Capturing heterogeneity in repeated measures data by fusion penalty

Area of Science:

• Statistics
• Biostatistics
• Data Analysis

Background:

• Clustered and longitudinal data often exhibit heterogeneity, which traditional fixed effects (FE) and random effects (RE) models may not fully capture.
• FE models can lack efficiency due to high degrees of freedom for heterogeneity.
• RE models may oversimplify heterogeneity, leading to potential biases in subject-specific effect estimates.

Purpose of the Study:

• To develop and evaluate a novel statistical approach, the "fused effects" model, for modeling heterogeneity in clustered or longitudinal data.
• To offer an intermediate approach between FE and RE models, accommodating an unknown number of distinct heterogeneity levels.
• To assess the performance of the fused effects model against traditional FE and RE methods through simulation studies.

Main Methods:

• The proposed fused effects model utilizes a fusion penalty approach for estimation and inference.
• This method assumes an unknown number of distinct levels of heterogeneity.
• Performance is evaluated via simulation studies and applied to real-world data.

Main Results:

• Simulation studies demonstrate the performance of the fused effects model compared to FE and RE models.
• The model successfully captured heterogeneity in the Ocular Hypertension Treatment Study.
• It specifically analyzed the progression rate of primary open-angle glaucoma in left and right eyes across different subjects.

Conclusions:

• The fused effects model provides a flexible and effective alternative for handling heterogeneity in clustered and longitudinal data.
• This approach offers improved efficiency and reduced bias compared to traditional FE and RE methods.
• The model has practical applications in analyzing complex health data, such as glaucoma progression.