Abstract
OBJECTIVE: To establish a prediction model using non-invasive clinical features for early discrimination of DM-ILD in clinical practice. METHOD: Clinical data of pediatric patients with JDM were retrospectively analyzed using machine learning techniques. The early discrimination model for JDM-ILD was established within a patient cohort diagnosed with JDM at a children's hospital between June 2015 and October 2022. RESULTS: A total of 93 children were included in the study, with the cohort divided into a discovery cohort (n = 58) and a validation cohort (n = 35). Univariate and multivariate analyses identified factors associated with JDM-ILD, including higher ESR (OR, 3.58; 95% CI 1.21-11.19, P = 0.023), higher IL-10 levels (OR, 1.19; 95% CI, 1.02-1.41, P = 0.038), positivity for MDA-5 antibodies (OR, 5.47; 95% CI, 1.11-33.43, P = 0.045). A nomogram was developed for risk prediction, demonstrating favorable discrimination in both the discovery cohort (AUC, 0.736; 95% CI, 0.582-0.868) and the validation cohort (AUC, 0.792; 95% CI, 0.585-0.930). Higher nomogram scores were significantly associated with an elevated risk of disease progression in both the discovery cohort (P = 0.045) and the validation cohort (P = 0.017). CONCLUSION: The nomogram based on the ESIM predictive model provides valuable guidance for the clinical evaluation and long-term prognosis prediction of JDM-ILD.