Poster 179: Unsupervised Machine Learning to Identify Clinically Meaningful Subgroups in Patients Undergoing Arthroscopic Rotator Cuff Repair

OBJECTIVES: Rotator cuff tears are estimated to affect 20.7% of the population, with the prevalence increasing with age. Surgery is indicated after non-response to nonoperative treatment, with arthroscopic rotator cuff repair (ARCR) as the current standard for full thickness tears. Clinically significant outcomes (CSOs) of ARCR can be measured in part with patient reported outcomes. Notably, Single Assessment Numerical Evaluation (SANE) scores and American Shoulder and Elbow Surgeon (ASES) scores are used to assess postoperative shoulder function. This study aims to determine predictors of achieving CSO thresholds following elective ARCR by utilizing machine learning (UML). METHODS: A retrospective case-cohort analysis of a prospectively collected database was performed to identify patients who underwent elective ARCR from 2017-2018. Tear dimensions were measured on MRI utilizing a validated technique. CSO achievements on the American Shoulder and Elbow Surgeon (ASES), the Single Assessment Numerical Evaluation (SANE), and the Constant Murley Subjective Score (CMS) at 2-years follow-up were calculated. An unsupervised random forest algorithm was utilized to partition patients into optimal and suboptimal CSO achievement subgroups and subsequently internally validate partitioning based on stability, connectivity, Dunn’s partition coefficient, and silhouette coefficients. This data, along with a total of 30 demographic, clinical, and preoperative PROs were assessed for prognostic value through a stepwise multivariable logistic regression. RESULTS: A total of 346 patients (male: 192, 55.5%; Age: 57.2±9.1, BMI: 30.1±5.4) were eligible for inclusion and followed for an average of 3.8 (range: 2.0 – 6.2 years). The random forest algorithm arrived at an optimal partition of 2 subgroups (Stability: 0.16; connectivity: 180.8; Dunn: 0.16; Silhouette: 0.05), with 176 patients in the optimal achievement subgroup and 157 patients in the suboptimal achievement subgroup. The two subgroups differed significantly (P≤0.004) in the likelihood of achievement of all CSOs. Stepwise multivariable logistic regression identified an increase of 1 mm in tear size in the sagittal dimension to predict a 10% increase in the probability of suboptimal achievement. Additional, additive risk factors for suboptimal CSO achievement included increased preoperative Constant-Murley shoulder score (OR: 1.11, 95% CI: 1.04-1.18, P<0.001), increasing number of tendons involved (OR: 14.07, 95% CI: 4.5-44.02, P<0.001) and subscapularis involvement (OR: 8.67, 95% CI: 2.45-30.71, P=0.01). Protective factors included performance of a subpectoral biceps tenodesis compared to biceps tenotomy (OR: 0.26, 95% CI: 0.05-0.92, P=0.03). CONCLUSIONS: Clinically meaningful subgroups were uncovered using a machine learning clustering approach in patients undergoing ARCR. Tear size, number of tendons involved, and subscapularis involvement were highly significant and additive predictors of suboptimal CSO achievement at 2-year minimum follow-up. Treatment of concurrent biceps pathology with tenodesis confers 74% increased likelihood of CSO achievement vs. tenotomy.