Optimal strategy for linkage of datasets containing a statistical linkage key and datasets with full personal identifiers

Abstract

BackgroundLinkage of aged care and hospitalisation data provides valuable information on patterns of health service utilisation among aged care service recipients. Many aged care datasets in Australia contain a Statistical Linkage Key (SLK-581) instead of full personal identifiers. We linked hospital and death records using a full probabilistic strategy, the SLK-581, and three combined strategies; and compared results for each strategy.MethodsLinkage of Admitted Patient Data for 2000–01 to 2008–09 and Registry of Births, Deaths and Marriages death registration data for 2008–09 for New South Wales, Australia, was carried out using probabilistic methods and compared to links created using four strategies incorporating a SLK-581. The Basic SLK-581 strategy used the SLK-581 alone. The Most Recent SLK-581, Most Frequent SLK-581, and Any Match SLK-581 strategies leveraged probabilistic links between hospital records drawn from the Centre for Health Record Linkage Master Linkage Key. Rates of hospitalisations among people who died were calculated for each strategy and a range of health conditions.ResultsCompared to full probabilistic linkage, the basic SLK-581 strategy produced substantial rates of missed links that increased over the study period and produced underestimates of hospitalisation rates that varied by health condition. The Most Recent SLK-581, Most Frequent SLK-581, and Any Match SLK-581 strategies resulted in substantially lower rates of underestimation than the Basic SLK-581. The Any Match SLK-581 strategy gave results closest to full probabilistic linkage.ConclusionsHospitalisation rates prior to death are substantially underestimated by linkage using a SLK-581 alone. Linkage rates can be increased by combining deterministic methods with probabilistically created links across hospital records.

Linkage of Australian national registry data using a statistical linkage key

“…These may be improved by the addition of further information, for example area of residence or postcode [ 14 ]. Our linkage had a higher rate of success compared to other studies [ 8 , 12 ]. This may be due to a much smaller cohort, the use of a single centre’s data, and the high quality of the collected data [ 15 , 16 ].…”

“…This is a 14-character code comprising the second, third and fifth characters of the family name, the second and third letters of the given name, the date of birth (DDMMYYYY) and sex [ 7 ]. This has shown to provide successful linkage in some datasets (for example a large residential aged care dataset), while its use has been less successful in others, particularly when a high rate of missing name data is present [ 8 , 9 ]. SLK-581 has recently been added to the Australian and New Zealand Intensive Care Society Adult Patient Database (ANZICS-APD).…”

Limited privacy protection and poor sensitivity

“…Several previous studies have provided absolute measures of linkage quality for both SLKs and Bloom filter–based record linkage; the SLK method, in particular, has received numerous evaluations due to its wide implementation. Privacy-preserving linkage using Bloom filters has been shown to achieve equal results to those found with linkage on un-encoded data (Randall et al, 2013), while poorer linkage quality has been found when SLKs have been tested against full probabilistic linkage (Bass and Garfield, 2002; Karmel, 2005; Taylor et al, 2014).…”

“…Linkage to death data also showed a lower recall (88.4%) when compared with that achieved through linkage with full named information, with poorer recall affecting research results (Karmel, 2005). The SLK-581 has also been shown to produce substantial rates of missed links that increase over time, resulting in underestimates of hospitalisation rates which vary by health condition (Taylor et al, 2014). Matching rules involving SLKs typically only identify matches where a large proportion of attributes are identical; as a result, low rates of false positive links have been observed (Ryan et al, 1999).…”

Centre for Health Record Linkage

“…The comprehensive linkage system contains a centralised repository of linked personal identifiers. Enduring links are stored in perpetuity in the CHeReL Master Linkage Key so that records do not need to be repeatedly matched for different studies [7], timeframes for accessing linked data are reduced [10] and linkage quality for new datasets with limited identifiers is improved through leveraging pre-linked arrays of personal identifiers for an individual [11,12].…”