Blinded Sample Size Recalculation in Longitudinal Clinical Trials Using Generalized Estimating Equations

Abstract

In clinical trials in which outcomes are measured repeatedly during the follow-up phase, data analysis is frequently performed using generalized estimating equations (GEEs). Sample size calculation is then especially challenging since in addition to the treatment effect, the intrasubject correlation and the variability of the model error term have to be specified. In this article, the authors investigated by Monte Carlo simulations whether a blinded midcourse estimation of these quantities in an internal pilot study design is feasible in such trials and whether nominal type I and type II error rates are preserved when the estimates are used for sample size recalculation. The actual type I error rates of the blinded sample size recalculation procedure turned out to agree well with the nominal levels. Furthermore, the simulated power was observed to be near the target value as long as the sample size of the internal pilot study was sufficiently high and the bound effects induced by the range of the correlation were limited. The proposed procedure is a helpful tool to achieve robustness of the power with respect to initial misspecifications in the planning stage in clinical trials analyzed by GEE.