Dynamic Regression of Longitudinal Trajectory Features.

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Chronic Disease Epidemiology

Background:

• Longitudinal studies in chronic diseases track biological and clinical markers over time.
• Understanding individual disease trajectories is crucial for assessing risk and status.
• Existing multi-level models often rely on restrictive distributional assumptions.

Purpose of the Study:

• To develop a novel dynamic regression framework for analyzing longitudinal data.
• To investigate heterogeneity in latent individual trajectories of disease progression.
• To link trajectory features with covariates without parametric assumptions.

Main Methods:

• Utilized multi-level modeling with pseudo B-spline functions for latent trajectories.
• Incorporated subject-specific random parameters for flexibility.
• Employed quantile regression to link latent features with observed covariates.
• Adapted conditional score principles for estimation and developed an efficient algorithm.

Main Results:

• The proposed framework effectively models heterogeneous patterns in longitudinal data.
• Estimators demonstrated desirable asymptotic properties and good finite-sample performance in simulations.
• The method provided valuable insights into cognitive decline heterogeneity in mild cognitive impairment (MCI) patients.

Conclusions:

• The dynamic regression framework offers a flexible approach to analyzing complex longitudinal disease data.
• This method avoids restrictive assumptions, enhancing applicability in biostatistics and epidemiology.
• The application to mild cognitive impairment (MCI) highlights its utility in understanding disease heterogeneity.