Subgroup Identification Of High-dimensional Longitudinal Data And Its Application

In biomedical statistics,the traditional homogeneous overall model is widely used to solve practical problems,but due to the heterogeneity between individuals,the efficiency of the simple homogeneous overall model is low.The method of subgroup identification allows researchers to relax the assumption of overall homogeneity to a more reasonable assumption of heterogeneity among groups,thereby obtaining more accurate coefficient estimates.Therefore,the strong adaptability of the subgroup identification method in the identification of heterogeneous data has received extensive attention from researchers,and subgroup analysis has become an important tool in various application scenarios such as clinical trials,personalized medicine,and market segmentation.However,few scholars construct subgroup identification methods based on the background of high-dimensional longitudinal data.At the same time,longitudinal data can be seen everywhere in biomedical research.Therefore,a reasonable and practical subgroup identification method is proposed to identify high-dimensional The subgroups in the longitudinal data are particularly important.In the context of high-dimensional longitudinal data modeling,this paper constructs a data-driven subgroup identification method,which creatively combines the maximumminimum concave penalty method and an improved binary segmentation method to identify subgroups The problem is transformed into the problem of identifying change points between regression parameters,and the improved binary segmentation method is used to identify the change points between regression coefficients,combined with the BIC criterion to select the threshold,and get the best subgroup model structure.Then through statistical simulation experiments,the constructed subgroup identification method is compared with the other six methods,and the performance of the constructed subgroup identification method is comprehensively investigated.Finally,the constructed subgroup identification is applied to industrial structure data and kangaroo growth data for case analysis.The results show that the subgroup identification method is better than the other six subgroup identification methods in terms of comprehensive performance.