Abstract: TH-PO015
Impact of Retinal Photography-Based Deep Learning System on Risk Stratification for CKD Progression
Session Information
- AI, Digital Health, Data Science - I
November 02, 2023 | Location: Exhibit Hall, Pennsylvania Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Augmented Intelligence, Digital Health, and Data Science
- 300 Augmented Intelligence, Digital Health, and Data Science
Authors
- Joo, Young Su, Yonsei University Institute of Kidney Disease, Seodaemun-gu, Seoul, Korea (the Republic of)
- Koh, Hee Byung, Yonsei University Institute of Kidney Disease, Seodaemun-gu, Seoul, Korea (the Republic of)
- Rim, Tyler Hyungtaek, Mediwhale, Seoul, Korea (the Republic of)
- Kang, Shin-Wook, Yonsei University Institute of Kidney Disease, Seodaemun-gu, Seoul, Korea (the Republic of)
- Park, Jung Tak, Yonsei University Institute of Kidney Disease, Seodaemun-gu, Seoul, Korea (the Republic of)
Background
We had previously developed a deep-learning-based risk evaluation system from retinal photographs, Reti-CKD, for stratifying chronic kidney development risk in kidney function preserved people. This study aims to evaluate whether Reti-CKD can improve risk assessment of kidney disease progression in diabetic patients with prevalent CKD.
Methods
Total of 5348 diabetic patients from two tertiary hospitals in Korea were evaluated. Patients with estimated glomerular filtration rate (eGFR) <90 ml/min/1.73m2 or albuminuria were included. Those with missing data for retinal photograph, serum creatinine, or albuminuria were excluded. Patients were categorized into low-risk, moderate-risk, and high-risk groups according to the KDIGO criteria for prognosis of CKD. The KDIGO groups were additionally dichotomized based on Reti-CKD score (Reti-CKD <20 and ≥20). CKD progression was compared between the categories using Cox regression models. Primary outcome was CKD progression, defined as incremental progression to a higher NKF-KDOQI CKD stage.
Results
The mean age of the patients was 62.4 ± 11.4 years and 60.6% were male. Mean eGFR was 86.6 ± 15.3 mL/min per 1.73 m2 and albuminuria was present in 46.9%. During a median follow-up of 5.0 (interquartile range, 2.5-7.8) years, primary outcome developed in 1379 (25.8%) patients. The primary outcome incidence rate gradually increased with higher KDIGO and Reti-CKD combined risk categories. The risk for CKD progression progressively increased in KDIG moderate-risk and high-risk groups compared to low-risk. When Reti-CKD was incorporated to the KDIGO category, significant stratification of CKD progression risk was noted in the KDIGO low-risk and moderate-risk groups. Additionally, the combination of KDIGO and Reti-CKD classification showed better discrimination power compared to the KDIGO only classification (delta c-statistics, 0.03; 95% CI 0.02 to 0.040).
Conclusion
Retinal photography-based deep learning system (Reti-CKD) further stratifies the risk of CKD progression and improves predictability in diabetic patients with reduced renal function.