A Novel Approach to Dual Feature Selection of Atrial Fibrillation Based on HC-MFS

Area of Science:

• Cardiology
• Medical Informatics
• Public Health

Background:

• Atrial fibrillation (AF) is a prevalent arrhythmia with complex risk factors.
• Identifying specific risk factors in diverse populations is crucial for targeted prevention and management.
• Previous studies have explored various indicators, but a comprehensive approach integrating clinical and environmental factors is needed.

Purpose of the Study:

• To identify and analyze risk factors for atrial fibrillation (AF) in Shanghai's Chongming District.
• To develop and evaluate a novel dual feature-selection methodology (HC-MFS) for improved AF risk prediction.
• To assess the utility of machine learning models in diagnosing AF predisposition.

Main Methods:

• Data from 678 patients treated for AF between 2020-2023 in Chongming District, Shanghai, were analyzed.
• A novel hierarchical clustering with Fisher scores (HC-MFS) feature selection method was developed and benchmarked.
• Five classification models were trained and evaluated using the HC-MFS approach, with performance validated on a test set.

Main Results:

• The HC-MFS method achieved superior performance, with the highest accuracy (0.9118) and lowest root mean square error (0.2970) in AF classification.
• Seasonal variations were identified as a primary risk factor for AF (correlation ranks: pr=0.31, FS=0.11, DCFS=0.33).
• Inflammatory markers (C-reactive protein) and lipid profiles (LDL, total cholesterol) along with platelet count were significant indirect risk factors for AF.

Conclusions:

• The novel HC-MFS feature selection method significantly enhances the accuracy of AF risk prediction models.
• Machine learning models, informed by pathological and climatic factors, can effectively aid clinicians in diagnosing AF predisposition.
• Seasonal variations and inflammatory markers are critical, often overlooked, risk factors for AF in the studied population.