Retinal photographs to predict life's essential 8 for cardiovascular risk stratification: a novel deep-learning-based...

Area of Science:

• Cardiology
• Ophthalmology
• Artificial Intelligence

Background:

• Cardiovascular diseases (CVDs) are a leading cause of mortality worldwide.
• Life's essential 8 (LE8) and retinal photographs are known indicators of CVD risk.
• Existing CVD risk stratification tools have limitations in accuracy and accessibility.

Purpose of the Study:

• To develop and validate RetiLE8, a novel deep learning tool for CVD risk stratification.
• To predict Life's essential 8 (LE8) scores using retinal photographs.
• To assess the performance of RetiLE8 compared to LE8 and the Pooled Cohort Equations (PCE).

Main Methods:

• Utilized UK Biobank prospective cohort data, including retinal photographs from over 36,000 participants.
• Developed a deep learning model to predict LE8 scores from retinal images, creating the RetiLE8 score.
• Employed Cox proportional hazards models and Harrell's concordance index (C-index) for outcome prediction and model comparison.

Main Results:

• RetiLE8 demonstrated a modest correlation with LE8 scores.
• A one SD increase in RetiLE8 score was associated with significantly lower all-cause mortality (13%) and CVD event risk (10%).
• RetiLE8 showed comparable discrimination to LE8 and PCE, significantly improving risk stratification beyond both.

Conclusions:

• RetiLE8 shows promise as a complementary tool for cardiovascular disease risk stratification.
• The tool effectively predicts CVD risk using retinal photographs, offering a novel approach.
• Further prospective studies are needed to evaluate RetiLE8's clinical utility in real-world settings.