Understanding randomized controlled trial generalizability through an embedded molecular diagnostics trial

Issues with randomized controlled trial generalizability are well described, but whether these issues result from differences in patient treatment across contexts remains unknown. We studied treatment of patients with high-risk prostate cancer after radical prostatectomy inside and outside a randomized controlled trial evaluating the impact of a genomic classifier on post-radical prostatectomy treatment decision making (Genomics in Michigan Impacting Observation of Radiation [G-MINOR]; ClinicalTrials.gov identifier NCT02783950).

G-MINOR enrolled 338 patients; propensity score-matched eligible but unenrolled patient cohorts (pretrial and trial contemporary) from the Michigan Urological Surgery Improvement Collaborative (MUSIC), in which the trial was embedded, were compared for rates and time to secondary treatment (adjuvant or salvage therapy) after prostatectomy.

Among 338 patients in the G-MINOR cohort, 69 (31 adjuvant, 38 salvage) received secondary treatment compared with 266 (183 adjuvant, 83 salvage) and 104 (60 adjuvant, 44 salvage) in the 1014 contemporary and 338 pretrial-matched MUSIC cohorts. Time to secondary treatment was much shorter in the MUSIC cohort across all comparisons. For example, matching G-MINOR to synchronous MUSIC patients demonstrated 84% vs 74% estimated 2-year treatment-free survival for trial and real-world patients, respectively (P < .001).

Controlling for key clinicopathologic factors, patients in the G-MINOR randomized controlled trial and MUSIC cohorts were treated differently, even after stratifying by genomic risk. These findings suggest that challenges in randomized controlled trial generalizability extend beyond the representativeness of trial participants. Differences in management may also explain why divergent patient outcomes are observed in randomized controlled trials vs real-world settings.