Digital Algorithm Outperforms Current Prediction Model for Kidney Failure, Death Risk in CKD


KDpredict uses age, sex, eGFR, and albuminuria to calculate 1-5 year risk of kidney failure and death in chronic kidney disease, outperforming the kidney failure risk equation.

Pietro Ravani, MD, PhD | Credit: Cochrane Kidney and Transplant

Pietro Ravani, MD, PhD

Credit: Cochrane Kidney and Transplant

A new tool for predicting patients’ risk of kidney failure and all-cause death may offer a promising alternative to the current benchmark risk prediction model for chronic kidney disease (CKD), consistently outperforming the kidney failure risk equation in external testing.1

The predictive algorithm and digital dashboard, known as KDpredict, was designed to simultaneously predict kidney failure and death over 1-5 year time horizons and support holistic decision-making for patients with moderate to severe CKD while addressing shortcomings of the kidney failure risk equation.1

“We’ve been able to show that KDpredict is consistently more accurate in predicting the risks of kidney failure and death in adults with moderate-to-severe chronic kidney disease than the risk prediction model currently in use,” Pietro Ravani, MD, PhD, professor and clinician scientist in the department of medicine and institute of public health at the University of Calgary, said in a press release.2

Acknowledging the shortcomings of the current kidney failure risk equation, investigators sought to build a tool to provide risk predictions for both kidney failure, accounting for the competing risk of death, and all-cause death in adults with newly documented moderate to severe CKD. Of note, the study proposes a strategy for prediction modeling to adapt to local settings rather than a one-size-fits-all model.1

The multinational, longitudinal, population-based cohort study included individuals with newly recorded chronic kidney disease stage G3b-G4, defined as estimated glomerular filtration rate (eGFR) 15-44 mL/min/1.73 m2. Population-based health data were linked to form 3 cohorts in Alberta, Denmark, and Scotland, with the Alberta data used to train KDpredict and tested against data from Denmark and Scotland.1

The super learner algorithm selected the best-performing regression models or machine learning algorithms based on their ability to predict kidney failure and mortality with minimized cross-validated prediction error based on the Brier score. Prespecified learners, including age, sex, eGFR, albuminuria, presence of diabetes, and cardiovascular disease, were ranked and combined in an ensemble or the learner with the lowest cross-validated prediction error was selected.1

The index of prediction accuracy, a measure of calibration and discrimination calculated from the Brier score, was used to compare KDpredict with the kidney failure risk equation. Outcomes included all-cause death and kidney failure, defined as maintenance kidney replacement treatment or eGFR of 10 mL/min/1.73 m2 sustained for more than 90 days, whichever came first.1

In total, 67,942 patients from Canada, 17,528 patients from Denmark, and 7740 patients from Scotland were included. Among the cohort, the median age was 77-80 years and the median eGFR was 39 mL/min/1.73 m2.1

Median follow-up times were 5-6 years in all cohorts. Kidney failure rates ranged from 0.8-1.1 per 100 person-years and death rates ranged from 10-12 per 100 person-years. Investigators noted KDpredict was more accurate than the kidney failure risk equation in predicting kidney failure risk. Specifically, the 5-year index of prediction accuracy was 27.8% (95% CI, 25.2% to 30.6%) for KDpredict versus 18.1% (95% CI, 15.7% to 20.4%) for the kidney failure risk equation in Denmark and 30.5% (95% CI, 27.8% to 33.5%) for KDpredict versus 14.2% (95% CI, 12.0% to 16.5%) for the kidney failure risk equation in Scotland.1

In hypothetical clinical scenarios, investigators pointed out predictions from the kidney failure risk equation and KDpredict differed substantially, potentially leading to diverging treatment decisions. In one example, an 80-year-old man with an eGFR of 30 mL/min/1.73 m2 and an albumin-to-creatinine ratio of 100 mg/g would receive a 5-year kidney failure risk prediction of 10% from the kidney failure risk equation but would receive a 5-year risk prediction of 2% for kidney failure and 57% for mortality from KDpredict.1

Investigators also noted individual risk predictions from KDpredict were accurate for both outcomes with 4 or 6 variables. Additionally, models retrained using older data provided accurate predictions when tested in temporally distinct, more recent data.1

Investigators highlighted several potential limitations to these findings, some of which include the use of data from 3 countries with predominantly White populations, the use of albuminuria measurements limited to albumin-to-creatinine ratio or protein-to-creatinine ratio in the external testing cohorts, the static nature of the prediction tool, and the need for testing in a randomized trial to validate its usefulness.1

“KDpredict is unique in its ability to provide accurate predictions of risk for both clinical outcomes in adults with this severity of CKD at the point of first onset, when a timely discussion should occur,” investigators concluded.1 “By presenting risk predictions of both kidney failure and death, KDpredict supports patient-centered care and holistic decision making.”


  1. Liu P, Sawhney S, Heide-Jørgensen U, et al. Predicting the risks of kidney failure and death in adults with moderate to severe chronic kidney disease: multinational, longitudinal, population based, cohort study. BMJ 2024; doi:
  2. University of Calgary. UCalgary researchers can predict kidney failure more accurately. EurekAlert! April 23, 2024. Accessed April 26, 2024.
Related Videos
Mark Barakat, MD: Stable IOP Outcomes After Aflibercept 8 mg in DME | Image Credit: Retina Macula Institute of Arizona
Noa Krugliak Cleveland, MD | Credit: University of Chicago
Video 6 - "Use of Oral Corticosteroids in Asthma"
Video 5 - "Thinking About Endotypes when Managing Asthma"
Ali Rezaie, MD | Credit: X
Remo Panaccione, MD | Credit: University of Calgary
Francisca Joly, MD, PhD | Credit: The Transplantation Society
Naim Alkhouri, MD | Credit: Naim Alkhouri on X
Paul Feuerstadt, MD | Yale School of Medicine
© 2024 MJH Life Sciences

All rights reserved.