Comparative Effectiveness of Prostate Cancer Treatment Options: Limitations of Retrospective Analysis of Cancer Registry Data

PURPOSE:Retrospective analyses of cancer registry and institutional data have consistently found better survival after radical prostatectomy versus radiation therapy, which contrasts with findings from a randomized trial. This is likely because of the inability of retrospective studies to fully account for comorbidity differences across treatment groups because of the lack of detailed data in the registries. We use a unique population-based data set with detailed data regarding comorbidities and functional limitations to assess whether this can provide valid comparisons of survival across prostate cancer treatment groups. METHODS AND MATERIALS:The Surveillance, Epidemiology, and End Results (SEER)-Medicare Health Outcomes Survey (MHOS) data set results from a linkage between the SEER database and the MHOS database, which includes detailed information regarding patient-reported comorbidity and functional limitations. We analyzed 3102 patients with prostate cancer in SEER-MHOS and used latent class analysis to identify the healthiest group with minimal comorbidity burden and functional limitations. Among the healthiest group, we examined overall survival across treatments using the Kaplan-Meier method. RESULTS:Three distinct health groups were identified using latent class analysis; the healthiest group comprised 57% of the cohort and had a 10-year overall survival of 67%. Other health groups had higher rates of comorbidities or functional limitations. Among the healthiest group, 10-year overall survival differed across treatment groups: no local treatment (55%), external beam radiation therapy (69%), brachytherapy (76%), and radical prostatectomy (85%). Survival curves for the 3 treated groups separated at 4 years of follow-up. CONCLUSIONS:Despite the detailed health status information available in SEER-MHOS, our retrospective analysis could not fully account for patient selection biases across prostate cancer treatment groups. These findings highlight an important limitation of retrospective studies using population-based data sets and serve as a reminder to interpret results with caution.