Truace
Subscribe

Four blood proteins plus AI improved kidney disease prediction in a 27,429-person study

Study title: Identification of plasma biomarkers for chronic kidney disease by integrating proteomics and machine learning

A prospective cohort study of 27,429 UK Biobank participants without chronic kidney disease at baseline used machine learning to find blood proteins that predict who develops the disease. Of 205 proteins significantly linked to incident CKD, the models narrowed the field to four: EDA2R, GDF15, HAVCR1 and SCARB2. Combined with demographic data, the panel reached an AUC of 0.830, up from 0.805 for the demographic model alone.

Read the study abstract

Background Chronic kidney disease (CKD) is a major global health burden, and early identification of individuals at risk remains challenging. This study aimed to identify plasma proteomic biomarkers associated with incident CKD and develop a machine learning-based prediction model integrating proteomic and clinical data. Methods We conducted a prospective cohort study of 27,429 UK Biobank participants without baseline CKD after excluding those with estimated glomerular filtration rate Results A total of 205 prot…

Truvace record2 min read1 sourceEvidence: Moderate · 65/100 — Observational study · n = 27,429 · 1 independent sourcePeer-reviewedTRV-2026-2410 · cite or check it
Identification of plasma biomarkers for chronic kidney disease by integrating proteomics and machine learning"Medical laboratories and clinics in Kinwat" by Mouryan, CC BY-SA 4.0.

In brief

Early identification of people at risk of chronic kidney disease remains difficult, and the study describes the condition as a major global health burden. A compact four-protein panel could give clinicians complementary information for risk stratification. The authors say the result is an internal validation, so external validation is needed before any clinical use.

Main points

  1. The study followed 27,429 UK Biobank participants who did not have chronic kidney disease at baseline.
  2. A total of 205 proteins were significantly associated with incident CKD.
  3. LASSO selection identified 23 candidate proteins, and LightGBM narrowed these to four: EDA2R, GDF15, HAVCR1 and SCARB2.
  4. The four-protein model alone performed comparably to the demographic model (AUC 0.796 vs 0.805; p = 0.14).
  5. Combining the proteins with demographic data improved discrimination to an AUC of 0.830 (95% confidence interval 0.806-0.855).
  6. SHAP and Kaplan-Meier analyses supported the predictive value of these proteins, and sensitivity analyses confirmed robustness.

The gain

In a cohort study of 27,429 UK Biobank participants, adding four plasma proteins to a demographic model raised discrimination for incident chronic kidney disease to an AUC of 0.830, from 0.805 for demographics alone.

The full story

Researchers used a prospective cohort design drawing on UK Biobank participants without baseline CKD. They tested which plasma proteins were associated with later kidney disease, finding 205 significant associations, then applied LASSO and LightGBM machine learning to select a smaller set.

The final four-protein panel (EDA2R, GDF15, HAVCR1 and SCARB2) matched a demographic model on its own, with AUCs of 0.796 and 0.805 (p = 0.14). Only the combination of proteins and demographic data improved discrimination, reaching 0.830, and it was assessed across prediction horizons up to 10 years.

SHAP and Kaplan-Meier analyses and sensitivity checks supported the findings. The authors conclude that the panel may add complementary information for risk stratification but that external validation is required before clinical application.

Sources

  1. Peer-reviewedKidney Research and Clinical Practice2026-11-02

The debate