Hospital or population controls for case-control studies of severe childhood diseases?

There are few empirical data to determine which control group seems best in a case-control study for a severe disease: population controls or hospital controls. The author conducted a case-control study of leukemia in children using two control groups, population and hospital controls (cancers other than leukemia and severe blood diseases), between 1980 and 1993 in Québec, Canada. Maternal, paternal, and child factors not known to be associated with leukemia as well as factors possibly associated were selected for analysis. Most factors were taken directly from parental interviews, but two factors related to parental occupational exposures were blindly coded by chemists. Hospital and population controls were compared using odds ratios estimated from logistic regression. Cases were compared with both types of controls with the same statistical method. Prevalence data from ongoing population surveys were compared with reported prevalence in controls. From the former comparisons and the distribution of socioeconomic variables, results suggested that study groups came from the same base population. Nevertheless, reported and coded exposures among hospital controls were closer to those of cases than to those of population controls. Although substantially different for only one factor, inferences using hospital controls in comparison with population controls resulted in odds ratios closer to the null value.