Harmonizing Logical Observation Identifiers Names and Codes (LOINC) Codes and Units in Real-World Oncology Data:...

Area of Science:

• Health Informatics
• Biomedical Data Science
• Clinical Research Data Management

Background:

• Multisource electronic health record (EHR) and claims data offer research opportunities but face challenges in semantic interoperability.
• Accurate mapping of laboratory tests to Logical Observation Identifiers Names and Codes (LOINC) is difficult, with 6-19% of tests unmappable.
• Existing systems struggle with absent, null, or incorrect source data strings, necessitating a scalable solution for data integrity.

Purpose of the Study:

• To present a universally applicable framework for identifying and correcting errors in quantitative laboratory results coded to LOINC.
• To standardize units of measure without relying on raw source data strings.
• To enhance the accuracy, conformance, consistency, and completeness of laboratory data.

Main Methods:

• A 2-step framework: 1) Correct LOINC codes using the associated unit of measure. 2) Adjust or populate units to match preferred units for the LOINC code.
• Quantitative results are validated against predefined acceptable ranges.
• The framework was applied to ~10 million cancer patient records from the ConcertAI database and 3 EHR subsets.

Main Results:

• The framework processing of 6.34 billion records demonstrated significant improvements in LOINC code-unit conformance and unit completeness.
• Correctly assigned units increased from 73.1% to 99.7% in the main dataset.
• Unit completeness improved from 92.7% to 99.8% in the main dataset, with similar gains across EHR subsets.

Conclusions:

• Laboratory data quality is critical for oncology systems, impacting therapy selection, monitoring, and disease assessment.
• The proposed solution offers a system-agnostic, scalable normalization process to address key laboratory data quality gaps.
• This framework represents a novel approach to enhancing multisource laboratory data quality across multiple dimensions.