Implementation of a Cohort Retrieval System for Clinical Data Repositories Using the Observational Medical Outcomes...

Area of Science:

• Health Informatics
• Clinical Research Informatics
• Data Science in Healthcare

Background:

• Electronic health records (EHRs) are increasingly used for secondary data analysis in research and healthcare.
• Natural language processing (NLP) and information retrieval (IR) techniques are vital for extracting insights from unstructured clinical text.
• Efficiently retrieving specific patient cohorts from large EHR datasets is crucial for clinical research.

Purpose of the Study:

• To present the development and implementation of a novel cohort retrieval system named CREATE (Cohort Retrieval Enhanced by Analysis of Text from Electronic Health Records).
• To demonstrate CREATE's capability to execute complex textual cohort selection queries using both structured and unstructured EHR data.
• To improve the performance and accuracy of patient cohort identification.

Main Methods:

• CREATE integrates structured queries with IR techniques applied to NLP-processed results from EHRs.
• The system utilizes the Observational Medical Outcomes Partnership Common Data Model for enhanced portability.
• A hierarchical index was developed to support efficient searching of common data model concepts within unstructured text.

Main Results:

• A case study involving 5 cohort identification queries evaluated CREATE's performance.
• CREATE achieved a mean precision at 5 of 0.90 at both patient and document levels.
• This significantly outperformed systems relying solely on structured data (mean precision at 5 of 0.54) or unstructured text (mean precision at 5 of 0.74).

Conclusions:

• The CREATE system effectively integrates structured and unstructured EHR data for improved cohort retrieval.
• Evaluation on Mayo Clinic Biobank data confirmed CREATE's superior performance over single-modality retrieval systems for complex queries.
• This approach enhances the feasibility of large-scale clinical research by enabling more precise patient cohort identification.