A phase II two stage clinical trial design to handle latent heterogeneity for a binary response.

摘要

Phase II clinical trial are generally single arm trial where a homogeneity assumption is placed on the response. In practice, this assumption may be violated resulting in a heterogeneous response. This heterogeneous or overdispersed response can be decomposed into distinct subgroups based on the etiology of the heterogeneity. A general classification model is developed to quantify the heterogeneity. The most common Phase II trial design used in practice is the Simon 2-stage design which relies on the assumption of response homogeneity. This design is shown to be flawed under the assumption of heterogeneity with errors exceeding the target trial errors. To correct for the error inflation, a modification is made to the Simon design if heterogeneity is detected after the first stage trial conduct. The trial sample size is increased using an empirical estimate for the variance inflation factor and the trial is then completed with design parameters constructed through the posterior predictive Beta-binomial distribution given the first stage results. The new design, denoted the 2-stage Heterogeneity Adaptive (2HA) design, is applied to a two subgroup problem under latent heterogeneity. Latent heterogeneity represents the most general form of heterogeneity, no information is known prior to trial conduct. The results, through simulation, show that the target errors can be maintained with this modification to the Simon design under a wide range of heterogeneity.