Causal Mediation Analysis With Recurrent Mediators And Survival Outcome

Recurrent event data are widely available in clinical trials and epidemiological studies,where individuals can repeatedly experience the same type of event during the study period.Recurrent events often accompany terminal events,which in turn hinder follow-up observation of recurrent events.When researchers are interested in the effects of treatment on terminal events and do not want the presence of recurrent events to interfere with final judgment,the recurrent event and its relationship to terminal events will complicate the solution of such effects.Currently,there are many methods for dealing with the recurrent events with dependent terminal event in the field of survival analysis,but there are few causal mediation analysis.For the problem of mediation analysis in which recurrent events are the mediators and survival time is the outcome,this paper discusses the causal diagram under the circumstance where the recurrent event process is a nonhomogeneous Poisson process.Taking into account the characteristics of recurrent events,a joint model for mediation analysis is established using the intensity of the recurrent event process as the mediator and the hazard function of the terminal event as the outcome.The regression parameters are estimated using methods such as B-spline and maximum likelihood estimation.In addition,estimates of natural direct effects and natural indirect effects with causal identification are given,and the consistency of these two types of effect estimates is demonstrated.Finally,this article demonstrates the effectiveness of the model parameter estimation and causal effect estimation through simulation,and applies the method proposed in this article to clinical trial data of the Veterans Administration Department of Urology.