Predicting hypertension risk using AI-derived retinal vascular complexity: a report from the Canadian Longitudinal Study on Aging (CLSA)

Background Retinal fractal dimension (Df), a quantitative metric of density and overall complexity of the retinal microvasculature, can be derived using deep learning models. Since the retinal and systemic vasculature share key anatomical, physiological and structural features, Df may offer a non-invasive measure of hypertension risk(1,2). Purpose We examined whether retinal vascular complexity, quantified through Df measures, is associated with incident hypertension, sought to identify circulating metabolites linked to Df and evaluated whether Df measures improve hypertension prediction beyond clinical risk factors. Methods Analysis included 30,097 participants from the Canadian Longitudinal Study on Aging (mean age 62.96 ± 10.25; 49% male). Digital fundus photography yielded 57,434 retinal images from 25,678 participants at baseline. Image quality (good/moderate/poor) and total, arterial, and venular Df values were assessed using the automated Vessel Assessment and Measurement Platform for Images of the Retina (VAMPIRE) software(3). Higher Df (per 1-unit increase) indicates greater vascular branching. Baseline plasma metabolomics profiling (808 metabolites; Metabolon platform) were available for 9,785 participants. The final analytic sample (n=3,093) had complete data for Df, metabolomics and incident hypertension (Figure 1). Incident hypertension was defined as average systolic >140 mmHg or diastolic BP >90 mmHg, and/or antihypertensive medication use at 3-year follow-up. Logistic regression assessed associations between Df and incident hypertension (adjusted for baseline systolic BP and diabetes). Age and sex adjusted linear regression assessed associations between Df and metabolites. Predictive performance of Df was evaluated against a clinical model (baseline systolic BP and diabetes) using Area Under the Curve of the Receiver Operating Characteristic (AUC-ROC). Results 250 participants (8.08%) developed hypertension. Higher total and arterial Df were associated with lower odds of incident hypertension (adjusted total Df Odds Ratio, OR=0.80, 95% Confidence Interval, CI:0.63-0.99; arterial Df OR=0.79, CI:0.68-0.93). Twenty-seven metabolites showed nominal associations with incident hypertension, including hexadecadienoate, which was also positively associated with total Df and lower hypertension risk (β=0.02 ± 0.01, p=0.02; OR=0.87; 95% CI: 0.76-0.99, p=0.04) (Figure 2). Predictive performance of Df alone was modest (AUC=0.54) compared with the clinical model (AUC=0.76). Conclusion Retinal vascular complexity was associated with incident hypertension but added limited predictive value beyond clinical risk factors. Associations between hexadecadienoate, microvasculature and hypertension, suggest a potential protective role of long-chain polyunsaturated fatty acid metabolism in vasodilation and vascular inflammation(4). As these fatty acids are commonly obtained through diet, nutrition may represent a modifiable pathway.