Statistical methods to harmonize electronic health record data across healthcare systems: case study and lessons...

Area of Science:

• Health Informatics
• Biostatistics
• Data Science

Background:

• Electronic Health Record (EHR) data facilitates multi-site research but often contains coding discrepancies between healthcare systems.
• Even with common data models, variations in coding practices persist, impacting data consistency.

Purpose of the Study:

• To present statistical methods for identifying and mitigating coding discrepancies in EHR data.
• To demonstrate these methods using real-world data from two healthcare systems.

Main Methods:

• Characterization of coding pattern differences between healthcare systems.
• Computation of a code mapping matrix to harmonize data.
• Application of methods to data from two FDA Sentinel data partners.

Main Results:

• Significant heterogeneity in coded EHR data was observed, even within a common data model.
• The developed data harmonization approaches effectively mitigate coding discrepancies.
• The study highlights the importance of data harmonization for downstream analyses.

Conclusions:

• Data harmonization is crucial for promoting semantic interoperability and enhancing data integration in multi-site EHR studies.
• The proposed methods improve the integrity of study conclusions by ensuring data quality.
• These approaches provide a foundational step for reliable comparative effectiveness research.