On changing a long-term clinical trial midstream

Clinical triallists are often reluctant to alter the protocol or the design of an ongoing study in part because of concern that changes may affect the integrity of the study. This paper encourages considering changes in long-term clinical trials when the medical environment changes or when the accruing data from the trial lead to questioning the assumptions underlying the original design. One must make such changes in a way that will not cast doubt on the integrity of the trial. An important aspect of the design is choice of sample size. Methodology for sample size recalculation has matured over the decade. Several techniques are now available for the usual types of endpoints - continuous, discrete and time-to-failure - as well as for longitudinal analysis. Published papers describe the choices of variance at the time of recalculation, approaches to estimation and testing at the end of the study, and the time of recalculation. In a study with a sponsor, a Data Safety Monitoring Board (DSMB) and an Executive Committee, one or more of these three groups is responsible for recommending increases in sample size when the accruing data indicate that the assumed variance underestimated the true variance. Sometimes a statistician independent of all three bodies performs the relevant calculations and sends the recommendation to the sponsor. This paper argues that the DSMB should not generally be the responsible body because knowledge of treatment effect can place it in an uncomfortable position.