Utility of adaptive design optimization for developing rapid and reliable behavioral paradigms for substance use disorders

摘要 物质使用障碍(SUD)的一个关键问题是它们的病因和功能异质性，这不是 很好地被当前的精神病因学捕捉到了。一个有影响力的基于神经科学的启发式框架， 成瘾神经临床评估(ANA)提出，为了解决这种异质性，评估 上瘾应该是多维的，并专注于三个关键领域：执行功能(EF)、激励 显著(IS)和负面情绪性(NE)，通过综合的自我报告和 神经行为任务。虽然计算工具增加了从这些任务中提取的知识， 令人惊讶的是，几乎没有高质量的分析方法来监测和表征这些领域。背负着 当前评估电池的管理可能需要长达10个小时，而且大多数评估工具缺乏 在识别潜在病因机制方面的精确性。最重要的是，大多数神经行为和神经成像 任务的重测信度较低，这限制了它们在生物标记物发现中的应用。为了解决这些限制， 我们建议将贝叶斯自适应设计优化(ADO；Myung&Pitt，2009)应用于下列既定任务 索引三个ANA结构域，目标是开发快速、强大和可靠的神经行为探针 这些域。ADO是一种优化数据的通用计算机器学习算法 收集并在尽可能少的试验中从参与者的反应中提取最大的信息。我们的 初步数据表明，ADO导致延迟贴现率在0.95或更高的重测信度下 1-2分钟的测试，在测试-重新测试的可靠性中捕获了大约10%的差异，并且是3-5倍 比传统评估方法更准确，效率高3-8倍(Ahn等人，2020年)。海流 一项研究建议开发和评估一系列基于ADO的任务、软件和移动应用程序，使用State- 最先进的计算方法将显著减少神经认知任务的时间 管理，同时提高任务的可靠性、精确度和效率。要捕捉到的异质性 这个电池将在神经典型个体和几个不同的不同人群中进行测试 三个国家(美国、韩国、保加利亚)的SUD类型(阿片类药物、兴奋剂、酒精和烟草) 我们已经为这类研究开发了基础设施。这一增值观点将对Out-Out有用 对我们的模型进行样本验证，并允许我们不仅解决ANA域的泛化问题 不同类型的SUD，还有跨文化领域的概括性，这一点还没有研究过。 本研究的具体目标是：(1)开发一套可靠、高效的基于ADO的神经行为 ANA领域的任务并评估其在神经典型个体中的重测可靠性；(2)评估预测 通过测试不同类型的SUD患者，新开发的ADO任务对SUD结果的效用；以及 (3)利用新开发的ADO任务设计基于网络的认知测量平台和移动应用程序， 并在开源软件平台上结合ADO等计算方法进行开发。

ABSTRACT A key problem in substance use disorders (SUD) is their etiological and functional heterogeneity, which is not well captured by the current psychiatric nosology. An influential neuroscience-based heuristic framework, Addictions Neuroclinical Assessment (ANA), proposes that to address this heterogeneity, the assessment of addictions should be multi-dimensional and focus on three key domains: executive function (EF), incentive salience (IS), and negative emotionality (NE), assessed with comprehensive batteries of self-report and neurobehavioral tasks. While computational tools have increased the knowledge extracted from these tasks, there are surprisingly few high-quality assays for monitoring and characterizing these domains. The burden of administration of current assessment batteries may take up to 10 hours and most assessment instruments lack precision in identifying underlying etiological mechanisms. Critically, most neurobehavioral and neuroimaging tasks have low test-retest reliability, which limits their utility for biomarker discovery. To address these limitations, we propose to apply Bayesian adaptive design optimization (ADO; Myung & Pitt, 2009) to established tasks that index the three ANA domains, with the goal of developing rapid, robust, and reliable neurobehavioral probes of these domains. ADO is a general-purpose computational machine-learning algorithm that optimizes data collection and extracts the maximal information from participant responses in the fewest possible trials. Our preliminary data show that ADO led to 0.95 or higher test-retest reliability of the delay discounting rate in under 1-2 minutes of testing, captured approximately 10% more variance in test-retest reliability, and was 3-5 times more precise and 3-8 times more efficient than conventional assessment methods (Ahn et al., 2020). The current study proposes to develop and evaluate a battery of ADO-based tasks, software, and mobile apps using state- of-the-science computational approaches that will significantly reduce the time for neurocognitive task administration, while increasing task reliability, precision, and efficiency. To capture the heterogeneity of addiction, this battery will be tested with neurotypical individuals and several diverse populations with different types of SUD (opioid, stimulant, alcohol, and tobacco) in three countries (USA, South Korea, Bulgaria) where we have developed infrastructure for this type of research. This value-added perspective would be useful for out- of-sample validation of our models and allow us to address not only the generalizability of the ANA domains to different types of SUD, but also the cross-cultural generalizability of the domains, which has not been examined. The specific aims of the study are to: (1) Develop a battery of reliable and efficient ADO-based neurobehavioral tasks of the ANA domains and assess its test-retest reliability in neurotypical individuals; (2) Assess the predictive utility of the newly developed ADO tasks for SUD outcomes by testing patients with different types of SUD; and (3) Design web-based platforms and mobile apps for measuring cognition with the newly developed ADO tasks, and open-source software platforms with the ADO and other computational methods we develop.