EXTERNAL VALIDATION, RECALIBRATION, AND CLINICAL UTILITY OF THE PROGNOSTIC MODEL KIDNEY FAILURE RISK EQUATION IN PATIENTS WITH CKD STAGES G3-4: A NATIONWIDE RETROSPECTIVE COHORT ANALYSIS IN PERU

Introduction:

Evidence on the external validity of the Kidney Failure Risk Equation (KFRE), a model for predicting kidney failure, is limited in Latin American countries with resource constraints. A previous study in Peru found KFRE miscalibrated, but no studies in Latin America have recalibrated the model or assessed its clinical utility. This study aimed to evaluate the external validation, recalibration, and clinical utility of KFRE in a national cohort of chronic kidney disease (CKD) patients in Peru, using data from a renal health program in primary care hospitals across all regions.

Methods:

A retrospective cohort study was conducted using data from the Renal Health Surveillance Program (VISARE) of EsSalud, covering all 25 regions of Peru. A total of 30,031 CKD stage G3-4 patients from 45 healthcare networks, diagnosed between January 1, 2013, and December 31, 2022, were included. The primary outcome was kidney failure, defined as dialysis initiation or nephrologist-confirmed end-stage renal disease. Calibration-in-the-large was assessed via the observed-to-expected (O/E) ratio, weak calibration using intercept and slope, and moderate calibration using calibration curves. Discrimination (C-index) of the original 4-variable KFRE was evaluated at 2 and 5 years. Recalibrated models were proposed, and net clinical benefit was assessed using decision curve analysis (DCA). The KFRE model at 5 years was compared against NICE 2014 and the local Peruvian guideline, which stratifies risk based on eGFR and albuminuria following KDIGO’s ACR/eGFR risk categories. NICE suggests nephrology referral for patients with a KFRE-predicted risk of ≥5% at 5 years, while KDIGO recommends referral for those with a KFRE risk of 3-5%. At 2 years, thresholds of 20-40% were evaluated, in line with KDIGO’s recommendation of 40% for RRT preparation, though lower thresholds are also used.

Results:

The original KFRE showed good discrimination with C-indices of 0.9 (95% CI 0.89-0.91) at 2 years and 0.89 (95% CI 0.88-0.9) at 5 years but demonstrated poor calibration in slope, underestimating risk in low-risk and overestimating in high-risk individuals (slope 0.62 at both time points) (Table 1).

The O/E ratio showed average overestimation at 2 years (1.42, 95% CI 1.37-1.46) and underestimation at 5 years (0.89, 95% CI 0.87-0.91). Calibration plots (Figure 1) revealed a complex pattern, with overpredictions at higher risks and underpredictions at lower risks, consistent with weak calibration indicated by the slope lower than 1. 

Recalibrated models improved calibration-in-the-large but retained weak and moderate calibration issues (Table 2)

DCA showed KFRE at 5 years provided higher net benefit for nephrology referrals across relevant thresholds compared to Peruvian guidelines and NICE 2014 (Figure 2). At 2 years, net benefit was observed within a threshold range of 20-30%, but diminished beyond 30%, indicating potential over-referral.

Conclusions:

KFRE, despite miscalibration, remains valuable for nephrology referrals in Peru, with recalibrated models offering improvements. The model also demonstrated strong discriminative ability, making it useful for CKD management. This is the first study in Latin America to evaluate KFRE’s utility, suggesting its applicability in resource-limited settings.

I have no potential conflict of interest to disclose.

I did not use generative AI and AI-assisted technologies in the writing process.