Background: There is little overall evidence behind clinical practice guidelines for diagnosis and treatment of rotator cuff repair. The purpose was to compare the accuracy of different machine learning models that use pre-operative data from an international and multicentric database to predict if a patient that underwent rotator cuff repair, could achieve the minimal important change (MIC) for single assessment numeric evaluation (SANE) at one year follow-up.Methods: This study analyzed prospectively collected data of patients undergoing rotator cuff repair. In total, 4683 cases were enrolled and divided into the training (80%) and validation (20%) datasets. Different machine learning models were trained (Extreme Gradient Boosting [XGBoost], Support Vector Machine [SVM], Multi-Layer Perceptron [MLP]), tested and compared using 229 valid pre-operative variables with a specificity set at 95%. The top ten features with the highest impact on the model output were then identified through SHapley Additive exPlanations (SHAP) and used to develop a simplified model. Privileged and unprivileged groups were designated to perform an exploratory fairness analysis.Findings: The performance of the best performing model (XGBShap10) revealed a precision of 0×587 (0×485-0×685), a recall of 0×320 (0×252-0×392), a balanced accuracy of 0×636 (0×593-0×679), a F1 -score of 0×413 (0×338-0×487), and an area under the receiver operating characteristic curve (AUC) of 0×667 (0×618-0×715).Interpretation: The present ML model performance was limited with a balanced accuracy of 64%. However, this model enabled the detection of one third of RCR failures. Using this model in clinical practice would theoretically avoid 4×4% of unnecessary operations while accepting to leave aside 5% of appropriate candidates for surgery.Funding: This project has received funding from the Fondation de Bienfaisance Pierre & Andrée Haas (Geneva, Switzerland), the insurance Foundation Groupe Mutuel (Martigny, Switzerland), FORE (Foundation for Research and Teaching in Orthopedics, Sports Medicine, Trauma, and Imaging in the Musculoskeletal System), Grant #2023-13, and TheArk Foundation (Valais, Switzerland).Declaration of Interest: Alexandre Lädermann is a paid consultant for Arthrex, Stryker, Medacta, and Enovis. He received royalties from Stryker and Medacta. He is the (co)founder of FORE, Med4Cast, and BeeMed. Philippe Collin receives royalties from and is a consultant and paid speaker for Stryker and Enovis. He is the co-founder of Med4Cast and Follow. He is on the board of SECEC and IBSES. Patrick Denard receives royalties from and is a consultant and paid speaker for Arthrex. He is the co-founder of Med4Cast. All others have nothing to declare. Ethical Approval: The registry used in this study as well as the research protocol were approved by ethical review boards (SOS #1) and ethics committee (EKNZ #2019-02076, CCER #2020-02670, Pred cohort study17) and registered at ClinicalTrials.gov (identifier NCT06145815) and at Registry of all Projects in Switzerland (RAPS identifier CCER 2020-02670). Every patient enrolled in this open-label clinical database provided a written informed consent for the use of their data.