Qualifying Missingness in Real-World Clinical Data for Secondary Use.

Area of Science:

• Data Science
• Health Informatics
• Biostatistics

Background:

• Clinical data warehouses (CDWs) frequently encounter missing data due to unsystematic collection practices.
• Understanding the characteristics of missing data is crucial for assessing data quality and interpreting real-world data.
• Existing methods for qualifying clinical data require systematic evaluation for their effectiveness.

Purpose of the Study:

• To identify and evaluate methods for characterizing missing data in clinical data warehouses.
• To develop and implement an automated pipeline for extracting data quality descriptors.
• To assess the utility of these descriptors in reflecting different missing data scenarios.

Main Methods:

• A scoping review identified relevant data qualification methods for clinical data.
• An automated pipeline was developed to extract data quality descriptors.
• Four distinct missing data scenarios were simulated from a complete dataset for evaluation.
• Statistical tests (Friedman, Cochran's Q) were used to analyze descriptor variability across scenarios.

Main Results:

• 230 descriptors were extracted per incomplete dataset.
• 37.4% of extracted descriptors showed statistically significant differences across simulated scenarios after correction.
• The most effective descriptors accurately mirrored the underlying structures of the simulated missing data patterns.

Conclusions:

• Descriptor-based qualification provides interpretable insights into data quality within clinical data warehouses.
• This approach aids in understanding the nature and extent of missingness in real-world clinical data.
• Automated extraction and analysis of these descriptors enhance the assessment of CDW data integrity.