Kidney Failure Risk Equation (KFRE)

Clinical background

The Kidney Failure Risk Equation (KFRE) meets a concrete need: to separate patients with chronic kidney disease who will actually require renal replacement therapy from the large majority who will not. eGFR and albuminuria alone give a crude risk stratification, but not enough to determine when referral to a nephrologist is warranted or when transplant assessment should begin. KFRE quantifies the probability of kidney failure requiring treatment within 2 and 5 years and thereby provides a basis for decision that is more precise than fixed eGFR thresholds.

The tool is used primarily for two decisions: the timing of referral to a nephrologist, and the timing of planning for vascular access and transplant assessment. KDIGO 2024 recommends KFRE as the basis for referral decisions and proposes a 5-year risk of 3 to 5 per cent as the threshold [1].

Calculating the KFRE

The calculator uses the four-variable version of the Kidney Failure Risk Equation. The variables are age, sex (male = 1, female = 0), eGFR and the urinary albumin-to-creatinine ratio (ACR). The formula first calculates a log risk score $L$ and then converts it into an absolute risk:

$$L = -0{.}2201 \times \left(\frac{\text{age}}{10} - 7{.}036\right) + 0{.}2467 \times (\text{male} - 0{.}5642) - 0{.}5567 \times \left(\frac{\text{eGFR}}{5} - 7{.}222\right) + 0{.}4510 \times (\ln(\text{ACR}) - 5{.}137)$$

$$\text{2-year risk} = 1 - 0{.}9832^{\exp(L)}$$

$$\text{5-year risk} = 1 - 0{.}9365^{\exp(L)}$$

ACR is given in mg/g. Values in mg/mmol are multiplied by 8.84 for conversion. eGFR can be calculated with any standard creatinine-based equation.

The derivation cohort consisted of 3,449 patients with chronic kidney disease stage 3 to 5 (eGFR 10 to 59 mL/min/1.73 m²) referred to nephrology in Ontario, Canada, between April 2001 and December 2008. The outcome was kidney failure defined as the need for dialysis or pre-emptive kidney transplantation. During follow-up, 386 patients (11 per cent) developed kidney failure. An independent validation cohort from British Columbia comprised 4,942 patients, of whom 1,177 (24 per cent) reached the outcome [2]. In the derivation cohort the eight-variable model achieved a c-statistic of 0.917 (95 per cent CI 0.901 to 0.933), while the four-variable model performed marginally less well. In the validation cohort the c-statistic for the eight-variable model was 0.841 (95 per cent CI 0.825 to 0.857).

Interpretation in practice

KFRE gives a continuous probability, but clinical guidelines and implementation studies have established thresholds for action. The following table summarises how the results can be translated into clinical action.

In an implementation study in Manitoba, Canada, a 5-year risk of 3 per cent was used as the threshold for triaging nephrology referrals. At this threshold the model was 97 per cent sensitive and 62 per cent specific for predicting kidney failure, with a negative predictive value of 99 per cent. The triage model reduced the median waiting time from 230 to 58 days, and 34 per cent of referrals were returned to primary care [3].

Validation and performance

The most comprehensive external validation is a meta-analysis of 31 cohorts in the CKD Prognosis Consortium, comprising 721,357 patients with chronic kidney disease stage 3 to 5 and 23,829 kidney failure events across four continents [4]. The original four-variable KFRE showed a pooled c-statistic of 0.90 (95 per cent CI 0.89 to 0.92) at 2 years and 0.88 (95 per cent CI 0.86 to 0.90) at 5 years. Discrimination was similar in subgroups defined by age, ethnicity and diabetes status.

Calibration was adequate in North American cohorts, but in several cohorts outside North America the original equation overestimated risk. A regional calibration factor that lowered the baseline risk by 32.9 per cent at 2 years and 16.5 per cent at 5 years improved calibration in 12 of 15 and 10 of 13 non-North American cohorts respectively (p = 0.04 and p = 0.02) [4]. This non-North American calibration factor is incorporated in several clinical implementations and in the web-based calculator at qxmd.com.

In a German validation of 403 patients with chronic kidney disease stage G2 to G4 referred to a nephrology tertiary care clinic, a c-statistic of 0.91 (95 per cent CI 0.83 to 0.99) was achieved [5]. Adding Doppler ultrasound parameters did not improve prediction.

A Swedish observational study from Stockholm (the SCREAM cohort) comprised 192,964 individuals with an eGFR below 60 mL/min/1.73 m² and 887,388 observations during follow-up [6]. The non-North American calibrated four-variable KFRE showed good discrimination and outperformed both traditional Swedish referral criteria and the KDIGO 2012 criteria in terms of sensitivity, specificity and net reclassification index. The study did, however, show that thresholds higher than the 3 to 5 per cent recommended by KDIGO gave the best combined sensitivity and specificity: 15 per cent for the non-North American calibrated equation and 9 per cent for a locally recalibrated version. Implementing a risk-based referral model would reduce the number of unnecessary referrals by 23 to 25 per cent without missing many cases genuinely in need of nephrology care [6].

Limitations

KFRE was derived and validated in patients referred to a nephrologist with chronic kidney disease stage 3 to 5 (eGFR 10 to 59 mL/min/1.73 m²). Accuracy outside this eGFR range is less well established. The calculator accepts an eGFR down to 1 mL/min/1.73 m², but at very low values the prediction becomes increasingly uncertain and clinically less meaningful, since these patients are already obvious candidates for renal replacement therapy.

The model does not apply to kidney transplant recipients, for whom a separate validation has been published but with differing performance. Nor does it apply to patients with acute kidney injury or rapidly progressive glomerulonephritis, in whom decisions on referral and treatment are governed by factors other than a long-term risk prediction.

An important limitation is that death was handled as censoring rather than as a competing risk in the original model, which leads to systematic overestimation of the risk of kidney failure, particularly in older patients with high mortality. The Swedish study from Stockholm showed that including death as a competing risk gave more realistic prognostic estimates [6].

The eight-variable model, which also includes serum calcium, phosphate, bicarbonate and albumin, gives marginally better discrimination but requires more inputs and is not implemented in this calculator. In the original validation cohort the improvement in the integrated discrimination index was 3.2 per cent (95 per cent CI 2.4 to 4.2 per cent) compared with the four-variable model [2].

External validation and choice of eGFR equation

In Swedish health care, eGFR is calculated automatically with the revised Lund–Malmö equation, not with CKD-EPI as used in the derivation cohort. This may affect calibration, since different eGFR equations give systematically different estimates, particularly at older ages. The Swedish study from Stockholm used Lund–Malmö and found good performance for the non-North American calibrated KFRE, but noted that a local recalibration improved the fit further [6].

Traditional Swedish referral criteria are based on fixed thresholds for age, eGFR and albuminuria. The Swedish study showed that a risk-based model using KFRE outperformed these criteria and could reduce the number of unnecessary nephrology referrals by approximately a quarter, while few cases genuinely in need of specialist care were missed [6]. KDIGO 2024 recommends a 5-year risk of 3 to 5 per cent as the referral threshold [1], but the Swedish study found that higher thresholds (9 to 15 per cent depending on calibration) gave the best combined sensitivity and specificity in a population with a low incidence of kidney failure.

References

1. Levin A, Ahmed SB, Carrero JJ, et al. Executive summary of the KDIGO 2024 Clinical Practice Guideline for the Evaluation and Management of Chronic Kidney Disease. Kidney Int 2024. PMID: 38519239
2. Tangri N, Stevens LA, Griffith J, et al. A predictive model for progression of chronic kidney disease to kidney failure. JAMA 2011. PMID: 21482743
3. Hingwala J, Wojciechowski P, Hiebert B, et al. Risk-based triage for nephrology referrals using the Kidney Failure Risk Equation. Can J Kidney Health Dis 2017. PMID: 28835850
4. Tangri N, Grams ME, Levey AS, et al. Multinational assessment of accuracy of equations for predicting risk of kidney failure: a meta-analysis. JAMA 2016. PMID: 26757465
5. Lennartz CS, Pickering JW, Seiler-Mußler S, et al. External validation of the Kidney Failure Risk Equation and re-calibration with addition of ultrasound parameters. Clin J Am Soc Nephrol 2016. PMID: 26787778
6. Caldinelli A, Faucon AL, Sjölander A, et al. Risk-based referral model to nephrologist specialist care in Stockholm. Nephrol Dial Transplant 2025. PMID: 40632493