John Simes

Informative censoring in maintenance therapy trials for advanced ovarian cancer: An empirical assessment of its impact on treatment benefit

A considerable proportion of patients in ovarian cancer maintenance trials may be censored in progression-free survival (PFS) analyses, the primary study endpoint. Such censoring is often informative, reflecting discontinuation due to toxicity, preference, or early switch to alternative therapies, potentially biasing results toward overestimating PFS benefit. We aimed to quantify the impact of informative censoring on PFS in these trials. Double-blind, placebo-controlled maintenance therapy trials were selected, and individual patient data reconstructed from published survival curves. A sensitivity analysis reclassified varying proportions of all censored events as progressions to model scenarios from 0 % to 100 % informative censoring. Hazard ratios (HRs) were re-estimated and compared with the originally reported values. Duration of therapy was compared with PFS. Twenty-two trial units (N = 8256) were included. Nineteen reported statistically significant results, falling to 14 (74 %) at the upper limit of analysis. HRs diminished progressively, with a 6 % reduction at 10 % censoring and 29 % at 100 %. In nine PARP inhibitor trials, treatment duration was shorter than PFS (mean of medians = 12.5 vs 17.6 months). Results were consistent when limited to PARP inhibitor studies. No correlation was observed between adverse events and censoring. Informative censoring can substantially distort PFS benefit estimates in ovarian cancer maintenance trials. Transparent reporting of censoring rates and their causes is essential for meaningful clinical interpretation and should be standard in all randomised maintenance therapy trials.