Authors of a comprehensive review examined how prognostic modeling in IgA nephropathy (IgAN) has evolved from traditional clinical and pathology-based tools to machine learning and deep learning approaches. Established models use factors such as eGFR, proteinuria, blood pressure, and biopsy findings to estimate progression risk, while newer approaches can incorporate longitudinal clinical data, digital pathology, biomarkers, and multi-omics information. These more advanced models may help identify patients at higher risk of progression and eventually support decisions about monitoring, biopsy timing, treatment intensity, and therapeutic selection.
However, greater algorithmic complexity does not necessarily translate into better clinical performance. Some machine learning models have failed to significantly outperform established prediction tools, and concerns remain about overfitting, interpretability, generalizability, and resource requirements. The authors emphasize that future models should be evaluated not only for predictive accuracy but also for usability and transparency, with multicenter and multiethnic validation needed before widespread clinical adoption. Integrating dynamic, interpretable models into clinical workflows could ultimately move IgAN care from one-time risk estimates toward continuously updated, individualized risk assessment.
Reference: Xu H, Ge S. Risk prediction in IgA nephropathy: from conventional models to machine learning, deep learning, and precision nephrology. Ren Fail. 2026 Dec;48(1):2613606. doi: 10.1080/0886022X.2026.2613606.
