Assessment of laboratory data quality in real-world data

Background: Increasingly, real-world data (RWD), especially from electronic health records (EHR) databases, includes laboratory data. These laboratory data play a key role in epidemiology research. However, the quality of laboratory data is often of concern and requires thorough evaluation.

Objectives: To develop a laboratory data quality assessment framework and associated algorithms and evaluate laboratory data quality in five RWD databases in the Observational Medical Outcomes Partnership (OMOP) common data model (CDM).

Methods: We first extracted laboratory data-related quality assessment checks from two RWD quality assessment resources (i.e., the PCORnet Data Characterization framework derived from the FDA’s Sentinel Data Quality Review and Characterization Programs and the Achilles and Data Quality Dashboard (DQD) framework from the OHDSI network) and then conducted a scoping literature review to identify RWD studies that carried out laboratory data quality assessments. The scoping review followed the PRISMA guideline, and search terms in PubMed include the domains of “real-world data”, “laboratory data” and “quality assessment”. We summarized the identified laboratory data quality assessment algorithms and further carried out quality assessments using the identified algorithms on selected lab tests of interest, e.g., complete blook count, oncology related blood tests, HPV tests, HbA1c, liver enzymes and lipid, using five commonly used RWD databases.

Results: We identified 104 laboratory data quality assessment algorithms from the two RWD quality assessment resources and the literature search, where we screened 144 published studies and eventually included 23 studies for full-text algorithm extraction. The domains laboratory data quality assessed include atemporal/temporal plausibility, consistency, completeness, conformance, information loss and degradation, and timeliness. After removing duplicated algorithms and expanding some for each target laboratory test of interest, we developed 216 laboratory data quality assessment checks. Of these, 55 (25%) relate to conformance, 50 (23%) to completeness and 111 (51%) to plausibility. The five RWD databases exhibited varied laboratory data quality issues, with common problems include data model mapping failures and a high percentage of null values of selected essential attributes, such as source value, unit, and ranges.

Conclusions: The quality of laboratory data across RWD databases varies and is suboptimal. Both general and study-specific laboratory data quality assessments can aid in identifying issues with lab data for downstream study analysis and should be employed to ensure continuous laboratory data quality improvement.