P864 - MACHINE LEARNING TO PREDICT WEIGHT LOSS WITH DULAGLUTIDE IN WOMEN WITH OVERWEIGHT OR OBESITY

P864

MACHINE LEARNING TO PREDICT WEIGHT LOSS WITH DULAGLUTIDE IN WOMEN WITH OVERWEIGHT OR OBESITY

J. Belkhouribchia^1,*, J. J. Pen²

¹Endocrinology, Endocrinology Center Hasselt, Hasselt, ²Endocrinology, UZ Brussel, Brussels, Belgium

Rationale: Several machine learning models are tested in order to predict whether a patient will achieve 10% weight loss on dulaglutide 1,5mg/week.

Methods: Retrospective data of 252 women in a Belgian endocrinology center were used. Women over 18 years old with a body mass index (BMI) of over 27kg/m² that were treated with dulaglutide once weekly subcutaneously, were included. Patients with diabetes and patients that dropped out in less than 6 months were excluded. Age, BMI, body fat percentage, skeletal muscle mass index (SMI), phase angle, homeostatic model of insulin resistance (HOMA-IR) and the McAuley index were designated independent input variables. An XGBoost (extreme gradient boost) classifier model as well as a neural network and a logistic regression model were tested.

Results: Of the 252 women, 139 (55%) reached 10% weight reduction. The XGBoost classifier achieved the best overall performance with an accuracy of 66.7%, a precision of 0.62, a recall of 0.70, and an F1-score of 0.65 on the test set. The Shapley additive explanations (SHAP) analysis revealed that the McAuley index, BMI, and HOMA-IR were the most influential features in predicting weight loss success, followed by SMI and body fat percentage. The neural network model, achieved a test accuracy of 61.1%, a precision of 0.43, a recall of 0.65, and an F1-score of 0.52. The logistic regression model showed the least favorable performance, with an accuracy of 58.8%, a precision of 0.55, a recall of 0.48, and an F1-score of 0.51. 

Conclusion: This study demonstrates that machine learning models, particularly the XGBoost classifier, can moderately predict which patients will achieve significant weight loss following treatment with dulaglutide based on anthropometric, metabolic, and bio-electrical impedance parameters. The XGBoost model outperformed both the other models. AI-based tools may support individualized decision-making in obesity management.

Disclosure of Interest: None declared