Using The Super Learner Algorithm To Predict Risk Of 30-day Readmission After Bariatric Surgery In The United States
*Matteo Torquati1, *Morgan Mendis2, *Huiwen Xu3, *Ajay Myneni4, *Katia Noyes4, *Aaron Hoffman4, Philip Omotosho5, *Adan Z Becerra5
1Boston College, Boston, MA;2Ayiti Analytics, Silver Spring, MD;3University of Rochester Medical Center, Rochester, NY;4University at Buffalo, Buffalo, NY;5Rush University Medical Center, Chicago, IL
Objective: Clinical risk prediction models that estimate individual patient probabilities of adverse events, such as readmission, are commonly deployed in bariatric surgery. We sought to validate a machine learning ensemble (Super Learner) model for predicting risk of 30-day readmission after bariatric surgery in comparison with a traditional logistic regression. Methods: This prognostic study used data from the Metabolic and Bariatric Surgery Accreditation and Quality Improvement Program. Patients aged 18-79 who underwent elective laparoscopic gastric bypass or laparoscopic sleeve gastrectomy between 2015-2018 were included. The Super Learner and logistic regression prediction models used 5-fold cross validation and were evaluated using the area under the receiver operating characteristic curve (AUC), the continuous net reclassification index (NRI) for both events and non events, and the integrated discrimination improvement (IDI). Results: The 30-day readmission rate among 393,833 patients (mean age, 45 years; mean preoperative BMI 45.2; 79.9% female) was 3.9%. Super Learner AUC was 0.674 (95% Confidence Interval (CI): 0.670-0.679), compared with 0.650 (95% CI: 0.645-0.654) for traditional logistic regression. The NRI was 0.239 (95% CI: 0.223-0.254), and 0.252 (95% CI: 0.249-0.255) for those who did and did not have a readmission within 30 days. Figure 1 compares the predicted risk for both models, stratified by those who did and did not have a readmission. The IDI was 0.0032 (95% CI: 0.0030, 0.0033). Conclusion: An ensemble machine learning algorithm, the Super Learner, outperformed traditional logistic regression in predicting risk of 30-day readmission after bariatric surgery. Risk prediction tools using the Super Learner may help target high-risk patients more optimally and prevent unnecessary readmissions after bariatric surgery.
Back to 2021 Abstracts