Regression Trees for Longitudinal Data with Baseline Covariates.

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Machine Learning in Healthcare

Background:

• Longitudinal data often exhibits heterogeneity, making traditional models insufficient.
• Existing regression tree methods have limitations in handling repeated measures or mixed variability.
• Understanding subgroup dynamics is crucial for personalized medicine.

Purpose of the Study:

• To develop a robust regression tree algorithm for longitudinal data.
• To identify and characterize homogeneous subgroups within longitudinal datasets.
• To apply the novel method to analyze choline level changes in HIV patients.

Main Methods:

• Proposed a longitudinal classification and regression tree (LongCART) algorithm.
• Utilized a two-step approach with a parameter instability test for variable selection.
• Employed a conditional inference framework to control for overfitting and bias.
• Evaluated performance through simulation studies and asymptotic results.

Main Results:

• The LongCART algorithm successfully identified homogeneous subgroups in longitudinal data.
• The parameter instability test effectively selected relevant partitioning variables.
• Simulations demonstrated the algorithm's superior performance compared to existing methods.
• Applied LongCART to reveal patterns in longitudinal choline levels among HIV-positive patients.

Conclusions:

• LongCART offers a powerful and flexible approach for analyzing heterogeneous longitudinal data.
• The method provides a type-I error controlled strategy for subgroup identification.
• This approach enhances understanding of disease progression and treatment response in patient populations.