Identifying patients at risk for short-term adverse events after hip arthroscopy: a machine learning analysis of a national database.
To develop and compare machine learning-based risk prediction models to identify patients at risk for short-term adverse outcomes (overnight admission, early complication, or readmission) after hip arthroscopy and to determine key predictive demographic and clinical factors.
Data from the American College of Surgeons National Surgical Quality Improvement Program (ACS-NSQIP) database were used to develop and compare risk prediction models aiming to: (1) identify patients likely to experience short-term adverse outcomes including overnight admission, early complication, or readmission; and (2) determine the most predictive demographic and clinical factors contributing to adverse outcomes following hip arthroscopy. Predictive models were developed using support vector machine, random forest, logistic regression, gradient boosting, and extreme gradient boosting methods.
A total of 1478 eligible patients were included (56.4% female, mean age 40.0 ± 14.9 years), of whom 214 (14.5%) experienced a short-term adverse event. Compared to patients with an uncomplicated outpatient surgical course, those experiencing a short-term adverse event exhibited higher rates of diabetes mellitus, hypertension requiring medication, COPD, bleeding disorder, wound class ⩾2, ASA class ⩾3, lower preoperative haematocrit, and longer operative times. Logistic regression produced the optimal model for predicting short-term adverse events (AUC = 0.763), with operative time, preoperative haematocrit, ASA class, surgical procedure (CPT code), and age identified as the strongest predictive features.
These findings demonstrate the value of ML and may assist in predicting surgical outcomes, guiding clinical decision-making, and managing patient expectations regarding their postoperative course.
Data from the American College of Surgeons National Surgical Quality Improvement Program (ACS-NSQIP) database were used to develop and compare risk prediction models aiming to: (1) identify patients likely to experience short-term adverse outcomes including overnight admission, early complication, or readmission; and (2) determine the most predictive demographic and clinical factors contributing to adverse outcomes following hip arthroscopy. Predictive models were developed using support vector machine, random forest, logistic regression, gradient boosting, and extreme gradient boosting methods.
A total of 1478 eligible patients were included (56.4% female, mean age 40.0 ± 14.9 years), of whom 214 (14.5%) experienced a short-term adverse event. Compared to patients with an uncomplicated outpatient surgical course, those experiencing a short-term adverse event exhibited higher rates of diabetes mellitus, hypertension requiring medication, COPD, bleeding disorder, wound class ⩾2, ASA class ⩾3, lower preoperative haematocrit, and longer operative times. Logistic regression produced the optimal model for predicting short-term adverse events (AUC = 0.763), with operative time, preoperative haematocrit, ASA class, surgical procedure (CPT code), and age identified as the strongest predictive features.
These findings demonstrate the value of ML and may assist in predicting surgical outcomes, guiding clinical decision-making, and managing patient expectations regarding their postoperative course.
Authors
Rudisill Rudisill, Hornung Hornung, McCormick McCormick, Streepy Streepy, Nho Nho, Chahla Chahla
View on Pubmed