Cancer deaths by city and county in Japan (1969–1971): A test of significance for geographic clusters of disease

Abstract

In geographic epidemiology, distribution of the categorized mortality or morbidity rates are visualized on a map. either based on actual land area or adjusted for the population density. Irrespective of the map used, visual study per se by no means indicates the statistical significance of the observed clusters, i.e. whether the geographic aggregations could occur by chance alone. We have developed an approach for assessing the deviation from chance expectation of the geographic pattern actually observed on a map and have described it in this paper.

A simple chi-square test is proposed, and the parameters required for the test are (1) total number of areas. (2) numbers of subareas for each mortality or morbidity category. (3) total number of geographically adjacent areas, and (4) observed numbers of adjacent areas having concordant category pairs.

When the test was applied to the geographic distribution of esophageal cancer mortality by city and county in Japan (1969–1971). the areas with high mortality were significantly clustered in both sexes, and those with low mortality in males.

There were no significant aggregations for breast cancer, though the areas with high mortality seemed distributed mainly in the northern half of the mainland. Japan. For uterus cancer low mortality showed significant clusters, and total geographic pattern was highly significant.

The validity of the proposed simple chi-square test of significance was substantiated by a Monte Carlo approach, which was derived analytically as a special case of Knox's test for space-time clustering.