Problems in the interpretation of small area analysis of epidemiological data: the case of cancer incidence in the West of Scotland.

Open Access
Publisher Website
Google Scholar
Abstract
STUDY OBJECTIVE--The aim was to examine the extent to which random variation alone will produce differences in observed incidence rates between small areas which will affect measures of spatial clustering and estimates of relative risk. DESIGN--This was a study of changes in the pattern of spatial concentration of cancer incidence over a five year time period. A comparison was made of observed incidence rates for 34 tumour sites with randomly generated values and, where possible, with expected values derived from known relative risks. SETTING--Twenty six local government districts in the West of Scotland. MAIN RESULTS--A statistically significant relationship was observed between sample size and the stability of a summary measure of spatial concentration. Almost all observed highest:mean rate ratios were within the 95% confidence interval of the simulated distribution of these values. In three cases examined, both observed and simulated highest:lowest rate ratios were larger than those expected on the basis of known exposures to risk. CONCLUSIONS--In the absence of a prior hypothesis, small area analysis of epidemiological data for periods of less than 10 years will almost always give misleading results for all but the most common diseases.