Optimal information size in trial sequential analysis of time-to-event outcomes reveals potentially inconclusive results because of the risk of random error

Abstract

ObjectivesThe current approach for evaluating the risk of random error in meta-analyses (MAs) using trial sequential analysis (TSA) can accommodate binary and continuous data but not time-to-event data. We conducted a TSA for time-to-event outcomes and applied the method to determine the risk of random error in MAs for treatments of multiple myeloma. Study Design and SettingLiterature search identified 11 systematic reviews consisting of 23 MAs. Of the 23 MAs, 13 had overall survival and 10 had progression-free survival as outcome; 48% (11 of 23) reported statistically significant treatment effects. We calculated the optimal a priori diversity–adjusted information size (APDIS) based on the relative risk reduction of 15% and 25%. We also calculated the optimal low-bias information size (LBIS) and low-bias diversity-adjusted information size (LBDIS). ResultsOverall, under APDIS15%, 48% (11 of 23) of MAs were false negative (FN) and 17% (4 of 23) of MAs were false positive. Under APDIS25%, 34% (8 of 23) of MAs were false negative and 4% (1 of 23) of MAs were false positive. LBIS identified 30% (7 of 23) as false negative MAs and 4% (1 of 23) as false positive MAs, whereas LBDIS identified 52% (12 of 23) as false negative MAs and 4% (1 of 23) as false positive MAs. ConclusionThe new method demonstrates the possibility of incorporating time-to-event outcomes into TSA and reveals that some MAs have potentially inconclusive results.