Automatic Methods to Extract New York Heart Association Classification from Clinical Notes.

Area of Science:

• Biomedical Informatics
• Clinical Natural Language Processing
• Cardiology

Background:

• Cardiac Resynchronization Therapy (CRT) is a key treatment for heart failure.
• New York Heart Association (NYHA) classification assesses patient response to CRT.
• NYHA classification is often missing from structured Electronic Health Record (EHR) data, residing instead in unstructured clinical notes.

Purpose of the Study:

• To investigate the use of Natural Language Processing (NLP) methods for identifying NYHA classification from clinical notes.
• To enable consistent, longitudinal tracking of heart failure progression and CRT response assessment.

Main Methods:

• Collected 6,174 clinical notes.
• Matched notes with hospital-specific NYHA class diagnosis codes.
• Evaluated machine-learning (ML) methods, including support vector machine (SVM) with n-gram features, against a rule-based method.

Main Results:

• Machine-learning methods showed comparable performance to the rule-based method.
• The best performing method was Support Vector Machine (SVM) with n-gram features, achieving a 93% F-measure.
• NLP effectively extracts NYHA classification from unstructured clinical text.

Conclusions:

• NLP methods, particularly SVM with n-gram features, are effective for identifying NYHA classification from clinical notes.
• This approach can improve the assessment of heart failure progression and CRT effectiveness using EHR data.
• Further validation of these findings is warranted.