A Machine-Learning Approach for Estimating Subgroup- and Individual-Level Treatment Effects: An Illustration Using the 65 Trial.
A Machine-Learning Approach for Estimating Subgroup- and Individual-Level Treatment Effects: An Illustration Using the 65 Trial.
复制标题
估计亚组和个人水平治疗效果的机器学习方法：使用65试验的说明。
DOI：
10.1177/0272989x221100717
复制
发表时间：
2022-10
影响因子：
3.6
通讯作者：
O'Neill, Stephen
中科院分区：
文献类型：
--
作者：
Sadique, Zia;Grieve, Richard;Diaz-Ordaz, Karla;Mouncey, Paul;Lamontagne, Francois;O'Neill, Stephen
个性化治疗建议或指南需要关于治疗效果异质性（HTE）的证据。机器学习（ML）方法可以通过考虑许多协变量（包括它们之间复杂的相互作用）来探索HTE。因果机器学习方法可以避免过拟合，当使用相同的数据集通过处理交互项来选择协变量以进行推断并减少对固定参数模型正确规格的依赖时，会出现过拟合。我们研究了因果森林（CF），这是一种基于改进决策树的机器学习方法，可以估计亚组和个人水平的治疗效果，而不需要正确的效果模型预规范。在65试验中，我们将CF与参数方法一起用于估计HTE，该试验评估了65岁或以上患有血管扩张性低血压的危重患者的90天死亡率，其中允许低血压策略与常规护理的效果。这里，CF方法对预先指定和事后亚组的治疗效果提供了与参数方法相似的估计，并且对总体HTE的测试结果显示出微弱的异质性证据。个体水平治疗效果的CF估计，即由协变量定义的亚群中个体治疗的预期效果，表明容许性低血压策略有望降低98.7%患者的90天死亡率，但95%置信区间包括71.6%患者的零。然后使用ML方法来评估与这些个体水平影响相关的患者特征，并帮助确定未来研究的目标，以确定干预对哪些患者亚组最有效。本文研究了一种因果机器学习方法，即因果森林（CF），用于探索治疗效果的异质性，而无需预先指定特定的功能形式。在65试验的再分析中考虑了CF方法，并发现它提供了与使用固定参数模型相似的亚组效应估计。CF方法还提供了个体水平治疗效果的估计，表明对于65试验中的大多数患者，干预有望降低90天死亡率，但存在很大的统计不确定性。该研究说明了如何分析个体水平的治疗效果评估，从而为进一步研究那些可能从干预中获益最多的患者产生假设。
Personalizing treatment recommendations or guidelines requires evidence about the heterogeneity of treatment effects (HTE). Machine-learning (ML) approaches can explore HTE by considering many covariates, including complex interactions between them. Causal ML approaches can avoid overfitting, which arises when the same dataset is used to select covariate by treatment interaction terms as to make inferences and reduce reliance on the correct specification of fixed parametric models. We investigate causal forests (CF), a ML method based on modified decision trees that can estimate subgroup- and individual-level treatment effects, without requiring correct prespecification of the effect model. We consider CF alongside parametric approaches for estimating HTE, within the 65 Trial, which evaluates the effect of a permissive hypotension strategy versus usual care on 90-d mortality for critically ill patients aged 65 y or older with vasodilatory hypotension. Here, the CF approach provides similar estimates of treatment effectiveness for prespecified and post hoc subgroups to the parametric approach, and the results of a test for overall HTE show weak evidence of heterogeneity. The CF estimates of individual-level treatment effects, the expected effects of treatment for individuals in subpopulations defined by their covariates, suggest that the permissive hypotension strategy is expected to reduce 90-d mortality for 98.7% of patients but with 95% confidence intervals that include zero for 71.6% of patients. A ML approach is then used to assess the patient characteristics associated with these individual-level effects, and to help target future research that can identify those patient subgroups for whom the intervention is most effective. This article examines a causal machine-learning approach, causal forests (CF), for exploring the heterogeneity of treatment effects, without prespecifying a specific functional form. The CF approach is considered in the reanalysis of the 65 Trial and was found to provide similar estimates of subgroup effects to using a fixed parametric model. The CF approach also provides estimates of individual-level treatment effects that suggest that for most patients in the 65 Trial, the intervention is expected to reduce 90-d mortality but with wide levels of statistical uncertainty. The study illustrates how individual-level treatment effect estimates can be analyzed to generate hypotheses for further research about those patients who are likely to benefit most from an intervention.