The first-order Markov conditional linear expectation approach for analysis of longitudinal data.

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Statistical Modeling

Background:

• Longitudinal discrete data often exhibit overdispersion, inflating variance beyond assumed distributions.
• Existing methods like generalized estimating equations (GEE) are semiparametric and may not fully leverage data structure.
• Unequally spaced time points add complexity to standard longitudinal analyses.

Purpose of the Study:

• To develop and evaluate a likelihood-based approach for analyzing longitudinal discrete data with overdispersion.
• To compare the performance of the proposed method against generalized estimating equations (GEE).
• To demonstrate the application of the new method in real-world scenarios.

Main Methods:

• Implementation of a likelihood-based method extending generalized linear models for longitudinal data.
• Assumption of subject independence, first-order antedependence within subjects, and exponential family distributions.
• Modeling linearity of expectations for conditional distributions.

Main Results:

• Simulations indicate benefits of the proposed likelihood-based approach for longitudinal discrete data analysis.
• The method was successfully applied to seizure count data and transplant center performance evaluation.
• Generalized estimating equations (GEE) demonstrated unexpectedly strong performance in simulations.

Conclusions:

• The novel likelihood-based method provides a valuable tool for analyzing complex longitudinal discrete data.
• The proposed approach offers a parametric alternative to semiparametric methods like GEE.
• Further research may explore extensions and applications of this likelihood-based framework.