ASN's Mission

To create a world without kidney diseases, the ASN Alliance for Kidney Health elevates care by educating and informing, driving breakthroughs and innovation, and advocating for policies that create transformative changes in kidney medicine throughout the world.

learn more

Contact ASN

1401 H St, NW, Ste 900, Washington, DC 20005

email@asn-online.org

202-640-4660

The Latest on X

Kidney Week

Please note that you are viewing an archived section from 2022 and some content may be unavailable. To unlock all content for 2022, please visit the archives.

Abstract: FR-PO909

Retinal Photograph-Based Deep Learning Predicts CKD Among People With Preserved Kidney Function

Session Information

Category: CKD (Non-Dialysis)

  • 2201 CKD (Non-Dialysis): Epidemiology‚ Risk Factors‚ and Prevention

Authors

  • Joo, Young Su, Yonsei University Institute of Kidney Disease, Seodaemun-gu, Seoul, Korea (the Republic of)
  • Lee, Geunyoung, Medi Whale, Seoul, Korea (the Republic of)
  • Park, Jung Tak, Yonsei University Institute of Kidney Disease, Seodaemun-gu, Seoul, Korea (the Republic of)
Background

Predicting kidney disease is challenging, especially in people with preserved kidney function. We developed a novel machine learning based risk scoring system in prediction of future risk of chronic kidney disease (CKD) using retinal photographs and the performance of this risk stratification system was externally validated.

Methods

We used 232,779 retinal photographs from three datasets from South Korea, and the UK to train and validate the algorithm. First, using a dataset from a Korean health-screening centre, we trained a deep learning algorithm to predict the probability of the CKD presence (i.e., deep-learning retinal CKD score, RetiCKD). Second, predictability of the RetiCKD was evaluated using Cox hazards models in two separate longitudinal cohorts for future CKD development. Those with eGFR <90 ml/min/1.73m2 or proteinuria at baseline were excluded. For the UK Biobank cohort, CKD development was defined with ICD-10 and OPCS-4 codes. For the Korean clinical cohort, CKD was defined as ≥2 occurrence of eGFR <60 ml/min/1.73m2.

Results

For the 33,814 participants in the UK Biobank cohort, mean age was 54.8 years and 13,670 (44.6%) were males. During a median follow-up of 10.8 years, 879 (1.2%) cases of CKD developed. When the participants were categorized to RetiCKD score tertile, the risk of CKD development increased in the highest tertile than the lowest tertile (Hazard ratio [95% CI], 2.50 [2.03-3.09]). Compared to the eGFR only model, eGFR with RetiCKD model showed superior concordance and reclassification performance (ΔC-statistics, 0.052 [95% CI, 0.035-0.069]; net reclassification index [NRI], 0.142 [95% CI, 0.096-0.187]). The Korean clinical cohort consisted of 4,050 diabetes patients. The mean age was 55.9 years and 2,193 (54.1%) were male. CKD occurred in 158 (3.9%) patients during a median follow-up of 6.1 years. CKD development risk was higher in the highest tertile than the lowest tertile (Hazard ratio [95% CI], 4.95 [2.42-10.1]). Predictability improvement was observed in the eGFR with RetiCKD model compared to the eGFR only model (ΔC-statistics, 0.039 [95% CI, 0.004-0.075]; NRI, 0.207 [95% CI, 0.049-0.267]).

Conclusion

A novel deep-learning and retinal photograph-derived CKD risk score successfully stratified future CKD risk among people with normal kidney function, in both the general population and diabetes patients.