<%BANNER%>

Baseline Health Related Quality of Life As Confounder in Observational Research for Elderly Men with Localized Prostate ...

Permanent Link: http://ufdc.ufl.edu/UFE0042698/00001

Material Information

Title: Baseline Health Related Quality of Life As Confounder in Observational Research for Elderly Men with Localized Prostate Cancer
Physical Description: 1 online resource (123 p.)
Language: english
Creator: HE,JINGHUA
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2011

Subjects

Subjects / Keywords: CONFOUNDING -- EFFECTIVENESS -- OBSERVATIONAL -- ONCOLOGY -- PROSTATE -- SURVIVAL
Pharmaceutical Outcomes and Policy -- Dissertations, Academic -- UF
Genre: Pharmaceutical Sciences thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Prostate cancer is the most concerning non-skin malignancy in elderly men. Most prostate cancers are diagnosed at a localized stage. Definitive treatment includes radical prostatectomy and radiation therapy. Conservative treatment includes expectant management and hormonal therapy. Patient overall health status is a key consideration in treatment planning. Selection of relatively healthy patients to receive definitive treatment results in strong confounding in observational study regarding the treatment effectiveness. Health related quality of life (HRQOL) measures multiple health dimensions. We conducted a cross-sectional study to examine the associations between a comprehensive range of baseline HRQOL variables and primary treatment selection in elderly men diagnosed with localized prostate cancer. The primary data source was the Cancer of Prostate Strategic Urologic Research Endeavor (CaPSURE). Three types of HRQOL were included, which were Karnofsky performance score, Medical Outcome Short-Form 36, and UCLA Prostate Cancer Index. Radical prostatectomy was found to be strongly associated with higher scores for all the HRQOL variables than the other primary treatments. Radiation therapy was moderately associated with higher HRQOL scores than both of the conservative treatments. The differences between expectant management and hormonal therapy were minor. Several HRQOL variables, particularly those related to physical functioning, remained associated with definitive treatment, after controlling for the baseline sociodemographic, cancer characteristic and clinical factors. We further investigated the confounding effect of the baseline HRQOL in elderly men diagnosed with low- or intermediate-risk prostate cancer. Both of the propensity score and weighting methods were used to control confounding. Radical prostatectomy and radiation therapy were found to be strongly associated with lower all-cause mortality than conservative treatment, after adjusting for commonly considered confounders. The associations were attenuated, but remained pronounced, after additional adjustment for the baseline HRQOL variables. Further analyses showed that the survival benefit was largely due to an implausibly lower risk of death from causes other than prostate cancer, indicating the persistence of substantial residual confounding. Additional adjustment for the baseline HRQOL variables is desirable, but insufficient, to rule out confounding bias. Randomized trial is warranted to evaluate the efficacy of primary treatment for localized prostate cancer.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by JINGHUA HE.
Thesis: Thesis (Ph.D.)--University of Florida, 2011.
Local: Adviser: Hartzema, Abraham G.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2013-04-30

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2011
System ID: UFE0042698:00001

Permanent Link: http://ufdc.ufl.edu/UFE0042698/00001

Material Information

Title: Baseline Health Related Quality of Life As Confounder in Observational Research for Elderly Men with Localized Prostate Cancer
Physical Description: 1 online resource (123 p.)
Language: english
Creator: HE,JINGHUA
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2011

Subjects

Subjects / Keywords: CONFOUNDING -- EFFECTIVENESS -- OBSERVATIONAL -- ONCOLOGY -- PROSTATE -- SURVIVAL
Pharmaceutical Outcomes and Policy -- Dissertations, Academic -- UF
Genre: Pharmaceutical Sciences thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Prostate cancer is the most concerning non-skin malignancy in elderly men. Most prostate cancers are diagnosed at a localized stage. Definitive treatment includes radical prostatectomy and radiation therapy. Conservative treatment includes expectant management and hormonal therapy. Patient overall health status is a key consideration in treatment planning. Selection of relatively healthy patients to receive definitive treatment results in strong confounding in observational study regarding the treatment effectiveness. Health related quality of life (HRQOL) measures multiple health dimensions. We conducted a cross-sectional study to examine the associations between a comprehensive range of baseline HRQOL variables and primary treatment selection in elderly men diagnosed with localized prostate cancer. The primary data source was the Cancer of Prostate Strategic Urologic Research Endeavor (CaPSURE). Three types of HRQOL were included, which were Karnofsky performance score, Medical Outcome Short-Form 36, and UCLA Prostate Cancer Index. Radical prostatectomy was found to be strongly associated with higher scores for all the HRQOL variables than the other primary treatments. Radiation therapy was moderately associated with higher HRQOL scores than both of the conservative treatments. The differences between expectant management and hormonal therapy were minor. Several HRQOL variables, particularly those related to physical functioning, remained associated with definitive treatment, after controlling for the baseline sociodemographic, cancer characteristic and clinical factors. We further investigated the confounding effect of the baseline HRQOL in elderly men diagnosed with low- or intermediate-risk prostate cancer. Both of the propensity score and weighting methods were used to control confounding. Radical prostatectomy and radiation therapy were found to be strongly associated with lower all-cause mortality than conservative treatment, after adjusting for commonly considered confounders. The associations were attenuated, but remained pronounced, after additional adjustment for the baseline HRQOL variables. Further analyses showed that the survival benefit was largely due to an implausibly lower risk of death from causes other than prostate cancer, indicating the persistence of substantial residual confounding. Additional adjustment for the baseline HRQOL variables is desirable, but insufficient, to rule out confounding bias. Randomized trial is warranted to evaluate the efficacy of primary treatment for localized prostate cancer.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by JINGHUA HE.
Thesis: Thesis (Ph.D.)--University of Florida, 2011.
Local: Adviser: Hartzema, Abraham G.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2013-04-30

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2011
System ID: UFE0042698:00001


This item has the following downloads:


Full Text

PAGE 1

1 BASELINE HEALTH RELATED QUALITY OF LIFE AS CONFOUNDER IN OBSERVATIONAL RESEARCH FOR ELDERLY MEN WITH LOCALIZED PROSTATE CANCER By JINGHUA HE A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2011

PAGE 2

2 2011 Jinghua He

PAGE 3

3 To my beloved family

PAGE 4

4 ACKNOWLEDGMENTS I would like to express my deepest gratitude to my advisor, Dr. Abraham Hartzema for his supervision and continuous support throughout my doctorate training. I sincerely thank the other supervisory committee members, Drs. Richard Segal, Almut Winterstein, Barbette Brumback, and Philipp Dahm for their gu idance throughout my dissertation work I would further extend my a ppreciation to all the faculty staff and students in the Department of Pharmaceutical Outcomes and Policy for their supports and friendships. S pecial thanks go to Dr. Joseph Dela ne y, w ho provided valuable technical advices to this dissertat ion work I would gratefully acknowledge the research team of Ca ncer of the P rostate S trategic U rologic R esearch E ndeavor (CaPSURE) in the Department of Urology University of California, San Francisco I would thank Dr. Peter Carrol l for kindly offe ring me access to the CaPSURE data. I would thank Dr. Natalia Sadetsky for facilitating data extraction. Finally, I would thank my parents, my paren ts in laws, and my wife for their unlimited love and encour agement throughout the years of my graduate study in the United States

PAGE 5

5 TABLE OF CONTENTS ACKNOWLEDGMENTS ...............................................................................................................4 page LIST OF TABLES ...........................................................................................................................7 LIST OF FIGURES .........................................................................................................................9 LIST OF ABBREVIATIONS ........................................................................................................10 ABSTRACT ...................................................................................................................................12 CHAPTER 1 INTRODUCTION ..................................................................................................................14 Background .............................................................................................................................14 Study Rationale .......................................................................................................................15 Health Related Q uality of Life (HRQOL) and Cancer Treatment Selection ..................15 The Confounding Effect of Baseline HRQOL ................................................................15 Purpose of the Study ...............................................................................................................17 Research Questions and Hypotheses ......................................................................................19 Research Question 1 ........................................................................................................19 Research Hypothesis 1 ....................................................................................................19 Research Question 2 ........................................................................................................19 Research Hypothesis 2 ....................................................................................................19 2 LITERATURE REVIEW .......................................................................................................20 Epidemiology of Prostate Cancer ...........................................................................................20 Primary Treatment Selection for Localized Prostate Cancer .................................................21 Primary Treatment Options .............................................................................................21 Conceptual Model for Cancer Treatment Decision Making ...........................................22 Factors influencing physician preference .................................................................22 Factors influencing patient preference .....................................................................25 Physician patient interaction ....................................................................................27 Theoretical Framework for Confounding ...............................................................................29 Confounding Definitions .................................................................................................29 Classical definition ...................................................................................................29 Collapsibility definition ............................................................................................30 Counterfactual definition ..........................................................................................31 Confounding and selection bias ...............................................................................32 The Rationale of Baseline HRQOL as Confounder ........................................................33

PAGE 6

6 3 METHODOLOGY .................................................................................................................37 Data Source .............................................................................................................................37 Study Design and Sam ple Selection .......................................................................................38 Descriptive Analysis ...............................................................................................................40 Multivariate Statistical Analysis .............................................................................................40 Missing Data Imputation .................................................................................................40 Multinom ial Logistic Regression ....................................................................................42 Cox Proportional Hazard Model .....................................................................................43 Informative Censoring Adjustment .................................................................................44 Propensity Score Method ................................................................................................45 Weightin g Methods .........................................................................................................46 4 RESULTS ...............................................................................................................................49 Association between HRQOL and Primary Treatment ..........................................................49 Descriptive Analysis ........................................................................................................49 Crude Associations between Baseline Variables and Primary Treatment ......................51 Adjusted Associations between Baseline HRQOL and Primary Treatment ...................52 The Confounding Effect of Baseline HRQOL .......................................................................55 Descriptive and Univariate Analysis ...............................................................................55 Propensity Score Comparisons ........................................................................................57 Confounding Balance by Propensity Score .....................................................................58 Propensity Score Quintile Specific Hazard Ratios ..........................................................59 Confounding Effect of Baseline HRQOL .......................................................................59 Definitive treatment vs. conservative treatment .......................................................60 Radical prostatectomy vs. conservative treatment ...................................................61 Radiation therapy vs. conservative treatment ..........................................................63 5 DISCUSSION .......................................................................................................................105 Association between HRQOL and Primary Treatment ........................................................105 The Confounding Effect of Baseline HRQOL .....................................................................107 Effect Modification ...............................................................................................................109 Comparison with Previous Studies .......................................................................................111 Limitations ............................................................................................................................113 Summary and Conclusions ...................................................................................................114 LIST OF REFERENCES .............................................................................................................116 BIOGRAPHICAL SKETCH .......................................................................................................123

PAGE 7

7 LIST OF TABLES Table page 41 Baseline sociodemographic variable distributions in Cancer of the Prostate Strategic Urologic Research Endeavor ( CaPSURE ) elderly men with localized prostate cancer ....65 42 Baseline clinical variable distributions in CaPSURE elderly men with localized prostate cancer ...................................................................................................................66 43 Baseline health related quality of life ( HRQOL ) variable distributions in CaPSURE elderly men with localized prostate cancer ........................................................................68 44 Crude estimates of associations between baseline sociodemographic variables and primary treatment ...............................................................................................................71 45 Crude estimates of associations between baseline clinical variables and primary treatment ............................................................................................................................73 46 Crude estimates of associations between baseline HRQOL variables and primary treatment ............................................................................................................................75 47 Estimates of associations between baseline HRQOL variables and primary treatment, adjusting for baseline sociodemographic and clinical variables ........................................77 48 Estimates of adjusted associations between baseline variables and primary treatment .....78 49 Baseline sociodemographic variable distributions in CaPSURE elderly men with low or intermediate risk prostate cancer ..........................................................................81 410 Baseline clinical variable distributions in CaPSURE elderly men with low or intermediate risk prostate cancer .......................................................................................82 411 Baseline HRQOL variable distributions in CaPSURE elderly men with low or intermediate risk prostate cancer .......................................................................................84 412 Crude associations between baseline sociodemographic variables and 10year survival outcomes ..............................................................................................................85 413 Crude associations between baseline clinical variables and 10year survival outcomes ............................................................................................................................86 414 Crude associations between baseline HRQOL variables and 10year survival outcomes ............................................................................................................................87 415 Associations between baseline HRQOL variables and primary treatment, conditioning on propensity scores: definitive treatment vs. conservative treatment .........88

PAGE 8

8 416 Associations between baseline HRQOL variables and primary treatment, conditioning on propensity scores: radical prostatectomy v s. co nservative treatment ......89 417 Associations between baseline HRQOL variables and primary treatment, conditioning on propensity scores: r adiation therapy vs. conservative treatment .............90 418 Propensity score quintile specific hazard ratios for all cause mortality before additional adjustment of the baseline HRQOL variables ..................................................91 419 Propensity score quintile specific hazard ratios for all cause mortality after additional adjustment of the baseline HRQOL variables ..................................................91 420 Adjusted associations between primary treatment and 10year mor tality outcomes: definitive treatment vs. conservative treatment .................................................................92 421 Adjusted associations between primary treatment an d 10year mortality outcomes: radical prostatectomy vs. conservative treatment ..............................................................93 422 Adjusted associations between primary treatment and 10year mortality outcomes: r adiaton therapy vs. conservative treatment .......................................................................94

PAGE 9

9 LIST OF FIGURES Figure page 21 Conceptual model for patient and physician cancer treatment selection. ..........................35 22 A directed acyclic gram under which the classic confounder cr iteria fail .........................36 41 Propensity score distributions for definitive treatment vs.conservative treatment ............95 42 Propensity score distributions for radical prostatectomy vs. conservative treatment ........95 43 Propensity score distributionsfor radiation therapy vs. conservative treatment ................96 44 Log log survival curves: definitive treatment vs. conservative treatment .........................97 45 Log log survival curves: radical prostatectomy vs. conservative treatment ......................98 46 Log log surviva l curves: radiation therapy vs. conservative treatment .............................99 47 Adjusted hazard ratios: definitive treatment vs. conservative treatment .........................100 48 Adjusted hazard ratios : radical prostatectomy vs. conservative treatment ......................100 49 Adjusted hazard ratios : radiation therapy vs. conservative treatment .............................101 410 Crude and weighted survival curves: definitive treatment vs. conservative treatment ....102 411 Crude and wei ghted survival curves: radical prostatectomy vs. conservative treatment ..........................................................................................................................103 412 Crude and wei ghted survival curves: radiation therapy vs. conservative treatment ........104

PAGE 10

10 LI ST OF ABBREVIATIONS BB Bowel bother BF Bowel function BP Bodily pain CaPSURE Cancer of Prostate Strategic Urologic Research Endeavor CGA C omprehensive geriatric assessment CI Confidence interval CT Conservative treatment DT Definitive treatment EM Expectant m anagement GH General health HS High school HT Hormonal therapy IPCW Inverse probability of censoring weighting IPTW Inverse probability of treatment weighting IQR Interquartile range KPS Karnofsky performance score MH Mental health NCCN National Comprehensive Cancer Network PCI UCLA prostate cancer index PF Physical functioning PS Propensity score RE Role emotional RO Role physical RP Radical prostatectomy

PAGE 11

11 RT Radiation therapy SB Sexual bother SC Some college SD Standard deviation SF Sexual fun ction SF 36 Medical outcome short form 36 SO Social functioning UB Urinary bother UF Urinary function VT Vitality

PAGE 12

12 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirem ents for the Degree of Doctor of Philosophy BASELINE HEALTH RELATED QUALITY OF LIFE AS CONFOUNDER IN OBSERVATIONAL RESEAR CH FOR ELDERLY MEN W ITH LOCALIZED PROSTATE CANCER By Jinghua He May 2011 Chair: Abraham Hartzema Major: Pharmaceutical Outcomes and Policy Prostate cancer is the most concerning non skin malignancy in elderly men. Most prostate cancer s are diagnosed at a localized stage. Definitive treatment includes radical prostatectomy and radiation therapy Conservative treatment includes expectant management and hormonal therapy. Patient o verall health status is a key consideration in treatment planning Selection of relatively healthy patients to receive definitive treatment results in strong confounding in observational study regarding the trea tment effectiveness. Health related quality of life (HRQOL) measures multiple health dimensions. We conducted a cross sectional study to examine the associations between a comprehensive range of baseline HRQOL variables and primary treatment selection in e lderly men diagnosed with localized prostate cancer. The primary data source wa s the Cancer of Prostate Strategic Urologic Research Endeavor (CaPSURE) Three types of HRQOL were included, which were Karnofsky performance score, Medical Outcome S hort F orm 36, and UCLA Prostate Cancer Index. R adical prostatectomy was found to be strongly associated with high er scores for all the HRQOL variables than the other primary treatments. R adiation therapy was moderately associated with high er HRQOL scores than both of the conservative treatment s The differences between

PAGE 13

13 expectant management and hormonal therapy were minor S everal HRQOL variables particularly those related to physical functioning, remained associated with definitive treatment, after controlling f or the baseline sociodemographic, cancer characteristic and clinical factors We further investigate d the confounding effect of the baseline HRQOL in elderly men diagnosed with low or intermediate risk prostate cancer Both of t he propensity score and weighting methods were used to control confounding Radical prostatectomy and radiation therapy w ere found to be strongly associated with low er all cause mortality than conservative treatment after adjusting for commonly considered confounders T he associations w ere attenuated, but remained pronounced, after a dditional adjustment for the baseline HRQOL variables Further analyses showed that t he survival benefit was largely due to a n implausibly lower risk of death from causes other than prostate cancer, indicat ing the persistence of substantial residual confounding. Additional adjustment for the baseline HRQOL variables is desirable, but insuf ficient, to rule out confounding bias Randomized trial is warranted to evaluate the efficacy of primary treatment for localized prostate cancer .

PAGE 14

14 CHAPTER 1 INTRODUCTION Background Prostate cancer has the highest incidence rate and the second highest death rate of all cancer types for American men. About 63% of all prostate cancer cases are diagnosed at age of 65 years old or above. More than 90% prostate cancers are diagnosed at the clinically localized stage .1 Radical prostatectomy and radiation therapy are guideline recommended definitive treatments for clinically localized prostate cancer .2 Because of the indolent nature of localized prostate cancer, patients are unlikely to benefit from definitive treatment when their life expectancies are predicted to be greater than 10 year s from diagnosis, and conservative treatment is considered more appropriate. Expectant management is a guideline recommended conservative treatment, by which patients are closely monitored and definitive treatment is withheld until disease progresses aggre ssively .2 3 Hormonal therapy is another conservative treatment. Although not recommended in current guidelines, it has been used in practice for palliative purpose in symptomatic patients whose life ex pectancy is too short to benefit from definitive treatment. Because uncertainties in life expectancy estimation increase dramatically with age, the treatment decision is more difficult to make for elderly patients than young patients .3 Health related quality of life (HRQOL) is a broad health concept incorporating multiple physical and psychosociological dimensions. These health dimensions are important prognostic factors for survival and often taken into consideration when planning treatment for elderly cancer patients .4 5 HRQOL may also associate with other cancer treatment selection determinants, such as quality of life preservation, patient treatment preference, socioeconomic background, physician patient communication etc.6 It is therefore reasonable to inquire whether

PAGE 15

15 and to what exte nt baseline HRQOL variables are associated with primary treatment selection in elderly men diagnosed with localized prostate cancer. The association between life expectancy and primary treatment selection leads to strong confounding in observational studie s that try to compare survival outcomes across different primary treatments. Adjustment for commonly considered confounders appears inadequate. Uncontrolled baseline HRQOL has been suspected to be a major source of residual confounding .7 8 Study Rationale H ealth R elated Q uality of Life (HRQOL) and Ca ncer Treatment Selection The concept of HRQOL was developed in accordance with the health definition of the W orld Health Organization (WHO) It embraces a wide range of clinically relevant dimensions, such as physical functioning, mental functioning, socia l functioning, and disease symptoms .9 These health dimensions often influence cancer treatment decision. From a physicians perspective, they are hallmarks of overall health status, and therefore prognostic for life expectancy and tolerance for treatment related adverse events .2 5 Physicians have been suggested to take HRQOL into consideration when planning treatment for localized prostate cancer. From a patients perspective, HRQOL measures perceived health state in the context o f culture and value system, which directly or indirectly reflect personal preference towards a particular treatment .9 11 Exploration of the association between baseline HRQOL variables and primary treatment will help understand the complex treatment selection process in the elde rly prostate cancer population. The Confounding E ffect of B aseline HRQOL Confounding control is a central issue in epidemiologi cal studies. A randomized, controlled trial (RCT) with a large sample size is the most ideal way to eliminate confounding.

PAGE 16

16 However, it is difficult to conduct RCTs for localized prostate cancer with regard to survival outcomes because the mortality rate is low and a long follow up time up to 10 years is required. Even though one such trial was completed and another trial is ongoing right now, they do not provide evidence to contemporary practice in a timely fashion. The Scandinavian Prostate Cancer Group St udy No 4 study is the only completed high quality randomized trial, which compared radical prostatectomy and expectant management in men with clinically detected prostate cancer between 1989 and 1999. The long term follow up results were released in 2005.12 This trial was conducted before the prostate specific antigen (PSA) screening became widely used. It is unknown whether the trial results apply to the current US prostate cancer population, most of who are diagnosed at an earlier stage of disease by the PSA blood test. The Prostate Cancer Intervention Versus Observation Trial (PIVOT) study is an ongoing trial comparing radical prostatectomy and expectant management. This trial was initiated in 1994 and the patient recruitment completed in 2002. Ninety percent of the included patients were diagnosed with the PSA blood test Unfortunately, the study results have not been released yet.13 Large clinical databases contain comprehensive longitudinal data collected from real world practice setting s They are valuable research resources that are convenient, easy to access and concurrent. Thus, observational studies analyzing large clinical databases have and will continue to play a key role in comparative effectiveness research for localized prostate cancer. Previous observational studies showed a positive association be tween treatment aggressiveness and survival .14 19 Nevertheless, unmeasured confounding, particularly unmeasured baseline HR QOL, was highly suspected to account for this observation instead of true treatment benefit. HRQOL incorporates multiple health domains that are not routinely available in most clinical databases, such as physical functioning, psychosociological functioning and disease symptoms. In the

PAGE 17

17 context of prostate cancer baseline HRQOL meets the classical definition of confounding. It associates with exposure (treatment selection), predicts outcome (survival), and cannot be affected by either exposure or outcome .20 We are interested to examine the potential confoundi ng effect of baseline HRQOL in a comparative observational study for localized prostate cancer treatment. The results of the proposed study will provide evidence and methodological recommendations to future research. Purpose of the Study We used the CaPSURE (Cancer of the Prostate Strategic Urologic Research Endeavor) database to investigate the association between baseline HRQOL variables and primary treatment selection in elderly men ( hypo thesized that some, if not all, baseline HRQOL variables are associated with primary treatment selection independent of sociodemographic and clinical factors. Specifically, m ultinomial logistic regression was used to model the associations between the bas eline HRQOL variables and primary treatment selection The four types of primary treatments were included, which were radical prostatectomy, radiation therapy, expectant management and hormonal therapy. The included HRQOL variables were measured by t hree H RQOL instruments, which were the Karnofsky performance scale (KPS), the RAND 36 item health survey 1.0 (SF 36) and the UCLA prostate cancer index (PCI) .21 KPS is a single scor e reflecting physician assessed functional status. SF 36 is an instrument for patient reported general HRQOL. It measures eight health domains: physical functioning (PF) role physical (RO), bodily pain (BP) general health (GH) vitality (VT), social func tioning (SO), role emotional (RE), and mental health (MH).22 23 PCI is an instrument for patient reported prostate cancer specific HRQOL. It measures six health domains: sexual functioning (SF) bowel functioning (BF) urinary functi oning (UF) sexual bother (SB) bowel bother (BB) and urinary bother (UB) .24

PAGE 18

18 Another purpose of the current study was to investigate whether and to what extent baseline HRQOL variables confound the association between primary treatment selection and survival outcomes in elderly men diagnosed with low or intermediate risk localized prostate cancer. Specifically, men treated with expectant management or hormonal therapy were pooled into a conservative treatment, and compared with men treated with radical prostatectomy or radiation therapy The primary outcome was 10 year all cause mortality The magnitude of confounding bias was quantified by using the change in estimate criterion which examines the relative changes in association estimates before and after adjustment for the baseline HRQOL variables in addition to commonly considered confounders.20 A change of greater than 10% was considered contextually important. Different confounding adjustment methods were used. T he propensity score method was used as it was done in the previous studies The weighting methods were used to examine t he confounding effect in accordance with the counterfactual outcome theory .25 27 The counterfactual outcome theory based confounding requires a clearly defined target population of inference. Since men treated definitive ly are very different from men treated conservatively three different weighting str ategies were used. When weighting to the overall sample, the target population was represented by the entire study sample, regardless of the primary treatments they actual ly received. When weighting to the definitive treatment, the target population was re presented by the men actually treated definitively. When weighting to the conservative treatment, the target population was represented by the men a ctually treated conservatively. To further illustrate the confounding effect of baseline HRQOL, 10year mort ality due to prostate cancer and other reasons were compared. The rationale was that definitive treatment for localized prostate cancer would unlikely affect other cause mortality. At least, the effect size

PAGE 19

19 would be smaller than that on prostate cancer spe cific mortality. We expected to see that the difference in 10 year other cause mortality, if there was any, diminished after additional HRQOL adjustment and the difference in 10year prostate cancer specific mortality remained .7 Research Questions and Hypotheses Research Q uestion 1 In CaPSURE elderly men ( diagnosed with localized prostate cancer, which baseline HRQOL variables are independently associated with primary treatment selection? Research Hypothesis 1 The null hypothesis is that, none of the baseline HRQOL variables are independently associated with primary treatment, conditioning on cancer characteristics sociodemographic and clinical variables. The alternative hypothesis is that, at least one baseline HRQOL variables are independently associated with primary treatment, conditioning on cancer characteristics sociodemographic and clinical variables. Research Question 2 In CaPSURE elderly men ( or intermediate risk prostate cancer, do baseline HRQOL variables confound the association between primary treatment and all cause survival, controlling for commonly considered confounders ? Resear ch Hypothesis 2 The null hypothesis is that the hazard ratio for all cause mortality does not change significantly (=<10%) after adjustment for the baseline HRQOL variables in addition to commonly considered confounders The alternative hypothesis is that, the hazard ratio for all cause mortality changes significantly (>10%) after adjustment f or the baseline HRQOL variables in addition to commonly considered confounders

PAGE 20

20 CHAPTER 2 LITERATURE REVIEW This chapter is d ivided into three sections: 1) e pidemiology of prostate cancer; 2) primary treatment selection for localized prostate cancer; 3) theoretical framework for confounding. Epidemiology of Prostate Cancer Prostate cancer is the most frequently diagnosed cancer in American men. About 1 in ever y 6 men in US will develop prostate cancer during their life time (life time risk of 15.6%). It has been estimated that 192,280 men in the United States were newly diagnosed with prostate cancer in 2009, which counted for approximately 25% of new non skin cancer cases. Prostate cancer is also the second most fatal cancer in American men with an estimation of 27,360 deaths in 2009, which counted for approximately 9% all type cancer death.1 Since PSA screening technology was introduced in early 1990s, most pr ostate cancers have been detected at early stage nowadays with low lifethreatening risks. More than 90% of all prostate cancers are discovered at the clinically localized stage without regional involvement of pelvic lymph nodes or distant metastases. With increasing life expectancy and the introduction of more advanced screening techniques, the number of localized prostate cancer cases is expected to keep rising in the future .1 Advanced age is a strong prognostic factor for prostate cancer. According to the Surveillance, Epidemiology and End Results (SEER) database of the USA National Cancer institute, the median age at diagnosis was 68 years old. Approximately 63% of prostate cancer cases were diagnosed in men aged 65 years old or above. More than 70% of deaths due to prostate cancer occur in men aged 75 years old and above. The other two well established risk factors are race/ethnicity and family history. African American has the highest prostate cancer incidence rate in the world. The disease is more common in North America and northwestern

PAGE 21

21 Europe than Asia and South America. Prostate cancer is also considered a hereditary disease. Recent genetic studies suggest that men who have a first degree relative (father, brother, son) with prostate cancer have a two fold to three fold increased risk of developing prostate cancer, compared to men with no family history .1 28 Primary Treatment Selection fo r Localized Prostate Cancer P rimary Treatment Option s Standard primary treatment options for localized prostate cancer include radical prostatectomy, radiation therapy (including both external beam radiation therapy and brachytherapy), and expectant management, which is also referred as watchful waiting or active surveillance.2 3 Hormonal therapy has been proposed as an alternative conservative treatment for men with clinically localized disease electing less aggressive management. Although not recommended by the current guidelines, its use has raised dramatically over the last decade in all risk categories .29 Conclusive evidence regarding treatment efficacy in elderly patients is absent in the current literature. Only one high qual ity randomized controlled trial was completed in Scandinavia countries, which compared radical prostatectomy vs. expectant management. The results favored radical prostatectomy over expectant management by showing a lower risk of cancer recurrence, cancer related death and all cause mortality after 10 year follow up. However, the survival benefit appeared limited to a younger subgroup aged less than 65 years old.12 Post treatment quality of life is another important clinical outcome besides survival. Both radical prostatectomy and radiation therapy can result in detrimental effects on urinary, bowel and sexual function. Radical prostatectomy risks the same complications from major surgery as well as urin ary incontinence and impotence. Radiation therapy imposes similar complications to patients with more significant bowel side effect and less severe urinary problems. Hormonal therapy also has side effects. It leads to hot flash, increased fatigue, difficul ties with erection, and

PAGE 22

22 declines in sexual interest and enjoyment .30 Expectant management avoids active treatment related side effects. However, it increases patient perceived uncertainty, anxiety, and other mental burdens .31 32 Conceptual Model for Cancer Treatment Decision Making Primary treatment selection for localized prostate cancer is a com plex process. Based upon a general cancer treatment decision making model developed by Zafar ,6 three parties are involved: physician, patient and patients family (F igure 1). Factors contributing to physician treatment preference include knowledge of individuals disease, general disease knowledge and personal experience. Factors contributing to patients treatment preference include sociodemographic characteristics, illness experience, quality of life concern, and disease status. Patients family influences the treatment selection through modifying patients preference. The final treatment decision is a result of active interactions between patients and physicians Factors i nfluencing p hysician p reference Tumor characteristics Tumor characteristics are the most important factor that physicians consider in primary treatment selection .33 Three quantifiable systems have been well established to characterize prostate cancer. The TNM system refers to Tumor, Nodes and Metastasis. T staging indicates the size of the tumor. A higher T stage indicates the involvement of a larger proportion of the prostate. N staging indicates whether or not the cancer has spread into nearby lymph nodes. M stagi ng indicates whether or not the cancer has metastasized. Clinically, localized prostate cancer is defined as T stage of T3a and below without evidence of lymph node involvement or metastatic spread. The Gleason grading system evaluates the microscopic patt ern of the cancer cells. Higher Gleason scores indicate higher cancer aggressiveness. The prostate specific antigen (PSA) is a biomarker, the level of which is

PAGE 23

23 positively associated with prostate cancer risks. Further, a risk stratification scheme for clin ically localized prostate cancer has been developed synthesizing the three systems into one system: Low risk: PSA <10 ng/mL and a Gleason score of 6 or less and T stage of T1T2a Intermediate risk: PSA between 10 to 20 ng/mL or a Gleason score of 7 or clinical stage T2b T2c High risk: PSA This risk stratification scheme has been widely adopted in current practice guidelines and serves as a basis for clinical decision making .2 3 Life expectancy. Localized prostate cancer usually progresses very slowly. When the life expectancy is short, patients are very likely to die from competing risks before localized prostate cancer becomes a cause of morbidity and mortality. Therefore, life expectancy is a key consideration for physician to avoid treatments that might decrease quality of life without prolonging survival. A 10year life expectancy rule has been generally accepted in the current clinical practice. According to the National Comprehensive Cancer Network (NCCN) guideline, radical prostatectomy is not recommended to patients with less than 10 years life expectancy .34 Life expectancy estimation can be very challenging, especially for elderly patients. In principle, chronological age is not a primary consideration in prostate cancer treatment decision. El derly men who receive stage appropriate therapy tend to do as well as their younger counterparts .34 35 Life expectancy estimation should be based on an overall health status evaluation of co morbidity, functional status, and other geriatric syndromes. Unfortunately, physicians have been criticized for being poor life expectancy estimators .36 The comprehensive geriatric assessment (CGA) has recently been developed to assist with the overall health status assessment for elderly patients. It assesses multiple geriatric dimensions, includi ng comorbidity, functional status, physical performance, cognitive status, social support,

PAGE 24

24 nutritional status, and polypharmacy .2 34 Out of the comprehensive CGA health dimensions, comorbidity is the only well documented treatment predictor in the current literature. Patients treated with radical prostatectomy are found to have lesser number of comorbidities or lower comorbidity index scores than those treated with nonsurgical interventions, after adjusting for cancer and socioeconomic characteristics .37 39 HRQOL overlaps CGA in multiple health dimensions. To our best knowledge, few studies have explore d whether and to what extent HRQOL variables independently influence primary treatment selection for localiz ed prostate cancer. Physician characteristics. Survey studies show that prolonging survival is the most important outcome that physicians consider. One study indicated that 90% of physicians defined effectiveness as extending expected survival, in contrast to 43% of patients .33 In the absence of solid evidence regarding survival superiority of one treatment over anot her, physicians expertise is one of the strongest predictors of treatment selection. A physician survey suggested that specialists prefer the treatment modality they deliver. That is, urologists almost universally identify radical prostatectomy as the bes t treatment option, while radiation oncologists almost universally identify radiation therapy as the best treatment option .40 A recent analysis of the SEER Medicare database revealed consistent results. It was fo und that for those who consulted urologist s alone only 5% underwent radiation therapy as primary treatment. For those who consulted both urologist s and radiologist s over 80% underwent radiati on therapy as primary treatment.41 However, the observation was interpre ted as a result of urologists referral decision rather than the greater persuasiveness of radiologist. It was believed that urologists usually perform the biopsies that lead to a diagnosis of prostate cancer, and

PAGE 25

25 therefore decide on the subsequent interve ntions. They would not refer a patient to radiation oncologist s unless they feel radiation therapy as the optimal treatment.42 Factors i nfluencing p atient p reference Subjective opinion. Patient centered treatment decision making has been advocated for cancer management. Large amount of qualitative studies we re conducted to explore the treatment decision making process in men with early stage prostate cancer. Zeliadt6 reviewed 69 studies and concluded that substantial variations existed in what patients consider important in primary treatment selection One example is that 18 different attributes of treatment were identified as being the most important in a small group of 60 patients .43 In general, eradication of cancer is a key issue that concerns almos t every patient upon diagnosis. However, patient treatment preference is often guided by personal belief and feeling rather than scientific evidence or logical reasoning. For example, radical prostatectomy is often preferred over radiation therapy by many patients because surgical removal of prostate is inaccurately believed to be the only way to cure cancer and radiation therapy is often inaccurately believed to be the backup plan. One study of elderly patients willingness to accept radical prostatectomy versus expectant management showed that nearly half of the participants preferred radical prostatectomy even when radical prostatectomy was presented with zero survival benefit and 1 2% operative mortality risks .44 Meanwhile, fear of surgery is one of the most popular reasons that patients choose radiation therapy and refuse radical prostatectomy .45 46 Treatment related side effect is another important issue that concerns a s ignificant proportion of patients .6 In one large study involving 1,000 prostate cancer patients, approximately 45% defined treatment effectiveness as preservation of quality of life .33 However, little is known how men tradeoff quality of life and quantity of life in their decisionmaking process. Current evidence shows that the treatment related side effect has little impact on final

PAGE 26

26 treatment decisio n, in spite of its significance stated by patients. One example is that more than 80% surveyed patients rated preservation of sexual function as a very important treatment goal, but only 3% of them indicated that having few side effects was the most im portant consideration in initiating therapy .33 Socioeconomic status. Analyses of large clinical databases revealed certain association patterns between socioeconomic characteristics and primary treatment of localized prostate cancer. Race/ethnicity has been identified to be a strong predictor for primary treatment selection. A common finding was that black men were l ess likely to receive radical prostatectomy than white men .47 54 The possible explanations for this phenomenon include poor physician patient communication and trust, more fear of surgery, more concerns about sexual functioning, more pessimism about the curability of prostate cancer, and lack of specifically designed decision aid tools for black men.50 55 Denberg50 51 analyzed the SEER database and found a s trong association between marital status and primary treatment, independent of cancer risks and other socioeconomic factors. Married men were more likely to receive aggressive treatments than unmarried men One explanation for this observation is that marr ied men or their wives advocate therapy that they perceive as likeliest to cure cancer, whereas unmarried men often lack social supports that would encourage aggressive interventions. Several qualitative studies also showed that patient wives tended to choose treatments that offered the best longevity and care less about treatment related side effects .56 58 The impacts of income and education on primary treatment selection are less clear in current literature. Independent associations were found between primary treatments and income/education levels in s everal SEER studies that used ecological measures based on census

PAGE 27

27 tract. Men living in lower income and education areas were more likely to use less aggressive treatments .50 53 54 59 However, Yan60 found no such associations when using individual level data from a prostate cancer screening program Illness experience. Patient treatment preference can be shaped by individuals prior health experience (co morbidity, sympt oms, psychological status, and so on). Current evidence is conflicting about how function and disease symptoms at diagnosis influences patients t reatment preference. Saigal11 assessed patient utilities for treatment associated side effects among 401 men underg oing prostate needle biopsy. It was concluded that treatment preference did not differ by baseline urinary, sexual and bowel functions. Instead, the general health seemed to affect treatment preference. Men perceiving better general health tended to place higher value on survival and choose more aggressive treatments. On the opposite, men suffering from poor general health tended to place higher value on quality of life and choose less aggressive tre a tments Zeliadt10 surveyed 593 patients with newly diagnosed localized prostate cancer to assess whether preexisting genitourinary symptoms at the time of diagnosis influenced patients prefere nces for radical prostatectomy vs. nonsurgical treatment. It was found that preexisting sexual dysfunction but not bowel or urinary dysfunctions was associated with considerations on treatment related side effects. Men without sexual dysfunction at diagnos is appeared to care less about the risks of side effect than men with sexual dysfunction. Physician p atient i nteraction The physician patient interaction is a critical stage in the conceptual cancer treatment decision making model. It assumes an acceptable balance between the patient and physician in terms of shared decision making .6 However, several qualitative studies showed that treatment selection process for localized prostate cancer is dominated by physicians. Patients often play a passive role, relying on physicians recommendations In one group of patients that recei ved

PAGE 28

28 definitive treatments, urologists are rated as an important source of information by 95% 100% responders and the most important source of information by 70% of responders .45 In a telephone survey of 421 randomly selected prostate cancer patients, 75% responders chose the first treatment that physicians ever recommended to them .61 Davison62 interviewed 25 men who decided to manage their low risk prostate cancer with expectant management. The specialists' description of the prostate cancer was identified to be the most influential factor on patients treatment decision making Quantitative studies of large clinical databases demonstrated strong associations between primary treatment received and specialist consulted, which indirectly confirmed the dominant role of physician in treatment selection.41 63 Lack of thorough physicianpatient communication is another reason for the passive involvement of patients in the treatment selection process. Physicians often do not provide patients with the treatmentrelated side effect information, or provide the side effect information in a way that patients do not understand. One study showed that only 16% of surveyed patients recalled discussing side effects with their physicians, whereas nearly all surveyed physicians reported that they had thorough discuss ions about treatment rela t ed side effects with all their patients .33 The inadequate communication about treatment related side effects helps explain why treatment related side effects concern patients but minimally influence final tre atment selection. Patient sociodemographic characteristics seem to influence the patient physician interaction. Qualitative studies showed that physicians selectively discussed the treatment options with patients. Black men and men from low socioeconomic c lass were often provided with less treatment options. These findings help explain the observed racial and socioeconomic disparities in primary treatment selection .64

PAGE 29

29 Theoretical Framework for Confounding Confounding D efinitions The concept of confounding is a central issue in epidemiology. Historically, the word confounding w as used with different underlying meanings .25 Three definitions of c onfounding are proposed in the current literature, which are the classical definition, collapsibility definition and counterfactual definition.65 Cla ssical d efinition The classical definition characterizes confounders by three criteria: 1) it must be a risk factor for the outcome among the unexposed; 2) it must be associated with the exposure in the source population from which the subjects arise; 3) it must not be affected by exposure or outcome .20 Confounding occurs when confounders are not adjusted. Although the classical definition is easy to understand and operate, it has several limitations. First, the observed association (or no association) between confounder and the outcome of interest or the e xposure of interest may be spurious due to unknown variables. Confounder identification by using data based statistical techniques can be misleading and should be reserved when little prior information is available .25 Secondly, a variable that satisfies the three criteria is not necessarily a confounder. Consider the dir ected acyclic diagram below (Figure 2): variable W satisfies all three criteria. It is associated with exposure X, predicts outcome Y, and not affected by either X or Y. However, W is not a confounder. On the contrary, controlling for W will introduce an non causal association between X and Y .20 This example implies that the three criteria are necessary but insuf ficient conditions for confounding .20 25 27 66 67 Confounder identification must be grounded on an understanding of the causal pathways that links the variables under study. The causal pathways should always be evaluated against background disease risk, knowledge of subject matter, logical argument, evidence from previous

PAGE 30

30 studies and other sources. Systematic tools have been developed to identify confounders such as the directed acyclic diagram .20 26 68 Collapsibility d efinition The second confounding definition popularly used in the current literature is based on collapsibility. Collapsibility means that the effect measure of an exposure on an outcome is homogeneous across the strata of a covariate and the stratum specific measure equals to the marginal measure. In the situation of noncollapsibility where the stratumspecific effect measure and marginal effect measure are unequal, the covariate is a collapsibility definition based confounder .20 25 65 67 Assume X is the exposure of interest, Y is the outcome of interest, and Z is a potential confounder. Let us further denote as the crude effect measure of X on Y, and denote as the stratum specific effect measure of X on Y, adjusting for Z. Z is a collapsibility based confounder when *. Change in estimate is the most popularly used criterion to quantify the confounding magnitude, which is a percentage difference between the crude effect measure and the stratum specific effect measure ( 100% ) 20 69 A quantitative cutoff value is often specified in advance to determine whether a confounder is contextually important or not. For example, a confounder may be considered unimportant unless the change in effect measure is greater than 10%. The collapsibility definition of confounding also has limitations. First, noncollapsibility by itself does not necessary imply confounding. Change in eff ect measures after adjustment for additional variables may be caused by reasons other than confounding. It may reflect adjusting away the exposure effect when the adjusted covariate is an intermediate variable in the causal pathway between exposure and out come; it may reflect the introduction of selection bias when the adjusted covariate is a common effect of both exposure and outcome ;68 70 it may also be an effect of measurement error rather than confounding .71 Hence, same with the classical

PAGE 31

31 definition, the collapsibility criterion has to be applied on top of a plausible causal linkage network. Secondly, the collapsibility definition depends on the chosen effect measures. Certain effect measures are not collapsible in nature, such as odds ratio, rate ratio and hazard ratio. The noncollapsibility nature reflects the inequality of marginal and stratumspecific effects, rather than confounding These effect measures could change upon adjustment for a covariate that is not a confounder or remain same upon adjustment for a covariate that is a confounder. Vice versa, a variable can be identified to be a confounder according to one effect measure b ut not another one.25 27 66 72 Thirdly, the noncollapsibility definition requires the homogeneity assumption, which assume s the effect measure to be constant across the strata of the potential confounder. It is unclear how to proceed when heterogeneity exists. Counterfactual d efinition The counterfactual definition of confounding is based on the counterfactual theory of caus ation, which views the effect of exposure as a contrast between the outcome in a target population when every unit is exposed and the outcome in the same population when every unit is unexposed. Because the same unit cannot be exposed and unexposed at the same time, one exposure condition must be hypothetical. The outcome under the hypothetical exposure condition is unobservable and called the counterfactual outcome. In practice, the counterfactual outcome is approximated from a substitute population that a ctually has the exposure condition. Confounding occurs if and only if the observed outcome from the substitute population differs from the counterfactual outcome. Confounders are any variables that cause the difference. Assuming the exposure is binary and the target population is exposed, let 1 represent the observed outcome of being exposed, 0 represent the counterfactual outcome had if the population been unexposed, and 0 represent the observed outcome in an unexposed group.

PAGE 32

32 Confounding occurs if and only if 0 0 *.20 25 27 66 To quantify the confounding bias under the counterfactual definition, the following deductions can be made: 1/ 0 = (1/ 0) (0 /0 *) (1/ 0 *)/(1/ 0) = (0 /0 *) 1/ 0 is the true effect measure, 1/ 0 is the biased estimate and 0 /0 is the measure of the bias due to confounding, which is sometimes called the confounding risk ratio.25 Counterfactual theory based confounding definition automatically meet the classical definition. It has to causally affect the outcome and distribute unevenly between the substitute population and the target population. The confounder would not be an intermediate variable either because the substitute population and the counterfactual population are under the same exposure condition. The counterfactual definition of confounding has recently been advocated for several reasons: 1) it is specific regarding to causal inference and has a direct causal interpretation; 2) it does not make reference to a part icular effect measure; 3) it offers solutions to time dependent confounding that the classical and collapsibility definitions cannot solve appropriately. 25 27 65 67 Confounding and s election b ias It should be noted that confounding and selection bias are sometimes used interchangeably in the currently literature when the bias is resulted from differential selection at start of follow up. The present study adopts the criteria developed by Rothman, et al .20: differential selection that occurs before exposure and disease leads to confounding, and can thus be controlled by adjustments for the factors responsibl e for the selection differences In contrast, selection bias was usually described in epidemi ology arises from selection affected by the exposure under study, and may be be yond any practical adjustment ( p. 138) Therefore, the bias resulting from the uncontrolled baseline HRQOL is referred as confounding rather than selection bias in the current study.

PAGE 33

33 The R ationale of B aseline HRQOL as C onfounder Previous observational studies reported a significant survival benefit of definitive treatment (radical prostatectomy or radiation therapy) relative to expectant management in elderly men diagnosed with low or intermediate risk localized prostate cancer .15 16 18 19 These findings contradicted the finding from the Scandinavia randomized trial, which showed no survival benefit of radical prostatectomy vs. expectant management in patients aged 65 years old and above .12 The implausible association was illustrated by Giordano7 who replicated a SEER Medicare study that found the definitive treatment associate with nearly 30% lower risks for prostate cancer specific mortality and all cause mortality relative to the expectant management In addition to the similar findings on prostate cancer specific mortality and allcause mortality, the replicated study found definitive treatment strongly associate d with s ignificantly lower mortality risks resulting from other diseases. Moreover, patients that received radical prostatectomy were found to have even better survival than a cancer free group. Given the assumption that prostate cancer interventions do not prolong survival from causes other than prostate cancer, these findings implied that the confounding bias was so strong that adjustment of SEER Medicare variables was inadequate. It was further speculated that unadjusted baseline HRQOL might contribute to the uncontrolled confounding .7 As discussed in the previous section, confounding identification requires prior knowledge of causal pathways. When the primary outcome is mortality, any independent mortality predictors at baseline besides primary treatment should be carefully considered as confounder candidates. The Medicare administrative database is primary developed for reimbursement purposes. It contains little health status information other than diagnostic codes. Many relevant health dimensions are omitted, such as functional status, social support, frailty etc. In addition, the generic co morbidity measures based diagnosis codes from administrative databases have

PAGE 34

34 been criticized for poorly reflecting the health status of the elderly population. First, under diagnosis is common in elderly patients. New symptoms may not be recognized or dismissed as manifestations of pre existing conditions .73 Secondly, current diagnosis coding systems have limited capability of reflecting disease severity. Comorbidity measures based on diagnosis code, such as number of comorbiditi es and comorbidity indices, assume that each constituent comorbid condition has the same impact and that comorbid conditions with the same diagnosis code are equally severe. Unfortunately, this is often not true. Clinical prognosis often varies greatly in patients carrying the same diagnosis codes .74 77 The potential co nfounding role of baseline HRQOL can be attributed to its nature as health status measures. In the present study, the general HRQOL is measured by SF 36. This instrument has been validated in a large sample of 9,897 elderly patients, aged from 65 to 104 ye ars old.4 22 23 34 76 78 80 It includes nine domains that address different aspects of health. For example, the physical functioning scale measures limitations in behavioral performance of ever yday physical activities; role functioning scale measures the extent of disability in everyday activity due to physical problems; the social functioning scale and role emotional scale measure limitations or disability due to personal or emotional problems. The HRQOL also measures disease symptoms. SF 36 has specific scales addressing body pain and fatigue, two common symptoms for many diseases. The PCI concerns prostate cancer related symptoms regarding urination, sexuality and bowel function. HRQOL often c orrelates with disease severity. For instance, Faller81 found that the SF 36 composite sco res independently predicted survival in patients with chronic heart failure. The prognostic values were mainly due to the strong correlations with the heart failure severity and depression severity. Furthermore, HRQOL correlates with patients income level social support,

PAGE 35

35 outlook, and attitudes. These factors often affect the patients ability to manage their cancer and comorbidities, but would not be captured in regular clinical examinations. Overall, baseline HRQOL variables measure multiple dimensions of health status. Adjustment of HRQOL variables in conjunction with commonly used demographic and comorbidity variables may help reduce the residual confounding. This effort will be particularly fruitful for older men who often simultaneously have multipl e comorbid conditions, impaired functional status, decreased social support and other characteristics for poor outcomes. Figure 21.Conceptual model for patient and physician cancer treatment selection. Source: Zafar et al 2006.

PAGE 36

36 Figure 22. A directed acyclic gram under which the classic confounder criteria fail. Source: Rothman et al. Modern Epidemiology Third Edition.

PAGE 37

37 CHAPTER 3 METHODOLOGY The methodology chapter consists of four parts: 1. description of the data source the CaPSURE database; 2. study design and sample selection; 3. descriptive analysis; 4. multivariate statistical analysis. Data Source The Cancer of Prostate Strategic Urologic Endeavor (CaPSURE) was the primary data source for the proposed study. CaPSURE is a large prostate cancer registry maintained by the Urology D epartment University of California, San Francisco since 1995. The primary goal of CaPSURE is to evaluate the impact of prostate cancer treatments on clinical, economic, and quality of life outcomes. It longitudinally follows biopsy proven prostate cancer patients in a naturalistic setting. Patients are enrolled from thirty seven community urology practices, three academic medical centers and three veteran affair hospitals throughout the United States. All patients with biopsy proven prostate cancer are invited to join the study by their urologists at the time of a clinical visit, irrespective of disease duration, severity or treatment type. Newly enrolled patients are followed prospectively until death or withdrawal from the study. Upon a form of consent is signed, data are collected from both urologists and patients. Urologists provide comprehensive clinical assessments of their patients at the time of first enrollment, gathering the data relevant to prost ate cancers, including laboratory test results, pathological staging, initial and following treatments used, and important dates. During the follow up time, urologists further collect treatment and clinical data at each clinic or office visit. Patients com plete questionnaires regarding their demographics, comorbidity, quality of life, satisfaction of care and healthcare resource utilization when they first join CaPSURE. During the follow up time, patients are mailed questionnaires every 6 months to update t heir quality of life,

PAGE 38

38 satisfaction with care, severity of co morbid conditions, and health care resource utilization. After patient mortality is reported by participating clinicians or the next of kin, a copy of the state death certificate is obtained and a determination of the cause of death (prostate cancer specific vs. other cause) is made by consensus of the study investigators. In general, death is considered to be prostate cancer specific if prostate cancer is listed as a primary, secondary, or tertia ry cause of death and no other malignancy was listed as a higher order cause. For patients whose state death certificate is unavailable, the National Death Index will be queried to identify date and cause of death. More details about CaPSURE have been publ ished elsewhere.21 Study Design and Sample Selection A cross sectional study design was used to investigate the associations between baseline HRQOL variables and primary treat ment selection in elderly men diagnosed with localized prostate cancer. The type and date of primary treatment received were recorded in CaPSURE. One advantage of CaPSURE was that the primary treatment was confirmed by doctors, which avoided potential trea tment misclassification. A 9 month time window (278 days) after diagnosis was used for primary treatment identification. Patients were classified into the radical prostatectomy group if they received the procedure during the time window, regardless of neo adjuvant therapy or adjuvant therapy. Patients were classified into the radiation therapy group when they received brachytherapy or external beam radiation as primary treatment during the time window, regardless of neoadjuvant therapy and adjuvant therapy Patients were classified into the hormonal therapy group if they received bilateral orchiectomy, luteinizing hormone releasing hormone, agonist, antiandrogen, diethylstilbesterol, or finasteride therapy alone or in any combination during the time window without any additional definitive treatment. The remaining patients were classified into the expectant management group as they did not receive

PAGE 39

39 any of the three interventions above and doctors checked the expectant management box in the questionnaire. The following sample selection criteria were further applied: Newly diagnosed with prostate cancer in year of 1995 or later. In CaPSURE, the diagnostic biopsy date is used as the diagnosis date, which is available for approximately 90% of enrolled patients .29 The requirement of diagnosis year of 1995 or later was to ensure prospective follow up since CaPSURE enrollment started in 1995. Newly diagnosed with prostate cancer at age of 65 years and older. No evidence of lymph node involvement (N1) or metastasis (M1) at diagnosis. Patients were excluded if no prostate cancer intervention was indicated and they were lost to follow up within 9 mont hs after diagnosis. Baseline HRQOL data availability. For radical prostatectomy, radiation therapy, and hormonal therapy groups, baseline was defined as the periodfrom90 days before the diagnosis date to the treatment date. For the expectant management g roup, baseline was defined as the period from 90 days before diagnosis date to 278 days after diagnosis. A retrospective cohort study design was used to investigate the potential confounding role of baseline HRQOL variables when comparing primary treatments with regard to survival outcomes in elderly men aged 65 years old or above. Because expectant management is not recommended for highrisk prostate cancer, this section further concentrated on elderly men with low or intermediate risk prostate cancer. The cancer risk level was determined by Gleason score and T stage as described in the previous studies .18 Patients were excluded if they had Gleason score sum more than 7 or T3/T4 stage. The radical prostatectomy and radiation therapy groups were collapsed into a definitive treatment g roup. The expectant management and hormonal therapy groups were collapsed into a conservative treatment group. These two groups were compared with regard to all cause mortality, prostate cancer specific mortality, and other cause mortality in 10 year follo w up. Because we found men treated with radical prostatectomy were quite different from men treated

PAGE 40

40 with radiation therapy in a wide range of baseline characteristics, both treatments were also compared with conservative treatment separately. Descriptive A nalysis Descriptive analyses were performed for the following variables: The primary treatment received Baseline sociodemographic variables Baseline prostate cancer variables Baseline BMI and comorbid variables Baseline Karnofsky performance score Baseline SF 36 scores Baseline PCI scores All these HRQOL variable were numeric, ranging from 0 100. Their distributions were summarized by mean SD and median with inter quartile. Because these HRQOL scores were highly skewed, they were dichotomized f or analyses. The SF 36 scores were dichotomized at population median for male aged 65 years old or above .22 The PCI scores and KPS were dichotomized at the sample median, as the population norms are not available. For those variables where the median was 100, patients were divided into two groups of 100 vs. <100. The crude associations between treatment and baseline variables were examined by using univariate logistic regression. The crude associations between the baseline variables and survival outcomes were examined by using univariate Cox regression analysis. Multivariate S tatistical A n alysis Missing D ata I mputation The CaPSURE collected data by using questionnaires completed by both patients and physicians. A small portion of baseline data was missing. To preserve statistical power, missing data were imputed by using the multiple imputa tion (MI) technique .82 MI assumes missing at random, which means that missing data depend on known values, and accounting for the known

PAGE 41

41 values produces unbiased results. Compared to the single imputation methods, MI not only reflects the natural variability in the missing data, but also incorporates the uncertainty of unknown missing val ue prediction. Further, multiple imputation has been shown robust to departures from normality assumptions and provides adequate results in the presence of low sample size or high rates of missing data. The method for imputation and subsequent analysis of the filled in data involved three steps: 1) imputing data to obtain 5 copies of the filledin data set; 2) carrying out standard statistical analyses on each dataset separately to obtain desired parameter estimates and standard errors; and 3) combining res ults of the five analyses by computing the overall estimations and a variance estimate that includes both a within imputation and an across imputation component. For the first step, the SAS procedure of PROC MI and the Markov chain Monte Carlo (MCTC) metho d were used to generate the five imputed datasets. As a general rule for building the imputation model, using all available information maximizes certainty and minimizes bias. All of the baseline variables together with the follow up time log transformatio n were included into the imputation model. SAS requires the imputed values to be numeric. Dummy variables were setup for all categorical variables before imputation. The imputed values remained unrounded in analysis to avoid potential bias since the primar y goal of imputation was to account for the missing data uncertainties rather than getting plausible values. Besides, none of the categorical variables was the primary interest of the present study .83 The planned statistical analyses were conducted in each of the five imputed dataset s The SAS procedure of PROC MIANALYZE was used to combines the results. More details regarding how to analyze the imputed datasets are described in the following sections

PAGE 42

42 Multinomial L ogistic R egression A multinomial logistic regression model was used to examine the associations between the baseline variables and primary treatments. The dependent variable was primary treatment with four categories: radical prostatectomy, radiation therapy, expectant management, and hormonal therapy. The independent variables were the baseline sociodemographic, clinical, and HRQOL variables. The formulation of the multinomial logistic regression is given by the following equation: Log jJ = + 1X1+ 2X2+ + iXi J denotes the number of outcome categories. J denotes the probability of the baseline outcome category. j denotes the probability of the comparison outcome category. X1 Xi denote independent variables that associated with the treatment selection. 1 i denote the association parameters for the independent variables. Bivariate analyses were performed to examine the crude associations between primary treatment selection an d individual baseline variables. Multivariate analyses were performed to examine the independent associations between baseline HRQOL variables and primary treatments, adjusting for the baseline sociodemographic and clinical variables. The multivariate anal ysis was completed in three steps. Step 1: A preliminary multinomial logistic regression analysis was constructed to select all baseline sociodemographic and clinical variables that was significantly associated with primary treatments. An automated backwar d variable selection procedure was used with stay criteria of P=0.05 while forcing in age and cancer risk level. Step 2: The HRQOL variables entered the preliminary model individually to determine their independent associations with the primary treatments by using log likelihood test. All the preselected sociodemographic and clinical variables were forced in. Step 3: All HRQOL variables entered the preliminary model at the same time and the most predictive HRQOL

PAGE 43

43 variables were determined by using the backwa rd elimination procedure with a stay criterion of P=0.05. All the preselected sociodemographic and clinical variables were forced in. Because the multiple imputation procedure produced five imputed datasets, all the variable selection processes were first performed in each dataset. The final selected variables had to retain in at least three datasets. Cox P roportional Hazard Model Cox proportional hazard model was used to compare the survival outcomes between the definitive treatment and the conservative treatment. Both propensity score and weighting methods were used to adjust for baseline confounders. All analyses were performed twice, without and with baseline HRQOL variables considered in addition to sociodemographic and clinical variables. The two haz ard ratios were compared to assess the confounding magnitude of the baseline HRQOL variables by using the changein estimate criterion. Both relative and absolute changes were calculated. Because of the immortal time concern, the starting time point was th e treatment date for radical prostatectomy, radiation therapy, and hormonal therapy. The starting time point for expectant management was the 278th day after the diagnosis date. Patients who did not have an event of interest were censored when they died fr om unrelated reasons, dropped out CaPSURE, or the end of the study period. The proportional hazards assumption was checked by plotting log log functions. The formulation of the Cox proportional hazard model is given below: ( t ) = 0 ( t ) exp ( txXtx+ 1X1+ 2X2+ + iXi) 0(t) denotes baseline hazard function at time t. Xtx denotes the dummy variables of individuals treatment category (conservative treatment was the reference category ).

PAGE 44

44 tx denotes the association parameters of definitive treatment. X1 Xi denote the covariates, which was the propensity score in current study. 1 i denote the association parameters for the covariates. Informative Censoring Adjustment Another ke y assumption for Cox proportional hazard model is random censoring, which means that an individuals censoring time is independent of his failure time. Nonrandom censoring would lead to selection bias .70 The inverse probability of censoring weighting (IPCW) method introduced by Robins was used to correct the selection bias due to informative censoring .13 T he IPCW weights were proportional to the inverse of the probability of surviving uncensored, conditioning all t he outcome predictors. In the current study, the 10year follow up time was divided into 120 intervals with 30 days in each. The probability of being uncensored was calculated for each interval by using logistic regression. The dependent variable was censo ring status (noncensored vs. censored), and the independent variables were all baseline sociodemographic, clinical and HRQOL variables. Three IPCW weights were calculated for all cause, prostate cancer specific, and other cause mortality, respectively. Be cause the localized prostate cancer interventions were point treatments, there was little concern on time dependent confounding. The 95% confidence intervals were obtained based on 1,000 bootstrap samples for all the Cox regression analyses. T he formulatio n to estimate stabilized IPCW weights can be simplified as below for point interventions without time dependent confounding: swi= Pr [C ( k ) = 0 | C ( k 1 ) = 0 Z = Zi ] Pr [C ( k ) = 0 | C ( k 1 ) = 0 Z = Zi, X = Xi] t k = 0 C denotes the censoring status, where 0 refers to noncensored and 1 refers to censored. k denotes the time intervals, which was 1120 in the current study Z denotes the primary treatment received (definitive treatment or conservative treatment).

PAGE 45

45 Xi denotes the risk factors for outcome. Propensity Score Method Propensity score method was used in the previous studies comparing localized prostate cancer treatments. It was introduced by Rosenbaum and Rubin, defined as the conditional probability of receiving a particular treatment given a set of observed covariates .84 Assuming that the treatment and outcome are conditionally independent given the covariates, the difference between the two compared groups is an unbiased estimate of the average treatment effect, controlling for the single propensity score. A major adva nta ge of propensity score is to synthesize a large collection of covariates to a single summary measure. This is very useful when outcome s are rare and treatment is common, which was exactly the situation in the current study. The propensity score was estimated by using multivariate logistic regression. The dependent variable was the treatment received. The independent variables were potential confounding variables. It has been recommended to include all risk factors for the outcome into the propensity score model, regardless of their associations with treatment.85 While previous studies included comorbidity indices (e.g. 0 vs. 1 vs. 2 vs. 3 or more) in their pr opensity score models, we found diabetes and neurological disease/stroke could not be balanced in that way. Both comorbidities remained associated with treatment (P<0.05). Therefore, all the 11 individual diseases and the presence of other disease (0 vs. a t least 1) were included in the propensity score model. Below is the equation for propensity score calculation. Pi= e + 1X1+ 2X2+ + iXi1 + e + 1X1+ 2X2+ + iXi Pi denotes individual is propensity score, which was the probability of receiving definitive treatment as primary treatment given baseline characteristics. X1 Xi denote covariates, which were the sociodemographic, clinical and HRQOL variables at baseline.

PAGE 46

46 1 i denote the association parameters for X1 Xi. Two propensity score models were developed separately. The reference model included all sociodemographic and clinical variables that were commonly used in the previous studies. The HRQOL model includ ed the dichotomized baseline HRQOL variables additionally. Propensity scores generated from either model were compared in terms of distributions and confounder balancing performance. There are three commonly used techniques to adjust confounding by propensity scores: matching, stratification, and regression. To be consistent with the previous studies, we chose the regression adjustment and included the propensity score as a continuous covariate in the Cox proportional hazard model. Because the propensity s cores were skewed, t he analyses were repeated adjusting for propensity score quintiles as a categorical variable, which avoided the linearity assumption .86 The hazard ratios for all cause mortality were also examined within each propensity score quintile, looking for effe ct heterogeneity. We did not use propensity score matching for two main reasons: 1) the propensity scores did not overlap substantially. Matching might lose significant amount s of patients that did not have matches. It would be difficult to characterize th e finally matched patients, and therefore would not know to which population the study conclusion might generalize to. 2) We intended to compare the reference propensity score vs. HRQOL propensity core. Matching by two different propensity scores might res ult in two different matched samples. It would be difficult to tell whether any differences were due to the additional HRQOL adjustment or due to the different characteristics of the two matched samples. Weighting Methods The hazard ratio obtained from propensity score regression adjustment is a maximum likelihood estimate, which aims to maximize the probability of the data under a homogeneity

PAGE 47

47 assumption. It does not necessarily produce a causal estimate based on the counterfactual theory of causation, part icularly when effect modification presents 20 84 87 The confounding role of baseline HRQOL variables was investigated according to the counterfactual theory by using two weighting methods, inverse probability of treatmentweighting (IPTW) and standardized mortality ratio weighting (SMRW) 30 31 88 89 Both weighting methods create pseudopopulations in which covariates included in weight calculations are n o longer associated with the treatment. When exposure is a point treatment, which is the case in the present study, both weighting methods can be interpreted as multivariable standardization methods that use different standard populations. Given no unmeasured confounders and correct weighting model specification, the IPTW method estimates the average treatment effect in a population, the risk factor distributions of which are equal to those in the entire sample. The SMRW method estimates the average treatme nt effect in a population, the risk factor distributions of which are equal to those in one comparison group. The propensity score, IPTW and SMRW estimates would be similar unless the comparison groups have very different risk factor distributions that modify the treatment effect 32 8 4 87 The weights can be easily calculated from the propensity score. The equations are given below. When weighting to the entire sample, the weights were proportional to the inverse probability of receiving the treatment that was actually received. One concern was that a few patients can have extremely large weights and dominate the weighted analysis. A stabilized weight was used instead to mitigate this situation, which timed the weights by the overall probability of receiving the treatment in the entire sample. When using the SMRW method, we estimated the average effect in both the definitive treatment and conservative t reatment groups separately. When weighting to the definitive treatment group, the weights were 1 for men in the

PAGE 48

48 definitive treatment group and the propensity odds for men in the conservative treatment group. When weighting to the conservative treatment gro up, the weights were the inversed propensity odds for the definitive treatment group and 1 for the conservative treatment group. The weight calculation formulations are given below: Weighting to the entire sample: wDT=P Pi ; wCT=1 P 1 Pi Weighting to the definitive treatment group: wDT= 1 ; wCT=Pi1 Pi Weighting to the conservative treatment group: wDT=1 PiPi ; wCT= 1 WDT denotes the weight for men in the definitive treatment group WCT denotes the weight for men in the conservat ive treatment group P denotes the unconditional probability of receiving definitive treatment in the entire sample Pi denotes the propensity to receive the definitive treatment Two set weights were generated separately. One set was calculated by the reference propensity score, considering all the baseline sociodemographic and clinical variables. The other set was calculated by the HRQOL propensity score, considering the baseline HRQOL variables additionally. Two hazard ratios obtained by using either weight were compared to assess the potential confounding magnitude of baseline HRQOL variables per target population. W eighted survival curves were plot ted following the method developed by Westreich 90 for all the three weighting methods.

PAGE 49

49 CHAPTER 4 RESULTS Association between HRQOL and Primary Treatment Descriptive Analysis The baseline HRQOL data were available for 3,775 CaPSURE men. Their enrollment times were between March 21st, 1995 and October 11th, 2007. After the selection criteria were applied, a cohort of 1, 924 men were selected. They were all newly diagnos ed with localized prostate cancer at age of 65 years and older, without lymph nodes involvement or metastasis. Based on the 9 month primary treatment determination window, 676 men received radical prostatectomy (34.8%), 815 men received radiation therapy ( 35.7%), and 172 men received hormonal therapy (8.4%). The remaining 261 men were classified into the expectant management group (21.1%). The average follow up time from prostate cancer diagnosis to death or censoring (end of follow up or loss to follow up) was 4.4 years (SD=2.5). The follow up time appeared to be similar across the four treatment groups. The length of time between prostate cancer diagnosis and baseline data collection was also examined. In the entire cohort, the mean length of time was 44 days (SD=38) and the median was 33 days (IQR: 18 56). The length of time was similar across the four treatment groups. Baseline sociodemographic characteristics are presented in Table 41. There were small amount s of miss ing data in each variable except fo r age and diagnosis year. The proportion of m i ssin gn ess ranged from 2.0% in race to 12.7% in annual household income. The missing data pattern was similar across all four treatment groups, with slightly higher rate s in the expectant management group. The m ean age of the entire cohort was 72 years (SD=5) and the median age was 71 years old (IQR=6875). Regardless of missingness, the included men were predominantly white, married, covered by Medicare, and nonsmokers. Their education and household income

PAGE 50

50 looked relatively high More than 80% men were above high school graduates, and 36.0% men were college graduates or above. Only 14% men reported their annual household income below $20,000. The $20,000$50,000 category accounted for 42.4%, and the > $50,000 ca tegory accounted for 30.6%. About 12.7% annual household income data were missing. Approximately two third of the entire sample were diagnosed after 2001. Baseline clinical characteristics are presented in Table 4 2. Approximately 25% clinical data were m issing. The missing data pattern was similar across the four treatment groups, with slightly higher rate s in the expectant management group. According to the NCCN guideline classification, low and intermediate risk prostate cancer accounted for 40.4% and 41.3% in the entire cohort, respectively. Only 13.3% men had highrisk prostate cancer at diagnosis, while 4.9% men had incomplete cancer information. Obesity (BMI 29.9) accounted for 19.2% and 48.0%. Normal weight (BMI<25) accounted for 29.0%, while 3.6% BMI data were missing. Approximately 91% men reported one or more comorbidit ies Men having 1, 2, and 3 or more comorbidities accounted for 24.0%, 27.8%, and 36.8%, respectively. Missing data accounted for 2.7%. Regarding specifi c disease type, the most prevalent ones were hypertension (49.7%), arthritis (45.0%), and heart disease (27.7%). Moreover, 90.1% men reported at least one other type diseases that were not on the comorbidity checklist. Baseline HRQOL data are presented in Table 4 3, which included the Karnofsky performance score (KPS), eight Medical Outcome Short Form 36 (SF 36) scores, and six UCLA prostate cancer index scores (PCI). All these HRQOL variables were numerical, ranging from 0 100. Substantial missingness was observed for KPS, which accounted for 65.2% in the entire cohort. The missingness in SF 36 and PCI was less than 6.0%. The missing data pattern was

PAGE 51

51 similar across all four treatment groups, with slightly higher rate in the expectant management group. R egardless of the missingness, the included men appeared to have fairly good HRQOL at diagnosis for their age group. The median KPS was 100. The median SF 36 scores were all above or equal to the population median. The median PCI urinary and bowl scores wer e 100 or near 100. The two sexual related scores were low, with median values around 50. All HRQOL variables were dichotomized into high score vs. low score. The KPS and PCI scores were dichotomized at the sample median. The SF 36 scores were dichotomized at the US population median .22 Crude Associations between Baseline Variables and Primary Treatment The crude associations between the baseline characteristics and primary treatment were examined by using bivariate multinomial logistic regressions. Missing data were imputed by using the multiple imputation technique as described in the m ethod s ection. Table 4 4 shows the crude odds ratios and 95% confidence intervals for baseline sociodemographic variables. Radical prostatectomy was strongly associated with younger age, married, higher education level, higher household income level, and being di agnosed in 2001 or after. Besides, the radical prostatectomy group was more likely to be white than the hormonal therapy group, and more likely to have Medicare coverage than radiation therapy. The differences among the other three groups were much less pr onounced. Compared with expectant management and hormonal therapy, radiation therapy was associated with younger age and being married. The radiation therapy group was also more likely to have higher education level than the expectant management group. The two conservative treatment groups did not significantly differ from each other. Table 4 5 showed the crude odds ratios and 95% confidence intervals for the baseline clinical variables. Radical prostatectomy and radiation therapy were associated with higher prostate cancer risk level than expectant management, while hormonal therapy was strongly

PAGE 52

52 associated with higher prostate cancer risk than all the other three treatments. High BMI appeared to associate with the two definitive treatments than the two cons ervative treatments. The radical prostatectomy group had the least number of comorbidities than the other three treatment groups. Regarding individual comorbidity, the radical prostatectomy group was less likely to have heart disease, neurological disease/ stroke, other cancer, diabetes, hypertension, and lung disease. The differences among the other three treatment groups were less pronounced. Table 4 6 shows the crude odds ratios and 95% confidence intervals for the dichotomized baseline HRQOL variables. R adical prostatectomy was associated with high KPS than radiation and hormonal therapy. Radical prostatectomy was also significantly or borderline significantly associated with high scores for nearly all the SF 36 and PCI variables. Radiation therapy was as sociated with high HRQOL scores than the conservative treatments for most baseline HRQOL variables, although not as significant as radical prostatectomy. Compared with expectant management, radiation therapy group was more likely to have high scores for three SF 36 variables: physical functioning, role physical and general health. Compared with hormonal therapy, the radiation therapy group was more likely to have high scores for all SF 36 variables except for role emotional and mental health, Radiation ther apy group was also more likely to have high scores for three PCI variables: bowel functioning, bowel bother, and sexual functioning. There was little difference between expectant management and hormonal therapy except that the expectant group was more likely to have high KPS and sexual functioning scores. Adjusted Associations between Baseline HRQOL and Primary Treatment Multivariate analyses were conducted to investigate the independent associations between the baseline HRQOL variables and primary treatmen t, adjusting for sociodemographic and clinical factors. First, a preliminary multinomial logistic model was fit with all baseline sociodemographic and clinical variables by using a backward elimination procedure, while

PAGE 53

53 forcing in age and prostate cancer ri sk level. Sociodemographic variables retained in the preliminary model were age, race, marital status, income level, insurance type, and diagnosis year. Clinical variable s retained in the preliminary model were cancer risk level, heart disease, neurologica l disease/stroke, diabetes, lung disease, kidney disease, and urinary condition. Next, log likelihood tests were used to investigate whether any dichotomized baseline HRQOL variables were significantly associated with primary treatment when added into the preliminary model. Six HRQOL variables were identified, including KPS, four SF 36 variables and one PCI variable. The four SF 36 variables were physical functioning, role physical, general health, and vitality. The one PCI variable was sexual functioning. Table 4 7 presents the adjusted odds ratios and 95% confidence intervals. For men with high KPS, the estimated odds of receiving radical prostatectomy was 1.60 (95% CI: 1.002.57) times the estimated odds of receiving radiation therapy, and 1.86 (95% CI: 1.162.99) times the estimated odds of receiving hormonal therapy. High KPS was associated with a higher likelihood of receiving expectant management as primary treatment than hormonal therapy (OR: 1.81, 95% CI: 1.003.27). Regarding the four SF 36 variab les, high scores were significantly associated with higher likelihood of receiving radical prostatectomy as primary treatment than the other three treatments. For men with high physical functioning score, the estimated odds of receiving radical prostatecto my as primary treatment was 1.96 (95% CI: 1.412.71), 2.22 (1.44, 3.42), and 2.78 (95% CI: 1.75 4.41) times the odds of receiving radiation therapy, expectant management, and hormonal therapy as primary treatment, respectively. For men with high role physi cal score, the three odds ratios were 1.36 (95% CI: 1.02 1.81), 1.94 (95% CI: 1.30 2.88), and 1.66 (95% CI: 1.072.58). For a high general health score, the three odds ratios were 1.55 (95% CI: 1.15 2.09), 2.04 (95% CI: 1.36 3.04), and 2.35 (95% CI: 1.503.69). For a high vitality score, the three odds

PAGE 54

54 ratios were 1.74 (95% CI: 1.322.28), 2.14 (95% CI: 1.453.17), and 2.62 (95% CI: 1.714.02). Men with high role physical were also 1.43 (95% CI: 1.021.99) times more likely to receive radiation than expectant management. Men with high general health and vitality scores were also 1.52 times (95% CI: 1.03 2.23) and 1.51 times (95% CI: 1.04219) more likely to receive radiation therapy than hormonal therapy, respectively. Expectant management and hormonal therapy did not differ in any SF 36 variables after covariate adjustment. Sexual function was the only PCI variable found to independently associate with treatment selection, adjusting for sociodemographic and clinical factors. Men with high sexual function sco re w ere 1.35 (95% CI: 1.061.72) times more likely to receive radical prostatectomy than radiation therapy, 1.77 (95% CI: 1.172.70) times more likely to receive radical prostatectomy than hormonal therapy, and 1.71 (95% CI: 1.09 2.67) time more likely to receive expectant management than hormonal therapy. Lastly, the most predictive HRQOL variables were determined by fitting all HRQOL variables into the preliminary model simultaneously. A backward elimination procedure was used, while forcing in all the pr eselected sociodemographic and clinical variables. Table 48 shows the odds ratios and 95% confidence intervals for all the baseline variables retained in the final model. Three baseline HRQOL variables were retained in the final model, which were KPS, phy sical functioning, and vitality. Their associations with primary treatment were similar to those obtained from the previous models. Men with high KPS scores were 1.75 (95% CI: 1.10 2.78) times more likely to receive radical prostatectomy than hormonal ther apy. Men with high physical functioning score were 1.62 (95% CI: 1.142.31), 1.74 (95% CI: 1.092.77), and 2.03 (95% CI: 1.24 3.33) times more likely to receive radical prostatectomy than radiation therapy, expectant management, and hormonal therapy, respe ctively. Men with high vitality score were

PAGE 55

55 1.49 (95% CI: 1.12 2.00), 1.83 (95% CI: 1.21 2.77) and 2.07 (95% CI: 2.011.243.27) times more likely to receive radical prostatectomy than the other three treatments, respectively. None of the three retained HRQ OL variables were significantly associated with treatments when choosing from radiation therapy, expectant management, or hormonal therapy. The C onfounding Effect of B aseline HRQOL The associations between definitive treatment, including both radical prost atectomy and radiation therapy, and 10 year all cause mortality were compared before and after adjustment of the baseline HRQOL variables in addition to comprehensive sociodemographic and clinical variables. Since substantial distinctions were founded betw een men treated with radical prostatectomy and radiation therapy in part I, radical prostatectomy and radiation therapy were further compared with conservative treatment separately. The associations with 10 year prostate cancer specific and other cause mor tality were also examined to assess the plausibility of the findings Descriptive and Univariate Analysis The final sample included 1,700 men aged 65 years old or above, who were newly diagnosed with low or intermediate risk prostate cancer. Out of the entire sample, 611 men received radical prostatectomy, 727 men received radiation therapy, 228 men received expectant management, and 134 men received hormonal therapy. The baseline characteristics distributions and crude associations with primary treatment were similar to the findings in Part I. The missingness on baseline characteristic variables was minor except for KPS. The missing data rate was similar across the groups. The entire sample was divided into a definitive treatment group and a conservative treatment group, and the baseline characteristics were compared. The definitive treatment group consisted of 1,338 men treated with either radical prostatectomy or radiation therapy. The

PAGE 56

56 conservative treatment group consisted of 362 men treated with either expectant management or hormonal therapy. The mean length of follow up since diagnosis was similar between the comparison groups, which was 4.4 years. The observed all cause mortality rate was 13.4% (228 cases) in the entire sample, 10.6% in the definitive treatment group (142 cases), and 23.8% (86 cases) in the conservative treatment group. Most deaths were due to reasons other than prostate cancer. The prostate cancer specific mortality rate was only 1.4% in the entire sample (24 cases), which accounted for 10.5 % all cause death. The definitive treatment group had a lower prostate cancer specific mortality rate than the conservative treatment group, which was 1.2% (16 cases) vs. 2.2% (8 cases). Tab le 4 9 and Table 410 summarize the descriptive statistic s for baseline sociodemographic and clinical variables. Their associations with treatment were examined by using logistic regressions after missing data were imputed by using the multiple imputation approach. The definitive treatment was associated with yo unger age, being married, higher education level, higher annual household income, higher tumor grade, and less comorbidities. Regarding specific disease types, men in the definitive treatment group were less likely to have heart disease, neurological disease/stroke, diabetes, and urinary c onditions. Table 411 summarizes the descriptive statistics for dichotomized baseline HRQOL variables, and their associations with treatment. Definitive treatment was strongly associated with high SF 36 scores except for m ental health. Definitive treatment was also strongly associated with high PCI scores except for urinary function and bowel function. Tabl e 4 12 and Table 413 summarize the results of univariate Cox regression analyses for all baseline sociodemographic an d clinical variables. With regard to the sociodemographic variables, older age and current smoker at diagnosis were significantly associated with increased

PAGE 57

57 hazards for all cause and other cause mortality. Higher annual household income was significantly as sociated with decreased hazard for all cause and other cause mortalities. Few sociodemographic variables were associated with prostate cancer specific mortality. Clinical variables that significantly associated with allcause and other cause mortality were heart disease, neurological disease/stroke, diabetes, and blood disease. Although not statistically significant, a dose response relationship was observed for the number of comorbidities. The hazards for all cause and other cause mortality increased with the number of comorbidities. The only variable that significantly associated with prostate cancer specific mortality was tumor grade. A higher grade was associated with higher hazard for prostate cancer specific mortality. Table 4 16 summarizes the univar iate Cox regression analysis results for the baseline HRQOL variables. High scores of KPS and all the eight SF 36 variables were strongly associated with decreased hazards for all cause and other cause mortality. High role emotional score was associated wi th decreased hazard for prostate cancer specific mortality. With regard to the PCI variables, high scores of bowel bother, sexual function, and sexual bother were significantly associated with decreased hazards for all cause and other cause mortality. High urinary bother score was also associated with decreased all cause mortality. Few baseline PCI variables were associated with prostate cancer specific mortality. Propensity Score Comparison s Two propensity score models were developed separately. The reference model included all the baseline sociodemographic and clinical variables. The HRQOL model included the same set of variables, plus the dichotomized baseline HRQO L variables. Figure 4 1 compares the propensity score distributions for definitive treatment vs. conservative treatment. As expected, the propensity scores for the definitive treatment group were higher than those for the conservative treatment group. The definitive treatment curves were skewed towards 1, while the

PAGE 58

58 conservative treatment curves s pread more evenly. There were some overlaps, regardless of the baseline HRQOL consideration. The mean reference propensity to receive definitive treatment in men actually treated definitively was 0.83 (SD=0.15), compared with 0.62 (SD=0.22) in men actually treated conservatively. The two HRQOL curves were slightly further apart after the two reference curves. The mean HRQOL propensity was 0.83 (SD=0.15) vs. 0.61 (SD=0.23). The shifting indicated that the reference model systematically underestimated the pro pensity discrepancy between the two comparison groups. Figure 4 2 compares the propensity scores for radical prostatectomy vs. conservative treatment. The propensity to receive radical prostatectomy was very different between these two groups, regardless of the baseline HRQOL consideration. The radical prostatectomy curves were highly skewed towards 1 while the conservative treatment curves were highly skewed towards 0. The overlaps were limited. The mean reference propensity was 0.82 (SD= 0.19) vs. 0.30 (S D=0.30). The HRQOL propensity curves were slightly further apart. The mean propensity was 0.83 (SD=0.19) vs. 0.28 (SD=0.29). Figure 43 compared the propensity scores for radiation therapy vs. conservative treatment. Opposite to figure 42, the propensity curves were much less skewed, regardless of the baseline HRQOL consideration. The overlaps were moderate. The mean reference propensity was 0.71 (SD=0.15) vs. 0.58 (SD=0.18). The mean HRQOL propensity was 0.72 (SD=0.15) vs. 0.57 (SD=0.19). Confounding Bal ance by Propensity Score Logistic regression models were used to determine whether the baseline characteristics were balanced by propensity score. It was found that all the baseline variables were well balanced by the HRQOL score for all the treatment comp arisons. The reference score balanced all the sociodemographic and clinical variables, but not a few baseline HRQOL variables. As shown in Table 415, the reference propensity score poorly balanced four SF 36 variables in

PAGE 59

59 either dichotomized or numeric for ms between definitive and conservative treatments, which were physical functioning, role physical, general health, and vitality (P<0.05). All the unbalanced HRQOL variables had odds ratios above 1.00, indicating that definitive treatment remained associated wit h high scores. Table 4 16 shows that more baseline HRQOL variables were unbalanced by the reference score when comparing radical prostatectomy and conservative treatment. Besides the four SF 36 variables mentioned above, the unbalanced variables include numeric KPS, dichotomized bodily pain, and numeric sexual function. Table 4 17 shows that the HRQOL variables were much better balanced by the reference score when comparing radiation therapy and conservative treatment. The only unbalance variables were role physical in both dichotomized and numeric forms, and numeric general health. Propensity Score Quintile Specific Hazard Ratios Table 4 18 and Table 419 show the propensity score quintile specific hazard ratios for all cause mortality. The hazard ratios were consistently below 1.00 when comparing definitive treatment and radiation therapy vs. conservative treatment. Heterogeneity was observed when comparing radical prostatectomy vs. conservative treatment. The first quintile hazard ratios were above 1.00 while the hazard ratios in the other quintiles were below 1.00. It indicated that the survival benefit of radical prostatectomy was modified by propensity scor e. Confounding Effect of Baseline HRQOL Cox proportional hazard model was used to compare the 10 year mortality outcomes between definitive and the conservative treatment. Radical prostatectomy and radiation therapy were also compared with conservative tr eatment separately. T he log log curves for each treatment comparison are presented from Figure 4 4 to Figure 46. The curves were parallel to each other for all the three mortality outcomes. The proportional hazard assumptions were satisfied. Potential bia s due to nonrandom censoring was controlled by using the IPCW method

PAGE 60

60 as described in the methodology section. The results obtained by different confounding adjustment methods are summarized in Table 4 20 to Table 422. The adjusted hazard ratios for 10ye ar all cause mortality are also summarized from Figure 4 7 to F igure 4 9. Definitive t reatment vs. c onservative t reatment The crude hazard ratio was 0.35 (95% CI: 0.16 0.77), indicating strong association between definitive treatment and longer survival. The hazard ratio increased toward 1.00 after adjusting for baseline sociodemographic and clinical variables, and further increased slightly after additional HRQOL adjustment. All confounding adjustment methods yielded similar results. Before and after the HRQOL consideration, the adjusted hazard ratio was 0.44 (95% CI: 0.191.05) vs. 0.46 (95% CI: 0.191.08) when adjusting the continuous propensity score; 0.43 (95% CI: 0.19 0.91 ) vs. 0.48 (95% CI: 0.23 1.00 ) when adjusting the propensity score quintiles; 0.43 (95% CI: 0.19 0.96 ) vs. 0.46 (95% CI: 0.20 01.04) when weighting to the entire sample; 0.42 (95% CI: 0.18 0.97 ) vs. 0.44 (95% CI: 0.191.05) when weighting to the definitive treatment group, and 0.50 (95% CI: 0. 24 1.05) vs. 0.51 (95% CI: .0.24 1.09 ) when weighting to the conservative treatment group. The relative changes in hazard ratio ranged from 1.9% to 9.3%. The absolute changes were minor, ranging from 0.01 to 0.04. The confounding effect is not significant. We separately compared the pros tate cancer specific mortality and other cause mortality, looking for clues of unmeasured confounding. The crude hazard ratio for prostate cancer specific mortality was 0.30 (95% CI: 0.081.07). The hazard ratio ranged from 0.19 to 0.35 after the baseline sociodemographic and clinical variables were adjusted, depending on the confounding adjustment method used. Additional HRQOL adjustment further increased the hazard ratios slightly. The relative changes ranged from 1.4% to 4.4%. The absolute differences we re less than 0.02. The hazard ratios for other cause mortality had values close to those for all cause mortality

PAGE 61

61 since it accounted for 90% all cause deaths. The crude hazard ratio was 0.36 (95% CI: 0.16 0.79). The hazard ratios obtained from different met hods were similar, ranging from 0.46 0.52 after adjusting the baseline sociodemographic and clinical factors. Additional HRQOL adjustment slightly increased the hazard ratios more towards 1.00. The relative changes ranged from 1.8% to 10.2%, and the abs olute changes ranged from 0.01 to 0.05. The adjusted hazard ratios for other cause mortality were more close to 1.00 than those for prostate cancer specific mortality, regardless of the confounding adjustment method used. The results above indicated that definitive treatment was strongly associated with allcause survival benefit relative to conservative treatment. Although attenuated, the association remained strong after confounding adjustment. Additional HRQOL adjustment had small impact. The relative ch anges were less than 10%, regardless of the confounding adjustment method used. The variation in hazard ratio resulting from different methods was small. The strength of the association between definitive treatment and survival was stronger for prostate cancer death than other cause death, which was plausible. Nevertheless, it was still questionable that definitive treatment for prostate cancer could reduce the hazard for other cause death for nearly 50%. Figure 4 10 presented the crude and weighted surviva l curves for 10 year all cause mortality. The gap between the definitive treatment and conservative treatment curves became slightly narrower after the baseline HRQOL variables were adjusted additionally. Radical p rostatectomy vs. c onservative t reatment The crude hazard ratio for all cause mortality was 0.27 (95% CI: 0.12 0.64). All hazard ratios were below 1.00 after adjustment of baseline sociodemographic and clinical variables. Additional HRQOL adjustment greatly increased all hazard ratios for 20 30% depending on the method used. The absolute changes ranged from 0.08 to 0.24. Different confounding adjustment methods produced somewhat different estimates. When weighting to the radical prostatectomy

PAGE 62

62 group, the hazard ratio was 0.36 (95% CI: 0.111.16) vs. 0.45 (0.131.51) before and after additional HRQOL adjustment. Similar hazard ratios were obtained when using the propensity score adjustment. Higher hazard ratio was obtained when weighting to the entire sample, which was 0.57 (95% CI: 0.182.78) vs. 0.81 (95% CI: 0.232.81) before and after additional HRQOL adjustment. The highest hazard ratio was obtained when weighting to the conservative treatment group, which was 0.83 (95% CI: 0.18 3.85) and 1.03 (94% CI: 0.21 5.19) before and after additional H RQOL adjustment. The crude hazard ratio for prostate cancer specific mortality was 0.19 (95% CI: 0.013.93). The hazard ratios ranged from 0.12 to 0.29 after adjusting for the baseline sociodemographic and clinical variables. The prostate cancer survival benefit was further strengthened after additional HRQOL adjustment as all hazard ratios decreased further away from 1.00. Although the relative changes appeared to be substantial, the absolute changes were minor ranging from 0.03 to 0.06. The crude hazar d ratio for other cause mortality was 0.28 (95% CI: 0.110.69). The adjusted hazard ratios from different methods had the same pattern with those for all cause mortality. All hazard ratios remained below 1.00 after adjusting for baseline sociodemographic a nd clinical variables. Additional HRQOL adjustment significantly increased all hazard ratios for 20% 30%. Radical prostatectomy associated other cause survival benefit diminished when weighting to the conservative treatment group. The association between radical prostatectomy and survival was stronger for prostate cancer specific death than other cause death, regardless of the confounding adjustment method used. The results above indicated that the baseline HRQOL variables confounded the association between radical prostatectomy and all cause mortality. Radical prostatectomy associated survival benefit was overestimated for approximately 20 30%, depending on the confounding adjustment

PAGE 63

63 method used. The overestimation seemed owing to the overestimated other cause survival but not prostate cancer specific survival. It was also noticed that different confounding adjustment methods yielded different estimates. After additional adjustment of the baseline HRQOL variables, the all cause survival benefit disappeared when weighting to the conservative treatment group (hazard ratio: 1.03, 95% CI: 0.215.19), but remained strong when weighting to the radical prostatectomy group (hazard ratio: 0.45, 95% CI: 0.131.51). The results revealed that the treatment effect varie d in the different target populations. Radical prostatectomy might prolong survival in the population represented by the radical prostatectomy group, but might not in the population represented by the conservative treatment group. Figure 4 11 presents the crude and weighted survival curves. The gap between the two treatment curves were narrower after the baseline HRQOL variables were adjusted additionally. Radiation t herapy vs. c onservative t reatment The crude hazard ratio was 0.43 (95% CI: 0.200.95). All confounding adjustment methods yielded very similar estimates after adjusting for baseline sociodemographic and clinical variables, which ranged from 0.47 to 0.50. Additional HRQOL adjustment had little impacts. All hazard ratios remained nearly unchanged. The relative differences were less than 2% and the absolute differences were equal or less than 0.01. The crude hazard ratio for prostate cancer mortality was 0.38 (95% CI: 0.101.42). The hazard ratio ranged from 0.24 to 0.41, adjusting for the baseline sociodemographic and clinical variables. Additional HRQOL adjustment had minor impacts. The relative changes were less than 7.1% and the absolute changes were less than 0.03. The crude hazard ratio for other cause mortality was 0.43 (95% CI: 0.191.00). The hazard ratio ranged from 0.47 to 0.52, adjusting for the baseline sociodemographic and clinical variables. Additional HRQOL adjustment had little

PAGE 64

64 impact. The adjusted hazard ratios for other cause mortality were closer to 1.00 than those for prostate can cer specific mortality, regardless of the confounding adjustment method used. The results indicated that the baseline HRQOL variables did not confound the association between radiation therapy and all cause mortality. Additional adjustment of HRQOL had little impact on the estimates. The strength of the association between radiation therapy and survival was stronger for prostate cancer death than other cause death, which was plausible. Nevertheless, it was questionable that radiation therapy could reduce ot her cause mortality for nearly 50%. Figure 4 13 presents the crude and weighted survival curves. The gap between the two treatment curves nearly unchanged after the baseline HRQOL variables were considered additionally .

PAGE 65

65 Table 4 1. Basel ine s ociodemographic variable distributions in CaPSURE elderly men with localized prostate cancer Overall, n (%) (N=1,924) RP, n (%) (N=676) RT, n (%) (N=815) EM, n (%) (N=261) HT, n (%) (N=172) Age Mean (SD), y 72 (5) 69 (3) 72 (5) 75 (5) 76 (6) Median (IQR), y 71 (68 75) 68 (66 70) 73 (69 76) 75 (71 79) 76 (71 80) 65 67 446 (23.2) 279 (41.3) 132 (16.2) 23 (8.8) 12 (7.0) 6 8 70 475 (24.7) 243 (35.9) 179 (22.0) 32 (12.3) 21 (12.2) 7 1 73 327 (17.0) 108 (16.0) 156 (19.1) 38 (14.6) 25 (14.5) 74 76 310 (16.1) 34 (5.0) 187 (22.9) 59 (22.6) 30 (17.4) > =77 366 (19.0) 12 (1.8) 161 (19.8) 109 (41.8) 84 (48.8) Race White 1752 (91.1) 628 (92.9) 745 (91.4) 227 (87.0) 152 (88.4) O ther 128 (6.7) 41 (6.1) 53 (6.5) 16 (6.1) 18 (10.5) Missing 44 (2.3) 7 (1.0) 17 (2.1) 18 (6.9) 2 (1.2) Marital status Married 1612 (83.8) 612 (90.5) 681 (83.6) 192 (73.6) 127 (73.8) S ingle 256 (13.3) 55 (8.1) 110 (13.5) 52 (19.9) 39 (22.7) Missing 56 (2.9) 9 (1.3) 24 (2.9) 17 (6.5) 6 (3.5) Ed ucation < HS 305 (15.9) 75 (11.1) 139 (17.1) 57 (21.8) 34 (19.8) HS / SC 872 (45.3) 297 (43.9) 369 (45.3) 120 (46.0) 86 (50.0) 693 (36.0) 295 (43.6) 282 (34.6) 67 (25.7) 49 (28.5) Missing 54 (2.8) 9 (1.3) 25 (3.1) 17 (6.5) 3 (1.7) An nual house hold Income <20K 270 (14.0) 61 (9.0) 130 (16.0) 47 (18.0) 32 (18.6) 2 0K 50K 816 (42.4) 277 (41.0) 365 (44.8) 104 (39.8) 70 (40.7) >50K 588 (30.6) 272 (40.2) 215 (26.4) 54 (20.7) 47 (27.3) M issing 250 (13.0) 66 (9.8) 105 (12.9) 56 (21.5) 23 (13.4) Insurance Medicare 1505 (78.2) 551 (81.5) 625 (76.7) 202 (77.4) 127 (73.8) O ther 355 (18.5) 103 (15.2) 172 (21.1) 45 (17.2) 35 (20.3) Missing 64 (3.3) 22 (3.3) 18 (2.2) 14 (5.4) 10 (5.8) Smoking status Current 117 (6.1) 46 (6.8) 46 (5.6) 11 (4.2) 14 (8.1) Not current 1746 (90.7) 618 (91.4) 742 (91.0) 230 (88.1) 156 (90.7) Missing 61 (3.2) 12 (1.8) 27 (3.3) 20 (7.7) 2 (1.2) Diagnosis year 1995 2000 607 (31.5) 173 (25.6) 273 (33.5) 100 (38.3) 61 (35.5) 2001 2007 1317 (68.5) 503 (74.4) 542 (66.5) 161 (61.7) 111 (64.5)

PAGE 66

66 Table 4 2. Baseline c linical variable distributions in CaPSURE elderly men with localized prostate cancer Overall, n (%) (N=1,924) RP, n (%) (N=676) RT, n (%) (N=815) EM, n (%) (N=261) HT, n (%) (N=172) Cancer risk Low 778 (40.4) 277 (41.0) 320 (39.3) 137 (52.5) 44 (25.6) I ntermediate 794 (41.3) 294 (43.5) 362 (44.4) 64 (24.5) 74 (43.0) High 255 (13.3) 72 (10.7) 109 (13.4) 26 (10.0) 48 (27.9) Missing 97 (5.0) 33 (4.9) 24 (2.9) 34 (13.0) 6 (3.5) BMI <25 558 (29.0) 174 (25.7) 232 (28.5) 95 (36.4) 57 (33.1) 25 29.9 923 (48.0) 349 (51.6) 396 (48.6) 105 (40.2) 73 (42.4) >=30 369 (19.2) 138 (20.4) 156 (19.1) 39 (14.9) 36 (20.9) Missing 74 (3.8) 15 (2.2) 31 (3.8) 22 (8.4) 6 (3.5) Heart dis ease Yes 533 (27.7) 117 (17.3) 265 (32.5) 87 (33.3) 64 (37.2) No 1340 (69.6) 549 (81.2) 532 (65.3) 156 (59.8) 103 (59.9) Missing 51 (2.7) 10 (1.5) 18 (2.2) 18 (6.9) 5 (2.9) Neurol ogic disease/stroke Yes 158 (8.2) 31 (4.6) 76 (9.3) 28 (10.7) 23 (13.4) No 1715 (89.1) 635 (93.9) 721 (88.5) 215 (82.4) 144 (83.7) Missing 51 (2.7) 10 (1.5) 18 (2.2) 18 (6.9) 5 (2.9) Other cancer Yes 312 (16.2) 94 (13.9) 146 (17.9) 36 (13.8) 36 (20.9) No 1561 (81.1) 572 (84.6) 651 (79.9) 207 (79.3) 131 (76.2) Missing 51 (2.7) 10 (1.5) 18 (2.2) 18 (6.9) 5 (2.9) Diabetes Yes 235 (12.2) 60 (8.9) 99 (12.1) 41 (15.7) 35 (20.3) No 1638 (85.1) 606 (89.6) 698 (85.6) 202 (77.4) 132 (76.7) Missing 51 (2.7) 10 (1.5) 18 (2.2) 18 (6.9) 5 (2.9) Kidney di sease Yes 104 (5.4) 35 (5.2) 41 (5.0) 24 (9.2) 4 (2.3) No 1769 (91.9) 631 (93.3) 756 (92.8) 219 (83.9) 163 (94.8) Missing 51 (2.7) 10 (1.5) 18 (2.2) 18 (6.9) 5 (2.9) Hypertension Yes 957 (49.7) 302 (44.7) 434 (53.3) 122 (46.7) 99 (57.6) N o 916 (47.6) 364 (53.8) 363 (44.5) 121 (46.4) 68 (39.5) Missing 51 (2.7) 10 (1.5) 18 (2.2) 18 (6.9) 5 (2.9)

PAGE 67

67 Table 4 2.Continued Overall, n (%) (N=1,924) RP, n (%) (N=676) RT, n (%) (N=815) EM, n (%) (N=261) HT, n (%) (N=172) Arthritis Yes 865 (45.0) 290 (42.9) 382 (46.9) 111 (42.5) 82 (47.7) No 1008 (52.4) 376 (55.6) 415 (50.9) 132 (50.6) 85 (49.4) Missing 51 (2.7) 10 (1.5) 18 (2.2) 18 (6.9) 5 (2.9) Blood disease Yes 48 (2.5) 12 (1.8) 26 (3.2) 4 (1.5) 6 (3.5) No 1825 (94.9) 654 (96.7) 771 (94.6) 239 (91.6) 161 (93.6) Missing 51 (2.7) 10 (1.5) 18 (2.2) 18 (6.9) 5 (2.9) GI disease Yes 302 (15.7) 105 (15.5) 123 (15.1) 47 (18.0) 27 (15.7) No 1571 (81.7) 561 (83.0) 674 (82.7) 196 (75.1) 140 (81.4) Missing 51 (2.7) 10 (1.5) 18 (2.2) 18 (6.9) 5 (2.9) Lung disease Yes 181 (9.4) 47 (7.0) 84 (10.3) 35 (13.4) 15 (8.7) No 1692 (87.9) 619 (91.6) 713 (87.5) 208 (79.7) 152 (88.4) Missing 51 (2.7) 10 (1.5) 18 (2.2) 18 (6.9) 5 (2.9) Urinary condition Yes 359 (18.7) 119 (17.6) 141 (17.3) 50 (19.2) 49 (28.5) No 1514 (78.7) 547 (80.9) 656 (80.5) 193 (73.9) 118 (68.6) Missing 51 (2.7) 10 (1.5) 18 (2.2) 18 (6.9) 5 (2.9) Other disease 0 140 (7.3) 52 (7.7) 62 (7.6) 11 (4.2) 15 (8.7) 1733 (90.1) 614 (90.8) 735 (90.2) 232 (88.9) 152 (88.4) Missing 51 (2.7) 10 (1.5) 18 (2.2) 18 (6.9) 5 (2.9) Total n umber of c omorbidity 0 169 (8.8) 83 (12.3) 56 (6.9) 16 (6.1) 14 (8.1) 1 461 (24.0) 198 (29.3) 185 (22.7) 54 (20.7) 24 (14.0) 2 535 (27.8) 197 (29.1) 238 (29.2) 66 (25.3) 34 (19.8) 708 (36.8) 188 (27.8) 318 (39.0) 107 (41.0) 95 (55.2) Missing 51 (2.7) 10 (1.5) 18 (2.2) 18 (6.9) 5 (2.9)

PAGE 68

68 Table 4 3. Baseline HRQOL variable distributions in CaPSURE elderly men with localized prostate cancer Overall, n (%) (N=1,924) RP, n (%) (N=676) RT, n (%) (N=815) EM, n (%) (N=261) HT, n (%) (N=172) KPS Mean (SD) 98 (5) 99 (4) 97 (6) 98 (4) 95 (7) Median (IQR) 100 (100 100) 100 (100 100) 100 (100 100) 100 (100 100) 100 (90 100) High score 535 (27.8) 184 (27.2) 213 (26.1) 90 (34.5) 48 (27.9) Low score 134 (7.0) 21 (3.1) 66 (8.1) 18 (6.9) 29 (16.9) Missing 1255 (65.2) 471 (69.7) 536 (65.8) 153 (58.6) 95 (55.2) S F 36 PF Mean (SD) 81 (22) 88 (16) 79 (23) 75 (25) 71 (26) Median (IQR) 90 (75 100) 95 (85 100) 90 (70 95) 80 (61 95) 80 (55 90) High score 1418 (73.7) 586 (86.7) 570 (69.9) 167 (64.0) 95 (55.2) Low score 462 (24.0) 78 (11.5) 223 (27.4) 91 (34.9) 70 (40.7) Missing 44 (2.3) 12 (1.8) 22 (2.7) 3 (1.1) 7 (4.1) RP Mean (SD) 74 (38) 83 (32) 72 (38) 64 (41) 62 (43) Median (IQR) 100 (50 100) 100 (75 100) 100 (50 100) 75 (25 100) 75 (25 100) High score 1322 (68.7) 539 (79.7) 542 (66.5) 146 (55.9) 95 (55.2) Low score 564 (29.3) 127 (18.8) 256 (31.4) 111 (42.5) 70 (40.7) Missing 38 (2.0) 10 (1.5) 17 (2.1) 4 (1.5) 7 (4.1) BP Mean (SD) 82 (21) 86 (19) 82 (21) 79 (22) 78 (23) Median (IQR) 90 (68 100) 90 (78 100) 90 (68 100) 90 (68 100) 90 (58 100) High score 1366 (71.0) 526 (77.8) 569 (69.8) 165 (63.2) 106 (61.6) Low score 523 (27.2) 142 (21.0) 230 (28.2) 88 (33.7) 63 (36.6) Missing 35 (1.8) 8 (1.2) 16 (2.0) 8 (3.1) 3 (1.7) GH Mean (SD) 71 (19) 76 (17) 70 (20) 66 (21) 64 (21) Median (IQR) 75 (60 85) 80 (65 90) 75 (60 85) 70 (50 80) 70 (45 80) High score 1353 (70.3) 547 (80.9) 553 (67.9) 157 (60.2) 96 (55.8) Low score 528 (27.4) 120 (17.8) 242 (29.7) 95 (36.4) 71 (41.3) Missing 43 (2.2) 9 (1.3) 20 (2.5) 9 (3.4) 5 (2.9)

PAGE 69

69 Table 4 3.Continued Overall, n (%) (N=1,924) RP, n (%) (N=676) RT, n (%) (N=815) EM, n (%) (N=261) HT, n (%) (N=172) RE Mean (SD) 84 (32) 89 (26) 83 (33) 80 (34) 79 (37) Median (IQR) 100 (100 100) 100 (100 100) 100 (83 100) 100 (67 100) 100 (67 100) High score 1440 (74.8) 545 (80.6) 600 (73.6) 177 (67.8) 118 (68.6) Low score 444 (23.1) 119 (17.6) 200 (24.5) 78 (29.9) 47 (27.3) Missing 40 (2.1) 12 (1.8) 15 (1.8) 6 (2.3) 7 (4.1) VT Mean (SD) 66 (19) 70 (18) 65 (20) 62 (19) 59 (21) Median (IQR) 70 (50 80) 75 (60 85) 70 (50 80) 60 (50 75) 60 (45 75) High score 1282 (66.6) 525 (77.7) 520 (63.8) 149 (57.1) 88 (51.2) Low score 616 (32.0) 145 (21.4) 283 (34.7) 109 (41.8) 79 (45.9) Missing 26 (1.4) 6 (0.9) 12 (1.5) 3 (1.1) 5 (2.9) SO Mean (SD) 88 (19) 91 (17) 88 (19) 86 (21) 84 (23) Median (IQR) 100 (75 100) 100 (88 100) 100 (75 100) 100 (75 100) 100 (75 100) High score 1205 (62.6) 463 (68.5) 500 (61.3) 150 (57.5) 92 (53.5) Low score 674 (35.0) 201 (29.7) 294 (36.1) 103 (39.5) 76 (44.2) Missing 45 (2.3) 12 (1.8) 21 (2.6) 8 (3.1) 4 (2.3) MH Mean (SD) 81 (15) 82 (15) 80 (15) 81 (15) 79 (17) Median (IQR) 84 (72 92) 84 (76 92) 84 (72 92) 84 (72 92) 84 (72 92) High score 1078 (56.0) 409 (60.5) 436 (53.5) 144 (55.2) 89 (51.7) Low score 817 (42.5) 261 (38.6) 365 (44.8) 114 (43.7) 77 (44.8) Missing 29 (1.5) 6 (0.9) 14 (1.7) 3 (1.1) 6 (3.5) UCLA PCI UF Mean (SD) 91 (14) 92 (13) 91 (13) 88 (18) 88 (17) Median (IQR) 100 (82 100) 100 (87 100) 100 (83 100) 100 (80 100) 100 (80 100) High score 1108 (57.6) 415 (61.4) 457 (56.1) 147 (56.3) 89 (51.7) Low score 772 (40.1) 249 (36.8) 336 (41.2) 112 (42.9) 75 (43.6) Missing 44 (2.3) 12 (1.8) 22 (2.7) 2 (0.8) 8 (4.7)

PAGE 70

70 Table 4 3.Continued Overall, n (%) (N=1,924) RP, n (%) (N=676) RT, n (%) (N=815) EM, n (%) (N=261) HT, n (%) (N=172) UB Mean (SD) 81 (26) 84 (24) 81 (26) 78 (28) 76 (30) Median (IQR) 100 (75 100) 100 (75 100) 100 (75 100) 100 (75 100) 100 (50 100) High score 1061 (55.1) 398 (58.9) 450 (55.2) 131 (50.2) 82 (47.7) Low score 821 (42.7) 267 (39.5) 346 (42.5) 127 (48.7) 81 (47.1) Missing 42 (2.2) 11 (1.6) 19 (2.3) 3 (1.1) 9 (5.2) BF Mean (SD) 87 (14) 89 (14) 87 (14) 87 (15) 84 (17) Median (IQR) 94 (81 100) 94 (85 100) 94 (80 95) 94 (81 100) 89 (75 94) High score 1085 (56.4) 448 (55.0) 448 (55.0) 140 (53.6) 79 (45.9) Low score 826 (42.9) 359 (44.0) 359 (44.0) 117 (44.8) 92 (53.5) Missing 13 (0.7) 8 (1.0) 8 (1.0) 4 (1.5) 1 (0.6) BB Mean (SD) 87 (22) 90 (19) 86 (22) 85 (22) 80 (27) Median (IQR) 100 (75 100) 100 (75 100) 100 (75 100) 100 (75 100) 100 (75 100) High score 1278 (66.4) 497 (73.5) 528 (64.8) 157 (60.2) 96 (55.8) Low score 623 (32.4) 175 (25.9) 272 (33.4) 101 (38.7) 75 (43.6) Missing 23 (1.2) 4 (0.6) 15 (1.8) 3 (1.1) 1 (0.6) SF Mean (SD) 41 (28) 49 (28) 38 (28) 37 (27) 28 (24) Median (IQR) 40 (17 66) 53 (27 75) 35 (14 61) 32 (10 61) 24 (7 46) High score 868 (45.1) 389 (57.5) 332 (40.7) 102 (39.1) 45 (26.2) Low score 997 (51.8) 273 (40.4) 455 (55.8) 149 (57.1) 120 (69.8) Missing 59 (3.1) 14 (2.1) 28 (3.4) 10 (3.8) 7 (4.1) SB Mean (SD) 54 (39) 59 (37) 52 (39) 48 (40) 51 (42) Median (IQR) 50 (25 100) 75 (25 100) 50 (0 100) 50 (0 100) 50 (0 100) High score 1097 (57.0) 434 (64.2) 450 (55.2) 126 (48.3) 87 (50.6) Low score 712 (37.0) 214 (31.7) 315 (38.7) 115 (44.1) 68 (39.5) Missing 115 (6.0) 28 (4.1) 50 (6.1) 20 (7.7) 17 (9.9)

PAGE 71

71 Table 4 4. Crude estimate s of association s between baseline sociodemographic variables and primary treatment (odds ratios and 95% confidence interval s ) RP vs. RT RP vs. EM RP vs. HT Age 65 67 (ref) 1.00 1.00 1.00 68 70 0.70 (0.52, 0.95)* 0.49 (0.32, 0.75) *** 0.50 (0.24, 1.03) 71 73 0.33 (0.24, 0.47)*** 0.25 (0.16, 0.39) *** 0.21 (0.10, 0.44) *** 74 76 0.09 (0.06, 0.14) *** 0.05 (0.03, 0.09) *** 0.05 (0.02, 0.11) *** 0.04 (0.02, 0.07) *** 0.01 (0.01, 0.03) *** 0.01 (0.00, 0.02) *** Race Other races (ref) 1.00 1.00 1.00 White 1.11 (0.72, 1.69) 1.06 (0.54, 2.13) 1.82 (1.02, 3.33)* Marital status Single (ref) 1.00 1.00 1.00 Married 1.79 (1.27, 2.50)** 3.03 (1.96, 4.55)*** 3.45 (2.22, 5.56)*** Education 50K 2.70 (1.92, 3.85)*** 3.85 (2.38, 5.88)*** 2.94 (1.75, 5.00)*** Insurance Other ins (ref) 1.00 1.00 1.00 Medicare 1.49 (1.14, 1.96)** 1.22 (0.83, 1.79) 1.52 (0.94, 2.44) Smoke Not current (ref) 1.00 1.00 1.00 Current smoker 1.18 (0.77, 1.79) 1.64 (0.82, 3.23) 0.84 (0.45, 1.56) Diagnosis year 1995 2000 (ref) 1.00 1.00 1.00 2001 2007 1.49 (1.19, 1.89)** 1.54 (1.18, 2.00) ** 1.59 (1.09, 2.27)* *P<0.05; **P<0.01; ***P<0.001;

PAGE 72

72 Table 4 4.Continued RT Vs. EM RT vs. HT EM vs. HT Age 65 67 (ref) 1.00 1.00 1.00 68 70 0.69 (0.44, 1.10) 0.71 (0.33, 1.49) 1.02 (0.45, 2.27) 71 73 0.74 (0.46, 1.18) 0.63 (0.30, 1.32) 0.85 (0.38, 1.92) 74 76 0.58 (0.37, 0.90)* 0.55 (0.27, 1.12) 0.95 (0.44, 2.04) 0.39 (0.26, 0.60) *** 0.20 (0.1, 0.38) *** 0.51 (0.25, 1.02) Race Other races (ref) 1.00 1.00 1.00 White 0.96 (0.50, 1.85) 1.64 (0.94, 2.94) 1.72 (0.81, 3.70) Marital status Single (ref) 1.00 1.00 1.00 Married 1.69 (1.16, 2.50)** 1.96 (1.3, 2.94)** 1.15 (0.72, 1.85) Education 50K 1.39 (0.90, 2.17) 1.09 (0.65, 1.85) 0.78 (0.45, 1.37) Insurance Other ins (ref) 1.00 1.00 1.00 Medicare 0.82 (0.57, 1.18) 1.01 (0.63, 1.61) 1.23 (0.71, 2.13) Smoke Not current (ref) 1.00 1.00 1.00 Current smoker 1.39 (0.68, 2.86) 0.71 (0.38, 1.33) 0.51 (0.22, 1.19) Diagnosis year 1995 2000 (ref) 1.00 1.00 1.00 2001 2007 1.02 (0.79, 1.32) 1.05 (0.73, 1.52) 1.03 (0.70, 1.52) *P<0.05; **P<0.01; ***P<0.001;

PAGE 73

73 Table 4 5. Crude estimates of associations between baseline clinical variables and primary treatment (odds ratios and 95% confidence intervals) RP vs. RT RP vs. EM RP vs. HT Cancer risk Low (ref) 1.00 1.00 1.00 Intermediate 0.93 (0.75, 1.16) 2.27 (1.64, 3.23)*** 0.67 (0.44, 1.02) high 0.74 (0.53, 1.04) 1.41 (0.85, 2.33) 0.24 (0.15, 0.39)*** BMI <25 (ref) 1.00 1.00 1.00 25 29.9 1.19 (0.94, 1.52) 1.75 (1.25, 2.50)** 1.59 (1.08, 2.33)* >=30 1.19 (0.88, 1.61) 1.92 (1.23, 3.03)** 1.32 (0.82, 2.13) Total n umber of comorbidities 0 (ref) 1.00 1.00 1.00 1 0.74 (0.50, 1.09) 0.75 (0.41, 1.35) 1.39 (0.68, 2.78) 2 0.56 (0.38, 0.83)** 0.61 (0.33, 1.10) 0.95 (0.49, 1.89) 3 or more 0.40 (0.28, 0.60)*** 0.35 (0.20, 0.63)*** 0.33 (0.18, 0.62)*** Heart d isease Yes vs. n o (ref) 0.43 (0.33, 0.55)*** 0.38 (0.28, 0.53)*** 0.34 (0.24, 0.49)*** Neurological disease/stroke Yes vs. no (ref) 0.46 (0.30, 0.71)** 0.37 (0.22, 0.63)*** 0.3 (0.17, 0.53)*** Other c ancer Yes vs. n o (ref) 0.72 (0.55, 0.96)* 0.94 (0.60, 1.47) 0.59 (0.38, 0.90)* Diabetes Yes vs. n o (ref) 0.69 (0.50, 0.98)* 0.48 (0.31, 0.76)** 0.37 (0.23, 0.58)*** Kidney disease Yes vs. n o (ref) 1.01 (0.64, 1.61) 0.52 (0.30, 0.88)* 2.22 (0.79, 6.25) Hypertension Yes vs. n o (ref) 0.69 (0.56, 0.85)** 0.83 (0.62, 1.11) 0.58 (0.41, 0.82)** Arthritis Yes vs. n o (ref) 0.84 (0.68, 1.03) 0.93 (0.70, 1.27) 0.83 (0.58, 1.16) Blood disease Yes vs. n o (ref) 0.52 (0.26, 1.06) 0.99 (0.30, 3.23) 0.47 (0.17, 1.28) GI disease Yes vs. n o (ref) 1.03 (0.77, 1.37) 0.76 (0.52, 1.11) 0.99 (0.61, 1.59) Lung disease Yes vs. n o (ref) 0.64 (0.44, 0.93)* 0.45 (0.28, 0.71)** 0.79 (0.43, 1.45) Urinary condition Yes vs. n o (ref) 1.01 (0.78, 1.33) 0.88 (0.61, 1.27) 0.54 (0.36, 0.79)** Other disease Yes vs. n o (ref) 1.01 (0.68, 1.49) 1.75 (0.93, 3.33) 0.83 (0.46, 1.52) *P<0.05; **P<0.01; ***P<0.001;

PAGE 74

74 Table 4 5. Continued RT Vs. EM RT vs. HT EM vs. HT Cancer risk Low (ref) 1.00 1.00 1.00 Intermediate 2.44 (1.75, 3.45)*** 0.71 (0.47, 1.09) 0.29 (0.18, 0.49)*** high 1.89 (1.18, 3.03)** 0.33 (0.21, 0.52)*** 0.17 (0.09, 0.31)*** BMI <25 (ref) 1.00 1.00 1.00 25 29.9 1.47 (1.06, 2.04)* 1.32 (0.91, 1.92) 0.89 (0.57, 1.41) >=30 1.64 (1.06, 2.50)* 1.11 (0.70, 1.75) 0.68 (0.39, 1.19) Number of comorbidities 0 (ref) 1.00 1.00 1.00 1 1.01 (0.55, 1.89) 1.89 (0.92, 3.85) 1.85 (0.79, 4.35) 2 1.08 (0.58, 1.96) 1.69 (0.85, 3.33) 1.59 (0.70, 3.57) 3 or more 0.87 (0.49, 1.56) 0.82 (0.44, 1.54) 0.94 (0.44, 2.00) Heart disease Yes vs. n o (ref) 0.90 (0.67, 1.22) 0.8 (0.57, 1.12) 0.88 (0.59, 1.33) Neurological disease/stroke Yes vs. no (ref) 0.81 (0.51, 1.28) 0.65 (0.40, 1.06) 0.81 (0.45, 1.45) Other cancer Yes vs. n o (ref) 1.30 (0.86, 1.96) 0.81 (0.54, 1.22) 0.63 (0.37, 1.06) Diabetes Yes vs. n o (ref) 0.69 (0.46, 1.04) 0.53 (0.35, 0.81)** 0.76 (0.45, 1.27) Kidney disease Yes vs. n o (ref) 0.51 (0.30, 0.86)* 2.22 (0.79, 6.25) 4.35 (1.47, 12.50)** Hypertension Yes vs. n o (ref) 1.20 (0.91, 1.61) 0.84 (0.60, 1.19) 0.70 (0.47, 1.03) Arthritis Yes vs. n o (ref) 1.12 (0.84, 1.49) 0.98 (0.70, 1.37) 0.88 (0.59, 1.30) Blood disease Yes vs. n o (ref) 1.89 (0.66, 5.56) 0.90 (0.37, 2.22) 0.47 (0.13, 1.69) GI disease Yes vs. n o (ref) 0.75 (0.51, 1.09) 0.96 (0.60, 1.52) 1.28 (0.74, 2.22) Lung disease Yes vs. n o (ref) 0.70 (0.46, 1.06) 1.23 (0.69, 2.17) 1.75 (0.93, 3.33) Urinary condition Yes vs. n o (ref) 0.86 (0.60, 1.23) 0.53 (0.36, 0.78)** 0.61 (0.39, 0.96)* Other disease Yes vs. n o (ref) 1.75 (0.93, 3.33) 0.83 (0.46, 1.49) 0.47 (0.22, 1.03) *P<0.05; **P<0.01; ***P<0.001;

PAGE 75

75 Table 4 6. Crude estimates of associations between baseline HRQOL variables and primary treatment (odds ratios and 95% confidence intervals) RP vs. RT RP vs. EM RP vs. HT KPS High vs. low (ref) 1.59 (1.05, 2.38)* 1.05 (0.76, 1.45) 2.00 (1.27, 3.13)** SF 36 PF High vs. low (ref) 2.78 (2.13, 3.70)*** 3.85 (2.78, 5.56)*** 5.26 (3.57, 7.69)*** RO High vs. low (ref) 2.00 (1.56, 2.50)*** 3.23 (2.38, 4.35)*** 3.13 (2.17, 4.55)*** BP High vs. low (ref) 1.52 (1.19, 1.92)** 2.04 (1.49, 2.78)*** 2.22 (1.54, 3.23)*** GH High vs. low (ref) 2.00 (1.56, 2.56)*** 2.78 (2.04, 3.85)*** 3.45 (2.38, 5.00)*** RE High vs. low (ref) 1.52 (1.16, 1.92)** 2.00 (1.45, 2.78)*** 1.89 (1.30, 2.78)** VT High vs. low (ref) 2.00 (1.56, 2.50)*** 2.63 (1.92, 3.57)*** 3.33 (2.33, 4.76)*** SO High vs. low (ref) 1.37 (1.10, 1.69)** 1.64 (1.22, 2.22)** 1.89 (1.35, 2.70)*** MH High vs. low (ref) 1.32 (1.06, 1.61)* 1.23 (0.93, 1.64) 1.39 (0.99, 1.96) UCLA PCI UF High vs. low (ref) 1.23 (0.99, 1.52) 1.27 (0.94, 1.69) 1.49 (1.06, 2.08)* UB High vs. low (ref) 1.14 (0.93, 1.41) 1.43 (1.06, 1.89)* 1.56 (1.11, 2.17)* BF High vs. low (ref) 1.30 (1.06, 1.61)* 1.39 (1.03, 1.85)* 1.89 (1.35, 2.63)*** BB High vs. low (ref) 1.49 (1.19, 1.89)** 1.85 (1.35, 2.50)*** 2.22 (1.56, 3.13)*** SF High vs. low (ref) 1.96 (1.59, 2.44)*** 1.96 (1.45, 2.63)*** 3.57 (2.50, 5.26)*** SB High vs. low (ref) 1.39 (1.12, 1.72)** 1.82 (1.35, 2.44)*** 1.61 (1.12, 2.27)** *P<0.05; **P<0.01; ***P<0.001;

PAGE 76

76 Table 4 6. Continued RT Vs. EM RT vs. HT EM vs. HT KPS High vs. low (ref) 0.67 (0.39, 1.15) 1.27 (0.74, 2.17) 1.89 (1.11, 3.23)* SF 36 PF High vs. low (ref) 1.39 (1.02, 1.85)* 1.89 (1.35, 2.63)*** 1.37 (0.93, 2.04) RO High vs. low (ref) 1.61 (1.22, 2.17)** 1.56 (1.12, 2.22)** 0.97 (0.65, 1.43) BP High vs. low (ref) 1.33 (0.99, 1.79) 1.47 (1.03, 2.08)* 1.09 (0.73, 1.64) GH High vs. low (ref) 1.41 (1.04, 1.89)* 1.72 (1.23, 2.44)** 1.22 (0.83, 1.82) RE High vs. low (ref) 1.33 (0.98, 1.82) 1.27 (0.88, 1.82) 0.95 (0.63, 1.45) VT High vs. low (ref) 1.33 (1.00, 1.79) 1.67 (1.19, 2.33)** 1.25 (0.85, 1.85) SO High vs. low (ref) 1.20 (0.90, 1.61) 1.39 (1.00, 1.96)* 1.15 (0.78, 1.72) MH High vs. low (ref) 0.94 (0.71, 1.25) 1.06 (0.76, 1.49) 1.12 (0.76, 1.67) UCLA PCI UF High vs. low (ref) 1.03 (0.78, 1.35) 1.20 (0.87, 1.69) 1.18 (0.80, 1.75) UB High vs. low (ref) 1.25 (0.94, 1.64) 1.37 (0.98, 1.92) 1.10 (0.74, 1.61) BF High vs. low (ref) 1.06 (0.80, 1.41) 1.45 (1.04, 2.04)* 1.37 (0.93, 2.04) BB High vs. low (ref) 1.23 (0.93, 1.64) 1.49 (1.06, 2.08)* 1.20 (0.82, 1.79) SF High vs. low (ref) 0.99 (0.75, 1.32) 1.82 (1.28, 2.63)** 1.85 (1.23, 2.78)** SB High vs. low (ref) 1.30 (0.98, 1.72) 1.15 (0.82, 1.61) 0.88 (0.60, 1.32) *P<0.05; **P<0.01; ***P<0.001;

PAGE 77

77 Table 4 7. Estimates of associations between baseline HRQOL variables and primary treatment, adjusting for baseline sociodemographic and clinical variables (odds ratios and 95% confidence intervals) RP vs. RT RP vs. EM RP vs. HT RT Vs. EM RT vs. HT EM vs. HT KPS High vs. low 1.60 (1.00, 2.57)* 1.03 (0.66, 1.59) 1.86 (1.16, 2.99)* 0.64 (0.35, 1.18) 1.16 (0.70, 1.92) 1.81 (1.00, 3.27)* PF High vs. low 1.96 (1.41, 2.71)*** 2.22 (1.44, 3.42)*** 2.78 (1.75, 4.41)*** 1.13 (0.80, 1.61) 1.42 (0.97, 2.08) 1.25 (0.80, 1.97) RO High vs. low 1.36 (1.02, 1.81)* 1.94 (1.30, 2.88)** 1.66 (1.07, 2.58)* 1.43 (1.02, 1.99)* 1.22 (0.83, 1.80) 0.86 (0.55, 1.33) GH High vs. low 1.55 (1.15, 2.09)** 2.04 (1.36, 3.04)** 2.35 (1.50, 3.69)*** 1.31 (0.94, 1.83) 1.52 (1.03, 2.23)* 1.16 (0.74, 1.80) VT High vs. low 1.74 (1.32, 2.28)*** 2.14 (1.45, 3.17)*** 2.62 (1.71, 4.02)*** 1.23 (0.89, 1.71) 1.51 (1.04, 2.19)* 1.22 (0.80, 1.87) SF High vs. low 1.35 (1.06, 1.72)* 1.04 (0.73, 1.48) 1.77 (1.17, 2.70)** 0.77 (0.56, 1.06) 1.32 (0.89, 1.95) 1.71 (1.09, 2.67)* *P<0.05; **P<0.01; ***P<0.001;

PAGE 78

78 Table 4 8. E stimates of adjusted associations between baseline variables and primary treatment (odds ratios and 95% confidence intervals) RP vs. RT RP vs. EM Age 65 67 (ref) 1.00 1.00 68 70 0.69 (0.50, 0.96)* 0.45 (0.29, 0.71)*** 71 73 0.29 (0.20, 0.42)*** 0.19 (0.12, 0.31)*** 74 76 0.08 (0.05, 0.12)*** 0.04 (0.02, 0.07)*** 0.03 (0.02, 0.06)*** 0.01 (0.00, 0.02)*** Race Other race(ref) 1.00 1.00 White 1.31 (0.80, 2.14) 1.50 (0.69, 3.28) Marital status Single (ref) 1.00 1.00 Married 1.68 (1.12, 2.53)* 3.24 (1.92, 5.48)*** Income <20K (ref) 1.00 1.00 20K 50K 1.26 (0.86, 1.84) 1.58 (0.94, 2.64) >50K 1.59 (1.04, 2.43)* 1.74 (0.98, 3.08) Insurance Other ins (ref) 1.00 1.00 Medicare 2.37 (1.73, 3.25)*** 2.06 (1.29, 3.29)** Diagnosis year 1995 2000 (ref) 1.00 1.00 2001 2007 1.49 (1.14, 1.94)** 2.06 (1.42, 2.98)*** Cancer risk Low (ref) 1.00 1.00 Intermediate 1.43 (1.1, 1.86)** 4.87 (3.2, 7.41)*** High 1.14 (0.75, 1.72) 3.48 (1.88, 6.45)*** Heart disease Yes vs. n o (ref) 0.49 (0.36, 0.65)*** 0.47 (0.32, 0.70)*** Neurological disease/stroke Yes vs. no (ref) 0.56 (0.33, 0.92)* 0.50 (0.27, 0.94)* Diabetes Yes vs. n o (ref) 0.75 (0.51, 1.12) 0.45 (0.26, 0.78)** Kidney disease Yes vs. n o (ref) 1.18 (0.69, 2.01) 0.62 (0.31, 1.25) Lung disease Yes vs. n o (ref) 0.74 (0.48, 1.13) 0.43 (0.25, 0.74)** Urinary condition Yes vs. n o (ref) 1.51 (1.09, 2.09)* 1.45 (0.90, 2.33) KPS High vs. low (ref) 1.55 (0.98, 2.46) 0.98 (0.64, 1.51) PF High vs. low (ref) 1.62 (1.14, 2.31)** 1.74 (1.09, 2.77)* VT High vs. low (ref) 1.49 (1.12, 2.00)** 1.83 (1.21, 2.77)**

PAGE 79

79 Table 4 8. Continued RP vs. HT RT vs. EM Age 65 67 (ref) 1.00 1.00 68 70 0.42 (0.20, 0.91)* 0.66 (0.41, 1.05) 71 73 0.16 (0.07, 0.34)*** 0.65 (0.40, 1.06) 74 76 0.03 (0.02, 0.08)*** 0.51 (0.32, 0.81)** 0.01 (0.00, 0.01)*** 0.31 (0.19, 0.48)*** Race Other race(ref) 1.00 1.00 White 2.66 (1.29, 5.49)** 1.14 (0.58, 2.27) Marital status Single (ref) 1.00 1.00 Married 3.5 (2.02, 6.08)*** 1.93 (1.28, 2.92)** Income <20K (ref) 1.00 1.00 20K 50K 1.28 (0.69, 2.39) 1.25 (0.80, 1.96) >50K 0.98 (0.51, 1.87) 1.09 (0.65, 1.85) Insurance Other ins (ref) 1.00 1.00 Medicare 2.91 (1.68, 5.03)*** 0.87 (0.57, 1.32) Diagnosis year 1995 2000 (ref) 1.00 1.00 2001 2007 1.55 (1.02, 2.37)* 1.39 (1.01, 1.91)* Cancer risk Low (ref) 1.00 1.00 Intermediate 1.37 (0.86, 2.19) 3.42 (2.35, 4.96)*** High 0.58 (0.32, 1.05) 3.05 (1.81, 5.16)*** Heart disease Yes vs. n o (ref) 0.46 (0.30, 0.71)*** 0.96 (0.69, 1.33) Neurological disease/stroke Yes vs. no (ref) 0.41 (0.21, 0.82)* 0.90 (0.55, 1.48) Diabetes Yes vs. n o (ref) 0.38 (0.22, 0.66)*** 0.59 (0.37, 0.95)* Kidney disease Yes vs. n o (ref) 2.93 (0.96, 8.91) 0.53 (0.30, 0.95)* Lung disease Yes vs. n o (ref) 1.04 (0.52, 2.10) 0.58 (0.37, 0.92)* Urinary condition Yes vs. n o (ref) 0.99 (0.61, 1.60) 0.96 (0.64, 1.43) KPS High vs. low (ref) 1.75 (1.10, 2.78)* 0.63 (0.35, 1.15) PF High vs. low (ref) 2.03 (1.24, 3.33)** 1.07 (0.74, 1.55) VT High vs. low (ref) 2.07 (1.30, 3.27)** 1.23 (0.87, 1.74)

PAGE 80

80 Table 4 8. Continued RT vs. HT EM vs. HT Age 65 67 (ref) 1.00 1.00 68 70 0.61 (0.28, 1.32) 0.93 (0.41, 2.12) 71 73 0.53 (0.24, 1.16) 0.81 (0.35, 1.87) 74 76 0.46 (0.22, 0.97)* 0.89 (0.40, 1.98) 0.17 (0.08, 0.35)*** 0.57 (0.27, 1.20) Race Other race(ref) 1.00 1.00 White 2.03 (1.08, 3.83)* 1.77 (0.80, 3.94) Marital status Single (ref) 1.00 1.00 Married 2.08 (1.32, 3.29)** 1.08 (0.65, 1.80) Income <20K (ref) 1.00 1.00 20K 50K 1.02 (0.57, 1.81) 0.81 (0.46, 1.45) >50K 0.62 (0.33, 1.15) 0.56 (0.30, 1.06) Insurance Other ins (ref) 1.00 1.00 Medicare 1.23 (0.74, 2.04) 1.41 (0.76, 2.62) Diagnosis year 1995 2000 (ref) 1.00 1.00 2001 2007 1.04 (0.72, 1.52) 0.75 (0.49, 1.16) Cancer risk Low (ref) 1.00 1.00 Intermediate 0.96 (0.63, 1.48) 0.28 (0.17, 0.48)*** High 0.51 (0.31, 0.85)** 0.17 (0.09, 0.32)*** Heart disease Yes vs. No. (ref) 0.94 (0.65, 1.37) 0.98 (0.64, 1.51) Neurological disease/stroke Yes vs. No. (ref) 0.74 (0.43, 1.28) 0.82 (0.44, 1.55) Diabetes Yes vs. No. (ref) 0.51 (0.32, 0.81)** 0.86 (0.48, 1.53) Kidney disease Yes vs. No. (ref) 2.49 (0.87, 7.11) 4.69 (1.56, 14.12)** Lung disease Yes vs. No. (ref) 1.42 (0.76, 2.64) 2.45 (1.24, 4.83)* Urinary condition Yes vs. No. (ref) 0.66 (0.43, 1.00)* 0.69 (0.42, 1.12) KPS High vs. low (ref) 1.13 (0.69, 1.85) 1.78 (1.00, 3.19) PF High vs. low (ref) 1.25 (0.83, 1.89) 1.17 (0.72, 1.90) VT High vs. low (ref) 1.38 (0.93, 2.06) 1.13 (0.72, 1.78) *P<0.05; **P<0.01; ***P<0.001;

PAGE 81

81 Table 4 9. Baseline sociodemographic variable distributions in CaPSURE elderly men with low or intermediate risk prostate cancer : DT vs. CT Overall, n (%) (N=1,700) DT, n (%) (N=1,338) CT, n (%) (N=362) Odds ratio Age 65 67 (ref) 410 (24.1) 377 (28.2) 33 (9.1) 1.00 68 70 432 (25.4) 383 (28.6) 49 (13.5) 0.68 71 73 287 (16.9) 233 (17.4) 54 (14.9) 0.38*** 74 76 275 (16.2) 198 (14.8) 77 (21.3) 0.23*** >=77 296 (17.4) 147 (11.0) 149 (41.2) 0.09*** Race Other (ref) 1,548 (91.1) 1,232 (92.1) 316 (87.3) 1.28 White 114 (6.7) 86 (6.4) 28 (7.7) 1.37 Missing 38 (2.2) 20 (1.5) 18 (5.0) -Marital status Single(ref) 1,425 (83.8) 1,160 (86.7) 265 (73.2) 1.00 Married 230 (13.5) 153 (11.4) 77 (21.3) 2.20*** Missing 45 (2.6) 25 (1.9) 20 (5.5) -Education 50K 520 (30.6) 437 (32.7) 83 (22.9) 2.08*** Missing 217 (12.8) 148 (11.1) 69 (19.1) -Insurance Other (ref) 1,320 (77.6) 1,048 (78.3) 272 (75.1) 1.00 Medicare 322 (18.9) 252 (18.8) 70 (19.3) 1.10 Missing 58 (3.4) 38 (2.8) 20 (5.5) -Smoking status Not current ( ref) 101 (5.9) 82 (6.1) 19 (5.2) 1.00 Current 1,548 (91.1) 1,225 (91.6) 323 (89.2) 1.11 Missing 51 (3.0) 31 (2.3) 20 (5.5) -Diagnosis year 1995 2000 (ref) 511 (30.1) 392 (29.3) 119 (32.9) 1.00 2001 2007 1,189 (69.9) 946 (70.7) 243 (67.1) 1.18 All odds ratio were calculated after the missing imputation.

PAGE 82

82 Table 4 10. Baseline clinical variable distributions in CaPSURE elderly men with low or intermediate risk prostate cancer : DT vs. CT Overall, n (%) (N=1,700) DT, n (%) (N=1,338) CT, n (%) (N=362) Odds ratio Tumor stage T1a T2a (ref) 1,344 (79.1) 1,044 (78.0) 300 (82.9) 1.00 T2b T2c 356 (20.9) 294 (22.0) 62 (17.1) 1.36 Tumor grade 2 4 (ref) 42 (2.5) 21 (1.6) 21 (5.8) 1.00 5 6 1,137 (66.9) 885 (66.1) 252 (69.6) 3.51 7 521 (30.6) 432 (32.3) 89 (24.6) 4.85*** PSA <10 ng/mL (ref) 1296 (76.2) 1,043 (78.0) 253 (69.9) 1.00 10 20 ng/mL 276 (16.2) 210 (15.7) 66 (18.2) 0.78 >20 ng/mL 67 (3.9) 44 (3.3) 23 (6.4) 0.45* Missing 61 (3.6) 41 (3.1) 20 (5.5) -BMI <25 (ref) 484 (28.5) 364 (27.2) 120 (33.1) 1.00 25 29.9 825 (48.5) 673 (50.3) 152 (42.0) 1.49* >=30 326 (19.2) 261 (19.5) 65 (18.0) 1.35 Missing 65 (3.8) 40 (3.0) 25 (6.9) -Comorbidity 0 (ref) 155 (9.1) 129 (9.6) 26 (7.2) 1.00 1 410 (24.1) 344 (25.7) 66 (18.2) 1.03 2 469 (27.6) 388 (29.0) 81 (22.4) 0.92 3 or more 624 (36.7) 455 (34.0) 169 (46.7) 0.55* Missing 42 (2.5) 22 (1.6) 20 (5.5) -Heart disease No (ref) 462 (27.2) 335 (25.0) 127 (35.1) 1.00 Yes 1,196 (70.4) 981 (73.3) 215 (59.4) 0.59*** Missing 42 (2.5) 22 (1.6) 20 (5.5) -Neurological disease/stroke No (ref) 145 (8.5) 99 (7.4) 46 (12.7) 1.00 Yes 1,513 (89.0) 1,217 (91.0) 296 (81.8) 0.54* Missing 42 (2.5) 22 (1.6) 20 (5.5) -Other cancer No (ref) 278 (16.4) 218 (16.3) 60 (16.6) 1.00 Yes 1,380 (81.2) 1,098 (82.1) 282 (77.9) 0.94 Missing 42 (2.5) 22 (1.6) 20 (5.5) -All odds ratio were calculated after the missing imputation.

PAGE 83

83 Table 4 10.Continued Overall, n (%) (N=1,700) DT, n (%) (N=1,338) CT, n (%) (N=362) Odds ratio Diabetes No (ref) 208 (12.2) 145 (10.8) 63 (17.4) 1.00 Yes 1,450 (85.3) 1,171 (87.5) 279 (77.1) 0.55*** Missing 42 (2.5) 22 (1.6) 20 (5.5) -Kidney disease No (ref) 93 (5.5) 68 (5.1) 25 (6.9) 1.00 Yes 1,565 (92.1) 1,248 (93.3) 317 (87.6) 0.67 Missing 42 (2.5) 22 (1.6) 20 (5.5) -Hyper tension No (ref) 837 (49.2) 653 (48.8) 184 (50.8) 1.00 Yes 821 (48.3) 663 (49.6) 158 (43.6) 0.86 Missing 42 (2.5) 22 (1.6) 20 (5.5) -Arthritis No (ref) 771 (45.4) 606 (45.3) 165 (45.6) 1.00 Yes 887 (52.2) 710 (53.1) 177 (48.9) 0.93 Missing 42 (2.5) 22 (1.6) 20 (5.5) -Blood disease No (ref) 44 (2.6) 35 (2.6) 9 (2.5) 1.00 Yes 1,614 (94.9) 1,281 (95.7) 333 (92.0) 0.97 Missing 42 (2.5) 22 (1.6) 20 (5.5) -GI disease No (ref) 261 (15.4) 204 (15.2) 57 (15.7) 1.00 Yes 1,397 (82.2) 1,112 (83.1) 285 (78.7) 0.95 Missing 42 (2.5) 22 (1.6) 20 (5.5) -Lung disease No (ref) 159 (9.4) 118 (8.8) 41 (11.3) 1.00 Yes 1,499 (88.2) 1,198 (89.5) 301 (83.1) 0.70 Missing 42 (2.5) 22 (1.6) 20 (5.5) -Urinary condition No (ref) 306 (18.0) 228 (17.0) 78 (21.5) 1.00 Yes 1,352 (79.5) 1,088 (81.3) 264 (72.9) 0.74* Missing 42 (2.5) 22 (1.6) 20 (5.5) -Other disease 0 (ref) 126 (7.4) 105 (7.8) 21 (5.8) 1.00 1 1,532 (90.1) 1,211 (90.5) 321 (88.7) 1.37 Missing 42 (2.5) 22 (1.6) 20 (5.5) -All odds ratio were calculated after the missing imputation.

PAGE 84

84 Table 4 11. Baseline HRQOL variable distributions in CaPSURE elderly men with low or intermediate risk prostate cancer : DT vs. CT Overall, n (%) (N=1,700) DT, n (%) (N=1,338) CT, n (%) (N=362) Odds ratio KPS Low score (ref) 470 (27.6) 354 (26.5) 116 (32.0) 1.00 High score 101 (5.9) 68 (5.1) 33 (9.1) 0.96 Missing 1,129 (66.4) 916 (68.5) 213 (58.8) -SF 36 PF Low score (ref) 1,276 (75.1) 1,049 (78.4) 227 (62.7) 1.00 High score 388 (22.8) 259 (19.4) 129 (35.6) 2.26*** Missing 36 (2.1) 30 (2.2) 6 (1.7) -R O Low score (ref) 1,191 (70.1) 985 (73.6) 206 (56.9) 1.00 High score 475 (27.9) 327 (24.4) 148 (40.9) 2.16*** Missing 34 (2.0) 26 (1.9) 8 (2.2) -BP Low score (ref) 1,215 (71.5) 988 (73.8) 227 (62.7) 1.00 High score 453 (26.6) 327 (24.4) 126 (34.8) 1.69*** Missing 32 (1.9) 23 (1.7) 9 (2.5) -GH Low score (ref) 1,211 (71.2) 994 (74.3) 217 (59.9) 1.00 High score 453 (26.6) 318 (23.8) 135 (37.3) 1.92*** Missing 36 (2.1) 26 (1.9) 10 (2.8) -RE Low score (ref) 1,293 (76.1) 1,041 (77.8) 252 (69.6) 1.00 High score 372 (21.9) 272 (20.3) 100 (27.6) 1.52* Missing 35 (2.1) 25 (1.9) 10 (2.8) -VT Low score (ref) 1,147 (67.5) 943 (70.5) 204 (56.4) 1.00 High score 530 (31.2) 378 (28.3) 152 (42.0) 1.84*** Missing 23 (1.4) 17 (1.3) 6 (1.7) -SO Low score (ref) 1,085 (63.8) 877 (65.5) 208 (57.5) 1.00 High score 576 (33.9) 430 (32.1) 146 (40.3) 1.43* Missing 39 (2.3) 31 (2.3) 8 (2.2) -MH Low score (ref) 961 (56.5) 761 (56.9) 200 (55.2) 1.00 High score 713 (41.9) 558 (41.7) 155 (42.8) 1.07 Missing 26 (1.5) 19 (1.4) 7 (1.9) -UCLA PCI UF Low score (ref) 996 (58.6) 794 (59.3) 202 (55.8) 1.00 High score 673 (39.6) 518 (38.7) 155 (42.8) 1.18 Missing 31 (1.8) 26 (1.9) 5 (1.4) -UB Low score (ref) 964 (56.7) 778 (58.1) 186 (51.4) 1.00 High score 701 (41.2) 534 (39.9) 167 (46.1) 1.32* Missing 35 (2.1) 26 (1.9) 9 (2.5) -BF Low score (ref) 968 (56.9) 779 (58.2) 189 (52.2) 1.00 High score 721 (42.4) 553 (41.3) 168 (46.4) 1.25 Missing 11 (0.6) 6 (0.4) 5 (1.4) -BB Low score (ref) 1,147 (67.5) 928 (69.4) 219 (60.5) 1.00 High score 533 (31.4) 393 (29.4) 140 (38.7) 1.49* Missing 20 (1.2) 17 (1.3) 3 (0.8) -SF Low score (ref) 788 (46.4) 655 (49.0) 133 (36.7) 1.00 High score 859 (50.5) 643 (48.1) 216 (59.7) 1.63*** Missing 53 (3.1) 40 (3.0) 13 (3.6) -SB Low score (ref) 984 (57.9) 805 (60.2) 179 (49.4) 1.00 High score 612 (36.0) 462 (34.5) 150 (41.4) 1.52* Missing 104 (6.1) 71 (5.3) 33 (9.1) -All odds ratio were calculated after the missing imputation.

PAGE 85

85 Table 4 12. Crude a ssociations between baseline sociodemographic v ariables and 10 year s urvival outcomes ( hazard r atio s and 95% c onfidence Interval s ) All cause death Prostate cancer death Other cause death Age 65 67 (ref) 1.00 1.00 1.00 68 70 1.43 (0.88, 2.31) 0.38 (0.1, 1.46) 1.8 (1.05, 3.07)* 71 73 1.65 (0.99, 2.76) 0.6 (0.16, 2.32) 2.02 (1.14, 3.57)* 74 76 1.86 (1.14, 3.06)* 0.93 (0.29, 2.92) 2.19 (1.25, 3.83)** 3.49 (2.25, 5.42)*** 1.01 (0.34, 3.03) 4.37 (2.66, 7.18)*** Race Other races (ref) 1.00 1.00 1.00 White 1.19 (0.63, 2.25) 0.63 (0.15, 2.71) 1.33 (0.65, 2.7) Marital status Single (ref) 1.00 1.00 1.00 Married 0.78 (0.53, 1.13) 0.48 (0.18, 1.24) 0.83 (0.56, 1.25) Education 50K 0.54 (0.35, 0.84)** 0.46 (0.14, 1.52) 0.56 (0.35, 0.88)* Insurance Other ins (ref) 1.00 1.00 1.00 Medicare 1.2 (0.82, 1.75) 4.5 (0.42, 48.42) 1.09 (0.75, 1.59) Smoking status Not current (ref) 1.00 1.00 1.00 Current 1.92 (1.24, 2.96)** 0.04 (0, 70.63) 2.19 (1.41, 3.39)*** Diagnosis year 1995 2000 (ref) 1.00 1.00 1.00 2001 2007 1.13 (0.85, 1.52) 0.58 (0.23, 1.44) 1.23 (0.9, 1.67) *P<0.05; **P<0.01; ***P<0.001;

PAGE 86

86 Table 4 13. Crude a ssociations between baseline c linical variables and 10 year s urvival outcomes ( h azard r atio s and 95% c onfidence i nterval s ) All cause death Prostate cancer death Other cause death Tumor stage T1 T2a (ref) 1.00 1.00 1.00 T2b T2c 1.33 (0.99, 1.79) 1.79 (0.78, 4.11) 1.28 (0.93, 1.75) Tumor grade 2 4 (ref) 1.00 1.00 1.00 5 6 1.08 (0.76, 1.55) 3.19 (1.29, 7.88)* 0.91 (0.62, 1.35) 7 1.62 (1.00, 2.62) 3.28 (0.91, 11.89) 1.48 (0.88, 2.50) PSA level <10 (ref) 1.00 1.00 1.00 10 20 1.46 (0.60, 3.58) 9 51 E6.00 (0, + ) 1.34 (0.55, 3.28) >20 1.86 (0.75, 4.59) 2 94 E6.00 (0, + ) 1.50 (0.60, 3.74) BMI <25 (ref) 1.00 1.00 1.00 25 29.9 0.91 (0.67, 1.24) 0.91 (0.38, 2.18) 0.91 (0.66, 1.26) >=30 0.86 (0.57, 1.28) 0.64 (0.17, 2.38) 0.88 (0.58, 1.36) Heart disease Yes vs. No (ref) 1.33 (1.00, 1.77)* 0.99 (0.41, 2.38) 1.38 (1.02, 1.86)* Neurological disease/stroke Yes vs. No (ref) 1.95 (1.35, 2.81)*** 2.04 (0.69, 5.98) 1.94 (1.32, 2.86)*** Other cancer Yes vs. No (ref) 1.06 (0.74, 1.52) 2.17 (0.9, 5.26) 0.95 (0.64, 1.41) Diabetes Yes vs. No (ref) 1.67 (1.17, 2.40)** 1.14 (0.34, 3.83) 1.75 (1.20, 2.54)** Kidney disease Yes vs. No (ref) 0.77 (0.39, 1.51) 1.5 (0.35, 6.39) 0.68 (0.32, 1.46) Hypertension Yes vs. No (ref) 1.21 (0.93, 1.59) 0.57 (0.24, 1.33) 1.33 (1.00, 1.77) Arthritis Yes vs. No (ref) 0.94 (0.72, 1.24) 1.62 (0.72, 3.65) 0.88 (0.66, 1.17) Blood disease Yes vs. No (ref) 1.99 (1.04, 3.84)* 0.04 (0, 4071) 2.24 (1.16, 4.32)* GI disease Yes vs. No (ref) 0.98 (0.69, 1.4) 0.38 (0.09, 1.64) 1.07 (0.74, 1.55) Lung disease Yes vs. No (ref) 1.27 (0.85, 1.91) 0.8 (0.19, 3.39) 1.34 (0.87, 2.05) Urinary condition Yes vs. No (ref) 1.09 (0.77, 1.54) 0.68 (0.2, 2.26) 1.15 (0.80, 1.65) Other disease Yes vs. No (ref) 0.98 (0.58, 1.65) 1.2 (0.28, 5.08) 0.95 (0.55, 1.67) Number of comorbidity 0 (ref) 1.00 1.00 1.00 1 1.09 (0.61, 1.95) 0.50 (0.11, 2.24) 1.23 (0.64, 2.34) 2 1.29 (0.73, 2.28) 1.00 (0.27, 3.69) 1.36 (0.72, 2.57) 3 or more 1.51 (0.86, 2.63) 0.69 (0.18, 2.60) 1.70 (0.91, 3.15) *P<0.05; **P<0.01; ***P<0.001

PAGE 87

87 Table 4 14. Crude a ssociations between baseline H RQOL variables and 10 year survival outcomes ( hazard r atio s and 95% c onfidence i nterval s ) All cause death Prostate cancer death Other cause death KPS High vs. low (ref) 0.61 (0.39, 0.95)* 1.21 (0.29, 5.03) 0.55 (0.37, 0.83)** SF 36 PF High vs. low (ref) 0.42 (0.32, 0.55)*** 0.93 (0.37, 2.34) 0.38 (0.29, 0.51)*** RO High vs. low (ref) 0.54 (0.41, 0.71)*** 0.81 (0.35, 1.86) 0.51 (0.38, 0.68)*** BP High vs. low (ref) 0.65 (0.49, 0.86)** 0.94 (0.38, 2.34) 0.62 (0.46, 0.83)** GH High vs. low (ref) 0.57 (0.43, 0.75)*** 0.99 (0.41, 2.39) 0.54 (0.40, 0.71)*** RE High vs. low (ref) 0.55 (0.42, 0.73)*** 0.25 (0.11, 0.57)*** 0.61 (0.46, 0.83)** V T High vs. low (ref) 0.54 (0.41, 0.70)*** 0.84 (0.36, 1.96) 0.51 (0.38, 0.68)*** S O High vs. low (ref) 0.72 (0.55, 0.94)* 0.77 (0.33, 1.77) 0.71 (0.53, 0.94)* MH High vs. low (ref) 0.66 (0.51, 0.87)** 0.86 (0.38, 1.98) 0.64 (0.48, 0.85)** UCLA PCI UF High vs. low (ref) 0.80 (0.61, 1.05) 1.16 (0.51, 2.64) 0.77 (0.58, 1.02) UB High vs. low (ref) 0.76 (0.58, 1.00)* 0.73 (0.32, 1.63) 0.77 (0.58, 1.02) BF High vs. low (ref) 0.85 (0.65, 1.11) 1.25 (0.55, 2.87) 0.81 (0.61, 1.08) BB High vs. low (ref) 0.65 (0.50, 0.85)** 0.50 (0.23, 1.12) 0.67 (0.50, 0.89)** SF High vs. low (ref) 0.61 (0.46, 0.82)*** 1.14 (0.51, 2.55) 0.56 (0.42, 0.77)*** SB High vs. low (ref) 0.65 (0.49, 0.86)** 0.55 (0.25, 1.22) 0.66 (0.49, 0.89)** *P<0.05; **P<0.01; ***P<0.001;

PAGE 88

88 Table 4 15. Associations between baseline HRQOL variables and primary treatment, conditioning on propensity score s : DT vs. CT Covariate Reference propensity score (not including HRQOL) HRQOL propensity score (including HRQOL) Odds Ratio P value Odds Ratio P value KPS High vs. low (ref) 0.86 0.37 0.98 0.89 10 points increase 1.03 0.83 1.09 0.51 SF 36 PF High vs. low (ref) 1.44 0.01* 1.05 0.77 10 points increase 1.08 0.01* 1.01 0.71 RO High vs. low (ref) 1.53 <0.01* 1.03 0.86 10 points increase 1.05 0.01* 1.00 1.00 BP High vs. low (ref) 1.29 0.08 1.03 0.85 10 points increase 1.06 0.09 1.00 0.98 GH High vs. low (ref) 1.49 0.01* 1.05 0.73 10 points increase 1.14 <0.01* 1.06 0.09 RE High vs. low (ref) 1.28 0.11 1.03 0.86 10 points increase 1.03 0.19 1.00 0.94 VT High vs. low (ref) 1.49 0.01* 1.03 0.86 10 points increase 1.08 0.02* 1.00 0.94 SO High vs. low (ref) 1.20 0.18 1.00 1.00 10 points increase 1.03 0.34 0.98 0.65 MH High vs. low (ref) 0.94 0.69 0.99 0.95 10 points increase 0.98 0.73 0.97 0.58 UCLA PCI UF High vs. low (ref) 0.99 0.97 1.02 0.89 10 points increase 1.07 0.16 1.06 0.19 UB High vs. low (ref) 1.11 0.44 1.03 0.86 10 points increase 1.03 0.24 1.01 0.62 BF High vs. low (ref) 1.09 0.51 1.02 0.90 10 points increase 1.04 0.45 1.01 0.88 BB High vs. low (ref) 1.11 0.45 1.01 0.95 10 points increase 1.03 0.30 1.01 0.84 SF High vs. low (ref) 1.10 0.51 1.04 0.78 10 points increase 1.03 0.28 1.02 0.57 SB High vs. low (ref) 1.26 0.10 1.02 0.86 10 points increase 1.02 0.33 0.99 0.71 *P<0.05

PAGE 89

89 Table 4 16. Associations between baseline HRQOL variables and primary treatment, conditioning on propensity scores : RP vs. CT Covariate Reference propensity score (not including HRQOL) HRQOL propensity score (including HRQOL) Odds Ratio P value Odds Ratio P value KPS High vs. low (ref) 1.21 0.48 1.02 0.92 10 points increase 1.68 0.04* 1.35 0.21 SF 36 PF High vs. low (ref) 2.49 <0.01* 1.07 0.80 10 points increase 1.21 <0.01* 1.04 0.46 RO High vs. low (ref) 2.05 <0.01* 1.07 0.76 10 points increase 1.08 <0.01* 1.00 0.97 BP High vs. low (ref) 1.65 0.02* 1.04 0.87 10 points increase 1.09 0.07 0.98 0.71 GH High vs. low (ref) 2.22 <0.01* 1.08 0.75 10 points increase 1.24 <0.01* 1.06 0.30 RE High vs. low (ref) 1.48 0.06 1.01 0.95 10 points increase 1.06 0.06 0.99 0.86 VT High vs. low (ref) 2.29 < 0.01* 1.08 0.73 10 points increase 1.17 <0.01* 0.99 0.87 SO High vs. low (ref) 1.30 0.21 1.03 0.90 10 points increase 1.05 0.37 0.96 0.51 MH High vs. low (ref) 1.12 0.58 0.98 0.93 10 points increase 1.06 0.38 0.99 0.85 UCLA PCI UF High vs. low (ref) 1.15 0.49 1.04 0.86 10 points increase 1.11 0.10 1.07 0.35 UB High vs. low (ref) 1.26 0.25 1.04 0.85 10 points increase 1.06 0.15 1.01 0.79 BF High vs. low (ref) 1.27 0.22 1.06 0.79 10 points increase 1.07 0.28 0.99 0.87 BB High vs. low (ref) 1.30 0.21 1.03 0.91 10 points increase 1.06 0.19 1.00 0.97 SF High vs. low (ref) 1.39 0.07 1.06 0.79 10 points increase 1.08 0.03* 1.02 0.64 SB High vs. low (ref) 1.46 0.07 1.05 0.83 10 points increase 1.05 0.06 1.01 0.80 *P<0.05

PAGE 90

90 Table 4 17. Associations between baseline HRQOL variables and primary treatment, conditioning on propensity scores : RT vs. CT Covariate Reference propensity score (not including HRQOL) HRQOL propensity score (including HRQOL) Odds Ratio P value Odds Ratio P value KPS High vs. low (ref) 0.75 0.11 0.98 0.88 10 points increase 0.89 0.44 1.04 0.79 SF 36 PF High vs. low (ref) 1.20 0.23 1.00 0.99 10 points increase 1.04 0.14 1.01 0.87 RO High vs. low (ref) 1.39 0.03* 1.00 1.00 10 points increase 1.04 0.03* 1.00 0.97 BP High vs. low (ref) 1.23 0.17 1.00 0.99 10 points increase 1.05 0.15 1.00 0.91 GH High vs. low (ref) 1.31 0.07 1.00 0.98 10 points increase 1.09 0.01* 1.04 0.30 RE High vs. low (ref) 1.22 0.21 1.01 0.96 10 points increase 1.02 0.38 0.99 0.79 VT High vs. low (ref) 1.33 0.05 1.00 0.97 10 points increase 1.06 0.14 0.99 0.82 SO High vs. low (ref) 1.15 0.34 0.99 0.94 10 points increase 1.02 0.49 0.99 0.76 MH High vs. low (ref) 0.90 0.48 0.99 0.94 10 points increase 0.96 0.43 0.97 0.49 UCLA PCI UF High vs. low (ref) 0.98 0.86 1.00 1.00 10 points increase 1.07 0.13 1.07 0.14 UB High vs. low (ref) 1.11 0.45 1.01 0.96 10 points increase 1.03 0.27 1.01 0.67 BF High vs. low (ref) 1.03 0.82 1.00 1.00 10 points increase 1.02 0.63 1.01 0.81 BB High vs. low (ref) 1.08 0.59 1.00 1.00 10 points increase 1.02 0.45 1.01 0.84 SF High vs. low (ref) 1.01 0.94 1.00 0.99 10 points increase 1.01 0.75 1.00 0.91 SB High vs. low (ref) 1.20 0.21 1.01 0.97 10 points increase 1.01 0.71 0.99 0.46 *P<0.05

PAGE 91

91 Table 4 18. Propensity score (PS) quintile specific hazard ratio s for all cause mortality before additional adjustment of the baseline HRQOL variables Quintile PS range Definitive treatment Conservative treatment Hazard ratio RP (n) RT (n) EM (n) HT (n) DT vs. CT 1 0.08 0.64 13 152 117 58 0.57 2 0.64 0.81 59 183 57 41 0.48 3 0.81 0.89 112 176 32 20 0.22 4 0.89 0.94 200 122 12 6 0.38 5 0.94 0.99 227 94 10 9 0.33 RP vs. CT 1 0.003 0.17 12 -119 64 1.17 2 0.17 0.66 82 -70 43 0.41 3 0.67 0.86 156 -21 17 0.18 4 0.86 0.94 179 -10 6 0.22 5 0.94 0.99 182 -8 4 0.49 RT vs. CT 1 0.09 0.51 -81 96 41 0.68 2 0.51 0.65 -140 47 31 0.64 3 0.65 0.75 -150 41 27 0.20 4 0.75 0.82 -170 31 17 0.45 5 0.82 0.97 -186 13 18 0.40 Table 4 19. Propensity score (PS) quintile specific hazard ratio s for all cause mortality after additional adjustment of the baseline HRQOL variables Quintile PS range Definitive treatment Conservative treatment Hazard ratio RP (n) RT (n) EM (n) HT (n) DT vs. CT 1 0.05 0.64 17 142 121 60 0.61 2 0.64 0.81 52 196 55 37 0.41 3 0.81 0.89 127 160 32 21 0.52 4 0.90 0.94 185 131 15 9 0.55 5 0.94 0.99 230 98 5 7 0.16 RP vs. CT 1 0.001 0.16 14 -118 63 1.26 2 0.16 0.64 76 -70 49 0.23 3 0.65 0.87 153 -26 15 0.43 4 5 0.87 0.99 368 -14 7 0.31 RT vs. CT 1 0.07 0.52 -81 95 42 0.59 2 0.52 0.65 -131 54 33 0.51 3 0.65 0.75 -154 40 24 0.29 4 0.75 0.83 -170 27 21 0.59 5 0.83 0.96 -191 12 14 0.31

PAGE 92

92 Table 4 20. Adjusted associations between primary treatment and 10year mortality outcome s : DT vs. CT Outcome Method Additional HRQOL adjustment Hazard ratio 95% CI Relative change Absolute change All cause mortality Crude N/A 0.35 0.16 0.77 --PS, continuous No 0.44 0.19 1.05 3.0% 0.01 Yes 0.46 0.19 1.08 PS, quintiles No 0.43 0.21 0.91 9.3% 0.04 Yes 0.48 0.23 1.00 Weighting, entire sample No 0.43 0.19 0.96 5.7% 0.03 Yes 0.46 0.20 1.04 Weighting, DT No 0.42 0.18 0.97 6.5% 0.03 Yes 0.44 0.19 1.05 Weighting, CT No 0.50 0.24 1.05 1.9% 0.01 Yes 0.51 0.24 1.09 Prostate cancer specific mortality Crude N/A 0.30 0.08 1.07 --PS, continuous No 0.26 0.06 1.14 1.4% <0.01 Yes 0.27 0.06 1.21 PS, quintiles No 0.28 0.07 1.18 0.5% <0.01 Yes 0.29 0.07 1.23 Weighting, entire sample No 0.31 0.06 1.51 3.9% 0.01 Yes 0.32 0.06 1.60 Weighting, DT No 0.35 0.06 2.09 4.4% 0.02 Yes 0.36 0.06 2.29 Weighting, CT No 0.19 0.05 0.81 3.0% 0.01 Yes 0.20 0.04 0.90 Other cause mortality Crude N/A 0.36 0.16 0.79 --PS, continuous No 0.46 0.19 1.10 3.0% 0.01 Yes 0.48 0.20 1.13 PS, quintiles No 0.45 0.21 0.94 10.2% 0.05 Yes 0.50 0.24 1.05 Weighting, entire sample No 0.44 0.20 1.00 5.8% 0.03 Yes 0.47 0.20 1.09 Weighting, DT No 0.42 0.18 1.01 6.6% 0.03 Yes 0.45 0.19 1.09 Weighting, CT No 0.52 0.25 1.11 1.8% 0.01 Yes 0.53 0.25 1.16 All estimates were obtained as IPCW weights applied; All 95% CIs were based on 1000 bootstrap samples;

PAGE 93

93 Table 4 21. Adjusted association s between primary treatment and 10year mortality outcome s: RP vs. CT Outcome Method Additional HRQOL adjustment Hazard ratio 95% CI Relative change Absolut e change All cause mortality Crude N/A 0.27 0.12 0.64 --PS, continuous No 0.37 0.11 1.21 19.9% 0.09 Yes 0.46 0.14 1.52 PS, quintiles No 0.34 0.12 0.95 18.6% 0.08 Yes 0.42 0.15 1.15 Weighting, entire sample No 0.57 0.18 1.78 29.4% 0.24 Yes 0.81 0.23 2.81 Weighting, DT No 0.36 0.11 1.16 20.3% 0.09 Yes 0.45 0.13 1.51 Weighting, CT No 0.83 0.18 3.85 19.7% 0.20 Yes 1.03 0.21 5.19 Prostate cancer specific mortality Crude N/A 0.19 0.01 3.93 --PS, continuous No 0.17 0.004 7.75 22.5% 0.03 Yes 0.14 0.003 6.05 PS, quintiles No 0.21 0.005 8.73 28.7% 0.05 Yes 0.16 0.004 6.09 Weighting, entire sample No 0.21 0.008 5.71 41.1% 0.06 Yes 0.15 0.005 4.11 Weighting, DT No 0.29 0.005 18.23 21.3% 0.05 Yes 0.24 0.004 15.79 Weighting, CT No 0.12 0.004 3.89 88.1% 0.06 Yes 0.06 0.002 2.47 Other cause mortality Crude N/A 0.28 0.11 0.69 --PS, continuous No 0.40 0.12 1.32 24.7% 0.13 Yes 0.52 0.16 1.69 PS, quintiles No 0.36 0.13 1.01 25.2% 0.12 Yes 0.48 0.18 1.28 Weighting, entire sample No 0.61 0.19 1.96 32.2% 0.29 Yes 0.90 0.25 3.23 Weighting, DT No 0.37 0.11 1.22 23.9% 0.12 Yes 0.48 0.14 1.72 Weighting, CT No 0.90 0.19 4.36 20.6% 0.23 Yes 1.14 0.22 5.87 All estimates were obtained as IPCW weights applied; All 95% CIs were based on 1000 bootstrap samples;

PAGE 94

94 Table 4 22. Adjusted association s between primary treatment and 10year mortality outcome s: RT vs. CT Outcome Method Additional HRQOL adjustment Hazard ratio 95% CI Relative change Absolute change All cause mortality Crude N/A 0.43 0.20 0.95 --PS, continuous No 0.47 0.20 1.08 1.5% 0.01 Yes 0.47 0.20 1.12 PS, quintiles No 0.46 0.22 0.98 0.6% <0.01 Yes 0.46 0.22 0.98 Weighting, entire sample No 0.47 0.21 1.04 0.6% <0.01 Yes 0.47 0.21 1.06 Weighting, DT No 0.46 0.19 1.12 1.6% 0.01 Yes 0.47 0.19 1.17 Weighting, CT No 0.50 0.23 1.10 1.8% 0.01 Yes 0.49 0.22 1.09 Prostate cancer specific mortality Crude N/A 0.38 0.10 1.42 --PS, continuous No 0.32 0.08 1.28 4.2% 0.01 Yes 0.33 0.08 1.42 PS, quintiles No 0.33 0.08 1.30 2.6% 0.01 Yes 0.34 0.08 1.42 Weighting, entire sample No 0.35 0.08 1.46 5.9% 0.02 Yes 0.37 0.09 1.62 Weighting, DT No 0.41 0.09 1.89 7.1% 0.03 Yes 0.45 0.09 2.14 Weighting, CT No 0.24 0.05 1.10 4.7% 0.01 Yes 0.25 0.05 1.36 Other cause mortality Crude N/A 0.43 0.19 1.00 --PS, continuous No 0.48 0.20 1.16 1.1% 0.01 Yes 0.49 0.20 1.19 PS, quintiles No 0.47 0.21 1.03 0.3% <0.01 Yes 0.47 0.21 1.03 Weighting, entire sample No 0.48 0.21 1.10 0.1% <0.01 Yes 0.48 0.21 1.12 Weighting, DT No 0.46 0.18 1.18 1.1% 0.01 Yes 0.47 0.18 1.22 Weighting, CT No 0.52 0.23 1.18 2.2% 0.01 Yes 0.51 0.23 1.17 All estimates were obtained as IPCW weights applied; All 95% CIs were based on 1000 bootstrap samples;

PAGE 95

95 Figure 4 1. Propensity score distributions for DT vs. CT in CaPSURE elderly men ( old) with low or intermediate risk prostate cancer ; Abbreviations: w/o QOL, without additional adjustment of baseline HRQOL variables; w/QOL, with additional adjustment of baseline HRQOL variables. Figure 42. Propensity score distributions for RP vs. CT in CaPSURE elderly men ( old) with low or intermediate risk prostate cancer ; Abbreviations: w/o QOL, without additional adjustment of baseline HRQOL variables; w/QOL, with additional adjustment o f baseline HRQOL variables.

PAGE 96

96 Figure 4 3. Propensity score distributions for RT vs. CT in CaPSURE elderly men ( old) with low or intermediate risk prostate cancer ; Abbreviations: w/o QOL, without additional adjustment of baseline HRQOL variables; w/QOL, with additional adjustment of baseline HRQOL variables.

PAGE 97

97 A B C Figure 4 4. Log log survival curves: DT vs. CT. A) all cause mortality; B) prostate cancer specific mortality; C) other cause mortality.

PAGE 98

98 A B C Figure 4 5. Log log survival curves: RP vs. CT. A) all cause mortality; B) prostate cancer specific mortality; C) other cause mortality.

PAGE 99

99 A B C Figure 4 6. Log l og survival curves : RT vs. CT A) all cause mortality; B) prostate cancer specific mortality; C) other cause mortality.

PAGE 100

100 Figure 4 7. Adjusted hazard ratios by different confounding adjustment methods: DT vs. CT; Abbreviations: None, no adjustment; PS, propensity score; WT, weighting; Ref, reference model; HRQOL, HRQOL model; Figure 4 8. Adjusted hazard ratios by different confounding adjustment methods RP vs. CT; Abbreviations: None, no adjustment ; PS, propensity score; WT, weighting; Ref, reference model; HRQOL, HRQOL model; 0.00 1.00 2.00 3.00 4.00 5.00 Ref HRQOL Ref HRQOL Ref HRQOL Ref HRQOL Ref HRQOL None PS, continuous PS, quintile WT, overall WT, DT WT, CT 0.00 1.00 2.00 3.00 4.00 5.00 Ref HRQOL Ref HRQOL Ref HRQOL Ref HRQOL Ref HRQOL None PS, continuous PS, quintiles WT, overall WT, RP WT, CT

PAGE 101

101 Figure 4 9. Adjusted hazard ratios by different confounding adjustment methods: RT vs. CT; Abbreviations: None, no adjustment ; PS, propensity score; WT, weighting; Ref, reference model; HRQOL, HRQOL model; 0.00 1.00 2.00 3.00 4.00 5.00 Ref HRQOL Ref HRQOL Ref HRQOL Ref HRQOL Ref HRQOL None PS, continuous PS, quintiles WT, overall WT, RT WT, CT

PAGE 102

102 A B C D Figure 4 10. Crude and weighted survival curves: DT vs. CT A) crude; B) weighting to the entire sample; C) weighting to the definitive treatment group; D) weighting to the conservative treatment group.

PAGE 103

103 A B C D Figure 4 11. Crude and weighted survival curves: RP vs. CT. A) crude; B) weighting to the entire sample; C) weighting to the radical prostatectomy group; D) weighting to the conservative treatment group.

PAGE 104

104 A B C D Figure 4 12. Crude and weighted survival curves: RT vs. CT. A) c rude; B) weighting to the entire sample; C) weighting to the radiation therapy group; D) weighting to the conservative treatment group.

PAGE 105

105 CHAPTER 5 DISCUSSION Association between HRQOL and P rimary T reatment To our best knowledge, this is the first study that explored the associations between a comprehensive range of HRQOL variables and primary treatment for localized prostate cancer. We found that men treated with radical prostatectomy had significantly bette r HRQOL than the others, and men treated with radiation therapy had moderately better HRQOL than men treated conservatively. There was no significant difference between men treated with expectant management and hormonal therapy. After adjusting sociodemogr aphic and clinical variables, six baseline HRQOL variables remained associated with primary treatment. KPS, physical functioning and vitality are the three most predictive HRQOL variables that retained in the final model. HRQOL measures important health di mensions and strongly predict s survival The strong association s between definitive treatment and high HRQOL scores w ere consistent with the current practice guidelines.2 KPS reflects physician as sessed functional performance. It is not surprising that high KPS was associated with radical prostatectomy since physician opinions dominate the treatment decision process.63 Four SF 36 variables were associated with radical prostatectomy independent to sociodemographic and clinical variables. They all directly measure or highly correlate with physical health status. High physical functioning score indicates little limitation in behavioral performance of everyday physical activities. High role physical score indicates no problem with work or other daily e veryday activities as a result of physical health. High general health score reflects good self evaluated overall health. High vitality score reflects no feeling of tiredness or wornout .22 Their strong associations with radical prostatectomy can be attributed to the fact that physical health status is the most prognostic factor for life expectancy

PAGE 106

106 and post surgical complication tolerance. Besides, elderly cancer patients are often screened by risk assessment tools. The most often used tools include Instrumental Activities of Daily Living, Brief Fatigue Inventory, ECOG Performance Status, and Americ an Society for Anesthesiologist Scale.91 92 All these instru ments include items that directly assess or highly correlate with patient physical functional status. Mental health, role emotional, and social functioning are the three SF 36 variables for mental health status. Their associations with primary treatment we re insignificant after adjusting for sociodemographic and clinical factors. It appears that mental health status has minor impact on treatment decisions, or such impact can be explained by sociodemographic and clinical factors. Out of the six PCI variable s, only sexual function was associated with primary treatment adjusting for baseline sociodemographic and clinical factors. Men with good sexual function were more likely to receive radical prostatectomy than radiation therapy and hormonal therapy. There w as no difference when choosing between radical prostatectomy and expectant management. This finding was consistent with previous studies. Zelidadt10 found that prostate cancer patients with good sexual function preferred surgical treatment than nonsurgical treatment Yan60 also found that prostate cancer patients without sexual dysfunction were more likely to receive radical prostatectomy than radiation therapy It has been speculated that men who were not experiencing sexual dysfunction at diagnosis may discount the treatment related side effect in the decision making process .10 Despite the strong associations we found, no firm conclusion can be made regarding the exact role of HRQOL in treatment decision making process HRQOL often highly correlates with a wide range of socioeconomic and health factors. Many HRQOL variables are also highly

PAGE 107

107 correlated with each other. The treatment decision s may be made based on one or several other unmeasured factors that highly correlate with HRQOL variables. Another limitation of the current study is unadjusted specialty bias. A recent study found that treatment selection was strongly associated with specialist consulted. Men solely evaluated by urologist were more lik ely to receive radical prostatectomy or conservative treatment than men consulted with radiation oncologist. Men consulting both radiation oncologist and urologist almost exclusively received radiation therapy .93 Nevertheless, the observed specialt y bias has been thought to be a consequence of urologist referral behavior. Urologists seldom refer patients to radiation oncologist unless they believe radiation therapy is more appropriate. If this is true and baseline HRQOL play a role in urologist eva luation, the specialty bias would have little impact on our findings. The C onfounding Effect of Baseline HRQOL It has been hypothesized that the observed association between definitive treatment and longer survival was confounded by uncontrolled baseline HRQOL. We found nearly all the fifteen baseline HRQOL variables included in our study were strongly associated w ith both primary treatment and survival. Thus, they were qualified confounders according to the classic definition. We further estimated the confounding magnitude of these baseline HRQOL variables collectively in addition to comprehensive baseline sociodemographic and clinical factors. Significant confounding effect was observed when comparing radical prostatectomy and conservative treatment. The association between radical prostatectomy and longer all cause survival was significantly attenuated after add itional HRQOL adjustment. The hazard ratios increased approximately 18% 29% depending on the confounding adjustment method used. It was also found that the attenuation was largely driven by attenuated association with other cause

PAGE 108

108 mortality. The plausible association with prostat e cancer specific mortality changed minimally The observed confounding effect was minor when comparing radiation therapy vs. conservative treatment. The hazard ratios for all cause, prostate cancer specific and other cause mortali ty remained nearly unchanged after additional HRQOL adjustment. Thus, the observed moderate confounding effect in the overall definitive treatment group lay between the significant confounding effect for radical prostatectomy and the nearly zeros confoundi ng effect for radiation therapy. These findings were consistent with the first part of our study that radical prostatectomy was more strongly associated with high HRQOL scores than radiation therapy Future observational studies evaluating radical prostate ctomy with regard to allcause survival should take HRQOL into consideration. However, several red flags for substantial residual confounding were noticed even after additional adjustment of the baseline HRQOL variables. Given no unmeasured confounding, w eighting to the radical prostatectomy group can be considered as a pseudorandomized clinic al trial that recruits men with health conditions permit ting radical prostatectomy. We obtained a hazard ratio of 0.45 (95% CI: 0.092.14) by using this method, the point estimate of which conflicted with the S candinavian trial subgroup analysis that radical prostatectomy did not benefit men older than 65 years. Moreover, men treated with radical prostatectomy or radiation therapy had nearly 50% lower risk for other cause death which w ere not plausible causal consequences of prostate cancer treatment. Considering the fact that other cause death accounted for approximately 90% all cause death, the observed survival benefit in the definitive treatment group was largely due to the lower other cause mortality. These findings indicated the persistence of substantial residual confounding. Miettinen91 pointed out that confounding may not be controllable in observational settings when treatment selection was highly influenced by

PAGE 109

109 the expected outcomes. In the case of localized prostate cancer treatment, longer life expectancy greatly increased the intention to treat definitively while shorter life expectancy greatly increased the intention to treat conservatively L ife expectan cy estimation is often based on a broad range of health indicators. This can be reflected by the fact that men treated definitively, particularly those treated with radical prostatectomy, had favorable characteristics for all the examined sociodemographic, clinical, and HRQOL variables. Conceivably, factors c ontributing to the life expectancy estimation are often too many too complex and too subtle to be fully captured and accurately quantified in observational studies H RQOL assessment seems to offer additional, but still insufficient, information capture all of them. Randomized trials are needed to evaluate the true efficacy of the primary treatment effect. Effect Modification Besides the confounding role of baseline HRQ OL, the current study compared several confounding adjustment methods. The adjusted hazard ratios varied when comparing radical prostatectomy and conservative treatment. When weighting to the radical prostatectomy group, the adjusted hazard ratio was 0.45 (95% CI: 0.131.51) It indicated that the observed survival in the radical prostatectomy group was longer than its counterfactual counterpart: what it would be had if these men received conservative treatment. When weighting to the conservative treatment group, the adjusted hazard ratio was 1.03 (95% CI: 0.215.19) It indicated that the observed survival in the conservative treatment group was about equal to its counterfactual counterpart: what it would be had if these men received radical prostatectomy. When weighting to the entire sample for this comparison, the hazard ratio was 0.81 (95% CI: 0.23 2.81) It resulted from the contrast between two counterfactual outcomes: what it would be if everyone received radical prostatectomy vs. what it would be if e veryone received conservative treatment.

PAGE 110

110 The diminished association when weighting to the conservative treatment is unlikely due to confounding because any unobserved risk factors are more likely to bias the results in favor of radical prostatectomy rather than conservative treatment The variation in estimates from the di fferent weighting methods possibly reflected a phenomenon of effect. That is, radical prostatectomy may benefit the population represented by the radical prostatectomy samples, but may not benefit the population represented by the conservative treatment samples. The effect modification can also be illustrated by the observed variation in propensity score quintile specific hazard ratios. Regardless of the additional HRQOL adjus tment, radi cal prostatectomy was not associated with all cause survival benefit for the first quintile with hazard ratios above 1.00. It was also noticed that nearly 50% conservatively treated men concentrated in the first quintile and nearly 50% definitively treatment men concentrated in the fifth quintile. This explained why the first quintile estimate dominated the overall estimation when weighting to th e conservative treatment group, and the fifth quintile estimate dominated the overall estimation when weighting t o the radical prostatectomy group. Discrepancy from different confounding adjustment methods in the presence of effect modification was previously reported by Kurth84 who estimated the effect of tissue plasminogen activator on mortality in stroke patients. Th ey found that weighting to the whole study sample yielded an odds ratio of 10.77, while weighting to the treated group yielded an odds ratio of 1.11. The authors pointed out that the treatment effect was modified by treatment contraindications. Weighting methods with different target populations might produce strikingly different effect estimates when treatment contraindication was common in the untreated group but rare in the treatment group. Adjustment of propensity score as continuous covariate or quinti le could not reveal the non uniform effect.

PAGE 111

111 In the context of localized prostate cancer treatment short life expectancy can be considered as contraindications that modified the effect of radical prostatectomy. The 10 year life expectancy rule basically sa ys radical prostatectomy offers survival benefit to men with at least 10 year Life expectancy but not to men with less than 10year life expectancy. Although we do not have a specific life expectancy variable, the descriptive analysis had showed that men t reated conservatively had multiple unfavorable characteristics, such as older age, more comorbidities, and worse HRQOL. These characteristics could collectively make radical prostatectomy in effective. No such effect modification was observed for radiation therapy. All confounding adjustment methods yielded similar results. This was because the radiation therapy group had risk factor distributions very similar to those in the conservative treatment group. The quintile specific estimates were also fairly homo geneous with a ll hazard ratios below 1.00. Comparison with Previous Studies Wong18 compared definitive treatment vs. expectant management in the SEER Medicare population. Comprehensive sociodemographic and clinical variables at baseline were controlled by adjusting propensity score as a continuous covariate. They reported an adjusted hazard ratio of 0.69 (95% CI: 0.660.72) for 10 year all cause mortality and an adjusted hazard ratio of 0.67 (95% CI: 0.58 0.77) for 10 year prostate cancer specific mortality The current study produced a hazard ratio of 0.44 (95% CI: 0.191.05) for all cause mortality and a hazard ratio of 0.26 (95% CI: 0.061.14) for prostate cancer specific mortality when using the same propensity score method to adjust a similar set of confounders. S everal reasons m ay explain why we observed stronger survival benefit than the re ference study: 1) Wong excluded men treated with hormonal therapy. Due to the limited sample size in CaPSURE, we combined both expectant management and hormonal therapy to compare with the def initive treatment. The hormonal therapy group appeared to have more advanced prostate

PAGE 112

112 cancer and worse health conditions than the expectant management group, which could make the definitive treatment appear more superior. 2) Wong identified primary treatme nt by using the ICD 9 and procedure codes in claims data within 6 months after diagnosis. The current study used a 9 month time window recommended by CaPSURE. The treatment type in CaPSURE was specified in urologist survey s which is supposed to be more accurate than claims codes. Moreover, we found that 12 men received radical prostatectomy and 120 men received radiation therapy between the 7th month and 9th month after diagnosis, account ing for approximately 10% of the definitive treatment group. If the t ime to treatment pattern observed in CaPSURE was generalizable, the SEER Medicare study misclassified approximately 10% definitively treated men into the expectant management group, which would have made the two comparison groups more similar, thus biased the treatment estimates towards the null. 3) Wong excluded men that died within 1 year after diagnosis while we did not use such an exclusion criterion. They claimed that more expectant management men were excluded due to this reason than definitive treatm ent men which favor ed the expectant management group. Liu19 separately compared different definitive treatments with expectant management in SEER Medicare men aged 65 75 years old. Baseline sociodemographic and clinical factors were adjusted by using multivariate Cox regression model. When radical prostatectomy was used alone, the adjusted hazard ratio for 10 year all cause mortality was 0.31 (95% CI: 0.250.37) When radical prosta tectomy was combined with radiation therapy, the hazard ratio w as 0.38 (95% CI: 0.28, 0.52) When radiation therapy was used alone, the hazard ratio w as 0.68 (95% CI: 0.560.81) Our study yielded similar results when adjusting the reference propensity score. After further adjusting the baseline HRQOL variables, the radical prostatectomy associated all cause survival benefit was attenuated with the hazard ratio was 0.46 (95% CI: 0.141.52) When

PAGE 113

113 comparing between radiation therapy and expectant management, we yielded a lower hazard ratio of 0.47 (95% CI: 0.211.06) Again, Liu used the 6month time window for primary treatment determination, which might had misclassified 10% radiation therapy patients into the expectant management group and biased their esti mates towards the null. Compared with the previous studies, our st udy has several advantages: 1) w e controlled for a number of baseline HRQOL variables in addition to comprehensive sociodemogr aphic and clinical factors. 2) m ore importantly, we evaluated t he treatment effects in clearly defined target populations by using the weighting methods, and revealed that radical prostatectomy might not be associated with survival benefit in the conservative treatment group had if these patients were treated with rad ical prostatectomy. The previous studies provided a single estimate without clearly defined target popula tion, which automatically hide the treatment effect heterogeneity It might result in a misleading perception that radical prostatectomy was superior t o conservative treatment regardless of patient characteristics. 3) w e would argue that it may not be appropriate to assess the effectiveness of radical prostatectomy and radiation therapy as a whole, since these two treatment population s have very distinct characteristics. Limitations The present study had several limitations. First, the observational nature made the current study inevitably vulnerable to unmeasured confou nding. No firm conclusion can be made regarding treatment superiority before large scale randomized trials are conducted. Secondly, approximately 70% of the CaPSURE participants do not have baseline HRQOL data available. Previous study showed that inability to complete the SF 36 questionnaire by older people was associated with poor physical and mental functioning .94 Physical and mental functioning are both key HRQOL domains. The missing data mechanism is therefore missing not at random and cannot be imputed correctly wit hout additional information. These patients

PAGE 114

114 had to be excluded from the current study. Completion of long health questionnaire i s a selection process. Men included in our study are presumably healthier than the general prostate cancer population. Thus, the generalizability of the current study may be limited. Thirdly, we might underestimate the confounding effect of the baseline HRQOL due to several study limitations The dichotomization of HRQOL might lead to oversimplification S ubtle yet important health information might be omitted The exclusion of men whose physical and mental status was too poor to complete the survey might reduce the variations of HRQOL in the study sample. The confounding effect of HRQOL is likely to be more pronounced in the general prostate cancer population. Nevertheless, these limitations are inherent in observational studies, reflecting the difficulties in observing and quantifying confounding variables. Lastly we used statistical models to account for uncertainties from missing data and informative censoring. We also used propensity score model and weighting model to control for the baseline confounding. The validity of the current study highly depends on the validity of the model assumptions such as missing at random censoring at random correct model specification etc. Given the data we had, s ome of these assumptions could not be validated. Summary and Conclusions Our study suggests that HRQOL is strongly associated with primary treatment selection in elderly men diagnosed with localized prostate cancer. Men treated with radical prostatectomy have significantly better HRQOL than men treated alternatively. Men treated with radiation therapy have moderately better HRQOL than men treated conservatively. The associations are more pronounced for physical health related HRQOL domains than the mental health domains. These findings are consistent with the current prostate cancer treatment guideline that definitive treatment is recommended to men with good health status and long life expectancy

PAGE 115

115 Selection of relatively healthy patients for definitive treatment causes strong confounding in observational studies assessing primary treatment effectivenes s with regard to all cause mortality. Our study show s that a dditional adjustment of dichotomized HRQOL variables besides commonly considered confounders is methodologically desirable in observational studies. However, the improvement in confounding reducti on appeared to be insufficient T he implausible association between definitive treatment and lower other cause mortality remained strong indicating the p ersistence of substantial residual confounding The effectiveness of definitive treatment tends to be overestimated Randomized trials are warranted to evaluate the efficacy of primary treatment for localized prostate cancer

PAGE 116

116 LIST OF REFERENCES 1. American Cancer Society Cancer f acts and f igures 2009. http://ww2.cancer.org/ downloads / STT/500809web.pdf Accessed Feburary 3, 2011. 2. National Comprehensive Cancer Network Prostate Cancer Clinical Practice Guidelines in Oncology V ersion 1.200 9. http://misc.medscap e. com/images/586/423/prostate.pdf Accessed Feburary 3, 2011. 3. Thompson I, Thrasher JB, Aus G, et al. Guideline for the management of clinically localized prostate cancer: 2007 update. J Urol. 2007;177:21062131. 4. Sadetsky N, Hubbard A, Carroll PR, Satariano W. Predictive value of ser ial measurements of quality of life on all cause mortality in prostate cancer patients: data from CaPSURE (cancer of the prostate strategic urologic research endeavor) database. Qual Life Res. 2009;18:10191027. 5. Lis CG, Gupta D, Grutsch JF. Patient sat isfaction with health related quality of life: implications for prognosis in prostate cancer. Clin Genitourin Cancer. 2008;6:9196. 6. Zeliadt SB, Ramsey SD, Penson DF, et al. Why do men choose one treatment over another?: a review of patient decision making for localized prostate cancer. Cancer. 2006;106:18651874. 7. Giordano SH, Kuo YF, Duan Z, Hortobagyi GN, Freeman J, Goodwin JS. Li mits of observational data in determining outcomes from cancer therapy. Cancer. 2008;112:24562466. 8. Litwin MS, Miller DC. Treating older men with prostate cancer: survival (or selection) of the fittest? JAMA. 2006;296:27332734. 9. Guyatt GH, Feeny DH Patrick DL. Measuring healthrelated quality of life. Ann Intern Med. 1993;118:622629. 10. Zeliadt SB, Ramsey SD, Potosky AL, et al. Association of Preexisting Symptoms with Treatment Decisions among Newly Diagnosed Prostate Cancer Patients. Patient. 2008;1:189. 11. Saigal CS, Gornbein J, Nease R, Litwin MS. Predictors of utilities for health states in early stage prostate cancer. J Urol. 2001;166:942946. 12. BillAxelson A, Holmberg L, Ruutu M, et al. Radical prostatectomy versus watchful waiting in early prostate cancer. N Engl J Med. 2005;352:19771984. 13. Wilt TJ, Brawer MK, Barry MJ, et al. The Prostate cancer Intervention Versus Observation Trial:VA/NCI/AHRQ Cooperative Studies Program #407 (PIVOT): design and baseline results of a randomized controlled trial comparing radical prostatectomy to watchful waiting for men with clinically localized prostate cancer. Contemp Clin Trials. 2009;30:8187. 14. Albertsen PC, Hanley JA, Penson DF, Barrows G, Fine J. 13year outcomes following treatment for clinically localized prostate cancer in a population based cohort. J Urol. 2007;177:932936.

PAGE 117

117 15. Merglen A, Schmidlin F, Fioretta G, et al. Short and long term mortality with localized prostate cancer. Arch Intern Med. 2007;167:1944 1950. 16. Tewari A, Raman JD, Chang P, Rao S, Divine G, Menon M. Long term survival probability in men with clinically localized prostate cancer treated either conservatively or with definitive treatment (radiotherapy or radical prostatectomy). Urology. 2006;68:12681274. 17. Tward JD, Lee CM, Pappas LM, Szabo A, Gaffney DK, Shrieve DC. Survival of men with clinically localized prostate cancer treated with prostatectomy, brachytherapy, or no definitive treatment: impact of age at diagnosis. Cancer. 2006;107:23922400. 18. Wong YN, Mitra N, Hudes G, et al. Survival associated with treatment vs observation of localized prostate cancer in elderly men. JAMA. 2006;296:26832693. 19. Liu L, Coker AL, Du XL, Cormier JN, Ford CE, Fang S. Long term survival after radical prostatectomy compared to other treatments in older men with local/regional prostate cancer. J Surg Oncol. 2008;97:583591. 20. Rothman KJ G reenland S, Lash TL. Modern epidemiology. 3rd Ed. Philadelphia ( PA ) : Lippincott, Williams & Wilkins.; 2008. 21. Lubeck DP, Litwin MS, Henning JM, et al. The CaPSURE database: a methodology for clinical practice and research in prostate cancer. CaPSURE Research Panel. Cancer of the Prostate Strategic Urologic Research Endeavor. Urology. 1996;48:773777. 22. Ware JE SK, Kosisnki M, Gandek B. SF 36 Health Survey Manual and Interpretation Guide. Boston ( MA ) : The Health Institute ; 1993. 23. Ware JE KM, Keller SD. SF 36 Physical and Mental health Summary Scales: A User's Manual. Boston ( MA ) : The Health Institue ; 1994. 24. Litwin MS, Hays RD, Fink A, Ganz PA, Leake B, Brook RH. The UCLA Prostate Cancer Index: development, reliability, and validity of a healthrelated quality of life meas ure. Med Care. 1998;36:10021012. 25. Greenland S, Morgenstern H. Confounding in health research. Annu Rev Public Health. 2001;22:189212. 26. Greenland S, Pearl J, Robins JM. Causal diagrams for epidemiologic research. Epidemiology. 1999;10:3748. 27. Greenland S, Robins JM. Identifiability, exchangeability and confounding revisited. Epidemiol Perspect Innov. 2009;6:4. 28. Lessick M, Katz A. A genetics perspective on prostate cancer. Urol Nurs. 2006;26:454460. 29. Cooperberg MR, Grossfeld GD, Lubeck DP, Carroll PR. National practice patterns and time trends in androgen ablation for localized prostate cancer. J Natl Cancer Inst. 2003;95:981989.

PAGE 118

118 30. Potosky AL, Reeve BB, Clegg LX, et al. Quality of life following localized prostate cancer treated initially with androgen deprivation therapy or no therapy. J Natl Cancer Inst. 2002;94:430437. 31. Wallace M. Uncertainty and quality of life of older men who undergo watchful waiting for prostate cancer. Oncol Nurs Forum. 2003;30:303309. 32. Johansson E, BillAxelson A, Holmberg L, Onelov E, Johansson JE, Steineck G. Time, symptom burden, androgen deprivation, and self assessed quality of life after radical prostatectomy or watchful waiting: the Randomized Scandinavian Prostate Cancer Group Study Number 4 (SPCG 4) clinical trial. Eur Urol. 2009;55:422430. 33. Crawford ED, Bennett CL, Stone NN, et al. Comparison of perspectives on prostate cancer: analyses of survey data. Urology. 1997;50:366372. 34. National Comprehensive Cancer Network Senior Adult Oncology Clinical Practice Guidelines in Oncolog. Version 2.2007. http://www.nccn.org/professionals/physician_gls/ PDF/senior.pdf Accessed May 16, 2008. 35. Thompson RH, Slezak JM, Webster WS, Lieber MM. Radical prostatectomy for octogenarians: how old is too old? Urology. 2006;68:10421045. 36. Walz J, Gallina A, Perrotte P, et al. Clinicians are poor raters of life expectancy before radical prostatectomy or definitive radiotherapy for localized prostate cancer. BJU Int. 2007;100:12541258. 37. Marr PL, Elkin EP, Arredondo SA, Broering JM, DuChane J, Carroll PR. Comorbidity and primary treatment for localized prostate cancer: data from CaPSURE. J Urol. 2006; 175:13261331. 38. Desch CE, Penberthy L, Newschaffer CJ, et al. Factors that determine the treatment for local and regional prostate cancer. Med Care. 1996;34:152162. 39. Schymura MJ, Kahn AR, German RR, et al. Factors associated with initial treatment and survival for clinically localized prostate cancer: results from the CDC NPCR Patterns of Care Study (PoC1). BMC Cancer. 2010;10:152. 40. Fowler FJ, Jr., McNaughton Collins M, Albertsen PC, Zietman A, Elliott DB, Barry MJ. Comparison of recommendations by urologists and radiation oncologists for treatment of clinically localized prostate cancer. JAMA. 2000;283:32173222. 41. Jang TL, Bekelman JE, Liu Y, et al. Physician visits prior to treatment for clinically localized prostate cancer. Arch Intern Med. 2010;170:440450. 42. Barry MJ. The prostate cancer treatment bazaar: comment on "Physician visits prior to treatment for clinically localized prostate cancer". Arch Intern Med. 2010;170:450452.

PAGE 119

119 43. Feldman Stewart D, Brundage MD, Van Manen L, Svenson O. Patient focussed decision making in early stage prostate cancer: insights from a cognitively based decision aid. Health Expect. 2004;7:126141. 44. Mazur DJ, Merz JF. How older patients' treatment pr eferences are influenced by disclosures about therapeutic uncertainty: surgery versus expectant management for localized prostate cancer. J Am Geriatr Soc. 1996;44:934937. 45. Hall JD, Boyd JC, Lippert MC, Theodorescu D. Why patients choose prostatectomy or brachytherapy for localized prostate cancer: results of a descriptive survey. Urology. 2003;61:402407. 46. Gwede CK, PowSang J, Seigne J, et al. Treatment decisionmaking strategies and influences in patients with localized prostate carcinoma. Cance r. 2005;104:13811390. 47. Underwood W, De Monner S, Ubel P, Fagerlin A, Sanda MG, Wei JT. Racial/ethnic disparities in the treatment of localized/regional prostate cancer. J Urol. 2004;171:15041507. 48. Moses KA, Paciorek AT, Penson DF, Carroll PR, Mas ter VA. Impact of ethnicity on primary treatment choice and mortality in men with prostate cancer: data from CaPSURE. J Clin Oncol. 2010;28:10691074. 49. Hoffman RM, Harlan LC, Klabunde CN, et al. Racial differences in initial treatment for clinically lo calized prostate cancer. Results from the prostate cancer outcomes study. J Gen Intern Med. 2003;18:845 853. 50. Denberg TD, Beaty BL, Kim FJ, Steiner JF. Marriage and ethnicity predict treatment in localized prostate carcinoma. Cancer. 2005;103:18191825. 51. Denberg TD, Glode LM, Steiner JF, Crawford ED, Hoffman RM. Trends and predictors of aggressive therapy for clinical locally advanced prostate carcinoma. BJU Int. 2006;98:335340. 52. Rose AJ, Backus BM, Gershman ST, Santos P, Ash AS, Battaglia TA. Predictors of aggressive therapy for nonmetastatic prostate carcinoma in Massachusetts from 1998 to 2002. Med Care. 2007;45:440447. 53. Klabunde CN, Potosky AL, Harlan LC, Kramer BS. Trends and black/white differences in treatment for nonmetastatic prostate cancer. Med Care. 1998;36:13371348. 54. Krupski TL, Kwan L, Afifi AA, Litwin MS. Geographic and socioeconomic variation in the treatment of prostate cancer. J Clin Oncol. 2005;23:78817888. 55. Boberg EW, Gustafson DH, Hawkins RP, et al. Assessing the unmet information, support and care delivery needs of men with prostate cancer. Patient Educ Couns. 2003;49:233242. 56. Lane T. Partner's influence on patient preference for treatment in early prostate cancer. BJU Int. 2004;93:887.

PAGE 120

120 57. Patel RH, Mirsadraee S, Emberton M. Partner's influence on patient preference for treatment in early prostate cancer. BJU Int. 2004;93:425. 58. Srirangam SJ, Pearson E, Grose C, Brown SC, Collins GN, O'Re illy PH. Partner's influence on patient preference for treatment in early prostate cancer. BJU Int. 2003;92:365369. 59. Polednak AP. Prostate cancer treatment in black and white men: the need to consider both stage at diagnosis and socioeconomic status. J Natl Med Assoc. 1998;90(2):101104. 60. Yan Y, Carvalhal GF, Catalona WJ, Young JD. Primary treatment choices for men with clinically localized prostate carcinoma detected by screening. Cancer. 2000;88:11221130. 61. Miles BJ, Giesler B, Kattan MW. Recall and attitudes in patients with prostate cancer. Urology. 1999;53:169174. 62. Davison BJ, Oliffe JL, Pickles T, Mroz L. Factors influencing men undertaking active surveillance for the management of low risk prostate cancer. Oncol Nurs Forum. 2009;36:8996. 63. Sommers BD, Beard CJ, D'Amico AV, Kaplan I, Richie JP, Zeckhauser RJ. Predictors of patient preferences and treatment choices for localized prostate cancer. Cancer. 2008;113:20582067. 64. Ziebland S, Herxheimer A. How patients' experiences contribute to decision making: illustrations from DIPEx (personal experiences of health and illness). J Nurs Manag. 2008;16:433439. 65. Newman SC. Commonalities in the classical, collapsibility and counterfactual concepts of confounding. J Clin Epidemiol. 2004;57:325329. 66. Greenland S, Robins JM. Identifiability, exchangeability, and epidemiological confounding. Int J Epidemiol. 1986;15:413419. 67. Nurminen M. On the epidemiologic notion of confounding and confounder identification. Scand J Work Envi ron Health. 1997;23:6468. 68. Hernan MA, Hernandez Diaz S, Werler MM, Mitchell AA. Causal knowledge as a prerequisite for confounding evaluation: an application to birth defects epidemiology. Am J Epidemiol. 2002;155:176184. 69. Maldonado G, Greenland S. Simulation study of confounder selection strategies. Am J Epidemiol. 1993;138:923936. 70. Hernan MA, Hernandez Diaz S, Robins JM. A structural approach to selection bias. Epidemiology. 2004;15:615625. 71. Greenland S, Robins JM. Confounding and misclassification. Am J Epidemiol. 1985;122:495506.

PAGE 121

121 72. Greenland S. Absence of confounding does not correspond to collapsibility of the rate ratio or rate difference. Epidemiology. 1996;7:498501. 73. Fried LP, S torer DJ, King DE, Lodder F. Diagnosis of illness presentation in the elderly. J Am Geriatr Soc. 1991;39 :117123. 74. Lash TL, Mor V, Wieland D, Ferrucci L, Satariano W, Silliman RA. Methodology, design, and analytic techniques to address measurement of comorbid disease. J Gerontol A Biol Sci Med Sci. 2007;62:281285. 75. Balducci L, Beghe C. The application of the princ iples of geriatrics to the management of the older person with cancer. Crit Rev Oncol Hematol. 2000;35:147154. 76. Geraci JM, Escalante CP, Freeman JL, Goodwin JS. Comorbid disease and cancer: the need for more relevant conceptual models in health servic es research. J Clin Oncol. 2005;23:73997404. 77. Boyd CM, Weiss CO, Halter J, Han KC, Ershler WB, Fried LP. Framework for evaluating disease severity measures in older adults with comorbidity. J Gerontol A Biol Sci Med Sci. 2007;62:286295. 78. Ware JE, Jr. SF 36 health survey update. Spine. 2000;25:31303139. 79. Parker SG, Peet SM, Jagger C, Farhan M, Castleden CM. Measuring health status in older patients. The SF 36 in practice. Age Ageing. 1998;27:1318. 80. Lis CG, Gupta D, Grutsch JF. Patient sat isfaction with health related quality of life: implications for prognosis in prostate cancer. Clin Genitourin Cancer. 2008;6:9196. 81. Faller H, Stork S, Schowalter M, et al. Is health related quality of life an independent predictor of survival in patie nts with chronic heart failure? J Psychosom Res. 2007;63:533538. 82. Rubin DB. Multiple Imputation for Nonresponse in Surveys New York: John Wiley & Sons, Inc; 1987. 83. Allison PC. Imputation of Categorical Variables with PROC MI Proceedings from SUGI 30: T he 30th M eeting of SAS Users Group International Conference ; Apr 1013, 2005; Philadelphia, PA. 84. Rubin DB. Estimating causal effects from large data sets using propensity scores. Ann Intern Med. 1997;127:757763. 85. Brookhart MA, Schneewei ss S, Rothman KJ, Glynn RJ, Avorn J, Sturmer T. Variable selection for propensity score models. Am J Epidemiol. 2006;163:11491156. 86. Glynn RJ, Schneeweiss S, Sturmer T. Indications for propensity scores and review of their use in pharmacoepidemiology. Basic Clin Pharmacol Toxicol. 2006;98:253259.

PAGE 122

122 87. Robins JM Information recovery and bias adjustment in proportional hazards regression analysis of randomized trials using surrogate markers In: Proceedings of the Biopharmaceutical Section, American Statistical Association; 1993 ; pp.2433. http://www.biostat.harvard.edu /~ robins/biopharm.pdf. Accessed Feburary 2, 2011. 88. Robins JM, Hernan MA, Brumback B. Marginal structural models and causal inference in epidemiology. Epidemiology. 2000;11:550560. 89. Hernan MA, Brumback B, Robins JM. Marginal structural models to estimate the causal effect of zidovudine on the survival of HIV positive men. Epidemiology. 2000;11:561570. 90. Westreich D, Cole SR, Tien PC, et al. Time scale and adjusted survival cu rves for marginal structural cox models. Am J Epidemiol. 2010;171:691700. 91. Miettinen OS. The need for randomization in the study of intended effects. Stat Med. 1983;2:267271. 92. Audisio RA, Pope D, Ramesh HS, et al. Shall we operate? Preoperative a ssessment in elderly cancer patients (PACE) can help. A SIOG surgical task force prospective study. Crit Rev Oncol Hematol. 2008;65:156163. 93. Jang TL, Bekelman JE, Liu Y, et al. Physician visits prior to treatment for clinically localized prostate cancer. Arch Intern Med.2010;170:440450. 94. Parker SG, Bechinger English D, Jagger C, Spiers N, Lindesay J. Factors affecting completion of the SF 36 in older people. Age Ageing. 2006;35:376381.

PAGE 123

123 BIOGRAPHICAL SKETCH Jinghua He was born and raised in Beijing, China. In 2001, he obtained a Bachelor of Science degree in pharmacology from the Department of Pharmacology, College of Pharmacy Peking University Health Science Center In 2005, he finished his Master of Science degree in biomedical science from the University of South Dakota. In the same year he joined the Department of Pharmaceutical Outcomes and Policy at the College of Pharmacy University of Florida as an Alumni Fellow to pursue a Doctor of Philosophy degree in pharmacoepidemiology.