Predicting atrial fibrillation and flutter using electronic health records

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Clinical Data Analysis

Background:

• Electronic Health Records (EHR) offer vast potential for predictive modeling of disease onset.
• Predicting the first episode of atrial fibrillation and/or atrial flutter (AFF) is crucial for timely intervention.

Purpose of the Study:

• To investigate the utility of free-text and coded data from EHRs for predicting AFF onset.
• To compare the performance of machine learning models using textual data, coded data, and their combination.

Main Methods:

• Trained and evaluated machine learning models (naïve Bayes, SVM, logistic regression, random forests) on EHR data.
• Utilized 10-fold cross-validation to assess model performance across various pre-onset time intervals (1, 3, 5, all years).
• Analyzed both free-text and coded data, individually and combined.

Main Results:

• Text-based datasets achieved an F-measure of 60.1% when used alone.
• Models utilizing coded data exclusively showed comparable performance to text-based models.
• Combining textual and coded data yielded performance similar to using each data type independently.

Conclusions:

• Free-text data from EHRs is a valuable resource for predicting disease onset, specifically AFF.
• Textual data can effectively complement coded data in machine learning models for disease prediction.
• The findings support the integration of diverse EHR data types for enhanced predictive accuracy.