Artificial intelligence-aided analysis of retinal imaging may predict cardiovascular mortality – 2 Minute Medicine

1. A model combining artificial intelligence-enabled analysis of retinal vascular images from a large database with several traditional risk factors performed comparably to Framingham risk scores in predicting stroke and myocardial infarction (MI) risk.

2. Addition of retinal vascular scores to Framingham risk scores did not improve prediction of incident stroke and MI.

Evidence Rating Level: 2 (Good)

Study Rundown: Changes in retinal vasculature, such as tortuosity, arteriolar narrowing, and branching, have been shown to be associated with cardiovascular disease severity. Retinal imaging thus has potential as a noninvasive prognostic tool. This study aimed to develop a predictive model using artificial intelligence (AI)-enabled analysis of fundus photographs then assess the models performance relative to existing risk scores. The AI model was developed using images and outcomes data from about 64,000 patients in the UK Biobank and tested on about 6,000 patients in another database. A model including retinal vessel metrics, age, smoking, and medical history [use of blood pressure and cholesterol medications, diabetes history, past stroke or myocardial infarction (MI)] predicted incident stroke or MI about as well as Framingham risk scores. Adding retinal data to Framingham risk scores did not improve their predictive performance. Patients in both databases had <1% incidence each of stroke, MI, and circulatory death, which limits the applicability of this predictive model to higher-risk populations. The overall performance data here suggests that retinal vascular metrics may perform a similar predictive role to serum lipid levels and blood pressure. Clinical integration of fundus photography in this setting would be challenging, but could one day provide a useful alternative.

Click to read the study in BJO

Relevant Reading: Retinal microvascular signs as screening and prognostic factors for cardiac disease: A systematic review of current evidence

In-Depth [retrospective cohort]: For both the UK Biobank and EPIC-Norfolk databases, retinal vessel analysis was performed on non-dilated 45 color fundus photographs centered on the fovea. The AI analysis system was fully automated and returned average vessel width, total vessel area, tortuosity, and variance of widths across both eyes for each patient. The optimism adjusted C-statistic for circulatory mortality from a model including retinal vasculometry (RV), smoking, and medical history was 0.774 for men and 0.748 for women in the EPIC-Norfolk database, with R2 values of 0.392 and 0.333, respectively. The C-statistics for incident stroke and MI ranged from 0.641-0.714 with R2 from 0.150-0.274. Framingham Risk score C-statistics for stroke and MI for the same population ranged from 0.682-0.732 with R2 from 0.199-0.309. The RV-based model placed 45.5% of men and 49.9% of women from EPIC-Norfolk with incident stroke or MI in the top quintile of risk scores, compared to 45.5% and 50.9% for Framingham risk score.

Image: PD

2022 2 Minute Medicine, Inc. All rights reserved. No works may be reproduced without expressed written consent from 2 Minute Medicine, Inc. Inquire about licensing here. No article should be construed as medical advice and is not intended as such by the authors or by 2 Minute Medicine, Inc.

More here:
Artificial intelligence-aided analysis of retinal imaging may predict cardiovascular mortality - 2 Minute Medicine