Logo image
Back
Classifying Adaptive and Maladaptive Exercise in Binge Spectrum Eating Disorders Using Passive Sensors and Ecological Momentary Assessment
Abstract   Peer reviewed

Classifying Adaptive and Maladaptive Exercise in Binge Spectrum Eating Disorders Using Passive Sensors and Ecological Momentary Assessment

Elizabeth W Lampe, Sasha Gorrell, David Conroy, Fengqing Zhang, Adrienne S Juarascio, Meghan L Butryn and Stephanie M Manasse
Annals of behavioral medicine, v 60(Supplement_1), pp S469-S469
01 Apr 2026
DOI: https://doi.org/10.1093/abm/kaag012#page=S469
url
https://doi.org/10.1093/abm/kaag012#page=S469View
Published, Version of Record (VoR) Open

Abstract

Binge eating Exercise
Up to 60% of individuals with binge-spectrum eating disorders (EDs) engage in maladaptive exercise which is intended to “compensate for” calories consumed and/or feels compulsive (e.g., continuing to exercise when injured). Engagement in maladaptive exercise further reinforces ED symptoms. Notably, up to 40% of individuals with binge-spectrum EDs also engage in exercise that is likely adaptive (i.e., neither compensatory, nor compulsive) and recent research suggests that adaptive exercise has the potential to decrease ED symptoms. Thus, a more nuanced understanding of both adaptive and maladaptive exercise is imperative to optimize ED treatment outcomes. However, existing literature 1) fails to account for the presence of adaptive exercise in ED populations, 2) is based on several disparate theoretical models, each of which may not capture the full spectrum of risk-factors on their own, and 3) relies on retrospective self-report of both exercise and associated behavioral drivers over long time periods (e.g., past 3 months). To address these limitations, we tested a supervised machine learning model to classify exercise episodes as adaptive or maladaptive in a sample of 30 individuals with binge-spectrum EDs. The supervised machine learning model leveraged 28 days of data on 26 transtheoretical drivers from wrist-worn passive sensors and ecological momentary assessment (EMA; 6 prompts per day). The model demonstrated excellent performance (AUC = 0.91), and good sensitivity (0.72) and excellent specificity (0.96). Interpretable machine learning techniques revealed that higher relative autonomous motivation, BMI, and perceived barriers, and lower perceived benefits and exercise duration were associated with adaptive exercise episode classification. Notably, the most important features were overwhelmingly derived from theoretical models of healthy activity promotion delivered outside the ED field. These constructs are seldom measured within ED studies, which suggests that our understanding of adaptive exercise engagement in the context of EDs is critically incomplete. Future work should empirically evaluate the role of such constructs (e.g., autonomous motivation) for helping individuals with EDs to exercise more adaptively and thus recover more completely.

Metrics

1 Record Views

Details

Logo image