The Evaluation of Matching in a Case-Control Study of Colorectal Cancer Using General Practice Lists

A crucial part of a case-control study is the selection of a sample of controls that represent the base-popula tion from which cases were drawn. Controls may be matched to cases by one or more potentially important confounding vari ables, such as socioeconomic status. In the United Kingdom, one method for control selection has been based on the pa tient list of the General Practice with whom the cases were registered, which we refer to as GP-matching. We aimed to ex plore whether GP-matching adequately control for the potential confounding effect of socioeconomic status. The Townsend index of deprivation was calculated for different two national census geography levels of Elec toral ward/Postcode Sector and Enumeration District/Output area for the three study areas of Dundee, Leeds and York. Con ditional logistic regression was used to estimate the association of cases with deprivation (based on the Townsend index) com pared with that of matched controls for the two geographical scales. At the largest geographical level (Electoral ward/Postcode Sector) there was no evidence of a difference in the dis tribution of deprivation scores between cases and controls. However, analysis at the smallest level (Enumeration Dis trict/Output area) showed that, despite GP matching, cases were more likely to live in deprived areas than matched controls. Using General Practice lists for the selection of controls for controlling the confounding effect of socioeco nomic status might not be an appropriate method for case-control studies conducted in the United Kingdom.