Build and Query Indexes of Clinical Documents with Easy-to-Reuse Pipelines

Area of Science:

• Biomedical Informatics
• Natural Language Processing

Background:

• Electronic Health Records (EHRs) contain valuable unstructured clinical text crucial for research.
• Current methods for extracting information from clinical text are often one-off and study-specific, limiting data reuse.
• This hinders the generation of real-world evidence and comprehensive patient phenotyping.

Purpose of the Study:

• To present medkit Seshat, an open-source Python pipeline designed for efficient processing and indexing of unstructured clinical text.
• To enable the secondary use and adaptation of extracted clinical data for diverse research objectives, particularly phenotyping.
• To provide a flexible web UI for exploring and utilizing the built indexes.

Main Methods:

• The pipeline ingests free text from Electronic Health Records.
• It recognizes and normalizes relevant clinical entities using OMOP vocabularies.
• An index is built for searching by concept or document, supported by a web UI for analysis and export.

Main Results:

• Successfully developed and demonstrated medkit Seshat, an open-source pipeline for clinical text analysis.
• Created an indexed repository of extracted clinical information, searchable by concept and document.
• Developed a web UI to showcase the utility of the indexed data for search and analysis.

Conclusions:

• medkit Seshat facilitates the efficient extraction and reuse of information from unstructured clinical text.
• The pipeline supports secondary data use for phenotyping and real-world evidence generation.
• The open-source nature and flexible UI promote adaptation and wider application in biomedical research.