Age-Adjusted Diabetes Mortality Rates Vary in Local Communities in a Metropolitan Area: Racial and Spatial Disparities and Correlates

Despite increased awareness of the problems associated with diabetes and increased efforts to bring this national epidemic under control, the prevalence of diabetes continues to rise in the U.S. In 2005–2008, diabetes prevalence among adults ≥20 years of age was 11%, up from 8% in 1988–1994. Diabetes has also been reported with increased frequency among children and adolescents over the last 2 decades (1). In the 5-year time period 2006–2010, diabetes was the seventh leading cause of death in the U.S. (2–6). In 2010, the age-adjusted diabetes mortality rate (AADMR) in the U.S. was 20.8 per 100,000 population (6).

Efforts to understand differences in diabetes morbidity and mortality have revealed variation by gender (7–12), age (8,11,13,14), socioeconomic status (15–18), and race/ethnicity (7–9,11–13,16,19,20). Nationally, for example, the highest diabetes mortality rates are found among African Americans (39.6), followed by Hispanics/Latinos (27.1) and then whites (18.2) (6).

These national data are important because they help to shape policy and inform the public about this growing epidemic. However, while national and sometimes even state- and city-level data may be relatively easy to find, community- or neighborhood-level data are less readily available. This is unfortunate because this is just the type of data needed to reveal variations that are often masked by averages, to inform local policies, and to engage communities in the pursuit of improved health (21).

The extensive race/ethnic variation observed at the national level and even at the city and county level (8,9,22) lends support to the notion that similar variation might be found among communities/neighborhoods within large urban areas. For example, the city-wide average for a place like Chicago, which is a highly segregated city (23), likely masks variation within the city in much the same way that the national average does among cities and states. The use of local-level data to determine which communities are particularly burdened with diabetes is imperative if we are to effectively address this problem. In this article we provide such data using Chicago as an example and urge the reader to consider the utility of such an analysis for his/her community.

Chicago is the third largest city in the U.S., with 2,700,000 residents. The racial/ethnic makeup of Chicago in 2010 was 29% Hispanic, 32% non-Hispanic (NH) black, and 32% NH white. In 2008, there were just under 20,000 deaths in the city (24). Chicago is divided into 77 officially designated community areas, which are the sums of census tracts and have historically been used to examine local variations in health data.

We analyzed death certificates from Illinois Vital Records supplied to us by the Chicago Department of Public Health (24) for all deaths of residents of the city of Chicago during the years 2006, 2007, and 2008 (2008 being the most recent year for which data were available). We used data from these years to calculate a 3-year average AADMR. These certificates also contain data on the decedent’s community area of residence at the time of death. These data were derived by the Chicago Department of Public Health from a record set supplied by the Illinois Department of Public Health (IDPH). IDPH specifically disclaims responsibility for any analysis, interpretations, or conclusions.

Deaths from diabetes were defined to consist of all deaths with an underlying cause, coded as E10–E14 under the International Classification of Diseases, 10th Revision. Data for denominators were drawn from the U.S. Census Bureau, 2000 and 2010. Using data from these two points, we performed linear interpolation to obtain the denominators for 2006, 2007, and 2008. All mortality rates were age-adjusted to the U.S. standard 2000 population.

Two demographic variables were included in the analysis: racial/ethnic makeup and median annual household income of each community area of residence. Data on the race/ethnic-specific populations in each community area were obtained from the U.S. Census Bureau, 2000 and 2010. Linear interpolation was used to calculate the 2006–2008 3-year average percentages for each race/ethnic group. Our analysis includes multiple racial/ethnic groups within the city of Chicago: NH white, NH black, and Hispanic (including Mexican and Puerto Rican, where possible). Data on median household income for each community area were obtained from the American Community Survey, 2010 (5-year estimates for 2006–2010).

For purposes of comparison, we present AADMRs at the national level as well. These data are provided annually via the National Vital Statistics Reports of the Centers for Disease Control and Prevention, but the reports do not include death data for Puerto Ricans and Mexicans. We thus abstracted numerator data for 2006, 2007, and 2008 from death files obtained from the National Center for Health Statistics (25) and used population data from the annual reports to calculate race/ethnic-specific 3-year averages for the U.S. (2–4).

Data were analyzed using SAS version 9.2 (26). We calculated the AADMR for each of the 77 community areas and then correlated these with the race/ethnicity and the median household income of the area. Spearman correlation coefficients were calculated using the SAS procedure PROC CORR. The t tests were used to determine whether the correlation coefficients were statistically significant. The z scores were calculated to test whether the Chicago-U.S. differences in AADMRs were significant and to test whether differences between racial/ethnic groups within Chicago were significant (5). A P value of ≤0.05 was considered statistically significant for all analyses.

Table 1 displays the AADMRs for Chicago and the U.S., in total and by race and ethnicity for the years 2006–2008 (3-year averages). The rate for the city of Chicago was 27.5 per 100,000 population, which was significantly higher than the national rate of 22.5 (P < 0.001). Within both the U.S. and the city of Chicago, the highest AADMRs are found among NH blacks (blacks), followed by Hispanics, and then NH whites (whites). The U.S. rate was significantly higher than the Chicago rate among blacks (43.6 vs. 35.9; P < 0.001) and about the same as the Chicago rate for whites (19.8 vs. 19.6; P > 0.05) and Hispanics (28.8 vs. 29.1; P > 0.05). The similarity of the “Hispanic” rates obscures important differences within this category. The rate among Puerto Ricans was higher in Chicago than for the nation (45.7 vs. 35.0; P = 0.0561), while the rate among Mexicans was significantly lower (27.5 vs. 34.6; P < 0.01).

Notably, within the city of Chicago, the highest AADMRs are found among Puerto Ricans (45.7), and the third highest rates are found among Mexicans (27.5).

In order to further examine these racial/ethnic differences in AADMRs, we calculated mortality rates at the community area level. Table 2 displays AADMRs, ranked from highest to lowest, along with racial/ethnic makeup and annual household income for each of the 77 community areas in Chicago. Figure 1 displays a map of these community areas with light to dark shading for lowest to highest AADMRs. The AADMR ranged from a low of 8.1 in the Loop to a high of 52.1 in Burnside. The community areas with the highest AADMRS (those in the 4th quartile) were overwhelmingly located on the west and south sides of the city, and had predominantly black populations.

Table 3 displays correlation coefficients for AADMRs and racial/ethnic makeup of a community area. There was a strong positive correlation between the proportion of black residents in a community area and the AADMR (0.64; P < 0.0001). Conversely, there was a strong negative relationship between the proportion of white residents and the AADMR (−0.68; P < 0.0001). The correlation was negative and smaller but statistically significant for Hispanic people (−0.34; P < 0.01).

Table 3 also displays correlation coefficients for AADMRs and median household incomes. Here we observed a strong negative relationship between household income and the AADMR for the entire city (−0.63; P < 0.0001). This relationship prevailed for the 28 community areas that were predominantly (≥60%) black (−0.52; P < 0.01). For the 13 predominantly white community areas, the correlation was negative, but weak and not statistically significant (−0.11; P = 0.72). For the 12 predominantly Hispanic community areas, the relationship was positive, but weak and not statistically significant (0.15, P = 0.63). For community areas in which there was no majority racial/ethnic group (n = 24), the relationship was weak, negative, and not statistically significant (−0.10, P = 0.65).

This analysis explores disparities in AADMRs across several geographic entities. First, the rate in Chicago (27.5) is significantly higher than the national rate (22.5), and there are important similarities and differences among racial/ethnic groups as well. For example, for both Chicago and the U.S., above average rates are found among blacks, Puerto Ricans, and Mexicans, while below average rates are observed among whites. However, the magnitude of these differences within these two places is what is really striking. The Puerto Rican rates (45.7 in Chicago and 35.0 in the U.S.) are greatly elevated at over 1.5 times the city and national rates (Table 1), and are also much higher than the Mexican rates.

Furthermore, we found that there is substantial variation in diabetes mortality rates across the 77 community areas of Chicago, ranging from a low of 8.1 to a high of 52.1, with substantial variation even among community areas that are predominantly black (Table 2). Finally, there are strong correlations between mortality rates and the racial, ethnic, and socioeconomic distributions among the community areas (Table 3).

Analyses at the community level are rare. For example, if we were to list the possible levels of analysis in order from most general to most specific, they would range across national, state, county, city, zip code, and finally community or neighborhood. We have located examples of each of these (1,8–9,22,27–30), but we find that as the level of analysis gets smaller, so does the number of analyses performed at this level. This is rather unfortunate as we have seen the positive effects that this type of data can have for motivating communities to make changes to overcome such challenges to health (21). For example, when we uncovered a high prevalence of and mortality from diabetes in a community in Chicago, the information led to an intervention, the development of a community-based organization to address the problem and a substantial National Institutes of Health grant (31). Preliminary analyses of our diabetes work, which are currently being prepared for publication, have revealed substantial reductions in HbA_1c level, meeting or exceeding those seen in other diabetes interventions in the literature. Our hope is that a similar response might come from other communities shown here to have particularly high diabetes mortality rates.

Our community-level analysis is very much tied to our racial/ethnic analysis, given the “hypersegregated” nature of the city of Chicago (23). While rates among racial/ethnic groups for Chicago and the U.S. are similar, what this analysis has revealed is that Puerto Ricans display much higher AADMRs (45.7) than any other racial/ethnic group in Chicago, much higher than the Hispanic average of 29.1. In a sense, breaking apart the category of Hispanic data is another way of examining variation, and in this case the results are startling, with Puerto Ricans having a diabetes mortality rate that is more than 1.5 times as high as that of Mexicans.

The notions that 1) the health profiles of Hispanic Americans might differ from those of their NH American counterparts or 2) there might be variation in health profiles across Hispanic subgroups are not recent. As far back as the late 1950s, researchers began examining death rates for Hispanic populations, despite the difficulties associated with generating such estimates. These early analyses focused on Mexican Americans and, because of data limitations, were geographically limited to areas with large Mexican American populations—specifically, Houston and San Antonio, TX, and California (32–37). The first analysis of Puerto Rican mortality was based on data for New York City and compared Puerto Ricans to non-Puerto Rican whites (38). Similar data for the Chicago metropolitan area (Cook County, IL) were published in 1987 (39) and presented mortality rates for Mexican- and Puerto Rican-born populations. In regard to diabetes mortality, the authors found elevated rates among Mexican-born male individuals and Puerto Rican-born female individuals compared with other whites.

More recently, Whitman et al. (27) presented AADMRs for Puerto Ricans and Mexicans in Chicago, using 1999–2001 3-year averages. They reported elevated rates among both Mexicans (41.7) and Puerto Ricans (70.9). This analysis contributes to the existing Hispanic origin literature by updating the data on diabetes mortality rates for Mexican and Puerto Rican populations in the city of Chicago. Both the Puerto Rican and Mexican rates appear to have declined by ∼30% during this 6-year time interval.

While the present analysis does not provide data on nativity or length of residence in the U.S., we recognize the importance of these variables in explaining the differences in mortality rates between Hispanic and NH populations in general, and within the Hispanic population, in particular. For instance, we know that Mexicans living in Chicago tend to be more recent immigrants (40), and research suggests that diabetes prevalence increases with length of residence in the U.S. (41). This might help explain the lower mortality rates observed among Mexicans compared with Puerto Ricans.

Finally, despite having a lower rate than at the national level (43.6), blacks in Chicago display higher rates of diabetes mortality (35.9) than their white counterparts (19.6), and we found that areas that are predominantly black have the highest rates of diabetes mortality in the city. Indeed, we found a strong positive correlation between the proportion of a community area that is black and the AADMR. Notably, even though the correlation is weak and not significant, we also observe some of the highest AADMRs in communities that have large Hispanic populations.

Analyses of diabetes mortality generally tend to include in the numerator deaths for which diabetes was listed as the underlying (or primary) cause of death. Our analysis followed this general paradigm for producing diabetes mortality rates. However, it is also possible to include deaths for which diabetes was listed as an underlying or contributing cause of death, and these data are usually presented as diabetes-related mortality. Smith and Barnett (20) provide an example of an analysis that examines diabetes-related mortality for Hispanic subgroups. The authors report elevated rates of diabetes-related mortality among Puerto Ricans (204) and Mexicans (251) compared with Cubans (101), and, notably, the rates were higher among Mexicans than Puerto Ricans. Future analyses should therefore also explore diabetes-related mortality rates in addition to the usual underlying cause analysis. While questions remain as to the reliability of death certificates in ascertaining diabetes-related mortality, this was not an issue for the present analysis given the focus on diabetes mortality (diabetes as the underlying cause of death vs. diabetes as any contributing cause of death).

When calculating rates, it is common to combine death data from ≥2 years to stabilize estimates. For this analysis, we used death data from 3 years (2006–2008) to estimate diabetes mortality. It is worth noting that even when data are pooled in this manner, there are still fluctuations from year to year that can cause rates to increase/decrease by a large degree and create a false sense of improvement/deterioration. For instance, while the Puerto Rican and Mexican rates reported here suggest a substantial improvement in diabetes mortality among these groups over the past 6 years, it should be noted that the number of deaths, while usually consistent from year to year, does drop for 1 or 2 years every few years. This was the case for Puerto Ricans in 2007 and 2008 when there were about 10 fewer deaths than is normal. A change like this may substantially alter the age-adjusted rate. The only way to be sure that rates are really changing is to examine trends over substantial periods of time.

Finally, as stated in the 1Introduction, diabetes mortality has been found to vary by gender. However, given the focus of our analysis on racial/ethnic differences and local variation, we did not examine differences by gender. We do not doubt that the findings presented here might differ if broken out by gender, and we certainly encourage future research on this topic, especially given its relevance to intervention work.

Whereas high-level aggregation provides invaluable insights into how the country as a whole, or even states, are faring with respect to diabetes mortality, disaggregation presents us with data that may facilitate targeted interventions and community engagement. The data presented in this article provide a helpful road map to where the worst diabetes mortality problems reside in Chicago. Our hope is that these data can be used to work toward the development of solutions to the very high diabetes mortality rates observed in several communities in Chicago and in similar communities throughout the U.S.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. B.R.H. researched data and wrote the manuscript. S.W. contributed the concept of the manuscript, and wrote, reviewed, and edited the manuscript. C.A.H. researched literature and contributed to the 1Introduction and 7Conclusions sections of the manuscript. B.R.H. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented in abstract form at the 139th Annual Meeting and Exposition of the American Public Health Association, Washington, D.C., 29 October–2 November 2011, and at the 140th Annual Meeting and Exposition of the American Public Health Association, San Francisco, CA, 27–31 October 2012.