Facilitating clinical research through automation: Combining optical character recognition with natural language...

Area of Science:

• Medical Informatics
• Clinical Research Data Management
• Natural Language Processing in Healthcare

Background:

• Performance status is vital for clinical research but often unstructured in electronic medical records.
• Scanned documents and image formats hinder data extraction and NLP application.
• Efficient retrieval of performance status data is challenging for research purposes.

Purpose of the Study:

• To develop and evaluate an optical character recognition (OCR) and natural language processing (NLP) pipeline for extracting performance status data.
• To improve the efficiency and accuracy of data abstraction from scanned electronic medical records.
• To reduce the time and effort required for clinical research data retrieval.

Main Methods:

• Utilized optical character recognition (OCR) software (ABBYY FineReader) to convert scanned medical record images into a searchable format.
• Applied natural language processing (NLP) software (Linguamatics i2e) to identify and extract performance status data elements.
• Evaluated the pipeline's accuracy and efficiency using a cohort of diffuse large B-cell lymphoma patients, comparing against manual abstraction.

Main Results:

• The OCR/NLP pipeline demonstrated high accuracy with less than 1% incorrect results (excellent precision, recall, and F score).
• Median review time for documents containing performance status data was reduced by one-third.
• Manual review time for documents without performance status information was reduced by 83% (18 minutes vs. 108 minutes).

Conclusions:

• The OCR/NLP pipeline significantly improves operational efficiency and reduces information retrieval time for clinical research.
• OCR effectively transforms scanned medical record images for NLP application, enabling highly accurate data abstraction.
• This pipeline facilitates research by enabling focused data review, eliminating unnecessary chart review, and freeing up time for other critical data elements.