A New Logistic Model With Subject-Specific and Serially Correlated Time-Specific Distribution-Free Random Effects on...

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Statistical Modeling

Background:

• Existing beta-binomial mixed effects models for longitudinal binary data often require strict parametric assumptions for beta and normal random effects.
• Current methods may compromise computational efficiency and clear interpretation when integrating normal random effects into beta-binomial models.

Purpose of the Study:

• To introduce a novel multiplicative model for longitudinal binary data.
• To incorporate distribution-free, subject-specific, and serially correlated time-specific random effects into logistic regression.
• To enhance the interpretability of random effects as risk modifiers and simplify model derivation and prediction.

Main Methods:

• Developed a new multiplicative logistic regression model.
• Incorporated distribution-free random effects on the unit interval, accounting for subject-specific and time-specific correlations.
• Employed a quasi-likelihood approach for model parameter estimation.

Main Results:

• The proposed model facilitates clear interpretation of random effects as risk modifiers.
• The multiplicative framework simplifies model derivation and random effects prediction.
• The quasi-likelihood estimation method yields results robust to the distribution of random effects.

Conclusions:

• The new multiplicative model offers a flexible and interpretable alternative for analyzing longitudinal binary data.
• The approach is computationally advantageous and robust, as demonstrated by its application to multiple sclerosis trial data.