Don Kendal

Researchers in spatial epidemiology who wish to employ small area aggregations of health and demographic data are confronted by a conundrum. Small area health data lack robustness and may be compromised by suppression and rounding rules, while larger areas increase heterogeneity which reduces the strength of the relationship between health outcomes and the underlying socioeconomic determinants of population health. This research project was designed to determine whether the use of geodemographic segmentation to create larger, more homogeneous clusters can overcome the small area problem and produce more robust, stronger associations.

Simcoe County, Ontario, was segmented into ten discontiguous clusters based on 48 key socioeconomic and demographic variables from the 1996 census. Counts of deaths in 1996 and 1997 due to all causes as well as ischemic heart disease, cerebrovascular disease and lung cancer, the number of low birthweight babies born in 1996 and 1997, and counts of all hospital admissions and in-patients treated for the same three specific diseases between fiscal year 1996/97 and 2000/01 were aggregated by geodemographic cluster. These health data were also aggregated by municipality for purpose of comparison. The data were standardized into four SMRs and five SIRs for each level.

In order to determine whether areas grouped by geodemographic clusters or by administrative units (CSDs) produce strong associations between socioeconomic conditions and health indicators, linear regression was performed using the nine standardized health ratios as dependent variables. Three separate simple linear regressions were performed using average household income, adult unemployment, and the best predictor among 26 key socioeconomic variables as the independent variable. In addition, stepwise multiple regression models were calculated for each health indicator and 26 key socioeconomic variables. A comparison of the explanatory power (R2) of each predictor and the parsimony of the multiple regression models led to the conclusion that stronger predictions of spatial variations in health outcomes could be made using aggregation of population into geodemographic clusters than into local administrative units.