Validation of the International IgA Nephropathy Prediction Tool in the Greek Registry of IgA Nephropathy

Background: Immunoglobulin A nephropathy (IgAN) is among the commonest glomerulonephritides in Greece and an important cause of end-stage kidney disease (ESKD) with an insidious chronic course. Thus, the recently published International IgAN prediction tool could potentially provide valuable risk stratification and guide the appropriate treatment module. This study aimed to externally validate this prediction tool using a patient cohort from the IgAN registry of the Greek Society of Nephrology.

Methods: We validated the predictive performance of the two full models (with or without race) derived from the International IgAN Prediction Tool study in the Greek Society of Nephrology registry of patients with IgAN using external validation of survival prediction models (Royston and Altman). The discrimination and calibration of the models were tested using the C-statistics and stratified analysis, coefficient of determination ( ${\text{R}}_{\text{D}}^2$ ) for model fit, and the regression coefficient of the linear predictor (β_PI), respectively.

Results: The study included 264 patients with a median age of 39 (30-51) years where 65.2% are men. All patients were of Caucasian origin. The 5-year risk of the primary outcome (50% reduction in estimated glomerular filtration rate or ESKD) was 8%. The ${\text{R}}_{\text{D}}^2$ for the full models with and without race when applied to our cohort was 39 and 35%, respectively, and both were higher than the reported ${\text{R}}_{\text{D}}^2$ for the models applied to the original validation cohorts (26.3, 25.3, and 35.3%, respectively). Harrel's C statistic for the full model with race was 0.71, and for the model without race was 0.70. Renal survival curves in the subgroups (<16th, ~16 to <50th, ~50 to <84th, and >84th percentiles of linear predictor) showed adequate separation. However, the calibration proved not to be acceptable for both the models, and the risk probability was overestimated by the model.

Conclusions: The two full models with or without race were shown to accurately distinguish the highest and higher risk patients from patients with low and intermediate risk for disease progression in the Greek registry of IgAN.