Empirical aspects of record linkage across multiple data sets using statistical linkage keys: the experience of the PIAC cohort study
- Equal contributors
1 Faculty of Health, University of Canberra, Canberra, Australian Capital Territory, Australia
2 Australian Institute of Health and Welfare, Canberra, Australian Capital Territory, Australia
3 School of Population Health, University of Queensland, Brisbane, Queensland, Australia
4 Lincoln Centre for Research on Ageing, Australian Institute for Primary Care, La Trobe University, Melbourne, Victoria, Australia
BMC Health Services Research 2010, 10:41 doi:10.1186/1472-6963-10-41Published: 18 February 2010
In Australia, many community service program data collections developed over the last decade, including several for aged care programs, contain a statistical linkage key (SLK) to enable derivation of client-level data. In addition, a common SLK is now used in many collections to facilitate the statistical examination of cross-program use. In 2005, the Pathways in Aged Care (PIAC) cohort study was funded to create a linked aged care database using the common SLK to enable analysis of pathways through aged care services.
Linkage using an SLK is commonly deterministic. The purpose of this paper is to describe an extended deterministic record linkage strategy for situations where there is a general person identifier (e.g. an SLK) and several additional variables suitable for data linkage. This approach can allow for variation in client information recorded on different databases.
A stepwise deterministic record linkage algorithm was developed to link datasets using an SLK and several other variables. Three measures of likely match accuracy were used: the discriminating power of match key values, an estimated false match rate, and an estimated step-specific trade-off between true and false matches. The method was validated through examining link properties and clerical review of three samples of links.
The deterministic algorithm resulted in up to an 11% increase in links compared with simple deterministic matching using an SLK. The links identified are of high quality: validation samples showed that less than 0.5% of links were false positives, and very few matches were made using non-unique match information (0.01%). There was a high degree of consistency in the characteristics of linked events.
The linkage strategy described in this paper has allowed the linking of multiple large aged care service datasets using a statistical linkage key while allowing for variation in its reporting. More widely, our deterministic algorithm, based on statistical properties of match keys, is a useful addition to the linker's toolkit. In particular, it may prove attractive when insufficient data are available for clerical review or follow-up, and the researcher has fewer options in relation to probabilistic linkage.