Continuous-time mediation analysis for repeatedly measured mediators and outcomes.

Area of Science:

• Causal inference
• Longitudinal data analysis
• Biostatistics

Background:

• Traditional mediation analysis is limited to fixed-time data.
• Longitudinal studies often involve continuous-time processes measured irregularly.
• Dementia research requires methods for time-varying mediators and outcomes.

Purpose of the Study:

• To develop a methodology for estimating causal mechanisms in continuous-time longitudinal data.
• To address time-dependent confounding in mediation analysis.
• To estimate natural and path-specific effects in complex longitudinal settings.

Main Methods:

• Propose a novel methodology for time-fixed exposure, mediator and outcome processes.
• Employ a multivariate mixed-effects model based on differential equations.
• Utilize identifiability assumptions for causal estimands.

Main Results:

• Methodology performance assessed through simulations.
• Illustrated with real-world examples from the 3C cerebral aging study.
• Applied to educational level, depressive symptomatology, cognitive functioning, genetic factors, vascular brain lesions, and neurodegeneration.

Conclusions:

• The proposed method enables causal mechanism estimation in continuous-time longitudinal data.
• It effectively handles time-dependent confounding.
• Applicable to complex health research, including dementia and aging studies.