Statistical and Machine Learning Methods for Integrating Clinical and Multimodal Imaging Data to Select Optimal Antidepressant Treatment

摘要：严重抑郁障碍(MDD)的公共健康负担是巨大的，目前的治疗方法 对于抗抑郁药物的选择治疗效果有限。据估计，只有不到三分之一的人 MDD患者会对他们开出的抗抑郁药有反应，对有效治疗的追求是 典型的以漫长的反复试验为特征的。需要确定患者的特征 (生物标志物)可以用来客观地选择个性化的抗抑郁药物治疗是明确的。 因此，大型临床研究，如NIMH资助的建立主持人和生物签名 临床治疗的抗抑郁反应(EMBARC)研究收集了大量的基线 包括来自各种神经成像来源的措施，希望其中一些可以被用来指导 抗抑郁药物治疗选择。这些数据带来了许多尚未解决的统计挑战 有效地解决了问题。这些挑战包括(1)处理高维数据，(2)处理数据 错失性，以及(3)确定如何最好地同时对来自 多种成像方式和感兴趣的反应。这个项目的目标是获得必要的 培训和经验，通过应对每一项挑战，在这一领域取得重大进展。目标 该项目的1将采用最先进的集成机器学习算法和有针对性的估计 使用分级临床、人口学和总结确定抗抑郁药物治疗效果的调节因素 包括EMBARC在内的抗抑郁药物临床试验的神经成像数据。处理策略 还将对这方面丢失的数据进行调查，并提出最佳做法准则。目标2 将扩展AIM 1中使用的方法，并开发用户友好的软件，以直接将高 将多模式神经影像数据维化成治疗决策规则。这一目标中将包括一个 在估计的情况下处理丢失的高维成像数据的最佳做法的研究 治疗决策规则。目标3将采用目标2中开发的新方法，并估计 治疗决策规则将被评估，并与目标1中制定的规则进行比较。我已经收集了一个 直接支持完成这些研究目标的培训计划。它包括指令， 指导和实践经验(1)精神病理学和精神障碍的神经基础和 这些障碍的治疗；(2)使用神经成像数据来了解抑郁和对 抗抑郁治疗；(3)使用现代算法存储、处理、处理和分析大 生物医学数据，如在多模式神经成像研究中出现的数据。这位K01指导研究科学家 发展奖将提供培训、时间和资源，以便能够在以下方面取得实质性进展 解决这一重要问题，并将提供在我的过渡中至关重要的技能和经验 交给一个独立的调查员。

Summary: The public health burden of major depressive disorder (MDD) is immense and current approaches for selecting antidepressant treatment have had limited success. By some estimates, fewer than one in three MDD patients will respond to their prescribed antidepressant and the quest for a treatment that will work is typically characterized by a lengthy course of trial-and-error. The need to identify patient characteristics (biomarkers) that can be used to objectively select personalized antidepressant treatment is clear. Accordingly, large clinical studies like the NIMH-funded Establishing Moderators and Biosignatures of Antidepressant Response for Clinical Care (EMBARC) study have collected massive amounts of baseline measures including those from various neuroimaging sources in the hope that some can be used to guide antidepressant treatment selection. These data bring with them many statistical challenges that have yet to be effectively addressed. These challenges include (1) dealing with high-dimensionality, (2) handling data missingness, and (3) determining how best to simultaneously model relationships between measures from multiple imaging modalities and the response of interest. The goal of this project is to acquire the essential training and experience to make significant progress in this area by addressing each of these challenges. Aim 1 of this project will employ state-of-the-art ensemble machine learning algorithms and targeted estimation to identify moderators of antidepressant treatment effect using scalar clinical, demographic, and summary neuroimaging data from clinical trials of antidepressant treatments, including EMBARC. Strategies for handling missing data in this context will also be investigated and guidelines on best practices will be proposed. Aim 2 will extend the methods used in Aim 1 and develop user-friendly software to directly incorporate high- dimensional multimodal neuroimaging data into treatment decision rules. Included in this aim will be an investigation into best practices for handling missing high-dimensional imaging data in the context of estimating treatment decision rules. Aim 3 will employ the novel methods developed in Aim 2 and the estimated treatment decision rules will be evaluated and compared with those developed in Aim 1. I have put together a training program that directly supports the completion of these research aims. It includes instruction, mentoring, and hands-on-experience (1) in psychopathology and the neural basis for psychiatric disorders and treatment for those disorders; (2) in the use of neuroimaging data to understand depression and response to antidepressant treatment; (3) in the use of modern algorithms to store, process, manipulate, and analyze big biomedical data like those arising in multimodal neuroimaging studies. This K01 Mentored Research Scientist Development Award will provide the training, time, and resources to be able to make substantial progress in addressing this important problem and will provide the skills and experience that will be crucial in my transition to an independent investigator.