具有部分依从现象的临床试验中因果效应的上下界估计  

Estimation of Boundary of Average Treatment Effects in Clinical Trials with Partial Compliance

作　　者：刘鸿翔[1] 耿直[2] 

机构地区：[1]湖北第二师范学院科研处,武汉430205 [2]北京大学概率统计系,北京100871

出　　处：《湖北第二师范学院学报》2010年第8期1-3,共3页Journal of Hubei University of Education

摘　　要：在随机处理——对照的临床试验中,除出现完全依从和完全不依从的现象外,还会出现部分依从的现象,即患者只服用部分药品。在仅出现完全依从和不依从情况时,Balke and Pearl利用线性规划的方法获得了ACE估计量的上下界,对于部分依从的情况,是将这些数据全部并入完全依从的数据,这样处理的合理性没有论述。同时,利用他们所提供的方法,有时会出现下界为负数,显然,这样的下界没什么实际意义。本文根据Angrist,Imbebns&Rubin讨论工具变量时所提出一些假设条件,导出了在部分依从情况下,计算ACE估计量的上下界的方法,并证明了其下界一定是非负的。In randomized controlled trials,non-compliance or partial compliance of the subjects may often occur,that is,subjects in the treatment group may only take a certain amount of prescribed dose.Balke and Pearl[4] have obtained the upper and lower boundary of ACE under the assumption of non-compliance or total compliance.By their method,they merge date of partial compliance into total compliance;meanwhile,lower boundary may sometimes be negative,which is abnormal.This paper deal with date of partial compliance,proposes a method to calculate estimates of the upper and lower boundary of ACE under the assumptions in Angrist,Imbebns and Rubin[2],and it shows that the lower boundary must be positive and that the interval between the lower and upper boundary is shorter than Balke and Pearl＇s.The paper has also discussed estimation of upper and lower boundary of ACE under partial compliance.

关 键 词：平均因果效应 不依从性 部分依从 上下界 

分 类 号：O21[理学—概率论与数理统计]