A novel block-coordinate gradient descent algorithm for simultaneous grouped selection of fixed and random effects in...

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Survival Analysis

Background:

• Joint models integrate longitudinal and time-to-event data to explore associations.
• Existing methods face challenges in simultaneously selecting fixed and random effects efficiently.
• A need exists for robust methods to handle complex covariate selection in joint models.

Purpose of the Study:

• To propose a novel algorithm for simultaneous selection of fixed and random effects in joint models.
• To address the research gap in efficient and effective covariate selection within joint modeling frameworks.
• To enhance the analysis of longitudinal and time-to-event data by improving covariate identification.

Main Methods:

• Developed a block-coordinate gradient descent (BCGD) algorithm for covariate selection.
• Employed a linear mixed-effects model for longitudinal processes with random intercepts and slopes.
• Utilized a proportional hazards model for the time-to-event submodel and penalized likelihood estimation.

Main Results:

• The proposed BCGD method effectively selects significant fixed and random effects covariates.
• Demonstrated excellent selection power and provided accurate empirical estimates for effects.
• Simulation studies confirmed the method's excellent performance and effectiveness.

Conclusions:

• The BCGD algorithm offers an efficient and effective solution for covariate selection in joint models.
• Successfully applied to real-world data, identifying risk factors for heart valve outcomes and primary biliary cholangitis.
• Highlights the utility of advanced statistical methods in uncovering complex health-related risk factors.