Automatic prediction of coronary artery disease from clinical narratives

Area of Science:

• Computational medicine and artificial intelligence in healthcare.
• Cardiovascular disease research and predictive analytics.

Background:

• Coronary Artery Disease (CAD) is the leading cause of death globally.
• Clinical free text in medical records contains valuable information for disease prediction.
• Existing research has focused on identifying CAD risk factors, not direct prediction from text.

Purpose of the Study:

• To develop and evaluate an automated system for predicting the development of Coronary Artery Disease (CAD) from clinical free text.
• To establish a novel approach for CAD prediction, marking the first attempt at automatic prediction using narrative medical histories.
• To address the challenge of overfitting in small datasets by employing an ontology-guided feature extraction method.

Main Methods:

• Development of a system to analyze narrative medical histories (clinical free text) for CAD prediction.
• Implementation of an ontology-guided approach for feature extraction to manage a limited feature set and prevent overfitting.
• Comparison of the proposed ontology-guided method with two traditional feature selection techniques on a corpus of diabetic patients.

Main Results:

• The proposed ontology-guided system achieved a state-of-the-art performance.
• The system demonstrated a 77.4% F1 score in predicting coronary artery disease development.
• The ontology-guided feature extraction proved effective in handling small datasets and improving prediction accuracy.

Conclusions:

• Automated prediction of Coronary Artery Disease (CAD) from clinical free text is feasible and effective.
• The ontology-guided feature extraction approach is a promising method for predictive modeling in small medical datasets.
• This system represents a significant advancement in leveraging unstructured medical data for cardiovascular disease prediction.