Comparison of Futility Monitoring Methods using Oncology Trials
Ed Zhang, Ph.D., American College of Radiology
March 20, 2012 @ 3:30 pm - 4:30 pm
Location: Blockley - Room 701
Biostatistics
Abstract: Futility monitoring is an important component in the conduct of clinical trials. An optimal rule would allow timely stopping if the new therapy is harmful or is unlikely to ultimately prove effective. Common methods proposed for futility monitoring include boundaries based on conditional power (CP), repeated confidence intervals (RCIs, adjusted for multiple looks), testing alternative hypothesis with small alpha (0.0025), and the linear 20% inefficacy boundary (LIB20). To compare among these futility rules and relative to no futility monitoring, we will select oncology trials conducted by NCI cooperative groups, include RTOG, National Surgical Adjuvant Breast and Bowel Project (NSABP), and GOG since 1990. These include multiple disease sites and cancer types with time to event as primary endpoints.
Total sample size and information for each trial are calculated using protocol specified effect size, type I, and type II error rates. Interim test statistics are reconstructed for each of the proposed futility analysis schedules. Operating characteristics of the futility monitoring rules under each analysis schedules are compared with respect to study duration and sample size, with impact of each rule (summarized as percent savings relative to no futility monitoring) on information, sample size, and calendar time scales over all trials.
Preliminary results show that testing alternative hypothesis and RCI rules yield less savings comparing to other rules. The CP10% rule is similar to LIB20 rule with respect to trial duration for more than 3 interim analyses, but saves less on information and sample size scales. The CP30% rule is too aggressive, stopping even when there is clinically meaningful treatment effect.
