Predictors of same-day discharge following bariatric surgery using machine learning

Area of Science:

• Bariatric Surgery Outcomes
• Health Services Research
• Machine Learning in Healthcare

Background:

• Same-day discharge (SDD) after bariatric surgery is a growing trend to reduce healthcare costs.
• Predictors for successful SDD are not consistently identified in existing literature.
• This study leverages machine learning to pinpoint SDD predictors and assess patient/procedure factors.

Purpose of the Study:

• To identify predictors of same-day discharge (SDD) following bariatric surgery using machine learning.
• To evaluate the relative importance of patient- and procedure-related factors in SDD.
• To compare the performance of various machine learning models against traditional logistic regression for predicting SDD.

Main Methods:

• Utilized data from the Metabolic and Bariatric Surgery Accreditation and Quality Improvement Program (2020-2023).
• Applied Synthetic Minority Oversampling Technique (SMOTE) due to the low incidence of SDD.
• Developed and evaluated Random Forest, Naïve Bayes, Neural Network, XGBoost, and CatBoost models, assessing feature importance with SHAP.

Main Results:

• Out of 768,744 bariatric surgery patients, 8.7% (66,809) had SDD.
• Operative duration was the strongest predictor of SDD, outweighing patient comorbidities.
• CatBoost achieved the highest predictive performance (AUC 0.80), significantly outperforming logistic regression (AUC 0.50).

Conclusions:

• A machine learning model was successfully developed to predict same-day discharge after bariatric surgery, surpassing logistic regression.
• Operative duration is the most critical factor influencing discharge status, indicating the significance of intraoperative events.
• Intraoperative factors appear more influential than preoperative patient comorbidities in determining same-day discharge decisions.