Passive mobile sensing and machine learning for the detection of drinking episodes

项目总结 动态评估(AA)技术(例如，生态瞬时评估、每日日记、 经验样本)为酒精使用障碍的发展理论提供了关键的测试 (AUD)通过确定个人内部过程(如负面或强化、压力暴露或社交 背景)，这可能会增加问题饮酒的风险，进而导致AUD。AA方法是领先的 推动个性化医疗的方法学方法，因为它提供了一个引人注目的平台 用于评估、诊断、实时监控和及时干预。然而，当前的实用程序 AUD风险个性化模型的AA是有限的，因为危险饮酒及其风险因素(例如 情绪、压力或社交环境的变化)在不同的时间尺度上变化。换句话说，即使是重的 饮酒者可能一周只喝几次酒，但他们的情绪、压力源和社交环境会发生多重变化 一天几次。目前的AA方法依赖于自我报告数据，必须足够频繁地采样才能灵敏 为了改变，足够长的时间来观察足够的饮酒事件，并在这样做的同时避免参与者 精疲力竭。被动移动传感，使用传感器(如GPS、加速度计、测光仪等)可用 在大多数智能手机上，已经在初步研究中显示了预测饮酒发作的可能性， 但这些研究使用的样本相对较少。目前的职业发展奖旨在发展 应聘者在被动移动传感和机器学习方法方面的专业知识 被动移动传感数据。这项研究计划将分析在一项 经常饮酒和吸食大麻的年轻人(18-22岁，n=500；95.2%饮酒者)的大样本， 作为家长R01(DA 047247)的一部分，他们将在连续8个周末使用AA进行跟踪。这项研究 目标是确定饮酒风险因素(压力、社交环境、睡眠、 情绪和冲动状态)，以及饮酒事件本身。候选人将会发展 在这些方法和模型方面的专门知识，将进一步发展旨在 为澳元开发特定人的风险模型。

PROJECT SUMMARY Ambulatory assessment (AA) techniques (e.g., ecological momentary assessment, daily diaries, experiencing sampling) have provided critical tests of theories about the development of alcohol use disorder (AUD) by identifying within-person processes (such as negative or reinforcement, stress exposure, or social context) that can raise the risk for problem drinking and in turn AUD. AA methods are the leading methodological approach in the push towards personalized medicine because it provides a compelling platform for assessment, diagnosis, real-time monitoring, and just-in-time interventions. However, the current utility of AA for personalized models of AUD risk is limited because risky drinking and the risk factors for it (such as changes in moods, stress, or social contexts) change at different scales of time. In other words, even heavy drinkers may only drink a few times a week, but their emotions, stressors and social contexts change multiple times a day. Current AA methods that rely on self-report data have to sample frequently enough to be sensitive to change, long enough to observe sufficient drinking episodes, and to do so while avoiding participant burnout. Passive mobile sensing, which uses sensors (such as GPS, accelerometer, light meter, etc.) available on most smartphones, has been shown in preliminary studies to predict the probability of drinking episodes, but those studies have used relatively small samples. The present career development award aims to develop the candidate’s expertise in passive mobile sensing and the machine learning methods used to analyze passive mobile sensing data. The research proposal will analyze passive mobile sensing data collected in a large sample of regular drinking and marijuana using young adults (age 18 – 22, n = 500; 95.2% who drink), who will be followed using AA over 8 successive weekends as part of a parent R01 (DA 047247). The research goal is to identify passive mobile sensing models of risk factors for drinking (stress, social contexts, sleep, mood, and impulsive states), as well as the drinking episodes themselves. The candidate will develop expertise in these methods and models that will further the development of a research program aimed at developing person specific models of risk for AUD.