Back to all papers

Deep learning-based retinal risk stratification for 10-year atherosclerotic cardiovascular mortality in patients with hypertension and diabetes: a retrospective cohort study.

September 1, 2026pubmed logopapers

Authors

Nam D,Seo J,Thakur S,Nusinovici S,Yoo TK,Jang M,Joe BH,Rim TH,Kim HK,Lee M,Hwang S,Lee YH,Lee CJ,Kim H,Park S

Affiliations (10)

  • Department of Internal Medicine, Graduate School, Yonsei University College of Medicine, Seoul, Republic of Korea.
  • Mediwhale Inc., Seoul, Republic of Korea.
  • Singapore Eye Research Institute, Singapore National Eye Centre, Singapore, Singapore.
  • Ophthalmology and Visual Sciences Academic Clinical Program (Eye ACP), Duke-NUS Medical School, Singapore, Singapore.
  • Ophthalmology, Bright St. Mary's Eye Center, Seoul, Republic of Korea.
  • Department of Internal Medicine, Yonsei University College of Medicine, Seoul, Republic of Korea.
  • Institute of Endocrine Research, Yonsei University College of Medicine, Seoul, Republic of Korea.
  • Department of Ophthalmology, Hanyang University Guri Hospital, Hanyang University College of Medicine, Guri, Republic of Korea.
  • Division of Cardiology, Severance Cardiovascular Hospital, Yonsei University College of Medicine, Seoul, Republic of Korea.
  • Institute of Vision Research, Department of Ophthalmology, Yonsei University College of Medicine, Seoul, Republic of Korea.

Abstract

Cardiovascular risk stratification in patients with hypertension and diabetes remains suboptimal. We evaluated the prognostic and incremental value of Dr. Noon CVD, a deep learning (DL)-based retinal biomarker, for 10-year atherosclerotic cardiovascular (ASCVD) mortality. This retrospective, single-center, primary prevention cohort study included 7,832 Korean patients with hypertension or diabetes diagnosed on or before retinal fundus photography (2005 to 2022). Patients with prior cardiovascular disease were excluded. The primary outcome of 10-year ASCVD mortality and a sensitivity analysis of 5-year mortality were stratified using 4-tier and 3-tier Dr. Noon CVD systems, respectively. Multivariable Cox proportional hazards models were adjusted for traditional risk factors and chronic kidney disease (CKD), and incremental predictive value was assessed using the change in the C-index (ΔC-index) and net reclassification improvement (NRI). In the overall population, the Dr. Noon CVD score was independently associated with 10-year ASCVD mortality (hazard ratio [HR] trend, 1.49; 95% confidence interval [CI], 1.21 to 1.85; <i>P</i> < 0.001) after full adjustment for traditional factors and CKD. Adding Dr. Noon CVD to the clinical model significantly improved discrimination (ΔC-index, 0.030; <i>P</i> = 0.001) and risk reclassification (NRI, 0.484; 95% CI, 0.233 to 0.666; <i>P</i> < 0.001). This independent prognostic value was consistent in both hypertension (10-year HR trend, 1.48; <i>P</i> = 0.006) and diabetes (10-year HR trend, 1.42; <i>P</i> = 0.003) subgroups. The reclassification improvement was greater in the diabetes than the hypertension subgroup (10-year NRI, 0.481 vs. 0.413). The score's prognostic value was independent of diabetic retinopathy status. DL-based retinal imaging provides prognostic value for 10-year ASCVD mortality independent of the available clinical risk factors (age, sex, hypertension, diabetes, and CKD) and consistent under adjustment for available laboratory measures, supporting its role as an accessible opportunistic screening tool for long-term primary prevention in cardiometabolic clinics.

Topics

Journal Article

Ready to Sharpen Your Edge?

Subscribe to join 11k+ peers who rely on RadAISlice. Get the essential weekly briefing that empowers you to navigate the future of radiology.

We respect your privacy. Unsubscribe at any time.