C. P. Fontanet1, H. A. Carlos1, J. E. Weiss1,3, M. C. Gil-Díaz1, A. P. Loehrer1,2 1Dartmouth, Geisel School Of Medicine, Lebanon, NH, USA 2Dartmouth-Hitchcock Medical Center, Department Of Surgery, Lebanon, NH, USA 3Dartmouth Cancer Center, Lebanon, NH, USA
Introduction: The use of geospatial analysis methods is increasingly common in health services research, especially as large datasets and software tools become more widely available. As we attempt to evaluate the role of geography in the receipt of surgical care, we need to understand the pitfalls of geospatial analysis as it applies to evaluating geographical differences in access to and receipt of cancer care. One such example is the “Modifiable Area Unit Problem” (MAUP), a type of ecological fallacy which can lead to different results due to the areal unit chosen for analysis. Unfortunately, few projects take the time to recognize and measure the potential impact of this problem. In this analysis, we demonstrate the impact of using differing areal units for evaluating disparities in late-stage presentation for patients with breast cancer.
Methods: We retrospectively identified patients with incident breast cancer within the Indiana State Cancer Registry from 2010 to 2015. To examine geospatial heterogeneity of health outcomes, we compared the results of our analysis at three different geographical levels: block group, census tract, and county. The Global Moran's I statistic was used to investigate the overall clustering of location. To illustrate the potential impact of using different areal units, maps of rates of late-stage presentation at each level (block group, census tract, and county) in the metropolitan area of Indianapolis, IN were compared visually.
Results: Our sample included 30,604 patients residing in 4,814 block groups, 1,511 census tracts, and 92 counties. We observed similar proportion of late-stage presentation at the level of block group (15.2%), census tract (15.3%), and county (14.5%). However, we observed decreasing variance with increasing size of area (block group: 19.1, census tract: 11.0, county: 3.8). Our analysis showed decreasing spatial autocorrelation with increasing area units, as represented by decreasing Moran’s I statistic z-scores and p-values (p-values for block group: <0.001, census tract: <0.001, and county: 0.19). At the block group level, low case counts within block groups (<6), led to unstable rates (Figure). At the county level, we were unable to appreciate local variation in late-stage presentation rates.
Conclusion: To evaluate variation in late-stage breast cancer rates in Indiana, we found that census tract captured local variation and may be the best compromise to capture local variation while avoiding unstable rates. Efforts to evaluate geographical variations in cancer care outcomes should consider areal units at varying scales to determine the most appropriate unit of analysis.
