Machine learning and deep learning to predict mortality in patients with spontaneous coronary artery dissection

Area of Science:

• Cardiology
• Medical Informatics
• Artificial Intelligence

Background:

• Spontaneous coronary artery dissection (SCAD) has a variable clinical course, lacking reliable mortality prediction methods.
• The utility of machine learning (ML) and deep learning (DL) for risk prediction in rare diseases with infrequent events remains uncertain.

Purpose of the Study:

• To evaluate the effectiveness of deep neural networks (DNNs) in predicting in-hospital mortality for patients with SCAD.
• To compare the predictive performance of DL and ML models against traditional methods using electronic health record (EHR) data.

Main Methods:

• A deep neural network was applied to EHR data from 375 SCAD patients to predict in-hospital mortality.
• Multiple ML models (logistic regression, random forest, AdaBoost, etc.) were trained and evaluated using cross-validation.
• Model performance was assessed using the area under the receiver-operator characteristic curve (AUC) and balanced accuracy.

Main Results:

• The best-performing DL algorithm achieved an AUC of 0.98 (95% CI 0.97-0.99) for predicting in-hospital mortality.
• DL models demonstrated significantly higher predictive accuracy compared to other ML models (P < 0.0001).
• AdaBoost achieved an AUC of 0.95, while random forest had an AUC of 0.50.

Conclusions:

• Deep learning models, particularly DNNs, offer superior predictive accuracy and discriminative power for identifying SCAD patients at risk of early mortality.
• This study supports the application of advanced ML/DL techniques for risk prediction in uncommon cardiovascular diseases like SCAD.
• EHR data combined with DL can enhance clinical decision-making for SCAD patient management.