Machine learning for risk prediction of acute coronary syndrome

Area of Science:

• Cardiology
• Data Science
• Machine Learning

Background:

• Acute coronary syndrome (ACS) presents a significant healthcare burden in the US, with high hospitalization rates and costs.
• Accurate diagnosis of ACS is challenging due to symptom overlap, leading to costly over- or under-treatment.
• Current risk stratification tools often lack sufficient accuracy and rely on curated data, necessitating improved predictive models.

Purpose of the Study:

• To develop and validate machine learning models for improved risk prediction in patients with suspected ACS.
• To compare the performance of random forest and elastic net models against existing clinical risk scores.

Main Methods:

• Utilized a dataset of 20,078 deidentified patient records with missing and noisy values.
• Developed predictive models using random forest and elastic net algorithms.
• Compared model performance using Area Under the Curve (AUC) metrics against TIMI and GRACE scores.

Main Results:

• The random forest model achieved an AUC of 0.848, outperforming elastic net (0.818), ridge regression (0.810), TIMI score (0.745), and GRACE score (0.623).
• Random forest demonstrated superior predictive accuracy compared to traditional scoring systems.
• The models effectively handled noisy and sparse data, achieving competitive performance.

Conclusions:

• Machine learning models, particularly random forest, offer enhanced accuracy for ACS risk prediction.
• These models show potential for clinical application, especially when dealing with real-world, imperfect datasets.
• The findings suggest a viable alternative to existing prognostic indices for early risk stratification in suspected ACS cases.