摘要
目的比较依从者的平均因果效应(CACE)、意向性分析(ITT)、遵循研究方案分析(PP)和接受干预措施分析(AT),在分析随机对照试验不依从数据的效果,探索各种方法的适用条件,为实际数据分析提供科学依据。方法通过SAS软件模拟产生不依从数据,处理措施的因果效应使用CACE、ITT、PP和AT进行估计,以平均偏倚、均方根误差、标准误和检验效能作为评价指标,比较各种方法的估计效果。结果在各种参数组合下,以平均偏倚、均方根误差和检验效能作为评价指标,CACE的估计效果均优于ITT、PP和AT。依从率低于50%时,CACE估计的标准误低于PP,高于ITT和AT;依从率高于50%时,CACE估计的标准误均低于ITT、PP和AT。结论当满足CACE模型假设时,CACE估计随机对照试验不依从数据因果效应的效果优于三种传统分析方法,能够提供更加稳健、无偏的处理效应估计值。
Objective To compare the performance of complier average causal effect（CACE） with that of intention-to-treat analysis （ ITT）, per-protocol analysis （ PP）, and as-treated analysis （AT） in estimating treatment effect of randomized con- trolled trial with non-compliance, to explore the application conditions of each method, and to provide scientific evidence for ana- lyzing practical data. Methods Non-compliance data was simulated using SAS. CACE, ITT, PP and AT were used to estimate treatment effect and their performances were evaluated using bias, root mean square error （ RMSE）, standard error, and pow- er. Results For all combinations of parameters, the performance of CACE was better than that of ITT, PP and AT in bias, RMSE and power. When compliant rate was less than 50% ,the performance of CACE was better than that of PP in standard er- ror,but worse than that of ITT and AT; when more than 50% ,CACE was the best. Conclusion When the model assumptions hold, CACE is better at estimating causal effect than ITT, PP and AT, and can provide an unbiased and robust estimation of treat- ment effect.
出处
《中国卫生统计》
CSCD
北大核心
2015年第4期594-597,共4页
Chinese Journal of Health Statistics
关键词
随机对照试验
不依从
CACE
ITT分析
PP分析
AT分析
Randomized controlled trial
Non-compliance
Complier average causal effect
Intention-to-treat analysis
Per-protocol analysis
As-treated analysis
