Comparison of Longitudinal Trajectories Using a High-dimensional Partial Linear Semiparametric Mixed-Effects Model

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Semiparametric Modeling

Background:

• Comparing longitudinal trajectories across groups is crucial in research.
• Existing methods may struggle with nonlinear patterns and high-dimensional data.

Purpose of the Study:

• To present a partial linear semiparametric mixed-effects model (PLSMM) for analyzing and comparing nonlinear longitudinal trajectories.
• To offer a flexible framework for complex temporal effects and high-dimensional covariates.
• To enable statistical inference on both linear and nonlinear components between groups.

Main Methods:

• Developed a partial linear semiparametric mixed-effects model (PLSMM).
• Employed a dictionary search strategy for automatic basis function selection to capture nonlinear trends.
• Introduced a novel debiasing procedure for post-selection inference on linear components.
• Utilized a bootstrap method for comparing nonlinear components.

Main Results:

• The PLSMM effectively handles complex temporal effects and high-dimensional covariates.
• The model successfully models nonlinear patterns without requiring prior functional form specification.
• Demonstrated capability in analyzing longitudinal data with irregular time points.
• Validated through simulations and application to a cohort study on oral Candida albicans concentration.

Conclusions:

• The PLSMM provides a robust framework for the analysis and comparison of nonlinear longitudinal trajectories.
• This method offers significant advantages in handling complex data structures and identifying group differences.
• The approach is valuable for research involving dynamic biological processes and diverse populations.