Truace
Subscribe

Machine learning adds little to standard kidney function equations, review finds

Study title: When machines estimate kidney function and predict risk: Clinical applications of machine learning

A review in Néphrologie & Thérapeutique examines two uses of machine learning in kidney care: estimating glomerular filtration rate (GFR) and predicting how chronic kidney disease progresses. Early studies using neural networks, random forests and ensemble methods looked encouraging for GFR estimation.

Read the study abstract

The use of artificial intelligence, and particularly machine learning, is rapidly expanding in medicine. In nephrology, these approaches have generated growing interest owing to the availability of large clinical, biological, and administrative datasets, as well as the need for more accurate assessment of kidney function and changes over time. Two major areas of application have emerged: estimation of glomerular filtration rate (GFR) and prediction of chronic kidney disease (CKD) progression. Initial studies foc…

By the Truvace desk1 min read1 sourceEvidence: Emerging · 30/100 — Commentary, opinion, or narrative review · 1 independent sourcePeer-reviewedTRV-2026-2293· cite & verify
When machines estimate kidney function and predict risk: Clinical applications of machine learning
Peritoneal dialysis machine by Ashashyou. CC BY-SA 3.0 · https://creativecommons.org/licenses/by-sa/3.0

In brief

Recent large multicenter work suggests the bottleneck is the biomarkers available rather than the algorithms, so smarter models do not fix weak inputs. Progression prediction looks more promising, but proof of clinical advantage over established tools remains limited, and routine use in nephrology is still a future direction.

Main points

  1. GFR estimation with machine learning gives only marginal improvement over EKFC or CKD-EPI when the same variables are used.
  2. Limits in GFR estimation are attributed to available biomarker information rather than algorithm capacity.
  3. Machine learning appears well suited to predicting kidney disease progression by integrating many clinical and laboratory variables.
  4. Few progression models have shown a clear clinical advantage over the Kidney Failure Risk Equation.

The problem

Large multicenter studies show machine learning improves kidney function estimation only marginally over EKFC or CKD-EPI, and few risk models clearly beat the Kidney Failure Risk Equation.

The rundown

The review covers supervised learning approaches such as artificial neural networks, random forests and ensemble methods applied to GFR estimation, motivated by large clinical, biological and administrative datasets and the need to track kidney function over time.

For progression, the models combine numerous clinical and laboratory variables to flag patients at high risk of rapid decline or kidney failure. The authors compare them with the Kidney Failure Risk Equation as the established benchmark.

The paper also discusses strengths, limitations and future directions for integrating these tools into nephrology practice.

Sources

  1. Peer-reviewedNéphrologie & Thérapeutique2026-10-27

The debate