Population-based data sources for chronic disease surveillance

Abstract

This study estimated agreement between population-based administrative and survey data for ascertaining cases of arthritis, asthma, diabetes, heart disease, hypertension and stroke. Chronic disease case definitions that varied by data source, number of years and number of diagnosis or prescription drug codes were constructed from Manitoba's administrative data. These data were linked to the Canadian Community Health Survey. Agreement between the two data sources, estimated by the kappa coefficient, was calculated for each case definition, and differences were tested. Socio-demographic and comorbidity variables associated with agreement were tested using weighted logistic regression. Agreement was strongest for diabetes and hypertension and lowest for arthritis. The case definition elements that contributed to the highest agreement between the two population-based data sources varied across the chronic diseases. Low agreement between administrative and survey data is likely to occur for conditions that are difficult to diagnose, but will be mediated by individual socio-demographic and health status characteristics. Construction of a chronic disease case definition from administrative data should be accompanied by a justification for the choice of each of its elements.