Predicting Binge and Purge Episodes from Passive and Active Apple Watch Data Using a Dynamical Systems Approach

点击翻译按钮获取中文摘要

PROJECT SUMMARY/ABSTRACT Bulimia nervosa (BN) and binge eating disorder (BED) are life-interrupting and associated with significant impairment. Via a unique opportunity that allowed us to adapt the widely used cognitive-behavioral based app Recovery Record for use on 1000 Apple Watches, we propose to optimize two domains of data being collected over a 30-day period in 1000 individuals with bulimia nervosa (BN) or binge-eating disorder (BED). This proposal augments a parent study [Binge Eating Genetics INitiative (BEGIN)], supported by NIMH (saliva kits for DNA at no cost). We will collect longitudinal passive sensor data via native applications in the Apple Watch and active data on binge-eating, purging, nutrition, mood, and cognitions using Recovery Record adapted for the Apple Watch. We will combine sensor-based measurements of autonomic nervous system (ANS) activity, actigraphy, and geolocation with active Recovery Record measures to characterize real world conditions under which individuals are more/less likely to binge and/or purge in their daily lives. Applying dynamical systems analytic approaches, both across and within individuals, we will identify stable, low-risk, and high-risk patterns that will enable the prediction of transition to high risk epochs that signal impending binge or purge episodes. Our work will provide an empirical foundation for transcending current cognitive- behavioral therapy approaches that are dependent on self-report (often retrospective) of high risk states, will enhance the understanding of eating disorders in terms of regulation, and will yield a personalized precision medicine approach to eating disorders treatment. Efficient and reliable quantitative characterization is the essential first step in the development of real-time interventions driven by automated recognition of individualized transitions into high-risk periods for disordered eating behaviors. Our aims are: 1) To predict the occurrence of binge eating and purging episodes in individuals with BN or BED with passive sensor data; 2) To test theoretically-derived regulatory models of binge eating and purging as reflected in differences in temporal patterns; and 3) To refine our capacity to predict high risk states by augmenting passive data with contextual factors collected by Recovery Record. This proposal optimizes the richness and longitudinal structure of the deep phenotypic data collected in BEGIN to lay the foundation for the next translational step in which we will develop personalized just-in-time interventions that can disrupt eating disorders behaviors in real time before they occur.