Minimal Patient Clinical Variables to Accurately Predict Stress Echocardiography Outcome: Validation Study Using...

Area of Science:

• Cardiology
• Machine Learning
• Medical Diagnostics

Background:

• Stress echocardiography is a standard diagnostic tool for suspected coronary artery disease (CAD).
• Patient variables like risk factors, medications, and anthropometrics are used to assess CAD probability.
• The relationship between stress echocardiography outcomes and these patient variables requires further investigation.

Purpose of the Study:

• To develop a machine learning framework for predicting significant CAD based on stress echocardiography results.
• To utilize patient anthropometrics, cardiovascular risk factors, and medication data as predictive variables.
• To enable clinical prioritization of patients with a high likelihood of CAD, optimizing clinician time and patient outcomes.

Main Methods:

• A four-stage machine learning framework was developed: feature extraction, preprocessing, feature selection, and classification.
• Mutual information was used for feature selection to identify informative variables for predicting stress echocardiography outcomes.
• Support Vector Machine (SVM) and random forest classifiers were employed, with data from 529 patients used for training and validation.

Main Results:

• Prior CAD diagnosis, sex, and specific medications (e.g., ACE inhibitors/ARBs) were identified as key predictors.
• SVM achieved the best sensitivity-specificity trade-off using three features, yielding 67.63% accuracy.
• For patients without prior CAD, two features (sex and ACE inhibitor/ARB use) achieved 70.32% accuracy.

Conclusions:

• Machine learning models can effectively predict stress echocardiography outcomes using a limited set of patient features.
• Key predictors include prior cardiac history, gender, and current medication.
• Further research with larger patient cohorts can enhance algorithm performance for improved patient selection and early intervention.