Machine Learning Methods for Identifying Atrial Fibrillation Cases and Their Predictors in Patients With Hypertrophic...

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning

Background:

• Hypertrophic cardiomyopathy (HCM) is linked to a high incidence of atrial fibrillation (AF) and stroke risk, irrespective of CHA2DS2-VASc scores.
• Understanding the pathophysiology of AF in HCM is crucial for risk stratification and management.
• Electronic health records (EHR) offer a valuable resource for data-driven insights into AF in HCM.

Purpose of the Study:

• To develop and apply a machine learning-based method for identifying AF cases in HCM patients.
• To identify key clinical and imaging features associated with AF in the HCM population.
• To improve the understanding of AF pathophysiology within the context of HCM.

Main Methods:

• A retrospective study utilizing EHR data from HCM patients.
• Patients were classified into AF cases (n=191) and No-AF cases (n=640) based on documented AF status.
• Feature selection involved 93 clinical variables, identified using t-tests and information gain; data imbalance was managed with over/undersampling.

Main Results:

• Identified 18 highly informative variables (11 positive, 7 negative correlations) for AF in HCM.
• An ensemble classifier (logistic regression and naive Bayes) achieved high performance (sensitivity=0.74, specificity=0.70, C-index=0.80) after addressing data imbalance.
• The model effectively distinguished AF from No-AF cases, demonstrating the utility of machine learning in this context.

Conclusions:

• The developed HCM-AF-Risk Model is the first machine learning approach for AF identification in HCM.
• The model exhibits robust performance and effectively handles data imbalance.
• Findings suggest a correlation between AF and a more severe cardiac phenotype in HCM patients.