Smartphone sensors to detect shifts toward healthy behavior during alcohol treatment

摘要 酗酒(4+/5+饮品/场合，女性/男性)会增加可预防的酒精相关风险 后果，特别是在年轻人(18-25岁)中。作为我们NIAAA资助机制的一部分 酒精治疗改变(匹配)短信干预随机对照试验(R01 AA023650)， 我们被资助(CTSI试点)收集(但不分析)智能手机传感器(例如，GPS、通信日志) 和关于饮酒行为的生态瞬时评估(EMA)数据在匹配的子样本(N=108)中。 此二次数据分析R21将利用在 一项酒精临床试验的背景，以获得对潜在行为变化过程的新见解。电话传感器 比赛期间收集的数据为一个人的日常旅行提供了细粒度的客观衡量 模式和访问的地方，以及社交能力(交流模式)。这些细粒度的数字痕迹或数字 表型提供了年轻人日常生活(例如，旅行、社交)如何改变的客观标志 与TM干预反应的关系。电话传感器数据提供了一种方法来客观地确定时间和 行为的变化如何与治疗效果相关，这将告知基于手机传感器的个人- 数字化干预的下一次迭代的实现。建议的二级分析重点放在比赛上 带有EMA数据和电话传感器数据的亚样本(n=93；71%女性，范围18-25)。电话传感器(例如， GPS、加速度计、通信日志)和EMA数据在14周内收集(2周磨合+12周- 每周干预)。为期两周的“磨合”提供了一个基准每日“例行公事”由电话传感器推断之前 干预。与我们以前的工作一样，我们确定干预“响应者”和“非响应者”类别；或探索 通过降低世界卫生组织的饮酒风险水平来定义应对措施。目标1比较治疗 在TM干预之前，应答者和非应答者的数字表型(例如，旅行、通信)。 目的2比较TM干预过程中应答者和非应答者的数字表型。探索性的 AIM使用分组迭代多模型估计(GIMME)来同时估计 特定于个人(具体分析)和组(响应者/非响应者)的选定电话传感器功能 TM干预期间的水平(以补充目标1和2中基于人口的分析)。探索性的 分析提供详细的个人层面的信息，以指导干预内容的个性化，而 还指示在组级别共享的关联。分析还将探索性别差异， 时间效应(例如，周末/工作日)和治疗臂。这些创新的辅助数据分析， 符合NIAAA推进个性化医疗的战略计划，将(1)确定时间和方式 年轻人饮酒行为的变化与TM干预有关，揭示了替代健康 应答器中的例程；以及(2)为使用电话传感器数据进行个性化的R01项目提供基础 该建议由酒精短信干预，以优化数字干预效果。

ABSTRACT Binge drinking (4+/5+ drinks/occasion for females/males) increases risk for preventable alcohol-related consequences, particularly among young adults (ages 18-25). As part of our NIAAA-funded Mechanisms of Alcohol Treatment Change (MATCH) text message intervention randomized controlled trial (R01 AA023650), we were funded (CTSI pilot) to collect (but not analyze) smartphone sensor (e.g., GPS, communication logs) and Ecological Momentary Assessment (EMA) data on drinking behavior in a MATCH subsample (N=108). This secondary data analysis R21 will leverage the unique combination of phone sensor data collected in the context of an alcohol clinical trial to gain new insight into processes underlying behavior change. Phone sensor data collected during MATCH provides fine-grained objective measures of a person's daily routine in travel pattern and places visited, and sociability (communication pattern). These fine-grained digital traces or digital phenotypes provide objective markers of how a young adult's daily routine (e.g., travel, sociability) changes in relation to response to TM intervention. Phone sensor data provide a means to objectively determine when and how shifts in behavior occur in relation to treatment effects, which will inform phone-sensor-based personal- ization of the next iteration of the digital intervention. Proposed secondary analyses focus on the MATCH subsample (n=93; 71% female, range 18-25]) with EMA data and phone sensor data. Phone sensor (e.g., GPS, accelerometer, communication logs) and EMA data were collected over 14 weeks (2-week run-in + 12- week intervention). The 2-week "run-in" provides a baseline daily "routine" inferred by phone sensors prior to intervention. As in our prior work, we identify intervention "responder" and "non-responder" classes; or explore defining response using reduction in World Health Organization risk drinking level. Aim 1 compares treatment responders and non-responders on digital phenotypes (e.g., travel, communication) prior to TM intervention. Aim 2 compares responders and non-responders on digital phenotypes during TM intervention. An exploratory aim uses group iterative multiple model estimation (GIMME) to simultaneously estimate associations between selected phone sensor features at person-specific (idiographic analysis) and group (responder/non-responder) levels during TM intervention (to complement population-based analyses in Aims 1 and 2). Exploratory analyses provide detailed individual-level information to guide personalization of intervention content, while also indicating associations that are shared at the group-level. Analyses also will explore gender differences, time effects (e.g., weekend/weekday), and treatment arm. These innovative secondary data analyses, which are in line with NIAAA's strategic plan to advance personalized medicine, will (1) determine when and how shifts in young adults' drinking behavior occur in relation to TM intervention, revealing alternative healthy routines in responders; and (2) provide the basis for an R01 project that uses phone sensor data to personalize the recommendation made by an alcohol text message intervention to optimize digital intervention effects.