Integrating Machine Learning and Dynamic Digital Follow-up for Enhanced Prediction of Postoperative Complications in Bariatric Surgery.

Abstract

Traditional risk models, such as POSSUM and OS-MS, have limited accuracy in predicting complications after bariatric surgery. Machine learning (ML) offers new opportunities for personalized risk assessment by incorporating artificial intelligence (AI). This study aimed to develop and evaluate two ML-based models: one using preoperative clinical data and another integrating postoperative data from a mobile application. A prospective study was conducted on 104 bariatric surgery patients at Saint-Pierre University Hospital (September 2022-July 2023). Patients used the "Care4Today" mobile app for real-time postoperative monitoring. Data were analyzed using ML algorithms, with performance evaluated via cross-validation, accuracy, F1 scores, and AUC. A preoperative model used demographic and surgical data, while a postoperative model incorporated symptoms and mobile app-generated alerts. A total of 104 patients were included. The preoperative model, utilizing Extreme linear discriminant analysis, achieved an accuracy of 75% and an AUC of 64.7%. The postoperative model, using supervised logistic regression with six selected features, demonstrated improved performance with an accuracy of 77.4% and an AUC of 71.5%. A user interface was developed for clinical implementation. ML-based predictive models, particularly those integrating dynamic postoperative data, improve risk stratification in bariatric surgery. Real-time mobile health monitoring enhances early complication detection, offering a personalized, adaptable approach beyond traditional static risk models. Future validation with larger datasets is necessary to confirm generalizability.