Unlocking echocardiogram measurements for heart disease research through natural language processing

Area of Science:

• Medical Informatics
• Natural Language Processing
• Cardiology

Background:

• Cardiovascular disease mechanisms in HIV patients require analysis of echocardiogram reports.
• A large, longitudinal, multi-center study is needed.

Purpose of the Study:

• To develop a natural language processing (NLP) system for extracting heart function measurements from clinical data.
• To enable large-scale retrospective analysis of clinical data for heart failure research.

Main Methods:

• Developed an NLP system using dictionary lookup, rules, and patterns to extract measurement-value pairs.
• Utilized curated semantic bootstrapping for custom dictionary creation and a novel disambiguation method.
• Built a scalable framework for processing large datasets.

Main Results:

• The system achieved high F-scores (0.844-0.877) and precision (0.936-0.982) across general clinic, echocardiogram, and radiology reports.
• Left ventricular ejection fraction (LVEF) was the most frequently extracted measurement with high precision (0.968-1.0).

Conclusions:

• Demonstrated the feasibility and effectiveness of large-scale information extraction from clinical data using NLP.
• The system facilitates addressing new clinical questions in heart failure through retrospective data analysis.