In a large, prospectively accrued cohort of men with prostate cancer, disease risk as measured by the CAPRA score does appear to be a major driver of treatment (), though the patterns observed are of some concern for both overtreatment of low-risk disease—identified in prior analyses of CaPSURE and other data sources14,15
—and undertreatment of high-risk localized disease.16
More men with low-risk disease should be candidates for active surveillance, whereas most with high-risk but clinically localized disease should be offered a chance at cure with multimodal therapy, including either radiation or surgery as the primary local modality.5
Interestingly, trends over time () suggest that overtreatment of low-risk disease may be ameliorating slightly, whereas potential undertreatment of high-risk disease appears to be a growing concern.
Aside from these observed trends over time and risk, this analysis confirms wide variation in practice patterns across clinical sites (). Indeed, a substantial proportion of variation in primary treatment selection for localized prostate cancer is attributable to practice site. The degrees of variation observed were similar when the analysis was restricted to men with low-risk disease only. These findings suggest that factors other than cancer risk and patient clinical and sociodemographic factors influence treatment decision making. In fact, treatment of localized prostate cancer is a model of what has been termed “preference-sensitive” health care, in which patient or clinician preferences, beliefs, or values drive decision making in the absence of strong scientific evidence.7
Indeed, clinicians and patients making decisions regarding localized prostate cancer do so in the setting of a relative dearth of high-quality data comparing outcomes following the various available treatments.4
Only one randomized trial of substantial size and quality has been reported,17
finding a survival benefit for prostatectomy over watchful waiting; however, the generalizability of the findings to contemporary patients, and to subgroups of various ages and levels of risk, has been the subject of ongoing debate. The Prostate Cancer Intervention Versus Observation Trial (PIVOT) study, also randomly assigning patients to surgery versus watchful waiting, completed accrual; results are expected during 2010.18
However, no trials that randomly assign patients among active treatment modalities have yet been completed. The Surgical Prostatectomy Versus Interstitial Radiation Intervention Trial (SPIRIT), intended to compare surgery with brachytherapy, closed because of poor accrual.19
The Prostate Testing for Cancer and Treatment (ProtecT) study successfully randomized men in the United Kingdom to a three-arm trial comparing prostatectomy, radiation, and surveillance, though results will not be available until the middle of the next decade.20
Previous analyses have documented substantial local variation for specific prostate cancer treatments. Using Medicare data from the mid-1990s, for example, the Dartmouth Atlas of Health Care project analyzed the 10 most commonly performed surgical procedures in the United States, including radical prostatectomy. Findings show that among the 10 procedures, prostatectomy was characterized by the greatest local variation: more than 12-fold greater than the procedure (hip fracture repair) with the least variation and more than eight-fold greater than colectomy for colon cancer. The absolute rates of prostatectomy, adjusted for prevalence of prostate cancer, varied by a factor of nearly 10, from 0.5 to 4.7 per 1,000 Medicare enrollees.8
Another recently published study focused on the use of androgen deprivation therapy, based on 1990s data from the Surveillance, Epidemiology, and End Results (SEER)–Medicare linked database. The investigators performed separate analyses for evidence-based androgen deprivation therapy—that is, therapy given together with radiation therapy for high-risk disease (stage T3 or stage T2/poorly differentiated)—and uncertain-benefit therapy, including all other uses. For the evidence-based setting, they found, based on ICC calculations, that disease characteristics accounted for 6.6% of variation, other patient characteristics explained 7.3%, and the treating urologist accounted for 25.4% of variation. For the uncertain benefit setting, the corresponding proportions were 5.3%, 5.0%, and 22.7%. From the early to the late part of the decade, the proportion of variation attributable to the urologist appeared to be increasing.9
The major limitation of these studies, aside from the somewhat older data used for analysis, was the lack of adequate cancer risk assessment available in the data sets analyzed: Medicare has no data on risk factors, and the 1990s-era SEER-Medicare data had only stage and differentiation but not PSA levels or Gleason scores. In addition, analyses based on Medicare data by definition are restricted to patients older than age 65 years. The present analysis of treatment variation comprises updated data, including all of the major treatment approaches for localized prostate cancer, and incorporates a robust, well-validated risk assessment score for case-mix adjustment.
The proportions of variation attributable to clinical site for androgen deprivation—13% and 14% for primary and neoadjuvant therapy, respectively—are slightly lower than those observed in the SEER-Medicare analysis,9
perhaps reflecting better control for patient and tumor factors in CaPSURE. What is striking, however, is the fact that site-specific variation was higher for every other treatment modality than for androgen ablation therapy (). Use of prostatectomy varied from 11% to 82%, nearly as great a range as the 10-fold variation observed in the Dartmouth Atlas of Health Care project.8
Explanations for the observed variation are speculative, but presumably reflect variable physician training, experience, and personal outcomes; payor mix, reimbursement patterns, and other financial incentives; the local medicolegal environment; uneven penetrance of novel technologies; impact of local culture on patient beliefs and preferences; and many other factors. An aggressive local or personal philosophy with respect to prostate cancer screening may also correspond to an increased tendency toward treatment, even controlling for varying stage at presentation. A similar observation has been made, for example, in correlating rates of coronary angiography with angioplasty and/or bypass surgery.8
The protracted time course of prostate cancer, the proliferation of treatment options, and the nature of the potential complications of treatment all likely contribute additionally to the problem in the case of prostate cancer.
Certainly the absence of high-quality comparative effectiveness data, along with controversy regarding interpretation of the data that do exist, creates a fertile substrate in which variation would be expected to thrive. Even where high-quality evidence exists, however, incorporation into clinical practice may be variable. Randomized trials supporting the use of androgen deprivation together with radiation therapy, for example, were reported in the late 1990s, and those finding no benefit for androgen deprivation given before prostatectomy were reported in the early 2000s.21
Within a few years, the use of androgen deprivation therapy with external-beam radiation for high-risk patients rose to 85%, whereas use of such therapy before prostatectomy appeared to be rising to more than 10% by the mid- to late-2000s.16,21
Caveats to this analysis must be noted, the most important of which is that although the practices in CaPSURE represent a range of practice sizes and geographic locations, they were not chosen at random and do not represent a statistically valid sample of the US population. We have previously compared the CaPSURE population to the SEER population of prostate cancer patients and found that the median age among patients is similar, though white men are relatively over-represented in CaPSURE compared with the general population.13
Previous studies have found that race and ethnicity have little impact on degrees of variation in health care.22
Men in CaPSURE also have slightly higher socioeconomic status on average than the overall population.11
It is reassuring that the findings with respect to variation in use of prostatectomy and androgen deprivation therapy are generally consistent with the previous analyses from Medicare and SEER-Medicare; certainly there is no reason to suspect that the current analysis underestimates the true extent of variation at the population level.
Data in CaPSURE are submitted only by patients and urologists; thus any treatments by other clinicians that are not reported by patients either to their urologists or in their resource utilization questionnaires may be missed. Extant quality assurance mechanisms, including medical records review of all hospital admissions, is expected to minimize this problem. Despite these cautionary notes, we believe our data provide the best available description of case-mix adjusted national practice patterns.
Examination of treatment patterns in a large, national disease registry confirms substantial practice-level variation in management of localized prostate cancer that cannot be explained by disease case-mix variability. A growing body of evidence suggests that improved decision support may not only improve decision quality23
and reduce decisional regret24
but may also be a means to reduce unwarranted variation in health care.7
Incorporation of such decision support into clinical practice, while challenging, should be a priority. Even more important is the need for better data on outcomes of prostate cancer treatment. The Institute of Medicine recently included treatment for localized prostate cancer among the 25 most important topics for future comparative effectiveness research.25
Only through such research, based on both prospective clinical trials and retrospective review of high-quality, clinically rich data sources, will clinicians and patient be able to assess more accurately the relative merits, risks, and costs of treatment alternatives and, by extension, to reduce variation in their selections among these alternatives.