优效性试验中存在补救治疗时基于假想策略的效应估计方法比较  

A Comparison of Causal Effect Estimation Methods based on Hypothetical Strategy for Superiority Trial with Rescue Treatment

作　　者：甘世林 黄清浩 仲子航 倪森淼 刘文 贺志强 尹健 王媛媛 耿睿 柏建岭 于浩 Gan Shilin;Huang Qinghao;Zhong Zihang(Department of Biostatistics,School of Public Health,Nanjing Medical University(211166),Nanjing)

机构地区：[1]南京医科大学公共卫生学院生物统计学系,211166

出　　处：《中国卫生统计》2023年第3期358-362,共5页Chinese Journal of Health Statistics

基　　金：国家自然科学基金(82273738,81773554);江苏高校“青蓝工程”。

摘　　要：目的探讨优效性临床试验中存在补救治疗时基于假想策略的效应估计方法。方法采用结构方程模型模拟存在补救治疗的优效性临床试验数据,在假想策略下采用重复测量的混合效应模型(MMRM)、基于加权的自然效应模型(WNEM)、基于填补的自然效应模型(INEM)和双稳健估计(DR)四种方法估计处理效应,比较其统计学性能,并应用于真实的糖尿病临床试验分析。结果当两组补救治疗发生比例不同时,WNEM估计偏倚较MMRM和INEM小,DR估计偏倚最小。当两组补救治疗发生比例相同时,四种方法估计偏倚均较小。不同情形下,前三种方法的标准误接近,DR的标准误较小。DR的检验效能优于前三种方法,但存在I类错误率膨胀。结论在满足潜在结果框架基本假设的情形下,自然效应模型方法合理利用了补救治疗后的数据信息,能更好地估计药物的真实疗效。Objective To evaluate the performance of three causal effect estimation methods under the hypothetical strategy in superiority clinical trials with rescue treatment.Methods We use the structural equation models to simulate the superiority clinical trials with rescue treatment.Subsequentially,the causal effects are estimated by mixed model for repeated measures(MMRM),weighting-based natural effect models(WNEM),imputation-based natural effect models(INEM)and doubly robust(DR)estimator under the hypothetical strategy.We further evaluate the performance of above-mentioned four methods and apply these to a real diabetes clinical trial.Results When the proportions of rescue treatment are different in two groups,we show that the causal effect estimation using WNEM are less biased than MMRM and INEM,while DR has a minimum bias.When the proportion of rescue treatment are same in two groups,all four methods have small bias.Additionally,the difference of the standard error of the first three methods is approximative in different scenarios and DR is smaller.Moreover,DR outperform in the power of hypothesis testing than the first three methods,but it′s type I error rate is inflation.Conclusion When the assumptions under the potential outcome framework are met,natural effect models reasonably utilize the information after rescue treatment and lead to a more accurate estimation of the causal effects.

关 键 词：估计目标 假想策略 补救治疗 自然效应模型 

分 类 号：R195.1[医药卫生—卫生统计学]