Using machine learning methods to predict prolonged operative time in elective total shoulder arthroplasty

BackgroundMachine learning (ML) is a form of artificial intelligence in which computer algorithms improve automatically with experience. Recently, ML has been utilized to predict operative characteristics and patient outcomes for orthopedic procedures, thereby allowing for better patient selection and preoperative planning. This study sought to develop ML models to aid in predicting operative time and 30-day postoperative complications for elective total shoulder arthroplasty and to compare them to regression models.MethodsThis cross-sectional national database study identified patients who underwent elective total shoulder arthroplasty from 2012 to 2018 in the American College of Surgeons National Surgical Quality Improvement Program registry. Boosted decision tree and artificial neural network (ANN) ML models were developed to predict prolonged operative time and 30-day postoperative complication rates. Model performance was measured using the area under the receiver operating characteristic curve and overall accuracy. Multivariate binary logistic regression analyses were also used to identify variables that predicted prolonged operative time and 30-day postoperative complication rates. ML model performance was then compared to the regression models in predicting outcomes.ResultsIn total, 21,544 elective total shoulder arthroplasty procedures met inclusion criteria. Variables associated with greater odds of prolonged operative time included male sex (odds ratio [OR] = 0.66; 95% confidence interval [CI] = 0.61-0.71; P < .001), obesity (OR = 1.19; 95% CI = 1.09-1.29; P < .001), age under 70 years (OR = 0.77; 95% CI = 0.71-0.85; P < .001), smoking history (OR = 1.16; 95% CI = 1.03-1.32; P = .022), and history of cancer (OR = 2.91; 95% CI = 1.52-5.54; P = .001). The boosted decision tree model yielded an area under the curve (AUC) of 0.642, with an overall accuracy of 85.6% for predicting prolonged operative time. The ANN model had an AUC of 0.906 and overall accuracy of 84.7%, while the regression model had an AUC of 0.590 with overall accuracy of 85.6%. Thirty-day complication rate (7.7% vs. 3.9%, respectively; P < .001) and reoperation rate (1.8% vs. 1.2%, respectively; P = .006) also differed significantly between the prolonged operative duration and normal operative duration cohorts.ConclusionThis is the first study to successfully develop ML models for predicting operative time in total shoulder arthroplasty and compare them to existing methods of data analysis. The ANN model was superior to the other models in predicting prolonged operative time. With regard to 30-day postoperative complications, both ML models displayed fair predictive capacity, compared to the regression model, which had poor predictive performance.Level of evidenceLevel III; Database Retrospective Comparative Cohort Study     

Risk factors for instability after reverse shoulder arthroplasty

BackgroundInstability after primary reverse total shoulder arthroplasty (rTSA) is a rare but serious complication, potentially resulting in revision surgery. The causes of instability after rTSA are multifactorial and sometimes unknown. The goal of this study is to analyze an international database of one-platform shoulder prosthesis and conduct a logistic multivariate regression analysis to identify the factors associated with instability after primary rTSA and quantify the 2-year minimum clinical outcomes of patients with and without instability.MethodsA total of 5631 primary rTSA patients were analyzed from the international database of single rTSA prosthesis to quantify clinical outcomes at 2-year minimum follow-up for patients with and without instability. rTSA patients were divided into 2 cohorts based on if they were stable or unstable, and a subanalysis was conducted for patients who were unstable early (<6 months) and also unstable late (>6 months). For both stable and unstable rTSA patients, univariate and multivariate analyses were performed to quantify the patient, implant, and operative risk factors associated with instability after rTSA.ResultsFifty-five of the 5631 primary rTSA shoulders were reported to be unstable, with an overall instability rate of 0.98%. Female patients had an instability rate of 0.60% (21/3496), which was significantly lower (P < .0001) than the 1.63% instability rate for male patients (34/2085). Patients with subscapularis repair had an instability rate of 0.45% (10/2222), which was significantly lower (P = .0052) than the 1.17% instability rate of patients without a subscapularis repair (37/3161). Multivariate analysis identified numerous risk factors for instability, including younger age at the time of surgery, the use of cemented humeral fixation, larger glenosphere diameters, expanded/lateralized center of rotation glenospheres, and not repairing the subscapularis.DiscussionOur study demonstrated that patients with instability had significantly worse clinical outcomes, more pain, and worse function and range of motion as compared to rTSA patients who were stable. The univariate and multivariate analyses identified numerous patient, implant, and operative risk factors associated with instability. A patient with 1 or more of these identified parameters has an increased risk for instability, and that recognition is useful for patient counseling and consideration of repair of the subscapularis, when possible.     

Failed prior rotator cuff repair is associated with worse clinical outcomes after reverse total shoulder arthroplasty

BackgroundThe purpose of this study is to determine the comparative risk profile and clinical outcomes for patients undergoing reverse total shoulder arthroplasty (RTSA) for cuff tear arthropathy (CTA) without failed prior rotator cuff repair (RCR) compared with RTSA for CTA with prior RCR.MethodsFrom 2006 to 2014, all patients who underwent RTSA by two surgeons after failed RCR with minimum 2-year follow-up were identified. Patients who underwent RTSA with failed prior RCR were matched in a 1:1 ratio to patients undergoing primary RTSA, while controlling for demographic factors, prosthesis design, and surgeon. Postoperative active forward elevation and active external rotation were recorded. Outcome measures included American Shoulder and Elbow Surgeons score, Visual Analog Scale (VAS), and Simple Shoulder Test. Perioperative complications and rates of secondary reoperation were noted, and comparative multivariate analysis was performed.ResultsOf 262 patients, 192 (73.3%) were available at minimum 2-year follow-up. The prior RCR group had a significantly higher complication rate (17.4%, n = 15) than the primary RTSA group (3.8%, n = 4) (P = .001), although no significant difference in periprosthetic infection (P = .469) or secondary revision rate (P = .136) was observed. At mean 36.3 ± 26.1-month follow-up, the prior RCR group had statistically worse American Shoulder and Elbow Surgeons score (P < .001), VAS (P = .001), Simple Shoulder Test (P < .001), and active forward elevation (P = .006). Patients with multiple failed RCR attempts (n = 38) before RTSA demonstrated no significant differences versus isolated failed RCR (n = 48; P > .05).ConclusionThis study demonstrated that patients with RTSA after prior failed RCR have significantly worse patient-reported outcomes and greater rate of perioperative complications than patients undergoing primary RTSA for CTA.     

Development of a predictive model for a machine learning–derived shoulder arthroplasty clinical outcome score

BackgroundWe use machine learning to create predictive models from preoperative data to predict the Shoulder Arthroplasty Smart (SAS) score, the American Shoulder and Elbow Surgeons (ASES) score, and the Constant score at multiple postoperative time points and compare the accuracy of each algorithm for anatomic total shoulder arthroplasty (aTSA) and reverse total shoulder arthroplasty (rTSA).MethodsClinical data from 2270 patients who underwent aTSA and 4198 patients who underwent rTSA were analyzed using 3 supervised machine learning techniques to create predictive models for the SAS, ASES, and Constant scores at 6 different postoperative time points using a full input feature set and the 2 different minimal feature sets. Mean absolute errors (MAEs) quantified the difference between actual and predicted outcome scores for each model at each postoperative time point. The performance of each model was also quantified by its ability to predict improvement greater than the minimal clinically important difference (MCID) and the substantial clinical benefit (SCB) patient satisfaction thresholds for each outcome measure at 2-3 years after surgery.ResultsAll 3 machine learning techniques were more accurate at predicting aTSA and rTSA outcomes using the SAS score (aTSA: ±7.41 MAE; rTSA: ±7.79 MAE), followed by the Constant score (aTSA: ±8.32 MAE; rTSA: ±8.30 MAE) and finally the ASES score (aTSA: ±10.86 MAE; rTSA: ±10.60 MAE). These prediction accuracy trends were maintained across the 3 different model input categories for each of the SAS, ASES, and Constant models at each postoperative time point. For patients who underwent aTSA, the XGBoost predictive models achieved 94%-97% accuracy in MCID with an area under the receiver operating curve (AUROC) between 0.90-0.97 and 89%-94% accuracy in SCB with an AUROC between 0.89-0.92 for the 3 clinical scores using the full feature set of inputs. For patients who underwent rTSA, the XGBoost predictive models achieved 95%-99% accuracy in MCID with an AUROC between 0.88-0.96 and 88%-92% accuracy in SCB with an AUROC between 0.81-0.89 for the 3 clinical scores using the full feature set of inputs.DiscussionOur study demonstrated that the SAS score predictions are more accurate than the ASES and Constant predictions for multiple supervised machine learning techniques, despite requiring fewer input data for the SAS model. In addition, we predicted which patients will and will not achieve clinical improvement that exceeds the MCID and SCB thresholds for each score; this highly accurate predictive capability effectively risk-stratifies patients for a variety of outcome measures using only preoperative data.Level of evidenceLevel III; Retrospective Comparative Study