Drago, Joshua Z.

Abstract Background Safety is a central consideration when choosing between multiple medications with similar efficacy. We aimed to evaluate whether adverse event (AE) profiles of 3 such drugs in advanced prostate cancer could be distinguished based on published literature. Methods We assessed consistency in AE reporting, AE risk in placebo arms, and methodology used for risk estimates and quantification of statistical uncertainty in randomized placebo-controlled phase III trials of apalutamide, enzalutamide, and darolutamide in advanced prostate cancer. Results Seven included clinical trials enrolled a total of 9215 participants (range = 1051-1715 per trial) across 3 prostate cancer disease states. Within disease states, baseline patient characteristics appeared similar between trials. Of 54 distinct AE types in total, only 3 (fatigue, hypertension, and seizure) were reported by all 7 trials. Absolute risks of AEs in the placebo arms differed systematically and more than twofold between trials, which was associated with visit frequency and resulted in different degrees of uncertainty in AE profiles between trials. No trial used inferential methodology to quantify statistical uncertainty in AE risks, but 6 of 7 trials drew overall conclusions. Two trials concluded that there was no elevated AE risk because of the intervention, including the trial of darolutamide, which had the greatest statistical uncertainty. Conclusions Rigorous comparison of drug safety was precluded by heterogeneity in AE reporting, variation in AE risks in the placebo arms, and lack of inferential statistical methodology, underscoring considerable opportunities to improve how AE data are collected, analyzed, and interpreted in oncology trials.

318 Background: Apalutamide, enzalutamide, and darolutamide were approved for non-metastatic castration-resistant prostate cancer (nmCRPC) based on three randomized placebo-controlled trials that showed similar efficacy. Adverse event (AE) profiles have been invoked to differentiate the drugs but have only been informally compared. Methods: We accounted for baseline characteristics, AE collection and reporting, and statistical uncertainty when comparing AE risks in the drug and placebo arms of SPARTAN (NCT01946204), PROSPER (NCT02003924), and ARAMIS (NCT02200614). Results: Patients in SPARTAN, PROSPER, and ARAMIS had a similar median age (74 years in all studies), median PSA doubling time (3.7-4.7 months) and ECOG performance status (68-80% with ECOG 0). AEs were gathered consistently using CTCAE 4.0 or 4.03. However, PROSPER and ARAMIS reported AEs occurring in ≥5% of patients; SPARTAN reported AEs occurring in ≥15%. Trials also reported 5 to 14 ‘AEs of interest.’ Of 34 AE types reported overall, only 10 were reported in all three trials. Absolute risks of adverse events in the placebo arms differed considerably. Compared to the placebo arm of SPARTAN, AEs were on average 44% less common in the placebo arm of PROSPER (95% CI, 28-56%) and 54% less common in the placebo arm of ARAMIS (95% CI, 41-64%), a difference not explained by length of follow-up. With lower event numbers, relative risk estimates were less precise with wider confidence intervals. For example, comparing treatment vs. placebo arms, the relative risk for fatigue, the most common AE, was 1.45 (95% CI, 1.16-1.80) in SPARTAN, 2.37 (95% CI, 1.86-3.04) in PROSPER, and 1.39 (95% CI, 1.01-1.91) in ARAMIS. Across all AE types, compared to SPARTAN, relative risks from PROSPER were 23% less precise and relative risks from ARAMIS were 30% less precise. Conclusions: While conducted in similar patient populations, these trials had remarkable differences in AE reporting and in absolute AE risks between placebo arms. Rather than indicating better safety, low absolute adverse event numbers decrease confidence in AE profiles. Published data are insufficient to differentiate the AE profiles of these agents in nmCRPC.