Mendelian Randomization With Longitudinal Exposure Data: Simulation Study and Real Data Application

Area of Science:

• Biostatistics
• Genetic Epidemiology
• Causal Inference

Background:

• Mendelian randomization (MR) traditionally uses cross-sectional data, limiting its ability to analyze time-varying effects.
• Estimating causal effects on the mean, slope, and within-individual variability of exposures over time requires advanced methodologies.

Purpose of the Study:

• To develop and validate a multivariable Mendelian randomization (MR) approach using longitudinal summary statistics for time-varying exposures.
• To assess the causal effects on the mean, slope, and within-individual variability of an exposure.

Main Methods:

• Utilized longitudinal summary statistics within a multivariable MR framework.
• Simulated 12 scenarios to evaluate power and type I error rates under varying conditions of shared instruments and regression models.
• Applied the approach to two real-world datasets (POPS and UK Biobank).

Main Results:

• Simulations demonstrated high power to detect causal effects on the mean and slope with strong instruments.
• Causal effects on within-individual variability were detected with low power, particularly when instruments were shared with the mean.
• Real data application identified significant causal estimates for the mean and slope, but weak instruments limited the detection of variability effects.

Conclusions:

• The developed MR approach shows promise for analyzing time-varying exposures, especially with strong genetic instruments.
• Accurate exposure regression model specification and sufficient genetic correlation are critical for reliable results.
• The scarcity of strong instruments in real-world data necessitates cautious interpretation of findings, considering biological context and exposure trajectories.