疾病预后研究的中介分析方法评价

Evaluation of Mediation Analysis Methods in Survival Data

目的对用于癌症预后研究的五种中介分析方法(VanderWeele法、Baron-Kenny法、Imai法、Sobel法和InverseWeight法)进行评价,为实例分析的方法选择提供依据。方法基于模拟试验,产生不同参数设置下的模拟数据,并评价五种方法的第一类错误、检验效能和分析时间。结果除InverseWeight法在相关系数较大时第一类错误有所膨胀外,其余四种方法的第一类错误在不同参数情况下均在0.05附近。五种方法的检验效能趋势一致,均随着样本量、中介比、总效应的增大而增大,随着删失比的增大而减小。在样本量较小(N=100)且中介比不大于30%的情况下,InverseWeight法的检验效能低于另四种方法。InverseWeight法、Baron-Kenny法和Imai法的分析效率远低于VanderWeele法和Sobel法。结论综合考虑一类错误控制、检验效能及分析效率,推荐VanderWeele法进行预后研究的中介分析。

Objective To evaluate the five different mediation analysis methods for survival data including Vander- Weele method, Baron and Kenny approach, Imai method, Sobel method and Inverse Weight method. Methods Statistical simulationswere performed under different parameter settings to evaluate the methods in type I error,power and analysis efficiency. Results Type I errors of all the five methods are close to the expected level（0. 05 ） except the Inverse Weight method. Increased power was observed according to the increase of sample size,total effect,and proportion of mediation effect,while which was decreased according to the increased censoringrate. However, Inverse Weight method was inferior to the other four methods at the situation of small sample size （such as 100 ）and low mediation proportion （such as 30% ）. The analysis efficiency of Inverse Weight method, Baron-Kenny approach and Imai method were far slower than the other two. Conclusion Considering the type I error, power and test efficiency, we recommend VanderWeele method to analysis survival data.