Constructing geographic areas for cancer data analysis: A case study on late-stage breast cancer risk in Illinois

Analysis of cancer data, particularly with a small geographic unit, often suffers from the small population (numbers) problem, which causes unstable rate estimates and data suppression in sparsely populated areas. This research proposes a regionalization approach to mitigate the problem by constructing larger areas in Geographic Information Systems (GIS) that are more coherent than geopolitical areas or arbitrary zip code area and census units in terms of attribute and spatial closeness. The method is applied to analysis of late-stage breast cancer risks in Illinois in 2000. Cancer rates in these newly-constructed areas have sufficiently large base population, and are thus more reliable and also conform to a normal distribution. This permits direct mapping, exploratory spatial data analysis, and even simple OLS regression. The method can be used to effectively mitigate the small population problem commonly encountered in analysis of public health data.