Subgroup Analysis for Longitudinal data via Semiparametric Additive Mixed Effect Model

TL;DR

This paper introduces a flexible semiparametric additive mixed effect model framework for subgroup analysis in longitudinal data, capable of identifying subgroups and estimating regression functions simultaneously, applicable to both balanced and unbalanced datasets.

Contribution

It develops a novel subgroup analysis method combining backfitting and k-means algorithms for longitudinal data, estimating the number of groups via Bayesian information criteria.

Findings

Method accurately identifies subgroups and estimates regression functions.

Numerical studies confirm the method's efficacy and precision.

Application to PBC data demonstrates practical usefulness.

Abstract

In this paper, we propose a general subgroup analysis framework based on semiparametric additive mixed effect models in longitudinal analysis, which can identify subgroups on each covariate and estimate the corresponding regression functions simultaneously. In addition, the proposed procedure is applicable for both balanced and unbalanced longitudinal data. A backfitting combined with k-means algorithm is developed to estimate each semiparametric additive component across subgroups and detect subgroup structure on each covariate respectively. The actual number of groups is estimated by minimizing a Bayesian information criteria. The numerical studies demonstrate the efficacy and accuracy of the proposed procedure in identifying the subgroups and estimating the regression functions. In addition, we illustrate the usefulness of our method with an application to PBC data and provide a meaningful partition of the population.