Medicaid claims data are collected for administrative rather than research purposes. The use of administrative claims codes, which reflect both covered services and reimbursement practices, may result in under- or over-estimates of the prevalence of diabetes in the study population. A broad range of diabetes diagnostic codes encompassing both type 1 and type 2 diabetes was used to define the eligible population. The contextual associations found in this study may vary by diabetes diagnosis. Future research would benefit from the use of clinical records to define diabetes by type and to account for different type-specific physical severity levels.
The mechanisms by which area-level socioeconomic disadvantage adversely affects health are complex and incompletely understood. Small-area socioeconomic deprivation, itself, may directly compromise individual health; alternatively, relatively poor health outcomes in socioeconomically disadvantaged areas may reflect reduced access to health care, limited social support, social disorder, exposure to hazardous environmental pollutants, and/or local discriminatory practices [42–44]. In this study, both county-level measures of socioeconomic disadvantage—persistent poverty and unemployment—were positively associated with diabetes prevalence among adult African American Medicaid recipients in bivariate models. Of the two measures, however, only persistent poverty was a significant bivariate predictor. The lack of a statistically significant association between persistent poverty and diabetes in the multivariate model may be due in part to correlation of the predictor variables (although model diagnostics indicated no collinearity). Additional studies are needed to clarify the impact of persistent poverty at the small area level on diabetes prevalence.
Rurality was significantly positively associated with diabetes prevalence in both bivariate and multivariate models. Compared to urban counties, primarily or completely rural counties had higher rates of diabetes among adult African American subjects, even after adjusting for local levels of unemployment, persistent poverty, and food establishment availability. These results are consistent with other investigations showing higher rates of diabetes in rural versus urban areas [11, 12]. Elevated rates of diabetes in rural regions might reflect diminished access to primary care , a lack of sidewalks or other safe places to walk [16, 17, 46], the relative inaccessibility of parks and recreational facilities [17, 46], low social support , and/or regionally-specific rural cultural norms that can undermine health [46, 47]. The observed association between diabetes prevalence and rurality has important implications for public health policy creation and health promotion planning. In particular, efforts to reduce the burden of diabetes among adult African Americans must extend beyond city boundaries and address in culturally relevant ways the specific health needs of rural African Americans in South Carolina and across the Southern "black belt ." Notably, the association between rurality and diabetes in this study was specific to adult African American Medicaid recipients and should not be generalized to non-African American or younger Medicaid recipients, or to the broader non-Medicaid population.
Neither measure of the built environment was significantly associated with diabetes prevalence. This result might partly reflect the study's reliance on a single electronic business directory to identify and spatially locate chain fast food restaurants and convenience stores. Like other commercial business directories, the Dunn and Bradstreet product used in this investigation may contain incomplete business listings, outdated information, and/or street address data for corporate headquarters rather than local places of business. In addition, the self-classification of business type in the Dunn and Bradstreet directory may lead to inconsistent classification across listings. Further investigations are needed to evaluate potential small-area associations between diabetes and the relative availability of unhealthy food outlets, the proximity of such food establishments (e.g., distance to the nearest fast food restaurant or convenience store) , and the frequency with which fast food or convenience-type food products are consumed . Future studies also might consider potential contextual associations between diabetes and access to such healthy food outlets as "green" or farmers' markets, roadside fruit and vegetable stands, and large supermarkets with extensive produce sections.
Recognition of the dependent relationship between spatial phenomenon and geographic unit of analysis—the so-called Modifiable Areal Unit Problem—is critical to the understanding of studies that employ spatially aggregated data . In this investigation, diabetes prevalence and environmental data were aggregated at the county level. The results obtained, therefore, reflect only county-level environmental influences on diabetes prevalence among adult African Americans. Associations between diabetes and environmental context may be different at larger (e.g., state or national) and smaller (e.g., census tract or census block group) geographic scales. County-level spatial analyses of health are appropriate when county-level agents (e.g., county health departments, park and recreation departments, regulatory commissions, and councils on aging) play direct roles in disease prevention/intervention and wellness promotion programs. Although county-level investigations may suggest potential associations between health and environment at different geographic scales, such associations must be evaluated separately using appropriate geographic units of analysis.
As this study shows, ring maps can highlight racial disparities in health, convey epidemiological uncertainty data (e.g., confidence interval data associated with standardized morbidity and mortality rates), and suggest small area-level associations between adverse health outcomes and characteristics of the socioeconomic and built environment. The ring maps presented here only begin to illustrate the potential utility of this visualization method for health geographers. For example, ring maps can depict multiple attributes at the census tract, census block group, ZIP code area, hospital catchment area, or public health service area level, in addition to the county level as shown in the figures [25, 27]. In addition, a ring map can be used to depict a single attribute at multiple geographic scales. For instance, a ring map might show diabetes prevalence rates for South Carolina at the census tract level in a base map and prevalence rates at the county level and public health region (multiple county) level in successive rings (in this case, the number of enumeration units in the inner ring would reflect the number of counties, and the number of enumerations units in the outer ring would reflect the number of health regions in the state). Ring maps also can display time-series data for a single variable of interest [25, 27]. A map with six rings might show annual incidence rates of cardiovascular disease over a six-year period, for example; alternatively, a ring map might depict the weekly incidence of cases associated with an influenza outbreak. Ring maps thus permit the exploration of relevant distributions, patterns, and associations across both space and time [25, 27]. Additional layers of data are easy to add to ring map visualizations, requiring only that new rings be drawn . Although the ring maps shown are circular, elliptical or even non-continuous rings can be drawn to accommodate irregularly shaped geographic regions [25, 27]. In short, ring maps provide sufficient flexibility in design and development to permit the visualization—and visual exploration—of spatiotemporal data across a wide range of health applications.
A distinct problem associated with ring map visualization is the loss of complete topology (i.e., information about the spatial relationships of geographic units) in the rings. Although a single county in South Carolina may have as many as nine adjacent neighbors, only two adjacent neighbors (spokes) exist in ring displays. Complete spatial topology is retained in the central base map, though, allowing users to determine adjacent relationships, relative direction, and the relative nearness or farness of geographic units. Another practical limitation is the ability of ring maps to display data for a large number of geographic units. It would be challenging to construct—and difficult to interpret—a ring map depicting the more than 800 census tracts in South Carolina, for instance. Ring maps—as they appear in print—also suffer limitations common to all static map products. They depict a set number of predetermined data layers, using a static data classification method, and an unchanging symbolization scheme.
Most of the graphic limitations associated with static ring map visualization, including the limited representation of spatial topology in rings, might be addressed in a dynamic ring mapping environment. In such an environment, a user could interactively select the geographic area of interest, establish the geographic scale of representation, choose data elements for exploration, assign attributes to rings, modify the classification and symbolization of data, reorder (and even resize and reshape) rings, and "click on" one or more ring attribute elements to see the corresponding geographic unit(s) highlighted on the base map in complete spatial topological context. A dynamic and fully interactive ring mapping application of this sort would represent a powerful information visualization tool with which to integrate and explore diverse data sets, frame questions, generate hypotheses, restructure problems, and achieve insights  critical to the protection and promotion of population health.
Further studies are needed to evaluate the relative strengths and weaknesses of ring maps—in both static and dynamic forms—as multivariate data visualization tools. Such studies might explore the optimal number of rings to display for specific data types and data type combinations, the maximum number of aggregation units that can be effectively depicted, and the effect of color symbolization on ring map interpretation. Finally, research is needed to demonstrate the relative effectiveness of ring maps versus small multiple map displays in the visualization of multivariate health data.