Background: Geographic information system (GIS)-based health studies require information on the physical location of data points, such as subject addresses. In a study of California women diagnosed with breast cancer between 1988 and 1997, we needed to locate the residential addresses of 4,537 women with post office boxes (POBs).
Methods: We investigated the feasibility of tracing street addresses for the POBs and examined potential selection biases and case attribute misclassifications introduced by different methods of handling POBs in GIS-based health studies.
Results: Our tracing method yielded street addresses for only 34% of POBs in our study. Examination of subjects' case characteristics revealed that boxholders were not representative of the full population. Geocoding using a POB's delivery-weighted five-digit zip code centroid, as a proxy for street address, resulted in case attribute misclassification for 81% of boxholders.
Conclusions: Disease registries should modernize their infrastructure to complement GIS technologies. Epidemiologists should understand GIS data limitations and consider potential biases introduced by incomplete or inaccurate geocoding.