A guide to evaluating linkage quality for the analysis of linked data

Area of Science:

• Health Informatics
• Epidemiology
• Biostatistics

Background:

• Linked datasets are vital for epidemiological and clinical research, but linkage errors can introduce bias.
• Data security often necessitates third-party linkage, hindering researchers' ability to assess data quality.
• There is a lack of clear guidance on quantifying the impact of linkage errors on study outcomes.

Purpose of the Study:

• To provide guidance on evaluating linkage quality for both data providers and researchers.
• To demonstrate methods for assessing the impact of linkage errors in linked datasets.
• To enhance the transparency and quality of research utilizing linked data.

Main Methods:

• Quantifying linkage error by applying algorithms to gold standard data subsets.
• Identifying potential bias by comparing linked and unlinked data characteristics.
• Assessing the sensitivity of research findings to variations in the linkage process.

Main Results:

• Demonstrated three practical approaches for evaluating linkage quality using a maternal and baby hospital record dataset.
• Showcased how to quantify linkage error and identify potential biases.
• Illustrated methods to test the robustness of results against linkage procedure changes.

Conclusions:

• The proposed methods improve understanding of linkage error impact in linked data.
• Researchers can select the most appropriate linkage procedures for specific analyses.
• Evaluating linkage quality enhances the reliability and transparency of epidemiological and clinical research.