Predicting individual responses to treatment for alcohol use disorder.

项目总结： 酒精使用障碍(AUD)的治疗特点是常见的复发，结果的异质性，以及 许多不同的干预措施显示出适度的疗效，但未能相互超越。研究的目的是 解释现有的异质性发现了许多显著的治疗效果调节因素，但很少有这些 具有足够大的效应大小，以表明它们应用于临床实践中的靶向治疗。 新的个性化医疗方法，使用机器学习算法来创建预测 考虑到多个预测因素，对AUD治疗的反应显示出早期的希望。这项研究 这项研究使用了11项随机临床试验，6项行为复发预防计划和5项行为复发预防计划的数据 减少大量饮酒的药物干预，开发和交叉验证个人预测 对酗酒的治疗效果。我们还将测试每个个体差异的重要性 干预并为个人描述他们对不同问题的预期反应提供预测间隔 干预措施。这项研究还旨在测试新的方法，将多个试验的数据结合起来，并 提高预测的精度，以便利用预测的个体治疗效果(PTITE) 框架在临床实践中更有用。 在这项研究结束时，将公布用于比较治疗效果预测的算法 多个治疗的新个体，这些效果的预测间隔，以及对内部 并在可能的情况下，对这些预测进行外部验证。这项工作强调结果的可复制性 通过交叉验证(这本身将通过模拟进行测试)，预测方法的先验规范 和协变量，并使用专家小组作出理论和文献知情的决定。这项研究是 旨在通过集成使治疗AUD的个性化药物在临床实践中使用 理论，临床咨询委员会带来的临床经验，并将结果清楚地传达给 临床观众。

Project summary: Treatment of alcohol use disorder (AUD) is characterized by common relapse, heterogeneity in findings, and many diverse interventions which show modest efficacy but fail to out perform each other. Research aiming to explain the existing heterogeneity has found many significant moderators of treatment effects but few of these have effect sizes large enough to indicate that they should be used in clinical practice for targeting treatments. New personalized medicine methods which use machine learning algorithms to create predictions of responses to AUD treatment which take into account multiple predictors show early promise. This research This research uses data from 11 randomized clinical trials, 6 of behavioral relapse prevention programs and 5 of pharmacological interventions to reduce heavy drinking, to develop and cross validate individual predictions of treatment effects on heavy drinking. We will also test the significance of individual differences for each intervention and provide predictive intervals for individuals describing their expected response to different interventions. The study also aims to test new approaches for combining data across multiple trials and for improving precision of predictions in order to make the use of the predicted individual treatment effects (PITEs) framework more useful in clinical practice. At the end of this study there will be published algorithms for comparing predictions of treatment effects for new individuals across multiple treatments, predictive intervals for those effects, and an assessment of internal and, where possible, external validation of those predictions. The work emphasizes replicability of results through cross-validation (which will itself be tested with simulations), a priori specification of predictive methods and covariates, and use of an expert panel to make theory and literature informed decisions. This research is designed to make personalized medicine for treatment of AUD usable in clinical practice through its integration of theory, clinical experience brought by the clinical advisory board, and clear communication of results to a clinical audience.