The combined utilization of predictors: a better approach to diagnose and predict rotator cuff tears than using a single predictor

Abstract

ObjectivesTo explore whether the combined utilization of two predictors is more efficient to predict and diagnose rotator cuff tears (RCTs) than using a single predictor, and to find out which combination is the best screening approach for RCTs.MethodsThis was a retrospective study and patients who visited our hospital and were diagnosed with or without RCTs via magnetic resonance imaging from January 2018 to April 2022 were enrolled and classified into two groups respectively. Four predictors, the critical shoulder angle (CSA), the acromion index (AI), the greater tuberosity angle (GTA) and the double-circle radius ratio (DRR) were picked to participate in the present study. Quantitative variables were compared by independent samples t tests and qualitative variables were compared by chi-square tests to find significant differences between groups. Binary logistic regression analysis was used to distinguish independent risk factors and calculate the increased odds of having a RCT, and construct discriminating combined models to further diagnose and predict RCTs. Receiver operating characteristic (ROC) curves were pictured and areas under the curve (AUC) were calculated to determine the overall diagnostic performance of the involved predictors and the combined models. For all tests a p value of < 0.05 was considered statistically significant.ResultsOne hundred and thirty-nine shoulders with RCTs and 57 shoulders without RCTs were included. The mean values of CSA (35.36 ± 4.57 versus 31.41 ± 4.09°, P = 0.000), AI (0.69 ± 0.08 versus 0.63 ± 0.08, P = 0.000), DRR (1.43 ± 0.10 versus 1.31 ± 0.08, P = 0.000) and GTA (70.15 ± 7.38 versus 64.75 ± 7.91°, P = 0.000) were significantly larger in the RCT group than for controls. Via ROC curves, we found the combined model always showed a better diagnostic performance than either of its contributors, improving the AUC from 0.668–0.823 to 0.751–0.883, and extending the upper boundary of diagnostic sensitivity from 71.9–81.3%, and the diagnostic specificity from 82.5–86.0%. Via logistic regression analysis, we found the values of both predictors over their cutoff values resulted in a dramatical and enormous increasement (20.169–161.214 folds) in the risk of having a RCT, which is far beyond that by using a single predictor only (2.815–11.191 folds).ConclusionThe combined utilization of predictors is a better approach to diagnose and predict RCTs than using a single predictor. With a comprehensive consideration on the diagnostic and predictive performance of the combined models, we conclude the CSA together with the DRR present the strongest detectability for a presence of RCTs.