Magnitude of effect and sample size justification in trials supporting anti-cancer drug approval by the US Food and Drug Administration

Abstract

Approval of drugs is based on randomized trials observing statistically significant superiority of an experimental agent over a standard. Statistical significance results from a combination of effect size and sampling, with larger effect size more likely to translate to population effectiveness. We assess sample size justification in trials supporting cancer drug approvals. We identified US FDA anti-cancer drug approvals for solid tumors from 2015 to 2019. We extracted data on study characteristics, statistical plan, accrual, and outcomes. Observed power (Pobs) was calculated based on completed study characteristics and observed hazard ratio (HRobs). Studies were considered over-sampled if Pobs > expected with HRobs similar or worse than expected or if Pobs was similar to expected with HRobs worse than expected. We explored associations with over-sampling using logistic regression. Of 75 drug approvals (reporting 94 endpoints), 21% (20/94) were over-sampled. Over-sampling was associated with immunotherapy (OR: 5.5; p = 0.04) and associated quantitatively but not statistically with targeted therapy (OR: 3.0), open-label trials (OR: 2.5), and melanoma (OR: 4.6) and lung cancer (OR: 2.17) relative to breast cancer. Most cancer drug approvals are supported by trials with justified sample sizes. Approximately 1 in 5 endpoints are over-sampled; benefit observed may not translate to clinically meaningful real-world outcomes.