Integrating a Retinal AI score into the NHS systemic cardiovascular prevention pathway: a validation study based on NICE Guidelines

Background/Introduction Primary prevention of cardiovascular disease (CVD) requires precise risk stratification, traditionally involving routine blood tests. However, universal blood testing is resource-intensive. A streamlined strategy using a novel artificial intelligence (AI)-based retinal cardiovascular score (Retinal AI score) could optimise resource allocation within National Health Service (NHS) without compromising patient safety. Purpose To evaluate the efficiency and clinical safety of a novel pre-screening triage strategy—exempting low-risk individuals from blood tests based on a Retinal AI score and diabetes status—compared to systemic standard care adapted from cardiovascular guidelines in the United Kingdom (UK). Methods Data were obtained from the UK Biobank, a prospective population-based cohort. Baseline clinical and retinal imaging data were collected between 2006–2010, with follow-up until February 2021 [1]. Individuals without prior CVD were included. The standard strategy was adapted from the primary risk thresholds of the National Institute for Health and Care Excellence (NICE) guidelines (NG238) [2]. It mandated full lipid and renal profiles to identify high-risk individuals using the QRISK3 algorithm (score ≥10%) [3] or an estimated glomerular filtration rate (eGFR) <60 mL/min/1.73m2. The Retinal AI-guided strategy employed a novel clinical pathway: patients with diabetes were triaged for mandatory standard testing, whereas non-diabetic individuals classified as low-risk by the Retinal AI score were safely exempted from blood tests. We compared the incidence rate (IR) of cardiovascular events between the safe cohorts of both strategies. Non-inferiority was evaluated using rate ratios (RR) and a Cox proportional hazards model. Results We analysed 40,850 individuals (55.2±8.2 years, 57.9% female) Over a median follow-up of 10.0 years, the Retinal AI-guided strategy successfully exempted 11,471 individuals (28.1%) from routine blood testing. The safe cohort defined by the Retinal AI-guided strategy (n=21,056) had a CVD IR of 2.91 per 1,000 person-years (PY), comparable to 2.70 per 1,000 PY in the standard safe by the Standard strategy (n=16,995). The RR was 0.931 (p=0.246), indicating statistical non-inferiority. The Cox proportional hazards model confirmed no significant difference in CVD risk between the Retinal AI-guided and standard safe cohorts (hazard ratio 1.08; 95% confidence interval 0.95-1.21; p=0.245). Conclusion(s) Integrating a Retinal AI score with clinical diabetes triage into the NHS systemic prevention pathway can safely reduce routine blood tests by over 28%. This approach is clinically non-inferior to NICE guideline-based risk stratification using the QRISK3 algorithm, offering a scalable method to ease the burden on healthcare systems.