Machine Learning Models for Predicting Rotator Cuff Tear Reparability

The study developed machine learning models to predict rotator cuff tears' reparability, addressing existing scoring systems' limitations. A cohort of patients with tears exceeding 1 cm underwent arthroscopic repair, with data split into training (70%) and validation (30%) sets and an external test set from surgeries by two different surgeons. An elastic-net logistic regression model performed best, achieving an AUROC of 0.847, surpassing an adjusted scoring system (AUROC of 0.786) and two previous systems (AUROCs of 0.757 and 0.687). The model was made available as an online tool. While promising, the study emphasized that clinicians should not rely solely on predictive models.