Title: Estimation of Controlled Direct Effects in Recurrent Event Processes

Abstract: Recurrent events, such as disease episodes or hospital admissions, are frequently observed in medical and epidemiological studies. Investigating how an exposure affects the occurrence of recurrent events is essential for understanding underlying causal mechanisms and making targeted interventions. In this talk, we introduce two mediation analysis methods for recurrent event processes, focusing on two settings: Poisson processes with external covariates only and dynamic modulated Poisson processes with internal covariates. We define the controlled direct effect of an exposure using intensity-based models of recurrent event processes, and provide the assumptions required to identify this effect for the aforementioned recurrent event processes. We demonstrate how sequential G-estimation can be applied to estimate direct exposure effects, and leverage its principle to introduce a novel one-stage estimation method based on the estimating equations framework. We provide asymptotic properties of the estimators and present the results of simulation studies conducted to evaluate the finite sample properties of the proposed estimators, and to assess their robustness against violations of the identifiability assumptions. Finally, we illustrate our method by analyzing hospital readmission data to estimate the controlled direct effect of sex on the readmission rates of colorectal cancer patients.

---

Location: HH-3017

Date and Time: Friday, Sept. 18 at 02:00 PM - 03:00 PM (NDT)

---