Background

Machine learning (ML) offers a novel, comprehensive approach to medical research, particularly in predicting postoperative outcomes in conversion bariatric surgeries. This study hypothesizes that ML can provide enhanced analysis of readmissions, reoperations, and complications within 30 days post-surgery compared to traditional regression analysis.Utilizing the MBSAQIP dataset from 2020-2022, we analyzed data from 38,220 patients (mean age 48.3 years, SD = 10.6; BMI 40.9 kg/m2, range 36.3 to 46.3) who underwent conversion bariatric surgeries. We employed Support Vector Machine and Random Forest models for ML analysis, interpreting results using Variable Importance, SHapley Additive exPlanations plots, and Receiver Operator Curve.Regression analysis revealed significant predictors of adverse outcomes: conversions from gastric bypass surgery notably increased the odds of readmission, reoperation, and overall complications (OR = 2.99, 2.46, 2.45, respectively). Extended operative times (>120 minutes) and therapeutic anticoagulation were also key risk factors. ML models, with an AUC of 61% to 64%, provided a more nuanced understanding by accommodating a wider range of variables affecting outcomes while aligning with regression analysis findings. Both methods identified increased risks in patients with a history of PE/DVT and/or BMI > 50. Interestingly, patients on dialysis showed a protective effect against these adverse outcome predictors. In conclusion, ML demonstrates significant potential in analyzing complex medical data, offering deeper insights than traditional methods. Future research, incorporating diverse data and novel variables, is crucial for enhancing the accuracy and applicability of ML in predicting surgical outcomes.