Kidney Week

Abstract: SA-PO0074

External Validation of a Fundus Photograph-Based Artificial Intelligence Model for Five-Year Cardiovascular Risk Prediction in Patients with CKD

Session Information

• Artificial Intelligence: Kidney Imaging, Pathology, and Precision Phenotyping
  October 24, 2026 | Location: Exhibit Hall A, Convention Center
  Abstract Time: 10:00 AM - 12:00 PM

Category: Artificial Intelligence, Digital Health, and Data Science

• 300 Artificial Intelligence, Digital Health, and Data Science

Authors

• Joo, Young Su, Yonsei University Institute of Kidney Disease, Seoul, Korea (the Republic of)

Young Su Joo, MD, PhD

• Seo, Jaewon, Mediwhale, Seoul, Korea (the Republic of)

Jaewon Seo, MSc

• Nam, Dongjin, Mediwhale, Seoul, Korea (the Republic of)

Dongjin Nam, MD

• Nusinovici, Simon, Mediwhale, Seoul, Korea (the Republic of)

Simon Nusinovici, PhD

• Thakur, Sahil, Mediwhale, Seoul, Korea (the Republic of)

Sahil Thakur, MBBS, MS, PhD, MBA

• Rim, Tyler, Mediwhale, Seoul, Korea (the Republic of)

Tyler Rim, MD

• Park, Jung Tak, Yonsei University Institute of Kidney Disease, Seoul, Korea (the Republic of)

Jung Tak Park, MD, PhD

Background

Patients with chronic kidney disease (CKD) face substantially elevated cardiovascular risk, yet accurate risk stratification in this population remains challenging. Dr. Noon CVD is a fundus photograph–based artificial intelligence model for cardiovascular risk prediction, but its performance in patients with CKD has not been previously evaluated. We assessed the external validity of Dr. Noon CVD for predicting 5-year major adverse cardiovascular events (MACE) in CKD cohort.

Methods

This retrospective cohort study included 9,606 adults with CKD (estimated glomerular filtration rate <60 mL/min/1.73 m² and/or albuminuria) treated at three hospitals affiliated with Yonsei University Health System between November 2002 and January 2026. The primary outcome was a 5-year composite of myocardial infarction, stroke, heart failure, coronary revascularization, or cardiovascular death. Risk stratification was evaluated across pre-defined Dr. Noon CVD risk categories using Cox proportional hazards models. Discrimination of a model integrating Dr. Noon CVD with age and sex was assessed by 5-year C-index.

Results

Among 9,606 participants (mean age 57.4 ± 15.1 years; 54.7% male), the cohort was right-shifted on the Dr. Noon CVD score distribution (median 43, IQR 36–49), with 61.0% classified as high or very high risk. Over a median follow-up of 5.16 years, 774 composite cardiovascular events occurred within 5 years (8.1%). Dr. Noon CVD stratified 5-year cardiovascular risk: event rates rose stepwise from 1.9% in the low-risk group to 5.2%, 9.4%, and 13.8% in moderate, high, and very-high-risk groups, respectively (P for trend <0.001). Compared with low risk, unadjusted hazard ratios were 2.95 (95% CI 1.95–4.45), 5.76 (3.93–8.46), and 8.58 (5.80–12.70). After adjustment for age and sex, hazard ratios were 2.03 (1.02–3.10), 3.14 (2.08–4.73), and 4.06 (2.64–6.23), respectively, indicating independent prognostic value beyond demographics. The 5-year C-index of the model integrating Dr. Noon CVD with age and sex was 0.678 (95% CI 0.661–0.698).

Conclusion

In this multicenter chronic kidney disease cohort, a fundus photograph–based artificial intelligence model independently stratified 5-year cardiovascular risk after adjustment for age and sex. These findings support the application of Dr. Noon CVD for cardiovascular risk assessment in CKD populations.

Funding

• Government Support – Non-U.S.