Abstract P348: Weight Loss of Machine Learning Identified Metabolic Subtypes in Response to Dietary Interventions

Introduction: Metabolic syndrome is a constellation of metabolic risk factors. However, various combinations of metabolic disorders may exhibit distinct responses to weight loss interventions. Objective: To identify metabolic subtypes using the machine learning method and assess their associations with weight loss response to dietary interventions. Methods: The study includes 645 participants from the POUNDS Lost trial. Five criteria for metabolic syndrome were used as the grouping factors. Hierarchical clustering was performed with an 8:2 train/test ratio. Results: Three metabolic subtypes were identified and validated among the POUNDS Lost participants. Cluster 1 was characterized by high proportions of central obesity and high blood pressure but low triglycerides; Cluster 2 showed central obesity with relatively low blood pressure; Cluster 3 had the lowest level of central obesity and blood pressure but the highest triglycerides level. Changes in body weight varied significantly across clusters (p=0.003 at 6 months and p< 0.001 at 2 years). At 6 months, adjusted least-square mean (SE) weights were: -6.5 (0.8) kg in cluster 1, -5.4 (0.6) kg in cluster 2, and -4.2 (0.7) kg in cluster 3. Participants regained weight after 6 months, but the difference across the clusters persisted: -5.5 (1.0) kg in Cluster 1, -3.1 (0.8) kg in Cluster 2, and -1.7 (0.8) kg in Cluster 3. Conclusion: We identified 3 metabolic subtypes that predict different responses to dietary weight loss interventions, which may contribute to subtype-specific precision medicine for obesity prevention and management.