|
Abstract:
|
In group sequential clinical trials, it is necessary to estimate the amount of information at interim analysis times relative to the information at the final analysis. This is often the ratio of sample sizes, if only one measurement is made per individual. However, as noted by Wu and Lan (1992, Biometrics, 48, 765-779), when the statistic of interest is a change over time such an approach overestimates the information. In this talk, we explore problems that can result when analysis of randomized controlled trials is based on standard linear models. Specifically, when the true pattern of dose response does not correspond exactly to the linear contrast used to measure treatment effect, information growth curves estimated with standard methods may be nonmonotonic. We give examples in which these standard estimates are inappropriate and demonstrate approaches that provide improved estimates.
|
