Machine learning models outperform traditional risk scores in bariatric surgery

Advanced machine learning techniques, particularly random forest, demonstrated superior performance in predicting 30-day complications in metabolic bariatric surgery compared to the established MBSAQIP risk score. The random forest model achieved an area under the receiver operating characteristic curve (AUROC) of 0.94 in training and 0.88 in validation. Key predictive features included serum alkaline phosphatase, platelet count, triglycerides, glycated hemoglobin, and albumin, highlighting potential for improved patient risk assessment.