临床试验不依从现象下相关数据的统计推断

Inference of Treatment Effects Based on Correlated Data from Randomized Trials with Non-compliance

假定在临床试验中必需考虑受试者服药状态相关性以及受试者协变量对治疗效果产生的影响。本文证明了治疗效果参数在这一假设下依然具有可识别性。利用依从者治疗效果参数的近似极大似然估计,本文构造了此参数的记分检验统计量。模拟试验表明,即使对较小样本,该检验方法犯第一类错误的概率也接近给定的检验水平。

Assume that in a trial, the compliances among patients are correlated, and the co-variables of patients may affect the evaluation of the treatment. In this paper, we prove the identifiability of the treatment effect. Based on an MLE of parameters, a score test is constructed for the treatment effect. Simulation results show that the score test can control type I error even for small samples.