Screening for geographic heterogeneity of disease rates: application to cancer incidence in Allegheny County, Pennsylvania, 1969-71.

A statistical technique for analyzing heterogeneity in rates proposed by Gail is potentially well-suited to the analysis of geographic patterns in disease incidence data. This was applied to 1969-71 cancer incidence in whites in 30 census tract aggregates within Allegheny County, Pennsylvania to ascertain its usefulness and to search for patterns. A product model, in which each area's risks are a constant multiple of one another across age groups, was found to adequately fit the observed data. The test for geographic heterogeneity indicated that 19 of 30 sex-site groupings manifested greater than chance variation across the county subunits, generally consistent with the number of outliers based on 95% confidence limits. Partitioning the heterogeneity in these sites by median income quintiles suggested a positive risk gradient with breast cancer (F) and negative risk gradients with rectal (M), pancreatic (F), lung (M), cervical (F), and bladder cancer (F). Percent foreign stock was positively associated with total cancer (M, F), colon (M, F), pancreatic (F), lung (M), and bladder cancer (M, F). The technique successfully demonstrated the accuracy of a product model, screened cancer sites meriting detailed study, and suggested several associations of median income and percent foreign stock with cancer risk.