Surgical Site Infections Following Ankle Fractures: A Machine Learning Prediction Model

Research Type: Level 3 - Retrospective cohort study, Case-control study, Meta-analysis of Level 3 studies Introduction/Purpose: Predictive models for Surgical Site Infections (SSIs), one of the most common complications in foot and ankle surgery, may help clinicians take preventive measures and lower the burden on patients and healthcare system. The goal of this study was to determine predictive factors for SSI and to assess the performance of machine learning models in prediction SSI after ankle fractures treated with open reduction internal fixation (ORIF). Methods: In this retrospective study, 114 patients with SSI following their ankle fracture ORIF were assigned to the treatment arm and 821 ankle fractures sans SSI were assigned to the control group. We conducted a multivariate analysis to determine the factors associated with SSI. We additionally developed five machine learning algorithms based on factors that showed significant correlation (p < 0.05) with SSI via independent univariate analyses. Results: Following multivariate logistic regression analysis, factors significantly correlated with SSI were age (OR=1.02, 95%CI 1.00-1.04, p=0.01, r=0.02), external fixation (OR=2.04, 95%CI 1.10-3.77, p=0.02, r=0.71), oral steroid use (OR=2.44, 95%CI 1.19-4.98, p=0.02, r=0.89), open ankle fracture (OR=0.82, 95%CI 1.39-3.75, p=0.001), and smoking (OR=1.72, 95%CI 1.04-2.85 p=0.03, r=0.55). The machine learning models showed areas under the curve (AUC) ranging from 0.61 to 0.85. Gradient-boosting-classifier model outperformed the other models with an AUC of 0.85, 0.91 accuracy, 0.89 F1-score, 0.91 precision, 0.88 sensitivity, and 0.77 specificity. Conclusion: Machine learning models have the potential to identify high risk patients, thereby enabling targeted use of health care resources to enhance patient outcomes. Our algorithms showed an acceptable accuracy in the prediction of post-ORIF SSI; however, improving the accuracy of the model and external validation should be considered to make these models more reliable and generalizable.