One of the main goals of sequential, multiple assignment, randomized trials (SMART) is to find the most efficacious design embedded dynamic treatment regimes. The analysis method known as multiple comparisons with the best (MCB) allows comparison between dynamic treatment regimes and identification of a set of optimal regimes in the frequentist setting for continuous outcomes, thereby, directly addressing the main goal of a SMART. In this paper, we develop a Bayesian generalization to MCB for SMARTs with binary outcomes. Furthermore, we show how to choose the sample size so that the inferior embedded DTRs are screened out with a specified power. We compare log-odds between different DTRs using their exact distribution without relying on asymptotic normality in either the analysis or the power calculation. We conduct extensive simulation studies under two SMART designs and illustrate our method's application to the Adaptive Treatment for Alcohol and Cocaine Dependence (ENGAGE) trial.
Confidence Adjusted Surprise Measure for Active Resourceful Trials (CA-SMART): A Data-driven Active Learning Framework for Accelerating Material Discovery under Resource Constraints
West Virginia University（西弗吉尼亚大学）；The University of Texas at San Antonio（德克萨斯大学圣安东尼奥分校）
Sample size estimation for comparing dynamic treatment regimens in a SMART: a Monte Carlo-based approach and case study with longitudinal overdispersed count outcomes
Sample size considerations for comparing dynamic treatment regimens in a sequential multiple-assignment randomized trial with a continuous longitudinal outcome
Robust Hybrid Learning for Estimating Personalized Dynamic Treatment Regimens
动态治疗方案（DTR）是在每个阶段根据可能随时间变化的患者特征和前一阶段观察到的中间结果定制的顺序决策规则。许多疾病和病症的复杂性、患者异质性和慢性性要求学习最佳DTR，其随着时间的推移最好地动态调整治疗以适应每个个体的反应。个性化数据的激增（例如，遗传和成像数据）提供了深度定制的机会以及对统计方法学的新挑战。在这项工作中，我们提出了一种强大的混合方法，称为增强多阶段结果加权学习（AMOL），将结果加权学习和Q学习相结合，以确定最佳DTR序列多分配随机试验（SMART）。我们推广了结果加权学习（O-learning; Zhao et al. 2012），以允许负面结果;我们提出了减少O学习中权重变化的方法，以实现数值稳定性和更高的效率;最后，对于多阶段SMART研究，我们将双重鲁棒增强引入到基于机器学习的O学习中，通过在每个阶段从基于回归模型的Q学习中提取信息来提高效率。即使Q学习模型被错误指定，所提出的AMOL仍然有效。我们建立了AMOL的理论性质，包括估计规则的一致性和收敛到最优值函数的速度。AMOL相对于现有方法的比较优势在广泛的模拟研究和两个SMART数据集的应用中得到了证明：注意力缺陷和多动障碍（ADHD）的两阶段试验和重度抑郁症（MDD）的STAR*D试验。
