Correcting for Measurement Error in Time-Varying Covariates in Marginal Structural Models

Area of Science:

• Epidemiology
• Biostatistics
• Causal Inference

Background:

• Marginal structural models (MSMs) are crucial for estimating causal effects with time-varying confounders.
• Measurement error in time-varying covariates can compromise the validity of MSM analyses.
• Existing research has limited focus on addressing measurement error within MSMs.

Purpose of the Study:

• To introduce and evaluate novel simulation-extrapolation (SIMEX) approaches for handling measurement error in time-varying covariates within MSMs.
• To compare the performance of direct and indirect SIMEX correction strategies.
• To assess the impact of measurement error on causal effect estimation in time-varying exposure scenarios.

Main Methods:

• Application of the simulation-extrapolation (SIMEX) procedure to address measurement error in time-varying covariates.
• Comparison of two SIMEX approaches: direct correction of outcome model parameters and indirect correction of inverse probability weights.
• Simulation studies under various clinically plausible assumptions to evaluate method performance.

Main Results:

• Measurement errors in time-dependent covariates can introduce substantial bias into MSM estimators of causal effects.
• Both proposed direct and indirect SIMEX approaches provide practically unbiased estimates in the presence of low-to-moderate measurement error.
• The methods were illustrated using data from the Canadian Co-infection Cohort Study.

Conclusions:

• Measurement error in time-varying covariates is a significant challenge for causal inference using MSMs.
• The proposed SIMEX-based methods offer effective solutions for mitigating bias caused by such errors.
• These novel approaches enhance the reliability of causal effect estimates in epidemiological studies with complex time-varying data.