gms | German Medical Science

Interest in personalized treatment and adaptive design strategies are increasing. Study outcomes are commonly continuous, binary or time to an event of interest. Those are often analysed using general linear models, logistic regressions or Cox proportional hazard models, depending on the type of outcome. These regression models allow for the adjustment for baseline covariates, which minimizes bias and increases efficiency, i.e., provides higher power for hypothesis tests [1].

Similar to the work by Placzek and Friede [2], we consider testing for treatment effects in several nested subgroups, which might for example be given by thresholds of a continuous biomarker. While Placzek and Friede considered normally distributed endpoints without adjustments for covariates, we aim to provide a general approach for testing subgroups in an adaptive trial setting using the inverse normal combination function described in Lehmacher and Wassmer [3].

Following to Mehta et al. [4] we divide the full population in disjoint subsets, in which the hypotheses of interest are tested. In a closed testing procedure the hypothesis tests for subgroups of certain stages consisting of one or more disjoint subsets are tested by combining the p-values for the relevant subsets to construct test statistic for subpopulations. We extend this approach to adaptive enrichment designs to allow for stage-wise testing while maintaining type I error rate control.

We investigate the properties of the proposed testing procedures using Monte Carlo simulations. Finally we close by discussing generalizations to non-nested subgroups.

The authors declare that they have no competing interests.

The authors declare that an ethics committee vote is not required.