#5059 EXTERNAL VALIDATION OF THE KFRE AND GRAMS PREDICTION MODELS FOR DEATH AND KRT IN A SPANISH POPULATION WITH ADVANCED CHRONIC KIDNEY DISEASE

Abstract

Abstract Background and Aims The Kidney Failure Risk Equation (KFRE) is used to predict the risk of kidney replacement therapy (KRT) initiation in the following 2 and 5 years, in patients with chronic kidney disease (CKD) stages G3-G5. The Grams model is applied exclusively to stage G4-G5 patients who have a higher risk of dying or requiring KRT. This model predicts both events, KRT and death, at 2 and 4 years, but unlike the KFRE, it considers the patient's death as a competing event with respect to the start of RRT. It has been suggested that the Grams model makes a more accurate prediction of the onset of KRT in the long term, since by including mortality as a competing event it does not overestimate the risk of KRT. However, there are few external validations of the Grams model to predict mortality in patients with CKD G4+. We performed an external validation of both models with respect to the start of KRT and mortality in a Spanish population with CKD G4+, followed up prospectively during 4-9 years. Method We conducted a prospective cohort analysis of incident patients followed in the Advanced Chronic Kidney Disease clinic of the Hospital Universitario Fundación Alcorcón, Spain between 1-1-2014 and 31-12-2018 who had CKD G4-G5. Clinical data was registered prospectively in a database at each visit to the clinic. Follow-up ended on 31-12-2022. For each patient, the follow-up time ranged from 4 to 9 years. The outcomes were: observed incidence of KRT (haemodialysis, peritoneal dialysis, or pre-emptive kidney transplantation) by 2, 4 and 5 years and death at 2 and 4 years before starting KRT. The predictors used in the study were calculated at the time of initial clinic referral and included the 2-year and 4-year Grams’ calculator risk scores (Grams-2 and Grams-4, respectively) for KRT and death and the 2-year and 5-year 4-variable KFRE scores for KRT (KFRE-2 and KFRE-5 respectively). We performed calibration and discrimination analyses to evaluate the performance of both predictive models. Discrimination was assessed by calculating the area under the ROC curve (C-statistic with 95% confidence interval). Calibration was assessed by using the Hosmer-Lemeshow test, through calculation of the slope and intercept, and visually by generating a calibration plot, evaluating the level of agreement between predicted probabilities versus observed outcomes. Overall model performance was evaluated using the Brier score as a composite measure of both model discrimination and calibration. Results We studied 347 patients, 31.1% were women, with a mean age of 72.1 ± 12.7 years, 52.4% with cardiovascular disease and 58.8% diabetics. At the start, the mean eGFR was 20.7 ± 5.0 ml/min and the median urine albumin-creatinine ratio was 327 (IQR 52-1118) mg/g. At 2, 4 and 5 years the percentage of patients who required KRT was 20.2%, 38.3% and 45.7% respectively, while 12.7% and 23.6% died at 2 and 4 years respectively, before starting KRT. For KTR both models had an excellent discrimination. For KFRE-2 and Grams-2 the AUC was 0.896 (95% CI 0.858-0.933) and 0.904 (95% CI 0.866-0.941) respectively, for Grams-4 the AUC was 0.842 (95% CI 0.801-0.884), and for KFRE-5 the AUC was 0.805 (95% CI 0.751-0.859). For death before KTR the Grams model demonstrated an acceptable discrimination, with an AUC of 0.726 (95% CI 0.646-0.806) and 0.749 (95% CI 0.690-0.809) for Grams-2 and Grams-4 respectively. There was excellent calibration for KRT for both models, the Hosmer-Lemeshow test presented a p>0.05 in all cases, the Brier score was less than 0.20. The adjustment was very precise and only in the event of estimated risks greater than 70% at 2 and 4 years is the observed risk overestimated. The calibration of the Grams model to estimate mortality before starting KTR was also excellent, although for estimated risks greater than 50% at 4 years the observed risk is underestimated. Conclusion In a Spanish cohort of patients with CKD G4+, the Grams and KFRE models adequately estimate the risk of KRT. The Grams model provides an acceptable estimate of the risk of death before starting KRT at both 2 and 4 years. It can be considered for treatment planning and information for patients with CKD G4+.