Personalized Networks and Sensor Technology Algorithms of Eating Disorder Symptoms Predicting Eating Disorder Outcomes - PROJECT SUMMARY/ABSTRACT Eating disorders (EDs) are severe mental illnesses. Efficacy rates of evidence-based treatments are low (<50% response) and relapse rates are high (>35% relapse after treatment). The low treatment response and high relapse rates are due, in part, to the fact that EDs are heterogeneous conditions. As such, idiographic (i.e., one person) models are needed that can predict and ultimately prevent, onset of both problematic ED behaviors (e.g., purging, binge eating) and remission/relapse. The current renewal application capitalizes on our existing data collection (N=120 ED) to both increase our sample size (N=140) and extend data collection to two years of follow-up. Our study goals are to: (1) characterize and predict shorter-and-longer-term relapse and remission (2) use real-time physiological data algorithms to predict onset of ED behaviors, relapse, and remission. We will use a multiple units of analysis approach combined with novel, cutting-edge advances in idiographic modeling. In our currently funded proposal, we collected intensive real-time data using mobile and sensor-technology from 120 individuals with a diagnosis of anorexia nervosa (AN), atypical AN, and bulimia nervosa across 30 days and assessed follow- up at 1-month and 6-months. In this renewal we will collect additional follow-ups at 18-month and 2-years and include behavioral assessments of body disturbances and behavioral avoidance. We will also collect a new subsample of participants (n=20) and include additional assessment of global positioning system (GPS), the sleep-wake cycle, and circadian rhythm disruption (CR). These additional assessments will improve characterization of relapse, capture a greater percentage of relapse events (~35% across two years), improve accuracy of prediction for ED behaviors, relapse, and remission, and identify which features (e.g., GPS, sleep-wake cycle) contribute to improved accuracy. Specific aims are to: (1) well-characterize longer-term (18 month and 2 years) relapse/remission in the existing sample of EDs, (2) test if both idiographic EMA and physiological (HR/HRV, EDA, ACC) features predict longer-term relapse/remission and (3) determine if the addition of GPS and sleep-wake ACC & CR data improve accuracy of our predictive algorithms. The proposed research uses highly innovative methods, combining intensive longitudinal data collection methods, all remote procedures, novel advances in idiographic modeling and sensor-technology, and state-of-the-art machine learning techniques. These data will lead directly to novel therapeutics such as just-in-time mobile and sensor alert systems that can provide guidance to both clinicians and patients on how to prevent problematic ED behaviors and ultimately increase remission and decrease relapse rates.
