A62 Sample size re-estimation in response to treatment noncompliance

424

Abstracts

A61 SAMPLE SIZE AND MONITORING CRITERIA FOR BAYESIAN PHASE II CLINICAL TRIALS Peter F. Thall and Richard Simon*

M.D. Anderson Cancer Center, University of Texas Houston, Texas *National Cancer Institute Bethesda, Maryland Thall and Simon (1992) propose a Bayesian approach to phase II clinical trials with binary endpoints and continuous monitoring. The efficacy of an experimental treatment E is evaluated relative to that of a standard treatment S based on the data from the trial of E, an informative prior for the success probability of S and a non-informativeprior for the success probability of E. The trial continues until E is shown with a high posterior probability to be either promising or not promising, or until a predetermined maximum sample size is reached. This paper provides sample size and early termination criteria for this design, with accompanying numerical guidelines for implementation. In addition, two extensions of the above decision structure are proposed. The first extension provides criteria for early termination of trials unlikely to yield conclusive results, based on the continuouslyupdated marginal (predictive) probability distribution for the observed success rate. The second extension is a version of the design which only terminates early if E is found to be not promising compared to S, and otherwise continues to the maximum sample size. Operating characteristics of each of these designs are evaluated numerically. In particular, we examine the effects of incomplete (discontinuous)sequential monitoring on the operating characteristics and achieved sample size, as compared to continuous monitoring. A62 SAMPLE SIZE RE-ESTIMATION IN RESPONSE TO TREATMENT NONCOMPLIANCE Dennis Cosmatos and T. Timothy Chen

Yanssen Research Foundation Titusville, New Jersey When patients entered onto a clinical trial do not complete treatment as prescribed in the protocol, difficulties surface regarding the manner of proper analysis of outcome information from those patients. Although it is widely accepted that an intent-to-treat analysis is the most appropriate, noncompliance may force a substantial loss of power. In a recent cancer clinical trial, lifts problem of treatment noncompliance was addressed by implementing a method for re-estimating the required sample size for the trial based on the rate of treatment noncompliance. This adjustment relies on the estimation of the observed compliance rate for each of the comparison treatment groups. Since the proposed method only considers adjustment of sample size, no "subgroup" analysis bias is introduced and analysis of patients by assigned treatment is maintained. Based on results from analyses of simulated data, sample size adjustment factors are presented for various levels of treatment differences and observed rates of treatment noncompliance. The simulation studies also demonstrated that there is no appreciable effect of this re-estimation procedure on Type I error whereas the procedure does dramatically improve the power of the study. Effects of dynamic sample size adjustment in response to changing rates of treatment compliance during the course of the study are also examined. It is suggested that the proposed sample size re-estimation procedures are incorporated into the initial study design.