Machine learning for predicting acute exacerbation and mortality in idiopathic inflammatory myopathy-associated interstitial lung disease

Objectives Idiopathic inflammatory myopathy-associated interstitial lung disease (IIM-ILD) is a heterogeneous condition with high mortality. This study aimed to develop a nomogram integrating high-resolution computed tomography (HRCT) radiomics with clinical indicators to predict poor prognosis (AE-ILD or mortality) in IIM-ILD. Methods 167 patients were randomized into training ( n = 118) and validation ( n = 49) cohorts. In the training cohort, mRMR and LASSO algorithms identified optimal radiomic features. A combined nomogram was constructed using multivariate logistic regression. Performance was evaluated via ROC analysis, calibration curves, and decision curve analysis (DCA). Results Independent predictors included rash ( p = 0.007), absolute neutrophil count ( p = 0.01), dyspnea ( p = 0.03), and Alveolar-arterial oxygen gradient ( p = 0.01). Twenty radiomic features were selected to calculate the Rad-score. The combined nomogram outperformed the clinical model in both cohorts. The training set achieved an AUC of 0.91 (sensitivity 0.90, specificity 0.84, and accuracy 0.86). The validation set showed an AUC of 0.88 (sensitivity 0.69, specificity 0.90, accuracy 0.82). Calibration curves ( p > 0.05) and DCA confirmed the model’s reliability and clinical utility. Conclusion The combined clinical-radiomics model effectively predicts poor prognosis in IIM-ILD, facilitating early identification and personalized intervention.