Progression-free survival estimates are shaped by specific censoring rules: Implications for PFS as an endpoint in cancer randomized trials

Kaplan-Meier analysis hinges on the assumption that patients who are censored– lost to follow-up, or only recently enrolled on the study– are no different, on average, than patients who are followed. As such, censoring these patients– omitting their future information and taking the average of those who were followed– should not dramatically change the overall estimate. Yet, in a recent clinical trial, two sets of censoring rules– one favored by trialists and one favored by the US Food and Drug Administration– were applied to a progression-free survival (PFS) estimate. In response, the PFS estimate changed dramatically, increasing the median in the experimental arm from 32 to 43 months, while the control arm was essentially unchanged. In this commentary, we explore the reasons why PFS changed so dramatically. We provide a broad overview of censoring in oncology clinical trials, and suggestions to ensure that PFS is a more reliable endpoint. •Censored data can substantially affect survival analysis results.•Different set of censoring rules distort progression-free survival estimates.•Informative censoring might benefit only patients in the experimental arm.•Trial endpoints measuring both clinical benefit and toxicity should be preferred.