<%BANNER%>

Prostate Cancer Clinical Practice Guidelines: Clinical and Economic Outcomes

Permanent Link: http://ufdc.ufl.edu/UFE0010098/00001

Material Information

Title: Prostate Cancer Clinical Practice Guidelines: Clinical and Economic Outcomes
Physical Description: Mixed Material
Copyright Date: 2008

Record Information

Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
System ID: UFE0010098:00001

Permanent Link: http://ufdc.ufl.edu/UFE0010098/00001

Material Information

Title: Prostate Cancer Clinical Practice Guidelines: Clinical and Economic Outcomes
Physical Description: Mixed Material
Copyright Date: 2008

Record Information

Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
System ID: UFE0010098:00001


This item has the following downloads:


Full Text

PAGE 1

PROSTATE CANCER CLINICAL PRACTICE GUIDELINES: CLINICAL AND ECONOMIC OUTCOMES By MICHAEL DENNIS TAYLOR A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2005

PAGE 2

Copyright 2005 by Michael Dennis Taylor

PAGE 3

This dissertation is dedicated to my gra ndparents, Lonnie Friday Persons and James Monroe Persons; and Margaret Miller Taylor and Lawrence Dennis Taylor.

PAGE 4

ACKNOWLEDGMENTS I thank my parents, John and Kay Taylor; and my sister, Kelli Taylor for the support they provided throughout my years as a graduate student. I would also like to thank the Faculty of the Department of Pharmacy Health Care Administration. I would like to acknowledge the guidance and support provided by Drs. Abraham Hartzema, Almut Winterstein, Richard Segal, Michael Daniels, and Charles Riggs. I will be forever grateful for the advice each provided throughout the dissertation process. Finally, I would like to acknowledge the financial support I received from the American Foundation for Pharmaceutical Education and the Perry A. Foote Small Grants Program in Health Outcomes Research. iv

PAGE 5

TABLE OF CONTENTS page ACKNOWLEDGMENTS.................................................................................................iv LIST OF TABLES............................................................................................................vii LIST OF FIGURES...........................................................................................................ix ABSTRACT.........................................................................................................................x TABLE 1 INTRODUCTION........................................................................................................1 Background...................................................................................................................1 Need for the Study........................................................................................................1 Purpose of the Study.....................................................................................................2 Specific Objectives.......................................................................................................3 Research Questions and Hypotheses............................................................................3 Research Question 1..............................................................................................4 Research Question 1 Hypothesis...........................................................................4 Research Question 2..............................................................................................4 Research Question 2 Hypothesis...........................................................................4 Summary.......................................................................................................................5 2 LITERATURE REVIEW.............................................................................................6 Prostate Cancer.............................................................................................................6 Prostate Cancer-Specific Mortality.......................................................................8 Expectant management versus radical prostatectomy....................................8 Radical prostatectomy versus radiotherapy....................................................9 Expectant management versus radical prostatectomy versus external beam radiotherapy....................................................................................10 Summary of prostate cancer-specific mortality literature............................12 Prostate Cancer Economics.................................................................................14 Guidelines...................................................................................................................20 Development of Guidelines.................................................................................21 Quality of Guidelines..........................................................................................23 National Comprehensive Cancer Network Prostate Cancer Guideline...............24 Cost Estimation...........................................................................................................26 v

PAGE 6

Cost-effectiveness................................................................................................31 3 METHODS.................................................................................................................33 Data Source.................................................................................................................33 Establishment of Comparison Cohorts.......................................................................34 Risk of Recurrence..............................................................................................34 Initial Treatment..................................................................................................36 Propensity Scores................................................................................................38 Clinical Effectiveness.................................................................................................39 Cost of Treatment.......................................................................................................40 Cost-effectiveness.......................................................................................................46 4 RESULTS...................................................................................................................50 Expectant Management vs. External Beam Radiotherapy.........................................53 Expectant Management vs. Brachytherapy................................................................54 Expectant Management vs. Prostatectomy.................................................................55 Expectant Management vs. External Beam Radiotherapy with Brachytherapy.........56 External Beam Radiotherapy vs. Brachytherapy........................................................57 External Beam Radiotherapy vs. Prostatectomy........................................................57 External Beam Radiotherapy vs. External Beam Radiotherapy with Brachytherapy.........................................................................................................59 Brachytherapy vs. Prostatectomy...............................................................................60 Brachytherapy vs. External Beam Radiotherapy with Brachytherapy.......................60 Prostatectomy vs. External Beam Radiotherapy with Brachytherapy........................61 5 DISCUSSION.............................................................................................................73 Limitations..................................................................................................................73 Prostate Cancer-Specific Mortality.............................................................................75 Prostate Cancer-Specific Cost....................................................................................76 Conclusions.................................................................................................................80 LIST OF REFERENCES...................................................................................................82 BIOGRAPHICAL SKETCH.............................................................................................88 vi

PAGE 7

LIST OF TABLES Table page 3-1 Radiotherapy (external beam or brachytherapy) CPT/HCPCS codes......................48 3-2 Androgen deprivation CPT/HCPCS codes..............................................................48 3-3 Androgen deprivation ICD9-CM procedure codes..................................................49 3-4 Other prostate cancer treatment CPT/HCPCS codes...............................................49 3-5 Prostatectomy CPT/HCPCS codes...........................................................................49 3-6 Prostatectomy ICD9-CM procedure codes..............................................................49 4-1 Race distribution by 5-year prostate cancer-specific survival status.......................62 4-2 Urban/rural classification distribution by 5-year prostate cancer-specific survival status.........................................................................................................................62 4-3 Tumor grade distribution by 5-year prostate cancer-specific survival status...........62 4-4 Tumor stage distribution by 5-year prostate cancer-specific survival status...........62 4-5 Diagnosis year by 5-year prostate cancer-specific survival status...........................63 4-6 Race distribution for patients treated with recommended and non-recommended strategies...................................................................................................................63 4-7 Urban/rural classification distribution for patients treated with recommended and non-recommended strategies.............................................................................63 4-8 Tumor grade distribution for patients treated with recommended and non-recommended strategies...........................................................................................63 4-9 Tumor stage distribution for patients treated with recommended and non-recommended strategies...........................................................................................64 4-10 Diagnosis year distribution for patients treated with recommended and non-recommended strategies...........................................................................................64 4-11 Race distribution by initial treatment strategy.........................................................64 vii

PAGE 8

4-12 Urban/rural classification distribution by initial treatment strategy.........................64 4-13 Tumor grade distribution by initial treatment strategy.............................................64 4-14 Tumor stage distribution by initial treatment strategy.............................................65 4-15 Diagnosis year distribution by initial treatment strategy.........................................65 4-16 Age and Charlson comorbidity score by initial treatment strategy..........................65 viii

PAGE 9

LIST OF FIGURES Figure page 4-1 Expectant management (EM) vs. external beam radiotherapy (EBR) log cumulative hazard vs. log time................................................................................66 4-2 Expectant management vs. external beam radiotherapy cumulative cost by year...66 4-3 Expectant management vs. brachytherapy cumulative cost by year........................67 4-4 Expectant management (EM) vs. prostatectomy (RP) log cumulative hazard vs. log time.....................................................................................................................67 4-5 Expectant management vs. prostatectomy cumulative cost by year........................68 4-6 Expectant management vs. external beam radiotherapy with brachytherapy cumulative cost by year............................................................................................68 4-7 External beam radiotherapy vs. brachytherapy cumulative cost by year.................69 4-8 External beam radiotherapy (EBR) vs. prostatectomy (RP) log cumulative hazard vs. log time...................................................................................................69 4-9 External beam radiotherapy vs. prostatectomy cumulative cost by year.................70 4-10 External beam radiotherapy vs. external beam radiotherapy with brachytherapy cumulative cost by year............................................................................................70 4-11 Brachytherapy vs. prostatectomy cumulative cost by year......................................71 4-12 Brachytherapy vs. external beam radiotherapy with brachytherapy cumulative cost by year...............................................................................................................71 4-13 Prostatectomy vs. external beam radiotherapy with brachytherapy cumulative cost by year...............................................................................................................72 ix

PAGE 10

Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy PROSTATE CANCER CLINICAL PRACTICE GUIDELINES: CLINICAL AND ECONOMIC OUTCOMES By Michael Dennis Taylor May 2005 Chair: Abraham Hartzema Major Department: Pharmacy Health Care Administration Prostate cancer has a significant impact on public health in the United States. Despite the high prevalence of prostate cancer, there is still no clear evidence that any one treatment is superior to all others. Furthermore, little is known about the cost of most prostate cancer treatment strategies beyond a 1-year time frame. We conducted a retrospective cohort study using the Surveillance, Epidemiology, and End Results (SEER)-Medicare database to examine the clinical and economic outcomes of prostate cancer treatment strategies over a 5-year time frame for patients at low and intermediate risk of prostate cancer recurrence. Clinical effectiveness of each treatment strategy was evaluated in terms of prostate cancer-specific survival. Kaplan-Meier survival analysis was used to estimate the survival function and the Cox proportional hazards model was used to test for differences. A modification of Kaplan-Meier survival using inverse probability weighting was used to estimate costs in the presence of censoring. x

PAGE 11

The study is subject to selection bias because of the lack of randomization. As a result of the selection bias, differences in study end-points may be due to differences in the comparison groups rather than differences in the treatments. However, the use of multivariate analyses allowed us to control for baseline differences in patient characteristics. We identified statistically significant differences in 5-year prostate cancer-specific cost of healthcare between treatment strategies recommended in prostate cancer practice guidelines, ranging from $630 (comparing external beam radiotherapy to brachytherapy) and $6740 (comparing external beam radiotherapy with brachytherapy to expectant management). We also found significant differences in prostate cancer-specific mortality when comparing external beam radiotherapy to expectant management, prostatectomy to expectant management, and prostatectomy to external beam radiotherapy. Finally, we were able to provide valuable information regarding the cost of preventing one death due to prostate cancer over a 5-year time period associated with the use of external beam radiotherapy rather than expectant management ($600,000), prostatectomy rather than expectant management ($646,670), and prostatectomy rather than external beam radiotherapy ($465,710). xi

PAGE 12

CHAPTER 1 INTRODUCTION Background Outcomes of health care are typically labeled as one of three types: clinical, humanistic, or economic. Clinical outcomes are results of an intervention or lack thereof expressed in clinical terms. An example of a clinical outcome in the treatment of a post-myocardial infarction patient with a -blocker would be the prevention of a second myocardial infarction. Humanistic outcomes are typically expressed as the impact of an intervention on an individuals quality of life (QOL); or more specifically, the individuals health related quality of life (HrQOL). An example of a humanistic outcome would be an increase in one of the domains of the patients SF-36 (a measure of health-related quality of life) score. Economic outcomes are the results of an intervention in monetary terms. Perhaps the simplest method of expressing an economic outcome is to say an intervention results in a net savings of x dollars. However, an economic outcome might also be expressed in terms of the additional cost of an intervention in dollars per unit of clinical improvement. Need for the Study Clinical practice guidelines are developed to provide health care practitioners with a guide for treating an illness in a manner most likely to result in the best possible outcomes. The question thus arises, what type of outcomes? All too often, guidelines are developed with a focus on only clinical outcomes. Humanistic and economic outcomes might be considered, but guideline development typically lacks formal consideration of 1

PAGE 13

2 humanistic and economic outcomes. Originally, guidelines were developed based upon expert opinion, a strategy fraught with potential bias. Today, guidelines are expected to be based on published evidence in peer-reviewed journals identified using systematic searches of the literature (1). Today, guideline-development groups may consider economic outcomes while formulating recommendations for inclusion in clinical practice guidelines. However, todays consideration of economic outcomes is akin to yesteryears consideration of clinical evidence; that is, it lacks formal evaluation. The lack of unlimited health care resources dictates a need for the inclusion of formal cost analyses in the development of clinical practice guidelines that shape much of the practice of medicine in todays society. The Institute of Medicine (2, p. 25) has called for clinical practice guidelines to be accompanied by explicit projections of the health and cost outcomes [emphasis added] expected from the use of particular services or procedures. In many clinical scenarios, multiple treatment strategies result in equivalent clinical outcomes. Such scenarios dictate a need for some other criterion on which to determine the preferred treatment strategy. Given that health care resources are limited, economic outcomes may be a good criterion for choosing treatment strategies. Purpose of the Study Our study used formal economic analysis methods to estimate the cost outcomes of interventions proposed by clinical practice guidelines. We applied economic analysis methodologies to the recommendations in the National Comprehensive Cancer Network (NCCN) prostate cancer clinical practice guideline for patients with cancer localized to the prostate.

PAGE 14

3 We chose prostate cancer guidelines for two reasons. First, prostate cancer is a significant public health concern, with 230,855 men expected to have been diagnosed with prostate cancer and 29,089 men expected to have died of prostate cancer in the United States during 2004 (3). Second, the guidelines are not able to recommend one treatment strategy over all others on the basis of clinical evidence for many stages of prostate cancer (4). Rather, the guidelines provide several treatment strategies as being acceptable for many prostate cancer clinical scenarios. Because of scenarios (such as those in the NCCN prostate cancer clinical practice guideline) in which one treatment strategy cannot be advocated above all others on the basis of clinical evidence, consideration of economic and/or humanistic outcomes is essential. Specific Objectives In demonstrating the use of formal economic analysis methods to estimate the cost outcomes of interventions included in clinical practice guidelines, our study addressed three specific objectives. First, we determined the clinical effectiveness of each NCCN prostate cancer clinical practice guideline treatment recommendation for patients at low and intermediate risk of prostate cancer recurrence (i.e., patients with cancer localized to the prostate) using observational data. Second, we determined the cost for each treatment option. Finally, we determined the minimum willingness-to-pay value resulting in a treatment that is both more effective and more expensive being cost effective. Research Questions and Hypotheses The first research question examined the clinical effectiveness of prostate cancer treatment strategies among men at least 65 years of age when diagnosed with prostate cancer using the SEER-Medicare linked database for the years 1992 to 1999. The

PAGE 15

4 clinical effectiveness of each strategy was defined as the treatment-specific survival function describing the time to death due to prostate cancer. Research Question 1 Is there a difference in clinical effectiveness between expectant management, 3-dimensional conformal radiotherapy (3D-CRT), prostatectomy, brachytherapy, and 3D-CRT with brachytherapy among newly diagnosed prostate cancer patients at least 65 years of age at low or intermediate risk of prostate cancer recurrence? Research Question 1 Hypothesis The null hypothesis was that the expectant management, 3D-CRT, prostatectomy, brachytherapy, and 3D-CRT with brachytherapy survival functions are all equal. The alternative hypothesis was at least one of the survival functions differed. The second research question examined the cost of prostate cancer treatment strategies among men at least 65 years of age when diagnosed with prostate cancer. Relevant costs included both the cost of the initial treatment and all prostate cancer-specific follow-up costs. Research Question 2 Is there a difference in prostate cancer-specific health care costs between expectant management, 3D-CRT, prostatectomy, brachytherapy, and 3D-CRT with brachytherapy among newly diagnosed prostate cancer patients at least 65 years of age at low or intermediate risk of prostate cancer recurrence? Research Question 2 Hypothesis The null hypothesis was the prostate cancer-specific health care costs for expectant management, 3D-CRT, prostatectomy, brachytherapy, and 3D-CRT with brachytherapy

PAGE 16

5 are all equal. The alternative hypothesis was the cost was different for at least one of the treatment strategies. Summary The long-term goal of our study was to demonstrate the use of formal economic analyses in evaluating treatment strategies being considered for inclusion in clinical practice guidelines. It is hoped that such a demonstration will foster guideline development panels to conduct formal economic analyses themselves, or to commission such analyses for use in the guideline development process and thus lead to the advancement of health care interventions based on clinical and economic impact, rather than clinical impact alone. Beyond mere consideration of economic outcomes, it is hoped the results of economic analyses of proposed guideline recommendations will be used to provide a supplemental grade to each recommendation, an idea proposed by Mason and colleagues (5). In other words, each recommendation in a guideline could have two grades, one based on clinical outcomes and a second based on economic outcomes. Or ultimately, perhaps, a system could be devised to consider all evidence (clinical, economic, and humanistic) and provide one overall grade.

PAGE 17

CHAPTER 2 LITERATURE REVIEW Prostate Cancer According to statistics provided by the American Cancer Society (3), the most common site of cancer among men in the United States is the prostate, with an estimated 230,110 men expected to have developed prostate cancer in 2004. Furthermore, prostate cancer is the second leading cause of cancer mortality, with an estimated 29,900 men expected to have died due to prostate cancer in 2004 (3). Prostate cancer is often described in terms of prostate-specific antigen (PSA) level, clinical stage, and histologic grade, as all three characteristics have been found to be predictive of prostate cancer outcomes (6-9). Khan and Partin (10) report that the best understanding of a patients extent of disease is arrived at by jointly considering the findings of the digital rectal examination (i.e., clinical stage), PSA, and Gleason score (i.e., histologic grade as determined by biopsy). In fact, studies conducted by Partin and others have led to the development, validation, and updating of nomograms, commonly referred to as Partin tables, which use the DRE findings, PSA level, and Gleason score to predict the true pathological stage of a patients prostate cancer. The better understanding of the true severity of a patients prostate cancer can then be used in making more educated decisions as to the appropriate treatment for the tumor. Prostate-specific antigen (PSA) is used both as a screening tool and a tumor marker to identify the response of prostate cancer to treatment and the relapse of prostate cancer following treatment (11). In terms of screening, a PSA value equal to 4.0 ng/mL has 6

PAGE 18

7 historically been viewed as the upper limit of normal. However, controversy surrounds this number as many patients with a PSA greater than 4.0 ng/mL do not have prostate cancer, and many patients with a PSA less than 4.0 ng/mL do have prostate cancer. The clinical stage of a malignant tumor of the prostate is typically classified using the 2002 American Joint Committee on Cancer (AJCC) Tumor, Node, and Metastasis (TNM) Staging System for Prostate Cancer (4). The AJCC TNM Staging System for Prostate Cancer uses characteristics of the tumor itself, involvement of lymph nodes, and presence and extent of metastases to clinically stage prostate cancer (12). The Gleason score is a histological summary of the tumor pathology computed by summing the Gleason grade for the tumors two most prevalent patterns of pathology. Each pathology pattern receives a grade ranging from one to five with one being well differentiated and 5 being undifferentiated. Thus the Gleason score ranges from two to ten (13). Treatments for localized prostate cancer include radical prostatectomy, external beam radiotherapy, brachytherapy, and expectant management (watchful waiting) (14-20). For localized prostate cancer, available clinical evidence has not demonstrated a clear superiority of one treatment over all others and its management is described by many as controversial (8,14,18-23). A patients treatment regimen could consist of any one or a combination of these treatment modalities depending on a variety of factors (16,24), including tumor stage (15,20,21,25,26), prostate-specific antigen level (15,16,20,25-27), histological grade of the tumor (15,16,20,21,25-28), patient age and life expectancy (15,21,26-28), comorbidities (15,21), cost(16), anticipated side effects, and patients preferences regarding the various side effects (15,16,19).

PAGE 19

8 Studies examining prostate cancer treatments have evaluated the various treatments using a number of endpoints. Endpoints evaluated include all-cause mortality, disease-specific mortality, and intermediate outcomes such as metastasis-free survival, clinical recurrence, and biochemical recurrence, among others. Our study focuses on prostate cancer-specific mortality. Prostate Cancer-Specific Mortality Four articles published since 1995 have compared the rate of prostate cancer-specific mortality of two or more of the three main treatment strategies for localized prostate cancer. Of the four studies comparing treatment strategies on the basis of prostate cancer-specific mortality data, one study compared expectant management and radical prostatectomy (18), one study compared radical prostatectomy and external beam radiotherapy (7), and two studies compared expectant management, radical prostatectomy, and radiotherapy (8,21). Expectant management versus radical prostatectomy Holmberg et al. report the results of a prospective trial in which patients with prostate cancer were randomized to either radical prostatectomy or watchful waiting (expectant management) (18). The study population consisted of 695 patients with prostate cancer classified as stage T0, T1 or T2 and Gleason score less than or equal to seven. Patients were on average approximately 65 years of age. The 5-year prostate cancer-specific mortality rate was 4.6% and 2.6% for patients randomized to watchful waiting and radical prostatectomy, respectively. The 8-year prostate cancer-specific mortality rate was 13.6% and 7.1% for patients randomized to watchful waiting and radical prostatectomy, respectively. The difference in 5-year prostate cancer-specific mortality rates between watchful waiting and radical prostatectomy was not statistically

PAGE 20

9 significant. However, the difference in 8-year prostate cancer-specific mortality rates was statistically significant. Comparison of prostate cancer-specific survival using Cox proportional hazards regression models revealed an unadjusted hazard ratio of 0.5 (95% confidence interval: 0.27,0.91) comparing radical prostatectomy to watchful waiting and a hazard ratio 0.45 (95% confidence interval: 0.25, 0.84) when adjusted for tumor stage, Gleason score, and patient age. Radical prostatectomy versus radiotherapy Fowler et al. examined 5and 10-year cause specific survival among 357 stage A2 or B prostate cancer patients treated with either radical prostatectomy or radiotherapy (7). The study was non-randomized and the choice of treatment typically followed an algorithm in which patients with a life expectancy greater than ten years, no pelvic lymph metastases, and no contraindication to surgery received radical prostatectomy and all others received radiotherapy. The treatment algorithm resulted in patients treated with radical prostatectomy and radiotherapy being similar in terms of tumor grade and prostate-specific antigen level but with radiotherapy patients being 4.5 years older on average than patients receiving radical prostatectomy. Five-year prostate cancer-specific mortality rates were 5% and 11% for patients treated with radical prostatectomy and radiotherapy, respectively. Ten-year prostate cancer-specific mortality rates were 14% and 33% for radical prostatectomy and radiotherapy, respectively. The difference in prostate cancer-specific mortality for patients treated with radical prostatectomy or radiotherapy was not significant at five or ten years.

PAGE 21

10 Expectant management versus radical prostatectomy versus external beam radiotherapy Barry et al. retrospectively assessed prostate cancer-specific survival among 2311 patients treated between 1971 and 1984 with expectant management, radical prostatectomy, or external beam radiotherapy (21). The 10-year prostate cancer-specific survival rates among patients with tumors graded as being Gleason scores of two to four were 94%, 83%, and 94% for patients treated with expectant management, external beam radiotherapy, and radical prostatectomy, respectively. The 10-year prostate cancer-specific survival rates for patients with tumors graded as being Gleason scores of five to seven were 75%, 72%, and 88% for expectant management, external beam radiotherapy, and radical prostatectomy, respectively. The 10-year prostate cancer-specific survival rates for patients with tumors graded as being Gleason score of eight to ten were 45%, 43%, and 64% for expectant management, external beam radiotherapy, and radical prostatectomy, respectively. Barry et al. did not perform analyses to determine if the survival rates of the treatments studied were statistically significantly different because of the incomparable nature of the cohorts in terms of age and comorbidities. Barry et al. also note the fact that the study population was diagnosed prior to the use of prostate-specific antigen testing and advancements in radiotherapy techniques, such as 3-dimensional conformal radiation. Lu-Yao et al. used the Surveillance, Epidemiology, and End Results (SEER) database to estimate prostate cancer-specific mortality in patients treated with radical prostatectomy, radiotherapy, or watchful waiting (8). The study cohort included 59,876 men ranging from 50 to 70 years of age and diagnosed with prostate cancer between 1983 and 1992. For each treatment type, prostate cancer-specific mortality was estimated

PAGE 22

11 within each of three grades of prostate cancer. The grades were defined on the basis of Gleason score with grade one including Gleason scores ranging from two to four, grade two including Gleason scores ranging from five to seven, and grade three including Gleason scores ranging from eight to ten. As with any non-randomized study comparing radical prostatectomy to other treatment strategies, the study by Lu-Yao et al. was subject to selection bias and confounding (8). However, Lu-Yao and colleagues limit the impact of this bias by using a form of intent-to-treat analysis. More specifically, all patients treated with radical prostatectomy or undergoing a lymph node dissection were analyzed in the radical prostatectomy group regardless of whether the patient was found during prostatectomy or lymph node dissection to have disease that had spread beyond the prostate. Meanwhile, among patients undergoing radiation therapy, only those thought to have localized cancer were included. The reason for the difference is that among patients undergoing radiation therapy there is not the opportunity to detect the spread of disease as there is with patients undergoing radical prostatectomy and/or lymph node dissection. The decision as to whether the cancer was localized must be made at the same time for patients receiving the different treatments. Thus, the classification of the cancer as localized must be made at the time in which the treatment choice was made (i.e., before the information on whether the cancer has spread is obtained in the radical prostatectomy patients). The 10-year prostate cancer-specific mortality rates determined using the intent-to-treat approach were 17% for radical prostatectomy, 24% for radiotherapy, and 18% for watchful waiting (8). In interpreting these results one must be reticent of the difference in mean age among the groups. Patients receiving radical prostatectomy, radiotherapy,

PAGE 23

12 and watchful waiting were on average 65.8, 70.4, and 70.7 years of age. The impact of age on results is likely limited by using prostate cancer-specific mortality rather than overall mortality, however, there is still potential for difference in age to be partially responsible for the differential findings in prostate cancer-specific mortality. Lu-Yao et al. do not report whether the mortality differences were statistically different. An additional important conclusion of Lu-Yao et al. (8) was the need for outcomes specific to each grade of prostate cancer. This conclusion was founded upon their finding that patients with grade one prostate cancer generally did well while patients with grade three prostate cancer generally did not do well. Lu-Yao et al. (8, p. 909) state conclusions drawn from data pooled across cancer grades are therefore too optimistic for patients with grade three cancer and too pessimistic for patients with grade one disease. Summary of prostate cancer-specific mortality literature Of the four studies described in which treatment strategies were compared in terms of prostate cancer-specific mortality, one was randomized and three were non-randomized. In assessing non-randomized studies, it is always necessary to consider the potential for a selection bias. Within the prostate cancer arena there is a particular concern that in non-randomized studies radical prostatectomy will errantly be found superior to other treatments as a result of a selection bias (i.e., patients with a better prognosis will receive prostatectomy and patients with a worse prognosis will receive radiotherapy, expectant management, or some other less physically demanding type of treatment because the individual with worse prognosis is perceived to not be able to withstand the physical stress of prostate surgery)(21). In addition to the need for awareness of a selection bias favoring radical prostatectomy, awareness must be

PAGE 24

13 maintained for an information bias that could result in findings that errantly favor radical prostatectomy over other treatment strategies (8). As for an information bias, there is often differential staging between patients undergoing radical prostatectomy and radiotherapy or watchful waiting (8). Typically only patients receiving radical prostatectomy receive pelvic lymph node dissection, which can identify involved nodes (indicating a more advanced stage of disease) and leading to a change in the course of treatment (i.e., the patient does not undergo prostatectomy and instead receives a different type of treatment). Patients treated with modalities other than radical prostatectomy do not typically undergo pelvic lymph node dissection and thus nodal involvement can be missed and result in a patient being classified with a less advanced disease than is actually the case. The resultant effect of this bias is that when comparing radical prostatectomy to other treatments among patients with disease that is localized, radical prostatectomy can appear amore effective than other treatments because the other treatment groups are infiltrated with patients with non-localized disease. As for a selection bias, patients receiving radical prostatectomy are typically healthier than patients undergoing radiotherapy or a watchful waiting strategy in terms of both general health (e.g., younger) and severity of prostate cancer (e.g., less advanced tumor stage) (29,30). Furthermore, as a result of the differential staging bias previously discussed, some patients scheduled to undergo radical prostatectomy will not undergo radical prostatectomy because of lymph node involvement, which is indicative of a more severe stage of disease and requires a different treatment strategy. Meanwhile, patients receiving other treatment options with more advanced disease will not be identified and

PAGE 25

14 thus there will not be the opportunity to adjust their treatment strategies to strategies more appropriate for their more advanced disease state. The potential for such a bias was demonstrated by Lu-Yao and colleagues in which the treatment-received survival rate for patients treated with radical prostatectomy was significantly higher than that for the intent-to-treat survival rate for patients treated with radical prostatectomy (8). Prostate Cancer Economics Seven studies published since 1995 have examined treatment-specific prostate cancer cost. The majority of these studies have examined costs for either the initial treatment itself or first-year costs. One study modeled the impact of treatment failure over a 5-year period. Penson et al. (31) used the CaPSURE database to assess the first-year treatment costs of treating prostate cancer with various treatment strategies. The CaPSURE database contains longitudinal data on the healthcare resources used by prostate cancer patients treated at 26 sites across the United States. In calculating the first-year treatment costs, Penson et al. (31) multiplied governmental unit reimbursement rates by the number of resource units consumed and summated across types of resources to arrive at the total first-year costs of each type of treatment. According to Penson et al. (31), the cost of first-year treatment of prostate cancer using radical prostatectomy, radiotherapy, and watchful waiting in 1996 dollars was $7,320, $7,430, and $484, respectively. Penson et al. (31) also examined the first-year costs of each treatment by stage of disease. The cost of radical prostatectomy was found to differ significantly across tumor stage, but the differences in cost of treatment across tumor stages for other treatment strategies failed to attain statistical significance. Penson et al. (31) note that the failure to attain statistically significant differences in costs

PAGE 26

15 between differing severities of tumor stage could have been the result of small sample sizes as opposed to being truly equivalent. The cost of radical prostatectomy for patients with stage T1c, T2a/b, and T2c were $6,881, $7,216, and $8,027, respectively. Brandeis et al. (29) examined charges associated with prostate cancer by type of treatment among patients at least 65 years of age using Medicare charge data made available by the Health Care Financing Administration (HCFA). Medicare charges included in the analysis included those for work-up, treatment, and complications over a 1-year period. A total of 10,107 patients diagnosed with prostate cancer between 1993 and 1996 and included in the 5% random sample of Medicare beneficiaries made available by HCFA were included in the analysis. Mean first-year charges for radical prostatectomy, external beam radiotherapy, brachytherapy, brachytherapy plus external beam radiotherapy, and radical prostatectomy plus external beam radiotherapy were $19,019, $15,937, $15,301, $24,407, and $31,329, respectively (29). The only charges not statistically significantly different were brachytherapy versus external beam radiotherapy. Not surprisingly, patients treated with combination therapy (i.e., brachytherapy plus external beam radiotherapy or radical prostatectomy plus external radiotherapy), had the highest mean charges. Burkhardt and colleagues (30) used the SEER-Medicare linked database to assess differences in costs associated with treatment of early-stage prostate cancer with either radical prostatectomy or external beam radiotherapy among Medicare beneficiaries enrolled in both Medicare Part A and Part B. Costs were assessed for a period of time beginning one month prior to diagnosis and ending nine months after diagnosis. The study included 4,956 radiotherapy patients in both the treatment-received and intent-to

PAGE 27

16 treat analyses, 4,847 radical prostatectomy patients in the treatment-received analysis, and 5,299 radical prostatectomy patients in the intent-to-treat analysis. A variety of adjustments were used to ensure patients in the radiotherapy and radical prostatectomy cohorts were similar in terms of age and comorbidities (30). Regardless of comorbidity adjustment mechanism, patients treated with radical prostatectomy were found to have greater costs over the 10-month period beginning one month prior to diagnosis than patients treated with external beam radiotherapy. Using a comorbidity adjustment strategy in which a patients average monthly Medicare expenditures were subtracted from each month in the 10-month cost assessment period, the cost of receiving either radical prostatectomy or external beam radiotherapy over a 10-month period were $15,082 and $11,377, respectively, using a treatment-received analysis. The intent-to-treat analysis resulted in the cost associated with receiving radical prostatectomy increasing to $15,205. Regardless of type of analysis conducted treatment-received or intent-to-treat and whether an adjustment was made for comorbidities or not, Medicare expenditures over a 10-month period were significantly higher for prostate cancer patients treated with radical prostatectomy than those treated with external beam radiotherapy. Perez et al. (32) examined reimbursement for treating patients with stage T1c or T2 prostate cancer using 3-dimensional conformal radiation therapy (3D-CRT), standard radiation therapy (SRT), and radical prostatectomy at a single institution. The average reimbursement for 3D-CRT, SRT, and radical prostatectomy during 1994 was $13,823, $10,864, and $12,250, respectively.

PAGE 28

17 Using a combination of institution-specific and literature-based efficacy data, Perez et al. (32) went a step further in assessing the cost of each treatment modality by incorporating the cost of five years of hormonal treatment consisting of a luteinizing hormone releasing hormone agonist and flutamide, a non-steroidal antiandrogen, for patients failing initial treatment. Assuming a 5% failure rate for patients treated 3D-CRT, and 20% for patients treated with SRT, and 10% for patients treated with radical prostatectomy, the cost of 3D-CRT, SRT, and radical prostatectomy increase to $15,173, $16,264, and $16,405, respectively. Wagner and colleagues (33) examined the hospital charges associated with treating patients with clinically localized prostate cancer with either radical prostatectomy or transperineal brachytherapy during 1996 and 1997. Among the 16 patients undergoing radical prostatectomy, the average total hospital charges were $15,097, and among the 19 patients undergoing brachytherpy, the average total hospital charges were $21,025. The resultant $5,928 difference in the cost of hospital-related charges favoring radical prostatectomy was statistically significant. In addition to comparing overall charges associated with radical prostatectomy and brachytherapy, Wagner et al. (33) compared the two treatments on a number of charge categories. The operative charges (i.e., anesthesia, operating room time, and operating room supplies) were found to be greater in patients undergoing radical prostatectomy than in those undergoing brachytherapy. Furthermore, the pharmacy and length of stay charges were significantly greater in radical prostatectomy patients. However, the charges for the radioactive seeds used in brachytherapy ($7,033) overcame the savings associated with brachytherapy in operative, pharmacy, and length of stay charges.

PAGE 29

18 Ciezki et al. (34) compared radical prostatectomy and brachytherapy in terms of perioperative costs. In assessing the perioperative costs among 404 patients treated with radical prostatectomy and 179 patients treated with brachytherapy between January 1, 1997 and October 30, 1998, Ciezki and colleagues (34) found brachytherapy to cost 85% to 105% more than radical prostatectomy. Similar to Wagner et al. (33), Ciezki et al. (34) found the operative, pharmacy, and length of stay related charges to be significantly lower for brachytherapy than for radical prostatectomy. Yet, just as in the study by Wagner et al. (33), any savings in operative, pharmacy, and length of stay costs were overcome by the considerable costs of the radioactive seeds used in brachtherapy. Kohan et al. (35) also compared the costs of radical prostatectomy and brachytherapy. Costs, including preoperative, operative, and postoperative costs, over a 1-year period were compared between 38 men with clinically localized prostate cancer undergoing radical prostatectomy and 22 men with clinically localized prostate cancer undergoing brachytherapy. All patients were treated at a single institution. Radical prostatectomy patients were younger (71.1 versus 61.2 years of age) and had less advanced tumors (52.6% stage T2 versus 68.1% stage T2). Furthermore, patients undergoing radical prostatectomy were more often treated with neoadjuvant hormonal therapy (45.5% versus 13.2%) that those undergoing radical prostatectomy. Kohan and colleagues (35) found patients undergoing radical prostatectomy to have significantly greater operative charges ($11,352.59 versus $9,818.17) and significantly lower postoperative charges ($1,007.20 versus $2,285.20) than brachytherapy patients. However, preoperative and total charges were not found to be significantly different.

PAGE 30

19 Total charges for radical prostatectomy and brachytherapy patients were $13,904.60 and $13,886.00, respectively. The finding of no significant difference in overall charges between radical prostatectomy and brachytherapy by Kohan et al. (35) stands in contrast to the findings of Wagner et al. (33) and Czieki et al. (34) in which brachytherapy was found to be more expensive than radical prostatectomy. Kohan et al. (35) note that the discrepancy could be due to the fact that reimbursement in their institution was based on a flat rate specific to a procedure. Grover and colleagues (36,37) developed a Markov model to forecast the clinical and economic burden of prostate cancer among Canadians. Data from the literature and population-based registries were used to estimate the efficacy and complications of prostate cancer for different tumor stages, histological grades, and treatment strategies (36). Canadian government fee schedules were used to apply costs to the resources used, and thus estimate the economic burden of prostate cancer among Canadian men (37). Unlike many of the other economic studies of prostate cancer, Grover et al. (37) estimated lifetime costs rather than costs for either just initial costs or just terminal costs. Unfortunately, Grover and colleagues (37) provide results only on an aggregate level; finding the total cost of prostate cancer among the population of Canadian men between ages 40 and 80 without prostate cancer in 1997 will cost $9.76 billion (in 1996 Canadian dollars) over the lifetime of these men. Burkhardt et al. (30, p. 2874) noted that their study [did] not include costs beyond the initial treatment period, and these would be expected to differ if one treatment approach generates more long-term side effects than the other. Given that the vast

PAGE 31

20 majority of studies examining treatment-specific costs have been limited to the initial treatment period, there is a need for further study of treatment-specific costs over a longer period of follow-up. Beyond the impact of differing side-effect profiles on long-term costs, there is a need to consider the long-term costs of treatment failure. The consideration of long-term costs is also necessary in any comparison of treatment-specific costs in which watchful waiting is a potential treatment strategy. Our study will provide a substantial contribution to the existing literature by providing information on the long-term treatment-specific costs of prostate cancer. Guidelines Clinical practice guidelines are defined by the Institute of Medicine (38, p. 38) as systematically developed statements to assist practitioner and patient decisions about appropriate health care for specific clinical circumstances. Practice guidelines are developed with the intent of improving patients health outcomes (39-42), and have been shown repeatedly to improve health outcomes (42-44). Based on the mass of new medical literature each week, it is impractical for any one clinician alone to assess all relevant literature for treating a condition. Thus, reviews summarizing evidence are necessary for a clinician to stay abreast of the relevant medical literature. Systematic reviews, such as Cochrane reviews, provide such summaries of the medical literature (45,46). Clinical guidelines go one step farther, providing recommendations for care based on the evidence summarized in systematic reviews (40,45,47,48). According to Cape and Barkham (41, p. 290), guidelines set out the specific clinical processes that will lead to optimal outcomes for the specific circumstances.

PAGE 32

21 The production of practice guidelines surged throughout the decade of the 1990s, as evidenced by a PubMed search revealing more than 4,500 publications classified as practice guidelines indexed in the PubMed literature database by the end of 1999. Guidelines are produced by a number of entities, including medical societies (general and specialty) and governments agencies (40,49,50). Guidelines are produced at local, regional, national, and international levels (49). Development of Guidelines The development of clinical practice guidelines is a significant undertaking in terms of both time and money (41,51-53); one report showed development costs of $200,000 per guideline in the United States (54). Historically, guidelines were developed based on expert opinion (48,54); however guideline development has evolved into a process that is based on systematic reviews of evidence reported in the medical literature (41,48,54). Guidelines typically include practice recommendations based on both scientific evidence and expert opinion when there are not formal studies available (47,55). Initially, a panel must be established to formulate the guidelines. The panel should be diverse in nature and representative of all parties the guidelines will impact, including users (e.g., physicians), patients, and experts in the clinical field (38,39,41,43,45,47,48,51,52,54). Additionally, guideline panels may include medical librarians, epidemiologists, statisticians, psychologists, communication specialists, health economists, and other experts in relevant methodologies (54). Perhaps the most important step in developing guidelines is to assess the evidence on which practice recommendations will be made. Assessment of evidence is the base on which the guideline will be built (38). As such, the development of clinical practice

PAGE 33

22 guidelines necessitates the performance of a systematic review of the evidence (43,51). Guidelines should include an overview of the evidence review, including how the evidence was located (e.g., PubMed search strategy) (38,39) and a description of inclusion criteria (i.e., study type) required for a study to be considered in the review of evidence (38,45,48). An integral step in developing guidelines is formulating and reaching consensus on the practice recommendations to be included in the guideline (43,47,52,54). Efforts to reach consensus can be formal or informal (47,54), with informal consensus more likely to result in practice recommendations that are not based on scientific evidence (47). Formal consensus can be reached through a variety of techniques, such as a nominal group technique or a Delphi panel (40,47). According to the Appraisal of Guidelines, Research, and Evaluation in Europe (AGREE) Collaborative Group Practice (49, p. 1040), recommendations should be explicitly linked to the evidence on which they are based (38,47,54), and each recommendation should include a grade indicating the strength of the evidence (39,43,45,47,52,54). For example, a recommendation based on multiple randomized clinical trials would have a higher grade than a recommendation based purely on expert opinion. Before disseminating guidelines, an external review process is undertaken (38,43,47,50,54). Shekelle et al. (52, p. 596) state that external review of guidelines is used to ensure content validity, clarity, and applicability. Ideally, the development of a guideline would also include a pre-testing phase (38). If a guideline is to remain valid over time, the guidelines must be updated as new evidence is discovered (51). Those in the guideline-development process should either determine when the guidelines will be

PAGE 34

23 updated or determine conditions necessitating the update of the guidelines (e.g., a well-designed controlled trial reveals evidence contradictory to the guideline recommendations) (38,43,53). Quality of Guidelines As stated previously, the development of guidelines increased rapidly throughout the 1990s (42). So many guidelines were produced, that improving and assessing the quality of guidelines has become challenging (42). Agencies such as the Agency for Healthcare Quality and Research (AHRQ) in the United States and the National Health System in the United Kingdom are funding efforts to improve the quality of guideline development (56). The Institute of Medicine (IOM) (38) developed a list of eight characteristics a clinical practice guideline would ideally meet: validity, reliability/reproducibility, clinical applicability, clinical flexibility, clarity, multidisciplinary process, scheduled review, and documentation (38). These characteristics are self-explanatory with the possible exceptions of validity and reliability/reproducibility. The IOM (38, p. 58) states that valid clinical practice guidelines lead to the health and cost outcomes projected for them. Valid guidelines should include both statements on the strength of evidence used in the formulation of recommendations and an estimate of the impact of these recommendations on both health and cost outcomes (2). According to the IOM, reliable guidelines would result in the same care for a patient by two separate practitioners, and a separate guideline-development panel using the same development strategy and scientific evidence would replicate those guidelines (2). Numerous studies have developed instruments to appraise the quality of a clinical practice guideline (2,39,40,42,57). The appraisal instruments range from a simple

PAGE 35

24 checklist of three items (39) to a 46-item instrument developed by an Institute of Medicine committee (2). The Grilli 3-item checklist assesses whether a description was provided of the individuals developing the guideline, whether the search for evidence was described, and whether guideline recommendations were graded based on the level of supporting evidence (39). More comprehensive in nature, the IOM appraisal instrument has items explicitly assessing the degree to which seven of the IOMs eight desirable characteristics of a practice guideline are fulfilled (2). The eighth desirable characteristic, documentation, is implicitly assessed through many of the 46 questions. National Comprehensive Cancer Network Prostate Cancer Guideline The National Comprehensive Cancer Network (NCCN) is an alliance of 19 cancer centers, which constructs cancer treatment guidelines (58). The NCCN released its most recent version of treatment guidelines for prostate cancer in January 2004 (4). The goal of the guideline is to extend life expectancy while minimizing treatment-related morbidity (58, p. 15). The NCCN prostate cancer clinical practice guideline appears to meet many of the characteristics desired in a guideline. The process used in the development of the guideline included clinical experts in the fields of medical and radiation oncology as well as urology; an extensive review and modification procedure was employed; recommendations were graded; evidence supporting the recommendation was cited; and an annual review process was planned and has been faithfully followed. One shortcoming of the guideline is its failure to outline the methods used to identify the literature that is used to support recommendations. The failure to outline the search process prevents the user from being able to judge whether the search for evidence was systematic and thorough.

PAGE 36

25 The guideline makes recommendations for prostate cancer treatment on the basis of risk of recurrence and life expectancy (4). Risk of recurrence is classified as a function of the prostate specific antigen level, 2002 American Joint Committee on Cancer (AJCC) Tumor Node Metastasis (TNM) stage, and Gleason score (58). Using the PSA value, clinical stage, and Gleason score, prostate cancer patients are classified as being at low, intermediate, high, or very high risk of recurrence (4). Patients with any positive nodes or any metastases are classified separately from patients without positive nodes or metastases and are not placed in a risk of recurrence group(4). Patients at low risk are those with a PSA less than 10 nm/mL, a stage T1 to T2a tumor, or a Gleason score of 2 to 6. Those at intermediate risk are those with a PSA of 10 to 20 ng/mL, a stage T2b to T2c tumor, or a Gleason score of 7. Those at high risk are those with a PSA greater than 20 ng/mL, a stage T3a tumor, or a Gleason score of 8 to 10. Those at very high risk are those with a stage T3b to T4 tumor. Treatment strategy recommendations for prostate cancer patients at low or intermediate risk of recurrence are further classified based on expected survival time (4). For patients at low risk of recurrence with a life expectancy less than ten years, the recommended treatment strategies are: expectant management; 3-dimensional conformal radiotherapy (3D-CRT); or brachytherapy. For patients at low risk of recurrence with a life expectancy greater than ten years, the recommended treatment strategies are: expectant management; 3D-CRT; brachytherapy; radical prostatectomy with pelvic lymph node dissection; or radical prostatectomy without pelvic lymph node dissection. For patients at intermediate risk of recurrence with a life expectancy less than ten years, the recommended treatment strategies are: expectant management; 3D-CRT; 3D-CRT

PAGE 37

26 with brachytherapy; radical prostatectomy with pelvic lymph node dissection; or radical prostatectomy without pelvic lymph node dissection. For patients at intermediate risk of recurrence with a life expectancy greater than ten years, the recommended treatment strategies are: 3D-CRT; 3D-CRT with brachytherapy; radical prostatectomy with pelvic lymph node dissection; or radical prostatectomy without pelvic lymph node dissection. The NCCN prostate cancer clinical practice guideline was used in our study to identify treatments recommended for patients with localized prostate cancer to be compared in terms of clinical effectiveness and cost. Cost Estimation One objective of our study was to estimate the difference in cost of treatment strategies for patients with localized prostate cancer (i.e. to compare treatment strategies recommended for patients at low or intermediate risk of prostate cancer recurrence according to the NCCN prostate cancer clinical practice guideline). In order for our findings regarding differences in cost to be valid, the methods used to estimate cost must be sound. Just as with effectiveness data, stochastic cost data collected as part of a randomized controlled trial or an observational study is often censored (59,60). As such, appropriate methods for estimating cost in the presence of censoring must be used. Several methods have been suggested for the estimation of costs in the presence of censoring. Three methods previously used to estimate costs in the presence of censoring have now been determined to be invalid. The first of these methods was the simple mean of observed costs (60-62). The mean of observed costs does not incorporate costs occurring after censoring takes place and is thus biased in a downward manner. The second method used was the average costs among only patients who were not censored (60,61). This measure is also biased because it is representative of patients with shorter

PAGE 38

27 survival times rather than the entire population of interest because patients with longer survival times are censored and not included in the estimate. The third method of estimating costs in the presence of censoring used standard survival analysis, replacing time to event with cost to event as the outcome being measured (60,61). The use of standard survival analysis methods to estimate costs in the presence of censoring was problematic because of a violation of the non-informative censoring assumption (59,60,63-65). The violation of the non-informative censoring assumption occurs because of a lack of independence between costs at the time of censoring and costs at the time of death (60,66). Lin (59, p. 36) notes the assumption is violated even if the underlying censoring mechanism is purely random because of the inherent patient heterogeneity with respect to cost accumulation entails that the cumulative cost at the censoring time is positively correlated with the cumulative costs at the endpoint of interest. In other words, there is a unique way in which each individual accumulates cost that results in costs at any time x being correlated with the costs at any time y for a given patient. And, as a result there is a correlation between a given individuals costs at the time of censoring and at the time of death if the patient had not been censored. Thus, there is a violation of the assumption of non-informative censoring. Fortunately, several methods have been developed to estimate costs in the presence of censoring that appear to work well. Two of the most commonly used methods are the Lin estimator and the Bang and Tsiatis estimator (62). The Lin estimator, also referred to as the direct method, divides the study period into k intervals, estimates the mean cost for patients alive at the start of the interval, and sums across the intervals the average cost of

PAGE 39

28 each interval multiplied by the Kaplan-Meier estimator of surviving until the beginning of the interval (60). The Bang and Tsiatis estimator is calculated using inverse weighting (65). The estimator is similar to the Lin estimator in that it divides the total time period into smaller intervals and estimates the cost within each interval. However, for patients dying within the interval, the cost is weighted by the inverse of the probability of not being censored at the time of death (64). Furthermore, if the patient survives until the end of the interval, the cost is weighted by the inverse of the probability of not being censored at the end of the interval. The direct method (Lins estimator) and the inverse-weighting method (Bang and Tsiatiss estimator) have been reported to be similar in terms of both variance and bias (64). Willan et al. (64) advocate use of the inverse-weighting method over the direct method because the direct method is more restrictive in terms of when censoring can take place. Similarly, provided that the intervals are relatively small, OHagan and Stevens (62) also recommend the use of the inverse-weighting method due to its consistency regardless of when censoring occurs. Equations 2-3 and 2-4, as reported by Willan et al. (64) estimate cost, vj, in the presence of censoring using inverse-weighting and the variance, V(vj), of the cost estimate, respectively. KkjkjCv1 (2-3) with, jjnijkijjkijkinijkijjkijkXGCYXGYC1*11*)()( and }{1kjijijijkiaXIY

PAGE 40

29 jniKkKlcjlicjkijZZvV111)()()( (2-4) with, jnljljkljijljljkijijkijjkjkijkijcjkiRBXXIBXGCCYnZ1*)(}{()(1 , and jnljijljiXXIR1}{ jnljkljjkjkljkljijkljijkiXGCCYXXIRB1**)()(}{1 In cases where there is a need to adjust cost for one or more covariates, cost can be assessed using regression techniques that also incorporate inverse probability weighting to account for censoring. Cost is the dependent variable and independent variables include the treatment variable and any other covariates that might impact cost. Equations 2-5 and 2-6, as reported by Willan et al. (67), are used to estimate the coefficients and variance-covariance matrix for the independent variables. In Equations 2-5 and 2-6, Zi is a p by 1 vector of individual is values for the p independent variables and is a 1 by p vector of the coefficients for the p independent variables in the kth time interval. k inikikikiiinikikikZXGCZZXG1**11** (2-5) n = number of individuals k = time-period (e.g., month or year) = 0 if individual i was censored during year k, otherwise = 1 *,ik *,ik Xi = min(time of event, censoring time) = min(Xi, k + 1) *,ikX = Kaplan-Meier estimate of probability of not being censored at *,ikXG *,ikX = observed cost for individual i during time-period k ikC,

PAGE 41

30 11111111111niiiniKkKmmikiniiiZZnnZZnnV (2-6) njkjjijjkiiikiikkikiFRXXIFZXGZCi1**}{ n = number of individuals k = time-period is from Equation 2-5 k = 0 if individual i was censored during time-period k, otherwise = 1 *,ik *,ik Xi = min(time of event, censoring time) for individual i Xj = min(time of event, censoring time) for individual j = min(Xi, k + 1) *,ikX = Kaplan-Meier estimate of probability of not being censored at *,ikXG *,ikX K1ki,kiCC i = 1, if patient i experienced the event of interest, otherwise i = 1 i = 1 i jnjkjjkkjkjikjikiZXGZCXXIRF 1***}{1 n = number of individuals k = time period is from Equation 2-6 k = 0 if individual j was censored during time-period k, otherwise = 1 *j,k *j,k Xi = min(time of event, censoring time) for individual I Xj = min(time of event, censoring time) for individual j = min(Xj, k + 1) *j,kX = Kaplan-Meier estimate of probability of not being censored at *j,kXG *j,kX = observed cost for individual j during time-period k j,kC j = 1, if patient j experienced event of interest, otherwise j = 1 j = 1 j

PAGE 42

31 njijiXXIR1}{ n = number of individuals Xi = min(time of event, censoring time) for individual i Xj = min(time of event, censoring time) for individual j Our study required the use of the regression techniques incorporating inverse weighting because it was not randomized (multiple confounding variables were identified) and the cost data was censored, and to the best of our knowledge is the first to estimate the cost of prostate cancer-specific healthcare using this technique. Cost-effectiveness Beyond summarizing differences between treatments in clinical effectiveness and cost, it can be useful to have a metric that jointly compares both differences in clinical effectiveness and cost. Cost-effectiveness is such a metric and is defined by Willan (68, p. 228) as a quantitative method for comparing a new medical treatment to a relevant alternative (e.g., standard care) in terms of both costs and effectiveness. Using the costs (in monetary units) and effects (in some clinical unit) of each treatment, comparisons are made to determine if one treatment dominates the other (69). Dominance occurs when one treatment is both more effective and less costly than the other (69,70). In cases where neither treatment dominates, an incremental cost-effectiveness ratio is calculated to compare the treatments in terms of cost-effectiveness (69). The incremental cost-effectiveness ratio is calculated as the ratio of the difference in the cost of the two interventions and the difference in the effectiveness of the two interventions (68,71). The calculated cost-effectiveness ratio is an estimate of the additional cost per unit of

PAGE 43

32 effectiveness gained associated with using the more costly, more effective treatment rather than the less costly, less effective treatment (69). Our study will not make any formal conclusions as to whether one treatment is cost-effective compared to another. Instead, for treatment comparisons identifying one treatment as both more effective and more costly, we will calculate and report the amount one must be willing to pay an additional unit of effectiveness for the more effective, more costly treatment to be cost-effective and will leave the decision as to whether the more effective, more costly treatment is cost-effective to the reader.

PAGE 44

CHAPTER 3 METHODS The methods of our study are presented in four parts. First, the methods used to establish cohorts for the comparison of treatment strategies recommended in the National Comprehensive Cancer Network (NCCN) prostate cancer clinical practice guideline for patients at low or intermediate risk of prostate cancer recurrence (i.e., patients with localized disease) are described. Second, the procedures used to assess and compare clinical effectiveness, in terms of prostate cancer-specific survival, of the strategies recommended for patients at low or intermediate risk of prostate cancer recurrence are presented. Third, the methods used to assess and compare the cost of prostate cancer-specific healthcare use the recommended treatment strategies are described. Finally, a description of the method used to assess the amount one must be willing to pay for a treatment that is both more effective and more costly to be cost-effective is presented. Data Source The source of data for our study was the linked Surveillance, Epidemiology, and End Results (SEER)-Medicare database. The SEER portion of the SEER-Medicare linked database is a compilation of data collected by population-based cancer registries throughout the United States. The SEER registries collect demographic, clinical stage, histological grade, treatment, and outcomes data on all cancer diagnoses in a geographic region. The National Cancer Institute (NCI) administers the SEER program, and each registry is subjected to data audits to ensure the reliability and validity of the data (72). The NCI, in conjunction with the Centers for Medicare and Medicaid Services (CMS), 33

PAGE 45

34 links the SEER cancer data with Medicare claims data for patients enrolled in Medicare. The linking of cancer registry data with Medicare claims data makes the SEER-Medicare database an ideal data source for studying the clinical and economic outcomes of cancer among an elderly population. Establishment of Comparison Cohorts Patients included in the study were: at low or intermediate risk of prostate cancer recurrence (defined below); managed with an initial treatment strategy recommended in the NCCN prostate cancer clinical practice guideline; diagnosed between 1992 and 1999; continuously eligible for Medicare (Parts A and B) from the time of diagnosis until either death or the end of the study period (i.e., December 31, 2002); and not enrolled in a Medicare health maintenance organization. After identifying patients meeting the inclusion criteria, the patients were divided into cohorts defined on the basis of the initial treatment strategy received. The initial treatment strategies compared were expectant management, external beam radiotherapy, brachytherapy, prostatectomy, and external beam radiotherapy with brachytherapy. Risk of Recurrence The National Comprehensive Cancer Network categorization of prostate cancer patients into risk of prostate cancer recurrence groups is based upon biochemical, clinical, and histological criteria (4). The biochemical, clinical, and histological criteria used are the prostate specific antigen (PSA) level, the 2002 American Joint Committee on Cancer (AJCC) Tumor, Node, and Metastasis (TNM) Staging System for Prostate Cancer classification, and the Gleason score. Patients at low risk are those with a PSA less than 10 ng/mL, a stage T1 to T2a tumor, or a Gleason score of 2 to 6. Those at intermediate

PAGE 46

35 risk are those with a PSA of 10 to 20 ng/mL, a stage T2b to T2c tumor, or a Gleason score of 7. The SEER database was used to classify patients risk of recurrence. The SEER database does not contain data on PSA values for patients diagnosed with prostate cancer prior to 1998. Thus, our study only used tumor stage and grade in classifying a patients risk of recurrence due to constraints imposed by the available data. The clinical extension component of the extent of disease variable in the SEER database provided the patients TNM stage. The grade component of the morphology variable in the SEER database provided a proxy for the patients Gleason score. Within the SEER database, the grade component of the morphology variable is classified as one for patients with a Gleason score of two, three, or four. If the Gleason score is five, six, or seven, the grade component of the morphology variable is classified as two. If the Gleason score is eight, nine, or ten, the grade component of the morphology variable is classified as three. As noted above, patients at low or intermediate risk of recurrence were combined for the comparison of treatments rather than comparing treatments separately for those at low risk of recurrence and those at intermediate risk of recurrence. The decision to group those at low and intermediate recurrence risk together before making treatment comparisons was made because the SEER database does not provide the actual Gleason score and the grade component variable used in our study does not allow for discrimination between the Gleason scores that would place an individual at low recurrence risk (i.e., 2 to 6) or intermediate recurrence risk (i.e., 7). An additional advantage of combining the low and intermediate recurrence risk patients was the

PAGE 47

36 increase in the power of the comparisons to find a difference between compared treatments because of the increased sample size. Initial Treatment The initial treatment strategy received by each prostate cancer (ICD-9-CM code 185.xx) patient was determined using both the treatment information provided in the SEER and Medicare claims databases. The treatment information provided in the SEER database is any treatment received within four months of diagnosis. Physician's Current Procedural Terminology (CPT), Health Care Financing Administration Common Procedure Coding System (HCPCS), and ICD9-CM procedure codes (Tables 3-1 to 3-6) on claims submitted to Medicare for services provided during the period beginning one month prior to and ending four months after prostate cancer diagnosis were also used in determining the initial treatment. The decision rules for determining the initial treatment strategy received were as follows: a patient was deemed to have received prostatectomy if the SEER data indicated the patient had a prostatectomy AND the patient did not have a claim during the period beginning one month prior to and ending four months after prostate cancer diagnosis with: A) a CPT/HCPCS code (Table 3-1) or ICD9-CM procedure code (92.2x: Therapeutic radiology and nuclear medicine) for radiotherapy (external beam or brachytherapy); B) a CPT/HCPCS code (Table 3-2) or ICD9-CM procedure code (Table 3-3) for androgen deprivation (surgical or chemical); or, C) a CPT/HCPCS code (Table 3-4) for any other type of prostate cancer treatment a patient was deemed to have received external beam radiation if the SEER data indicated the patient received external beam radiotherapy AND the patient did not have a claim during the period beginning one month prior to and ending four months after prostate cancer diagnosis with: A) a CPT/HCPCS code (Table 3-5) or ICD9-CM procedure code (Table 3-6) for prostatectomy; B) a CPT/HCPCS code (Table 3-2) or ICD9-CM procedure code (Table 3-3) for androgen deprivation (surgical or chemical); or, C) a CPT/HCPCS code (Table 3-4) for any other type of prostate cancer treatment a patient was deemed to have received brachytherapy if the SEER data indicated the patient received brachytherapy AND the patient did not have a claim during the period beginning one month prior to and ending four months after prostate cancer

PAGE 48

37 diagnosis with: A) a CPT/HCPCS code (Table 3-5) or ICD9-CM procedure code (Table 3-6) for prostatectomy; B) a CPT/HCPCS code (Table 3-2) or ICD9-CM procedure code (Table 3-3) for androgen deprivation (surgical or chemical); or, C) a CPT/HCPCS code (Table 3-4) for any other type of prostate cancer treatment a patient was deemed to have received external beam radiation with brachytherapy if the SEER data indicated the patient received external beam radiotherapy with brachytherapy AND the patient did not have a claim during the period beginning one month prior to and ending four months after prostate cancer diagnosis with: A) a CPT/HCPCS code (Table 3-5) or ICD9-CM procedure code (Table 3-6) for prostatectomy; B) a CPT/HCPCS code (Table 3-2) or ICD9-CM procedure code (Table 3-3) for androgen deprivation (surgical or chemical); or, C) a CPT/HCPCS code (Table 3-4) for any other type of prostate cancer treatment a patient was deemed to have received expectant management if the SEER data indicated the patient did not receive prostatectomy, external beam radiation, or brachytherapy AND the patient did not have a claim during the period beginning one month prior to and ending four months after prostate cancer diagnosis with: A) a CPT/HCPCS code (Table 3-1) or ICD9-CM procedure code (92.2x: Therapeutic radiology and nuclear medicine) for radiotherapy (external beam or brachytherapy); B) a CPT/HCPCS code (Table 3-5) or ICD9-CM procedure code (Table 3-6) for prostatectomy; C) a CPT/HCPCS code (Table 3-2) or ICD9-CM procedure code (Table 3-3) for androgen deprivation (surgical or chemical); or, D) a CPT/HCPCS code (Table 3-4) for any other type of prostate cancer treatment While the NCCN prostate cancer clinical practice guideline calls for the use of 3-dimensional conformal radiotherapy rather than 2-dimensional radiotherapy, it was not possible based on the information included in the SEER database to distinguish between the two variations of external beam radiotherapy. Furthermore, while procedure codes used to bill Medicare distinguish between 2and 3-dimensional radiotherapy, examination of the data led to a lack of confidence in the reliability of the coding to distinguish between 2and 3-dimensional radiotherapy. Thus, 2and 3-dimensional beam radiotherapy were grouped together as external beam radiotherapy in all analyses. Comparisons of prostate cancer-specific survival and cost of prostate cancer-specific healthcare use were based only on the initial treatment strategy (i.e., treatment received in the first four months following diagnosis). Treatment received after the first

PAGE 49

38 four months did not impact the initial treatment strategy groupings used for comparing prostate cancer-specific survival and cost of prostate cancer-specific healthcare use. Propensity Scores As with any study without randomization, our study was subject to selection bias and could have been influenced by multiple observed and non-observed confounders. Our study used a multivariate analysis with an individuals propensity score (explained below) included as a covariate to control for the effect of variables believed to be confounders. The propensity score is the probability of a patient receiving treatment x rather than treatment y given the patients observed values on the characteristics thought to be potential confounders (73). The propensity score was estimated using multiple logistic regression with the dependent variable being treatment received and the independent variables being hypothesized confounding variables. The hypothesized confounders used in the logistic regression as independent variables were: SEER registry (an indicator of geographic area), age at diagnosis, race, year of prostate cancer diagnosis, tumor stage, tumor grade, urban/rural classification, and Charlson comorbidity score. A confounding variable is a variable related to both the treatment received and the outcome of interest. Each of the variables believed to be confounders included in the logistic regression estimating the propensity score were related to both the treatment received and prostate cancer-specific survival, except for the Charlson comorbidity score (see Chapter 4). Following estimation of the model, each individual patients propensity score was calculated. A separate propensity score model was developed for each pair of treatments compared: expectant management vs. external beam radiotherapy expectant management vs. brachytherapy

PAGE 50

39 expectant management vs. prostatectomy expectant management vs. external beam radiotherapy with brachytherapy external beam radiotherapy vs. brachytherapy external beam radiotherapy vs. prostatectomy external beam radiotherapy vs. external beam radiotherapy with brachytherapy brachytherapy vs. prostatectomy brachytherapy vs. external beam radiotherapy with brachytherapy prostatectomy vs. external beam radiotherapy with brachytherapy The propensity score calculated for a comparison of two treatments can be used to control for differences between groups in three ways: matching; stratification; and inclusion as a covariate in a regression model. In order to retain maximum power to detect a difference in prostate cancer-specific survival and cost of prostate cancer-specific healthcare use between treatment strategies, the propensity score was used as a covariate in the regression models assessing the relation between treatment and outcome because inclusion of the propensity score in a multivariate model does not decrease sample size, which occurs with both matching and stratification. Clinical Effectiveness The clinical effectiveness of each treatment strategy was defined by the estimated survival function of the treatment strategy at 5-years (a standard time frame for examining cancer survival). The methods of Kaplan and Meier (74) were used to estimate each treatment-specific survival function. Equation 3-1 is the Kaplan-Meier estimate of the survival curve with dj being the number of deaths due to prostate cancer at time j and nj being the number of individuals at risk of death at time j. kjjjndtS1)/exp()( (3-1) In estimating the survival function, patients with prostate cancer listed as the cause of death in the SEER database were treated as events, and patients dying due to

PAGE 51

40 causes other than prostate cancer or alive at the end of 5-year follow-up period were treated as censored observations. Patients dying due to prostate cancer were identified by the cause of death listed in the SEER database, which uses state death certificates to classify a patients cause of death. (75). The SEER database classifies patients with ICD-9 code 185 or ICD-O-2 code C619 on the death certificate as dying due to prostate cancer. The effect of the initial treatment strategy received on prostate cancer-specific survival was determined using the Cox proportional hazards model. Equation 3-2 presents the Cox proportional hazards model used in comparing each pair of treatments. In Equation 3-2, trti is a dummy variable indicating individual is initial treatment strategy and psi is individual is calculated propensity score. )()**exp()(0thpstrtthiPSitrti (3-2) The Wald statistic was used to test whether a statistically significant difference in the hazard of death due to prostate cancer existed between the treatments being compared controlling for the variables included in the propensity score (i.e., SEER registry, urban/rural classification, race, tumor stage, tumor grade, year of diagnosis, age at diagnosis, and Charlson comorbidity score). The a priori significance level was set at 0.05. Cost of Treatment The cost of interest was that of prostate cancer-specific healthcare use over the 5-year period beginning with prostate cancer diagnosis. Several key tasks had to be accomplished in estimating this cost. First, the amount of all healthcare costs incurred by prostate cancer patients that are attributable to prostate cancer was determined. In other

PAGE 52

41 words, prostate cancer-related and non-prostate cancer-related costs were separated. Second, in order for the testing of differences in costs between two treatment strategies to be fair, the groups had to similar in terms of factors that might influence the accrual of costs. Third, the estimation of costs had to take into account the presence of censoring in the cohort of prostate cancer patients being used to estimate the cost of prostate cancer-specific healthcare use. Initially, the methods proposed by Brown and colleagues (76), in which total healthcare costs are determined for both cancer patients and matched controls were used to estimate prostate cancer attributable costs. However, the estimates did not appear stable, and an alternate approach (described below) was instead used to estimate the cost of prostate cancer-specific healthcare use. Rather than using prostate cancer and matched non-cancer patients to estimate the cost of prostate cancer-specific healthcare use, the alternate approach used only prostate cancer patients in the estimation of costs. The alternate approach used the ICD-9-CM diagnosis codes included as support for a claim when submitted to Medicare to determine if the cost associated with the claim was related to prostate cancer. If a claim included ICD-9-CM code 185.xx (the diagnosis code for prostate cancer), the cost associated with the claim was deemed to be related to prostate cancer. The cost of each claim was adjusted to year 2004 US dollars using the medical component of the Consumer Price Index. In order to balance the groups on important covariates and account for the presence of censoring, the regression techniques proposed by Lin (59) and Willan et al. (67) were used. The regression method described by Lin (59) and Willan et al. (67) divides the

PAGE 53

42 time period of interest into smaller time intervals for which cost data is available and estimates a separate regression for each of the time intervals. The method allows an individuals error terms to be correlated across each of the separate regressions for different time intervals(59). Our study estimated five regressions equations (i.e., one regression for each of five 1-year time intervals). Each equation regressed an individuals observed cost for prostate cancer-specific healthcare use on the treatment received and the individuals calculated propensity score. Equation 3-3 is an example of one of the regression equations. Ck,i = icpt,k*1 + trt,k trti + ps,k psi + k,i (3-3) In Equation 3-3, Ck,i is individual is observed cost of prostate cancer-specific healthcare use in year k, trti is a dummy variable indicating individual is initial treatment strategy, and psi is individual is calculated propensity score. Prior to carrying out the regression, a variable was created and set to equal for each individual. The creation of the variable set to equal for all individuals is used to obtain an intercept value (icpt,k) which provides an estimate of the kth year cost of prostate cancer-specific healthcare use associated with being initially treated with the treatment coded as zero on the trt dummy variable and having a propensity score of zero. The kth year cost for prostate cancer-specific healthcare use for an individual receiving the treatment coded one and having a propensity score of zero is the sum of icpt,k and trt,k. The kth year cost of prostate cancer-specific healthcare use for any given propensity score can be calculated by adding the product of the individuals propensity score and ps,k to icpt,k if the patient received the treatment coded zero or to the sum of icpt,k and trt,k if the patient received the treatment coded one. Equation 3-4 is used to calculate the kth year cost (C0,k)

PAGE 54

43 for an individual receiving the treatment coded as zero but having a propensity score indicating the individual was equally likely to have received the treatment coded as one. Equation 3-5 is used to calculate the kth year cost (C1,k) for an individual receiving the treatment coded as one but having a propensity score indicating the individual was equally likely to have received the treatment coded as zero. C0,k = icpt,k + (0.5* ps,k) (3-4) C1,k = icpt,k + trt,k + (0.5* ps,k) (3-5) As noted above, the same regression was also performed for years two, three, four, and five. The 5-year cost (C0) for an individual receiving the treatment coded zero but having a propensity score indicating the individual was equally likely to have received the treatment coded as one is calculated using Equation 3-6. The 5-year cost (C1) for an individual receiving the treatment coded one but having a propensity score indicating the individual was equally likely to have received the treatment coded as zero is calculated using Equation 3-7. C0 = icpt,1 + (0.5* ps,1) + icpt,2 + (0.5* ps,2) + icpt,3 + (0.5* ps,3) + icpt,4 + (0.5* ps,4) + icpt,5 + (0.5* ps,5) (3-6) C1 = icpt,1 + trt,1 + (0.5* ps,1) + icpt,2 + trt,2 + (0.5* ps,2) + icpt,3 + trt,3 + (0.5* ps,3) + icpt,4 + trt,4 + (0.5* ps,4) + icpt,5 + trt,5 + (0.5* ps,5) (3-7) If is defined as the {icpt,k trt,k ps,k} vector and Z is defined as the {1, trti, psi} vector, the coefficients in Equation 3-3 can be estimated using Equation 3-8 from Willan et al. (67), and the variance-covariance matrix for can be estimated using Equation 3-9 from Willan and colleagues (67). The value of *,*,/ikikXG in Equation 3-8 is the inverse probability of censoring weight described in Chapter 2 that has been shown to result in a consistent and non-biased estimate of cost in the presence of censoring.

PAGE 55

44 iiniikikikiiiiniikikkpstrtXGCpstrtpstrtXG,,1,,1,,11*,,*,11*,*, (3-8) n = number of individuals k = year = 0 if individual i was censored during year k, otherwise = 1 *,ik *,ik Xi = min(year of prostate cancer-specific death, censoring year) = min(Xi, k + 1) *,ikX = Kaplan-Meier estimate of probability of not being censored at *,ikXG *,ikX = observed cost for individual i during year k ikC, trti = individual is value for the trt dummy variable psi = individual is propensity score 111ABAnV (3-9) n = number of individuals is defined below A is defined below B niiiiipstrtpstrtnA11,,1,,1 n = number of individuals trti = individual is value for the trt dummy variable psi = individual is propensity score n1iK1kK1mmiki1nB n = number of individuals K = number of time periods (5 in our study) is defined below ki n1jj,kjijji,kiiii*i,kiiki*i,kkiFR}XX{IFps,trt,1XGps,trt,1C n = number of individuals

PAGE 56

45 k = year is from Equation 3-6 k = 0 if individual i was censored during year k, otherwise = 1 *,ik *,ik Xi = min(year of prostate cancer-specific death, censoring year) for individual i Xj = min(year of prostate cancer-specific death, censoring year) for individual j = min(Xi, k + 1) *,ikX = Kaplan-Meier estimate of probability of not being censored at *,ikXG *,ikX K1ki,kiCC i = 1, if patient i died due to prostate cancer, otherwise i = 1 i = 1 i trti = individual is value for the trt dummy variable psi = individual is propensity score Rj is defined below Fk,i is defined below jjn1j*j,kjjkj,k*j,ki*j,kii,kps,trt,1XGps,trt,1C}XX{IR1F n = number of individuals k = year is from Equation 3-6 k = 0 if individual j was censored during year k, otherwise = 1 *j,k *j,k Xi = min(year of prostate cancer-specific death, censoring year) for individual I Xj = min(year of prostate cancer-specific death, censoring year) for individual j = min(Xj, k + 1) *j,kX = Kaplan-Meier estimate of probability of not being censored at *j,kXG *j,kX = observed cost for individual j during year k j,kC j = 1, if patient j died due to prostate cancer, otherwise j = 1 j = 1 j trtj = individual js value for the trt dummy variable psj = individual js propensity score Ri is defined below }{1njijiXXIR

PAGE 57

46 n = number of individuals Xi = min(year of prostate cancer-specific death, censoring year) for individual i Xj = min(year of prostate cancer-specific death, censoring year) for individual j The treatment-specific 5-year cost of prostate cancer-specific healthcare use reported in Chapter 4 for each treatment was calculated to be representative of that for an individual equally likely to receive either of the two treatments in the comparison (i.e., the propensity score was set to equal 0.5). The Wald statistic was used to test whether a statistically significant difference in the cost of prostate cancer-specific healthcare use existed between the treatments being compared controlling for the variables included in the propensity score. The a priori significance level was set at 0.05. Cost-effectiveness For any treatment comparison in which one treatment was both more effective and more costly, we calculated the incremental cost-effectiveness ratio (ICER) using Equation 3-8. As noted in Chapter 2, the incremental cost-effectiveness ratio provides an understanding of how much one must be willing to pay for an additional unit of effectiveness for the more effective, more costly treatment to be cost effective compared to the less effective, less costly treatment. We chose not to make a conclusion as to whether a more effective, more costly treatment is cost effective. Instead, we have reported the incremental cost effectiveness ratio in terms of the cost one must be willing to pay to prevent one death due to prostate cancer over a 5-year period in order to make the more effective, more costly treatment cost effective. It is up to the reader to determine whether the cost to prevent one death due to prostate cancer over a 5-year period makes the more effective, more costly treatment cost effective or not. In Equation 3-8, was the difference in the 5-year cost of prostate cancer-specific healthcare use c

PAGE 58

47 between the treatments being compared. The difference in effectiveness, was determined using the estimate of the 5-year survival function for the less effective treatment and the hazard ratio that adjusts for differences in the characteristics included in the propensity score (SEER registry, urban/rural classification, race, tumor stage, etc.). e ecICER (3-8)

PAGE 59

48 Table 3-1. Radiotherapy (external beam or brachytherapy) CPT/HCPCS codes CPT/HCPCS Description 76078 radiographic absorptiometry 76950 ultrasonic guidance for placement of radiation therapy fields 76965 ultrasonic guidance for interstitial radioelement application 77261-77264 therapeutic radiology treatment planning 77280, 77285, 77290, 77295 therapeutic radiology simulation-aided field setting 77299 unlisted procedure, therapeutic radiology clinical treatment planning 77300 basic radiation dosimetry calculation 77305, 77310, 77315 teletherapy, isodose plan 77321 special teletherapy port plan 77326-77328 brachytherapy isodose plan 77331 special dosimetry 77332-77334 treatment devices, design and construction 77336 continuing medical physics consultation 77370 special medical radiation physics consultation 77380, 77520 proton treatment delivery 77399 unlisted procedure, medical radiation physics consultation 77400-77416 radiation treatment delivery 77417 therapeutic radiology port films 77419, 77420, 77424, 77425 weekly radiation therapy management 77427, 77430-77435 radiation treatment management 77470 special radiation treatment procedure 77499 unlisted procedure, therapeutic radiology treatment management 77750 intracavitary radiation source application 77761-77263 intracavitary radiation source application 77776-77778 interstitial radiation source application 77781-77784 remote afterloading high intensity brachytherapy 77789 surface application of radiation source 77790 supervision, handling, loading of radiation source 77799 unlisted procedure, clinical bracytherapy 79900 provide therapeutic radiopharmaceuticals Table 3-2. Androgen deprivation CPT/HCPCS codes HCPCS/CPT Description 54520, 54521, 54529, 54530 remove testis J9202 goserelin J1950, J9217, J9218 leuprolide

PAGE 60

49 Table 3-3. Androgen deprivation ICD9-CM procedure codes ICD9-CM Procedure Code Description 62 Operations on testes 62.0 Incision of testis 62.2 Excision or destruction of testicular lesion 62.3 Unilateral orchiectomy 62.4x Bilateral orchiectomy Table 3-4. Other prostate cancer treatment CPT/HCPCS codes CPT Code Description 53850 transurethral destruction of prostate tissue, by microwave thermotherapy 53852 transurethral destruction of prostate tissue, by radiofrequency thermotherapy 55860 exposure of prostate, any approach, for insertion of radioactive substance 77600 hyperthermia, externally generated; superficial 77605 hyperthermia, externally generated; deep 77610 hyperthermia generated by interstitial probes (5 or fewer applicators) 77615 hyperthermia generated by interstitial probes (more than 5 applicators) Table 3-5. Prostatectomy CPT/HCPCS codes CPT/HCPCS Description 52601 transurethral electrosurgical resection of prostate 52612, 52614, 52620, 52630 transurethral resection of prostate 52647 non-contact laser coagulation of prostate 52648 contact laser vaporization with or without transurethral resection of prostate 52650 transurethral prostate surgery 55810, 55812, 55815, 55840, 55842, 55845 extensive prostate surgery 55801, 55821, 55831 removal of prostate G0160 cryosurgical ablation of localized prostate cancer Table 3-6. Prostatectomy ICD9-CM procedure codes ICD9-CM Procedure Code Description 60 Operations on prostate and seminal vesicles 60.0 Incision of prostate 60.2x Transurethral prostatectomy 60.3 Suprapubic prostatectomy 60.4 Retropubic prostatectomy 60.5 Radical prostatectomy 60.6x Other prostatectomy 60.8x Incision or excision of periprostatic tissue 60.9x Other operations on prostate

PAGE 61

CHAPTER 4 RESULTS A total of 161,557 men residing in Surveillance, Epidemiology, and End Results (SEER) geographic areas were diagnosed with prostate cancer between 1992 and 1999. The elimination of individuals without continuous eligibility, individuals participating in a Medicare health maintenance organization, individuals with periods of time in which the individual was not enrolled in Medicare between the time of diagnosis and death, and those not at either low or intermediate risk of recurrence reduced our study sample to 38,334 individuals. The average age among those patients at either low or intermediate risk of prostate cancer recurrence was 74.0 (SD = 5.6). Of the 38,334 prostate cancer cases, 85.7% of the individuals were Caucasian, 8.5% were African-American, 1.5% were Hispanic, 2.5% were Asian, 0.2% were Native American, and 1.6% were classified as other or unknown. Seven hundred sixteen individuals died due to prostate cancer within five years of diagnosis. SEER registry (an indication of the individuals geographic location), race (Table 4-1), urban/rural classification (Table 4-2), tumor grade (Table 4-3), tumor stage (Table 4-4), diagnosis year (Table 4-5), and age at diagnosis (77.3 for patients dying due to prostate cancer within five years of diagnosis and 73.9 for patients not dying due to prostate cancer within five years of diagnosis) were found to be related to prostate cancer-specific death (p < 0.05 for each). The difference in Charlson comorbidity score, which increases with the number illnesses a patient has, for those dying due to prostate 50

PAGE 62

51 cancer (0.333) and those not dying due to prostate cancer (0.329) within five years of diagnosis was not significantly different (p = 0.07). Only 12,292 of the 38,334 prostate cancer patients at low or intermediate risk of recurrence were initially treated with one of the treatment strategies recommended in the National Comprehensive Cancer Network (NCCN) prostate cancer clinical practice guideline for patients at low or intermediate recurrence risk. Of the 12,292 patients treated with one of the recommended strategies, 1,993 were treated with expectant management (EM), 3,313 were treated with external beam radiotherapy (EBR), 567 were treated with brachytherapy (BT), 5,954 were treated with prostatectomy (RP), and 465 were treated with external beam radiotherapy plus brachytherapy (EBRwBT). The remaining 26,042 patients received some other type of treatment or a combination of the recommended treatments. Of the 26,042 individuals not receiving one of the recommended treatment strategies, 633 (2.4%) died due to prostate cancer within five years of diagnosis. Of the 12,944 individuals treated with a recommended treatment strategy, 83 (0.7%) died due to prostate cancer within five years of diagnosis. Patients treated with a recommended treatment strategy and patients not treated with a recommended treatment strategy differed significantly (p < 0.05 for each comparison) in terms of geographic area (SEER registry), race (Table 4-6), urban/rural classification (Table 4-7), tumor grade (Table 4-8), tumor stage (Table 4-9), year of diagnosis (Table 4-10), age at diagnosis (72.0 for patients treated with a recommended treatment strategy and 74.9 for patients treated with a non-recommended treatment strategy), and Charlson comorbidity score (0.25 for

PAGE 63

52 patients treated with a recommended treatment strategy and 0.37 for patients treated with a non-recommended treatment strategy). As discussed in Chapter 3, the propensity score was used to control for differences between treatment groups on potential confounders. In order for a variable to be a confounder, the variable must be related to both the treatment received and the outcome of interest. Omnibus tests for differences between treatment strategies for the distributions of registry, race (Table 4-11), urban/rural classification versus rural area (Table 4-12), tumor grade (Table 4-13), tumor stage (Table 4-14), year of diagnosis (Table 4-15), age at diagnosis (Table 4-16), and Charlson comorbidity score (Table 4-16) by initial treatment strategy identified significant differences for each of the hypothesized confounders (p < 0.05 for each). The sections below report the results of the Cox proportional hazard and cost regressions for each comparison of treatment pairs. The Cox proportional hazard model was not used for comparisons including brachytherapy or external beam radiotherapy with brachytherapy because there were no deaths due to prostate cancer within 5 years of diagnosis for patients receiving either treatment strategy. While the lack of deaths due to prostate cancer in patients treated with either brachytherapy or external beam radiotherapy with brachytherapy could lead to a temptation to consider these treatment strategies superior, any such temptation must be tempered by the fact that these treatment strategies had significantly fewer patients diagnosed prior to 1999 leading to less opportunity for those patients treated using brachytherapy or external beam radiotherapy with brachytherapy to be observed for at least 5 years. The study sample included 1,208, 2,675, and 5,228 patients diagnosed prior to 1999 initially managed with expectant

PAGE 64

53 management, external beam radiotherapy, and prostatectomy, respectively. Meanwhile, the study sample included only 287 and 260 patients diagnosed prior to 1999 receiving brachytherapy and external beam radiotherapy with brachytherapy, respectively. Expectant Management vs. External Beam Radiotherapy The comparison of patients receiving expectant management and external beam radiotherapy included 1,993 patients initially managed using expectant management and 3,313 patients initially treated using external beam radiotherapy. Of the 1,993 patients treated using expectant management, 20 died due to prostate cancer within five years of diagnosis. The 5-year KM estimate of the prostate cancer-specific survival function for expectant management was 0.9872. Of the 3,313 patients treated using external beam radiotherapy, 37 died due to prostate cancer within five years of diagnosis. The 5-year KM estimate of the prostate cancer-specific survival function for external beam radiotherapy was 0.9874. The hazard ratio comparing expectant management to external beam radiotherapy was 0.505 (95% confidence interval: 0.263 0.969), indicating patients treated with external beam radiotherapy were only 50.5% as likely to die due to prostate cancer at any time as patients treated with expectant management controlling for the hypothesized confounders (i.e., SEER registry, urban/rural classification, race, tumor grade, tumor stage, age at diagnosis, Charlson comorbidity score, and year of diagnosis) through inclusion of the propensity score in the Cox proportional hazards model. The assumption of proportional hazards seems reasonable based on inspection of the plot of the log cumulative hazard versus log time presented in Figure 4-1 (i.e., the curves are roughly parallel). Additionally, neither the interaction term for treatment and time (p = 0.245) nor the interaction term for the propensity score and time (p = 0.449) were significant. As

PAGE 65

54 such, it is reasonable to use the Cox proportional hazard model to compare expectant management and external beam radiotherapy. The cost of prostate cancer-specific health care use in the first five years following diagnosis was $15,460 for patients initially treated with expectant management and $19,060 for patients initially treated with external beam radiotherapy. The $3,600 difference in 5-year cost was statistically significant (Wald statistic = 97.7, p < 0.01). Cumulative cost by initial treatment strategy for patients initially managed using expectant management or external beam radiotherapy is presented in Figure 4-2. The minimum willingness-to-pay value to make treatment with external beam radiotherapy cost-effective compared to expectant management was $600,000 per death due to prostate cancer prevented over the course of five years. Expectant Management vs. Brachytherapy The comparison of patients receiving expectant management and brachytherapy included 1,993 patients initially managed using expectant management and 567 patients initially treated using brachytherapy. Of the 1,993 patients treated using expectant management, 20 died due to prostate cancer within five years of diagnosis. The 5-year KM estimate of the prostate cancer-specific survival function for expectant management was 0.9872. Of the 567 patients treated using brachytherapy, none died due to prostate cancer within five years of diagnosis. The 5-year KM estimate of the prostate cancer-specific survival function for brachytherapy was 1. The cost of prostate cancer-specific health care use in the first five years following diagnosis was $15,780 for patients initially treated with expectant management and $18,990 for patients initially treated with brachytherapy. The $3,210 difference in 5-year cost was statistically significant (Wald statistic = 13.9, p < 0.01). Cumulative cost by

PAGE 66

55 initial treatment strategy for patients initially managed using expectant management or brachytherapy is presented in Figure 4-3. Expectant Management vs. Prostatectomy The comparison of patients receiving expectant management and prostatectomy included 1,993 patients initially managed using expectant management and 5,954 patients initially treated using prostatectomy. Of the 1,993 patients treated using expectant management, 20 died due to prostate cancer within five years of diagnosis. The 5-year KM estimate of the prostate cancer-specific survival function for expectant management was 0.9872. Of the 5,954 patients treated using prostatectomy, 26 died due to prostate cancer within five years of diagnosis. The 5-year KM estimate of the prostate cancer-specific survival function for prostatectomy was 0.9955. The hazard ratio comparing expectant management to prostatectomy was 0.307 (95% confidence interval: 0.137 0.690), indicating patients treated with prostatectomy were only 30.7% as likely to die due to prostate cancer at any time as patients treated with expectant management controlling for the hypothesized confounders through inclusion of the propensity score in the Cox proportional hazards model. The assumption of proportional hazards seems reasonable based on inspection of the plot of the log cumulative hazard versus log time presented in Figure 4-4 (i.e., the curves are roughly parallel). Additionally, neither the interaction term for treatment and time (p = 0.199) nor the interaction term for the propensity score and time (p = 0.728) were significant. As such, it is reasonable to use the Cox proportional hazard model to compare expectant management and prostatectomy. The cost of prostate cancer-specific health care use in the first five years following diagnosis was $15,640 for patients initially treated with expectant management and

PAGE 67

56 $21,460 for patients initially treated with prostatectomy. The $5,820 difference in 5-year cost was statistically significant (Wald statistic = 158.1, p < 0.01). Cumulative cost by initial treatment strategy for patients initially managed using expectant management or prostatectomy is presented in Figure 4-5. The minimum willingness-to-pay value to make treatment with prostatectomy cost-effective compared to expectant management was $646,670 per death due to prostate cancer prevented over the course of five years. Expectant Management vs. External Beam Radiotherapy with Brachytherapy The comparison of patients receiving expectant management and external beam radiotherapy with brachytherapy included 1,993 patients initially managed using expectant management and 465 patients initially treated using external beam radiotherapy with brachytherapy. Of the 1,993 patients treated using expectant management, 20 died due to prostate cancer within five years of diagnosis. The 5-year KM estimate of the prostate cancer-specific survival function for expectant management was 0.9872. Of the 465 patients treated using external beam radiotherapy with brachytherapy, none died due to prostate cancer within five years of diagnosis. The 5-year KM estimate of the prostate cancer-specific survival function for patients initially treated using external beam radiotherapy with brachytherapy was 1. The cost of prostate cancer-specific health care use in the first five years following diagnosis was $14,970 for patients initially treated with expectant management and $21,710 for patients initially treated using external beam radiotherapy with brachytherapy. The $6,740 difference in 5-year cost was statistically significant (Wald statistic = 25.0, p < 0.01). Cumulative cost by initial treatment strategy for patients

PAGE 68

57 initially managed using expectant management or external beam radiotherapy with brachytherapy is presented in Figure 4-6. External Beam Radiotherapy vs. Brachytherapy The comparison of patients receiving external beam radiotherapy and brachytherapy included 3,313 patients initially managed using external beam radiotherapy and 567 patients initially treated using brachytherapy. Of the 3,313 patients treated using external beam radiotherapy, 37 died due to prostate cancer within five years of diagnosis. The 5-year KM estimate of the prostate cancer-specific survival function for patients initially treated using external beam radiotherapy was 0.9874. Of the 567 patients treated using brachytherapy, none died due to prostate cancer within five years of diagnosis. The Kaplan-Meier (KM) estimate of the 5-year prostate cancer-specific survival function for patients initially treated with brachytherapy was 1. The cost of prostate cancer-specific health care use in the first five years following diagnosis was $17,890 for patients initially treated with external beam radiotherapy and $17,260 for patients initially treated with brachytherapy. The $630 difference in 5-year cost was statistically significant (Wald statistic = 8.1, p < 0.01). Cumulative cost by initial treatment strategy for patients initially managed using external beam radiotherapy or brachytherapy is presented in Figure 4-7. External Beam Radiotherapy vs. Prostatectomy The comparison of patients receiving external beam radiotherapy to those receiving prostatectomy included 3,313 patients initially managed using external beam radiotherapy and 5,954 patients initially treated using prostatectomy. Of the 3,313 patients treated using external beam radiotherapy, 37 died due to prostate cancer within five years of diagnosis. The 5-year KM estimate of the prostate cancer-specific survival

PAGE 69

58 function for patients initially treated using external beam radiotherapy was 0.9874. Of the 5,954 patients treated using prostatectomy, 26 died due to prostate cancer within five years of diagnosis. For patients treated with prostatectomy the 5-year KM survival function estimate was 0.9955. The hazard ratio comparing external beam radiotherapy to prostatectomy was 0.432 (95% confidence interval: 0.224 0.833), indicating patients treated with radical prostatectomy were only 43.2% as likely to die at any time as patients treated with external beam radiotherapy controlling for the hypothesized confounders through inclusion of the propensity score in the Cox proportional hazards model. The assumption of proportional hazards seems reasonable based on inspection of the plot of the log cumulative hazard versus log time presented in Figure 4-8 (i.e., the curves are roughly parallel). Additionally, the interaction term for propensity score and time (p = 0.498) was not significant. The interaction term for treatment and time (p = 0.005) was significant. The significant interaction between treatment and time could be an indication of a violation of the proportional hazards assumption. Despite the significant interaction between treatment and time, the parallel nature of the log cumulative hazard versus log time plot (Figure 4-8) provides sufficient support for the assumption of proportional hazards and supports the validity of using the Cox proportional hazards model to compare external beam radiotherapy and prostatectomy. The cost of prostate cancer-specific health care use in the first five years following diagnosis was $20,190 for patients initially treated with external beam radiotherapy and $23,460 for patients initially treated with prostatectomy. The $3,260 difference in 5-year cost was statistically significant (Wald statistic = 91.8, p < 0.01). Cumulative cost by

PAGE 70

59 initial treatment strategy for patients initially managed using external beam radiotherapy or prostatectomy is presented in Figure 4-9. The minimum willingness-to-pay value to make treatment with prostatectomy cost-effective compared to external beam radiotherapy was $465,710 per death due to prostate cancer prevented over the course of five years. External Beam Radiotherapy vs. External Beam Radiotherapy with Brachytherapy The comparison of patients receiving external beam radiotherapy and external beam radiotherapy with brachytherapy included 3,313 patients initially managed using external beam radiotherapy and 465 patients initially treated using external beam radiotherapy with brachytherapy. Of the 3,313 patients treated using external beam radiotherapy, 37 died due to prostate cancer within five years of diagnosis. The 5-year KM estimate of the prostate cancer-specific survival function for patients initially treated using external beam radiotherapy was 0.9874. Of the 465 patients treated using external beam radiotherapy with brachytherapy, none died due to prostate cancer within five years of diagnosis. The 5-year KM estimate of the prostate cancer-specific survival function for patients initially treated using external beam radiotherapy with brachytherapy was 1. The cost of prostate cancer-specific health care use in the first five years following diagnosis was $17,730 for patients initially treated with external beam radiotherapy and $20,400 for patients initially treated using external beam radiotherapy with brachytherapy. The $2,680 difference in 5-year cost was statistically significant (Wald statistic = 34.9, p < 0.01). Cumulative cost by initial treatment strategy for patients initially managed using external beam radiotherapy or external beam radiotherapy with brachytherapy is presented in Figure 4-10.

PAGE 71

60 Brachytherapy vs. Prostatectomy The comparison of patients receiving brachytherapy and prostatectomy included 567 patients initially managed using brachytherapy and 5,954 patients initially treated using prostatectomy. Of the 567 patients treated using brachytherapy, none died due to prostate cancer within five years of diagnosis. The 5-year KM estimate of the prostate cancer-specific survival function for brachytherapy was 1. Of the 5,954 patients treated using prostatectomy, 26 died due to prostate cancer within five years of diagnosis. The 5-year KM estimate of the prostate cancer-specific survival function for prostatectomy was 0.9955. The cost of prostate cancer-specific health care use in the first five years following diagnosis was $18,430 for patients initially treated with brachytherapy and $17,100 for patients initially treated using prostatectomy. The $1,340 difference in 5-year cost was statistically significant (Wald statistic = 33.5, p < 0.01). Cumulative cost by initial treatment strategy for patients initially managed using brachytherapy or prostatectomy is presented in Figure 4-11. Brachytherapy vs. External Beam Radiotherapy with Brachytherapy The comparison of patients receiving brachytherapy and those patients receiving external beam radiotherapy with brachytherapy included 567 patients initially managed using brachytherapy and 465 patients initially treated using external beam radiotherapy with brachytherapy. Of the 567 patients treated using brachytherapy, none died due to prostate cancer within five years of diagnosis. The 5-year KM estimate of the prostate cancer-specific survival function for brachytherapy was 1. Of the 465 patients treated using external beam radiotherapy with brachytherapy, none died due to prostate cancer within five years of diagnosis. The 5-year Kaplan-Meier estimate of the prostate cancer

PAGE 72

61 specific survival function for treatment using external beam radiotherapy with brachytherapy was 1. The cost of prostate cancer-specific health care use in the first five years following diagnosis was $18,380 for patients initially treated with brachytherapy and $22,160 for patients initially treated using external beam radiotherapy with brachytherapy. The $3,780 difference in 5-year cost was statistically significant (Wald statistic = 6.07, p < 0.01). Cumulative cost by initial treatment strategy for patients initially managed using brachytherapy or external beam radiotherapy with brachytherapy is presented in Figure 4-12. Prostatectomy vs. External Beam Radiotherapy with Brachytherapy The comparison of patients receiving prostatectomy and external beam radiotherapy with brachytherapy included 5,954 patients initially managed using prostatectomy and 465 patients initially treated using external beam radiotherapy with brachytherapy. Of the 5,954 patients treated using prostatectomy, 26 died due to prostate cancer within five years of diagnosis. The 5-year KM estimate of the prostate cancer-specific survival function for prostatectomy was 0.9955. Of the 465 patients treated using external beam radiotherapy with brachytherapy, none died due to prostate cancer within five years of diagnosis. The 5-year KM estimate of the prostate cancer-specific survival function for treatment using external beam radiotherapy with brachytherapy was 1. The cost of prostate cancer-specific health care use in the first 5 years following diagnosis was $17,720 for patients initially treated with prostatectomy and $20,330 for patients initially treated using external beam radiotherapy with brachytherapy. The $2,600 difference in 4-year cost was statistically significant (Wald statistic = 56.9, p <

PAGE 73

62 0.01). Cumulative cost by initial treatment strategy for patients initially managed using prostatectomy or external beam radiotherapy with brachytherapy is presented in Figure 4-13. Table 4-1. Race distribution by 5-year prostate cancer-specific survival status Patients with prostate cancer-specific death within 5 years (%) Patients without prostate cancer-specific death within 5 years (%) White 85.8 82.7 Black 8.4 12.2 Asian 2.6 1.7 Hispanic 1.5 1.1 Native North American 0.2 0.4 Other/unknown 1.6 2.0 Table 4-2. Urban/rural classification distribution by 5-year prostate cancer-specific survival status Patients with prostate cancer-specific death within 5 years (%) Patients without prostate cancer-specific death within 5 years (%) Big metro 61.1 52.0 Metro 21.0 25.7 Urban 7.2 11.5 Less urban 9.0 9.6 Rural 1.8 1.3 Table 4-3. Tumor grade distribution by 5-year prostate cancer-specific survival status Patients with prostate cancer-specific death within 5 years (%) Patients without prostate cancer-specific death within 5 years (%) Well differentiated 19.6 14.4 Moderately differentiated 80.4 85.6 Table 4-4. Tumor stage distribution by 5-year prostate cancer-specific survival status Patients with prostate cancer-specific death within 5 years (%) Patients without prostate cancer-specific death within 5 years (%) Clinically unapparent 37.2 26.4 Clinically/ radiographically apparent 33.2 36.3 Clinically unapparent or clinically/ radiographically apparent 29.6 37.3

PAGE 74

63 Table 4-5. Diagnosis year by 5-year prostate cancer-specific survival status Patients with prostate cancer-specific death within 5 years (%) Patients without prostate cancer-specific death within 5 years (%) 1992 97.2 2.8 1993 97.2 2.8 1994 97.5 2.5 1995 97.7 2.3 1996 98.0 2.0 1997 98.9 1.1 1998 99.2 0.8 1999 99.7 0.3 Table 4-6. Race distribution for patients treated with recommended and non-recommended strategies Recommended treatment strategy (%) Non-recommended treatment strategy (%) White 85.8 85.4 Black 8.0 9.5 Asian 2.7 2.1 Hispanic 1.5 1.6 Native North American 0.2 0.2 Other/unknown 1.8 1.2 Table 4-7. Urban/rural classification distribution for patients treated with recommended and non-recommended strategies Recommended treatment strategy (%) Non-recommended treatment strategy (%) Big metro 58.0 67.3 Metro 23.2 16.5 Urban 7.5 6.6 Less urban 9.5 8.0 Rural 1.8 1.6 Table 4-8. Tumor grade distribution for patients treated with recommended and non-recommended strategies Recommended treatment strategy (%) Non-recommended treatment strategy (%) Well differentiated 23.0 12.2 Moderately differentiated 77.0 87.8

PAGE 75

64 Table 4-9. Tumor stage distribution for patients treated with recommended and non-recommended strategies Recommended treatment strategy (%) Non-recommended treatment strategy (%) Clinically unapparent 38.8 33.4 Clinically or radiographically apparent 31.9 36.0 Clinically unapparent or clinically or radiographically apparent 29.4 30.6 Table 4-10. Diagnosis year distribution for patients treated with recommended and non-recommended strategies Recommended treatment strategy (%) Non-recommended treatment strategy (%) 1992 29.3 70.7 1993 25.9 74.1 1994 26.2 73.8 1995 26.5 73.5 1996 25.2 74.8 1997 25.1 74.9 1998 53.3 46.7 1999 51.2 48.8 Table 4-11. Race distribution by initial treatment strategy EM (%) EBR (%) BT (%) RP (%) EBRw/BT (%) White 82.0 80.2 89.1 89.0 86.9 Black 11.9 16.1 6.5 5.4 9.0 Asian 2.8 1.7 1.4 2.2 2.2 Hispanic 1.8 0.9 1.6 2.0 1.3 Native North American 0.2 0.1 0.2 0.2 0.0 Other/unknown 1.3 1.1 1.2 1.3 0.6 Table 4-12. Urban/rural classification distribution by initial treatment strategy EM (%) EBR (%) BT (%) RP (%) EBRw/BT (%) Big metro 65.1 80.1 69.5 59.7 79.8 Metro 14.4 10.6 17.3 20.8 11.2 Urban 8.8 4.0 4.2 7.8 4.1 Less urban 10.2 4.4 8.3 9.6 3.9 Rural 1.5 0.9 0.7 2.1 1.1 Table 4-13. Tumor grade distribution by initial treatment strategy EM(%) EBR (%) BT (%) RP (%) EBRw/BT (%) Well differentiated 19.3 12.7 7.9 10.4 7.7 Moderately differentiated 80.7 87.3 92.1 89.6 92.3

PAGE 76

65 Table 4-14. Tumor stage distribution by initial treatment strategy EM (%) EBR (%) BT (%) RP (%) EBRw/BT (%) Clinically unapparent 36.6 32.0 42.3 32.2 34.4 Clinically/radiographically apparent confined to prostate 26.4 36.8 28.2 39.8 33.1 Clinically unapparent or clinically/radiographically apparent confined to prostate 37.0 31.2 29.5 28.1 32.5 Table 4-15. Diagnosis year distribution by initial treatment strategy EM (%) EBR (%) BT (%) RP (%) EBRw/BT (%) 1992 2.9 27.5 0.2 69.4 0.1 1993 3.8 30.7 0.4 64.9 0.2 1994 4.9 28.5 0.6 65.2 0.8 1995 6.7 25.3 0.8 66.3 0.8 1996 8.4 23.5 2.5 64.7 1.0 1997 8.5 20.3 2.8 65.3 3.1 1998 31.0 26.0 8.0 27.6 7.5 1999 29.8 24.2 10.6 27.6 7.8 Table 4-16. Age and Charlson comorbidity score by initial treatment strategy EM Mean (sd) EBR Mean (sd) BT Mean (sd) RP Mean (sd) EBRw/BT Mean (sd) Age at diagnosis 76.0 y (5.7) 73.4 y (4.3) 72.0 y (3.9) 69.9 y (3.2) 72.3 y (4.0) Charlson comorbidity score 0.382 (0.796) 0.334 (0.686) 0.252 (0.636) 0.164 (0.485) 0.217 (0.551)

PAGE 77

66 Figure 4-1. Expectant management (EM) vs. external beam radiotherapy (EBR) log cumulative hazard vs. log time 05000100001500020000 yr 1yr 2 yr 3yr 4yr 5 expectant management external beam radiotherapy Figure 4-2. Expectant management vs. external beam radiotherapy cumulative cost by year

PAGE 78

67 05000100001500020000 yr 1yr 2 yr 3yr 4yr 5 expectant management brachytherapy Figure 4-3. Expectant management vs. brachytherapy cumulative cost by year Figure 4-4. Expectant management (EM) vs. prostatectomy (RP) log cumulative hazard vs. log time

PAGE 79

68 0500010000150002000025000 yr 1yr 2yr 3yr 4yr 5 expectant management prostatectomy Figure 4-5. Expectant management vs. prostatectomy cumulative cost by year 0500010000150002000025000 yr 1yr 2yr 3yr 4yr 5 expectant management external beam radiotherapy with brachytherapy Figure 4-6. Expectant management vs. external beam radiotherapy with brachytherapy cumulative cost by year

PAGE 80

69 020004000600080001000012000140001600018000 yr 1yr 2yr 3yr 4yr 5 external beam radiotherapy brachytherapy Figure 4-7. External beam radiotherapy vs. brachytherapy cumulative cost by year Figure 4-8. External beam radiotherapy (EBR) vs. prostatectomy (RP) log cumulative hazard vs. log time

PAGE 81

70 0500010000150002000025000 yr 1yr 2yr 3yr 4yr 5 external beam radiotherapy prostatectomy Figure 4-9. External beam radiotherapy vs. prostatectomy cumulative cost by year 0500010000150002000025000 yr 1yr 2yr 3yr 4yr 5 external beam radiotherapy external beam radiotherapy with brachytherapy Figure 4-10. External beam radiotherapy vs. external beam radiotherapy with brachytherapy cumulative cost by year

PAGE 82

71 05000100001500020000 yr 1yr 2yr 3yr 4yr 5 brachytherapy prostatectomy Figure 4-11. Brachytherapy vs. prostatectomy cumulative cost by year 0500010000150002000025000 yr 1yr 2 yr 3yr 4yr 5 brachytherapy external beam radiotherapy with brachytherapy Figure 4-12. Brachytherapy vs. external beam radiotherapy with brachytherapy cumulative cost by year

PAGE 83

72 0500010000150002000025000 yr1yr2 yr3yr4yr5 prostatectomy external beam radiotherapy with brachytherapy Figure 4-13. Prostatectomy vs. external beam radiotherapy with brachytherapy cumulative cost by year

PAGE 84

CHAPTER 5 DISCUSSION Limitations Our study has several significant limitations. First, the limited number of patients initially treated using either brachytherapy or external beam radiotherapy with brachytherapy prevented us from being able to compare the clinical effectiveness of treatment using either brachytherapy or external beam radiotherapy with brachytherapy to the clinical effectiveness of any of the other recommended treatment strategies. However, it was possible to compare treatment using brachytherapy or external beam radiotherapy with brachytherapy to all other recommended treatment strategies in terms of the 5-year cost of prostate cancer-specific healthcare use. Second, the findings of our study must be interpreted in light of the studys observation design. Given the lack of randomization, the potential exists for any differences in the effectiveness or cost of the treatments to be the result of a selection bias leading to non-equivalent groups rather than a true difference in the effectiveness or cost of the treatments being compared. Indeed, comparison of treatment groups for differences in SEER registry (an indicator of geographic area), urban/rural classification, race, tumor stage, tumor grade, year of diagnosis, age at diagnosis, and Charlson comorbidity score found significant differences for each between the treatment groups. Furthermore, each of these characteristics (except the Charlson comorbidity score) was related to prostate cancer-specific death. Despite the presence of a selection bias, confidence in our results is strengthened by the use of a multivariate analysis 73

PAGE 85

74 incorporating the calculated propensity score, which controls for differences between treatments in the previously listed characteristics related to the treatment strategy a patient received, to compare prostate cancer-specific mortality and 5-year cost of prostate cancer-specific healthcare. While the use of an observational study design is a limitation, it is also a strength of our study because it allows for the assessment of whether differences in effectiveness exist between the compared treatments as opposed to a difference in efficacy. Additionally, the observational design provides a more realistic estimate of the cost of treatment because an experimental design typically influences the pattern of care through the experiments protocol requirements (e.g., additional follow-up visits and tests). Information on effectiveness and costs in a non-protocol-driven environment can be very useful seeing as most individuals are not treated in the clinical trial setting. Instead, most individuals are treated in the real world setting with all of its potential effect modifiers and without protocol-driven follow-up requirements. Classification of the cause of death was based on the death certificate-coded cause of death. It is possible that relying on the inaccurate coding of the cause of death on the death certificated resulted in some patients classified in our study as dying due to prostate cancer having instead died due to some other cause. Similarly, it is possible that some patients treated as censored observations in our study did indeed die due to prostate cancer but the death certificate did not reflect prostate cancer as the cause of death. In terms of the estimation of the cost of prostate cancer-specific healthcare use, we relied upon ICD9-CM diagnosis codes on claims submitted to Medicare to determine if the claim was related to prostate cancer. It is possible that reliance upon billing diagnosis

PAGE 86

75 codes caused us to classify some non-prostate cancer-related costs as related to prostate cancer. Additionally, the reliance upon billing diagnosis codes could have resulted in us classifying some prostate cancer-related costs as not being related to prostate cancer. Furthermore, our study was not able to account for the cost of prostate cancer-related healthcare (e.g., oral prescription medications) that is not covered by Medicare. Prostate Cancer-Specific Mortality We identified four studies published in the last ten years comparing the rate of prostate-cancer specific mortality between two or more of the treatment strategies recommended in the National Comprehensive Cancer Network prostate cancer clinical practice guideline. Similar to our study, two of the published studies examined 5-year prostate cancer-specific mortality. Our study found the 5-year prostate cancer-specific mortality rate for patients treated with expectant management, external beam radiotherapy, brachytherapy, prostatectomy, and external beam radiotherapy with brachytherapy to be 1.3%, 1.3%, 0%, 0.4%, and 0% respectively. However, as noted in Chapter 4, the rates for brachytherapy and external beam radiotherapy with brachytherapy must be interpreted with caution due to the relatively small number of patients with the potential for five years of follow-up receiving brachytherapy or external beam radiotherapy with brachytherapy. Holmberg et al. (18) found the 5-year prostate cancer-specific mortality rate for patients treated with expectant management (EM) or prostatectomy (RP) to be 4.6% and 2.6% respectively in a prospective randomized trial with a hazard ratio of 0.45 (95% CI: 0.250.84) favoring prostatectomy. The hazard ratio in our study also favored prostatectomy (HR = 0.307, 95% CI: 0.137 0.690). Differences in survival rates between the study by Holmberg et al. and our study (EM: 4.6% vs. 1.3%; RP: 2.6% vs.

PAGE 87

76 0.4%) are likely due to patients included in our study having tumors of a less advanced stage. Fowler et al. (7) found the 5-year prostate cancer-specific mortality rate for patients treated with radical prostatectomy or external beam radiotherapy (EBR) to be 5% and 11%, respectively. Fowler et al. did not report a hazard ratio, but stated the difference in prostate cancer-specific survival was not significant. Our study did find a significant difference in survival (HR = 0.66, 95% CI: 0.47-0.91). Differences in survival rates between the study by Fowler et al. and our study (RP: 5.0% vs. 0.4%; EBR: 11.0% vs. 1.3%) are likely due to patients included in our study having less advanced tumors in terms of both stage and grade than patients in the study by Fowler and colleagues. Prostate Cancer-Specific Cost Our study significantly bolsters the available information on the cost of treating prostate cancer. As noted in Chapter 2, seven studies published since 1995 have addressed prostate cancer treatment-specific costs. The only study examining costs beyond the initial treatment period was carried out by Perez et al. (32) and used a modeling approach to estimate the accrual of cost for five years beyond the initial treatment phase. Our study has provided 5-year prostate cancer-specific healthcare use cost estimates for treatment strategies recommended in the National Comprehensive Cancer Network prostate cancer clinical practice guideline for patients with localized prostate cancer (i.e. patients at either low or intermediate risk of prostate cancer recurrence). Each estimate was obtained using patient-level data. As such, the cost data was stochastic in nature, which allowed for the testing of differences in cost between treatments using inferential statistics. Our study appears to be the first study of prostate cancer-specific health care cost to properly account for censoring of cost data.

PAGE 88

77 The wide range of the cost estimates for the same treatment strategy obtained when comparing a specific treatment to each of the other recommended treatments is slightly surprising. For example, there is a difference of $6,360 between estimates of the 5-year cost of prostatectomy depending on whether the estimate was made in the comparison with brachytherapy or external beam radiotherapy. The cost of prostatectomy was $17,100 when compared to brachytherapy, and the cost of prostatectomy was $23,460 when compared to external beam radiotherapy. The range of the cost estimates for the other treatment strategies were not as dramatic. The ranges for expectant management, external beam radiotherapy, brachytherapy, and external beam radiotherapy with brachytherapy were $810, $2,460, $1,730, and $1,830, respectively. The likely explanation for the differences in the cost estimates for the same treatment is the difference in the values of the variables included in the propensity score (SEER registry, urban/rural classification, tumor stage, etc.) between a patient equally likely to receive expectant management or prostatectomy (i.e., a patient with a set of values for the variables in the propensity score resulting in the propensity score calculated in the expectant management versus prostatectomy comparison equaling 0.5) and a patient equally likely to receive external beam radiotherapy or prostatectomy (i.e., a patient with a set of values for the variables in the propensity score resulting in the propensity score calculated in the external beam radiotherapy versus prostatectomy comparison equaling 0.5). A wide range of the cost estimates for the same treatment is expected in light of the realization that each estimate is for a specific set of values for the variables included in the propensity score. As such, conclusions about the cost of treatments estimated using the methods of our study should be limited to relative

PAGE 89

78 differences in cost between treatment strategies rather than the provision of an estimate to be used for the purposes of estimating the absolute cost of providing care using a specific treatment. An absolute estimate of the cost of a treatment would depend on the type of patient (geographic location, urban/rural classification, tumor stage, tumor grade, etc.) for whom the estimate is desired. Six studies published within the last ten years compare the cost of prostate cancer-specific healthcare use. The study conducted by Perez et al. (32) was the only study to examine cost beyond one year. However, in contrast to our study, the estimates by Perez et al. for costs beyond the initial treatment phase were based on a model developed using data from the literature rather than patient-level data. Therefore, Perez et al. were not able to test whether the difference in 5-year cost was statistically significant. Perez et al. found the 5-year cost of an initial treatment strategy using prostatectomy to be $140 to $1,230 less than treatment with external beam radiotherapy. Our study also found prostatectomy to be more expensive, but the cost differential identified in our study was larger ($3260). While our study focused on 5-year costs, some comparisons to the five studies published within the last ten years examining differences in the first year cost of prostate cancer treatment strategies can be made. Both our study and the study by Penson et al. (31) found the first year cost of prostatectomy to be greater than expectant management. Penson et al. found prostatectomy $6,840 more than expectant management, and we found prostatectomy to cost an additional $10,020. Penson et al. found external beam radiotherapy to be $6,950 more than expectant management. Similarly, we found

PAGE 90

79 treatment with external beam radiotherapy to be $6,360 more than treatment with expectant management. Comparing external beam radiotherapy to brachytherapy, Brandeis et al. (29) found a non-significant $640 difference. We found external beam radiotherapy cost $640 more than brachytherapy, but we did not test whether the difference was statistically significant. Though not statistically significant, external beam radiotherapy was also the more expensive treatment in the study by Brandeis and colleagues. In the same study, Brandeis et al. (29) found external beam radiotherapy with brachytherapy to be $9,110 more than brachytherapy. Our study also found external beam radiotherapy with brachytherapy to be more expensive than brachytherapy but by a smaller amount ($4,180). Comparing external beam radiotherapy with brachytherapy to prostatectomy, Brandeis et al. found external beam radiotherapy with brachytherapy to be $5,390 more expensive while our study again found a smaller difference ($750). Comparing external beam radiotherapy with brachytherapy to external beam radiotherapy, Brandeis et al. found external beam radiotherapy with brachytherapy to be $8,470 more than external beam radiotherapy. Again, we found a similar but smaller difference ($3,160). Three of the previously published studies compared the cost of external beam radiotherapy to that of prostatectomy. Burkardt et al. (30) and Brandeis et al. (29) found prostatectomy to be more expensive by $3710 and $3080, respectively. Our study found prostatectomy to be $5,150 more than external beam radiotherapy. Penson et al. (31) did not find a difference in the cost of prostatectomy and external beam radiotherapy.

PAGE 91

80 Three of the previously published studies also compared the cost of brachytherapy and prostatectomy. Similar to our study in which brachytherapy was $3,560 more than prostatectomy, Wagner et al. (33) found brachytherapy to be more expensive than prostatectomy but by a smaller amount ($5,930). In contrast, Brandeis et al. (29) found prostatectomy to cost $3,720 more than brachytherapy. Kohan et al. (35) failed to find a difference in the cost of prostatectomy and brachytherapy. Conclusions Our study identified a number of differences in the treatments recommended in the NCCN prostate cancer clinical practice guideline. First, statistically significant differences in the 5-year cost of prostate cancer-specific healthcare were found in each comparison of recommended treatments. The differences in 5-year cost ranged between $630 (comparing external beam radiotherapy to brachytherapy) and $6740 (comparing external beam radiotherapy with brachytherapy to expectant management). The final determination of the degree to which these differences in cost are meaningful is left to the reader. In terms of clinical effectiveness, the findings regarding differences in the risk of death due to prostate cancer comparing external beam radiotherapy to expectant management, prostatectomy to expectant management, and prostatectomy to external beam radiotherapy are significant contributions to the literature. However, one must consider the fact that the absolute differences in survival were small when evaluating the clinical significance of the statistically significant differences in prostate cancer-specific mortality between the treatments. Finally, we provided valuable information regarding the cost of preventing one death due to prostate cancer over a 5-year time period associated with the use of external beam radiotherapy rather than expectant management

PAGE 92

81 ($600,000), prostatectomy rather than expectant management ($646,670), and prostatectomy rather than external beam radiotherapy ($465,710).

PAGE 93

LIST OF REFERENCES 1. The National Guidelines Clearinghouse. Fact Sheet. AHRQ Publication No.00-0047, July 2000. Agency for Healthcare Research and Quality, Rockville, MD. http://www.ahrq.gov/clinic/ngvfact.htm. Last accessed: April 2005. 2. Institute of Medicine. Guidelines for Clinical Practice: From Development to Use. Washington, D.C.: National Academy Press, 1992. 3. American Cancer Society. Cancer Facts & Figures 2004. American Cancer Society, Atlanta, GA. http://www.cancer.org/downloads/STT/CAFF_finalPWSecured.pdf. Last accessed: April 2005. 4. NCCN Prostate Cancer Panel. Clinical Practice Guidelines in Oncology Prostate Cacer. January 2004. National Comprehensive Cancer Network, Jenkintown, PA. http://www.nccn.org/professions/physcian_gls/PDF/prostate/pdf. Last accessed: December 2004. 5. Mason J, Eccles M, Freemantle N, Drummond M. A framework for incorporating cost-effectiveness in evidence-based clinical practice guidelines. Health Policy 1999;47:37-52. 6. Perrotti M, Rabbani F, Farkas A, Ward WS, Cummings KB. Trends in poorly differentiated prostate cancer 1973 to 1994: observations from the Surveillance, Epidemiology and End Results database. J Urol 1998;160:811-5. 7. Fowler JE, Jr., Braswell NT, Pandey P, Seaver L. Experience with radical prostatectomy and radiation therapy for localized prostate cancer at a Veterans Affairs Medical Center. J Urol 1995;153:1026-31. 8. Lu-Yao GL, Yao SL. Population-based study of long-term survival in patients with clinically localised prostate cancer. Lancet 1997;349:906-10. 9. Kanamaru H, Arai Y, Akino H, Suzuki Y, Oyama N, Yoshida H, Okada K. Long-term treatment results of elderly patients with prostate cancer in Japan: an analysis of prognostic factors. Jpn J Clin Oncol 1999;29:151-5. 10. Khan MA, Partin AW. Partin tables: past and present. BJU Int 2003;92:7-11. 11. Hernandez J, Thompson IM. Prostate-specific antigen: a review of the validation of the most commonly used cancer biomarker. Cancer 2004;101:894-904. 82

PAGE 94

83 12. AJCC Cancer Staging Manual. New York: Springer-Verlag, 2002. 13. Epstein JI, Potter SR. The pathological interpretation and significance of prostate needle biopsy findings: implications and current controversies. J Urol 2001;166:402-10. 14. Abdalla I, Basu A, Hellman S. An evidence-based analysis of the management of localized prostate cancer. Cancer J 2002;8:40-6. 15. Jani AB, Hellman S. Early prostate cancer: clinical decision-making. Lancet 2003;361:1045-53. 16. Eng TY, Thomas CR, Herman TS. Primary radiation therapy for localized prostate cancer. Urol Oncol 2002;7:239-57. 17. Norderhaug I, Dahl O, Hoisaeter PA, Heikkila R, Klepp O, Olsen DR, Kristiansen IS, Waehre H, Bjerklund Johansen TE. Brachytherapy for prostate cancer: a systematic review of clinical and cost effectiveness. Eur Urol 2003;44:40-6. 18. Holmberg L, Bill-Axelson A, Helgesen F, Salo JO, Folmerz P, Haggman M, Andersson SO, Spangberg A, Busch C, Nordling S, Palmgren J, Adami HO, Johansson JE, Norlen BJ. A randomized trial comparing radical prostatectomy with watchful waiting in early prostate cancer. N Engl J Med 2002;347:781-9. 19. Parker CC, Dearnaley DP. Radical radiotherapy for prostate cancer. Cancer Treat Rev 2003;29:161-9. 20. Kupelian PA, Elshaikh M, Reddy CA, Zippe C, Klein EA. Comparison of the efficacy of local therapies for localized prostate cancer in the prostate-specific antigen era: a large single-institution experience with radical prostatectomy and external-beam radiotherapy. J Clin Oncol 2002;20:3376-85. 21. Barry MJ, Albertsen PC, Bagshaw MA, Blute ML, Cox R, Middleton RG, Gleason DF, Zincke H, Bergstralh EJ, Jacobsen SJ. Outcomes for men with clinically nonmetastatic prostate carcinoma managed with radical prostactectomy, external beam radiotherapy, or expectant management: a retrospective analysis. Cancer 2001;91:2302-14. 22. Carter CA, Donahue T, Sun L, Wu H, McLeod DG, Amling C, Lance R, Foley J, Sexton W, Kusuda L, Chung A, Soderdahl D, Jackmaan S, Moul JW. Temporarily deferred therapy (watchful waiting) for men younger than 70 years and with low-risk localized prostate cancer in the prostate-specific antigen era. J Clin Oncol 2003;21:4001-8. 23. Klein EA, Kupelian PA. Localized prostate cancer. Curr Treat Options Oncol 2000;1:433-45.

PAGE 95

84 24. Cadeddu JA, Partin AW, Epstein JI, Walsh PC. Stage D1 (T1-3, N1-3, M0) prostate cancer: a case-controlled comparison of conservative treatment versus radical prostatectomy. Urology 1997;50:251-5. 25. D'amico AV, Moul J, Carroll PR, Sun L, Lubeck D, Chen MH. Cancer-specific mortality after surgery or radiation for patients with clinically localized prostate cancer managed during the prostate-specific antigen era. J Clin Oncol 2003;21:2163-72. 26. Merrick GS, Butler WM, Lief JH, Dorsey AT. Is brachytherapy comparable with radical prostatectomy and external-beam radiation for clinically localized prostate cancer? Tech Urol 2001;7:12-9. 27. Schmid HP, Adolfsson J, Aus G. Active monitoring (deferred treatment or watchful waiting) in the treatment of prostate cancer. A review. Eur Urol 2001;40:488-94. 28. Peschel RE, Colberg JW. Surgery, brachytherapy, and external-beam radiotherapy for early prostate cancer. Lancet Oncol 2003;4:233-41. 29. Brandeis J, Pashos CL, Henning JM, Litwin MS. A nationwide charge comparison of the principal treatments for early stage prostate carcinoma. Cancer 2000;89:1792-9. 30. Burkhardt JH, Litwin MS, Rose CM, Correa RJ, Sunshine JH, Hogan C, Hayman JA. Comparing the costs of radiation therapy and radical prostatectomy for the initial treatment of early-stage prostate cancer. J Clin Oncol 2002;20:2869-75. 31. Penson DF, Schonfeld WH, Flanders SC, Henke CJ, Warolin KL, Carroll PR, Litwin MS. Relationship of first-year costs of treating localized prostate cancer to initial choice of therapy and stage at diagnosis: results from the CAPSURE database. Urology 2001;57:499-503. 32. Perez CA, Michalski J, Ballard S, Drzymala R, Kobeissi BJ, Lockett MA, Wasserman TH. Cost benefit of emerging technology in localized carcinoma of the prostate. Int J Radiat Oncol Biol Phys 1997;39:875-83. 33. Wagner TT, III, Young D, Bahnson RR. Charge and length of hospital stay analysis of radical retropubic prostatectomy and transperineal prostate brachytherapy. J Urol 1999;161:1216-8. 34. Ciezki JP, Klein EA, Angermeier KW, Ulchaker J, Zippe CD, Wilkinson DA. Cost comparison of radical prostatectomy and transperineal brachytherapy for localized prostate cancer. Urology 2000;55:68-72. 35. Kohan AD, Armenakas NA, Fracchia JA. The perioperative charge equivalence of interstitial brachytherapy and radical prostatectomy with 1-year followup. J Urol 2000;163:511-4.

PAGE 96

85 36. Grover SA, Coupal L, Zowall H, Rajan R, Trachtenberg J, Elhilali M, Chetner M, Goldenberg L. The clinical burden of prostate cancer in Canada: forecasts from the Montreal Prostate Cancer Model. CMAJ 2000;162:977-83. 37. Grover SA, Coupal L, Zowall H, Rajan R, Trachtenberg J, Elhilali M, Chetner M, Goldenberg L. The economic burden of prostate cancer in Canada: forecasts from the Montreal Prostate Cancer Model. CMAJ 2000;162:987-92. 38. Institute of Medicine. Clinical Practice Guidelines: Directions for a New Program. Washington, DC: National Academy Press, 1990. 39. Grilli R, Magrini N, Penna A, Mura G, Liberati A. Practice guidelines developed by specialty societies: the need for a critical appraisal. Lancet 2000;355:103-6. 40. Shaneyfelt TM, Mayo-Smith MF, Rothwangl J. Are guidelines following guidelines? The methodological quality of clinical practice guidelines in the peer-reviewed medical literature. JAMA 1999;281:1900-5. 41. Cape J, Barkham M. Practice improvement methods: Conceptual base, evidence-based research, and practice-based recommendations. British Journal of Clinical Psychology 2002;41:285-307. 42. Graham ID, Calder LA, Hebert PC, Carter AO, Tetroe JM. A comparison of clinical practice guideline appraisal instruments. Int J Technol Assess Health Care 2000;16:1024-38. 43. Eccles M, Clapp Z, Grimshaw J, Adams PC, Higgins B, Purves I, Russell I. North of England evidence based guidelines development project: methods of guideline development. BMJ 1996;312:760-2. 44. Graham ID, Beardall S, Carter AO, Glennie J, Hebert PC, Tetroe JM, McAlister FA, Visentin S, Anderson GM. What is the quality of drug therapy clinical practice guidelines in Canada? CMAJ 2001;165:157-63. 45. Natsch S, van der Meer JWM. The role of clinical guidelines, policies and stewardship. Journal of Hospital Infection 2003;53:172-6. 46. Chalmers I, Haynes B. Reporting, updating, and correcting systematic reviews of the effects of health care. BMJ 1994;309:862-5. 47. Rycroft-Malone J. Formal consensus: the development of a national clinical guideline. Quality in Health Care 2001;10:238-44. 48. Cates JR, Young DN, Guerriero DJ, Jahn WT, Armine JP, Korbett AB, Bowerman DS, Porter RC, Sandman TD, King RA. Evaluating the quality of clinical practice guidelines. J Manipulative Physiol Ther 2001;24:170-6.

PAGE 97

86 49. Guideline development in Europe. An international comparison. Int J Technol Assess Health Care 2000;16:1039-49. 50. Burgers JS, Cluzeau FA, Hanna SE, Hunt C, Grol R. Characteristics of high-quality guidelines: evaluation of 86 clinical guidelines developed in ten European countries and Canada. Int J Technol Assess Health Care 2003;19:148-57. 51. Browman GP. Improving clinical practice guidelines for the 21st century. Attitudinal barriers and not technology are the main challenges. Int J Technol Assess Health Care 2000;16:959-68. 52. Shekelle PG, Woolf SH, Eccles M, Grimshaw J. Clinical guidelines: developing guidelines. BMJ 1999;318:593-6. 53. Grol R, Dalhuijsen J, Thomas S, Veld C, Rutten G, Mokkink H. Attributes of clinical guidelines that influence use of guidelines in general practice: observational study. BMJ 1998;317:858-61. 54. Burgers JS, Grol R, Klazinga NS, Makela M, Zaat J. Towards evidence-based clinical practice: an international survey of 18 clinical guideline programs. Int J Qual Health Care 2003;15:31-45. 55. Cook D, Giacomini M. The trials and tribulations of clinical practice guidelines. JAMA 1999;281:1950-1. 56. Fletcher SW, Fletcher RH. Development of clinical guidelines. Lancet 1998;352:1876. 57. Cluzeau FA, Littlejohns P, Grimshaw JM, Feder G, Moran SE. Development and application of a generic methodology to assess the quality of clinical guidelines. Int J Qual Health Care 1999;11:21-8. 58. Scherr D, Swindle PW, Scardino PT. National Comprehensive Cancer Network guidelines for the management of prostate cancer. Urology 2003;61:14-24. 59. Lin DY. Linear regression analysis of censored medical costs. Biostatistics 2000;1:35-47. 60. Lin DY, Feuer EJ, Etzioni R, Wax Y. Estimating medical costs from incomplete follow-up data. Biometrics 1997;53:113-28. 61. Zhao H, Tian L. On estimating medical cost and incremental cost-effectiveness ratios with censored data. Biometrics 2001;57:1002-8. 62. O'Hagan A, Stevens JW. On estimators of medical costs with censored data. J Health Econ 2004;23:615-25.

PAGE 98

87 63. Willan AR, Lin DY. Incremental net benefit in randomized clinical trials. Stat Med 2001;20:1563-74. 64. Willan AR, Lin DY, Cook RJ, Chen EB. Using inverse-weighting in cost-effectiveness analysis with censored data. Stat Methods Med Res 2002;11:539-51. 65. Bang H, Tsiatis AA. Estiamteing medical costs with censored data. Biometrika 2000;87:329-43. 66. Etzioni RD, Feuer EJ, Sullivan SD, Lin D, Hu C, Ramsey SD. On the use of survival analysis techniques to estimate medical care costs. J Health Econ 1999;18:365-80. 67. Willan AR, Lin DY, Manca A. Regression methods for cost-effectiveness analysis with censored data. Stat Med 2005;24:131-45. 68. Willan AR. Analysis, sample size, and power for estimating incremental net health benefit from clinical trial data. Control Clin Trials 2001;22:228-37. 69. Willan AR, O'Brien BJ. Confidence intervals for cost-effectiveness ratios: an application of Fieller's theorem. Health Econ 1996;5:297-305. 70. Stinnett AA, Mullahy J. Net health benefits: a new framework for the analysis of uncertainty in cost-effectiveness analysis. Med Decis Making 1998;18:S68-S80. 71. Briggs A, Fenn P. Confidence intervals or surfaces? Uncertainty on the cost-effectiveness plane. Health Econ 1998;7:723-40. 72. Warren JL, Klabunde CN, Schrag D, Bach PB, Riley GF. Overview of the SEER-Medicare data: content, research applications, and generalizability to the United States elderly population. Med Care 2002;40:IV-3-IV-18. 73. Cepeda MS, Boston R, Farrar JT, Strom BL. Comparison of logistic regression versus propensity score when the number of events is low and there are multiple confounders. Am J Epidemiol 2003;158:280-7. 74. Kaplan EL, Meier P. Nonparametric-Estimation from Incomplete Observations. Journal of the American Statistical Association 1958;53:457-81. 75. Bach PB, Guadagnoli E, Schrag D, Schussler N, Warren JL. Patient demographic and socioeconomic characteristics in the SEER-Medicare database applications and limitations. Med Care 2002;40:IV-19-IV-25. 76. Brown ML, Riley GF, Schussler N, Etzioni R. Estimating health care costs related to cancer treatment from SEER-Medicare data. Med Care 2002;40:IV-104-IV-117.

PAGE 99

BIOGRAPHICAL SKETCH Michael D. Taylor was raised in Chipley, Florida before moving to Tallahassee, Florida, where he graduated magna cum laude from Leon High School in 1995. He enrolled at the University of Florida in 1995 as a pre-pharmacy student and participated in the Honors Program. Michael received a Doctor of Pharmacy with High Honors from the University of Florida in 2001. As a graduate student in the Pharmacy Health Care Administration Department, Michael received the Rho-Chi Schering Plough American Foundation for Pharmaceutical Education First-Year Graduate Scholarship, a University of Florida Alumni Fellowship, and an American Foundation for Pharmaceutical Education Pre-Doctoral Fellowship. 88