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

Area of Science:

• Bariatric Surgery Outcomes
• Health Informatics
• Predictive Analytics in Healthcare

Background:

• Same-day discharge (SDD) after bariatric surgery is underutilized due to patient selection challenges.
• The Metabolic and Bariatric Surgery Accreditation and Quality Improvement Program (MBSAQIP) database is a potential source for predicting SDD candidacy.
• This study evaluates the predictive capability of MBSAQIP data for SDD in bariatric surgery patients.

Purpose of the Study:

• To assess if the MBSAQIP database contains sufficient information to predict same-day discharge (SDD) candidacy after bariatric surgery.
• To evaluate the performance of various machine learning models in identifying SDD candidates.

Main Methods:

• Retrospective cohort study using the 2023 MBSAQIP database.
• Identified average-risk patients undergoing gastric bypass or sleeve gastrectomy.
• Defined SDD candidacy by discharge within 24 hours and no 30-day readmission.
• Applied five predictive models: logistic regression, random forest, gradient boosting, XGBoost, and light gradient boosting machine.

Main Results:

• Out of 47,071 patients, 70.1% of gastric bypass and 80.4% of sleeve gastrectomy patients met SDD criteria.
• SDD candidates were more likely to be male with fewer comorbidities.
• All five predictive models demonstrated poor performance in identifying SDD candidacy, with the best models achieving only marginal predictive power (AUROC < 0.54).

Conclusions:

• Over half of average-risk bariatric surgery patients may be suitable for same-day discharge.
• Predictive artificial intelligence models using MBSAQIP data were not reliable for identifying SDD candidacy in 2023.
• Further research is needed to develop accurate predictive tools for SDD in bariatric surgery.