Derivation and independent validation of kidneyintelX.dkd: A prognostic test for the assessment of diabetic kidney disease progression

Girish N. Nadkarni, Sharon Stapleton, Dipti Takale, Katherine Edwards, Kara Moran, Gohar Mosoyan, Michael K. Hansen, Michael J. Donovan, Hiddo J.L. Heerspink, Fergus Fleming, Steven G. Coca

Research output: Contribution to journalArticlepeer-review


Aims: To develop and validate an updated version of KidneyIntelX (kidneyintelX.dkd) to stratify patients for risk of progression of diabetic kidney disease (DKD) stages 1 to 3, to simplify the test for clinical adoption and support an application to the US Food and Drug Administration regulatory pathway. Methods: We used plasma biomarkers and clinical data from the Penn Medicine Biobank (PMBB) for training, and independent cohorts (BioMe and CANVAS) for validation. The primary outcome was progressive decline in kidney function (PDKF), defined by a ≥40% sustained decline in estimated glomerular filtration rate or end-stage kidney disease within 5 years of follow-up. Results: In 573 PMBB participants with DKD, 15.4% experienced PDKF over a median of 3.7 years. We trained a random forest model using biomarkers and clinical variables. Among 657 BioMe participants and 1197 CANVAS participants, 11.7% and 7.5%, respectively, experienced PDKF. Based on training cut-offs, 57%, 35% and 8% of BioMe participants, and 56%, 38% and 6% of CANVAS participants were classified as having low-, moderate- and high-risk levels, respectively. The cumulative incidence at these risk levels was 5.9%, 21.2% and 66.9% in BioMe and 6.7%, 13.1% and 59.6% in CANVAS. After clinical risk factor adjustment, the adjusted hazard ratios were 7.7 (95% confidence interval [CI] 3.0-19.6) and 3.7 (95% CI 2.0-6.8) in BioMe, and 5.4 (95% CI 2.5-11.9) and 2.3 (95% CI 1.4-3.9) in CANVAS, for high- versus low-risk and moderate- versus low-risk levels, respectively. Conclusions: Using two independent cohorts and a clinical trial population, we validated an updated KidneyIntelX test (named kidneyintelX.dkd), which significantly enhanced risk stratification in patients with DKD for PDKF, independently from known risk factors for progression.

Original languageEnglish
Pages (from-to)3779-3787
Number of pages9
JournalDiabetes, Obesity and Metabolism
Issue number12
StatePublished - Dec 2023


  • cohort study
  • diabetic nephropathy
  • machine learning
  • observational study
  • primary care


Dive into the research topics of 'Derivation and independent validation of kidneyintelX.dkd: A prognostic test for the assessment of diabetic kidney disease progression'. Together they form a unique fingerprint.

Cite this