Determining the Importance of Lifestyle Risk Factors in Predicting Binge Eating Disorder After Bariatric Surgery Using Machine Learning Models and Lifestyle Scores

• Published in: Obesity surgery Vol. 35; no. 4; pp. 1396 - 1406
• Main Authors: Mousavi, Maryam, Tabesh, Mastaneh Rajabian, Moghadami, Seyyedeh Mahila, Saidpour, Atoosa, Jahromi, Soodeh Razeghi
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.04.2025; Springer Nature B.V
• Subjects: Adult; Bariatric Surgery - adverse effects; Binge eating; Binge-Eating Disorder - epidemiology; Binge-Eating Disorder - etiology; Binge-Eating Disorder - psychology; Bulimia; Eating disorders; Exercise; Female; Gastrointestinal surgery; Humans; Life Style; Lifestyles; Machine Learning; Male; Medicine; Medicine & Public Health; Middle Aged; Obesity, Morbid - psychology; Obesity, Morbid - surgery; Risk Factors; Surgery; Lifestyle score; Binge eating disorder; Lifestyle risk factors; Machine learning
• ISSN: 0960-8923; 1708-0428; 1708-0428
• DOI: 10.1007/s11695-025-07765-0

Background This study was conducted to assess the association between lifestyle risk factors (LRF) and odds of binge eating disorder (BED) 2 years post laparoscopic sleeve gastrectomy (LSG) using lifestyle score (LS) and machine learning (ML) models. Methods In the current study, 450 individuals who had undergone LSG 2 years prior to participation were enrolled. BED was assessed using BES questionnaire. The collected data for LRF included smoking, alcohol consumption, physical activity (PA), fruit and vegetable intake, overweight/obesity, and percentage excess weight loss (EWL%). ML models included: logistic regression (LG), KNN, decision tree (DT), random forest (RF), SVM, XGBoost, and deep learning or artificial neurol network (ANN). Additionally, accumulative LRF was assessed using LS. Results One hundred and twenty-two subjects (26.1%) met the criteria for BED 2 years after LSG. Participants who were in the highest quartile of the lifestyle score (nearly worst) had significantly three times higher odds of BED compared to the lowest quartile (nearly optimal) ( p trend = 0.01). Furthermore, RF, LG, SVM, and ANN had the highest accuracy (about 75%) in predicting BED compared to other ML models (between 60 and 72%). Among the lifestyle risk factors, insufficient PA, lower vegetable consumption, a higher level of BMI, and lower EWL% were independently associated with BED ( p  < 0.05). Conclusions Our findings indicate that poor lifestyle patterns are associated with the development of BED, in contrast to non-BED individuals. Given the prevalence of this disorder among LSG participants, lifestyle risk factors must receive special attention after BS.