Incorporating Pathomics Into IgAN Risk Prediction

IgA nephropathy (IgAN) is the most common glomerulonephritis worldwide, with heterogeneous progression rates. The International IgAN Prediction Tool (IIgAN-PT) enables individualized risk prediction using clinical characteristics and histopathological scoring. However, traditional histopathological scoring is semiquantitative and prone to interobserver variability. Deep learning-based, automated quantification of digitized histopathology enables the extraction of quantitative morphological biomarkers (pathomics). Here, we evaluated the added prognostic value of pathomics over traditional histopathology and its potential integration into the IIgAN-PT.

We analyzed a multicenter European cohort of 555 adults with biopsy-proven IgAN (VALIGA), including centralized histopathology assessment. We assessed the potential of 25 clinically interpretable candidate pathomics predictors across 2 clinically relevant settings by (i) comparing pathomics with traditional histopathology (MEST-C [mesangial hypercellularity (M), endocapillary hypercellularity (E), segmental glomerulosclerosis (S), tubular atrophy and/or interstitial fibrosis (T) and crescents (C)]) and (ii) extending the IIgAN-PT with pathomics. Model performance was evaluated by overall fit, discrimination, reclassification, and calibration.

In a direct comparison, pathomics slightly outperformed the MEST-C score when used alone (Δ C-statistic 0.04 (95% confidence interval [CI]: 0.03-0.06)) but was marginally worse after adjusting for clinical predictors (Δ C-statistic -0.034 [95% CI: -0.042 to -0.027]). Patient stratification into clinical risk groups was consistent between models. Combining pathomics and traditional histopathology led to minor improvements over the current IIgAN-PT (without ethnicity, Δ C-statistic 0.024 [95% CI: 0.019-0.028]). Pathomics variables captured chronic morphological changes including glomerular deformation, tubular shrinkage, and arterial luminal narrowing.

Our findings indicate that pathomics-derived digital tissue biomarkers provide an additional layer of information and are noninferior to MEST-C, supporting more standardized, robust, and reproducible risk stratification in IgAN.