Precise risk-prediction model including arterial stiffness for new-onset atrial fibrillation using machine learning...

Area of Science:

• Cardiology
• Medical Informatics
• Preventive Medicine

Background:

• Atrial fibrillation (AF) is a common arrhythmia and a major risk factor for stroke.
• Existing risk prediction models for AF often lack comprehensive data on arterial stiffness and demographic diversity.
• Annual health check-ups in Japan provide a valuable dataset for studying cardiovascular health.

Purpose of the Study:

• To develop and validate a novel machine learning-based risk prediction model for new-onset atrial fibrillation (AF).
• To incorporate electrocardiogram data, demographic factors, hypertension, and arterial stiffness into the AF prediction model.
• To identify key predictors of new-onset AF in a large Japanese population undergoing regular health examinations.

Main Methods:

• Utilized machine learning techniques, specifically eXtreme Gradient Boosting and Shapley Additive Explanation values, for risk model development.
• Employed a dataset of 13,410 individuals from annual health check-ups (2005-2015), with 110 new-onset AF cases.
• Randomly split data into 80% training and 20% testing sets to build and validate the predictive model.

Main Results:

• The developed risk prediction model achieved an area under the receiver operator characteristic curve of 0.789 in the test set.
• Key predictors for new-onset AF included age, cardio-ankle vascular index (arterial stiffness measure), estimated glomerular filtration rate, and sex.
• Other significant predictors identified were body mass index, uric acid, liver enzymes, triglycerides, and systolic blood pressure.

Conclusions:

• A new machine learning model effectively predicts new-onset atrial fibrillation using readily available health check-up data, including arterial stiffness.
• The model's performance suggests its utility in identifying high-risk individuals for targeted AF prevention strategies.
• Integrating arterial stiffness measures into risk prediction models enhances their accuracy for forecasting AF development in the general population.