Advanced Non-linear Modeling and Explainable Artificial Intelligence Techniques for Predicting 30-Day Complications...

Area of Science:

• Bariatric Surgery Outcomes
• Machine Learning in Healthcare
• Surgical Risk Prediction

Background:

• Metabolic bariatric surgery (MBS) is key for severe obesity management.
• Accurate prediction of post-operative complications is vital for patient safety and selection.
• Current risk scores, like MBSAQIP, assist but may have limitations.

Purpose of the Study:

• To evaluate machine learning (ML) models for predicting 30-day post-operative complications after MBS.
• To compare the performance of various ML models against the established MBSAQIP risk score.

Main Methods:

• Retrospective analysis of 424 MBS patients (2006-2020).
• Evaluation of ML models (logistic regression, SVM, random forest, k-NN, MLP, XGBoost) for predicting 30-day complications (Clavien-Dindo Classification).
• Performance assessment using Area Under the Receiver Operating Characteristic Curve (AUROC) and comparison with MBSAQIP score.

Main Results:

• Random forest model demonstrated superior predictive performance with the highest AUROC (0.94 training, 0.88 validation).
• ML models, particularly random forest, significantly outperformed the MBSAQIP score (AUROC 0.64) in predicting complications (p < 0.001).
• Key predictive features identified by random forest included serum alkaline phosphatase, platelet count, triglycerides, glycated hemoglobin, and albumin.

Conclusions:

• Developed ML models effectively identify patients at risk for 30-day complications post-MBS.
• The random forest model is the top performer, surpassing the current MBSAQIP risk score.
• This advanced ML approach can enhance the pre-operative identification of high-risk individuals for MBS.