Causal Inference
Novel g-computation algorithm for time-varying actions with recurrent and semi-competing events Alena Sorensen D’Alessio* Alena D’Alessio D’Alessio D’Alessio D’Alessio D’Alessio University of North Carolina at Chapel Hill
Background: A core aspect of epidemiology is determining the impacts of various interventions in public health. Suppose we were interested in how preventing cigarette smoking throughout young-adulthood and mid-adulthood might impact later-life hypertension. Because cigarette smoking is a time-varying action, we must address two challenges: 1) time-varying confounding; and 2) death before end of follow-up. Standard methods do not address both time-varying confounding and semi-competing events (where a terminal event (e.g., death) precludes an intermediate event (e.g., hypertension), but not the reverse).
Methods: We propose an iterated conditional expectation g-computation algorithm for causal effects with Time-varying Actions with Semi-competing Events (TASE). To explore the performance of our novel TASE g-computation estimator, we conducted a Monte Carlo simulation study. We also applied TASE g-computation in the context of cigarette smoking prevention and prevalent hypertension using data from Waves III (aged 18-26 years) – VI (aged 39-51 years) of the National Longitudinal Study of Adolescent to Adult Health.
Results: Our simulations show the TASE g-computation estimator has little bias and appropriate confidence interval coverage (Figure 1). It outperformed existing alternative estimators across sample sizes. In the illustrative application, TASE g-computation identified a small reduction in prevalence of hypertension and risk of death in mid-life had all cigarette smoking been prevented across follow-up compared to the natural course of smoking. The alternative estimators had reduced precision and diverging estimates.
Conclusion: As long-running cohorts progress in age, death within the study sample will become an increasing concern for studies of aging-related outcomes, life course analyses, and investigations into chronic disease development. TASE g-computation provides a solution for both time-varying confounding and semi-competing events.
