Objective To investigate the clinical and imaging factors affecting the re-tear after arthroscopic repair in massive rotator cuff tears, and to develop a nomogram prediction model. Methods A retrospective analysis was conducted on 190 patients with massive rotator cuff tears who met the selection criteria and underwent arthroscopic repair between January 2022 and December 2024. The cohort comprised 79 males and 111 females, with an age of (60.0±9.6) years. Baseline data and radiographic parameters of rotator cuff were collected for each patient. All participants were randomly divided into a training set (n=133) and a test set (n=57) at a ratio of 7∶3. Univariate analysis was initially employed to identify potential influencing factors associated with re-tear after repair. Subsequently, an integrated feature selection strategy combining LASSO regression, the Boruta algorithm, and recursive feature elimination (RFE) was applied to determine risk factors for re-tear and to construct a nomogram prediction model. The predictive performance of the model was evaluated using the receiver operating characteristic (ROC) curve, with the area under the curve (AUC) calculated to assess discriminative ability. Decision curve analysis (DCA) was also performed to evaluate the clinical utility and net benefit of the model. Results Among the patients included in the study, 81 (42.63%) had re-tears after repair. Univariate analysis showed that the age, body mass index (BMI), smoking history, Goutallier grade, modified Patte grade, acromial index (AI), acromiohumeral distance (AHD), lateral acromial angle (LAA), and critical shoulder angle (CSA) was the influencing factor of re-tear (P<0.05). There was no significant difference in the above variables and the incidence of re-tear between the training set and the test set (P>0.05). Based on the combined screening strategy, 7 consistent variables were obtained, including BMI, AHD, CSA, LAA, Goutallier classification, modified Patte classification, and smoking history, to construct the prediction model. The model achieved an AUC of 0.959, accuracy of 0.955, sensitivity of 0.942, and specificity of 0.963 in the training set. In the test set, the corresponding values were AUC 0.967, accuracy 0.842, sensitivity 0.897, and specificity 0.786. The optimal prediction threshold was 0.298. The DCA based on test set showed that the net benefit of clinical decision-making based on the model was higher than that of the “all intervention” strategy within the threshold probability range of 0.02-0.40, and was better than that of the “no intervention” strategy under each threshold, suggesting that the model had good clinical practicability and decision support value. ConclusionBased on the LASSO-Boruta-RFE combination screening strategy, seven key factors were identified from multiple clinical and imaging variables, including Goutallier grade, modified Patte grade, smoking history, AHD, CSA, LAA, and BMI. The constructed nomogram model shows ideal discrimination efficiency and calibration in the training set and the test set, and there is no sign of overfitting, which has good clinical applicability and decision support value.
