Department of Population Health Science and Policy, Icahn School of Medicine at Mount Sinai, New York, NY, USA; Tisch Cancer Institute, Icahn School of Medicine at Mount Sinai, New York, NY, USA; Department of Biostatistics, Columbia University, New York, NY, USA.
Department of Biostatistics, Columbia University, New York, NY, USA.
Contemp Clin Trials. 2019 Oct;85:105830. doi: 10.1016/j.cct.2019.105830. Epub 2019 Aug 27.
This article proposes a method to overcome limitations in current methods that address multiple comparisons of adaptive interventions embedded in sequential multiple assignment randomized trial (SMART) designs. Because a SMART typically consists of numerous adaptive interventions, inferential procedures based on pairwise comparisons of all may suffer a substantial loss in power after accounting for multiplicity. Meanwhile, traditional methods for multiplicity adjustments in comparing non-adaptive interventions require prior knowledge of correlation structures, which can be difficult to postulate when analyzing SMART data of adaptive interventions. To address the multiplicity issue, we propose a likelihood-based omnibus test that compares all adaptive interventions simultaneously, and apply it as a gate-keeping test for further decision making. Specifically, we consider a selection procedure that selects the adaptive intervention with the best observed outcome only when the proposed omnibus test reaches a pre-specified significance level, so as to control false positive selection. We derive the asymptotic distribution of the test statistic on which a sample size formula is based. Our simulation study confirms that the asymptotic approximation is accurate with a moderate sample size, and shows that the proposed test outperforms existing multiple comparison procedures in terms of statistical power. The simulation results also suggest that our selection procedure achieves a high probability of selecting a superior adaptive intervention. The application of the proposed method is illustrated with a real dataset from a depression management study.
本文提出了一种方法，以克服当前方法在解决序贯多重分配随机试验 (SMART) 设计中嵌入的自适应干预措施的多重比较方面的局限性。由于 SMART 通常由许多自适应干预措施组成，因此在考虑多重性后，基于所有干预措施的两两比较的推断程序可能会大大丧失功效。同时，用于比较非自适应干预措施的多重性调整的传统方法需要预先了解相关结构的知识，而在分析自适应干预措施的 SMART 数据时，这可能很难假设。为了解决多重性问题，我们提出了一种基于似然的总体检验，该检验同时比较所有自适应干预措施，并将其用作进一步决策的门控检验。具体而言，我们考虑了一种选择程序，只有当提出的总体检验达到预定的显著性水平时，才会选择具有最佳观察结果的自适应干预措施，以控制假阳性选择。我们推导出了基于样本量公式的检验统计量的渐近分布。我们的模拟研究证实，渐近逼近在中等样本量下是准确的，并且表明所提出的检验在统计功效方面优于现有的多重比较程序。模拟结果还表明，我们的选择程序能够以高概率选择出优越的自适应干预措施。通过对来自抑郁管理研究的真实数据集的应用，说明了所提出方法的应用。
