Causal inference for longitudinal environmental mixtures

Talk outline

This talk has two main parts. First, it will introduce the methodological challenges of estimating the health effects of longitudinal environmental mixtures, including correlated exposures measured repeatedly over time, time-varying confounding, and feedback between prior exposures and subsequent confounders. These challenges motivate moving beyond cross-sectional mixture models toward longitudinal causal inference frameworks. This part will introduce the causal structure, target estimand, identification assumptions, and methodological framework underlying Time-Varying Quantile G-computation.

Second, the talk will present a case study using Canadian environmental epidemiology data to estimate the joint causal effect of five time-varying PM_2.5 components, black carbon, sulfate, nitrate, ammonium, and organic matter, on nonaccidental mortality. Using the R package tvcQGComp, the hands-on implementation will demonstrate how to:

1. prepare longitudinal data and specify the required models;
2. estimate the cumulative joint effect of simultaneous quantile increases in multiple exposures across repeated time points;
3. adjust for time-varying confounders affected by prior exposure;
4. evaluate effect-measure modification by baseline characteristics; and
5. conduct model diagnostics by comparing simulated natural-course trajectories with observed data and assessing positivity assumptions over time.

Participants will also learn how to interpret and communicate the resulting joint-effect estimates and patterns of differential vulnerability across population groups. The session will conclude with a brief discussion of future directions for causal inference in longitudinal environmental and multi-omics research, followed by an interactive Q&A.