Causal inference using multivariate generalized linear mixed-effects models.

Area of Science:

• Biostatistics
• Epidemiology
• Precision Medicine

Background:

• Dynamic prediction of causal effects is crucial for personalized treatment strategies.
• Observational studies present challenges due to unknown treatment assignment and effect mechanisms.
• Unmeasured confounding factors can bias treatment effect estimations.

Purpose of the Study:

• To develop a robust statistical framework for estimating subgroup-specific treatment benefits in dynamic treatment regimes.
• To address the challenge of unmeasured time-invariant confounders in observational data.
• To investigate the efficacy of continuous mycophenolate use in scleroderma patient subgroups.

Main Methods:

• A multivariate generalized linear mixed-effects model was employed.
• A Bayesian g-computation algorithm was utilized to compute posterior distributions of treatment benefits.
• Subject-specific random effects were incorporated to account for unmeasured time-invariant factors.

Main Results:

• The proposed method successfully estimates subgroup-specific intervention benefits.
• Simulation studies demonstrated the method's performance in handling unmeasured confounding.
• Application to scleroderma data revealed differential efficacy of continuous mycophenolate use across subgroups.

Conclusions:

• The developed statistical model and algorithm effectively estimate dynamic causal effects in precision medicine.
• The approach accounts for unmeasured heterogeneity, improving treatment effect estimation from observational studies.
• Findings provide insights into personalized treatment strategies for scleroderma patients.