ACCESS extracts electronic patient records from a national network of clinical and laboratory services in order to report on key indicators for sexually transmissible infections (STIs) and blood borne viruses (BBVs). The records include demographics, test outcomes and prescriptions for BBVs and STIs and are extracted without patient identifiers. Instead, extraction software, GRHANITE®, creates a series of linkage keys that are generated at the site using the available patient identifiers. These linkage keys are then extracted alongside the patient clinical records. The linkage keys generated by GRHANITE® are unique to individuals and are used to match patient records while maintaining patient privacy.
A possible limitation of sentinel surveillance programs, such as ACCESS, is the duplication introduced by patients presenting to multiple services for testing and/or treatment. This can introduce error and bias when trying to answer key population-level research questions or accurately describe population-level trends. For this reason, the GRHANITE Linkage Tool® is used in ACCESS to de-duplicate and improve data quality by linking patient records for a single individual within and between ACCESS services via their unique linkage keys.
To evaluate the patient record linkage within ACCESS we created two gold-standard datasets, where true match status could be identified, and conducted a sensitivity and specificity analysis comparing the observed linkage results using the GRHANITE Linkage Tool®.
Sensitivity and specificity of the linkage was 100% when applying the GRHANITE Linkage Tool® to the smaller gold standard dataset which had a high level of completeness of the data used to generate the linkage keys. In the larger gold standard dataset generated from matched pathology and clinical records with a lower level of data completeness, the sensitivity was over 94% and the estimated specificity was over 90%.
This evaluation suggests that the GRHANITE Linkage Tool® used in the production of ACCESS datasets is a strong and accurate tool for linking patient records and underpins the ability for ACCESS to be confidently used for public health surveillance.
This is a summary of Long Nguyen's first author paper "Evaluating privacy-preserving record linkage within a public health surveillance system that uses de-identified records". Published in the Journal of Medical Internet Research (JIMR) available pre print here