Validation of a deep-learning-based retinal biomarker (Reti-CVD) in the prediction of cardiovascular disease: data from UK Biobank

Table 2 Risk of cardiovascular events by the deep-learning-based retinal CVD biomarker (Reti-CVD) in all participants

Risk predictor	N	Cases	Person-years	Incidence (95% CI)	Unadjusted hazard ratio (95% CI)
Reti-CVD
Low	19,304	545	209,186	2.6 (2.4–2.8)	1 (reference)
Moderate	26,543	1900	280,021	6.8 (6.5–7.1)	2.62 (2.38–2.88)
High	2413	321	24,508	13.1 (11.7–14.6)	5.11 (4.45–5.86)
					Adjusted HR trend^a
					1.41 (1.30–1.52)
QRISK3 score
≥ 0 to < 5	26,136	702	283,604	2.5 (2.3–2.7)	1 (reference)
≥ 5 to < 10	14,932	1099	157,442	7.0 (6.6–7.4)	2.84 (2.58–3.12)
≥ 10 to < 15	5668	695	57,723	12.0 (11.2–13.0)	4.93 (4.44–5.48)
≥ 15 to < 20	1207	196	11,988	16.3 (14.2–18.8)	6.75 (5.76–7.91)
≥ 20	317	74	2957	25.0 (19.9–31.4)	10.50 (8.27–13.35)
Total	48,260	2766	513,714	5.4 (5.2–5.6)

^aBased on multivariable model after adjusting QRISK3 five groups. Incidence per 1000 person-years. CI confidence interval. CVD cardiovascular disease. HR hazard ratio. N number at risk. Reti-CVD deep-learning-based retinal CVD biomarker

ISSN: 1741-7015