Taming Big Data: An Information Extraction Strategy for Large Clinical Text Corpora

Area of Science:

• Clinical Informatics
• Natural Language Processing
• Health Informatics

Background:

• Clinical text in electronic medical records contains unevenly distributed concepts.
• Efficiently accessing relevant clinical information is crucial for research and clinical decision support.

Purpose of the Study:

• To identify filtering techniques for selecting 'high-yield' clinical documents.
• To increase the efficacy and throughput of information extraction from clinical text.

Main Methods:

• Utilized two large corpora of clinical text.
• Applied filtering techniques to identify 'high-yield' document sets in distinct clinical domains.

Main Results:

• Identified 'high-yield' document sets for homelessness and indwelling urinary catheters.
• For homelessness, high-yield documents included homeless program and social work notes.
• For urinary catheters, high-yield documents were primarily from hospitalized patients' nursing notes.

Conclusions:

• Filtering techniques can effectively identify relevant clinical documents.
• Customizable information extraction pipelines can be refined using these methods.
• This approach facilitates the extraction of key concepts for clinical decision support and other applications.