In this study, the CAPRA score was shown to be an accurate predictor of metastasis, cancer-specific mortality, and all-cause mortality across a variety of primary treatment approaches. The strengths of the associations between CAPRA score and metastasis or cancer-specific mortality were similar to those for pathological and biochemical endpoints as calculated in earlier studies. In these studies (6
), the risk of biochemical recurrence roughly doubled with each 2-point increase in CAPRA score; the present analysis demonstrated a similar increase in the risk of metastasis (HR for metastasis = 1.47) and cancer-specific mortality (HR for death = 1.39), again consistent with a doubling of risk with each 2-point increase in score. The smaller incremental increase in risk for all-cause mortality (HR for death = 1.13) was expected, given the impact of patient age and multiple competing causes of mortality among men diagnosed with prostate cancer (14
The accuracy of the CAPRA score for prediction of all three endpoints in this CaPSURE cohort (c
-index = 0.78, 0.80, and 0.71 for bone metastases, prostate cancer–specific mortality, and all-cause mortality, respectively) was markedly superior to the accuracy in the original development study for the biochemical recurrence endpoint (c
-index = 0.66) (6
). The accuracy was somewhat lower among patients who were treated with radiation therapy (c
-index = 0.68) than among those treated with radical prostatectomy (c
-index = 0.72) or primary androgen deprivation therapy (c
-index = 0.79), which likely reflects the heterogeneity of radiation dose and technique over the years and over the multiple treatment sites represented in the CaPSURE registry.
Counseling men with a new diagnosis of prostate cancer entails many challenges, including presentation of realistic likelihoods of disease progression and mortality. These likelihoods, together with patient comorbidity, life expectancy, and preferences for treatment, should help guide planning of a risk-adapted treatment strategy. Men with low-risk prostate cancer are now eligible for at least a trial period of active surveillance at a growing number of institutions (15
). Men with low- to intermediate-risk disease are well managed by local monotherapy, while those with higher risk disease generally require aggressive multimodal treatment. Finally, men with high-risk tumors are treated systemically for presumptive micrometastatic disease and/or, ideally, should be offered clinical trial enrollment, given the high rates of recurrence and progression with extant standard therapies.
The menu of instruments to help guide decision making has grown rapidly in the 10 years since publication of the original preoperative nomogram by Kattan et al. (16
), to 111 instruments for various prostate cancer scenarios by one recent count (4
). Most instruments intended for use at time of diagnosis predict biochemical recurrence after one specific form of treatment—for example, radical prostatectomy, external beam radiotherapy, or brachytherapy (4
). However, most have not been well validated, and comparison across instruments is difficult, given the concurrent profusion of published definitions of biochemical recurrence (17
). Moreover, biochemical recurrence predicts clinical endpoints with various degrees of precision, depending on factors including tumor grade and PSA kinetics after treatment (2
). Notable exceptions include a nomogram published by Kattan et al. (18
), shown to predict metastases after external beam radiotherapy and the three-level classification by D’Amico et al. (19
), which predicts cancer-specific mortality after radical prostatectomy or external beam radiotherapy.
The CAPRA score is among the most extensively and independently validated risk assessment tools available for localized prostate cancer, and it performs well in terms of accuracy, calibration, generalizability, and parsimony (5
). The score has previously been evaluated as a predictor of pathological and biochemical outcomes in community-based and academic cohorts of radical prostatectomy patients in both the United States and the Europe. In these studies, the accuracy of the instrument was generally good (c
-index range = 0.66 to 0.81) and was higher among the academic validation studies (6
). The accuracy of the CAPRA score in these studies was consistently comparable with the Kattan nomogram (c
-index range = 0.68 to 0.78) (9
). To our knowledge, however, the CAPRA score has not been assessed before this study as a predictor of distal endpoints or examined in cohorts of non–radical prostatectomy patients. Indeed, no validated multivariable instrument yet published has been demonstrated to predict mortality outcomes from time of diagnosis across multiple primary treatment types.
Yossepowitch et al. (22
) recently reviewed the accuracy of eight definitions of high-risk disease in predicting distant outcomes, including cancer-specific mortality after radical prostatectomy. These definitions included several simple definitions of risk grouping and a score of 50% or less on the updated preoperative nomogram of Stephenson et al. (23
). None of these measures were able to identify a group with greater than a 12% likelihood of cancer-specific mortality at 10 years after treatment. Of note, in the analysis of Yossepowitch et al., a PSA velocity of greater than 2 ng/mL per year, which was previously identified as a strong predictor of cancer-specific mortality (24
), was the weakest indicator of risk (22
). By contrast, the high-risk group that was identified by a CAPRA score of 6–10 in this analysis had a cancer-specific mortality at 10 years of 20.9%, compared with 2.9% and 8.4%, respectively, for the low-risk and intermediate-risk groups that were defined by CAPRA scores of 0–2 and 3–5, respectively. Moreover, individuals in the high-risk group that was defined by a CAPRA score of 6–10 can be substratified, with actuarial cancer-specific mortality rates ranging from 16.8% to 27.6%.
A particular strength of the CaPSURE database is its large numbers of patients undergoing different primary treatments with uniform ascertainment of follow-up assessment, PSA levels, and clinical endpoints, regardless of initial treatment. Pooling or comparing patients undergoing radical prostatectomy and radiation therapy is difficult in studies with biochemical endpoints given variations in the definitions of biochemical recurrence (17
). By analyzing metastases, cancer-specific mortality, and all-cause mortality, we circumvented this problem. Moreover, these distant endpoints are ultimately more relevant to patients than either pathological or biochemical outcomes. Finally, the CAPRA score can be calculated without paper nomograms, lookup tables, or computer software, and, therefore, is easily applied in clinical and research settings alike. Better and more consistent application of risk assessment techniques should be expected to reduce overtreatment of low-risk disease and undertreatment of high-risk disease, phenomena that appear to have diminished the potential benefits of prostate cancer screening (25
This study had several limitations. The number of metastasis and cancer-specific mortality events was relatively small, particularly for the secondary analysis by primary treatment type. Additional follow-up should provide more events, including those from patients managed with watchful waiting/active surveillance or cryotherapy. Ascertainment of cancer-specific mortality from a review of death certificates is inherently limited by the quality of information on the certificates; these may be completed by any physician who may have variable familiarity with prostate cancer and with the patient's history. Mortality that is caused by side effects of treatment, in particular, is likely to be underestimated. For example, the death of a patient with prostate cancer who dies of bladder cancer due to pelvic radiation (26
), coronary artery disease accelerated by androgen deprivation therapy (27
), or sequelae of a hip fracture attributable to osteoporosis that was accelerated by androgen deprivation therapy (28
) will likely not be attributed on the death certificate to prostate cancer. Underestimation of cancer-specific mortality may, in fact, partially explain the better-than-expected success of the CAPRA score in predicting all-cause mortality.
The CaPSURE practice sites are distributed across the United States but were not chosen at random and do not represent a statistically significantly valid sample of the population. Comparing the present cohort with the Surveillance Epidemiology and End Results (SEER) sample (29
) reveals some relatively minor demographic differences. The median age at diagnosis of prostate cancer patients in the SEER areas was 68 years in the period from January 1, 2001, through December 31, 2005, compared with 66 years in CaPSURE for the same period. In addition, for the same period, African Americans constituted 12.1% of the prostate cancer patients in SEER but only 10.3% of those in CaPSURE, whereas patients of other ethnicities constituted 12.9% of those in SEER but only 3.6% of those in CaPSURE. CaPSURE patients also tend to have slightly higher socioeconomic status on average than the overall population (11
A total of 3113 (22.6%) of the cohort of 13
740 patients were excluded from the analysis, with roughly one-third excluded because of missing data. This limitation likely reflects the large number of clinicians contributing data to the registry. Imputation of the CAPRA scores for those with only a single missing variable ameliorated the problem to some extent. The similar distribution of CAPRA scores among those with fully calculated and imputed scores was reassuring, as were results of the sensitivity analysis that excluded those patients with imputed scores. Furthermore, we had no reason to suspect that the missing data were not missing at random.
Patients in CaPSURE are treated by many clinicians in a variety of practice settings. Details of surgery, radiation therapy, and androgen deprivation therapy vary considerably with time and geographic location, and controlling adequately for this variability was not practical with the data available. However, we expect that this unmeasured variability would tend to artificially weaken rather than strengthen the accuracy of the instrument. Indeed, in previous studies (8
), the CAPRA score performed better in the academic series with fewer clinicians and more consistent treatment patterns than in CaPSURE and the Shared Equal Access Regional Cancer Hospital database (6
), both of which include multiple sites and clinicians. Future validation studies of the CAPRA score that use data from these and other databases will be important as more patients in these registries reach distal endpoints. Finally, in this analysis, we analyzed patients across multiple treatment approaches because, to date, outcomes have not been proven to be different between these approaches (3
). The question of differential risk-adjusted mortality outcomes across primary treatments will be addressed in future CaPSURE studies.
The CAPRA score, which has been well validated in multiple contexts to predict pathological and biochemical endpoints (6
), is, to our knowledge, the first instrument that uses information available at time of diagnosis to predict accurately the development of metastases, cancer-specific mortality, and all-cause mortality, irrespective of primary treatment. These findings were obtained by use of data from a diverse multi-institutional registry but should still be validated in other cohorts. The impact of primary and secondary therapy will be investigated in further detail in CaPSURE as more patients reach these distal endpoints. Given its high degree of accuracy and ease of calculation, the CAPRA score may prove an increasingly valuable tool for risk stratification in both the clinical practice and the research setting.