Use of Machine Learning Theory to Predict the Need for Femoral Nerve Block Following ACL Repair

Abstract

Objective. We report on a classification approach using machine learning (ML) algorithms for prediction of postoperative femoral nerve block (FNB) requirement following anterior cruciate ligament (ACL) reconstruction. Background. FNBs are commonly performed for ACL reconstruction to control postoperative pain. Ideally, anesthesiologists would target preoperative FNB only to ACL reconstruction patients expected to experience severe postoperative pain. Perioperative factors associated with postoperative FNB placement following ACL reconstruction remain unclear, may differ among separate surgical facilities, and render such predictions difficult. Methods. We conducted a chart review of 349 patients who underwent ACL reconstruction at a single outpatient surgical facility. Standard perioperative data commonly available during routine preoperative examination were recorded. ML classifiers based on logistic regression, BayesNet, multilayer perceptron, support vector machine, and alternating decision tree (ADTree) algorithms were then developed to predict which patients would require postoperative FNB. Results. Each of the ML algorithms outperformed traditional logistic regression using a very limited data set as measured by the area under the receiver operating curve, with ADTree achieving the highest score of 0.7 in the cross-validated sample. Logistic regression outperformed all other classifiers with regard to kappa statistics and percent correctly classified. All models were prone to overfitting in comparisons of training vs cross-validated samples. Conclusion. ML classifiers may offer improved predictive capabilities when analyzing medical data sets compared with traditional statistical methodologies in predicting severe postoperative pain requiring peripheral nerve block. Wiley Periodicals, Inc..