Identifying Personality-Related Behavioral Phenotypes for Binge Drinking Using Smartphone Sensors and Machine Learning

点击翻译按钮获取中文摘要

PROJECT SUMMARY/ABSTRACT Binge drinking in young adults is a significant public health problem. A major barrier to increasing the efficacy of binge drinking interventions is the heterogeneity between people in predictors of alcohol use/misuse. Treatment can be improved by matching people to interventions based on personality traits that increase risk for binge drinking, but a better understanding of the everyday behaviors linking traits to drinking episodes is needed for such interventions to be effective. Theories of alcohol use/misuse specify multiple behavioral pathways through which personality traits influence problematic drinking, including tendencies to engage broadly in high-risk behavior, self-select into high-risk social drinking contexts, and regulate emotions with alcohol. Such contextualized behavior patterns are key risk factors that can be modified with more personalized treatment. The proposed study will use machine learning methods to identify naturalistic, personality-related behavioral phenotypes that predict binge drinking from smartphone sensor data (e.g., GPS, text/call activity). Data for this project will be drawn from an ongoing NIAAA-funded study of young adults that regularly binge drink (anticipated N = 300). Daily alcohol use and continuous, unobtrusive tracking of smartphone sensor data are collected from participants in the parent study’s 120-day ambulatory assessment protocol. Towards the long-term objective of developing more targeted interventions, this study has 3 specific aims: (1) clarify who is at risk for binge drinking and addressing the problem of recall bias that affects prior research reliant on retrospective reports of alcohol use by establishing associations between personality traits and drinking assessed at the daily level, (2) uncover passively sensed behavioral/contextual risk factors related to personality traits that predict binge drinking with machine learning methods, (3) quantify how much of the relationship between personality traits and binge drinking is explained by passively sensed behavioral phenotypes. The proposed research and training activities will be conducted at the University of Pittsburgh. This fellowship will provide specialized training necessary for the applicant to become an impactful independent clinical scientist. Training will focus on three goals: (1) enhance knowledge of alcohol use etiology/maintenance mechanisms with regular mentor meetings, guided readings, seminars, and journal clubs, (2) gain expertise in applying ambulatory assessment for tracking alcohol use by assisting with the parent study data collection, attending lab meetings, and guided applied practice, and (3) learn machine learning techniques for analyzing passive sensing data with mentored application of methods, relevant courses, workshops, and seminars. Results of the proposed study will advance precision medicine by identifying behavioral markers that can inform development of interventions based on a person’s unique characteristics (NIAAA Strategic Plan Objective 4A).