新药临床试验中，有可能出现受试者没有完全按照随机化分组接受干预，即出现不依从的情况，这时通常会导致结论偏倚。对于非劣效试验中的不依从问题，目前国内外均缺乏有效的解决方法，主要原因在于ITT原则用于非劣效评价时不保守。项目申请者已经提出了一种解决方案，但仅适用于生存数据，无法满足实际需要。本项目将基于Bayes因果模型，给出一种新的因果推断模型及相应的非劣效评价方法。该方法通过不同先验分布类型的选择，对任意分布的连续型和二分类型终点指标均能适用，具有一定的通用性。研究模型针对潜在依从性与疗效结果的联合概率分布建模，通过共轭分布和MCMC方法推断后验分布，得出依从者平均因果效应（CACE）估计量作为干预效应估计结果，避免不依从者的影响；最后针对CACE估计量确定出非劣效界值，构建原假设与备择假设，并给出相应的新统计推断方法，同时将这种新的模型用在正在进行的天麻苄醇酯片的Ⅱ期和Ⅲ期临床试验上。

In clinical trials, non-compliance occurred when subjects failed to comply with their randomized treatment, and non-compliance usually tends to resulting in conclusion bias. In the presence of non-compliance, the ITT analysis is well accepted as being conservative for superiority trials, however, ITT tends to be anti-conservative in demonstrating non-inferiority. Currently, there are rarely papers that have investigated the correcting methods for non-compliance in non-inferiority trials, while there is an urgent need for this. We have developed a method for survival data with non-compliance in non-inferiority trials, however, it cannot be used for data of more general type. In this study we will propose a three-arm Bayesian causal model, where different types of prior distributions for the outcome variables may be chosen. Therefore, this model can be used for non-inferiority trials with continuous endpoints and binary endpoints. A jointly probability distribution modelling approach for the potential compliance and efficacy outcomes is presented; and then basing on Bayes methods, the posterior distribution of CACE causal estimands can be computed. The CACE estimands can then be used to calculate the non-inferiority margin, built and alternative hypotheses, and then complete the non-inferiority assessment. In this study, we will develop a new statistical model and the corresponding methodology for dealing with non-compliance in three-arm designed non-inferiority clinical trials with continuous or binary endpoints. The proposed method will be applied to the phase II and phase III clinical trials of Gastrodia benzyl alcohol ester.