Spatial clusters of diabetes: individual and neighborhood characteristics in the ELSA-Brasil cohort study

This study identified spatial clusters of type 2 diabetes mellitus among participants of the Brazilian Longitudinal Study of Adult Health (ELSA-Brasil) residing in two cities and verified individual and neighborhood socioeconomic environmental characteristics associated with the spatial clusters. A cross-sectional study was conducted with 4,335 participants. Type 2 diabetes mellitus was defined as fasting blood glucose ≥ 126mg/dL (7.0mmol/L), oral glucose tolerance test ≥ 200mg/dL (11.1mmol/L), or glycated hemoglobin ≥ 6.5% (48mmol/L); by antidiabetic drug use; or by the self-reported medical diagnosis of type 2 diabetes mellitus. Neighborhood socioeconomic characteristics were obtained from the 2011 Brazilian census. A spatial data analysis was conducted with the SaTScan method to detect spatial clusters. Logistic regression models were fitted to estimate the magnitude of associations. In total, 336 and 343 participants had type 2 diabetes mellitus in Belo Horizonte, Minas Gerais State (13.5%) and Salvador, Bahia State (18.5%), respectively. Two cluster areas showing a high chance of type 2 diabetes mellitus were identified in Belo Horizonte and Salvador. In both cities, participants living in the high type 2 diabetes mellitus cluster area were more likely to be mixed-race or black and have a low schooling level and manual work; these were also considered low-income areas. On the other hand, participants in the low type 2 diabetes mellitus cluster area of Salvador were less likely to be black and have low schooling level (university degree) and live in a low-income area. More vulnerable individual and neighborhood socioeconomic characteristics were associated with living in clusters of higher type 2 diabetes mellitus occurrence , whereas better contextual profiles were associated with clusters of lower prevalence.

This study was developed with 3,115 participants living in the city of Belo Horizonte (Minas Gerais State) and 2,029 participants living in the city of Salvador (Bahia State) at baseline.Among them, those with missing geographic coordinates (n = 59), those with geographic coordinates out of the municipalities of Belo Horizonte and Salvador (n = 570 and n = 84, respectively), and participants who self-reported as Indigenous (n = 69), due to the very few observations, were excluded.The final sample size was 2,486 individuals for Belo Horizonte and 1,849 for Salvador.
Belo Horizonte is the state capital of Minas Gerais, with Human Development Index (HDI) of 0.810 16 and an estimated population of 2,375,151 inhabitants, being the sixth most populous municipality in Brazil 17 .Salvador is the state capital of Bahia.with a 0.759 HDI 16 and an estimated population of 2,675,656 inhabitants, being the third most populous municipality in Brazil 17 .
The data collection team at the ELSA-Brasil Research Centers was trained and certified for the collection of anthropometric measurements, blood samples, and application of questionnaires.Collection in all centers was standardly conducted according to the procedures described in the specific Operation Manuals for interviews and exams.
The study outcome was to live inside or outside a detected prevalence cluster of type 2 diabetes mellitus.This variable was fitted in two steps.
Participants were considered as having type 2 diabetes mellitus when they presented the following parameters: fasting blood glucose ≥ 126mg/dL (7.0mmol/L), oral glucose tolerance test (OGTT) ≥ 200mg/dL (11.1mmol/L), glycated hemoglobin (HbA1c) ≥ 6.5% (48mmol/L), or by antidiabetic drug use; or by the self-reported medical diagnosis of diabetes.Blood samples were collected after a 10 to 14-hour fasting, stored in a freezer at -80°C and sent to the certified central laboratory in São Paulo (Brazil).An OGTT was administered to all participants without a known diabetes diagnosis.Blood glucose level was measured using the enzymatic colorimetric method (ADVIA 1200, https://www.diamonddiagnostics.com),and HbA1c was measured using high-pressure chromatography (HPLC, https://www.bio-rad.com).
Initially, the database from ELSA-Brasil was obtained in georeferenced formats from the residential address informed at baseline.The census sectors were identified from the geographic coordinates (X and Y coordinates of the study participants) of each point of residence of the individuals in the study who were georeferenced in the census tract.
The first step to remove out-of-region points is defining the shape of each capital city, which in this case was defined as the shape obtained from the geobr R package (http://www.r-project.org) for each municipality, constituted of all the shapes of its census tracts.
After data preparation, a spatial data analysis was conducted with the SaTScan (http://www.satscan.org)method of cluster detection of punctual data, which tests for the existence of an area with multiple clusters of any size up to 20% of the total population for the outcome.The SaTScan detects a spatial cluster on different maps and computes its significance based on Monte Carlo simulations.The Bernoulli spatial scan statistic 18 was used to detect spatial clustering of tye 2 diabetes mellitus among the participants of ELSA-Brasil.Analyses were performed at the individual level to maximize spatial heterogeneity 19 .For statistical inference the null hypothesis of complete spatial randomness was rejected at a p-value ≤ 0.05.
The covariates of this study were divided into: individual characteristics of the participants and neighborhood characteristics of the area where the participants lived.
Individual characteristic variables included sex; age; race/skin color; schooling level; nature of occupation.This study also included tobacco use; alcohol consumption measured by the sum of doses of each type of alcoholic beverage consumed turned into grams, with excessive consumption being ≥ 210g/week of alcohol for men and ≥ 140g/week for women; leisure physical activity was assessed using the long-modified version of the International Physical Activity Questionnaire (IPAQ) and categorized according to time spent at different intensities of physical activity (light: < 600 MET-min/ week; moderate: 600-3,000 MET-min/week; or vigorous: ≥ 3,000 MET-min/week) 20 ; waist circum-Cad.Saúde Pública 2023; 39(5):e00138822 ference (WC) measured at the midpoint between the last rib and the iliac crest, using a non-elastic anthropometric tape (Mabis, https://www.livehealthsmart.com)with 0.1cm precision; and abdominal obesity, which was classified according to IDF cut-off points: WC ≥ 90cm for men and WC ≥ 80cm for women 21 .
The neighborhood socioeconomic environmental variables were obtained from the 2010 Brazilian Demographic Census 17 .The unit of analysis to build up neighborhood variables was the census tracts.This study included: (1) neighborhood household income per capita (in Brazilian Reais -BRL) stratified in tertiles, the highest area income was used as the reference in data analysis; (2) percentage of neighborhoods with adequate housing (dwellers living in properly identified adequate houses/ dwellers in owned homes); (3) percentage of neighborhoods with litter (dwellers with an accumulation of litter on the public roads/dwellers in owned homes); (4) percentage of neighborhoods with sidewalks (dwellers that have sidewalks/dwellers in owned homes); due to the asymmetric distribution, these last three variables were divided according to the best possible statistical participation using two categories: low and high (cut-off points = 67th percentile for percentage of neighborhoods with adequate housing, 85th percentile for percentage of neighborhoods with littering, 66th percentile for percentage of neighborhoods with sidewalks in the city of Belo Horizonte).For the city of Salvador, the only variable with two categories was the percentage of neighborhoods with exposed trash (cut-off points = 80th percentile), with the others divided by tertiles.
Then, a frequency distribution of individuals and socioeconomic environmental characteristics was performed (Tables 1 and 2), fitting the logistic regression models to estimate the magnitude of the associations of the individuals and neighborhood characteristics between participants living inside and those living outside the clusters.The models were adjusted for sex, age, race/skin-color, schooling level, smoking, alcohol consumption, nature of occupation, leisure physical activity, waist circumference, and neighborhood socioeconomic environmental variables.
All analyses and plots presented and discussed in this study were produced using the R programming language version 3.5.1 with the RStudio IDE v1.3.125 and the SaTScan program version 9.6.0.

Results
Tables 1 and 2 describe individual and neighborhood socioeconomic characteristics.The characteristics that predominated in the study population were female sex, aged 41 to 60 years, self-reported race/skin color as white (Belo Horizonte) or mixed-race/black (Salvador), higher education, nonmanual work, never smoked, moderate alcohol consumption, and light leisure physical activity.Prevalence of type 2 diabetes mellitus was higher among women, participants with manual jobs, and those with abdominal adiposity (Table 1).In Belo Horizonte, the prevalence of type 2 diabetes mellitus was higher in neighborhoods with low-income household and in neighborhoods with a high percentage of littering.In Salvador, the prevalence of type 2 diabetes mellitus was higher in neighborhoods with the poorest environmental socioeconomic characteristics for all indicators included in this study (Table 2).
Figure 1 shows the cluster area with the highest prevalence of type 2 diabetes mellitus in Belo Horizonte (circle; n = 132).This area was in the northeastern region of the city, with a 4.8km radius.The prevalence ratios of type 2 diabetes mellitus were 28.8% and 13.5%, respectively, among those who lived inside and outside the cluster area.
Figure 2 shows the cluster area with the highest prevalence of type 2 diabetes mellitus in Salvador (n = 355).This area was in the northern region of the city, with a 8.5km radius.The prevalence of type 2 diabetes mellitus was 27.6% and 18.5%, respectively, among those who lived inside and outside the cluster area.The Figure 2 also shows the cluster area with a low prevalence of type 2 diabetes mellitus (n = 71).This area was in the southern region of the city, with a 0.92km radius.The prevalence of type 2 diabetes mellitus was 1.4%.
Tables 3, 4, and 5 present the comparison of the individual characteristic of the participants who lived inside and outside the cluster area of type 2 diabetes mellitus.
After adjusting for covariables, living inside the cluster area with a high prevalence of type 2 diabetes mellitus was associated with increased odds of self-reporting mixed-race and, primarily, black Cad.Saúde Pública 2023; 39(5):e00138822 ND: no data.
Cad. Saúde Pública 2023; 39(5):e00138822 race/skin color, a lower education level, and manual work occupation than those living outside the cluster (Tables 3 and 4).Table 6 shows the comparison of the socioeconomic neighborhood environmental characteristics, after adjustment for all covariables, from the area inside and outside of the type 2 diabetes mellitus cluster area.
In Belo Horizonte, 90.1% of the area within the cluster presented low-income neighborhoods.Thus, this location is characterized as an impoverished area, which is the only socioeconomic environmental factor independently related to the cluster type 2 diabetes mellitus area.In Salvador, the cluster with the highest prevalence of type 2 diabetes mellitus had a substantial and high variation in odds ratio (OR) estimated for each covariable (area income, neighborhoods with adequate housing, neighborhoods with litter, and neighborhoods with sidewalks).In the same municipality, participants living inside the cluster areas with a low prevalence of type 2 diabetes mellitus had lower odds of reporting mixed-race or black race/skin color and a lower schooling level, and higher odds of excessive alcohol consumption than participants living inside, after adjusting for covariables (Table 5).When we analyzed the OR, the cluster with a low prevalence of type 2 diabetes mellitus in Salvador exhibited geographical areas with higher incomes.

Discussion
This study used spatial scan methods to identify clusters of type 2 diabetes mellitus and verified individual and neighborhood characteristics associated with spatial clusters of type 2 diabetes mellitus in ELSA-Brasil participants living in Belo Horizonte and Salvador.One spatial cluster of type 2 diabetes mellitus was found in Belo Horizonte (higher prevalence of type 2 diabetes mellitus), whereas two spatial clusters of type 2 diabetes mellitus were found in Salvador (higher and lower prevalence of      type 2 diabetes mellitus).At the individual level, black and mixed-race individuals and lower schooling levels were associated with greater chances of belonging to the cluster area with a high prevalence of type 2 diabetes mellitus in Belo Horizonte and Salvador, even after adjustments for the individual covariables.In the cluster identified with a low prevalence of type 2 diabetes mellitus in Salvador, this study found a minimal chance of black and low schooling level participants living in this area.In both cities, the low levels of neighborhood income per capita were related to high spatial clusters of type 2 diabetes mellitus.In Salvador, the best level of these variables (neighborhood income per capita and pertcentage of neighborhoods with adequate housing) were associated with the low spatial cluster of type 2 diabetes mellitus.
Our findings indicate a high prevalence of type 2 diabetes mellitus in participants of ELSA-Brasil from Belo Horizonte (13.5%) and Salvador (18.5%).In Belo Horizonte and Salvador, the occurrence of type 2 diabetes mellitus was greater in the cluster areas located, respectively, in the northeastern and northern regions compared to other areas.These cluster areas were characterized as being impoverished locations, with lower neighborhood income and higher proportions of mixed-race/black people, lower schooling levels, and manual work occupations.By contrast, the southern region of Salvador was a cluster area with a lower chance of type 2 diabetes mellitus.This area was characterized as a wealthier location, with a higher neighborhood income, and with a high schooling level and low proportion of black people .Finally, in both cities, we identified that, regardless of the individual characteristics, poor socioeconomic status of neighborhoods increased the chance of type 2 diabetes mellitus (low income, low percentage of adequate housing, high percentage of litter, and low percentage of sidewalks).
In summary, our results showed a relationship between poverty and social inequality and type 2 diabetes mellitus.Poorer people are more exposed to illness from NCDs, including type 2 diabetes mellitus, and to the worsening of their clinical conditions, due to the difficulty in meeting their health demands 22 .Furthermore, low schooling levels make it difficult to understand health promotion, disease prevention, and treatment actions, limiting the subject's empowerment regarding self-care 22,23 .
Regarding skin color, some studies have suggested that the racial and ethnic disparities observed in type 2 diabetes mellitus may reflect differences in socioeconomic factors associated with skin color rather than genetic issues 24,25 .
Furthermore, our results indicated that the areas with the highest and lowest chance of the occurrence of type 2 diabetes mellitus had territorial characteristics that highlighted social inequities.The areas with the highest concentration of people with type 2 diabetes mellitus had higher percentages of low-income families and neighborhoods with litter, and lower percentages of neighborhoods with adequate housing and neighborhoods with sidewalks.This profile was the opposite in the area with Cad.Saúde Pública 2023; 39(5):e00138822  the lowest concentration of people with type 2 diabetes mellitus.Previous studies corroborate our findings, as they also showed that the areas with the greatest chance of the occurrence of type 2 diabetes mellitus were characterized as areas with worse socioeconomic conditions 9,11 .A survey conducted in Argentina found that the diabetes mortality rate was higher in provinces with low socioeconomic development.Such results could be a consequence of structural problems and health service coverage 26 .
The distribution of population and economic activities in the urban space of cities is not homogeneous, but rather very uneven in several aspects.This article exemplifies this spatial inequality, associated with the prevalence of type 2 diabetes mellitus.The literature on regional and urban economics indicates two major forces that act on this distribution of economic activities 27 .The first is a specific hierarchy in the complexity of products and services offered in urban environments.
The effect on land rent, or rents, in this spatial distribution of economic activities on urban space will also affect the spatial location of the residences or dwellings of a city's population.Homes with more space available and more equipment associated with quality of life and leisure will be attracted to the proximity to these centers and the "concentric rings" formed in their surroundings.This trend will be reproduced by real estate developers who seek to maximize the income earned from the land they develop.Thus, the supply of high-standard housing is usually not homogeneous across space, but similar to economic activities concentrated in neighborhoods close to urban centers and constituents of so-called noble areas 28 .
The constitution of this "urban fabric" produces a socioeconomic segregation not only by income level but is also reproduced in the urban space.The families with the highest income are in the same neighborhoods and regions, which, due to the logic of distribution of economic activities, are the neighborhoods closest to complex and diversified urban services.This dynamic creates a differentiation in the quality of life of these populations compared to those who live in peripheral areas 28,29 .
Contemplating policies that do not consider this socioeconomic segregation indicated by the spatial distribution reinforce the tendency of increasing spatial heterogeneity associated with the occurrence of chronic diseases 30 .
This study used the circular scan statistic of Kulldorff 18 to indicate the spatial structure of the point process associated with type 2 diabetes mellitus.This is a well-known tool in spatial analysis, able to handle the major problems that have been faced by scan-type statistics in general: multiple hypothesis tests associated with a scanning window that moves through many regions of varying shapes and sizes, and the presence of non-negligible spatial correlations in the underlying point process.In fact, statistical inference under these circumstances tends to be problematic: the correlations that arise between the different tests due to factors such as proximity and adjacency invalidate standard procedures such as Bonferroni-type corrections for p-values.The method proposed by Kulldorff 18 solves both problems by considering a maximum likelihood ratio test based on a single p-value associated with the most likely cluster area.From an inferential point of view, this is an important advantage of this approach and one of the main reasons for it becoming a standard in investigations such as the one we presented here.
The cross-sectional design of this study limits the possibility of establishing temporal relation between variables.The association observed between the neighborhood and individual characteristics with type 2 diabetes mellitus inside and outside the clusters may be subject to unmeasured sources of confounding factors.However, generalization in epidemiological studies may also be based on the plausibility of associations 31 to guide, for example, prevention and control programs to the different areas identified.

Figure 1
Figure 1The probable cluster with a high chance (red circle) of the presence of type 2 diabetes mellitus by the circular spatial scan statistic of punctual data.Belo Horizonte, Minas Gerais State, Brazil.ELSA-Brasil 2008/2010.

Figure 2
Figure 2The probable clusters of high (red circle) and low (blue circle) chances of the presence of type 2 diabetes mellitus by the circular spatial scan statistic of punctual data.Salvador, Bahia State, Brazil.ELSA-Brasil 2008/2010.

Table 3
Individual characteristics of the participants who live inside and outside the cluster with higher chance of type 2 diabetes mellitus and their association.Belo Horizonte, Minas Gerais State, Brazil.ELSA-Brasil 2008/2010 (n = 2,486).

Table 4
Individual characteristics of the participants who lived inside and outside the cluster with higher chance of type 2 diabetes mellitus and their association.Salvador, Bahia State, Brazil.ELSA-Brasil 2008/2010 (n = 1,849) * Adjusted for sex, age, race/skin color, schooling level, tobacco use, alcohol consumption, type of occupation, leisure physical activity, and waist circumference.Cad.Saúde Pública 2023; 39(5):e00138822

Table 5
Individual characteristics of the participants who lived inside and outside the cluster with low chance of type 2 diabetes mellitus and their association.Salvador, Bahia State, Brazil.ELSA-Brasil 2008/2010 (n = 1,849).

Table 6
Neighborhood socioeconomic environmental characteristics from census tracts of participants who lived inside and outside the clusters and their association.Belo Horizonte, Minas Gerais State and Salvador, Bahia State, Brazil.ELSA-Brasil 2008/2010.

Table 6 (continued)
Note: no participants were in this category living inside the cluster area of type 2 diabetes.* Ajusted for sex, age, race/skin color, schooling level, tobacco use, alcohol consumption, type of occupation, leisure physical activity, waist circumference, and each socioeconomic environmental characteristic.