An Innovative Method to Account for Non-Adherence in Treatment Evaluation

描述（由申请人提供）：将个体随机分配至两个或多个治疗组的实验研究设计在临床研究中很常见，并被视为金标准。如果研究参与者没有完全遵守分配的治疗方案，就会出现严重的问题，因为这会破坏对治疗风险和获益的准确评估。在研究酒精依赖治疗效果的试验中，不完美的依从性很常见，需要使用统计模型和方法，这些模型和方法可以正确地将依从行为和健康结果结合起来。有几种方法可用，但它们通常依赖于一些统计假设，这些假设不一定现实或可能无法验证，但会强烈影响估计结果。本研究将开发一种替代和创新的贝叶斯方法，用于在治疗依从性不佳的情况下估计治疗疗效，并将相对于竞争方法对其进行评估。该方法不基于一些常用的统计假设，因此在疗效估计中具有较小的偏倚风险。此外，研究人员可以选择将外部信息（例如来自其他试验的信息）纳入评估中，这可以导致更精确的推断。该方法将应用于对联合收割机研究数据的实证分析，该研究是治疗酒精依赖的最大和最知名的随机试验之一，并包含有关个人依从性的详细信息。总之，本研究有以下两个具体目标：（1）开发一种创新的贝叶斯方法，用于估计治疗依从性不佳的随机研究中的治疗疗效，并比较其相对于竞争方法的性能;（2）在联合收割机研究中应用该方法估计药物和行为干预对酒精依赖的疗效。研究团队由贝叶斯方法、联合收割机研究和酒精研究以及临床试验设计和分析方面的专家组成，能够成功完成拟议的研究。这项工作的结果将为研究人员提供一个有价值的和实用的新工具，在治疗依从性不完善的情况下评估治疗效果。

DESCRIPTION (provided by applicant): Experimental study designs in which individuals are randomly assigned to two or more treatment arms are common in clinical research and considered the gold standard. A serious problem arises if study participants do not fully adhere to their assigned treatment regimen, because this undermines the accurate evaluation of the risks and benefits of treatment. Imperfect adherence is common in trials studying the efficacy of treatments for alcohol dependence and requires the use of statistical models and methods that can properly incorporate both adherence behavior and health outcomes. Several methods are available, but they often rely on a number of statistical assumptions that are not necessarily realistic or may not be verifiable, yet can strongly influence the estimation results. This study wll develop an alternative and innovative Bayesian method for estimating treatment efficacy in the presence of imperfect treatment adherence and will evaluate it relative to competing approaches. This method is not based on some of the commonly invoked statistical assumptions, and therefore carries a smaller risk of bias in the efficacy estimates. Additionally, researcher has the option to incorporate external information, for example from other trials, into the evaluation, which can lead to more precise inference. The method will be applied in an empirical analysis of data from the COMBINE study, which is one of the largest and best known randomized trials for treatment of alcohol dependence and contains detailed information about individual adherence. In summary, this study has the following two specific aims: (1) to develop an innovative Bayesian method for estimating treatment efficacy in randomized studies with imperfect treatment adherence and to compare its performance relative to competing approaches; and (2) to apply this method to estimate the efficacy of pharmacological and behavioral interventions for alcohol dependence in the COMBINE study. The research team, consisting of experts in Bayesian methods, the COMBINE study and alcohol studies research, and clinical trial design and analysis, is in an excellent position to successfully complete the proposed research. The results from this work will provide researchers with a valuable and practical new tool for evaluating treatment efficacy in the presence of imperfect treatment adherence.