Efficient Record Linkage for Healthcare Data
Author Information
Author(s): Finney John M, Walker A Sarah, Peto Tim EA, Wyllie David H
Primary Institution: NIHR Biomedical Research Centre, John Radcliffe Hospital, Oxford, UK
Hypothesis
Can a two-step record linkage algorithm effectively detect identifier errors in large healthcare datasets?
Conclusion
The proposed method efficiently links over 250 million health records, achieving high accuracy in identifying individual patients.
Supporting Evidence
- The algorithm successfully clustered 3.6 million records with 99.8% accuracy in identifying individual patients.
- Linkage was completed in about 30 minutes, demonstrating high computational efficiency.
- Logistic regression models detected over 97% of bad clusters with a low false positive rate.
Takeaway
This study shows a fast way to connect health records, helping to make sure that doctors have the right information about each patient.
Methodology
A two-step algorithm was used to identify high-cardinality identifiers for initial exact matching, followed by a graph-based method to detect and resolve identifier errors.
Potential Biases
Potential for mislinkage if identifiers are incorrectly entered, leading to identity collisions.
Limitations
The algorithm requires at least one exact identifier for each record, which may not be available in all datasets.
Participant Demographics
Data from five sources within a large UK hospital group, covering about 1% of the UK population.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website