Predictability of Same-Day Discharge Candidacy After Bariatric Surgery with Machine Learning

Despite growing interest, same-day discharge (SDD) after bariatric surgery remains uncommon due to challenges with patient selection. In this study, we seek to evaluate whether Metabolic and Bariatric Surgery Accreditation and Quality Improvement Program (MBSAQIP) database provides sufficient information for predicting SDD candidacy METHODS: In this retrospective cohort study, we identified average-risk patients who underwent gastric bypass or sleeve gastrectomy in the 2023 MBSAQIP database. We assigned SDD candidacy to patients who were discharged home within a day and experienced no 30-day readmission. We used five predictive modeling approaches: logistic regression, random forest, gradient boosting, extreme gradient boosting (XGBoost), and light gradient boosting machine. Hyperparameter optimization was conducted with a Bayesian optimization framework.

Among the 47,071 included patients, about 70.1% of gastric bypasses and 80.4% of sleeve gastrectomies were deemed SDD candidates. SDD candidates were more commonly males and had less comorbidities. Despite extensive tuning and optimization, all five models performed poorly in predicting SDD candidacy. For gastric bypass, XGBoost performed the best, with an Area Under ROC curve (AUROC) of 0.519, sensitivity/specificity of 0.690/0.329, and Kappa of 0.018. For sleeve gastrectomy, logistic regression performed the best, with an AUROC of 0.538, sensitivity/specificity of 0.609/0.451 and Kappa of 0.042.

As of 2023, more than half of average-risk bariatric patients may be safe for SDD. However, predictive artificial intelligence could not reliably identify SDD candidacy using MBSAQIP data.