統計学輪講 第4回

In clinical trials, it is of particular importance to judge whether there exists a subgroup, characterized by some covariates, whose treatment effect is enhanced in comparison with the other. In the literature, such subgroup identification is carried out in the form of hypothesis testing (e.g. [1] and [2]). This testing-based approach, however, often requires bootstrap, which can possibly be computationally costly. To combat this problem, this research proposes a novel, simple alternative for subgroup identification with a Logistic-Normal mixture model. The proposed method judges the existence of a subgorup based on the value of particular parameters which shrinks toward zero when there is no subgroup. Some theoretical properties of the method is also discussed.

[1]Shen, J. and He, X. (2015) “Inference for Subgroup Analysis with a Structured Logistic-normal Mixture Model,” Journal of the American Statistical Association, 110, 303-312.
[2]Fan, A., Song, R., and Lu, W. (2017) “Change-plane Analysis for Subgroup Detection and Sample Size Calculation,” Journal of the American Statistical Association, 112, 769-778.