CHINA·77779193永利(集团)有限公司-Official website

题目：Subgroup Analysis for Heterogeneous Samples

报告人：Baosheng Liang （Department of Statistics and Actuarial Science, University of Hong Kong）

摘要：Detections for the subgroups of patients with similar characteristics is animportant issue, which is essential to the development of personalized (or precision) medicine. In practice, this task is often challenging due to the lack of priorknowledge for the grouping and the complex heterogeneous structure of thesubgroups in feature space. In this article, we develop a hybrid framework tomodel the impact of subgroup structure and the effect heterogeneity of covariate onregression. The proposed method is tailored for handling the situations where thesample is not homogeneous in the sense that the response variables in differentdomains of feature space are generated through different mechanisms. For suchscenario, the observations can be viewed as a mixture of several homogeneous datasets. Directly applying the traditional regression method would produce inaccurateestimates and lead to misleading inference results. Compared to the conventionalsubgroup analysis methods which usually detect the subgroups based on differenceof subjects in feature space, the proposed method can detect the subgroups usingboth the difference of subjects in feature space and the effect difference in thecovariates. The key step of our method incorporates the K-means clustering intothe regression framework with pairwise concave fusion so that the regression andsubgroup detection tasks can be performed simultaneously. An efficient algorithmof subgroup detection with robust adjustment of the estimated coefficients isdeveloped. Through extensive simulation studies, we demonstrate superiorperformance of the proposed method in terms of the accuracy rate of subgroupdetection and prediction as well as the accuracy of the estimated regressioncoefficients for each subgroup.

时间：12月8日（星期五）下午1：30-2：30

地点：首师大校本部新教二楼527

欢迎全体师生积极参加！