因果森林在医学处理效应异质性评价中的基本原理与应用

Causal forest in the evaluation of heterogeneity of treatment effects in medicine:basic principles and application

随机对照试验是评估干预措施效果的金标准,主要用于估计研究人群整体的平均干预效果。但同一种干预对不同特征人群的效果可能存在显著差异,即存在处理效应异质性。传统的亚组分析和交互作用分析对处理效应异质性的检验功效通常较低,难以有效发现异质性来源。近年来,随着机器学习技术的发展,因果森林成为评估处理效应异质性的新方法,可有效解决传统分析的局限性。然而,该方法在医学研究领域的应用尚处于起步阶段。为促进该方法的合理使用,本文在医学干预效果评价的情境下,介绍因果森林方法的用途、基本原理和实现步骤,并结合实例解读及代码实现样例,讨论使用因果森林方法的注意事项。

Randomized controlled trials are the gold standard for evaluating the effects of medical interventions,primarily providing estimates of the average effect of an intervention in the overall study population.However,there may be significant differences in the effect of the same intervention across sub-populations with different characteristics,that is,treatment heterogeneity.Traditional subgroup analysis and interaction analysis tend to have low power to examine treatment heterogeneity or identify the sources of heterogeneity.With the recent development of machine learning techniques,causal forest has been proposed as a novel method to evaluate treatment heterogeneity,which can help overcome the limitations of the traditional methods.However,the application of causal forest in the evaluation of treatment heterogeneity in medicine is still in the beginning stage.In order to promote proper use of causal forest,this paper introduces its purposes,principles and implementation,interprets the examples and R codes,and highlights some attentions needed for practice.