Interpretable Machine Learning Approaches for Identification of Acute Aortic Dissection in Chest Pain Patients

Area of Science:

• Biomedical Informatics
• Cardiovascular Medicine
• Machine Learning in Healthcare

Background:

• Accurate and timely diagnosis of acute aortic dissection (AAD) is critical for patient outcomes.
• Differentiating AAD from other causes of sudden chest pain, such as acute myocardial infarction (AMI), acute pulmonary embolism (APE), and abdominal aortic aneurysm (AAA), presents a clinical challenge.

Purpose of the Study:

• To develop and validate interpretable machine learning models for identifying patients with AAD.
• To differentiate AAD from other critical conditions causing chest pain using routine laboratory biomarkers and clinical data.

Main Methods:

• Utilized a cohort of 832 patients (515 with AAD) with data split into training and testing sets.
• Employed LASSO regression for variable selection and nine machine learning algorithms for model development.
• Applied the SHAP method for feature importance analysis and model interpretability.

Main Results:

• LASSO regression identified eight significant variables, including N.L and age, for AAD prediction.
• Machine learning models achieved an Area Under the Curve (AUC) between 0.72 and 0.77 on the test set.
• The Random Forest model demonstrated strong performance in sensitivity, specificity, and F1 Score.

Conclusions:

• An interpretable machine learning model can effectively aid in differentiating AAD from other causes of chest pain.
• Prioritization of key features through interpretable AI supports prompt diagnosis and treatment of AAD.
• This approach holds significant potential for improving clinical decision-making in emergency settings.