Machine learning approaches that use clinical, laboratory, and electrocardiogram data enhance the prediction of...

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Traditional pretest probability (PTP) models for obstructive coronary artery disease (ObCAD) may overestimate risk.
• Integration of standard laboratory findings and electrocardiogram (ECG) raw data into PTP estimation requires further evaluation.

Purpose of the Study:

• To develop and evaluate an ensemble model using machine learning (ML) and deep learning (DL) for ObCAD assessment.
• To incorporate clinical, laboratory, and ECG data for improved PTP estimation.

Main Methods:

• Extracted data from electronic medical records of 7907 patients with suspected ObCAD undergoing coronary angiography (2008-2020).
• Developed an ML model using 27 clinical and laboratory variables.
• Developed a DL model using ECG waveform data.
• Combined models into an ensemble for ObCAD prediction.

Main Results:

• The clinical-laboratory model achieved an AUC of 0.747.
• The ECG model achieved an AUC of 0.685.
• The ensemble model demonstrated the highest AUC of 0.767, with sensitivity, specificity, and F1 score of 0.761, 0.625, and 0.696, respectively.

Conclusions:

• The ensemble model shows superior predictive performance for ObCAD compared to traditional PTP models.
• This approach may enable personalized ObCAD assessment and reduce overestimation of pretest probability.
• Integration of ML/DL with clinical, laboratory, and ECG data offers a promising strategy for cardiovascular risk stratification.