age, de pending on the study), becomes a variable of interest in the receipt of appropriate post su rgical radio therapy. The incomplete use of breast conserving therapy (BCS + RT) in any one group based on age, race or residence is in conflict with the NIH Consensus Statement and raises quality of care issues (Gilligan 2002). In a 1992 study of geographic variation in th e treatment of localized breast cancer that used SEER program data, Farrow and associates found that older women across the United States and Black women in the Atlant a and Detroit areas were significantly less likely to receive RT following BCS. The radi ation use decreased initially after age 65 and more steeply after age 75 (Farrow 1992). Similarly, Ballard-Barbash found that frequency of radiation thera py did not vary significantly by racial group; however it did increase in areas with higher educational leve ls and decrease significantly with each fiveyear age group (Ballard-Barbash 1996). In addition studies by Roetzheim and associates and Marrow and associates see no racial dispar ity for radiation therapy based on race, but

PAGE 23

12 rather by age (Marrow 2001, Roetzheim 2000). Using NBCCEDP linked data from three Western states, Richardson and associat es also found no association between race/ethnicity and the receipt of RT (Richardson 2001). On the other hand, Bradley and associat es found that Black women were less likely to receive radiation following BCS than white women (OR .74, 95% CI: .61-.88). However, when logistic regression models controlled for poverty via residence in a census tracts with poverty levels of 13% or greater, age, race, marital status, cancer stage and Medicaid status, Bradley found that Bl ack women did not have a statistically significant difference in the use of radiati on therapy after breast conserving surgery. (Bradly 2002). In a ten-year, national study using SEER data, Joslyn found that 29.9% of Black women versus 24.2% of White wome n in the study did not receive radiation therapy post-BCS. Black women were significantly less likely to receive radiation at each age category except the group of 85years of age or older (Joslyn 2002). Additionally, there are access is sues created by distance to radiation facilities that may serve as barriers to appropriate use of radiation therapy for pos t-breast conserving surgery cases. The availability of transpor tation and the relative convenience of such facilities may also account for slower rates of adoption of breast conserving therapy versus modified radical mastectomy for early stage breast cancer (Nattinger 2001). Nattingers study found that women living more than 15 miles from a hospital with a radiotherapy center had a significantly re duced likelihood of unde rgoing BCT (OR .52, CI 0.46-0.58). Parviz and associ ates used geocoding software to examine the effect of distance from home to therapy site on initia l surgical treatment and found that there was no statistically significant difference in BCT rates when looking at distance as a

PAGE 24

13 continuous variable. However, statistic al significance existed when cases were categorized dichotomously based on greater th an or less than 40 miles from the radiation therapy center. Weaknesses of the Parviz st udy include that it was conducted on a small sample size treated by the same team of surg eons at the same teaching facility (Parviz 2003). In a U.S. wide study conducted in 2002, Gilligan found that there was greater use of breast conserving surgery and radiation ther apy in urban areas compared to rural areas (Gilligan 2002). Study Contribution to the Literature The state of Florida presents itself as th e ideal setting in which to undertake this study. Florida has the highest crude incide nce rate of cancer in the nation with a 14,000,000 population residing in 67 counties. Two hundred forty-seven hospitals report over 120,000 cases annually, which when undup licated, translate into approximately 80,000 newly diagnosed cases per year (F CDS 2003). With both a large elderly population and a large minority representation, Florida is a bell-weather state for the graying population of the entire United States. In Florida in 2001, there were 18,403 incident cases of breast can cer out of 103,587 analytic, incident all-type cancer cases (FCDS Monograph 2003). This study will contribute to the vast body of breast cancer research by providing an up-to-date picture of the breast cancer tr eatment differences or lack of differences between Black and White women in Florida for both surgical options and the appropriate use of radiation therapy postbreast conserving surgery. Also, this study will differ from other studies by adjusting for and evaluating th e influence of rural versus urban residence

PAGE 25

14 in the use of surgical option and radia tion therapy by employing the new continuum developed by the United States Department of Agriculture (USDA 2003). Limitations Limitations are factors outside of the re searcher's control that can affect the outcome of the study. Several limitations exist within this study design. The choice to conduct secondary data analysis is beneficial in many ways (economic cost, scope of sample size) but can hinder a researcher in te rms of limiting the ability to select variables to be included in multivariate modeling. There is a necessary gap between the type of data the researcher would lik e to collect for inclusion in the study and the information that is available in the pre-existing dataset (Iezzoni 2003, Nachmias & Nachmias 1998). Additionally, the use of the Florida Ca ncer Data System tumor registry information inherently imposes several lim itations. The registry only includes cancer cases treated within Florida and may limit the generalizability of findi ngs to other states or regions of the country. When using a c oded data source the issue of reliability revolves around how accurately and completely the coded information from the medical records was abstracted. Last, the validity of using cancer registry data for evaluating quality of cancer care has not been well studied. Malin and a ssociates compared medical records, the "gold standard," to California Cancer Registry data and found that th e validity of registry data varied across the settings of care. As cancer care moved from the in-patient setting the capture of data decreased as follows: surgery, radiation therapy, chemotherapy and hormonal therapy. Thus, although registry data may be valid for use in studying surgery and radiation therapy, it is not equally valid for studying the use of chemotherapy and the

PAGE 26

15 prescription of adjuvant therapy (like Tamoxife n) which are more likely to take place in an ambulatory setting or phys ician's office (Malin 2002). Hypotheses Hypothesis One. Black women in Florida are statistically significantly less likely than White women to receive breast conser ving surgery rather than mastectomy for treatment of locali zed breast cancer. Hypothesis Two. Black women in Florida are statistically significantly less likely than White women to receive radiation therapy after breast conserving surgery.

PAGE 27

16 Research Design Data Source This study proposes to answer the ab ove hypotheses by conducting secondary data analysis using the Florida Cancer Da ta System information from the year 2001. FCDS is Florida's Statewide Population-Ba sed Cancer Registry. The result of a collaboration between the Florida Depa rtment of Health and the Sylvester Comprehensive Cancer Center (SCCC) at th e University of Miami School of Medicine designed and implemented the registry in 1978. FCDS has been coll ecting incidence data since 1981. In October 1994, the Florida Cancer Data System became part of the National Program of Cancer Registries (NPCR) admi nistered by the Centers for Disease Control (CDC). Through this program the CDC provides funding for states, such as Florida, to enhance their existing registry to meet national standa rds for completeness, timeliness and data quality set forth by the North American Association of Central Registries (NAACCR), the American College of Surg eons, Commission on Cancer (ACoS/CoC) and the Surveillance, Epidemiology and End Results (SEER) reporting program of the National Cancer Institute (NCI). The legal statutes that govern the creation of the database and the participation of facilities providing cancer care are numerous. Florida Statute 385.202 provides for the establishment of a statew ide cancer registry. All facilities licensed under Florida Statute 395 and each freestanding radiation therapy center as defined in Florida Statute 408.07

PAGE 28

17 shall report to the Department of Hea lth, through FCDS, such cancer incidence information as specified by Rule 64D-3.006 which includes, but is not limited to, diagnosis, stage of disease, medi cal history, laboratory data, ti ssue diagnosis, radiation, or surgical treatment and either method of dia gnosis or treatment for each cancer diagnosed or treated by the facility or center. Confidentiality FCDS continues to adhere to all Flor ida Statues and Department of Health guidelines regarding patient and instituti onal confidentiality. No unique, patient identifying information will be used in the conduct of this study. Approach First univariate analysis describing the characteri stics of the population was conducted. This included unadjusted mean values of continuous variables and frequency analysis of categorical variable s. The results of the univariat e analysis are given in tables in the Results chapter. Next, bivariat e analysis was conducted to describe the relationship between the predicto r variable of interest, Race, and each of the covariates. Next, two separate, multivariate logistic regression models were developed. Logistic regression is typically used to model dichotomous dependent variables and performs well when compared to more complex modeling approaches (Feinstein, Wells and Walter 1990, Selker 1995, Iezzoni 2003). The first model dealt with the outcome/dependent variable of type of surgery and race, and the second mode l evaluated the recommended use of radiation therapy and race. A descri ption of the variables used in the study is provided below.

PAGE 29

18 In developing the models an iterative process of determining which of the covariates are useful to the explanation of the dependent variable was followed. Model fit statistics were used and interaction terms were evaluated for inclusion. All models were developed first to determine unadjusted odds ratios and, next, refined to determine adjusted odds ratios. Also, to remain consis tent with the literature, the conventional, nominal levels for alpha, p-value less th an or equal to 0.05, a nd ninety-five percent confidence intervals were developed in order to determine statistica l significance of study results. These results are presented in the ne xt chapter using a variet y of tables. Finally, all multivariate logistic regression modeli ng was performed using SAS version 8.00 for Windows (SAS Institute, Cary, NC). Inclusion and Exclusion Criteria Initially localized breast carcinoma case s occurring in women reported to the Florida Cancer Data System during 2001 will be considered for inclusion in this study. Additional inclusion and exclusion criteria are detailed below. Gender. Men and women differ chromosomally, anatomically, physiologically, hormonally and reproductively. Socioeconomic circumstances influence the lives of the two genders differently (Iezzoni 2003). Breas t cancer is 100 times more likely to occur in women than in men, yet each year in the United States about 400 men die of breast cancer (0.22 percent of all cancer deaths). B ecause of the small size of men's breasts and location of the primary tumor is most frequent ly in the center of the breast, modified radical and radical mastectomy are the most common surgical options for men. Because of the differences in surgical treatment (outcome variable) and because male breast cancer is a rare disease, men are exclude d from the study (Elk 2003). Additionally,

PAGE 30

19 subjects coded as "Intersexed," "Transgende red" or "Unknown" are also excluded from the sample. Thus, only subjects coded as "Femal e" will be included in the data set to be analyzed for the study. Stage of disease. The disease to be studied exclus ively in this study is breast cancer. The initial data set to be examined will be all breast cancer cases within the FCDS for 2001 (coded as C50.0-C50.9). Iezzon i writes that research targeting one particular disease will not need to adjust for the presence of the disease but rather for the disease specific seve rity (Iezzoni 2003). Recent long-term randomized clinical trials have reaffirmed that breast conserving surgery is appropria te for women with local st age breast cancer (Fisher 1995, Veronesi 1995). Rather than recording Am erican Joint Commission on Cancer staging information, the FCDS reports SEER Summary Stage information listed in Table 1. Table 1 SEER Summary Stage Code Description 0 in situ 1 Local 2 Regional/Direct Extension 3 Regional/Nodes Only 4 Regional/Direct Extension and Nodes SEER Summary Stage is based on a comb ination of pathologic, operative and clinical assessments. The Summary Stage in the FCDS data set is based on all information available through completion of surg ery in the first course of treatment or

PAGE 31

20 within four months of dia gnosis in the absence of dis ease progression, whichever is longer (FCDS-DAM 2003). For the purpose of answering Hypotheses #1 and #2, only women with Local stage breast cancer, which incl udes lymph node-negative findings will be included in the analysis (Farrow 1992, Roetzheim 2000, Richardson 2004). Missing Values Cases with missing variables will not be included in analysis because such missing variables may produce biased results. This common approach for dealing with subject records in a data file that are incomplete is called complete-subject analysis. However, there is the potential to lose st atistical power should a significant number of any variable be missing from the data set or be more likely to be missing from one category. Thus, there may be the need to re -evaluate the inclusion of cases with missing information (Rothman 1998). If there are differences in the pattern of missing variables, Tabachnick and Fidell recommend five methods of dealing with the variables: 1) delete the cases or variable; 2) estimate the missing data using a mean value; 3) use a missing data correlation matrix; 4) treat missing data as data; and 5) repeat analyses with and without missing data (Tabachnick 1989). Dependent (Outcome) Variables Surgery type. Surgery type will be categorized as a dichotomous variable. In the comparison of Mastectomy versus Breast Conserving Therapy, BCS will be defined as lumpectomy, segmental mastectomy, quadrantectomy, tylectomy, wedge resection, nipple resection, excisional bi opsy, and partial mastectomy no t otherwise specified. BCS may or may not have included axillary lym ph node dissection and/or follow-up radiation.

PAGE 32

Mastectomy will be considered as simple and subcutaneous mastectomies, radical mastectomies and modified radical mastectomies. Surgical procedures within the data set are coded using SEER program and COC/ACoS surgery identifiers (FCDS-Data Acquisition Manual Appendix H 2003). The surgical codes are divided into two possible groups (0/1) as previously described. Although the FCDS includes categories for "No Surgery," "Surgery, Not Otherwise Specified," and "Unknown," cases with those entries in the Primary Surgery Summary field of the database will be excluded from this study. Radiation Therapy. Radiation Therapy will be collapsed into a dichotomous variable. The FCDS defines Radiation Therapy as listed below in Table 2. Table 2 FCDS Radiation Therapy Code Description 0 No radiation 1 Beam radiation 2 Radioactive implant 3 Radioisotopes 4 Combinations of 1 with 2 and/or 3 5 Radiation, NOS (Not Otherwise Specified) 6 Patient or patients guardian refused 7 Radiation therapy recommended, unknown if administered 8 Unknown if radiation therapy administered For the purpose of this study radiation will be broken down into two categories: No radiation (consisting of code 0) and any type of radiation received or planned (Codes 21

PAGE 33

22 1, 2, 3, 4, 5, and 8). Subjects whose radiati on treatment status is unknown (codes 7 and 9) will not be included in the analysis (Bradley 2002). Predictor Variable of Interest Race. Race will be defined as a dichotomous, categorical variable with the case's race being defined as White or Black. Other racial/ethnic groups such as Asian, Native American and Hispanic are excluded from this study because they vary widely in their cancer outcomes, and combining such disparat e groups into the study can lead to biased results (Bradley 2001). A broad-based populatio n outcomes study by Studnicki, et al., set in one county in Florida concluded that th ere are research constraints in the limited number of ethnically identified indicators a nd, especially for Hispanics, problems in the accuracy and consistency of the assignment to racial categories and subsequent reporting (Studnicki 2004). Also, subjec ts who are self-identified as "Multiracial" in the FCDS data set will be excluded from the study. Covariates Age. Iezzonie writes that age is "simple straight forward[has] good face validityis almost always available, and [we] expect age to be in the model" (2003). It is standard in all risk adjustment modeling. Age is of particular in terest in the study of breast cancer because a woman's risk of de veloping this disease increases during her lifetime. Some 80% of breast cancers o ccur in women older than 50. By age 70, a woman's chances of developing breast cance r are 1 in 24 (Elk 2003). In the study of breast cancer disparities age plays a signifi cant role because black women younger than 40 have slightly higher rates of breast cancer diagnosis th an white women in the same age group (Elk 2003).

PAGE 34

23 Age by itself is a continuous variable and in the descriptive anal ysis of the data will at first be evaluated as such using univariate analysis; however, for the purpose of responding to the hypotheses, Age will become a categorical variable, stratified into four age groups depending on age at time of diagnos is. Subjects with missing or incomplete age will be categorized as Unknown and excluded from the study. Because of findings by Farrow and Ballard-Barbash that stress age as an indicator for declining use of radio therapy post-breast conserving surgery, age as a variable will be evaluated categorically to best present the difference across st rata (Farrow 1992, Ballard-Barbash 1996). The subjects will be categorized into the following groups as listed in Table 3. Table 3 Age Categories and Descriptions Code Description 1 Young (<50) (Reference Group) 2 Middle Age (50-64) 3 Old (65-79) 4 Very Old (80+) Although other studies have divided age groups by ten-year intervals or created an open-ended 65+ age group, this study will use th e aforementioned four categories for age. Using ten-year age groups pr oduces a large number of categories so th at the observations are stretched over too many categories and patt erns may not be easily discerned in the resulting cross tabulation. For the later, 65+ age group (used in studies that relate pre-, during and post-menopausal status to breast cancer), Rothman says that a category with

PAGE 35

24 "no upper limit allows for a considerable range of variability within which the desired homogeneity of exposure or risk may not be achieved" (206). Thus the creation of the 80+ group will bring to light the experience of the increasing number of "oldest old." One study shows that although elderly cancer patients often receive less aggressive treatment, they spend more time discussing limitations of treatments with their physicians (Iezzoni 2003, Rose 2000). Payer. Primary payer of health care for breas t cancer treatment is considered an imported independent variable by many resear chers. In the case of BCS versus Mastectomy, previous research has contraindications of what are the effects of Medicaid insurance. Studies by Bradley and Morrow show Medicaid as a great equalizer of surgical option used, where as Luther found that women in Florida were twice as likely to receive mastectomies if Medicaid was the primary insurance payer. (Bradley 2002, Morrow 2001, Luther 2001). Primary payer is a categorical variable. The FCDS database includes a multitude of categories which will be divided into five categories as shown in Table 4.

PAGE 36

25 Table 4 Primary Payer Categories and Descriptions Code Description 1 Private insurance and Mana ged Care: (Reference group) Managed care provider, NOS Health Maintenance Organization (HMO) Preferred Provider Organization (PPO) Insured, Type Unknown Unknown insurance Champus Military 2 Medicare (Federally funded insurance types) Medicare Veterans Administration Indian Health Services Public Health Services 3 Medicaid State funded, NOS Medicaid Welfare 4 Uninsured: Not Insured, NOS Not insured, charity write-off Not insured, self-pay 5 Unknown Champus and Military insurance were cat egorized in the Private insurance group because previous studies have found that women in these groups share numerous demographic characteristics as privately in sured women. These ch aracteristics include age, race, and stage at diagnosis (Richardson 2004). Rurality. The FCDS includes the case's count y of residence and zip code of residence at the time of diagnosis. Luther a nd Studnicki found that residing in a rural rather than an urban area was significantly associated with the use of mastectomy over BCS (Luther 2001). As shown in the revi ew of the literature section, studies by

PAGE 37

26 Nattinger and Parviz have evaluated the effect of rurality in terms of distance to a treatment center for radia tion therapy (Nattinger 2001, Pa rviz 2003). The rurality measure to be employed in this project is ad apted from the United States Department of Agriculture's (USDA) Rural-Urban Conti nuum Codes which were produced by the Economic Research Service. The application of a continuum rather than a dichotomous break down of Urban versus Rural will enable isomorphic comparisons across categories. Rural-Urban Continuum Codes form a clas sification scheme that distinguishes metropolitan (metro) counties by the populati on size of their metro area, and Non Metropolitan (Non Metro) count ies by degree of urbanization and adjacency to a metro area or areas. The metro and Non Metro categor ies have been subdivided into three metro and six Non Metro groupings, resulting in a nine-part county codi fication. The codes allow researchers working with county data to break such data into finer residential groups beyond a simple metro-Non Metro dich otomy, particularly for the analysis of trends in Non Metro areas that may be relate d to degree of rurality and metro proximity. All U.S. counties and county equivalents are grouped according to their official metro-Non Metro status announced by the Office of Management and Budget (OMB) in June 2003, when the population and worker co mmuting criteria used to identify metro counties were applied to results of the 2000 Census. Metro counties are distinguished by population size of the Metropolit an Statistical Area of whic h they are part. Non Metro counties are classified accord ing to the aggregate size of their urban population. Within the three urban size categories, Non Metro c ounties are further id entified by whether or not they have some functional adjacency to a metro area or areas. A Non Metro county is defined as adjacent if it physically adjoins one or more metro areas, and has at least 2

PAGE 38

27 percent of its employed labor force commuting to central metro counties. Non Metro counties that do not meet these criteria are classed as nonadjacent (USDA 2003). All nine of the categories of rural-urban do not apply to the state of Florida. Some adaptation of the Rural-Urban Continuum code s is necessary in order to maintain the statistical power of this study. One area that wa s changed was a collapsing of the last two categories that apply to Florida because of small sample size within the last stratum (Levels 6 and 8). The original US DA codes are listed in Table 5. Table 5 USDA Rural-Urban Continuum Codes Code Description Metro counties: 1 Counties in metro areas of 1 million population or more. 2 Counties in metro areas of 250,000 to 1 million population. 3 Counties in metro areas of fewer than 250,000 population. Non Metro counties : 4 Urban population of 20,000 or more adjacent to a metro area. 5 Urban population of 20,000 or more not adjacent to a metro area. 6 Urban population of 2,500 to 19,999, adjacent to a metro area. 7 Urban population of 2,500 to 19,999, not adjacent to a metro area. 8 Completely rural or less than 2,500 urban population, adjacent to a metro area. 9 Completely rural or less than 2,500 ur ban population, not adjacent to a metro area. (USDA 2003) It is noteworthy that none of the "Not adjacent to metro area" codes apply to Florida, which may be an artifact of the state's geographic lay out as a long narrow

PAGE 39

28 peninsula and panhandle. The USDA codes as they apply to Florida's sixty-seven counties are listed in Table 6. Table 6 Florida Counties by USDA Rural-Urban Continuum Codes Code Counties 1 Baker, Broward, Clay, Duval, Hernando, Hillsborough, Lake, MiamiDade, Nassau, Orange, Osceola, Palm Beach, Pasco, Pinellas, St. Johns, Seminole (Reference Group) 2 Brevard, Collier, Escambia, Gadsden, Jefferson, Lee, Leon, Manatee, Marion, Martin, Polk, St. Lucie, Sant a Rosa, Sarasota, Volusia, Wakulla 3 Alachua, Bay, Charlotte, Gilchr ist, Indian River, Okaloosa 4 Citrus, Flagler, Hendry, Highlands, Monroe, Okeechobe, Putman, Sumter 6 Bradford, Calhoun, Columbia, DeSot o, Dixie, Franklin, Glades, Gulf, Hamilton, Hardee, Holmes, Jackson, Madison, Suwannee, Taylor, Union, Walton, Washington 8 Lafayette, Liberty 88 Out of State, Out of Country 90 Unknown In order to create a more parsimonious m easure of rurality and to increase the statistical power of the cate gories during analysis, it become s necessary to further group the counties in Florida in to the following categories described in Table 7.

PAGE 40

29 Table 7 Urban-Rural Categories and Descriptions Used in Analysis Category Description Metro Includes USDA Code 1. Counties in metro areas of 1 million population or more. Small Metro Includes USDA Codes 2 and 3. Counties in metro areas of fewer than 1 million population. Non Metro Includes USDA Codes 4,6 and 8. Ur ban population of 20,000 or less adjacent to a metro area or complete ly rural adjacent to a metro area. Unknown FCDS Codes 88 and 90. Out of Stat e, Out of Country and Unknown. Cases that fall into the Unknown cate gory are excluded from further study because they fail to contribute to the overall pi cture of disparities in breast cancer surgical treatment and radiation therapy us e in the state of Florida. Marital status. Although much has been written on the impact of marital status on the increased use of breast cancer screen ing and stage of diagnosis (Miller 2002, Mandelblatt 1991, Suarez 1994, Roetzheim 1999), little has been written on the effects of marital status and the appropriate use of br east conserving surgery a nd radiation therapy. Marital status is used to represent a degree of social support available to the case. This project will evaluate the impact of marital st atus on the dependent variables. Information in the FCDS provides for four possibl e categories as described in Table 8.

PAGE 41

30 Table 8 Marital Status Codes and Descriptions Code Description 1 Married (Reference group) 2 Never Married (Single) 3 Other (Divorced, Separated, Widowed) 4 Unknown Nachmias and Nachmias point out the po ssible lack of exhaustiveness in this enumeration of categories (Frankfurt-Nach mias 1996). Respondents who are "living together" with an opposite sex or same sex partner or an elderly parent who lives with an adult child do not fit accurately into this scheme, yet might exhibit the same level of social support as subjects in the "Married" category.

PAGE 42

31 Results Overview This chapter presents the results of the data analysis conducted for the study. The chapter begins with a detailed explanation of the reduction of the initial data set including all records from FCDS for 2001 to the final data set used in the anal ysis. As previously outlined, the next set of information enco mpasses the entire eligible population for Hypothesis One, which looks for a Black/White racial disparity in the use of Breast Conserving Surgery. First, the univariate anal ysis presents descriptive statistics for the entire population. Next, the bivariate analysis shows the relationship between the predictor variable of interest Race, when evaluated according to each of the covariates. Next information is presented about the iterative processes for determining which of the covariates are most useful in the explanation of the depende nt variable, BCS use, and the unadjusted and adjusted odds ratios (ORs) are given as well as the 95% Confidence Intervals. After the results from Hypothesis One are fully outlined, then the process is repeated for Hypothesis Two, which focuses on the smaller population of BCS recipients and the receipt of Radiation Therapy. Data Reduction Based upon the criteria previously detailed in the Methods chapte r, a considerable amount of data reduction was necessary in orde r to produce the final data set for analysis. The process is illustrated in Figure 2.

PAGE 43

32 Each record in the FCDS dataset repr esents one tumor. The data reduction process began by identifying tumors exclusiv ely identified as bei ng breast cancer (C50.0C50.9), which consisted of 20,969 tumors (Box A). Next, 172 tumor records were removed because of a gender identity othe r than female, leaving 20,797 (Box B). There were no tumors occurring in the same person th at were coded as female in one record and some other gender in another record. With respect to ethnicity, there were 2,393 tumors identified as belonging to Hispanics, and they were removed (Box C) leaving 18,404. Next, there were 283 tumors identified as belonging to individuals with a race other than Black or White, and they were removed, leaving 18,121 (Box D). Seven (7) tumors were removed for discrepant race entries (i.e., tumors on the same person reported the person as Black in one record, and White in another), leaving 18,114 (Box E). To limit tumors to those with local state breast cancer, 9,238 were removed, leaving 8,876 (Box F). Those coded as not fa lling into either the BCS or Mastectomy group were removed (n=217), leaving 8,659 (Box G). In the case of multiple tumor surgeries for the same individual coded for BCS and Mastectomy, the BCS records were removed (n=117), as Mastectomy would be th e most definitive surgical option. This exclusion left 8,542 tumors (Box H). With respect to Radiation Therapy (RT), when there were multiple tumors on the same patient, only the most recent record w ould theoretically report RT because as the individual moved through th e continuum of cancer care, Radiation Therapy would be among the last steps registered in the FCDS. In this case, tumors reporting treatment with RT were kept, and the others removed (n=510), leaving 8,032 (Box J).

PAGE 44

33 In the case multiple tumor records for th e same individual reporting a different age for the individual, the row with the earliest age of diagnosis was retained (n = 8,017, Box K), and the rest removed (n=15). To conduct the planned analysis, it was necessary to reduce the data set to one tumor per person. To accomplish this, tumors on individuals who still had more than one tumor represented in the data set were eval uated for missing data. Those with the most missing data were removed (n=53), and 7,964 re cords were retained (Box L). As the planned analysis involved an evaluation of the impact of insura nce type, those with multiple tumors reporting inconsistent insu rance were removed (n=81), leaving 7,883 (Box M). Finally, for those individuals still representing multiple tumors in the data set, tumors were randomly removed (n=487) so that each individual c ontributed 1 tumor to the data set (n=7,396, Box N). Because the analysis is limited to indivi duals in Florida, individuals coded as outside the state were removed (n=217), leaving a final data set of 7,179 individuals representing one tumor each (Box P).

PAGE 45

Figure 3. Data Reduction Strategy. Male = 150 Other (Intersexed) = 1 Transsexual = 2 Unknown/not stated = 19 Mexican = 43 Puerto Rican = 89 Cuban = 396 South or Central American = 221 Other Spanish/Hispanic Origin = 33 N OS, Spanish/Hispanic Origin = 1,264 Spanish surname only = 81 Unknown whether Spanish or not = 266 American Indian = 12 Chinese = 11 Japanese = 10 Filipino = 17 Hawaiian = 1 Korean = 3 Asian Indian/Pakistani = 24 Vietnamese = 11 Thai = 2 Asian non-specified = 30 Pacific Islander non-specified = 6 Other = 54 U nkn own = 102 Box B Tumors on women only n=20,797 Tumors not on women n=172 Box C Tumors on non-Hispanics only n=18,404 Tumors not on Hispanics n=2,393 Box D Tumors on Blacks or Whites n=18,121 Tumors not on Blacks or Whites n=283 Box E Tumors on cases with consistent race reporting n=18,114 Tumors on cases with discrepant race reporting n=7 Box F Tumors in Local stage n=8,876 Continued on next page Tumors not in Local stage n=9,238 Box A BCA tumors n=20,969 In situ = 3,022 Regional/direct extension = 195 Regional/nodes only = 3,405 Regional/direct extension and nodes =357 Regional, non-specified = 45 Distant/systemic disease = 631 Unknown = 1 583 34

PAGE 46

Figure 3. (Continued). Box L Where multiple tumors, most complete tumor records retained n=7,964 Least complete tumor records removed n=53 Box M Where multiple tumors, those with consistent insurance reporting retained n=7,883 Removal of multiple tumors with inconsistent insurance reporting n=81 Box K Where inconsistent, tumors reporting earliest age of diagnosis retained n=8,017 Tumors reporting no RT on individuals with tumors reporting RT n=510 Tumors reporting later age of diagnosis on individuals with discrepant age reporting n=15 Tumors with other surgery values n=217 N o surgery = 205 Surgery not specified = 12 BCS-treated tumors on individuals who reported Mastectomy-treated tumors Box G Tumors with BCS or Mastectomy surgery n=8,659 Box H Where inconsistent, Mastectomy-treated tumors retained n=8,542 Box J Where inconsistent, tumors reporting RT retained n=8,032 Box N Where multiple tumors, one tumor per individual retained n=7,396 Random records removed to leave 1 per individual n=487 Cases outside Florida n=217 Out of state = 194 Out of country = 23 Box P Final Dataset Tumors/cases in state of Florida n=7 179 Continued from previous page 35

PAGE 47

36 Hypothesis One, Breast Conserving Surgery Use Descriptive Statistics, Univariate. Hypothesis One asked if Black women in Florida in 2001 were statistically significan tly less likely than White women to receive breast conserving surgery rather than a ma stectomy for treatment of localized breast cancer. Table 9 summarizes the univariat e descriptive information for the study population who had Local stage breast cancer a nd were eligible for the study based on the previously discussed criteria. The total numbe r of cases included in this study was 7,179. Of those cases, 66 percent (4,763) underw ent breast conserving surgery and the remaining 44 percent (2,416) had mastecto mies. The same table shows the exact numbers and proportions of women in the to tal population for each covariate further stratified by the outcome variab le. The racial distribution of the study shows that Black cases constituted 6 percent of the tota l population (460 Black v. 6719 White). The proportion of participants in each age category continues to increase with the exception of the Very Old category. The mean age for cases was 64.95 with a st andard deviation of 13.54 years. For the variable Marital Status, 56 percen t (n = 4054) or the majority of the population in this data set fall into the Marr ied category with the next populous category, Other, being comprised of 31 percent (n = 2243) of the population. Insurance Type had 1 percent of the population on Medicaid, 3 percent Uninsured, 48 per cent on Medicare, 45 percent had Private insurance and 3 percent were Unknown. In the initial ve rsion of the USDA Rural-Urban Continuum 4,164 (58%) of the population lived in a Metro area larger than 1 million people; 2,170 (30%) li ved in a Metro area with 250K-1 million people; 416 (6 %) lived in a Metro area with less than 250K; 126 (2 %) lived in a Non

PAGE 48

Metro area with greater than or equal to 20K; and 15 cases (0 %) lived in Non Metro areas with less than 2.5K people. In order to make meaningful comparisons for populations living in all areas, the description of the final Rural-Urban codes used in analysis were as follows: 4,164 (58%) in Metro areas; 2,586 (36%) in Small Metro areas; and 423 (6%) in Non Metro. Table 9 also looks at the Outcome variable, Surgical Treatment for the entire population. Sixty-six percent of the population had Breast Conserving Surgery. For the Age categories, Middle Age (55-64) and Old (65-79) higher percentages of cases undergoing Breast Conserving Surgery and Mastectomy than in either of the two other Age categories. Table 9 Descriptive Statistics for Local Stage Breast Cancer Population. Total BCS Mastectomy N Percentage N Percentage N Percentage Total N 7179 100% 4763 100% 2416 100% Surgery Type BCS 4763 66% XX XX XX XX Mastectomy 2416 34% XX XX XX XX Race White 6719 94% 4477 94% 2242 93% Black 460 6% 286 6% 174 7% Age Groups Young >50 1132 16% 715 15% 417 17% Middle Age 50-64 2143 30% 1443 30% 700 29% Old 65-80 2857 40% 1955 41% 902 37% Very Old 80+ 1047 15% 650 14% 397 16% Mean, SD: Young: 43.4, 13.5 Middle: 57.3, 4.2 Old: 72.1, 4.23 Very Old: 84.2, 3.78 Marital Groups Married 4054 56% 2748 58% 1306 54% Single 672 9% 413 9% 259 11% Other 2243 31% 1455 31% 788 33% Unknown 210 3% 147 3% 63 3% 37

PAGE 49

Table 9 (Continued) Total BCS Mastectomy N Percentage N N Percentage N Insurance Type Medicaid 99 1% 51 1% 48 2% Medicare 3478 48% 2314 49% 1164 48% Private 3235 45% 2192 46% 1043 43% Uninsured 182 3% 76 2% 106 4% Unknown 185 3% 130 3% 55 2% Rural/USDA-all Metro >= 1 mill 4164 58% 2784 58% 1380 57% Metro 250k-1 mill 2170 30% 1480 31% 690 29% Metro <250k 416 6% 258 5% 158 7% Non Metro >=20k 282 4% 164 3% 118 5% Non Metro 2.5-20k 126 2% 65 1% 61 3% Non Metro <2.5k 15 0% 9 0% 6 0% Unknown/Out of State 6 0% 3 0% 3 0% Rural/USDA-Collapsed Metro >= 1 mill 4164 58% 2784 58% 1380 57% Small Metro 2586 36% 1738 36% 848 35% Non Metro 423 6% 238 5% 185 8% Unknown/Out of state 6 0% 3 0% 3 0% Descriptive Statistics, Bivariate. Table 10 provides a more in-depth overview of the local stage breast cancer population in the study by detailing the number of cases and the appropriate percentages when evaluating the relationship between the predictor variable of interest, Race, according to each of the covariates. White cases constituted 6719 of the total 7,179 cases used in the study, or 93.59 percent. There were 460 Black cases or 6.4 percent. In the Age categories, White cases fell predominantly into the Middle Age (50-64 years) and Old Age (65-80 years old) groups with 1,976 (29%) in Middle Age and 2,735 (41%) in Old Age. The highest percentage categories for Black cases were Young (134 cases, 29%) and Middle Age (167 cases, 36%). Table 11 38

PAGE 50

39 presents a more detailed look at differences within Age categories according to Race. For White cases the average age at diagnosis was 65.38 years with a st andard deviation of 13.4 years. Black cases had a range of 27-96 years of age wi th a mean of 58.64 years of age at diagnosis. The Black standard de viation was 14.06 years. Overall, the Black category had younger mean ages within age cate gories and larger intr a-category standard deviations than the White category. Figure 3 gr aphically illustrates the differences in age distribution between Blac k and White categories. Returning to Table 10, under Marital Stat us, the majority, 58%, of White cases fell into the Married categor y (n = 3,877). Black cases ha d a more level distribution among the three categories: Married, 38% ; Single, 24%; and Other 33%. Health Insurance Type had White cases broken down as follows: Medicare, 50%; Private, 45%, Uninsured, 2%; and Medicaid, 1%. Black cases fell into the following categories for Insurance Type: Private, 53%, Medicare, 33%, Uninsured, 6% and Medicaid, 5%. Rurality for White cases was divided as follows: 57% in Metro areas, 37% in Small Metro areas, and 6% in NonMetro areas. Black cases fell into the following categories: 69% in Metro areas, 26% in Small Metro area s, and 5% in Non-Me tro areas. Figure 3 visually presents the variati on between Black and White case s in a series of histograms for each of the covariates. Table 10 also describes the crude relations hip between the vari able of interest, Race, and the outcome variable, Surgery T ype, with 67% of White cases and 62% of Black cases undergoing Breast Conserving Surgery. Percentage-wise, Black women were less likely to receive Breast Conservi ng Surgery, but the diffe rence was not quite statistically significant (Crude OR=.823, CI=.677-1.00).

PAGE 51

Table 10 Bivariate Descriptive Statistics, Categorical Covariates by Race White Black N % N % Total N 6719 100 460 100 Age Groups Young >50 998 15 134 29 Middle Age 50-64 1976 29 167 36 Old 65-79 2735 41 122 27 Very Old 80+ 1010 15 37 8 Marital Groups Married 3877 58 177 38 Single 563 8 109 24 Other 2091 31 152 33 Unknown 188 3 22 5 Surgery Type BCS 4477 67 286 62 Mastectomy 2242 33 174 38 Insurance Type Medicaid 77 1 22 5 Medicare 3326 50 152 33 Private 2990 45 245 53 Uninsured 155 2 27 6 Unknown 171 3 14 3 Rural/USDA-All Metro >= 1 mill 3845 57 319 69 Metro 250k-1 mill 2073 31 97 21 Metro <250k 395 6 21 5 Non Metro >=20k 275 4 7 2 Non Metro 2.5-20k 111 2 15 3 Non Metro <2.5k 15 0 0 0 Unknown/Out of State 5 0 1 0 Rural/USDA Collapsed Metro >= 1 mill 3845 57 319 69 Small Metro 2468 37 118 26 Non Metro 401 6 22 5 Unknown/Out of State 5 0 1 0 40

PAGE 52

Table 11 Mean, SD, and Range of Age by Race. White Black N Mean SD Range N Mean SD Range Total N 6719 65.38 13.40 21-100 460 58.64 14.06 27-96 Young >50 998 43.54 4.63 21-49 134 42.37 5.16 27-49 Middle 50-64 1976 57.40 4.19 50-64 167 56.60 4.24 50-64 Old 65-79 2735 72.16 4.23 65-79 122 71.32 4.23 65-79 Very Old 80+ 1010 84.24 3.77 80-100 37 85.00 3.79 81-96 Proportion of Participants Receiving BCS by Race in Each Age Category49%(18/37)66%(80/122)64%(107/167)60%(81/134)63%(632/1010)69%(632/2735)68%(1336/1976)64%(634/998)0%10%20%30%40%50%60%70%80%Young (<50)Middle (50-64)Old (65-79)Very Old (80+)Age CategoryPercentag e Black White Proportion of Participants Receiving BCS by Race in Each Marital Status Category64%(14/22)61%(92/152)61%(66/109)64%(114/177)71%(133/188)65%(1363/2091)62%(347/563)68%(2634/3877)54%56%58%60%62%64%66%68%70%72%MarriedSingleOtherUnknownMarital StatusPercentage Black White Figure 4. Histograms of Bivariate Analysis of BCS by Race and Category 41

PAGE 53

Proportion of Participants Receiving BCS by Race in Each Insurance Category79%(11/14)56%(15/27)55%(12/22)61%(92/152)64%(156/245)70%(119/171)39%(61/155)51%(39/77)67%(2222/3326)68%(2036/2990)0%10%20%30%40%50%60%70%80%90%PrivateMedicareMedicaidUninsuredUnknownType of InsurancePercentage Black White Proportion of Participants Receiving BCS by Race in each Urban/Rural Category100%(1/1)41%(9/22)62%(73/118)64%(204/319)60%(3/5)57%(229/401)67%(1665/2468)67%(2580/3845)0%20%40%60%80%100%120%MetroSmall MetroNon-metroUnknown/Out of StateUrban/Rural CategoryPercentage Black White Figure 4. (Continued) Modeling. The next items presented are the iterative processes for determining which of the covariates are useful to the explanation of the Outcome Variable, BCS use in its relation to Race. Multiple combinations of covariates and interaction terms were developed. The most relevant of the models that explain the impact of Race on BCS are illustrated in Appendix A. Also, the role of the other covariates in terms of disentangling the impact of Race on BCS is illustrated. The table clearly shows which of the covariates were statistically significant for each model as well as the Odds Ratio and 95% 42

PAGE 54

43 Confidence Interval in models where Black is a statistica lly significant covariate. The value of the -2 Log Likelihood function and de grees of freedom (df) are also given for each equation in order to make meaningful comparisons while attempting to disentangle the effects of Race on the outcome BCS. Log Likelihood functions are frequently used to compare the support given to competing hypotheses by the data (Rothman 1998, Goodman 1988). Then Appendix B displa ys the comparison of models based on percentage change in Black covariate, th e difference in Log Likelihood function, degrees of freedom and the chi-square p-value. Table 12 contains the parameter estim ates, p-values, odds ratios and 95% confidence intervals for the crude model. Table 13 contains the same values for the full model. Table 14 presents the full model with all possible interaction terms including the predictor variable of interest, Race. The crude odds rati o for Black is 0.8230 and the term is barely statistically si gnificant with a 95% Confid ence Interval of 0.6670-1.000. The full model pushes Black into not being si gnificant. In Table 13, the statistically significant terms include Single, Medicaid, Uninsured and Non-Metro. Other Married and Small Metro border on sta tistical significance with 95% confidence interval ranges of 0.802-1.008 and 0.889-1.099 respectively. In Table 14 none of the possible interaction terms between the cova riates and the predictor variab le of interest prove to be statistically significant. Appendix B, Comparison of Models for Br east Conserving Surgery, illustrates the variation between the models that include various elements of Race, Age, Insurance Type, Marital Status and Rurality. By calcu lating the -2 Log Li kelihood differences and degrees of freedom and running a chi-square distribution, improvements in subsequent

PAGE 55

44 logistic models can be determined. The extreme Chi-square p-values resulting from comparisons of the A-series of models show that including other covariates proves useful in explaining the relationship between Race and Surgery Type. Table 12 Crude Model for Surgical Treatment Parameter Estimate OR P-value 95% CI Crude Model Black -0.1947 0.8230 0.0506 0.6770 1.000 Table 13 Full Model for Surgical Treatment Parameter Estimate OR P-value 95% CI Full Model Black -0.1067 0.8990 0.2989 0.735 1.099 Young (<50) reference reference reference reference Middle (50-64) 0.1618 1.1760 0.0389 1.008 1.371 Old (65-79) 0.2026 1.2250 0.0412 1.008 1.488 Very Old (80+) -0.0607 0.9410 0.5992 0.751 1.180 Married reference reference reference reference Single -0.2142 0.8070 0.0155 0.679 0.960 Other -0.1059 0.9000 0.0689 0.802 1.008 Private reference reference reference reference Medicare -0.0654 0.9370 0.4151 0.800 1.096 Medicaid -0.5406 0.5820 0.0092 0.388 0.875 Uninsured -1.0412 0.3530 <0.0001 0.260 0.479 Metro (>=1 million) reference reference reference reference Metro (<1 million) -0.0117 0.9880 0.8288 0.889 1.099 Non-metro -0.4507 0.6370 <0.0001 0.519 0.783

PAGE 56

45 Table 14 Full Model with Interaction Terms for Surgical Treatment Parameter Estimate OR P-value 95% CI Interaction Model Black -0.1152 xx 0.6274 xx Young (<50) reference reference reference reference Middle (50-64) 0.1610 xx 0.0521 xx Old (65-79) 0.1925 xx 0.0644 xx Very Old (80+) -0.0557 xx 0.6436 xx Married reference reference reference reference Single -0.2297 xx 0.0156 xx Other -0.1028 xx 0.0877 xx Private reference reference reference reference Medicare -0.0574 xx 0.4927 xx Medicaid -0.6056 xx 0.0096 xx Uninsured -1.1564 xx <0.0001 xx Metro (>=1 million) reference reference reference reference Metro (<1 million) -0.0058 xx 0.9177 xx Non-metro -0.4211 xx 0.0001 xx Young (<50) Black reference reference reference reference Middle (50-64) Black 0.0103 xx 0.9683 xx Old (65-79) Black 0.1336 xx 0.7092 xx Very Old (80+) Black -0.2783 xx 0.5423 xx Married Black reference reference reference reference Single Black 0.1186 xx 0.6586 xx Other Black 0.0314 xx 0.8984 xx Private Black reference reference reference reference Medicare Black -0.1603 xx 0.5944 xx Medicaid Black 0.3111 xx 0.5492 xx Uninsured Black 0.7264 xx 0.1024 xx Metro (>=1 million) Black reference reference reference reference Metro (<1 million) Black -0.0784 xx 0.7354 xx Non-metro Black -0.4498 xx 0.3405 xx Tables 15 and 16 show the parameters of the best equations that explain the relationship between the covari ates and the outcome are mode ls C8 and C9. Both models include the covariates for Mi ddle Age (50-64), Old Age (6579), Very Old Age (80+),

PAGE 57

46 Medicaid, Single, Other Married and Non-Metr o. However, only model C8 contains the variable of interest, Black. In order to better explain th e relationship between the covariates and the outcome, Breast Conserving Surgery, all covariates were included in an iterative process of testing for interaction terms. None of the attemp ts proved to be statistically significant. The results of the modeling process can be seen in Appendix C. Table 15 Candidate for Best BCS Explanatory Model with Black, C8 Parameter Parameter Estimate SE OR 95% LL 95% UL P value Intercept 0.6574 0.0664 XX XX XX <.0001 BLACK -0.1326 0.1019 0.876 0.717 1.069 0.1931 NON METRO -0.4600 0.1023 0.631 0.517 0.771 <.0001 MIDAGE 0.1681 0.0776 1.183 1.016 1.377 0.0303 OLDAGE 0.2087 0.0760 1.232 1.062 1.430 0.0060 VERYOLD -0.0543 0.0936 0.947 0.788 1.138 0.5619 MCAID -0.4690 0.2069 0.626 0.417 0.938 0.0234 SINGLE -0.2262 0.0878 0.798 0.671 0.947 0.0100 OTHMAR -0.1138 0.0580 0.892 0.796 1.000 0.0498 Table 16 Candidate for BCS Best Explanatory Model without Black, C9 Parameter Parameter Estimate SE OR 95% LL 95% UL P value Intercept 0.6451 0.0657 XX XX XX <.0001 NON METRO -0.4580 0.1022 0.6325 0.5561 0.7195 <.0001 MIDAGE 0.1726 0.0775 1.1884 1.0209 1.3833 0.0259 OLDAGE 0.2176 0.0756 1.2431 1.0719 1.4416 0.0040 VERYOLD -0.0432 0.0931 0.9579 0.7979 1.1494 0.6429 MCAID -0.4846 0.2065 0.6159 0.4109 0.9232 0.0189 SINGLE -0.2396 0.0872 0.7869 0.6633 0.9336 0.0060 OTHMAR -0.1186 0.0579 0.8882 0.7929 0.9949 0.0406

PAGE 58

Hypothesis Two, Radiation Therapy Use Data Reduction. Hypothesis 2 asks if Black women in Florida are statistically significantly less likely than White women to receive radiation therapy after breast conserving surgery. Thus, a subset of cases from the larger data set used to answer the first hypothesis is needed in order to properly assess the relationship between Race and Radiation Therapy use. Figure 4 details the additional data reduction steps used to answer Hypothesis Two. The initial size of the sample is 4,763 cases that had breast conserving surgery. Of these BCS cases, 2,066 had radiation therapy; 2,678 did not have radiation therapy; and 19 cases were unknown for radiation therapy. 47 Breast Conserving Surgery Total Population N = 4,763 Radiation Therapy No N = 2,678 Unknown Radiation Therapy N = 19 Radiation Therapy Yes N = 2,066 Figure 5. Data Reduction for Radiation Therapy Analysis Descriptive Statistics, Univariate. Table 17 summarizes the univariate descriptive information for the study population who had breast conserving surgery and were otherwise eligible for inclusion in the study based on the previously discussed criteria. The total number of cases included in this portion of the study was 4,763. All of these cases underwent breast conserving surgery for treatment of localized breast cancer. Of those cases, 43 percent (2,066) had radiation therapy, 56 percent (2,678) reported no

PAGE 59

radiation therapy, and 3 percent (19) were unknown for radiation therapy status. Cases reporting unknown status will be excluded from further portions of this study. The same table shows the exact numbers and proportions of women in the total population for each covariate further stratified by the outcome variable. The racial distribution shows that Black cases constituted 6 percent of the BCS population (286 Black v. 4,477 White). For Age groups, each age category continues to increase with the exception of the Very Old category. Table 17 also shows the mean age 64.98 and standard deviation, 13.24 years, for the population included in this analysis. The age range for the study was 21100 years. Table 17 Descriptive Statistics of Breast Conserving Surgery Population Total Radiation Therapy-Yes No Radiation Therapy N Percentage N Percentage N Percentage Total N 4763 100% 2066 100% 2678 100% Radiation Therapy Yes 2066 43% XX XX XX XX No 2678 56% XX XX XX XX Unknown 19 0% XX XX XX XX Race White 4477 94% 1948 94% 2512 94% Black 286 6% 118 6% 166 6% Age Groups Young >50 715 15% 262 13% 447 17% Middle Age 50-64 1443 30% 645 31% 794 30% Old 65-79 1955 41% 936 45% 1012 38% Very Old 80+ 650 14% 223 11% 425 16% Mean, SD: Young: 43.4, 4.67 Middle: 57.4, 4.15 Old: 72.05, 4.23 Very Old: 84.18, 3.79 Marital Groups Married 2748 58% 1242 60% 1496 56% Single 413 9% 158 8% 251 9% Other 1455 31% 603 29% 848 32% Unknown 147 3% 63 3% 83 3% 48

PAGE 60

Table 17. (Continued) Total Radiation Therapy-Yes No Radiation Therapy N Percentage N Percentage N Percentage Insurance Type Medicaid 51 1% 21 1% 30 1% Medicare 2314 49% 1032 50% 1275 48% Private 2192 46% 930 45% 1250 47% Uninsured 76 2% 26 1% 50 2% Unknown 130 3% 57 3% 73 3% Rural/USDA-all Metro >= 1 mill 2784 58% 1126 55% 1644 61% Metro 250k-1 mill 1480 31% 741 36% 736 27% Metro <250k 258 5% 115 6% 141 5% Non Metro >=20k 164 3% 54 3% 110 4% Non Metro 2.5-20k 65 1% 24 1% 41 2% Non Metro <2.5k 9 0% 5 0% 4 0% Unknown/Out of State 3 0% 1 0% 2 0% Rural/USDA-Collapsed Metro >= 1 mill 2784 58% 1126 55% 1644 61% Small Metro 1738 36% 856 41% 877 33% Non Metro 238 5% 83 4% 155 6% Unknown/Out of state 3 0% 1 0% 2 0% Table 17 also shows that for the variable Marital Status, 58 percent (n = 2,748), the overwhelming majority of the population in this data set fall into the Married category with the next populous category, Other, being comprised of 31 percent (n = 2243) of the population. Insurance Type had 1 percent of the population on Medicaid, 2 percent Uninsured, 49 percent on Medicare, 46 percent had Private insurance and 3 percent were Unknown. In the initial version of the USDA Rural-Urban Continuum 2,784 (58%) of the population lived in a Metro area larger than 1 million people; 1,480 (31%) lived in a Metro area with 250K-1 million people; 258 (5 %) lived in a Metro area with less than 250K. For Non Metro populations, 164 (3%) lived in a Non Metro area with greater than or equal to 20K; 65 cases (1 %) lived in Non Metro areas with between 2.5-20,000; 9 49

PAGE 61

50 cases of were in areas with less than 2,500 a nd 3 cases fell into the Unknown/Out of State category and were eliminated from further analysis. The Collapsed Rural-Urban categorical breakdown used in analysis is as follows: 2,784 (58%) in Metro areas; 1,738 (36%) in Small Metro areas; and 238 (5%) in Non Metro. In terms of Radiation Therapy Use, wh ich is the Outcome variable for this hypothesis, Table 17 shows that of the 4,763 total cases, 2,066 (43%) had Yes for Radiation Therapy in the FCDS data se t and 2,678 (56%) had No for Radiation Therapy. There were 19 cases (0%) that were Unknown for Radiation Therapy Use and were excluded from further statistical analysis. Descriptive Statisti cs, Bivariate. Table 18 provides a more in-depth overview of the breast conserving surgery population in the study by detailing the number of cases and the appropriate percentages when evalua ting the relationship between the predictor variable of interest, Race, according to each of the covariates. White cases constituted 4,477 of the total 4,763 cases used in this portion of the study, or 94 percent. There were 286 Black cases or 6 percent.

PAGE 62

Table 18 Descriptive Statistics, Categorical Covariates by Race for BCS White Black N Percentage N Percentage Total N 4477 100 286 100 Age Groups Young >50 634 14 81 28 Middle Age 50-64 1336 30 107 37 Old 65-79 1875 42 80 28 Very Old 80+ 632 14 18 6 MaritalGroups Married 2634 59 114 40 Single 347 8 66 23 Other 1363 30 92 32 Unknown 133 3 14 5 RadiationTherapy Yes 1948 44 118 41 No 2512 56 166 58 Unknown 17 0 2 1 InsuranceType Medicaid 39 1 12 4 Medicare 2222 50 92 32 Private 2036 45 156 55 Uninsured 61 1 15 5 Unknown 119 3 11 4 Rural/USDA All Metro >= 1 mill 2580 58 204 71 Metro 250k-1 mill 1420 32 60 21 Metro <250k 245 5 13 5 Non Metro >=20k 160 4 4 1 Non Metro 2.5-20k 60 1 5 2 Non Metro <2.5k 9 0 0 0 Unknown/ Out of State 3 0 0 0 Rural/USDA Collapsed Metro >= 1 mill 2580 58 204 71 Small Metro 1665 37 73 26 Non Metro 229 5 9 3 Unknown/ Out of state 3 0 0 0 51

PAGE 63

In the Age categories, White cases fell predominantly into the Middle Age (50-64 years) and Old Age (65-80 years old) groups with 1,336 (30%) in Middle Age and 1,875 (42%) in Old Age. The highest percentage category for Black cases was Middle Age (107 cases, 37%). Young (>50) and Old Age (65-79) had equal parts of the Black cases with 81 and 80 cases respectively or an even 28% of the population for each category. There was no statistical significance in terms of placement in the Young age category between Black and White cases (OR: 2.39, 95% CI: 1.9808-3.0282). Table 19 presents a more detailed look at differences in Race according to Age. Table 19 Mean, SD, and Range of Age by Race for BCS Population White Black N Mean SD Range N Mean SD Range Total N 4477 65.38 13.13 21-100 286 58.79 13.46 28-96 Young >50 634 43.52 4.67 21-49 81 42.79 4.61 28-49 Middle Age 1336 57.49 4.14 50-64 107 57.01 4.34 50-64 Old 65-79 1875 72.06 4.22 65-79 80 71.49 4.23 65-79 Very Old 80+ 632 84.16 3.78 80-100 18 84.89 4.17 81-96 For White cases the average age at diagnosis was 65.38 years with a standard deviation of 13.13 years. The White range was 21-100 years. Black cases had a range of 28-96 years of age with a mean of 58.64 years of age at diagnosis. The Black standard deviation was 13.46 years. Overall, the Black category had younger mean ages within age categories with the exception of the Very Old category where Black cases had a mean age of 84.89 versus 84.16 from White cases. 52

PAGE 64

53 Returning to Table 18, under Marital Stat us, the majority, 59%, of White cases fell into the Married category (n = 2,634). 8% (n=347) into the Single category and 30% (n-1363) into the Other category with 3% (n=133) Unknown. Black cases were distributed more equally across the Marita l categories: Married, 40%; Single, 23%; and Other 32%. Health Insurance Type had White cases broken down as follows: Medicare, 50%; Private, 45%, Uninsured, 1%; and Medi caid, 1%. Black cases fell into the following categories for Insurance Type: Pr ivate, 55; Medicare, 32%; Uninsured, 5%; and Medicaid, 4%. Rurality for White cases was divided as follows: 58% in Metro areas, 37% in Small Metro areas, and 5% in Non Metro areas. Black cases fell into the following categories: 71% in Metro areas, 26% in Small Metro areas, and 3% in Non Metro areas. Table 18 also shows the crude relationship between the variable of interest, Race, and the outcome variable, Radi ation Therapy. According to the information in the data set, for White cases 44% and for Black cases 41% received Radiation Therapy. Percentage-wise, Black women were less lik ely to receive Radiation Therapy following Breast Conserving Surgery, but the difference was not statistically significant (Crude OR= 0.917, CI=0.719-1.17). Modeling. The next items presented are the iterative processes for determining which of the covariates are useful to the e xplanation of the Outcome Variable, Radiation Therapy (RT-Yes) use in its relation to Race. Numerous models were developed in order to accomplish this task using all combinations of covariates and in teraction terms. The most relevant of the models that explain the impact of Race on receipt of radiation therapy (RT-Yes) are illustrated in Appendix D. Also, the role of the other covariates in

PAGE 65

54 terms of teasing out the impact of Race on RT-Y es is illustrated. The table clearly shows which of the covariates were statistically significant for each model as well as the Odds Ratio and 95% Confidence Interval for equa tions that include Black. The -2 Log Likelihood value and degrees of freedom (df) ar e also given for each equation in order to make meaningful comparisons between m odels. Appendix E then displays the comparison of models based on pe rcentage change in Black cova riate, the difference in -2 Log Likelihood function, degrees of free dom and the chi-square p-value. Tables 20, 21 and 22 present summaries of the crude and full models as well as the full model with all possible interaction terms relating Black to the other covariates. In Table 20, the crude OR for Black is 0.917. The relationship is not statistically significant, 95% CI: 0.719-1.170. The full m odel given in Table 21 has the following significant covariates: Middle Age (50-64), Old (65-79) and Non-Metro. The two Marital Status terms Single and Other were out side the range of st atistical significance with upper limits to the 95% Confidence Inte rval ranges of 1.03. Table 22 is the full model with all possible covariates interacting with the predictor variable of interest, Race. Statistical significance in this model is base d on the p-value for each possible covariate or interaction term. Significant covariates incl uded Middle Age (50-64), Old (65-79). Small Metro and Non-Metro was just out of range w ith a p-value of .0609. The interaction term between Black and Medicaid proved to be statistically significant.

PAGE 66

55 Table 20 Crude Model for RT Crude Model Parameter Estimate OR P-value 95% CI Black -0.0866 0.9170 0.4856 0.7190 1.1700 Table 21 Full Model for RT Full Model Parameter Estimate OR P-value 95% CI Black 0.0116 1.0120 0.9278 0.788 1.299 Young (<50) reference reference reference reference Middle (50-64) 0.3197 1.3770 0.0008 1.143 1.659 Old (65-79) 0.4157 1.5150 0.0005 1.200 1.913 Very Old (80+) -0.1156 0.8910 0.4236 0.617 1.182 Married reference reference reference reference Single -0.1879 0.8290 0.0901 0.667 1.030 Other -0.1002 0.9050 0.1443 0.791 1.035 Private reference reference reference reference Medicare 0.0417 1.0430 0.6587 0.867 1.254 Medicaid 0.0545 1.0560 0.8520 0.595 1.872 Uninsured -0.2697 0.7640 0.2772 0.469 1.242 Metro (>=1 million) reference reference reference reference Small Metro 0.3331 1.3950 <0.0001 1.235 1.576 Non-metro -0.2845 0.7520 0.0466 0.569 0.996

PAGE 67

56 Table 22 Full Model with Interaction Terms for RT Interaction Model Parameter Estimate OR P-value 95% CI Black 0.0910 xx 0.7629 xx Young (<50) reference reference reference reference Middle (50-64) 0.3347 xx 0.0009 xx Old (65-79) 0.4288 xx 0.0006 xx Very Old (80+) -0.0890 xx 0.5528 xx Married reference reference reference reference Single -0.2239 xx 0.0609 xx Other -0.1051 xx 0.1375 xx Private reference reference reference reference Medicare 0.0518 xx 0.5971 xx Medicaid -0.2873 xx 0.4075 xx Uninsured -0.3875 xx 0.1707 xx Metro (>=1 million) reference reference reference reference Small Metro 0.3360 xx <0.0001 xx Non-metro -0.2730 xx 0.0613 xx Young (<50) Black reference reference reference reference Middle (50-64) Black -0.1624 xx 0.6181 xx Old (65-79) Black -0.1809 xx 0.6824 xx Very Old (80+) Black -1.0751 xx 0.1512 xx Married Black reference reference reference reference Single Black 0.2891 xx 0.3937 xx Other Black 0.2322 xx 0.4438 xx Private Black reference reference reference reference Medicare Black -0.2590 xx 0.4946 xx Medicaid Black 1.5022 xx 0.0452 xx Uninsured Black 0.5283 xx 0.3905 xx Metro (>=1 million) Black reference reference reference reference Small Metro Black -0.0970 xx 0.7539 xx Non-metro Black -0.3826 xx 0.6248 xx In order to better explain th e relationship between the covariates and the outcome, receipt of Radiation Therapy, all covariates were included in an iterative process of testing for interaction terms. The J-series of models in Appendix D show the significant covariates and the -2 Log Likelihood and degrees of freedom for each model. The

PAGE 68

interaction of Black and Medicaid proved to be statistically significant. The initial results of the comprehensive interactive modeling process can be seen in Appendix F. Proportion of BCS Participants Receiving Radiation Therapy by Race in Each Insurance Category36%(4/11)47%(7/15)67%(8/12)37%(34/92)42%(65/156)45%(53/119)31%(19/61)33%(13/39)45%(998/2222)42%(865/2036)0%10%20%30%40%50%60%70%PrivateMedicareMedicaidUninsuredUnknownType of InsurancePercentage Black White Figure 6. Histogram of Bivariate Analysis by Race and Payer for RT The interaction term between Black and Medicaid provides a better explanatory model and can be seen in Appendix E by the comparison between models C10 and J16. The same interaction results are clearly visible in Figure 6, Histogram of Bivariate Analysis by Race and Payer for RT. The significance of the chi-square distribution p-value demonstrates that the interaction term must be included. Tables 23 and 24 show the parameters of the two best equations that explain the relationship between the covariates and the outcome are models J15 and J16. Both models include the covariates for Middle Age (50-64), Old Age (65-79), Very Old Age (80+), Medicaid, Single, Other Married, Metro, Small Metro, Non-Metro and the interaction term of Black*Medicaid. Due to the paucity of covariates for inclusion in the models as well as the importance of Marital Status in the literature, Model J16 is the best explanatory model for the 57

PAGE 69

58 relationship between Race and receipt of Radiation Therapy. Among the other models that are in the J-series of models and include interaction terms, when J18 and J19 are compared to model J16, there proves to be no significant difference when removing the two Marital Status covariates Single and Other Married. Table 23 Candidate for Best Explanatory Model, J15 Parameter Parameter Estimate SE P value OR 95% LL 95% UL Intercept -0.6365 0.0837 <.0001 xx xx xx BLACK -0.0760 0.1293 0.5567 0.9268 0.8144 1.0547 MIDAGE 0.3204 0.0947 0.0007 1.3780 1.1440 1.6590 OLDAGE 0.4484 0.0911 <.0001 1.5660 1.3100 1.8720 VERYOLD -0.1090 0.1147 0.3417 0.8970 0.7160 1.1230 MCAID -0.3446 0.3455 0.3186 0.7085 0.5015 1.0009 SMALL METRO 0.3379 0.0622 <.0001 1.4020 1.2410 1.5840 NON-METRO -0.2818 0.1428 0.0485 0.7540 0.5700 0.9980 BLACK*MCAID 1.5046 0.7159 0.0356 4.5024 2.2005 9.2119 Table 24 Candidate for Best Explanatory Model, J16 Parameter Parameter Estimate SE P value OR 95% LL 95% UL Intercept -0.5908 0.0866 <.0001 xx xx xx BLACK -0.0451 0.1303 0.7295 0.9559 0.8391 1.0889 MIDAGE 0.3157 0.0951 0.0009 1.3710 1.1380 1.6520 OLDAGE 0.4472 0.0927 <.0001 1.5640 1.3040 1.8760 VERYOLD -0.0845 0.1194 0.4789 0.9190 0.7270 1.1610 MCAID -0.3043 0.3462 0.3793 0.7376 0.5218 1.0428 SMALL METRO 0.3319 0.0623 <.0001 1.3940 1.2330 1.5740 NON-METRO -0.2916 0.1430 0.0414 0.7470 0.5640 0.9890 SINGLE -0.1889 0.1112 0.0895 0.8280 0.6660 1.0300 OTHMAR -0.1001 0.0687 0.1451 0.9050 0.7910 1.0350 BLACK*MCAID 1.4861 0.3738 0.0380 4.4198 3.6872 9.1958

PAGE 70

59 Table 25 details the interaction effect of Black and Medicaid in the best explanatory model, J16. Black women on Me dicaid were twice as likely to have radiation therapy as White women not on Medi caid (OR: 2.95); four times as likely to have radiation therapy as White cases on Me dicaid (OR: 4.17); and three times as likely to have radiation therapy as Black cases not on Medicaid (OR: 3.19). Table 25 Interactive Effect of Black and Medicaid on Model J16 OR's for Interaction BLACK MCAID BLACK* MCAID Black on Medicaid compared to -0.0451 -0.3043 1.4861 White not on Medicaid 0.0000 0.0000 0.0000 Total para. Est. Total -0.0451 -0.3043 1.4861 1.1367 OR 3.1165 95% CI: 1.49786.4840 Black on Medicaid compared to -0.0451 -0.3043 1.4861 White on Medicaid 0.0000 -0.3043 0.0000 Total para. Est. Total -0.0451 0.0000 1.4861 1.441 OR 4.2249 95% CI: 2.03068.7902 Black on Medicaid compared to -0.0451 -0.3043 1.4861 Black not on Medicaid -0.0451 0.0000 0.0000 Total para. Est. Total 0.0000 -0.3043 1.4861 1.182 OR 3.2602 95% CI: 1.56726.7845

PAGE 71

60 Discussion and Conclusion Breast Conserving Surgery Hypothesis Hypothesis One asked if Black women in Florida in 2001 were statistically significantly less likely than White women to receive breast conserving surgery rather than a mastectomy for treatment of localized breast cancer. Based on results reviewed in the previous chapter the answer to this research question is no. The results of statistical analysis of the relationship between Black and Breast Conserving Surgery are not significant. The Odds Ratio is 0.823 a nd the 95% Confidence Interval has a range of 0.677-1.000. Consistent with the lit erature, age, insurance status, rural residence and marital status were signi ficant contributors in the explanation of breast conserving surgery use. There were differences between the Black and White categories. Of all the local stage breast cancer cases included in the study, Black women were more than two times as likely as White women to be less than 50 years of age (OR: 2.356, 95% CI: 2.143-2.568) and 1.4 times as likel y to have Private health insurance (OR: 1.421, 95% CI: 1.284-1.555). When all cont ributory covariates were included in modeling, Black race alone did not improve th e explanation of the outcome variable, Breast Conserving Surgery. Extending insurance coverage to the poor seems to alleviate disparity for differences in breast cancer treatment, but cl early Medicaid is not the panacea for all diseases. Poverty as measured by Medicaid status is not associated with Breast Conserving Surgery when other va riables are controlled for.

PAGE 72

61 In terms of the state of Floridas rate of BCS use as an alternative to Mastectomy for local breast cancer, Florida exceeds natio nal geographic variati ons addressed in the Review of the Literature by having a combined Black and White proportion of 66%. However, there are incalculable difficultie s with examining breast conserving surgery versus mastectomy as a quality of cancer ca re outcome measure. The optimum rate of breast conserving therapy use is not 100%. Specifically, surgical treatment as an outcome alone focuses on a well-recorded, easily accessi ble data set that in many ways negates the salient and complex series of decisions and processes that precede surgery. Examples include womens preferences supporting life-style such as a quick versus prolonged treatment duration, the role of clinician (especially surge on) attitude in the decisionmaking process, womens post-surgical satis faction and peace of mind. The role of patient preference in maintaining high rates of mastectomy has been relatively neglected in research (Morrow 2003). A better measur e of quality of care, however difficult to obtain, might be to study the degree to which women are being fully informed of their surgical treatment options and the degree to which they are involved in the treatment decision-making process. Radiation Therapy Hypothesis Hypothesis Two asked if Black women in Florida were statistically significantly less likely than White women to receive radi ation therapy after breast conserving surgery. Based on results presented in the previ ous chapter, again the answer is no. The results of statistical analysis of the relationship between Black and Radiation Therapy use are not significant. The O dds Ratio is 0.9170 and the 95% Confidence Interval has a range of 0.7190-1.170. Consistent with the lit erature, age, insurance status,

PAGE 73

62 rural residence and marital status were signi ficant contributors in the explanation of radiation therapy use. There were differences between the Black and White categories. When all contributory covariates were in cluded in modeling, Black race alone was not necessary as an explanatory variable in the outcome, receipt of Radiation Therapy. However, the true number of post-breast conserving surgery cas es that actually received radiation therapy remains unknow n. As Shavers, Richardson and Button reported, there is a drop-off in reporting of cancer care across the continuum of treatment such that in this study only 43% of cases re ceived radiation thera py. Du reported that SEER data under recorded radiation therapy use by 26% (Du 1999). According to FCDS statistical analyst and data base manager, Jackie Button, all Radia tion Therapy that is recorded is listed under one facility number, usually that of the hospital (Button 2004). Additionally, radiation treatment facilities are listed by the Florida Cancer Data System in a separate Facility Layout that does not match to the Cancer File Layout. Thus, not only is there no mandatory reporting of radiat ion therapy receipt, in the event that treatment is recorded it has no match to pati ent records, making impossible the ability to geocode or to otherwise calcu late distance from residence of the patient to treatment facility. Our findings are consistent with the literature, Black women on Medicaid were more likely to receive radiation therapy than those women not insured by Medicaid (Bradley 2002). In order to determine if the statistically significant interaction term between Black and Medicaid for receipt of radiation therapy post-breast conserving surgery was an artifact of the manner in whic h data is collected, a separate analysis was run that compared Metro to a collapsed Sm all Metro and Non Metro populations. There

PAGE 74

63 was no statistically significant difference in the results, which reaffirms that Black women on Medicaid have the highest rate of radiation therapy use. Implications and Recommendations While this study adds to the knowledge of the relationship between race and breast conserving surgery and radiation therapy use for localized breast cancer, a number of limitations should be acknowledged. Consis tent with most of the studies in the literature, this study presente d analysis based on observationa l rather than experimental data. The use of the FCDS dataset made it im possible to fully account for differences in patient characteristics that may have effected selection into the various categories of covariates. Another of the us ual limitations is geographic scope which may be acceptable because the region covered is large in elderl y population size and important with respect to the evolution of geriatric health care. Last, co-morbidity data that may have been counter indications to the use of breast conserving surger y and radiation therapy were absent so that an unknown number of patient s may have been otherwise ineligible for inclusion in the study. An impressive array of organizations f und Breast Cancer research on a myriad of levels. On the national level the leader is the National Cancer Inst itute followed by the National Institutes of Health. There is also the federally funded Department of Defense Breast Cancer Research Program Other national and state leve l organizations include the American Cancer Society, the Susan G. Ko men Breast Cancer Research Fund and the AVON Breast Cancer Organization. Each of these organizations should fund research that seeks not only to find new and innovative breast cancer cures (tr eatment) but also to fund research that tracks disparities.

PAGE 75

64 With breast cancer firmly entrenched on th e national research agenda, the policies that enable poor and underserved women to access screening and treatment should be subjected to further research so that furthe r policy recommendations can be made based on Adays principles of effectiveness, effi ciency and equity (Aday 2004). Should the NBCCEDP be successfully evaluated as achie ving its healthcare system objective, then further policy ideas could be generated to e xpand screening and tr eatment services to other diseases such as colon cancer or prostate cancer. Although those policy recommendations may not be received or acted upon immediatel y, the ideas could circulate as research generated alternatives in Kingdons policy primeval soup where in they can await the confluence of the streams and the opening of a policy window (Kingdon 1995). However, in the absence of furt her health services research pertaining to the expected percentages of appropriate use of breast c onserving surgery and radiation therapy, it would be unwise to dictate a policy standard such as a HEDIS measure. The findings in this study have impo rtant implications for policy makers, clinicians and researchers that call for and coul d aid in additional initi atives and studies to provide improved understanding of the quality of breast cancer treatment in Florida. 1. Other variables could have been useful in explaining the relationship between the variable of interest and either of the outcomes, breast conserving surgery and radiation therapy use. Examples of these that are not collected nor included in the FCDS dataset are Body Mass Index (BMI), co-morbidities, annual household income, and level of education. 2. The FCDS should act to c ounter the severe limitations in the ability to analyze the appropriate receipt of radiation therapy after breast conserving surgery.

PAGE 76

65 The first is the lack of mandatory reporting requirements by free-standing treatment facilities. The second is FC DS failure to include location of treatment in the tumor registry. 3. Research efforts should focus on the imp act of patient choice, clinician bias and informed consent in the decision making process, and their complex relation to mastectomy versus breast conserving therapy use for local stage breast cancer. 4. Researchers and policymakers must conti nue to observe and monitor trends in breast cancer incidence, screening, tr eatment and mortality in order to accurately assess changes and make pr ogress toward diminishing breast cancer disparities in all population subgroups. New therapies such as shortterm, intense radiation therapy treatmen t may make breast conserving therapy an acceptable option for rural residence. 5. Health advocates and hea lth policymakers should take note of the effect of Medicaid as a social equalizer for breast cancer treatment specifically through the vehicle of the National Breast and Cervical Cancer Early Detection Program (NBCCEDP). Future initiatives could seek to provide screening and treatment through the Medicaid system fo r other diseases which could help to dramatically eliminate disparities in other areas.

PAGE 77

66 References Aday L, Begley C, Lairson D, Balkrishnan R. (2004) Evaluating the healthcare system. (3 rd ed.) Chicago: Health Administration Press. American Cancer Society. (2003). Cancer facts & figures for African Americans Atlanta. Ballard-Barbash R, Potosky A, Harlan L, Nayfield S, Kessler L. (1996) Factors associated with surgical and radiation therapy for early stage breast cancer in older women. JNCI 88 (11), 716-26. Bickell, N. (2002). Race, ethnicity, and disp arities in breast ca ncer: victories and challenges. Women's Health Issues 12(5), 238-251. Blinchert-Toft M, Brincker H, Andersen J, et al. (1988). A Danish randomized trail comparing breast-preserving therapy w ith mastectomy in mammary carcinoma: preliminary results. Acta Oncol 27, 671-677. Boyer-Chammard A, Taylor T, Anton-Culver H. (1999) Survival differences in breast cancer among racial/ethnic groups: a population based study. Detect Prev 23, 463-473. Bradley C, Given C, Roberts C. (2002) Race, socioeconomic status, and breast cancer treatment and survival. JNCI 94(7), 490-496. Brawley O. (2002) Disaggregating the eff ects of race and poverty on breast cancer outcomes. JNCI 94(7), 471-473. Breen N, Wesley M, Merrill R, Johnson K. (1999) The relationship of socio-economic status and access to minimum expected therapy among female breast cancer patients in the National Cancer Instit ute Black-White Cancer Survival Study. Ethnic Dis 9, 111-125. Button, Jackie (2004, May 15) Personal interview. Centers for Disease Contro l and Prevention. (2003). National Breast and Cervical Cancer Early Detection Program. www.cdc.gov/cancer/nbccedp accessed 6 December 2003

PAGE 78

67 Diehr P, Yergan J, Chu J, et al. (1989) Treatment modality and quality differences for black and white breast-cancer patient s iterated in community hospitals. Med Care, 27, 942-958. Du X, Freeman J, Goodwin J. (1999) Informa tion on radiation treatment in patients with breast cancer: the advantages of th e linked Medicare and SEER data. J Clin Epidemiology 52, 463-70. Dunmore C, Plummer P, Regan G, Mattingly D, Jackson S, Milikan R. (2000). Re: Race and differences in breast can cer survival in a manage d care population [letter]. JNCI 92, 1690-1. Elk R, Morrow M. (2003) Breast Cancer for Dummies Hoboken, NJ: Wiley. Farrow D, Hunt W, Samet J. (1992). Geographi c variation in the tr eatment of localized breast cancer. N Engl J Med, 326, 1097-1101. Feinstein A, Wells C, Walter S. (1990). A comparison of multivariable mathematical methods for predicting survival--I. in troduction, rationale, and general strategy. J Clinical Epi, 43 (4), 339-47. Fisher B, Anderson S, Redmond C, et al. (1995) Reanalysis and results after 12 years of followup in a randomized clinical trail comparing total mastectomy with lumpectomy with or without irradiation in the treatment of breast cancer. N Engl J Med 333(22), 1456-1461. Fisher B, Redmond C, Poisson R, et al. (1989) Eight-year results of a randomized clinical trial comparing total mastectomy and lump ectomy with or without irradiation in the treatment of breast cancer. N Engl J Med, 320, 822-828. Florida Cancer Data System Home. (2003) Retrieved November 21, 2003 from http://fcds.med.miami.edu/welcome.html Florida Cancer Data System. (2003) Data Acquisition Manual II. Retrieved 10 March 2004 from FCDS website: http://fcds.med.miami.edu/downloads/dam/DAM_SECTIONII.pdf Florida Cancer Data System. (2003) Monograph of Cancer in Florida: Volume 2, 2001Admissions--Analytic. Retrieved November 19, 2003 from FCDS web site: http://fcds.med.miami.edu/welcome.html Frankfurt-Nachmias C, Nachmias D (1996) Research Methods in the Social Sciences (5 th ed) New York: St. Martin's Press.

PAGE 79

68 Gilligan M, Kneusel R, Hoffmann R, Greer A, Nattinger, A. (2002) Persistent differences in sociodemographic determinants of br east conserving treatment despite overall increased adoption. Medical Care, 40 (3), 181-189. Goodman S, Royall R. Eviden ce and scientific research. Am J Epi, 1988, 78, 1568-1574. Hewitt M, Simone J. (1999) Ensuring quality cancer care National Cancer Policy Board, Institute of Medicine. Washington, D.C.: National Academies Press. Iezzoni L. (2003) Risk Adjustment for Measur ing Health Care Outcomes (3 rd ed.). Chicago: Health Administration Press. Institute of Medicine. (2003). Unequal treatment: confronting racial and ethnic disparities in healthcare (B. Smedley, A. Stith & A. Nelson, Eds.), Committee on Understanding and Eliminating Racial and Ethnic Disparities in Health Care. Washington, DC: National Academies Press. Johantgen M, Coffey R, Harris D, Levy H, C linton J. (1995). Treating early-stage breast cancer: hospital characteristics associ ated with breast-conserving surgery. Am J Pub Health 85, 1432-4. Jones L, Chilton J. (2002) Rural health and women of color. Am J Pub Health 93 (4), 539-42. Joslyn, S. (2002) Racial differences in trea tment and survival from early-stage breast carcinoma. Cancer 95(8), 1759-66. Keppel, Kenneth G. (2004, February 20). Pe rsonnel correspondence with Tracey Perez Koehlmoos. Kingdon J. (1995) Agendas, Alternatives, and Public Policies. (2 nd ed.) New York: Addison Wesley Longman. Lantz P, Zemencuk J, Katz S. (2002) Is mastectomy overused? A call for an expanded research agenda. Health Serv Research, 37 (2), 417-32. Lazovich D, Solomon C, Thomas D, et al (1999) Breast conserva tion therapy in the United States following the 1990 Nationa l Institutes of Health Consensus Development Conference on the treatment of patients with early stage breast carcinoma. Cancer 86, 628-637. Lee-Feldstein A, Feldstein P, Buchmuelle r T, Katterhagen G. (2001) Breast cancer outcomes among older women. JGMI, 16, 189-199.

PAGE 80

69 Lichter, A, Lippman M, Danforth D, et al (1992) Mastectomy vers us breast-conserving therapy in the treatment of Stage I and II carcinoma of the breast: A randomized trial at the National Cancer Institute. J Clin Oncol 10, 976-983. Luther S, Studnicki J. (2001) Physician pr actice volume and alternative surgical treatment for breast cancer in Florida. Health Services Research 36 (6), 166178. Mandelblatt, J, Andrews H, Kerner J, Lauber A, Burnett W. (1991) Determinants of late stage diagnosis of breast and cervical cancer : the impact of age, race, social class and hospital type. AJPH 81, 646-50. Malin J. Kahn K, et al. (2002) Validity of cancer registry data for measuring the quality of breast cancer care. JNCI 94 (11), 835-44. Miller B, Hankey, B, Thomas, T. (2002) Impact of sociodemographic factors, hormone receptor status, and tumor grade on ethnic differences in tumor stage and size for breast cancer in US women. Am J Epi 155(6), 534-45. Morrow M. (2003) Breast conser ving therapy: who and why? J Surg Oncol 84: 55-56. Morrow M, White J, Moughan J, et al. ( 2001) Factors predicting the use of breastconserving therapy in stage I and II breast carcinoma. J Clin Oncol 19, 2254-62. Muss H, Hunter C, Wesley M, et al. (1992) Treatment plans for black and white women with stage II node-positive breast can cer: the national Cancer Institutes Black/White Cancer Survival Study experience. Cancer 70, 2460-7. National Cancer Institute (2002) Plans and priorities for cancer research, planning national agendas in dise ase specific research. Retrieved from: http://plans2002.cancer .gov/diseaselinks.htm Accessed National Institutes of Health Consensus Deve lopment Panel Consensus Statement. (1992) Treatment of early-stage breast cancer. J National Cancer Inst. Monogr 11, 1120. Nattinger A, Gottlieb M, Veum J, Yahnke D, Goodwin J. (1992). Geographic variation in the use of breast-conserving treatment for breast cancer. N Engl J Med, 326, 1102-7. Nattinger A, Kneusel R, Hoffman R, et al (2001) Relationship of distance from a radiotherapy facility and initial breast cancer treatment. J Natl Cancer Inst 93, 1344-6.

PAGE 81

70 Newman L, Kuerer H, Hunt K, Singh G, Ames F, Feig B, et al. (1999) Local recurrence and survival among black women with early-stage breast can cer treated with breast-conservation therapy or mastectomy. Ann Surg Oncol 6, 241-8. Osteen R, Karnell L. (1994). The National Cancer Data Base report on breast cancer. Cancer 73, 1994-2000. Parviz M, Cassel J, Kaplan B, Karp S, Ne ifeld J, Penberthy L, Bear H. (2003). Breast conservation therapy rates ar e no different in medically indigent versus insured patients with early stage breast cancer. J Surg Oncol 84, 57-62. Richardson L (2004, January 27). Personal interview. Richardson L, Schulman J, Sever L, Lee N, Coates R. (2001) Early-stage breast cancer treatment among medically underserved wome n diagnosed in a national screening program, 1992-1995. Breast Cancer Research and Treatment 69, 133-42. Ries L, Eisner M, Korsary C, Hankey B, Miller B, Clegg L, et. al (1998). Seer Cancer Statistics Review, 1973-1998. National Ca ncer Institute. Accessed on-line: http://seer.cancer.gov /publications/CSR 1979-1998/breast.pdf Roetzheim R, Gonzalez E, Ferrante, J, et al (2000) Effects of health insurance and race on breast carcinoma treatments and outcomes. Cancer 89 (11), 2202-13. Rose J, O'Tolle E, et al. (2000). Age diffe rences in care practices and outcomes for hospitalized patients with cancer. J Am Geriatric Society 48 (5, Suppl), S25-32. Rothman K, Greenland S (1998) Modern Epidemiology (2 nd ed.) Philadelphia: Lippincott, Williams & Wilkins. Sarrazin D, Le M, Arriagada R, et al. ( 1989) Ten-year results of a randomized trial comparing a conservative treatment to mastectomy in early breast cancer. Radiother Oncol 14, 177-184. Selker H, Griffith J, Patil S, Long W, D'Agostino R. (1995) A comparison of performance of mathematical predic tive methods for medical diagnosis: Identifying acute cardiac ischemia among emergency department patients. J Investigative Medicine, 43(5), 468-76. Shavers V, Brown M. (2002) Racial and ethn ic disparities in th e receipt of cancer treatment. JNCI 94(5), 334-357. Shinagawa, S. (2000) The excess burden of breast carcinoma in minority and medically underserved communities. Cancer Supplement, 88 (5), 1217-23.

PAGE 82

71 Studnicki J, Berndt D, Luther S, et al. (2004) Hispanic Heal th Status in Orange County, Florida. U.S. Bureau of the Census. (2000). Projections of the resi dent population by race, Hispanic origin and nativity: middle series, 2050-2070 http://www.census.gov/population/projec tions/nation/summary/np-t5-g.pdf accessed 6 December 2003. U.S. Department of Agriculture (2003) Rural-Urban Continuum. Retrieved 14 April 2004 from http://www.ers.usda.gov/Data/RuralUrbanContinuumCodes/ U.S. Department of Health and Human Services. (2000a). Healthy People 2010: Understanding and Improving Health 2 nd ed. Washington, DC: U.S. Government Printing Office. U.S. Department of Health and Human Services (2003). National Healthcare Disparities Report Retrieved 20 February 2004 fr om: www.qualitytools.ahrq.govdisparitiesreport.doc Van Dongen J, Bartelink H, Fentiman I, et al. (1992) Factors infl uencing local relapse and survival and results of salvage treat ment after breast-conserving therapy in operable breast cancer: EO RTC Trial1081, breast conservation compared with mastectomy in TNM Stage 1 and II breast cancer. Eur J Cancer 28A, 801-805. Velanovich V, Yood M, Bawle U, Nathanson S, Strand V, talpos G, et al. (1999) Racial differences in the presentation and surg ical management of breast cancer. Surgery, 125, 375-9. Veronesi U, Saccozzi R, DelVecchio M, et al. (1981) Comparing radical mastectomy with quadratectomy, axillary dissection a nd radiotherapy in patients with small cancers of the breast. N Engl J Med, 305, 6-11. Veronesi U, Slavadori B, Luini A, et al. (1995) Breast conservation is a safe method in patients with small cancer of the breast Long term results of three randomized trials on 1,973 patients. Eur J Cancer 31A, 1574-1579. Winchester D, Cox J. (1992). Standards for breast-conservation treatment. CA Cancer J Clin 42,134-176.

PAGE 83

72 Bibliography Agresti A. (1996) An Introduction to Categorical Data Analysis New York: John Wiley & Sons. American Joint Committee on Cancer. (2002) AJCC Cancer Staging Manual (6 th ed.) Ann Arbor, MI: Edwards Brothers. Kahn H, Sempos C. (1989) Statistical Methods in Epidemiology. New York: Oxford University Press. Kingdon J. (1995) Agendas, Alternatives, and Public Policies. (2 nd ed.) New York: Addison Wesley Longman. Neter J, Kutner M, Nachtsheim C, Wasserman W. (1996) Applied Linear Statistical Models (4 th ed.) Boston: WCB McGraw-Hill. Rosen P. (2001) Rosens Breast Pathology. (2 nd ed.) Philadelphia, PA: Lippincott, Williams & Wilkins. Rubin B. (2000) A Citizens Guide to Politics in America (2 nd ed.) Armonk, New York: M.E. Sharpe. Szeko M, Nieto F. (2000) Epidemiology Beyond the Basics. Gaithersburg, MD: Aspen.

PAGE 84

73 Appendices

PAGE 85

Appendix A. Summary of Logistic Regression Models for Breast Conserving Surgery Model Identity Covariates Included BLACK par. Est. BLACK OR OR 95% LL OR 95% UL BLACK P value Significant Covariates Neg2LogL df A BLACK -0.1947 0.8230 0.6770 1.0000 0.0506 9166.8580 1 A2 BLACK, MIDAGE, OLDAGE,VERYOLD -0.1805 0.8350 0.6860 1.0160 0.0722 MIDAGE, OLDAGE 9147.4550 4 A3 BLACK, NOIN, MCAID, MCARE -0.1414 0.8680 0.7120 1.0590 0.1624 NOIN, MCAID 9109.7890 4 A4 BLACK, SINGLE, OTHMAR -0.1514 0.8590 0.7060 1.0470 0.1324 SINGLE, OTHMAR 9154.5590 3 A6 BLACK, SMETU, NON METRO -0.1997 0.8190 0.6730 0.9960 0.0456 BLACK, NON METRO 9146.8810 3 A6A BLACK, NON METRO -0.2009 0.8180 0.6730 0.9950 0.0439 BLACK, NON METRO 9146.9200 2 B BLACK, NON METRO, MIDAGE, OLDAGE, VERYOLD -0.1865 0.8300 0.6810 1.0110 0.0636 NON METRO, MIDAGE, OLDAGE 9126.4310 5 C BLACK, NON METRO, MIDAGE, OLDAGE, VERYOLD, NOIN, MCAID, MCARE, SINGLE, OTHMAR -0.1056 0.9000 0.7360 1.1000 0.3034 NON METRO, MIDAGE, OLDAGE, NOIN, MCAID, SINGLE, OTHMAR 9066.0920 10 C3 BLACK, NON METRO, MIDAGE, NOIN, MCAID, SINGLE, OTHMAR -0.1097 0.8960 0.7340 1.0950 0.2825 NON METRO, NOIN, MCAID, SINGLE, OTHMAR (MIDAGE IS .08) 9079.9710 7 74

PAGE 86

Appendix A. (Continued) Model Identity Covariates Included BLACK par. Est. BLACK OR OR 95% LL OR 95% UL BLACK P value Significant Covariates Neg2LogL df C4 BLACK, NON METRO, MIDAGE, OLDAGE, VERYOLD, NOIN, MCAID, SINGLE, OTHMAR -0.1064 0.8990 0.7350 1.0990 0.2997 All except BLACK and VERYOLD 9066.7900 9 C5 NON METRO, MIDAGE, OLDAGE, VERYOLD, NOIN, MCAID, SINGLE, OTHMAR xx xx xx xx xx all except VERYOLD and OTHMAR barely (0.0573) 9067.8580 8 C6 BLACK, NON METRO, MIDAGE, OLDAGE, VERYOLD, NOIN, MCAID, SINGLE, OTHMAR -0.1591 0.8530 0.6920 1.0510 0.1359 NON METRO, MIDAGE, OLDAGE, MCAID, SINGLE, OTHMAR 8423.7990 9 C7 BLACK, NON METRO, MIDAGE, OLDAGE, VERYOLD, MCAID, SINGLE, OTHMAR -0.1630 0.8500 0.6890 1.0470 0.1261 All except VERYOLD 8424.3930 8 C8 BLACK, NON METRO, MIDAGE, OLDAGE, VERYOLD, MCAID, SINGLE, OTHMAR -0.1326 0.8760 0.7170 1.0690 0.1931 All except VERYOLD 9111.5920 8 C9 NON METRO, MIDAGE, OLDAGE, VERYOLD, MCAID, SINGLE, OTHMAR xx xx xx xx xx All except VERYOLD 9113.2700 7 75

PAGE 87

Appendix B. Comparison of Models for Breast Conserving Surgery Model BlackCovariate diff % change in Black Covariate Neg 2 L Diff Df diff Chisq Notes A to A2 -0.0142 0.0729 19.4030 3 0.0002 Age categorical (VERYOLD) better than crude A to A3 -0.0533 0.2738 57.0690 3 0.0000 Insurance better than crude, large change in covariate A to A4 0.3461 -1.7776 12.2990 2 0.0021 Marital better than crude, large change in covariate A to A6 0.0050 -0.0257 19.9770 2 0.0000 USDA better than crude A to A6A 0.0062 -0.0318 19.9380 1 0.0000 Just NON METRO better than crude A6 to A6A 0.0012 -0.0060 0.0390 1 0.8434 Model same with or without SMETU A6A to B -0.0144 0.0717 20.4890 3 0.0001 Adding age category (VERYOLD) to a model with BLACK and NON METRO improves fit. Medium change in covariate. A2 to B 0.0060 -0.0332 21.0240 1 0.0000 Adding NON METRO to a model with age category (VERYOLD) and BLACK improves fit. A2 to C -0.0749 0.4150 81.3630 6 0.0000 Adding all insurance and marital to age category (VERYOLD) and BLACK improves fit. Large change in covariate A3 to C -0.0358 0.2532 43.6970 6 0.0000 Adding age category (VERYOLD) to all insurance and all marital improves fit. Large change in covariate. A4 to C -0.0458 0.3025 88.4670 7 0.0000 Adding insurance, Non Metroity and age catgory (VERYOLD) improves the fit of a model with just BLACK and marital in it. A6A to C -0.0953 0.4744 80.8280 8 0.0000 Age cat (VERYOLD), all insurance, and marital improve the fit of a model with only BLACK and NON METRO in it. B to C -0.0809 0.4338 60.3390 5 0.0000 Adding insurance and marital improves the model with just age category (VERYOLD), NON METRO, BLACK. C4 to C -0.0008 0.0075 0.6980 1 0.4035 MEDICARE makes no difference in the C model C8 to C4 -0.0262 0.1976 44.8020 1 0.0000 It is better to exclude NOIN from model. C9 to C8 xx xx 1.6780 1 0.1952 BLACK makes no difference in the C8 model. 76

PAGE 88

Appendix C. Results of Interaction Term Modeling for Hypothesis One. Model Identity Covariates Included Notes J1 BLACK, NON METRO, MIDAGE, OLDAGE, VERYOLD, MCAID, SINGLE, OTHMAR, BLACK*MCAID interaction not significant (p= 0.5213) J2 BLACK, NON METRO, MIDAGE, OLDAGE, VERYOLD, MCAID, SINGLE, OTHMAR, BLACK*SINGLE, BLACK*OTHMAR interaction not significant (p= 0.5800, 0.8302) J3 BLACK, NON METRO, MIDAGE, OLDAGE, VERYOLD, MCAID, SINGLE, OTHMAR, BLACK*NON METRO interaction not significant (p=0.4141) J4 BLACK, NON METRO, MIDAGE, OLDAGE, VERYOLD, MCAID, SINGLE, OTHMAR, BLACK*MIDAGE, BLACK*OLDAGE, BLACK*VERYOLD interaction not significant (p= 0.8994, 0.9209, 0.2575) J5 BLACK, NON METRO, AGE, MCAID, SINGLE, OTHMAR, BLACK*AGE interaction not significant (p= 0.8607) J6 BLACK, NON METRO, MIDAGE, OLDAGE, VERYOLD, MCAID, SINGLE, OTHMAR, NON METRO*MIDAGE, NON METRO*OLDAGE, NON METRO*VERYOLD interaction not significant (p= 0.8215, 0.6184, 0.4965) J7 BLACK, NON METRO, MIDAGE, OLDAGE, VERYOLD, MCAID, SINGLE, OTHMAR, NON METRO*SINGLE, NON METRO*OTHMAR interaction not significant (p= 0.8047, 0.5991) J8 BLACK, NON METRO, MIDAGE, OLDAGE, VERYOLD, MCAID, SINGLE, OTHMAR, MIDAGE*SINGLE, OLDAGE*SINGLE, VERYOLD*SINGLE, MIDAGE*OTHMAR, OLDAGE*OTHMAR, VERYOLD*OTHMAR interaction not significant (p=0.7644, 0.2605, 0.1547, 0.3366, 0.7362, 0.5986) J9 BLACK, NON METRO, MIDAGE, OLDAGE, VERYOLD, MCAID, SINGLE, OTHMAR, NON METRO*MIDAGE, NON METRO*OLDAGE, NON METRO*VERYOLD, NON METRO*SINGLE, NON METRO*OTHMAR, SINGLE*MIDAGE, SINGLE*OLDAGE, SINGLE*VERYOLD, OTHMAR*MIDAGE, OTHMAR*OLDAGE, OTHMAR*VERYOLD, NON METRO*SINGLE*MIDAGE, NON METRO*SINGLE*OLDAGE, NON METRO*SINGLE*VERYOLD, NON METRO*OTHMAR*MIDAGE, NON METRO*OTHMAR*OLDAGE, NON METRO*OTHMAR*VERYOLD interactions not significant (p=0.7880, 0.8435, 0.9238, 0.5861, 0.2621, 0.6146, 0.2824, 0.1555, 0.3781, 0.9209, 0.4542, 0.3855, 0.7655, 0.9076, 0.8238, 0.2454, 0.2355) 77

PAGE 89

Appendix C. (Continued) Model Identity Covariates Included Notes J10 BLACK NON METRO MIDAGE OLDAGE VERYOLD MCAID SINGLE OTHMAR BLACK*NON METRO BLACK*MCAID NON METRO*MCAID BLACK*NON METRO*MCAID interactions not significant (p=0.3058, 0.5295, 0.5028, 0.7954) J11 BLACK NON METRO MIDAGE OLDAGE VERYOLD MCAID SINGLE OTHMAR NON METRO*MIDAGE NON METRO*OLDAGE NON METRO*VERYOLD NON METRO*SINGLE NON METRO*OTHMAR SINGLE*MIDAGE SINGLE*OLDAGE SINGLE*VERYOLD OTHMAR*MIDAGE OTHMAR*OLDAGE OTHMAR*VERYOLD NON METRO*SINGLE*MIDAGE NON METRO*SINGLE*OLDAGE NON METRO*SINGLE*VERYOLD NON METRO*OTHMAR*MIDAGE NON METRO*OTHMAR*OLDAGE NON METRO*OTHMAR*VERYOLDNON METRO*MIDAGE*SINGLE NON METRO*MIDAGE*OTHMAR NON METRO*OLDAGE*SINGLE NON METRO*OLDAGE*OTHMAR NON METRO*VERYOLD*SINGLE NON METRO*VERYOLD*OTHMAR interactions not significant (p=0.7880, 0.8435, 0.9238, 0.5861, 0.2621, 0.6146, 0.2824, 0.1555, 0.3781, 0.9209, 0.4542, 0.3855, 0.7655, 0.9076, 0.8238, 0.2454, 0.2355) J12 BLACK NON METRO MIDAGE OLDAGE VERYOLD MCAID SINGLE OTHMAR BLACK*MIDAGE BLACK*OLDAGE BLACK*VERYOLD BLACK*SINGLE BLACK*OTHMAR MIDAGE*SINGLE MIDAGE*OTHMAR OLDAGE*SINGLE OLDAGE*OTHMAR VERYOLD*SINGLE VERYOLD*OTHMAR BLACK*MIDAGE*SINGLE BLACK*MIDAGE*OTHMAR BLACK*OLDAGE*SINGLE BLACK*OLDAGE*OTHMAR BLACK*VERYOLD*SINGLE BLACK*VERYOLD*OTHMAR interactions not significant (p=0.9473, 0.5772, 0.6388, 0.8363, 0.5177, 0.8070, 0.2843, 0.2988, 0.9497, 0.1265, 0.4491, 0.9531, 0.6908, 0.8849, 0.3911, 0.3887, 0.6801) 78

PAGE 90

Appendix D. Summary of Logistic Regression for Receipt of Radiation Therapy Model Identity Covariates Included BLACK par. Est. BLACK OR OR 95% LL OR 95% UL BLACK P value Significant Covariates Neg2LogL df A BLACK -0.0866 0.9170 0.7190 1.1700 0.4856 6496.9150 1 A1 BLACK, AGE -0.0771 0.9260 0.7250 1.1830 0.5376 6496.4980 2 A2 BLACK, MIDAGE, OLDAGE,VERYOLD -0.0577 0.9440 0.7380 1.2070 0.6456 MIDAGE, OLDAGE 6445.0810 4 A3 BLACK, NOIN, MCAID, MCARE -0.0582 0.9430 0.7380 1.2060 0.6420 6492.3670 4 A4 BLACK, SINGLE, OTHMAR -0.0436 0.9570 0.7490 1.2240 0.7280 SINGLE, OTHMAR 6487.4880 3 A6 BLACK, SMETU, NON METRO -0.3751 0.9500 0.7440 1.2140 0.6821 SMETU highly sig, NON METRO close (p= 0.0801) 6456.3190 3 A6A BLACK, NON METRO -0.0945 0.9100 0.7130 1.1610 0.4470 NON METRO 6489.0040 2 B BLACK, NON METRO, MIDAGE, OLDAGE, VERYOLD -0.0659 0.9360 0.7320 1.1980 0.5999 NON METRO,MIDAGE, OLDAGE 6435.9530 5 C BLACK, NON METRO, MIDAGE, OLDAGE, VERYOLD, NOIN, MCAID, MCARE, SINGLE, OTHMAR -0.0239 0.9760 0.7610 1.2530 0.8514 NON METRO,OLDAGE, VERYOLD, SINGLE 6428.4930 10 C3 BLACK, NON METRO, MIDAGE, NOIN, MCAID, SINGLE, OTHMAR -0.0434 0.9580 0.7480 1.2260 0.7312 NON METRO,SINGLE, OTHMAR 6475.9230 7 79

PAGE 91

Appendix D. (Continued) Model Identity Covariates Included BLACK par. Est. BLACK OR OR 95% LL OR 95% UL BLACK P value Significant Covariates Neg2LogL df C4 BLACK, NON METRO, MIDAGE, OLDAGE, VERYOLD, NOIN, MCAID, SINGLE, OTHMAR -0.0227 0.9780 0.7620 1.2550 0.8584 NON METRO, MIDAGE, OLDAGE 6429.0240 9 C5 NON METRO, MIDAGE, OLDAGE, VERYOLD, NOIN, MCAID, SINGLE, OTHMAR XX XX XX XX XX NON METRO, MIDAGE, OLDAGE 6429.0560 8 C6 BLACK, NON METRO, MIDAGE, OLDAGE, VERYOLD, NOIN, MCAID, SINGLE, OTHMAR -0.0097 0.9900 0.7640 1.2840 0.9415 NON METRO, MIDAGE, OLDAGE, SINGLE, OTHMAR 6061.0090 9 C7 BLACK, NON METRO, MIDAGE, OLDAGE, VERYOLD, MCAID, SINGLE, OTHMAR -0.0154 0.9850 0.7600 1.2760 0.9074 NON METRO, MIDAGE, OLDAGE, SINGLE 6061.9960 8 C8 BLACK, NON METRO, MIDAGE, OLDAGE, VERYOLD, MCAID, SINGLE, OTHMAR -0.0315 0.9690 0.7550 1.2430 0.8044 NON METRO, MIDAGE, OLDAGE, SINGLE, OTHMAR 6430.3970 8 C9 NON METRO, MIDAGE, OLDAGE, VERYOLD, MCAID, SINGLE, OTHMAR XX XX XX XX XX NON METRO, MIDAGE, OLDAGE, SINGLE 6430.4580 7 80

PAGE 92

Appendix D. (Continued) Model Identity Covariates Included BLACK par. Est. BLACK OR OR 95% LL OR 95% UL BLACK P value Significant Covariates Neg2LogL df C10 BLACK MIDAGE OLDAGE VERYOLD MCAID SMETU NON METRO SINGLE OTHMAR 0.0072 1.0070 0.7840 1.2930 0.9551 MIDAGE, OLDAGE,SMETU, NON METRO (SINGLE IS CLOSE AT 0.0915) 6402.0040 9 J15 BLACK MIDAGE OLDAGE VERYOLD MCAID SMETU NON METRO BLACK*MCAID XX XX XX XX XX MIDAGE, OLDAGE SMETU NON METRO BLACK* MCAID 6401.6970 8 J16 BLACK MIDAGE OLDAGE VERYOLD MCAID SMETU NON METRO SINGLE OTHMAR BLACK*MCAID XX XX XX XX XX MIDAGE, OLDAGE SMETU NON METRO BLACK* MCAID (SINGLE IS CLOSE AT 0.0895) 6397.4190 10 J18 BLACK MIDAGE OLDAGE VERYOLD MCAID SMETU NON METRO SINGLE BLACK*MCAID XX XX XX XX XX MIDAGE, OLDAGE SMETU NON METRO BLACK* MCAID 6399.5440 9 81

PAGE 93

Appendix D. (Continued) Model Identity Covariates Included BLACK par. Est. BLACK OR OR 95% LL OR 95% UL BLACK P value Significant Covariates Neg2LogL df J19 BLACK MIDAGE OLDAGE VERYOLD MCAID SMETU NON METRO OTHMAR BLACK*MCAID XX XX XX XX XX MIDAGE, OLDAGESMETU NON METRO BLACK* MCAID 6400.3280 9 J20 BLACK MIDAGE OLDAGE VERYOLD MCAID SMETU NON METRO SINGLE OTHMAR BLACK*MCAID XX XX XX XX XX MIDAGE, OLDAGESMETU BLACK* MCAID 6024.1040 10 82

PAGE 94

83 Appendix E. Comparison of Models for Receipt of Radiation Therapy Model Black Covariate diff % change in Black Covariate Neg 2 L Diff Df diff Chisq A to A1 -0.0095 0.1097 0.4170 1 0.5184 A to A2 -0.0289 0.3337 51.8340 3 0.0000 A to A3 -0.0284 0.3279 4.5480 3 0.2080 A to A4 0.1302 -1.5035 9.4270 2 0.0090 A to A5 0.0202 -0.2333 10.6430 1 0.0011 A to A6 0.2885 -3.3314 40.5960 2 0.0000 A to A6A 0.0079 -0.0912 7.9110 1 0.0049 A6 to A6A -0.2806 0.7481 32.6850 1 0.0000 A6A to B -0.0286 0.3026 53.0510 3 0.0000 A2 to B 0.0082 -0.1421 9.1280 1 0.0025 A6A to B1 0.4168 -4.4106 14.7130 2 0.0006 A2 to C -0.0338 0.5858 16.5880 6 0.0109 A3 to C -0.0343 0.5893 63.8740 6 0.0000 A4 to C -0.0197 0.4518 58.9950 7 0.0000 A6A to C -0.0706 0.7471 60.5110 8 0.0000 B to C -0.0420 0.6373 7.4600 5 0.1886 A3 to C1 -0.0557 0.9577 30.1820 5 0.0000 A4 to C1 -0.0411 0.9436 25.3030 6 0.0003 A6A to C1 -0.0920 0.9740 26.8190 7 0.0004 B1 to C1 -0.5088 0.9952 12.1060 5 0.0334 C4 to C 0.0012 -0.0529 0.5310 1 0.4662 C8 to C4 -0.0088 0.2794 1.3730 1 0.2413 C9 to C8 xx xx 0.0610 1 0.8049 C10 to J16 xx xx 4.5850 1 0.0323 J15 to J16 xx xx 4.2780 2 0.1178 J18 to J16 xx xx 2.1250 1 0.1449 J19 to J16 xx xx 2.9090 1 0.0881

PAGE 95

84 Appendix F. Results of Interaction Term Modeling for Breast Conserving Surgery Population. Model Identity Covariates Included Notes J1 BLACK, NON METRO, MIDAGE, OLDAGE, VERYOLD, MCAID, SINGLE, OTHMAR, BLACK*MCAID interaction significant (p= 0.0384) J2 BLACK, NON METRO, MIDAGE, OLDAGE, VERYOLD, MCAID, SINGLE, OTHMAR, BLACK*SINGLE, BLACK*OTHMAR interaction not significant (p=0.2970, 0.8458) J3 BLACK, NON METRO, MIDAGE, OLDAGE, VERYOLD, MCAID, SINGLE, OTHMAR, BLACK*NON METRO interaction not significant (p= 0.9366) J4 BLACK, NON METRO, MIDAGE, OLDAGE, VERYOLD, MCAID, SINGLE, OTHMAR, BLACK*MIDAGE, BLACK*OLDAGE, BLACK*VERYOLD significance of interaction debatable (p=0.5031, 0.1947, 0.0755) J5 BLACK, NON METRO, AGE, MCAID, SINGLE, OTHMAR, BLACK*AGE significance of interaction debatable (p= 0.1855) J6 BLACK, NON METRO, MIDAGE, OLDAGE, VERYOLD, MCAID, SINGLE, OTHMAR, NON METRO*MIDAGE, NON METRO*OLDAGE, NON METRO*VERYOLD interaction not significant (p=0.8467, 0.7225, 0.2678) J7 BLACK, NON METRO, MIDAGE, OLDAGE, VERYOLD, MCAID, SINGLE, OTHMAR, NON METRO*SINGLE, NON METRO*OTHMAR interaction not signficant (p=0.6892, 0.9522) J8 BLACK, NON METRO, MIDAGE, OLDAGE, VERYOLD, MCAID, SINGLE, OTHMAR, MIDAGE*SINGLE, OLDAGE*SINGLE, VERYOLD*SINGLE, MIDAGE*OTHMAR, OLDAGE*OTHMAR, VERYOLD*OTHMAR interaction not significant (p=0.2091, 0.1851, 0.2944, 0.7146, 0.9508, 0.9901)

PAGE 96

Appendix F. (Continued) 85 J9 BLACK, NON METRO, MIDAGE, OLDAGE, VERYOLD, MCAID, SINGLE, OTHMAR, NON METRO*MIDAGE, NON METRO*OLDAGE, NON METRO*VERYOLD, NON METRO*SINGLE, NON METRO*OTHMAR, SINGLE*MIDAGE, SINGLE*OLDAGE, SINGLE*VERYOLD, OTHMAR*MIDAGE, OTHMAR*OLDAGE, OTHMAR*VERYOLD, NON METRO*SINGLE*MIDAGE, NON METRO*SINGLE*OLDAGE, NON METRO*SINGLE*VERYOLD, NON METRO*OTHMAR*MIDAGE, NON METRO*OTHMAR*OLDAGE, NON METRO*OTHMAR*VERYOLD interaction not significant (p=0.5599, 0.5893, 0.1738, 0.4946, 0.9543, 0.1946, 0.1482, 0.2736, 0.4871, 0.7053, 0.7313, 0.7143, 0.4182, 0.9846, 0.9533, 0.9536, 0.9515) J10 BLACK NON METRO MIDAGE OLDAGE VERYOLD MCAID SINGLE OTHMAR BLACK*NON METRO BLACK*MCAID NON METRO*MCAID BLACK*NON METRO*MCAID interaction between BLACK and MCAID interesting, but whole model is not significant (p=0.9306, 0.0391, 0.7144, 0.6676) J11 BLACK NON METRO MIDAGE OLDAGE VERYOLD MCAID SINGLE OTHMAR NON METRO*MIDAGE NON METRO*OLDAGE NON METRO*VERYOLD NON METRO*SINGLE NON METRO*OTHMAR SINGLE*MIDAGE SINGLE*OLDAGE SINGLE*VERYOLD OTHMAR*MIDAGE OTHMAR*OLDAGE OTHMAR*VERYOLD NON METRO*SINGLE*MIDAGE NO N METRO*SINGLE*OLDAGE NON METRO*SINGLE*VERYOLD NON METRO*OTHMAR*MIDAGE NON METRO*OTHMAR*OLDAGE NON METRO*OTHMAR*VERYOLDNON METRO*MIDAGE*SINGLE NON METRO*MIDAGE*OTHMAR NON METRO*OLDAGE*SINGLE NON METRO*OLDAGE*OTHMAR NON METRO*VERYOLD*SINGLE NON METRO*VERYOLD*OTHMAR interaction not significant (p=0.5599, 0.5893, 0.1738, 0.4946, 0.9543, 0.1946, 0.1482, 0.2736, 0.4871, 0.7053, 0.7313, 0.7143, 0.4182, 0.9846, 0.9533, 0.9536, 0.9515)

PAGE 97

Appendix F. (Continued) 86 J12 BLACK NON METRO MIDAGE OLDAGE VERYOLD MCAID SINGLE OTHMAR BLACK*MIDAGE BLACK*OLDAGE BLACK*VERYOLD BLACK*SINGLE BLACK*OTHMAR MIDAGE*SINGLE MIDAGE*OTHMAR OLDAGE*SINGLE OLDAGE*OTHMAR VERYOLD*SINGLE VERYOLD*OTHMAR BLACK*MIDAGE*SINGLE BLACK*MIDAGE*OTHMAR BLACK*OLDAGE*SINGLE BLACK*OLDAGE*OTHMAR BLACK*VERYOLD*SINGLE BLACK*VERYOLD*OTHMAR interaction not significant (p=0.9052, 0.9718, 0.9645, 0.1697, 0.8731, 0.5617, 0.6364, 0.5831, 0.9742, 0.4248, 0.9667, 0.3551, 0.6002, 0.1902, 0.7275, 0.9655, 0.9676) J13 BLACK NOIN MCAID MCARE BLACK*NOIN BLACK*MCAID BLACK*MCARE interaction compelling, especially for MCAID (p=0.2508, 0.0473, 0.2814) J15 BLACK MIDAGE OLDAGE VERYOLD MCAID SMETU NON METRO BLACK*MCAID interaction significant (p= 0.0356), overall better than J14 J16 BLACK MIDAGE OLDAGE VERYOLD MCAID SMETU NON METRO SINGLE OTHMAR BLACK*MCAID interaction significant (p= 0.0380), single and othmar may contribute to model. J17 BLACK MIDAGE OLDAGE VERYOLD MCAID SMETU NON METRO SINGLE OTHMAR BLACK*MCAID BLACK*SINGLE BLACK*OTHMAR SINGLE*MCAID OTHMAR*MCAID BLACK*MCAID*SINGLE BLACK*MCAID*OTHMAR Interactions not significnant (p=0.4454, 0.3418, 0.8943, 0.4037, 0.3621, 0.7381, 0.9521) J18 BLACK MIDAGE OLDAGE VERYOLD MCAID SMETU NON METRO SINGLE BLACK*MCAID J16 with just SINGLE. Interaction significant (p= 0.0343), J19 BLACK MIDAGE OLDAGE VERYOLD MCAID SMETU NON METRO OTHMAR BLACK*MCAID J16 with just OTHMAR. Interaction significant (p= 0.0387),

PAGE 98

About the Author Tracey Lynn Perez Koehlmoos received a Bachelor of Arts cum laude from Davidson College in 1991 and a Master of Hea lth Administration from the University of South Florida, College of Public Health in 2002. She served as an Administrative Fellow at Tampa General Hospital in 2002-2003 wh ile simultaneously completing her course work as a doctoral student at th e University of South Florida. Mrs. Perez Koehlmoos is a former Army officer and mother of three who won numerous awards while at the College of Public Health including the nationally distinguished American College of Health care Executives Hill-Rom Management Essay Competition in both 2001 and 2002. She completed her dissertation research and writing while living in Kathmandu, Nepal.


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 001681164
003 fts
005 20060215104653.0
006 m||||e|||d||||||||
007 cr mnu|||uuuuu
008 051012s2005 flu sbm s000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0001085
035
SFE0001085
040
FHM
c FHM
090
2005
1 100
Koehlmoos, Tracey.
0 245
Racial disparities in breast cancer surgical treatment and radiation therapy use
h [electronic resource] /
by Tracey Koehlmoos.
260
[Tampa, Fla.] :
b University of South Florida,
2005.
502
Thesis (Ph.D.)--University of South Florida, 2005.
504
Includes bibliographical references.
516
Text (Electronic thesis) in PDF format.
538
System requirements: World Wide Web browser and PDF reader.
Mode of access: World Wide Web.
500
Title from PDF of title page.
Document formatted into pages; contains 98 pages.
520
ABSTRACT: This study explores the relationship between race and surgical treatment and radiation therapy use for localized breast cancer patients in the state of Florida in 2001. The study will be useful in raising awareness of the relationship between Black race and appropriate breast cancer treatment within the Florida Cancer Data System. The Healthy People 2010 initiatives call to eliminate racial disparities and the high placement of breast cancer on the national research agenda make this study timely and insightful for health policymakers, clinicians and other health researchers.Also, the study evaluates the effect of other health system and patient related factors such as insurance provider and rural versus urban residence, to the appropriate use of cancer therapy in order to present an up-to-date and accurate picture of the quality of breast cancer care for women in the state of Florida.The study used multivariate logistic regression modeling and chi-square distribution to compare models in order to disentangle the effects of age, rural residence, marital status and primary health insurance provider from race and to determine how these factors influenced breast conserving surgery versus mastectomy use.Further, the second research question exclusively focused on the population that received breast conserving surgery in order to examine the impact of race and the other covariates as explanatory measures of appropriate receipt of radiation therapy.The first hypothesis found that there was no statistically significant difference between Black and White women in terms of receipt of breast conserving surgery for treatment of localized breast cancer. The second hypothesis, which focused on appropriate receipt of radiation therapy following breast conserving surgery, found that there was a statistically significant interaction between Black race and Medicaid as primary health insurance provider.The study concludes by examining possible areas of improvement in data collection in the State of Florida.
590
Adviser: Dr. James Studnicki.
Co-adviser:
653
Insurance.
Health access.
Women's health.
Lumpectomy.
Rurality.
690
Dissertations, Academic
z USF
x Health Policy and Management
Doctoral.
773
t USF Electronic Theses and Dissertations.
4 856
u http://digital.lib.usf.edu/?e14.1085