Using Smartphone Assessments for Personalized Prediction of Problematic Alcohol Use

项目摘要/摘要 危险饮酒，如酗酒(男性/女性每2小时饮酒5+/4+)和高强度饮酒(2-3倍 酗酒的比率在年轻人中非常普遍，与严重的急性和 较长期的负面行为和健康后果。然而，考虑到它的流行，参与的个人 在这类活动中包括一个不同的群体。研究人员很难辨认出不同的 预测与酒精相关的负面后果和有问题的轨迹的行为过程 跨越时间。预测谁会继续发展成持久的问题，以及谁的高风险饮酒行为 发展受限尤其具有挑战性。部分障碍来自于以下方法： 目前用于研究这种不同的行为。特别是，研究人员经常使用横断面研究来 看一看个人。虽然这获得了关于个人之间的差异的宝贵知识 他们的饮酒模式，它并没有揭示有助于维持这种行为的动态过程 或者让一个人更有可能产生负面后果。然而，越来越多的假设与这些有关 动态过程。这需要对个体的情绪和行为进行量化描述 流程。科学可以朝着更细微的理解不同的 通过得出有效的酒精使用描述而导致有问题的酒精使用的机制 个人级别的(即个性化的)流程。 我们建议在以下四个方面向个性化定量模型迈进：1)开发一个知情的 密集的纵向研究设计，能够在每天的基础上跨时间获取相关变量 和一年的时间跨度；2)使用创新的测量技术，实现客观评估 与饮酒相关的上下文功能；3)使用最先进的电话应用程序收集数据，使 用户不需要接口的自我报告和被动数据采集；4)实现裁剪- 边缘机器学习算法，可以可靠地得到个体级别的检测和预测模型 这可以作为未来适时适应性干预的基础。我们将通过以下方式实现这一目标 招募N=300名年轻成年危险饮酒者，他们将完成120天的动态评估方案 完成对也配备了被动传感器和应用程序的智能手机的调查，然后 提供一年内关于酒精使用和相关变量(例如后果)的4波数据。 最后，我们的努力将创建新的方法来测量和建模相关的行为过程 到抓住每个参与者个性的高风险饮酒。这些努力将为我们提供 准确检测和预测日常饮酒和长期问题饮酒的框架 支持未来科学和临床工作的轨迹。

PROJECT SUMMARY/ABSTRACT Risky drinking, such as binge (5+/4+ drinks per 2-hour occasion for males/females) and high-intensity (2-3x the rates of binge) drinking, is highly prevalent among young adults and associated with severe acute and longer-term negative behavioral and health outcomes. However, given its prevalence, individuals who engage in such activities comprise a heterogeneous group. Researchers have had a hard time identifying the varied behavioral processes that are predictive of negative alcohol-related consequences and problematic trajectories across time. Predicting who will go on to develop lasting problems and whose risky alcohol use behavior is developmentally-limited is especially challenging. Part of the hindrance comes from the methods that are currently used to study this diverse behavior. In particular, researchers often use cross-sectional studies to look across individuals. While this has reaped invaluable knowledge regarding differences among individuals in their drinking patterns, it does not reveal the dynamic processes that contribute to maintaining such behaviors or make one more likely to have negative consequences. However, increasingly hypotheses pertain to these dynamic processes. This requires arriving at quantitative descriptions of individuals' emotional and behavioral processes. The science can move towards a more nuanced understanding of the varied mechanisms contributing to problematic alcohol use by arriving at valid descriptions of individual-level (i.e., personalized) processes. We propose to make advances towards personalized quantitative models in four ways: 1) develop an informed intensive longitudinal research design that enables acquisition of relevant variables across time on a daily basis and across the span of one year; 2) use innovative measurement technologies that enable objective assessment of contextual features related to drinking; 3) collect data using state-of-the-art phone applications that enable self-report and passive data collection where the user does not need to interface; and 4) implement cutting- edge machine learning algorithms that can reliably arrive at individual-level detection and predictive models that can be used as the foundation for future just-in-time adaptive interventions. We will accomplish this by enrolling N=300 young adult risky drinkers who will complete a 120-day ambulatory assessment protocol completing surveys on smartphones that are also equipped with passive sensors and applications, and then provide 4 waves of data on alcohol use and associated variables (e.g., consequences) over one year. In the end, our endeavors will create novel approaches to measuring and modeling behavioral processes related to high-risk drinking that capture the individuality of each participant. These endeavors will provide the framework for accurate detection and prediction of daily drinking and long-term problematic alcohol use trajectories that support future scientific and clinical efforts.