Development and external validation of a machine learning risk model for pediatric IgA vasculitis nephritis in the post-pandemic era

To investigate shifts in clinical characteristics and risk factors for IgA vasculitis nephritis (IgAVN) in children before and after the COVID-19 pandemic, and to develop a risk prediction model adapted to the post-pandemic era. Clinical data from children with IgAV were retrospectively collected from two cohorts: pre-pandemic period (January 2017 to January 2019, n = 123) and post-pandemic period (January 2023 to January 2025, n = 216). Univariate analysis and LASSO regression identified candidate predictors. A machine learning model was transferred to evaluate temporal generalization, and a new Random Forest ( RF ) model was constructed and externally validated. Post-pandemic children exhibited higher BMI, lower albumin, and elevated complement levels. The risk factor spectrum for IgAVN shifted; the pre-pandemic model showed decreased performance (∆AUC = -0.136, P = 0.014). The post-pandemic RF model demonstrated favorable predictive performance and potential clinical net benefit in external validation. Rash duration, cystatin C, erythrocyte sedimentation rate (ESR), and BMI were the principal contributors to model predictions. Clinical characteristics and predictor distributions differed between the two calendar-based cohorts. The Random Forest model developed in the later cohort demonstrated promising performance in regional external validation and may support future risk-stratification research. Further prospective multicenter validation is required before routine clinical implementation.