Table 3 Summary of studies in the current literature (chronic kidney disease)
From: Prognostic potentials of AI in ophthalmology: systemic disease forecasting via retinal imaging
Author | Study type | Study year | Key findings | Retinal biomarker | Dataset | Total of events/ participants | Adjusted variable | Study design |
|---|---|---|---|---|---|---|---|---|
Zhang et al. [21] | CKD | September 2019 to November 2019 | DL models analysing retinal photos and clinical data effectively detect CKD and T2DM, predict key health markers like eGFR and blood-glucose levels, and risk-stratify patients. With AUCs of 0.85–0.93, these models accurately identify CKD and T2DM, showcasing potential in disease progression risk stratification. | Retinal Fundus Images | CC-FII | Cross-sectional Cohort CC-FII: 3,156 Longitudinal dataset CC-FII-L: 10,269 External longitudinal test set: 3,376 External test set 1: 8,059 External test set 2: 3,081 | Age Gender Blood pressure Height Weight BMI Hypertension T2DM | Prospective, Retrospective, Longitudinal |
Joo et al. [22] | CKD | UK Biobank (each participant was followed up to 11.6Â years from their initial visit to the last date of visit Feb 28, 2021) Korean Diabetic Cohort (each participant was followed up to 14Â years from the data of initial visit to the last date of visit Feb 28, 2022) | Higher Reti-CKD scores correlate with increased CKD risk and outperform eGFR methods in stratifying future risk among those with normal kidney function. In the highest vs. lowest quartile, hazard ratios were 3.68 (UK Biobank) and 9.36 (Korean Diabetic Cohort), indicating superior Reti-CKD predictive accuracy. | Reti-CVD | UK Biobank Korean Diabetic Cohort | 30,477 UK Biobank 5,014 Korean Diabetic Cohort | Age Gender Diabetic status Hypertension status eGFR | Observational cohort study |
Zhang et al. [23] | CKD | 2006 to 2010 | Every one-year increase in the retinal age gap (model-based retinal age vs. chronological age) is associated with a 10% higher kidney failure risk, highlighting its potential as a non-invasive biomarker. This is based on a DL model analysing retinal images against chronological age. | Retinal age gap | UK Biobank | 500,000 | Age Gender Ethnic background Townsend deprivation indices (TDI) Smoking status Alcohol consumption Physical activity level General health status Diabetic status Blood pressure BMI Cholesterol level eGFR | Prospective cohort study |