Causal Mediation Analysis Of Survival Outcome With Multiple Mediators

Mediation analysis has been a popular approach to investigate the effect of an exposure on an outcome through a mediator and provides a way to deeper understanding of the mechanism behind human diseases.In medical research,a large number of empirical articles established the mediation model for analysis.Due to the complexity of the researches,one often needs multiple mediators to explain the effect of the exposure on the outcome.In the analysis of survival data,the research and application of multi-mediator models have just begun.There are three methods for multi-mediator analyses using survival models:Aalen additive hazard models,Cox proportional hazard models,and accelerated failure time(AFT)models.Mediation analysis using Cox models under the rare outcome assumption and Aalen additive hazard models consider effects on log hazard ratio and hazard difference,respectively;analysis using accelerated failure time(AFT)models considers effects on the ratio of mean survival times.The study builds multiple mediation models on survival outcome.We compare the type I error and power of the three models through Monte-Carlo simulations.Scenarios are set in different correlation coefficients,effect ratio,censoring rate and so on.We compare the results to provide suggestions for the selection of methods.The simulation results show that it is easier to find the mediating effect of the mediator M2 with the larger correlation coefficient r2.The censoring rate and effect ratio have larger influence on Aalen additive hazard models than the other two models.With the censoring rate decreasing and effect ratio increasing,the type I error of the Aalen models is obviously inflated,so the Aalen models apply only to the high censoring rate and the low effect ratio.The larger the sample size,the more powerful test of three models and tends to stabilize.Cox models are available only under the rare outcome assumption and it is the limitation of its application.With different parameters,AFT models is the most efficient,followed by Cox models,and finally Aalen models.Conclusion:AFT models is superior to the other two methods,and it is recommended for the mediation analysis of survival outcome with multiple mediators.The sample size should be large enough when using mediation analysis.