Validation of the kidney failure risk equation in European CKD patients.

BACKGROUND Patients with chronic kidney disease (CKD) are at risk for progression to kidney failure. Using data of Canadian CKD patients, Tangri et al. recently developed models to predict the progression of CKD stages 3-5 to kidney failure within 5 years. We validated this kidney failure risk equation (KFRE) in European CKD patients. METHODS We selected non-transplanted patients with CKD stages 3-5 who participated in the MASTERPLAN study, a randomized controlled trial in patients with CKD. Kidney failure was defined as the initiation of chronic dialysis or kidney transplantation within 5 years. Patients who died before kidney failure were censored. Patients followed for <5 years, who did not develop kidney failure and did not die, were excluded. The 5-year kidney failure risk was predicted using three different models developed by Tangri et al. and compared with the actual kidney failure rate in MASTERPLAN. Model performance was evaluated using the area under the receiver operating characteristic curve (ROC-AUC), the net reclassification index (NRI) and by comparing the observed and predicted rates of kidney failure. RESULTS A total of 595 patients were included; 114 developed kidney failure. (Overall observed kidney failure risk in our cohort was 5% lower than in the Canadian validation cohort.) Discrimination of the eight-variable model [including age, sex, estimated glomerular filtration rate (eGFR), albuminuria, calcium, phosphate, bicarbonate, albumin] was similar to that of the four-variable model (including age, sex, eGFR, albuminuria) and the three-variable model (including age, sex, eGFR); ROC-AUCs were 0.89 [95% confidence interval (CI) 0.86-0.92], 0.88 (95% CI 0.85-0.91) and 0.88 (95% CI 0.85-0.92), respectively. Using the NRI, the eight-variable model slightly outperformed the four-variable model (NRI 6.5%) and the three-variable model (NRI 12.4%). The mean differences between the observed and predicted kidney failure risk were -4.0, -7.1 and -7.4% for the eight-, four-, and three-variable model, respectively. CONCLUSIONS The KFRE accurately predicted the progression to kidney failure in European CKD patients. Discrimination of the three models was similar. Calibration of the eight-variable model was slightly better than that of the simpler models. We question whether this outweighs its added complexity.