An information extraction framework for cohort identification using electronic health records

Area of Science:

• Clinical Informatics
• Natural Language Processing
• Biomedical Data Science

Background:

• Information Extraction (IE) is crucial for clinical applications, enabling automated systems and secondary use of Electronic Health Records (EHRs).
• Developing high-performance IE systems is challenging due to the complexity and variability of clinical language.
• Existing IE methods often struggle with the nuances of clinical text.

Purpose of the Study:

• To present a novel knowledge-driven Information Extraction (IE) framework for cohort identification using EHR data.
• To demonstrate a flexible framework adaptable by subject matter experts for specific clinical information needs.
• To enhance the utility of EHRs for clinical and translational research through improved data extraction.

Main Methods:

• Developed a knowledge-driven IE framework under the Unstructured Information Management Architecture (UIMA).
• Employed expert knowledge engineering to define externalized knowledge resources for information extraction.
• Focused on cohort identification as a primary application within the EHR domain.

Main Results:

• The framework facilitates the construction of IE systems tailored to specific clinical information requirements.
• Enables subject matter experts to contribute their knowledge directly to the IE system development.
• Demonstrates a viable approach for extracting structured information from unstructured EHR text for research.

Conclusions:

• The proposed knowledge-driven IE framework offers a robust solution for cohort identification in EHRs.
• This approach empowers domain experts to build effective IE systems, overcoming language complexity challenges.
• The framework supports the secondary use of EHRs, advancing clinical and translational research capabilities.