三臂非劣效临床试验基于干预依从性的效应估计方法  

Estimating Treatment Effects in Three-arm Non-inferiority Clinical Trials based on Compliance of Active Treatments

作　　者：吴研鹏 陈平雁 吴莹 Wu Yanpeng;Chen Pingyan;Wu Ying(Key Laboratory of Public Health Safety of Ministry of Education,Department of Epidemiology,School of Public Health,Fudan University(200032),Shanghai)

机构地区：[1]复旦大学公共卫生学院流行病学教研室公共卫生安全教育部重点实验室,200032 [2]南方医科大学公共卫生学院生物统计学系

出　　处：《中国卫生统计》2024年第2期213-217,222,共6页Chinese Journal of Health Statistics

基　　金：国家自然科学基金(81703322,82273732);广东省医学科研基金资助(A2019438)。

摘　　要：目的临床试验中干预不依从问题通常无法避免,分析方法不当会导致结果偏倚,非劣效研究尤其如此。本研究针对三臂非劣效临床试验,建立Bayes因果模型,实现对干预依从人群的效应估计。方法基于主分层思想,按依从性类型将试验人群分层,从而转化为不同依从性人群的结局混合分布识别问题。通过构建Bayes模型和使用DA(data augmentation)算法,实现结局参数后验分布估计和统计推断。通过模拟研究,比较本研究提出方法与传统的意向性治疗(intention-to-treat,ITT)、遵循研究方案(per-protocol,PP)及实际接受治疗(as-treated,AT)分析的统计性能。结果当依从类型与结局相关时,传统的ITT、PP及AT分析均存在较大偏倚;本研究提出方法在依从类型与结局相关和不相关两种情况下,均能够将偏倚控制在较小的范围。结论对于存在较严重干预不依从问题的非劣效临床试验,本研究提出方法较传统分析可以更好地控制分析偏倚。Objective Non-compliance of active treatments occurred in clinical trials is usually unavoidable,and improper use of standard approaches may lead to biased results,especially for non-inferiority trials.Thus,for three-arm non-inferiority clinical trials,we established a Bayes causal model to estimate causal effects in the presence of non-compliance.Methods Based on the framework of principal stratification,population was stratified according to types of compliance,and the issue was transformed into mixed-distribution identification.Bayes causal model was constructed and data augmentation(DA)algorithm were employed to calculate the posterior distribution of parameters of interest and complete statistical inference.Through simulation,we evaluated performances of our approach,compared with traditional methods including ITT,PP and AT.Results The method of ITT,PP,and AT all had a significant bias when the type of compliance associated with outcomes.The method proposed in this study both had a good performance whether the type of compliance associated with outcomes or not.Conclusion For non-inferiority trials with a high proportion of non-compliance,the method in this article has a better control of the bias.

关 键 词：不依从 三臂非劣效性临床试验 Bayes因果模型 DA算法 

分 类 号：R195[医药卫生—卫生统计学]