Assessing the harmonization of structured electronic health record data to reference terminologies and data...

Area of Science:

• Health Informatics
• Data Science
• Clinical Data Management

Background:

• Structured electronic health record (EHR) data is crucial for clinical research.
• Harmonization to reference terminologies and assessing data completeness are vital for data quality.
• Variability in EHR data across different data partners (DPs) necessitates rigorous evaluation.

Purpose of the Study:

• To assess the harmonization of laboratory results and medications in EHR data to reference terminologies.
• To characterize the severity of harmonization issues and identify data completeness problems.
• To evaluate the impact of data provenance on completeness assessments.

Main Methods:

• Distributed harmonization and completeness queries to three Data Partners (DPs).
• Examined top 200 laboratory results and medications for harmonization outliers and summary statistics.
• Analyzed data completeness for four conditions of interest, stratified by time, care setting, and provenance.

Main Results:

• Median code associations generally met expectations, but DP-specific harmonization outliers were identified.
• Significant differences in patient concept percentages were observed based on data provenance.
• Harmonization queries identified mapping errors, overly specific codes, and 'null' code issues.

Conclusions:

• Harmonization errors, though not always widespread, exist in clinical data models (CDMs).
• Multiple data provenance types provide more robust completeness results than single sources.
• Provenance information is key to identifying EHR data completeness issues if represented in the CDM.