Estimate Time-Varying Exposure Effects via Ensemble Learning-Based Marginal Structural Model With Application to...

Area of Science:

• Epidemiology
• Biostatistics
• Machine Learning

Background:

• Evaluating time-varying exposures is crucial in longitudinal studies.
• Estimating effects with numerous time-dependent confounders presents significant challenges.
• Existing methods may be sensitive to model misspecification.

Purpose of the Study:

• To propose a robust estimator for marginal structural models (MSMs) in longitudinal settings.
• To develop a model that is less sensitive to misspecification by integrating multiple machine learning algorithms.
• To address challenges posed by hundreds of time-dependent confounders and potential nonlinear effects.

Main Methods:

• Developed the Marginal Structure Ensemble Learning Model (MASE).
• MASE integrates multiple machine learning algorithms for propensity score and conditional outcome mean modeling.
• Employed extensive simulation analysis to compare MASE with benchmark methods (MSM, G-computation, Targeted Maximum Likelihood).

Main Results:

• MASE demonstrated superior performance over benchmark methods in simulations.
• The proposed model yielded smaller estimation bias and improved inference accuracy.
• Application to adolescent cognitive development showed an aggregated negative impact of insufficient sleep on cognitive performance.

Conclusions:

• MASE offers a robust approach for estimating effects in longitudinal studies with complex time-varying confounding.
• The ensemble learning strategy enhances model stability and reduces the risk of inconsistent estimation.
• Findings highlight the detrimental effect of insufficient sleep on youth cognitive development, underscoring the need for interventions.