Kidney failure risk equation to predict kidney transplant failure in individuals with an eGFR2: a UK Renal Registry...

To externally validate and subsequently repurpose/recalibrate the easily accessible kidney failure risk equation (KFRE) for a prevalent transplant population with an estimated glomerular filtration rate (eGFR)<30 mL/min/1.73 m². It also aims to evaluate the recalibrated KFRE's potential to improve clinical practice with a clinical utility assessment.

Retrospective cohort study using UK Renal Registry data.

68 adult UK kidney centres.

4092 patients with grafts at least 2 years old and eGFR<30 mL/min/1.73 m² from 2009 to 2018.

Death-censored allograft failure at 2 years, defined as dialysis initiation, re-transplantation or, in the absence of the former two, the recorded date of transplant failure.

The KFRE was calculated at baseline using the 2-year, 8-variable non-North American KFRE, and performance was assessed using Harrell's C-statistic and calibration plots. The model was recalibrated using Cox Regression (2009-2013 cohort) and temporally validated using the 2014-2018 cohort. Clinical utility was assessed using decision-curve analysis, estimating per-100-patient gains in timely planning and reductions in unnecessary interventions compared with eGFR triggers.

The original KFRE had excellent discrimination but was miscalibrated, underpredicting graft failure. Temporal validation demonstrated that the performance of the recalibrated KFRE could be maintained across time periods (Harrell's C-index of 0.81 (95% CI 0.80 to 0.83); O/E (Observed/Expected events) ratio 1.00 (95% CI 0.93 to 1.07). It identified 9/100 more patients for timely intervention and 13/100 more for whom intervention could be delayed compared with a late clinical trigger of an eGFR<15 mL/min/1.73 m².

While there are other prognostic models, this is the first study to focus on the understudied and clinically important cohort of patients with an eGFR<30 mL/min/1.73 m², offering more relevant 2-year predictions. For this underserved population, we present an easily accessible model that is robust to temporal and case-mix differences, demonstrates clinical utility and has the potential to improve daily clinical practice.