JMIR Publications

ABSTRACT

Background:

The growing availability of multi-sourced real-world data from EHRs and claims offers immense potential for advancing research, drug discovery, and clinical decision support. While data standards like LOINC ensure interoperability for lab data, properly assigning these concepts to strings or mapping to them from local coding systems presents a challenge. Studies show that 6% to 19% of lab tests cannot be accurately mapped to LOINC. Although systems have been proposed to address these errors, they often depend on source data strings and other input features that may be absent, null, or incorrect. This underscores the need for a solution that is adaptable to multisource lab data and capable of correcting LOINC assignments, improving unit consistency, and ensuring data integrity.

Objective:

This paper introduces a universally applicable framework that identifies and corrects observable errors in quantitative lab results that have been coded to LOINC for the observations and SNOMED for the unit of measure without relying on raw source data strings. The process seeks to improve accuracy, conformance, consistency, and completeness of lab data while maintaining complete provenance.

Methods:

A framework is proposed that involves a two-step process. First, the LOINC code will be corrected by using information contained in the associated unit of measure. Next, the unit will be adjusted or filled to conform to a preselected preferred unit for that LOINC. Both steps will use the quantitative result to inform if the transformation can proceed by comparison to a predetermined acceptable value range for that LOINC. The framework is executed using three knowledge artifacts that contain logic for both steps. The framework is then applied, and results are evaluated in datasets derived from the ConcertAI database, which covers approximately 10 million cancer patients. The analysis will look at improvements in LOINC-unit conformance and unit completeness. The framework and analysis are run on four datasets with independent LOINC coding extracted from the ConcertAI data pipeline. Datasets include the full ConcertAI dataset, and three subsets of source data transmitted to ConcertAI, grouped by data source or by EHR vendor, representing the top contributors by volume and diversity.

Results:

All four datasets were processed using the proposed framework. Application of the framework increased LOINC-unit conformance across all evaluated records from 73.1% to 99.7% in the ConcertAI dataset. A similar improvement was observed in the other three datasets. The framework raised the unit completion rate from 92.7% to 99.9% across all four datasets.

Conclusions:

Lab data quality is crucial in oncology systems for therapy selection, monitoring, and disease progression assessment. This proposed solution is a first-of-its-kind, system-agnostic, and scalable normalization process that addresses key gaps in lab data quality across multiple dimensions. Clinical Trial: Not applicable