Kidney Failure Prediction Models: A Comprehensive External Validation Study in Patients with Advanced CKD.

Area of Science:

• Nephrology
• Epidemiology
• Biostatistics

Background:

• Existing kidney failure prediction models lack head-to-head comparisons and validation in advanced CKD.
• Most models do not account for competing risks, such as death.

Purpose of the Study:

• To externally validate 11 kidney failure prediction models in patients with advanced CKD.
• To assess model performance considering the competing risk of death.

Main Methods:

• Included patients from the European Quality Study (EQUAL) and Swedish Renal Registry (SRR).
• Assessed model performance using discrimination and calibration metrics.
• Kidney failure defined as end-stage kidney disease requiring renal replacement therapy (RRT-treated ESKD).

Main Results:

• 1580 patients from EQUAL and 13,489 from SRR were included.
• Average c-statistic was 0.74 (EQUAL) and 0.80 (SRR), lower than previous validations.
• Shorter-term models (2-year KFRE, 4-year Grams) showed good calibration and discrimination.
• Longer-term models (5-year KFRE) overestimated risk.

Conclusions:

• Accurate kidney failure prediction is possible in advanced CKD.
• The Kidney Failure Risk Equation (KFRE) is effective for short-term (2-year) predictions.
• The Grams model is suitable for longer-term (4-year) predictions, accounting for competing risks.