[Establishing a predictive model for the activity of idiopathic inflammatory myopathy based on MRI and clinical features]
Z J Wang,Z R Tian,Y Q Wang,B Tian,R Gong,S S Chi
DOI: https://doi.org/10.3760/cma.j.cn112137-20240805-01790
2024-09-24
Abstract:Objective: To analyze MRI and clinical characteristics of idiopathic inflammatory myopathy (IIM) activity and construct a prediction model. Methods: A retrospective analysis was conducted on 326 patients with IIM from December 2019 to December 2023 at General Hospital of Ningxia Medical University, including 112 males and 214 females, aged(53.7±15.3) years. According to histopathology and electromyography, they were divided into active phase group(n=86) and inactive phase group (n=240). The two groups were randomly divided into the training set and the verification set according to the ratio of 7∶3. The single factor analysis, least absolute shrinkage and selection operator (Lasso), random forest algorithm, and multivariate logistic regression model were used to screen the risk factors of IIM activity and construct a prediction model. Receiver operating characteristic (ROC) curve and calibration curve were used to evaluate the performance of prediction model. Results: There were significant differences in gender, age, T1 value, T2 value, creatine kinase-MB(CKMB), creatine kinase (CK) and lactate dehydrogenase (LDH) between the two groups(all P<0.05). Lasso and random forest algorithm screened 5 variables for analysis, age (λ=-0.009), T2 value (λ=-2.564), CKMB (λ=-0.256), CK (λ=-0.492), LDH (λ=-2.786) respectively. Multivariate logistic regression model showed that age (OR=1.603, 95%CI: 1.030-1.096), T2(OR=352.269, 95%CI: 13.303-9 328.053), CKMB (OR=2.470, 95%CI: 1.497-4.075), CK(OR=4.973, 95%CI: 2.583-9.575), LDH(OR=1 155.247, 95%CI: 152.387-8 757.954) were risk factors for active IIM patients. A prediction model nomograms were drawn with the above risk factors included. The area under the ROC curve (AUC) of the prediction model for the training set MRI combined with clinical indicators was higher than that of the clinical indicator model [0.914 (95%CI: 0.873-0.955) vs 0.901 (95%CI: 0.858-0.945), P<0.001], with sensitivity of 88.3% and 90.7%, and specificity of 81.7% and 75.0%, respectively. The AUC of the prediction model for the validation set MRI combined with clinical indicators was higher than that of the clinical model [0.982 (95%CI: 0.873-0.955) vs 0.934 (95%CI: 0.858-0.945), P<0.001], with sensitivity of 97.2% and 88.5%, and specificity of 100.0% and 92.3%, respectively. The calibration curves plotted in the training set and test set, respectively, fit well with the ideal curve. Conclusion: The nomogram model of MRI combined with clinical indicators can effectively predict the activity of IIM.