Variable Selection for Illness-Death Processes Under Dual Observation Schemes.

Area of Science:

• Biostatistics
• Epidemiology
• Chronic Disease Research

Background:

• The illness-death process is crucial for chronic disease research, but often progression times are not directly observed.
• Disease progression is frequently recorded at intermittent assessments, leading to interval-censored data.
• Simultaneous modeling of disease progression and death is essential for comprehensive analysis.

Purpose of the Study:

• To develop a statistical method for variable selection in joint illness-death models with interval-censored progression data.
• To address the challenges of dual observation schemes in chronic disease progression and mortality studies.
• To provide a flexible framework for analyzing complex disease trajectories using penalized likelihood methods.

Main Methods:

• A penalized observed data likelihood approach for variable selection in multiplicative intensity-based models.
• Utilized different penalty functions for distinct sets of regression coefficients.
• An expectation-maximization algorithm was developed for optimization, compatible with existing software.

Main Results:

• Simulation studies confirmed the method's good finite sample performance.
• The approach effectively handles interval-censored progression times and right-censored death data.
• Variable selection was demonstrated to be robust in the context of joint modeling.

Conclusions:

• The proposed penalized likelihood method offers a powerful tool for variable selection in joint illness-death models.
• This approach enhances the understanding of chronic disease progression and mortality.
• Application to dementia data (National Alzheimer's Coordinating Center) highlights its practical utility.