EcoSta 2024: Start Registration
View Submission - EcoSta2024
A0696
Title: Inference on potentially identified subgroups in clinical trials Authors:  Xinzhou Guo - The Hong Kong University of Science and Technology (Hong Kong) [presenting]
Abstract: When subgroup analyses are conducted in clinical trials with moderate or high dimensional covariates, candidate subgroups often need to be identified from the data, and the potentially identified subgroups are evaluated in a replicable way. The classical statistical inference applied to the potentially identified subgroups, assuming the subgroups are the same as what is observed from the data, might suffer from bias issues when the regularity assumption that the boundaries of the subgroups are negligible is violated. A shift-based method is proposed to address the nonregularity bias issue and combine it with cross-fitting and subsampling to develop a de-biased inference procedure for potentially identified subgroups. The proposed method is model-free and asymptotically efficient whenever it is possible and can be viewed as an asymmetric smoothing approach. The merits of the proposed method are demonstrated by re-analyzing the ACTG 175 trial.