Extraction of left ventricular ejection fraction information from various types of clinical reports

Area of Science:

• Medical Informatics
• Natural Language Processing (NLP)
• Cardiology

Background:

• Congestive heart failure (CHF) management relies on quality measures to ensure guideline-concordant care.
• Accurate extraction of left ventricular ejection fraction (LVEF) from clinical notes is crucial for CHF quality assessment.
• Clinical notes vary in structure, posing challenges for automated information extraction.

Purpose of the Study:

• To identify and extract left ventricular ejection fraction (LVEF) data from diverse clinical note formats.
• To evaluate the effectiveness of NLP modules for LVEF information extraction in heart failure quality measurement.
• To assess the impact of training data size on NLP model performance for LVEF extraction.

Main Methods:

• Analyzed annotation differences across corpora, including Echocardiography, Radiology, and Text Integrated Utility (eXtensible Markup Language) notes.
• Developed and compared two sequence-tagging NLP modules for LVEF extraction, one utilizing prior extraction module predictions.
• Conducted experiments varying training data size to determine its effect on extraction accuracy.

Main Results:

• Information extraction accuracy is higher with less training data for highly structured reports.
• Combining predictions from existing LVEF extraction modules significantly improves accuracy for less structured reports.
• NLP modules demonstrate enhanced performance when dealing with varied vocabulary in clinical notes.

Conclusions:

• NLP-based LVEF extraction is a viable method for improving heart failure quality measurement.
• The developed NLP modules enhance the accuracy of LVEF data retrieval from diverse clinical documentation.
• Adaptable NLP strategies are essential for robust information extraction from heterogeneous clinical text.