Probabilistic Graphical Modeling for Estimating Risk of Coronary Artery Disease: Applications of a Flexible...

Area of Science:

• Computational biology and bioinformatics
• Medical informatics and decision support systems

Background:

• Coronary artery disease (CAD) is a leading global cause of mortality and morbidity.
• Current risk prediction models often lack transparency and struggle with complex causal relationships.
• There is a need for advanced models that can improve personalized diagnosis and therapy selection for CAD.

Purpose of the Study:

• To utilize Bayesian networks (BNs) for modeling and predicting the risk of coronary artery disease (CAD).
• To demonstrate the utility of BNs in incorporating background knowledge, handling missing data, and enabling adaptive decision-making under uncertainty.
• To compare the performance of BNs against traditional machine learning classifiers for CAD risk prediction.

Main Methods:

• Bayesian networks (BNs) were employed to model CAD risk using the Z-Alizadehsani dataset, comprising 303 Iranian patients.
• The study explored BNs' capabilities in managing incomplete observations and facilitating adaptive decision-making.
• Performance was evaluated using the area under the receiver-operating characteristic curve (AUC) via 10-fold cross-validation.

Main Results:

• BNs achieved a mean 10-fold AUC of 0.93 ± 0.04, comparable to logistic regression, SVM, and ANN.
• BNs demonstrated superior probability calibration compared to other machine learning classifiers.
• The study successfully illustrated the application of BNs for prediction with missing data and adaptive prognostic value calculation.

Conclusions:

• Bayesian networks are powerful and versatile tools for risk prediction and health outcomes research.
• BNs complement traditional statistical methods, especially in medical domains with uncertain or incomplete information.
• The interpretability and robustness of BNs make them particularly valuable for personalized CAD diagnosis and treatment decisions.