Abstract: PO1582
External Validation of Two New IgA Risk-Prediction Tools in a Norwegian Cohort
Session Information
- Glomerular Diseases: Clinicopathological Features and Outcomes in IgAN, Lupus Nephritis, and Vasculitis
November 04, 2021 | Location: On-Demand, Virtual Only
Abstract Time: 10:00 AM - 12:00 PM
Category: Glomerular Diseases
- 1203 Glomerular Diseases: Clinical, Outcomes, and Trials
Authors
- Haaskjold, Yngvar Lunde, Haukeland Universitetssjukehus, Bergen, Norway
- Lura, Njål, Haukeland Universitetssjukehus, Bergen, Norway
- Bjoerneklett, Rune, Haukeland Universitetssjukehus, Bergen, Norway
- Bostad, Lars Sigurd, Haukeland Universitetssjukehus, Bergen, Norway
- Knoop, Thomas, Haukeland Universitetssjukehus, Bergen, Norway
Background
Recently two prediction tools for IgA nephropathy (IgAN) have been developed combining clinical and histopathological parameters. Barbour and colleagues developed the International IgAN Prediction Tool, to predict the risk for 50 % decline in estimated glomerular filtration rate (eGFR) or end stage renal disease (ESRD) up to 80 months after diagnosis. Schena and colleagues developed the IgA Nephropathy Clinical Decision Support System (CDSS), using artificial neural networks (ANN) to estimate the risk for ESRD. In the present study we aim to externally validate both prediction tools using a Norwegian cohort with long-term follow-up.
Methods
We included 306 patients with biopsy-proven primary IgAN. Histopathologic samples were retrieved from the Norwegian Kidney Biopsy Registry and reclassified according to the Oxford classification. The probability of the primary outcome, ESRD, was calculated by plotting the data from our cohort into the respectively online tools. The predicted outcome probabilities from the models were handled as a prognostic index in the validation analysis. We used principles for external validation of prognostic models: discrimination (concordance statistics), calibration (cox models and survival estimates), and model fit (akaike information criterion).
Results
Mean patient follow-up was 16.5 years, and a total of 61 (20%) patients reached ESRD.
The cumulative dynamic time dependant receiving operating characteristics (tdROC) analysis showed no difference between the models with an area under curve (AUC) of 0.88 and 0.85 at 15 years for Barbour and Schena respectively. Integrated AUC (iAUC) at 20 years was 0.79 for Schena and 0.75 Barbour (p =0.2). Incident dynamic ROC-analysis at 5 and 20 years showed no significant difference in Scena’s tool (AUC 0.80 and 0.74 respectively, p=0.08), but there was but there was a significant decrease in Barbour’s tool (AUC 0.80 and 0.65 respectively, p<0.001). Concordance index was 0.85 for Barbour, and 0.82 for Schena (p=0.03).
Conclusion
Both prediction tools perform well and could become helpful tools for clinicians to identify patients at risk. Barbour’s tool seem to loose prognostic discriminative value at a faster rate than Schena’s over time.