Using Machine Learning to Predict Clinical Outcomes After Shoulder Arthroplasty with a Minimal Feature Set.

Abstract Background A machine learning analysis was conducted on 5,774 shoulder arthroplasty patients to create predictive models for multiple clinical outcome measures after anatomic Total Shoulder Arthroplasty (aTSA) and reverse Total Shoulder Arthroplasty (rTSA). The goal of this study is to compare the accuracy associated with a full feature set predictive model (e.g. full model =291 parameters) and a minimal feature set model (e.g. abbreviated model =19 input parameters) to predict clinical outcomes in order to assess the efficacy of using a minimal feature set of inputs as a shoulder arthroplasty clinical decision-support tool. Methods Clinical data from 2,153 primary aTSA patients and 3,621 primary rTSA patients were analyzed using the XGBoost machine learning technique to create and test predictive models for multiple outcome measures at different postoperative timepoints using a full and abbreviated model. Mean absolute errors (MAE) quantified the difference between actual and predicted outcomes, and each model also predicted if a patient would experience clinical improvement greater than the minimal clinically important difference (MCID) and substantial clinical benefit (SCB) patient satisfaction anchor-based thresholds for each outcome measure at 2-3 years after surgery. Results Across all postoperative timepoints analyzed, the full and abbreviated models had similar MAE for the American Shoulder and Elbow Surgeons (ASES) (full model = ±11.7 vs. abbreviated model = ±12.0), Constant (±8.9 vs. ±9.8), Global Shoulder Function (±1.4 vs. ±1.5), Visual Analog Scale (VAS) pain (±1.3 vs. ±1.4), active abduction (±20.4o vs. ±21.8o), forward elevation (±17.6o vs. ±19.2o), and external rotation (±12.2o vs. ±12.6o). Marginal improvements in MAE were observed for each outcome measure prediction when the abbreviated model was supplemented with implant size/type data and measurements of native glenoid anatomy. The full and abbreviated models each effectively risk-stratified patients using only preoperative data by accurately identifying patients with improvement greater than the MCID and SCB thresholds. Discussion Our study demonstrated the full and abbreviated machine learning models achieved similar accuracy to predict clinical outcomes after aTSA and rTSA at multiple postoperative timepoints. These promising results demonstrate an efficient utilization of machine learning algorithms to predict clinical outcomes. The use of a minimal feature set of only 19 preoperative inputs suggests that this tool may be easily used during a surgical consultation to improve decision-making related to shoulder arthroplasty.