False Discovery Estimation in Record Linkage.

Area of Science:

• Data Science
• Biostatistics
• Epidemiology

Background:

• Integrating diverse datasets offers research advantages but lacks unique identifiers due to privacy and varied collection methods.
• Record linkage (RL) algorithms probabilistically link records using identifying variables, but imperfect matches necessitate assessing false discoveries.
• The false discovery proportion (FDP) is critical for validating linked data reliability in subsequent analyses.

Purpose of the Study:

• To introduce a novel method for estimating the FDP in RL for two overlapping datasets.
• To provide a reliable approach for assessing and improving the quality of linked data across various RL techniques and settings.
• To highlight the importance of accounting for linkage errors in healthcare record analysis.

Main Methods:

• A novel FDP estimation method using synthetic data generated from empirical distributions alongside real data.
• Synthetic records, unable to link with real entities, quantify falsely linked pairs.
• The method is applicable to all RL techniques, especially in complex scenarios with poorly discriminative variables.

Main Results:

• The proposed method effectively estimates FDP in RL, enabling assessment and improvement of linked data reliability.
• Evaluated performance using established RL algorithms and benchmark datasets.
• Successfully applied to link siblings in the Netherlands Perinatal Registry, confirming its practical utility.

Conclusions:

• The developed method provides a robust way to estimate FDP in RL, enhancing data reliability.
• Accurate FDP estimation is vital for trustworthy research outcomes derived from linked datasets.
• Accounting for linkage errors is essential, particularly in sensitive healthcare data studies like mother-child dynamics.