BACKGROUND: This study compares registry and non-registry approaches to linking 2006 Census of Population data for Manitoba and Ontario to hospital data from the Discharge Abstract Database (DAD). DATA AND METHODS: Using a probabilistic linkage, the registry approach linked the census data to provincial health insurance registries, followed by a deterministic linkage to the DAD based on health insurance number (HIN). The non-registry approach used hierarchical deterministic exact matching based on three variables common to both files to link census data to the DAD. The approaches were compared in terms of linkage and coverage rates, sensitivity and specificity, and consistency of HINs on the linked records. RESULTS: Results of the registry and non-registry linkage approaches were similar. In Manitoba, 7% and 6% of census long-form respondents linked to the DAD with the registry and non-registry linkage approaches, respectively; in Ontario, the linkage rate was 5% for both approaches. With the registry approach, the linked census-DAD data represented 84% (weighted) of hospital admissions in the 2006/2007 DAD in both provinces, compared with 82% in Manitoba and Ontario with the non-registry approach. INTERPRETATION: In the absence of access to provincial health insurance registries with which census data can be linked, a non-registry approach can be used to create a research-quality dataset.
BACKGROUND: This study compares registry and non-registry approaches to linking 2006 Census of Population data for Manitoba and Ontario to hospital data from the Discharge Abstract Database (DAD). DATA AND METHODS: Using a probabilistic linkage, the registry approach linked the census data to provincial health insurance registries, followed by a deterministic linkage to the DAD based on health insurance number (HIN). The non-registry approach used hierarchical deterministic exact matching based on three variables common to both files to link census data to the DAD. The approaches were compared in terms of linkage and coverage rates, sensitivity and specificity, and consistency of HINs on the linked records. RESULTS: Results of the registry and non-registry linkage approaches were similar. In Manitoba, 7% and 6% of census long-form respondents linked to the DAD with the registry and non-registry linkage approaches, respectively; in Ontario, the linkage rate was 5% for both approaches. With the registry approach, the linked census-DAD data represented 84% (weighted) of hospital admissions in the 2006/2007 DAD in both provinces, compared with 82% in Manitoba and Ontario with the non-registry approach. INTERPRETATION: In the absence of access to provincial health insurance registries with which census data can be linked, a non-registry approach can be used to create a research-quality dataset.
Keywords:
Coverage; data collection; data linkage; databases; hospital records; medical records; meta-analysis
Authors: Danielle A Southern; Matthew T James; Stephen B Wilton; Lawrence DeKoning; Hude Quan; Merril L Knudtson; William A Ghali Journal: Int J Popul Data Sci Date: 2018-11-12