Censoring in trials testing immunotherapy in advanced cancers. A systematic review and a meta-research study

Background Informative censoring affects interpretation of trials results. We investigated censoring rates in randomized controlled trials (RCTs) of immune checkpoint inhibitors (ICIs).Methods We searched articles of RCTs testing ICIs in advanced cancers, published up to December 2023. For both progression-free survival (PFS) and overall survival (OS) Kaplan-Meier (K-M) curves, we collected the rates of censored patients at the first (T1), median PFS/OS (TmPFS/OS), and last (T2) study intervals. We calculated the unweighted difference in censoring rates (Delta C-E) and the weighted difference adjusted for enrollment size (w Delta C-E) in control (C) vs experimental (E) arm at T1, Tm, and T2.Results Of the selected 140 trials, censoring data at T1, Tm, and T2 were available for 53/140 (37.8%) and 55/140 (39.2%) trials for PFS and OS K-M curves, respectively. Rates of censoring in C and E were as follows: at T1, 8.19% and 4.92%, for PFS; TmPFS, 15.5% and 12.5%; T1, 2.33% and 1.16%, for OS; TmOS, 20.1% and 21.3%; T2, 23.29% and 26.34%, for PFS; T2, 33.3% and 39.49%, for OS. Analysis of w Delta C-E revealed more censoring in C at T1 (PFS = 1.32; OS = 0.40) and in E at T2 (PFS = -2.61; OS = -5.23). Finally, at T1, we found larger rates of censoring in C of open-label compared with double-blinded RCTs.Conclusions Multiple RCTs of ICIs did not report censoring data. The rate of censoring is higher in C at the start and increases in E over the course of the trial. Further studies might elucidate the role of censoring on survival outcomes.