Exploring the Application of the Observational Medical Outcomes Partnership Common Data Model to Multi-site Stroke...

Area of Science:

• Rehabilitation research
• Data interoperability
• Artificial intelligence in healthcare

Background:

• Artificial intelligence and machine learning (AI/ML) can enhance precision rehabilitation.
• Large volumes of rehabilitation data exist but lack interoperability for aggregation.
• Common Data Models (CDMs) like OMOP improve healthcare data interoperability.

Purpose of the Study:

• Evaluate the Observational Medical Outcomes Partnership (OMOP) Common Data Model (CDM) coverage for research-generated rehabilitation data.
• Assess OMOP CDM's suitability for aggregating data from the Enhancing NeuroImaging Genetics through Meta-Analysis Stroke Recovery (ENIGMA-SR) database.
• Identify challenges in applying OMOP CDM to harmonized, multi-site rehabilitation research data.

Main Methods:

• Two independent raters mapped ENIGMA-SR variables (demographics, medical history, assessments) to OMOP standard concepts.
• Assessed initial rater agreement using Gwet's agreement coefficient (AC).
• Analyzed OMOP inclusion, granularity of complex assessments, and mapped concept characteristics after reconciling differences.

Main Results:

• Good initial agreement for OMOP inclusion and assessment concept mapping (Gwet's AC: 0.79-0.89).
• Higher OMOP inclusion for demographics/medical history (84.8%) compared to rehabilitation assessments (58.9%).
• Limited OMOP coverage for assessment subscales/items (9.4%-19.2%) and frequent mapping to multiple OMOP domains.

Conclusions:

• OMOP CDM offers good demographic data coverage but moderate and limited coverage for top-level and granular rehabilitation assessments, respectively.
• Uneven OMOP coverage and mapping variability challenge the aggregation of clinical and research rehabilitation data for AI/ML applications.
• Enhancing OMOP's rehabilitation concept catalogue, creating cross-walks to research standards, and adapting computational tools are crucial for improved data aggregation.