Non-invasive chronic kidney disease risk stratification tool derived from retina-based deep learning and clinical...

Area of Science:

• Nephrology
• Ophthalmology
• Artificial Intelligence

Background:

• Chronic kidney disease (CKD) prevention is crucial, but identifying at-risk individuals, especially those with preserved kidney function, remains challenging.
• Early detection and intervention are vital for managing CKD progression and improving patient outcomes.

Purpose of the Study:

• To develop and validate a novel predictive risk score for CKD using deep learning analysis of retinal photographs.
• To assess the performance of this score, termed Reti-CKD, in predicting CKD incidence, particularly in individuals with preserved kidney function.

Main Methods:

• A deep learning algorithm was employed to derive the Reti-CKD score from retinal fundus images.
• The score's predictive performance was validated in two independent longitudinal cohorts: the UK Biobank and the Korean Diabetic Cohort.
• Validation focused on individuals with estimated glomerular filtration rate (eGFR) ≥90 mL/min/1.73 m² and no proteinuria at baseline.

Main Results:

• The Reti-CKD score demonstrated significant association with future CKD events in both cohorts.
• Individuals in the highest Reti-CKD score quartile exhibited substantially increased hazard ratios for CKD development compared to the lowest quartile.
• The Reti-CKD score showed a superior concordance index for CKD prediction compared to conventional eGFR-based methods.

Conclusions:

• The Reti-CKD score, derived from retinal images, effectively stratifies future CKD risk in individuals with preserved kidney function.
• This novel deep learning-based approach offers improved predictive performance over traditional eGFR-based methods for CKD incidence.
• Retinal image analysis presents a promising non-invasive tool for early CKD risk assessment.