Machine-learning-based models to predict cardiovascular risk using oculomics and clinic variables in KNHANES

Area of Science:

• Cardiovascular disease risk prediction
• Oculomics and clinical data integration
• Machine learning applications in healthcare

Background:

• Established correlation between triglyceride-glucose (TyG) index and atherogenic index of plasma (AIP) with cardiovascular disease (CVD) risk.
• Identified a gap in research for non-invasive and rapid CVD risk prediction methods.
• Need for advanced predictive models utilizing accessible patient data.

Purpose of the Study:

• To develop and validate a machine learning model for predicting cardiovascular risk.
• To utilize oculomics measurements and clinical questionnaires as input variables.
• To assess the model's efficacy in predicting elevated TyG-index or AIP levels.

Main Methods:

• Utilized data from the Korean National Health and Nutrition Examination Survey (KNHANES) (2008-2012).
• Trained 25 machine learning algorithms on oculomics and clinical data for 32,122 participants.
• Evaluated model performance using Area Under the Receiver Operating Characteristic Curve (AUC), accuracy, precision, recall, and F1 score.

Main Results:

• The best-performing model identified TyG-index cut-offs (8.0, 8.75, 8.93) and AIP cut-offs (0.318, 0.34) with high AUCs (0.809-0.911).
• Internal and external validation demonstrated consistent predictive capacity for both TyG-index and AIP.
• Observed significant gender-based variations in predictive accuracy for certain cut-offs, with near-identical performance at TyG-index 8.93.

Conclusions:

• A simple, effective, and non-invasive machine learning model was developed for cardiovascular risk prediction.
• The model demonstrates significant clinical value for identifying individuals at elevated risk in the general population.
• Oculomics combined with clinical data offers a promising avenue for rapid and non-invasive health assessments.