Employing machine learning to predict adverse acute post-surgical outcomes following elective ulnar collateral ligament reconstruction

Abstract

Background: Ulnar collateral ligament reconstruction ameliorates valgus elbow instability in various patient populations, including overhead athletes, patients with acute UCL rupture following high energy trauma, and those with chronic, subclinical elbow laxity. This study aims to explore machine learning algorithms to identify risk factors in patients undergoing elective UCL reconstruction in the ambulatory setting to predict postoperative outcomes. Methods: RStudio was used to create a filtering code to identify adult patients who underwent elective UCL reconstruction from 2008 to 2018 in the American college of surgeons national surgical quality improvement program database. Patients were analyzed using six ML algorithms, which were trained to predict outcomes such as extended length of stay, non-home discharge, and adverse events based on various patient characteristics and surgical variables. Algorithmic performance was then assessed and top performing algorithms underwent further analysis to determine relative feature importance using a permutation feature importance method. Results: ML exhibited excellent performance in predicting LOS, with an average AUC of 0.953, similar to that of logistic regression. Regarding NHD, ML demonstrated a 60.8% increase in AUC compared to LR. In predicting AAE, ML achieved an average AUC that was 12.7% higher than LR. Conclusions: The highly predictive capability of ML indicates the possibility to represent a procedure-specific complementary tool for the preoperative risk stratification process. This study provides a comprehensive analysis of UCL reconstruction in the management and outcomes of any patient, regardless of age or activity level.