Informative missingness: What can we learn from patterns in missing laboratory data in the electronic health record?

Area of Science:

• Computational epidemiology
• Health informatics
• Clinical data science

Background:

• Missing laboratory test results in electronic health records are understudied but can reflect disease progression and clinical concerns.
• Patterns of missingness may offer insights into patient conditions and healthcare system data quality.

Purpose of the Study:

• To identify informative patterns of missing laboratory data among COVID-19 inpatients across multiple international healthcare sites.
• To explore the relationship between missingness patterns, clinical outcomes, and hospital treatment capacity.

Main Methods:

• Analysis of demographic, diagnosis, and laboratory data for 69,939 COVID-19 patients across 15 healthcare sites in three countries.
• Investigated missingness patterns, stratification by demographics, temporal trends, lab correlations, and clustering based on missingness.

Main Results:

• Identified data mapping and collection nuances at seven of 15 sites.
• Temporal trends in missingness may help identify severe COVID-19 cases.
• Missingness patterns revealed relationships between laboratory tests reflecting clinical behaviors.

Conclusions:

• Missing laboratory data patterns can be informative for analyses, potentially indicating changes in patient condition or clinical outcomes.
• Computational approaches highlight heterogeneity in COVID-19 data across sites and time, suggesting missing data should be considered valuable.
• Findings aid researchers in identifying suitable sites for specific COVID-19 research questions and understanding data quality.