PADT indicates primary androgen deprivation therapy. Error bars represent 95% confidence intervals (CIs) at 5 and 10 years. For cancer-specific survival, results were adjusted for age, race, income, marital status, urban residence, comorbidity status, year of diagnosis, and cancer stage. Prostate cancer–specific survival was lower in high- vs low-PADT use areas among men with moderately differentiated cancer (bootstrap P < .001). Prostate cancer–specific survival was borderline statistically different between high- and low-PADT use areas among men with poorly differentiated cancer (bootstrap P = .049). Overall survival was similar in high- and low-PADT use areas among men with moderately differentiated cancer; median overall survival was 89 and 90 months for high- and low-use areas, respectively (bootstrap P = .67). Difference in overall survival between high- and low-PADT use areas among men with poorly differentiated cancer did not reach statistical significance (bootstrap P = .13). Median overall survival was 57 and 54 months for high- and low-PADT use areas, respectively. The difference in median overall survival between high- and low-PADT use areas was 3 months (95% CI, −1 to 7 months).
Customize your JAMA Network experience by selecting one or more topics from the list below.
Lu-Yao GL, Albertsen PC, Moore DF, et al. Survival Following Primary Androgen Deprivation Therapy Among Men With Localized Prostate Cancer. JAMA. 2008;300(2):173–181. doi:10.1001/jama.300.2.173
Author Affiliations: Department of Environmental and Occupational Medicine, Robert Wood Johnson Medical School (Dr Lu-Yao), and Department of Biostatistics, The School of Public Health (Drs Moore, Shih, and Lin), University of Medicine and Dentistry of New Jersey, Piscataway; Department of Medicine, University of Medicine and Dentistry of New Jersey, Robert Wood Johnson Medical School, New Brunswick (Drs Lu-Yao, DiPaola, and Yao); Cancer Institute of New Jersey, New Brunswick (Drs Lu-Yao, Moore, Shih, Lin, DiPaola, and Yao); Department of Surgery (Urology), University of Connecticut, Farmington (Dr Albertsen); and The Dean and Betty Gallo Prostate Cancer Center, New Brunswick, New Jersey (Drs Lu-Yao and DiPaola).
Context Despite a lack of data, increasing numbers of patients are receiving primary androgen deprivation therapy (PADT) as an alternative to surgery, radiation, or conservative management for the treatment of localized prostate cancer.
Objective To evaluate the association between PADT and survival in elderly men with localized prostate cancer.
Design, Setting, and Patients A population-based cohort study of 19 271 men aged 66 years or older receiving Medicare who did not receive definitive local therapy for clinical stage T1-T2 prostate cancer. These patients were diagnosed in 1992-2002 within predefined US geographical areas, with follow-up through December 31, 2006, for all-cause mortality and through December 31, 2004, for prostate cancer–specific mortality. Instrumental variable analysis was used to address potential biases associated with unmeasured confounding variables.
Main Outcome Measures Prostate cancer–specific survival and overall survival.
Results Among patients with localized prostate cancer (median age, 77 years), 7867 (41%) received PADT, and 11 404 were treated with conservative management, not including PADT. During the follow-up period, there were 1560 prostate cancer deaths and 11 045 deaths from all causes. Primary androgen deprivation therapy was associated with lower 10-year prostate cancer–specific survival (80.1% vs 82.6%; hazard ratio [HR], 1.17; 95% confidence interval [CI], 1.03-1.33) and no increase in 10-year overall survival (30.2% vs 30.3%; HR, 1.00; 95% CI, 0.96-1.05) compared with conservative management. However, in a prespecified subset analysis, PADT use in men with poorly differentiated cancer was associated with improved prostate cancer–specific survival (59.8% vs 54.3%; HR, 0.84; 95% CI, 0.70-1.00; P = .049) but not overall survival (17.3% vs 15.3%; HR, 0.92; 95% CI, 0.84-1.01).
Conclusion Primary androgen deprivation therapy is not associated with improved survival among the majority of elderly men with localized prostate cancer when compared with conservative management.
Prostate cancer is the most common nonskin cancer and the second most common cause of cancer death among men.1 For the majority of men with incident prostate cancer (approximately 85%), disease is diagnosed at localized (T1-T2) stages,2 and standard treatment options include surgery, radiation, or conservative management (ie, deferral of treatment until necessitated by disease signs or symptoms).
Although not standard or sanctioned by major groups or guidelines, an increasing number of clinicians and patients have turned to primary androgen deprivation therapy (PADT) as an alternative to surgery, radiation, or conservative management, especially among older men.3,4 For example, in a 1999-2001 survey, PADT had become the second most common treatment approach, after surgery, for localized prostate cancer.3
Randomized clinical trials support the use of early androgen deprivation therapy (ADT) as an adjunct to surgery or radiation for patients with high-risk cancer.5-10 In 1 trial,5,8 early ADT reduced mortality by approximately 50% when used with radiation in high-risk disease (poorly differentiated T1-T2 or T3-T4); whereas, in another trial,9 mortality was reduced by approximately 60% in patients with nodal disease identified at surgery. Consequently, many investigators have concluded that the early use of ADT is appropriate for patients with higher-risk or intermediate-risk disease in conjunction with local therapy, but studies that assess the use of ADT alone, as primary therapy, or in lower-risk settings are sparse.
The importance of determining the appropriate application of ADT has recently increased, because a growing body of literature now demonstrates that chronic ADT use has been associated with approximately 10% to 50% increases in the risks of fracture, diabetes, coronary heart disease, myocardial infarction, and sudden cardiac death, in addition to adverse effects on fat mass, cholesterol, and quality of life.11-16 In the Prostate Cancer Outcomes Study (PCOS),17 the risk of gynecomastia and hot flashes increased 500% and a 267% increase in impotence was observed after 1 year of treatment. In addition, medical ADT is costly. The costs associated with ADT medication use in the United States reached $1.2 billion in 200318 and ADT drugs represented the second highest Medicare Part B drug expenditure.
A randomized clinical trial would provide the data needed to determine the usefulness of PADT vs conservative management in localized disease. However, because of the relatively indolent nature of most cases of localized prostate cancer, such a trial would take more than a decade to complete and, given current treatment practices and resources, would probably not be feasible. Observational studies are often used to provide insight under such circumstances, although they may be more subject to biases.
Instrumental variable analysis (IVA) techniques have been applied successfully to observational medical studies19,20 to help minimize many of these biases so that the results of randomized clinical trials may often be mimicked with observational data.21 Instrumental variable analysis is a method of capturing the random component of patient treatment choice and using it to balance treatment groups with respect to measured and unmeasured confounders. We used this approach to assess the association between PADT and disease-specific survival and overall survival in men with T1-T2 prostate cancer.
Data were obtained from the population-based Surveillance, Epidemiology, and End Results (SEER) program database and linked Medicare files. The SEER regions encompass approximately 14% of the US population before 2001 and 26% thereafter. The Medicare database covers approximately 97% of US persons aged 65 years or older, and linkage to the SEER database was complete for approximately 93% of the patients.22 The study received institutional review board approval from the University of Medicine and Dentistry of New Jersey, as well as the SEER program, and the Center for Medicare & Medicaid Services. Informed consent was waived by the institutional review board because the data did not contain personal identifiers.
The study cohort consisted of 89 877 men aged 66 years or older who were SEER residents and diagnosed with T1-T2 cancer in 1992-2002. Men who died within 180 days of diagnosis were excluded (n = 1761) (inclusion of patients dying within 180 days did not significantly alter the results). Patients receiving definitive local therapy (eg, prostatectomy or radiation) within 180 days of diagnosis were also excluded (n = 31 485). To ensure that the database accurately documented the patient's clinical course, patients without both Medicare Part A (hospitalization) and Part B (physician and outpatient) as their primary health care insurance coverage during the study period were excluded (n = 33 987). Patients with missing data (n = 2995), unknown cancer grade (n = 255), or initiation of ADT before cancer diagnosis (n = 123) were also excluded. Therefore, a total cohort of 19 271 men were included in our analysis.
Patients undergoing PADT received ADT as primary cancer therapy (eg, no surgery or radiation) during the first 180 days following diagnosis. Patients in the conservative management group were those that did not receive surgery, radiation, or PADT during this time. A previous study demonstrated that patients generally start primary therapy within 6 months of diagnosis.23 Using a previously described algorithm,11 Medicare physician, inpatient and outpatient claims were used to identify orchiectomy (Healthcare Common Procedure Coding System codes 54520, 54521, 54522, 54530, or 54535, or the International Classification of Diseases, Ninth Revision code 624) and the use of luteinizing hormone-releasing hormone agonists (Healthcare Common Procedure Coding System codes J1950, J9202, J9217, J9218, or J9219). Luteinizing hormone-releasing hormone agonists and orchiectomy were combined because previous studies have shown these treatments to be essentially equivalent.24
Overall and prostate cancer–specific survival was available through December 31, 2006, and December 31, 2004, respectively. Underlying cause of death was determined from data in the SEER records. Studies have shown that cause of death in the SEER data confirm information available in medical records in 87% to 88% of cases.25,26
Cox proportional hazards regression model covariates included age at diagnosis, race (self-determined by the patients and included as a variable because race can be associated with outcomes in prostate cancer), zip code, income, SEER region, urban area, marital status, cancer grade, clinical tumor (T) stage, Charlson comorbidity score, and year of diagnosis. Charlson score, a powerful predictor of longevity in men with localized prostate cancer27-29 was derived from Medicare claims during the year before prostate cancer diagnosis by using a validated algorithm.30,31 For cancer grade, Gleason score 2 to 4, 5 to 7, and 8 to 10 corresponded to well-differentiated, moderately differentiated, and poorly differentiated disease, respectively. We used clinical extension information provided by SEER to determine cancer stage (T1, T2).
A health service area (HSA) is defined as 1 or more counties that are relatively self-contained with respect to the provision of routine hospital care.32 The instrumental variable was constructed by first calculating the proportion of patients who received PADT in each HSA. Because some HSAs had small numbers of prostate cancer cases, each HSA with less than 50 cases was combined with the nearest (in terms of distance between geographic centers) HSA with 50 or more cases. The threshold of 50 or more cases was chosen because lower thresholds were associated with more imbalances in patient characteristics in high-PADT and low-PADT use areas. The algorithm produced 66 utilization areas. High-use and low-use areas corresponded to the top and bottom tertiles of PADT utilization and were used as the (binary) instrumental variable for the (binary) treatment assignment.
Previous studies have demonstrated that PADT use is highly influenced by nonmedical factors,33 with tumor characteristics accounting for only 9.7% of the total variance in use.34 Our data confirmed that PADT use varied widely across HSAs, a key requirement of an instrumental variable. An instrumental variable must be associated with outcomes primarily through its correlation with treatment status and not through any other independent effect. We verified this assumption by comparing baseline characteristics, including age at diagnosis, cancer stage, and grade at diagnosis, and found these factors comparable between low-PADT and high-PADT areas.
Instrumental variable analysis methods based on the Rubin Causal Model21 were used to account for both measured and unmeasured (eg, prostate-specific antigen, family history, diet, weight) confounders. Covariates in the IVA models included age, race, comorbidity status, cancer stage, cancer grade, income status, urban residence, marital status, and year of diagnosis. All IVA results were derived from the same models. We examined all the required assumptions listed above to ensure the validity of our IVA. Traditional Cox proportional hazards regression model results were also reported for comparison with the IVA results. Analyses were conducted by using SAS version 9.1 (SAS Institute Inc, Cary, North Carolina) and R version 2.7.0 (R Foundation for Statistical Computing, Vienna, Austria). Cancer grade was a predefined measured covariate. We calculated PADT utilization for each cancer grade so that it was not necessary to assume that the patterns of PADT utilization were the same for all cancer grades within the same area. Results for well-differentiated cancer (Gleason score 2-4) are not shown separately because results were unstable due to the limited sample size.
Patients who differ in the likelihood of receiving PADT are compared and the treatment effect on the marginal population is estimated. The marginal effect (local average treatment effect)21 of PADT was calculated as
where Hi = a geographic area in the upper tertile of PADT use, Lo = a geographic area in the lower tertile of PADT use, Pr(PADT Hi/Lo) = estimated probability of PADT use among men who had localized prostate cancer and did not have surgery or radiation as their primary cancer therapy in high/low use region, and Adjusted OutcomesHi/Adjusted OutcomesLo = estimated survival probability at a particular time (eg, 5- or 10-year survival) among men who had localized prostate cancer and did not have surgery or radiation as their primary cancer therapy in high/low use region.
To compute the population-adjusted survival curves, we substituted the population means (for continuous covariates) into the Cox proportional hazards regression model for each combination of the categorical covariates to derive an adjusted hazard function. Then, a weighted average of these adjusted hazard functions was computed with weights proportional to the numbers of patients in each class. In addition, the population-adjusted survival curve was computed from the weighted hazard function.35 Estimates of 5- and 10-year overall and cancer-specific survival for men at average risk were derived from these adjusted curves. Confidence intervals (CIs) were obtained by computing these adjusted survival curves for each of 10 000 bootstrap samples of the original data. P values and 95% CIs were derived from the bootstrap estimates. Testing was 2-sided, with α = .05. Analyses were repeated for different age groups but results were similar across age groups and the interaction between age and PADT use was not significant; therefore, all age groups were combined.
Power calculations for determining the difference in survival between high- and low-use HSAs were performed by using simulations. Overall, the study had 80% power to detect a 7% difference in overall survival between high- and low-use PADT areas.
The total cohort consisted of 19 271 men aged 66 years or older with localized prostate cancer diagnosed in 1992-2002. By definition, none of these men received definitive local therapies (eg, radiation or surgery) in the first 180 days following diagnosis; 41% received PADT. The median age of the study cohort was 77 years and the median follow-up for overall survival was 81 months. As expected, patients receiving PADT and patients receiving conservative management differed in many characteristics, suggesting that there could be differences in unmeasured characteristics that might not be adjusted for by conventional statistical methods (Table 1).
PADT utilization within 180 days varied widely across HSAs (31%-53%) (Table 2). When we extended the window for defining PADT from 180 days to 18 months, the high-use and low-use patterns remained the same. Duration of PADT use was also longer in high-use areas.
There were 1560 prostate cancer deaths and 11 045 deaths from all causes in the study cohort. Unadjusted and adjusted prostate cancer–specific survival and overall survival were worse for patients treated with PADT when analyses were conducted using a traditional Cox multivariate model (Table 3). The Cox proportional hazards regression model approach, however, is unable to adjust for unmeasured confounders and selection biases (eg, higher-risk patients may be preferentially selected for PADT, thus yielding apparently adverse outcomes for this group). When IVA was used (Table 3, Table 4, Table 5, and Figure), PADT was still associated with increased unadjusted and adjusted prostate cancer–specific mortality (hazard ratio [HR], 1.17; 95% CI, 1.03-1.33), but there was no significant associated effect on unadjusted, and adjusted median overall survival (82 months vs 82 months; HR, 1.00; 95% CI, 0.96-1.05). Results were similar when analyses were restricted to men with comorbidity scores of 0 or without other cancers, suggesting that the results were independent of comorbidity.
In preplanned analyses by cancer grade, PADT was associated with either no effect or an adverse effect on prostate cancer–specific survival and overall survival for poorly differentiated or moderately differentiated cancer, respectively, in unadjusted and adjusted Cox proportional hazards regression model analyses. Evaluation by IVA, however, revealed a borderline improvement in unadjusted and adjusted median prostate cancer–specific survival in patients with poorly differentiated cancer (Table 3 and Figure), although the associated effect on median overall survival was not significant (57 months; 95% CI, 56-61 months vs 54 months; 95% CI, 53-57 months, respectively). Similar patterns were observed for 5- and 10-year prostate cancer–specific survival and overall survival (Table 4). The marginal effect of PADT on prostate cancer–specific survival was 15.1% (95% CI, 0%-30.5%) at 5 years and 24.5% (95% CI, 0.1%-49.6%) at 10 years for men with poorly differentiated cancer (Table 5). Similar benefit, however, was not observed in men with moderately differentiated cancer (Table 5).
Most patients received PADT for extended periods. Among PADT users, only 1.1% received 1 month of treatment; whereas, 75% received PADT for at least 18 months and 50% received PADT for more than 30 months. Longer durations of PADT utilization were associated with lower overall and cancer-specific survival among 5826 PADT users who survived at least 3 years (Table 6). Similar patterns were observed for all cancer grades. Sensitivity analyses restricted to patients with comorbidity scores of 0 yielded similar results.
Despite the widespread use of PADT in localized (T1-T2) prostate cancer,3,4 there is little information regarding the clinical outcomes associated with this practice. Our study was designed to evaluate the association between PADT and prostate cancer–specific survival and overall survival in men who did not initially receive definitive therapy (eg, surgery or radiation) for localized prostate cancer.
Using IVA as one of the best available means of controlling for both measured and unmeasured confounding variables, we found no overall survival benefit for elderly men with localized prostate cancer receiving PADT. Results obtained with a traditional Cox proportional hazards regression model that adjusts only for measured confounding factors differed from those with the NA approach. These observations suggest that there is significant unaccounted residual bias associated with traditional analytical methods in this setting and that the NA approach may be particularly advantageous. In addition, 1 potential advantage of this study over clinical trials is that it includes real-world patients that would often be excluded from clinical trials, even though these patients would receive the treatment in practice.
In our study, cancer-specific survival but not overall survival appeared worse for men with lower risk cancer treated with PADT. This observation has been previously documented in a randomized controlled study of PADT in men with T0-T4 disease.36 The authors suggested several possible explanations for this finding, including competing causes of death, misclassification, and statistical variation.36 Another possibility could be that suppression of moderately or well-differentiated cells not destined to harm a patient's overall survival may allow for the establishment or overgrowth of more rapidly growing malignant clones (as observed in preclinical models)37-39 that increase the probability of death due to prostate cancer instead of a competing cause of death. As shown in the Figure, the likelihood of death from competing causes normally exceeds the risk of death from prostate cancer in this population; this balance may be altered if PADT preferentially allows for the establishment or overgrowth of a more malignant fraction of a tumor.
Our study had some limitations. The study was limited to men aged 66 years or older; therefore, the results could differ for younger men. The SEER-Medicare database does not capture information on antiandrogen use. Therefore, patients using antiandrogens only might be misclassified into the conservative management cohort and, because a previous study40 suggested that antiandrogens may result in adverse outcomes in these patients, it is possible that the conservative management group performed unusually poorly. However, previous data from another large database (CaPSURE)41 showed that the use of antiandrogens as sole treatment for localized prostate cancer is relatively uncommon (approximately 2%) and it is unlikely that this small subset could alter the outcomes of the conservative management group overall.
Just like the success of a randomized study is dependent on factors such as the attainment of a sufficient sample size to balance both measured and unmeasured characteristics in different treatment groups, the use of IVA to balance treatment group characteristics (eg, prostate-specific antigen levels, family history, diet, body mass) depends on finding a suitable, partly random, varying factor (instrumental variable) that can be used to balance treatment groups. Our instrumental variable (high- and low-PADT use HSAs) had excellent properties. However, as in randomized studies, it is possible that some unmeasured factors still may have been imbalanced between groups. Nonetheless, sensitivity analyses, using various geographic-based instruments and removing patients with other cancers or comorbidity scores of more than 0, yielded similar results and suggested that the analyses were robust. However, it is still possible that the use of an NA approach in this setting does not adequately control for unknown confounding variables; therefore, if possible, a randomized trial should be considered.
There are few data comparing PADT with conservative management, or any other established treatment option (eg, surgery or radiation), in men with localized (T1-T2, NO, M0) prostate cancer, even though the popularity of PADT has increased in this population by 2- to 3-fold in recent years.3 Published studies have generally not provided data specific for localized (T1-T2) disease and have had limited sample sizes.42,43 The largest published study describing PADT use among patients with T1-T2 disease was descriptive, noncomparative, and had limited follow-up.41 The randomized Early Prostate Cancer trial44 had a large subset of patients with T1-T2 disease but a nontraditional form of PADT (bicalutamide) was used. Results from this trial revealed a trend toward decreased overall survival in patients treated with PADT.40 The Veterans Administration Co-operative Urological Research Group study45 also used a nonconventional form of PADT (diethylstilbestrol). Results were inconsistent, with benefit in T2 disease but harm in T1 disease. In a related randomized study, European Organization for Research and Treatment of Cancer trial 30891,36 which included patients with both localized and advanced disease (eg, T0-4, N0-2), a modest overall survival benefit was found in favor of PADT but further analyses suggested that this benefit was associated with a group of patients with high-risk disease.46 Studies by the Medical Research Council,47 the European Organization for Research and Treatment of Cancer trial 30846,10 and the Swiss Group for Clinical Cancer Research48 focused on patients with more advanced disease. In general, although the designs, therapies, and settings vary significantly from our study, the findings of these previous studies are inferentially consistent with our documentation of a lack of overall benefit, and some suggestion of potential benefit in high-risk or advanced-disease subgroups.
In conclusion, our analyses suggest that PADT is not associated with improved survival among the majority of elderly men with T1-T2 prostate cancer. The significant adverse effects and costs associated with PADT, along with our finding of a lack of overall survival benefit, suggest that clinicians should carefully consider the rationale for initiating PADT in elderly patients with T1-T2 prostate cancer.
Corresponding Author: Siu-Long Yao, MD, Cancer Institute of New Jersey, Robert Wood Johnson Medical School, 195 Little Albany St, Room 5544, New Brunswick, NJ 08901 (firstname.lastname@example.org).
Author Contributions: Dr Lu-Yao had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Lu-Yao, Albertson, Yao.
Acquisition of data: Lu-Yao.
Analysis and interpretation of data: Lu-Yao, Albertson, Moore, Shih, Lin, DiPaola, Yao.
Drafting of the manuscript: Lu-Yao, Lin, Yao.
Critical revision of the manuscript for important intellectual content: Lu-Yao, Albertson, Moore, Shih, Lin, DiPaola, Yao.
Statistical analysis: Lu-Yao, Moore, Shih, Lin, Yao.
Obtained funding: Lu-Yao.
Administrative, technical, or material support: Lu-Yao, Albertson.
Study supervision: Lu-Yao, Shih, DiPaola.
Financial Disclosures: None reported.
Funding/Support:This study was supported in part by award DAMD17-01-1-0755 from the US Army Medical Research Acquisition Activity, Fort Detrick, Maryland, and by the Cancer Institute of New Jersey; award W81XWG-05-1-0235 from the Department of Defense, Ohl Foundation; and by grants R01 CA116399 from the National Cancer Institute and NCI CA-72720-10 from the Cancer Institute of New Jersey.
Role of Sponsors: The Department of Defense did not play any role in the design and conduct of the study, in the collection, management, analysis, and interpretation of the data, or in the preparation, review, or approval of the manuscript. The National Cancer Institute (NCI) sponsored the collection and management of the data, and reviewed and approved the study protocol and manuscript. However, NCI did not play any role in the analysis and interpretation of the data.
Disclaimers: This study used the Linked SEER-Medicare Database. The interpretation and reporting of these data are the sole responsibility of the authors. The content of the information does not necessarily reflect the position or the policy of the US Government, and no official endorsement should be inferred.
Additional Contributions: We thank the Applied Research Branch, Division of Cancer Prevention and Population Science, National Cancer Institute; Office of Information Services and Office of Strategic Planning, Centers for Medicare & Medicaid Services; Information Management Services Inc; and the SEER program tumor registries in the creation of the SEER-Medicare database. Thanusha Puvananayagam, MPH (Cancer Institute of New Jersey assistant staff) provided outstanding administrative and technical assistance. Ms Puvananayagam did not receive any compensation.