Development and validation of an artificial intelligence-based model for cardiovascular disease prediction using...

Area of Science:

• Machine learning applications in healthcare
• Cardiovascular disease epidemiology
• Predictive modeling in medicine

Background:

• Cardiovascular disease (CVD) poses a significant mortality challenge in Iran.
• Machine learning (ML) models are being explored to identify CVD risk factors.
• Evaluating deep learning and mixed-effects models for predicting 10-year CVD incidence.

Purpose of the Study:

• To determine significant predictive factors for cardiovascular disease (CVD) events in Iran.
• To assess the effectiveness of deep learning models (LSTM, GRU) and mixed-effects logistic models.
• To predict 10-year CVD incidence using longitudinal data.

Main Methods:

• Analysis of 4,872 adults (≥30 years) without prior CVD, followed for 10 years.
• Utilized demographic, behavioral, and biochemical data as input features.
• Employed Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) deep learning models to analyze risk factor dynamics.

Main Results:

• 545 participants (11.2%) experienced CVD events during follow-up.
• The GRU model outperformed the LSTM model in both sexes (AUC ~0.738-0.739).
• GRU model performance was comparable or superior to mixed-effects models, using only 21 variables.

Conclusions:

• Deep learning models, particularly GRU, effectively predict future cardiovascular disease events.
• These models leverage longitudinal medical data efficiently.
• Comparable performance achieved with a limited feature set highlights model efficacy.