EP8.87 Individualized Forecasting of the Ten-Year Risk of Revision Hip Arthroscopy: A Machine Learning Analysis
Abstract
Abstract Background Recent literature has explored whether arthroscopic hip preservation surgery can produce sustainable outcomes at long-term follow-up; however, these reports remain limited by population-level evaluation of outcomes and do not allow for individualized determination of clinical outcome improvement or survivorship. The purpose of the current study was to forecast the patient-specific risk of undergoing revision hip arthroscopy at a minimum of 10-year follow-up. Methods Consecutive patients undergoing hip arthroscopy for femoroacetabular impingement syndrome (FAIS) between 2012-2015 at a large, tertiary academic hospital were included. A total of 57 potential predictive features from a secure registry were evaluated for inclusion in six supervised machine learning (ML) models trained on the dataset. The primary outcome was the incidence of revision hip arthroscopy at 10-year follow-up. Algorithm performance was evaluated through discrimination, calibration, decision-curve analysis, and Brier score. Results A total of 584 patients were included of which 46 (7.9%) underwent revision hip arthroscopy. A combination of five features optimized algorithm predictions: On the independent testing set, a stochastic gradient boosting (SGB) model demonstrated superior performance among all ML models (C-statistic of 0.84, calibration intercept and slope of 0.19 and 1.03, and Brier score 0.061) and to a traditional logistic regression. Age less than 18 years and greater than 50 years, greater than two prior surgeries, medial joint space with (JSW) less than 4.5 millimeters, and greater than one preoperative corticosteroid injection were specifically associated with increased risk of revision hip arthroscopy at 10-year follow-up. Tönnis angle demonstrated variable associations in this model, with angles approaching 0 and greater than 10 demonstrating consistently increased risk. Conclusions A prognostic ML algorithm demonstrated good performance for predicting the long-term risk of revision hip arthroscopy. Long-term risk of revision may be influenced by age, prior hip surgery, number of injections, medial joint space width, and measures of hip dysplasia.