Causal inference in longitudinal comparative effectiveness studies with repeated measures of a continuous...

Area of Science:

• Causal inference
• Longitudinal data analysis
• Comparative effectiveness research

Background:

• Nonrandomized longitudinal studies present challenges in assessing causal effects.
• Intermediate variables, like glucose levels, complicate treatment effect estimation.
• Principal stratification offers a framework to address these complexities.

Purpose of the Study:

• To propose a principal stratification approach for causal inference in longitudinal comparative effectiveness studies.
• To extend existing methods to handle binary endpoints and continuous intermediate variables.
• To compare the effectiveness of glucose-lowering medications in type 2 diabetes patients.

Main Methods:

• A three-step modeling procedure is proposed.
• Hybrid growth mixture modeling identifies principal strata based on intermediate variables.
• Pseudoclass technique and propensity score weighting are used to estimate stratum-specific treatment effects.

Main Results:

• The method allows for causal effect estimation adjusting for baseline covariates and intermediate variable trajectories.
• It enables comparison of medication effects on cardiovascular disease hospitalization and all-cause mortality.
• The approach accounts for heterogeneity in compliance and glucose levels over time.

Conclusions:

• The proposed principal stratification method provides a robust framework for causal inference in complex longitudinal studies.
• It facilitates a more accurate assessment of treatment effects by stratifying on intermediate outcomes.
• This approach is particularly valuable for comparative effectiveness research in chronic diseases like type 2 diabetes.