Retinal vascular fingerprints predict incident stroke: findings from the UK Biobank cohort study.

Abstract

Objective: To investigate the associations between a comprehensive set of retinal vascular parameters and incident stroke to unveil new associations and explore its predictive power for stroke risk.

Methods: Retinal vascular parameters were extracted from the UK Biobank fundus images using the Retina-based Microvascular Health Assessment System. We used Cox regression analysis, adjusted for traditional risk factors, to examine the associations, with false discovery rate adjustment for multiple comparisons. Receiver operating characteristic (ROC) curves were used to assess their predictive values.

Results: During a median follow-up of 12.5 years, 749 incident strokes occurred among 45 161 participants. The analysis identified 29 significant parameters associated with stroke risk, with a notable dominance of density parameters (over half). Each SD change in these parameters increased stroke risk by 9.8% to 19.0%. For identified calibre parameters, each SD change was associated with an increased risk (ranging from 10.1% to 14.1%). For identified complexity parameters and arterial inflection count tortuosity, each SD decrease was linked to an increased risk (ranging from 10.4% to 19.5%). The introduction of retinal vascular parameters improved the area under the ROC curve to 0.752, significantly outperforming the model using only traditional risk factors (0.739, p<0.001).

Conclusions: Retinal vascular analysis, a non-invasive screening approach for stroke risk assessment, performed better than traditional risk stratification models. The 29 novel retinal indicators identified offer new avenues for stroke pathophysiology research.