Figures
Abstract
The outbreak of COVID-19 has led to there being a worldwide socio-economic crisis, with major impacts on developing countries. Understanding the dynamics of the disease and its driving factors, on a small spatial scale, might support strategies to control infections. This paper explores the impact of the COVID-19 on neighborhoods of Recife, Brazil, for which we examine a set of drivers that combines socio-economic factors and the presence of non-stop services. A three-stage methodology was conducted by conducting a statistical and spatial analysis, including clusters and regression models. COVID-19 data were investigated concerning ten dates between April and July 2020. Hotspots of the most affected regions and their determinant effects were highlighted. We have identified that clusters of confirmed cases were carried from a well-developed neighborhood to socially deprived areas, along with the emergence of hotspots of the case-fatality rate. The influence of age-groups, income, level of education, and the access to essential services on the spread of COVID-19 was also verified. The recognition of variables that influence the spatial spread of the disease becomes vital for pinpointing the most vulnerable areas. Consequently, specific prevention actions can be developed for these places, especially in heterogeneous cities.
Citation: Souza APGd, Mota CMdM, Rosa AGF, Figueiredo CJJd, Candeias ALB (2022) A spatial-temporal analysis at the early stages of the COVID-19 pandemic and its determinants: The case of Recife neighborhoods, Brazil. PLoS ONE 17(5): e0268538. https://doi.org/10.1371/journal.pone.0268538
Editor: Fausto Cavallaro, Universita degli Studi del Molise, ITALY
Received: October 6, 2021; Accepted: April 30, 2022; Published: May 17, 2022
Copyright: © 2022 Souza et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: The datasets analysed during the current study are available in the SEPLAG (Secretaria de Planejamento e Gestão de Pernambuco) repository, https://dados.seplag.pe.gov.br/apps/corona_dados.html.
Funding: This research was partially supported by the Universidade Federal de Pernambuco (UFPE- 06/2020) (website: https://www.ufpe.br). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Introduction
The World Health Organization (WHO) declared the coronavirus disease 2019 (COVID-19) pandemic in March 2020 [1]. It is a vascular disease caused by the severe acute respiratory syndrome coronavirus 2 (SARS-CoV2) virus, which perturbs multiple organ systems and prompts clinical manifestations in the lungs, heart, and kidneys, in particular [2,3]. In the early stages of the pandemic, the spread of the disease was mainly mitigated by non-pharmaceutical interventions [4], which were then complemented with vaccination programs [5] as technology evolved. However, changes of intervention measures [6] along with the emergence of highly contagious variants of the virus [7] have affected the rates and the patterns of the spread of COVID-19.
Spatial-temporal analysis might be a valuable tool to capture the dynamic behavior of the pandemic. One of the most regularly applied approaches in this context is spatial analysis, in which cluster and hotspot analysis, interpolation, and space–time scan statistics have been the main techniques used [8]. For instance, in China, spatial clustering patterns, over time, of cumulative cases of COVID-19 at the city [9] and county-levels [10] showed a shift from hotspots of larger coverage to few specific places. In global terms, evidence suggests that there is a sort of spread of COVID-19 cases from developed countries to less affluent countries [11].
South America saw its geographic centroid for COVID-19 cases generally shifting from west to east in Brazilian areas, up to January 2021 [12]. In Brazil, the centroids of cases and deaths started in São Paulo state (Southeast region), then progressively moved north until early May 2020 [13]. The COVID-19 epidemic increased rapidly across northeastern Brazil up to May 2020, at which time it was the metropolitan areas (mostly in the coastal areas) that were reported as having clusters of cases, even though there was an already relevant spread towards the countryside [14]. Understanding the transmission dynamics within local communities is important in order to develop strategies to combat the spread of the disease.
Additionally, another valuable analysis is investigating risk factors and determinants of COVID-19. Spatial regression tools have been used as a means to analyze the influence of these factors on the number of COVID-19 cases, incidence rate, the mortality rate and the case-fatality rate across space [15–17]. Some demographic indicators have been identified as crucial factors for the spreading of COVID-19, including the positive associations of cases of the disease with the elderly population [16,18,19], the percentage of people between 15 and 64 years-old [20], and the density of the population [16,19,21]. Different socio-economic factors are also found to positively impact COVID-19 incidence rates, such as income per capita [15], household income [16,22], life expectancy at birth [15], and the area deprivation index [18], whereas others affect them inversely including access to education [23]. Specifically, racial/ethnic minority relationships varied from positive with COVID-19 cases in regions of the USA [16,18] to negative with COVID-19 deaths in England [17]. In terms of health conditions, COVID-19 cases have positive associations with diabetes rates [16,19], while there are negative associations with smokers on the global scale [20] and with morbidity in South Korea [23]. Healthcare access [23] and available hospital beds [19] are also negatively related to the spread of COVID-19 cases. Finally, the built environment can be positively associated with COVID-19 cases with regard to the average height of buildings [21], and the density of commercial facilities and of roads [24], likewise negatively with the average street length [21].
Nonetheless, these relationships seem to vary according to the spatial scale, the study area and the phase of the outbreak [8]. On a global scale, population density, older populations, and household size are crucial predictors in early weeks of the reported spread of COVID-19, whereas the impacts of interpersonal contact and the globalization of trade are greater over time [11]. The mobility of the population also has a valuable role in this scenario, on varying spatial scales, since highly connected places [25] and areas with great service from bus stops and subway stations [26] showed a significant effect on the spread of the disease. Finally, an increase of mobility in the state of New York—due to the relaxation of restrictive measures—moved the COVID-19 clusters from counties of high population density to low ones [27]. Knowledge about the determinants and the community spread of the disease may support the development of specific sanitary strategies for the pandemic scenario. Strategies need to be created, evaluated, controlled and sustained by information on local characteristics.
In this regard, Brazil has been one of the most severely affected countries throughout the pandemic [12]. Brazil is a very large and complex country which is unequal in terms of public health and socio-economic matters, so there is not a unique explanation for the spatial spread of the virus across states [13]. Mortality risks tend to be higher for older men in Brazilian municipalities, whereas young adults and women are prone to higher risks of infection [28]. Also at the municipality-level in Brazil, a spatial regression model identified 13 indicators positively associated with the COVID-19 incidence rate, including the activity rate of people aged 10 to 14 years; the percentage of people aged six to 14 years who do not attend school; and the percentage of employed persons aged 18 or over who have completed elementary school [15]. In fact, the COVID-19 pandemic has been most severe in the poorest and most unequal regions of Brazil, such as the states of the Northeast region [6]. A recent study found that these states have suffered from a large number of infections and also a high mortality risk due to the deprived socioeconomic status of the poor and unsatisfactory health care conditions [28].
Considering the singular characteristic of the Northeastern states of Brazil, this study analyses the spatial patterns and driving factors of the spread of COVID-19 in the city of Recife, capital of Pernambuco state. So far, we found that previous studies about these topics in Recife were restricted to the association of the number of COVID-19 cases with the number of users of public transportation systems [29] and socio-economic factors [30]. Therefore, in this study, we adopted a three-stage methodology, in which we have investigated the spatial-temporal trend of the disease within the city and the impact on COVID-19 rates of local facilities that continued to function even during lockdown periods. Subsequently, we investigated the joint effect caused by socio-economic factors and the continuation of essential services on the spatial spread of COVID-19 confirmed cases and on the case-fatality rate on a neighborhood-level scale.
Data and methods
Design
This study develops a spatial evaluation to comprehend the behavior of the outbreak of coronavirus across the neighborhoods of Recife, Brazil. Statistical approaches (correlation, quartile and regression analysis) were combined with GIS-based methods (hotspots and spatial regression analysis) for the purpose of exploring local characteristics that may make places more susceptible to the spatial spread of COVID-19. A brief summary of the methodology is set out in Fig 1.
The city of Recife is the capital of the third most populated state in the Northeast region of Brazil (Pernambuco). Recife covers an area of 218 km2 and the estimated population is 1.55 million [31]. The city is located on the coast, in the Southern Hemisphere, and below the Equator. Recife has a tropical and humid climate, with an annual average temperature of about 26°C and a small variation of 5° C, approximately. It has the highest GDP per capita and is the second-most densely populated city in NE Brazil [31]. Moreover, the city had the highest Municipal Human Development Index (MHDI) among Northeast state capitals in 2010 (0.772), but it had just the 13th best performance out of 27 state capitals and was the 210th best city index nationally [32]. On the other hand, in 2019, Recife had the highest Gini index (0.612) among Brazilian state capitals, which reveals its heterogeneity and the extreme inequality of the distribution of income [33]. For instance, some neighborhoods are considered wealthy, but they also contain sizeable favelas. In some cases, a neighborhood is surrounded by others with almost the opposite socio-economic conditions even although they are in the same zone of the city [34]. There are two zones in Recife which have a very high HDI: one in the North and the other in the South. The latter mainly covers the Boa Viagem neighborhood. On the other hand, two-thirds of the districts (namely 42 out of 62) are classified as having either a low or average HDI [34]. Fig 2 shows the location of Recife and its neighborhoods.
Sources: Brazilian Institute of Geography and Statistics (IBGE) 2021, and software ArcGIS 10.4.1.
Data collection
The numbers of reported cases and data on deaths were obtained from the Planning and Management Secretariat of the state of Pernambuco, Brazil [35]. Daily records on Recife were provided at the city-level, but there was a lack of reliable data at finer scales. Georeferenced data were available only at the neighborhood-level, from 16 April to 3 July 2020, in cumulative form, and were published approximately once a week.
Socio-economic and demographic factors were selected to represent the main characteristics of the neighborhoods in terms of population structure and living conditions. In this regard, a set of 15 census indicators was extracted from the 2010 Brazilian Census [36], and is described in Table 1. We extended this set to characterize the impact of the presence of non-stop services (the so-called essential services) during the pandemic, since their locations have the potential for crowds to form and for premises to become overcrowded. So, six environmental facilities—that represent each of these services—were considered based on the classification proposed by the Planning and Management Secretariat of the state of Pernambuco, Brazil [37]. Georeferenced point data concerning these places were extracted from Google Maps and Google Earth platforms. These facilities are spatially described in Table 2 by means of Kernel density estimation [38]. However, they were analyzed throughout the paper in the form of counting in the polygons that delimit the neighborhoods of Recife.
Data analysis
An initial investigation presented Recife’s epidemiological situation regarding confirmed cases over time starting with when the first cases were confirmed in March 2020. A 7-day moving average was applied to the data on daily infections at the city-level, thereby seeking to reduce sudden variability due to reporting biases such as the lack of testing and the delay in recording cases and deaths [39]. The impact of restricting and relaxation measures imposed by the State Government of Pernambuco and the City Council of Recife on the variability of the rate of new infections was also discussed. These government actions include the following: closing of non-essential commercial activities; mandatory use of masks; strict quarantine; reopening of building supply stores, beauty salons, suburban retailers, malls, and places of worship [37].
In order to comprehend disparities in the spread of COVID-19 on a small scale, neighborhoods were explored using a spatial cluster analysis as applied for other diseases [40,41]. Georeferenced data obtained from ten specific days between 16 April and 3 July 2020 at intervals of roughly one week were considered. Nevertheless, we were able to represent the ascending, peaking and descending behaviors of the curve of infections.
At first, for each of the ten dates, we checked if the cumulative confirmed COVID-19 cases and the case-fatality rate (which means the total number of deaths divided by the total number of cases) were spatially dependent across the study area. In this regard, the spatial statistic Global Moran’s I [42,43] was performed considering spatial weights obtained from inverse distance and contiguity of edges and borders among neighborhoods. Then, Local Moran’s I [44] was implemented for cumulative confirmed cases and the case-fatality rate—which means the total number of deaths divided by the total number of cases. This statistic assessed the spatial autocorrelation associated with each neighborhood of the study region in terms of a few surrounding spatial units. It was calculated for each area i = 1,…,n in: (1) where y means the COVID-19 reported cases or case-fatality rates for the ith area or its jth neighboring areas, and Wij means a weight that represents proximity for the pair of areas i and j (in this case, the inverse distance between them). As a result, the location of statistically significant spatial clusters and outliers could be identified, represented in maps and further explored. These clusters are called High-High (high values surrounded by high values; or hotspots) and Low-Low (low values surrounded by low values), whereas the outliers are the High-Low (high values surrounded by low values) and Low-High (low values surrounded by high values) entities [45].
We also investigate patterns of driving factors within spatial clusters and outliers to understand the spread of COVID-19 in the study area. Thus, we emphasized distinctive community characteristics associated with highlighted clusters, at different stages of the pandemic. The investigation of the determinants aims to test if the pattern of the spread of COVID-19 in Recife follows the global pattern [11]. Moreover, we explore the impact of the local demography, expecting that the population [16,19,21] and the economically active population [20] factors would be highly associated with the occurrence of the disease. Finally, as COVID-19 cases are positively associated with the density of commercial facilities [24], we expect an analogous behavior regarding the presence of essential services.
For this purpose, 15 socio-economic factors were submitted to quartile analysis, and Spearman’s rank correlation tests were conducted in relation to case-fatality rates only. Additionally, the relevance of the dataset of 6 essential services was ratified due to its linear connection with case-fatality rates noticed in scatterplots, which justified the use of a Pearson’s product-moment correlation.
Subsequently, we have performed several regression analyses. Regression methods are widely applied in health-related studies [15–22,46,47]. The Ordinary Least Squares (OLS) regression technique [48] was used to reveal the strength of the relationships between the dependent and the most significant explanatory variables throughout different stages of the pandemic. A regression analysis was applied using as a dependent variable only the cumulative data on COVID-19 infections gathered from the latest date then available, July 3rd, 2020. To further our exploratory concern, initially, essential services and socio-economic factors were treated separately, thus assuring that relevant relationships among explanatory variables from each dataset would not be discarded. An OLS regression was run twice, one per dataset. Correlated variables were eliminated when they had a variance inflation factor (VIF) greater than 7.5 as previously verified in other COVID-19 health studies [49,50]. We used a stepwise method based on the Akaike Information Criterion (AIC) in order to reduce both sets of determinants to their non-redundant cores [51].
An extended investigation was undertaken by performing regression analysis for both datasets, socio-economic and non-stop services facilities, as a way of understanding their likely synergistic effect for predictions. The OLS method was adopted, but now considering the same ten dates from which georeferenced data on infections at the neighborhood-level were available. The reason is to understand what determinants became statistically significant over time as the pandemic evolved and government measures were in force. In this study, multi-source data can be used since we have complementary full datasets for the populations studied [52], on which a one-to-one linkage procedure was applied for complete observations, instead of sample-based datasets [53].
A k-fold cross-validation procedure [54] was applied to evaluate possible overfitting of the OLS regression models. The dataset of COVID-19 cases regarding the neighborhoods was split into randomly selected k = 5 folds of similar size. A fold of the data was held-out for final validation (testing), while the remaining k-1 formed the training set (learning). For these remaining subsets, k-1 iterations of training and validation were performed on them and the RMSE (root mean square error) was measured for each fold. The average RMSE of these iterations was compared to the same metric of the initially held-out fold. If the results are found to be close according to the context, i.e. a maximum difference of around 4 units, then there is no significant overfitting.
We are dealing with geospatial data, so Geographically Weighted Regression (GWR) [55] was used as a way of relaxing OLS assumptions that the observations and the error terms are independent and constant over the study region [45]. GWR takes into account the spatial autocorrelation and allows relationships among variables to vary over space and to be determined for each location [55]. In this method, a kernel function with a bandwidth parameter is used to calculate a local weights matrix in terms of the distance between each pair of spatial units [56]. Considering i = 1,…,n as each sample location, GWR is mathematically denoted by [56]: (2) where yi is the dependent variable of COVID-19 confirmed cases at neighborhood i, xik is the value of the kth explanatory variable at location i, βi0 is the intercept, βik is the regression coefficient for the kth explanatory variable, p is the number of regression terms, and εi is the random error at location i. Cumulative count data has been applied in studies of COVID-19, especially at a small-scale level such as neighborhood or county-level [57] and grid [58]. The study of [59] predicted cumulative confirmed and cured cases of COVID-19 at a province-level, while [60] considered the number of deaths. Furthermore, the use of count data as a dependent variable instead of some sort of rates at the neighborhood-level is advocated by [61], and it is widely applied in the context of studies of crime [62].
OLS and GWR statistical performances were compared, thereby taking account of the reduced set of determinants found by using data acquired from July 3rd, 2020. Then we determined how GWR outputs explain the spread of COVID-19 in each neighborhood, and we also elucidated how every relevant factor impacts on the prominence of hotspots for new cases. According to the spatial context, an adaptive Gaussian kernel function was chosen with a view to adjusting the weighting for the density of infections. The kernel bandwidth parameter was estimated by means of a corrected Akaike Information Criterion (AICc) approach, based on which the optimal number of neighboring areas was found. Finally, in order to check if there is no spatial dependence on GWR residuals, Global Moran’s I [42,43] was applied to them. The weighting was once again based on the contiguity of edges and borders among neighborhoods.
Spatial analysis was conducted on ArcGIS 10.4.1 software, while the statistical tests and OLS regressions were performed on the R 3.6.1 platform.
Results
An evolution of cases over time in the city of Recife was plotted on Fig 3, including relevant actions taken by the State Government and the City Council. It was noted that these authorities took an early decision to close facilities when first cases were reported. Only a few types of facilities were allowed to open, including supermarkets, grocery stores, bakeries, and pharmacies [37]. As daily infections were increasing, the state governor issued an edict obliging the population to wear masks in public places. However, as fines were not imposed on shoppers but on the owners of the commercial facilities they entered, adherence to the measure became more dependent on the willingness of the general public to wear masks and intense supervision at the entrances to commercial premises.
These measures were not sufficient to flatten the increase in cases of infection, so in mid-May, a 15-day strict quarantine was imposed in five municipalities in Pernambuco, including Recife. People were only allowed to leave their homes to seek essential services, for which they had to show proof, and vehicles were only allowed on roads according to a vehicle rotation system based on the final number of the number plate [34]. Transmissions of the virus peaked in late May 2020 and the cases seemed to have been stabilized at a low level for that moment (July 8th, 2020), even though restrictions were then gradually relaxed and places in which people congregate such as shopping malls and commercial premises were reopened.
Spatial clustering analysis of cumulative COVID-19 cases and case-fatality rate
A significant spatial dependence (p < 0.05) was found in Recife neighborhoods for the case-fatality rate on the majority of dates, according to Global Moran’s I, with an increasing tendency of significance as the pandemic evolved. On the other hand, the clustering of reported cases was significant for the first date analyzed and the last two dates. Consequently, spatial clusters for cases and the case-fatality rate tend to be formed across the study region.
The locations of these clusters were found by means of Local Moran’s I. The results for the total number of reported cases per neighborhood are shown in Fig 4. We noted that cases of the disease at first were concentrated in the South Zone of the city, specifically in the neighborhood called Boa Viagem (Fig 1). Furthermore, it was the only significant spatial cluster on April 23rd, 2020. As cases increased, other High-High clusters and High-Low spatial outliers were found in the South, the North and the West Zones of the city.
Sources: Brazilian Institute of Geography and Statistics (IBGE) 2021, and software ArcGIS 10.4.1.
Moreover, a quartile analysis was developed in the entire city using the 15 census indicators [36] that are listed and described in Table 1 (see Section 2 above). This analysis concerned the socio-economic and demographic factors that define the residents’ environment in terms of average income, population size, extent of home ownership, government support and by age group. Results revealed social disparities in the clusters shown in red in Fig 4 compared to Boa Viagem. While Boa Viagem is highly populated and yet well-developed in terms of its access to sanitation, garbage collection, and of levels of income and literacy, the other hotspots were found to have indices for these variables that were lower than at least 50% of all other neighborhoods in Recife.
Case-fatality rate (CFR) hotspots due to COVID-19 were also examined as plotted and shown in Fig 5. A different spatial pattern arises when these are compared to the clusters of cases. Over time, the CFR ones remained mostly stable so that High-High clusters usually formed in the North and Southwest zones, while Low-Low ones were seen in the North and West zones.
Even though Boa Viagem has been a significant cluster for reported cases since the first date analyzed, this has not occurred for the case-fatality rate. This neighborhood presents a high number of cases, but the CFR has been growing in a lower proportion. The first death in Boa Viagem was confirmed on April 10th, 2020 after 63 cases had been recorded, which represented 15% of the COVID-19 cases in Recife at that time. Boa Viagem led the number of deaths in the city from April 27th until July 8th, 2020. However, this neighborhood did not become the hotspot of CFR in any of the transitional periods, i.e., Boa Viagem had the highest number of cases but it has had a low number of deaths per cases. On July 3rd, 2020 the neighborhood reached its maximum CFR, 18%, which, however, is considerably lower than the hotspots of CFR, for which the percentages were between 34% and 54%.
The results from quartile analysis show that hotspots of case-fatality rate usually present areas with similar environmental characteristics to those in hotspots of confirmed cases, disregarding Boa Viagem. For instance, most of these areas have a precarious public service provision and a low-income population. The opposite characteristics were verified when analyzing the Low-Low clusters. It has also been observed that neighborhoods in Low-Low clusters (of both case and case-fatality rate) have fewer residents per household than 75% of all other neighborhoods.
Spearman correlation tests were applied to the case-fatality rate and socio-economic factors using data from July 3rd, 2020. We detected that CFR is positively associated with residents per household, whereas it is negatively associated with household income, the literacy rate, access to the sewage system and to garbage collection, and the total of people over 60 years old.
Comparing previous results, we noted that some places tend to suffer from fewer deaths due to COVID-19 when residents in such places have a level of income and literacy that is well above average for Recife. These wealthier areas also have more access to public services and their number of residents per household is lower than elsewhere in Recife. In contrast to what was expected, the results also showed that the elderly population was higher in Low-Low clusters of case-fatality. In other words, although the elderly are more prone to catching severe forms of COVID-19 [63], the incidence of such cases was less in clusters with a high concentration of this population.
We also examined the presence of essential services (Table 2) in spatial clusters of case-fatality from July 3rd, 2020 by means of Pearson’s product-moment correlation. After exploring the Low-Low clusters for the case-fatality rate, the ‘number of banks’ was the only factor significantly associated with the CFR (negative). Six out of ten neighborhoods were placed in the fourth quartile of this determinant.
Association between environmental factors and COVID-19 cases
Local determinants for COVID-19 were explored in greater depth by using spatial regression analysis. A set of 15 explanatory variables was compiled from census indicators (Table 1), whereas places that are typically conducive to attracting crowds of people and that operated even during strict quarantine (Table 2)—and so could have become centers of SARS-CoV2 infections—formed another set.
As an initial exploration, the determinants were processed separately in two regression models, according to their datasets of origin. We used the cumulative data of COVID-19 confirmed cases from July 3rd, 2020 as a response variable in each model based on the OLS method and reduced the number of determinants to non-redundant sets. Both regression models were significant (p < 0.001). With regard only to essential services, bakeries, grocery stores, banks and pharmacies remained in the final set of determinants with a high adjusted R2 of 0.8659. It was noted that ‘bakeries’ is associated with 77% of the variability in confirmed cases. Considering only the socio-economic factors, five determinants formed the significant final set (R2 of 0.9521): ‘people between 0 and 9 years old’ and ‘people older than 60 years old’ (extreme opposites among the age groups), ‘piped water’, ‘garbage collection’ and ‘residents per household’. The total of residents (factor ‘population’) is associated with nearly 90% of the variability in COVID-19 cases.
New assessments were made based on an aggregated set of 21 driving factors, regardless of the origins of this data, to understand how they can interact throughout space. OLS regression models were built for ten different days between April and July 2020, taking the cumulative reported cases of COVID-19 as the dependent variable. A smaller and significant set of explanatory variables was identified for each day by excluding correlated variables and using the stepwise method. Findings are summarized in Table 3 and, although determinants were analyzed together, the final sets were shown separately to clarify patterns. The column called GM, government measures, indicates which decisions were being imposed on each date by local authorities in their attempt to gradually reduce the number of people on the streets or to enable more people to be on the streets.
Every designed model was found significant (p < 0.001). The adjusted determination coefficient R2 value reached a high level from the first date analyzed. As time progressed in the pandemic, the correlation of demographic and socioeconomic factors to cumulative cases increased, which prompted the same pattern for the adjusted R2. For instance, evaluating the contribution of each determinant in the final model (Table 3) separately in April 16th, ‘bakeries’ is associated with 76% of the variability in cases, followed by lottery shops (47%) and ‘grocery stores’ (39%). This same procedure was conducted in July 3rd, resulting in ‘children from 0 to 9 years old’ being associated with 80% of the variability in cases, followed by ‘bakeries’ (77%) and ‘pharmacies’ (73%). However, this increasing tendency for the adjusted R2 remained only until early June 2020, and was followed by a stabilisation pattern. It is likely this change was due to the scenario of the slow growth of cases in Recife as seen in Fig 3.
Even though the adjusted R2 metrics are high, it is not justified by overfitting. This hypothesis was discarded after each OLS model had been cross-validated based on partitions of the COVID-19 cases dataset and a comparison of RMSE (root mean square error) indexes. As ‘bakeries’ and/or the children’s age group are part of the models from the first date analyzed, they appear to be one of the main sources of the high adjusted R2. Other determinants are frequently repeated in the final sets, including ‘income’ and ‘grocery stores’, which reveals their importance for predicting cases. Furthermore, the variables ‘people older than 60 years old’ and ‘population’ (both highlighted in the initial analysis using separated databases) were disregarded after reducing the set of determinants to their non-redundant cores. This likely happened due to their strong collinearity with ‘bakeries’, namely a Pearson’s correlation of 0.85 (p < 0.001) for the total of residents and 0.93 (p < 0.001) for the elderly. Conclusively, all of the highlighted determinants in Table 3 (except for ‘bus terminals’) and the factors ‘population’ and ‘people aged over 60’ are positively associated with COVID-19 cases.
Spatial associations between environmental factors and COVID-19 cases.
Data from the last day from which there is available georeferenced data at the scale studied, July 3rd, 2020, were kept for a subsequent evaluation using a spatial regression approach, GWR. So, the following set of relevant explanatory variables was considered: number of grocery stores, number of pharmacies, number of bakeries, the average income of residents and the total number of residents aged 0 to 9 years old. We used the cumulative confirmed cases as the dependent variable. According to the AICc approach, the optimal number of 54 neighboring areas per neighborhood was taken as the bandwidth parameter to compute the spatial weights. Finally, the GWR model resulted in a R2 of 0.960. A Global Moran’s I test applied to GWR residuals showed an index of -0.028 (p = 0.63), which provided evidence that the residuals are randomly distributed, and thus the model is adequate. Note that there is a slight improvement from the OLS results (global analysis) to those of the GWR (local analysis) since adjusted R2 increased from 0.926 to 0.944. Also, AIC statistics reduced from 935.99 to 916.19, indicating a significant improvement in the quality of the model [64].
GWR concerns a local prediction to elucidate spatial variations all over the region of interest [55], so the distribution of local R2 in each Recife neighborhood is illustrated in Fig 6. Values were found to be remarkably high since the minimum one explains 82.8% of reported cases. It was also observed that the areas with a higher local performance (the red ones) coincide with hotspots for cases (identified in Fig 4), indicating where COVID-19 infections are concentrated.
Sources: Brazilian Institute of Geography and Statistics (IBGE) 2021, and software ArcGIS 10.4.1.
A further exploration was made seeking to clarify how every relevant contribution of a determinant to spatial regression modelling could influence a prediction of the number of cases. Fig 7 classifies their coefficients on regression local equations into five categories by the Jenks Natural Breaks algorithm [65]. When the previously mentioned hotspots are analyzed, it is noted that the Southern ones were impacted most by the presence of bakeries and their residents’ average income. The existence of the Western cluster is best explained by the presence of local pharmacies and grocery stores. Moreover, the strongest influence on the incidence of COVID-19 in the Northern hotspot came from bakeries, pharmacies and children aged from 0 to 9.
Sources: Brazilian Institute of Geography and Statistics (IBGE) 2021, and software ArcGIS 10.4.1.
Discussion
Recife started applying deterrent measures in order to reduce contagion immediately after the first patients were confirmed in March 2020, thereby preventing a collapse in the provision of hospital care. After the first registered infection, cases took around a month to start a phase of sustained increase, which may have indicated an initial acceptance by the general public of these measures. Then a natural relaxation due to fatigue and ignorance about the consequences of COVID-19 was reflected in a wider local transmission of the virus. The public authorities’ initiative to tighten the quarantine had excellent results. That could be verified as the peak of contamination was reached when quarantine was still in force, after which the tendency was for the number of new cases to fall and stabilize.
Relationships concerning the explored determinants imply that some neighborhoods are generally more susceptible to the spread of COVID-19, essentially due to the direct influence of specific socioeconomic and environmental scenarios–that might be worsened when combined. COVID-19 cases are positively related to better socioeconomic conditions (household income and public services), a large population (mainly children and elderly) and the presence of some essential services (especially those connected to daily routine, such as bakeries and grocery stores). In this context, the size of the population acts as an enabler of more social contact even when social distancing measures are in force, which reiterates what was found by [16,21]. As to reported cases, hotspots were first verified in the wealthy and densely-populated neighborhood of Boa Viagem. In contrast, since then, other hotspots with worse socio-economic conditions have emerged. A similar pattern was found in the city of São Paulo, Brazil, but concerning COVID-19 deaths: what was detected was a shift of high risk from the areas with the best socio-economic conditions to those with the worst conditions [66]. This situation could have happened in Recife because, according to socio-economic indicators, a substantial proportion of the population in the Boa Viagem hotspot had the resources to travel more frequently to Brazilian metropolises, or even to other countries. Thus, these people were the first to be infected elsewhere, and, on their return to Recife, they might have contributed to spreading the disease to those around them. On the other hand, people who live in less privileged places consequently have less infrastructure in their neighborhoods, nor do their living conditions in their homes enable them to follow advice on social isolation and personal care. This includes not having enough money to buy preventive health supplies; there being a lack of constant access to piped water; and they do not have the option of working from home. All of these factors are aggravated by the social impacts of COVID-19 [67].
Bakeries, grocery stores and pharmacies seem to strongly influence the spatial spread of COVID-19 as observed in the regression analysis results. These findings were similar to a previous study in a prefecture in China, where these points-of-interest influenced the increase of the COVID-19 cluster size in the surrounding neighborhoods [68]. Going to and entering bakeries and grocery stores are part of the daily routine of regular citizens in the metropolitan region of Recife, because there is a local habit of buying fresh bread every day in the morning or late afternoon. According to a study in the city of São Paulo, Brazil, people who live in areas with a great mix of destinations within 500 m of their residences, including supermarkets, food stores and bakeries, were more inclined to walk outside home [69]. In a COVID-19 pandemic context, even during strict quarantine restrictions, residents still have to buy primary groceries and medicines, probably near to where they live. This is particularly true in low-income neighborhoods of Recife since a significant part of their population cannot afford public transportation, which restricts their routine to places they can reach on foot (or riding bicycles at most) [70]. The number of these commercial facilities is significant for predicting COVID-19, but per se this does not necessarily imply longer lines or crowded spaces that help to transmit SARS-CoV2. This consequence also depends on the level of demand at specific times and may be affected by the population of the neighborhood. Finally, the strong collinearity between the number of bakeries and residential population indicates that these facilities are located where the people are.
The age groups highlighted during the regression analysis reveal that places with a large number of children (between 0 and 9 years old) and/or seniors (over 60 years old) tend to present the largest number of reported cases. A similar result was found regarding the population aged above 65 at the global scale, specifically in the early weeks of the outbreak [11]. Studies affirm that most asymptomatic cases of COVID-19 are verified in children [71], so there is a high chance of their not being submitted to tests in Brazil because there is a lack of testing capacity [72]. Moreover, schools and daycare centers had been closed in Recife since mid-March 2020 [52], which stimulated those younger groups to stay at home and, therefore, it was likely that they would spread the virus to their relatives. But studies from China show that children have a lower incidence of SARS-CoV2 and are less prone than other groups to being infected by it [73]. Hence, a more in-depth exploration needs to be carried in order to determine whether this pattern also happened in Brazil, even though our results imply the opposite. On the other hand, there should be a focus on the elderly since a study specified that an increase in coronavirus infection among elderly people had a direct correlation with the risk of infections among other age groups [74]. Therefore, tightening social distancing for the elderly and other measures, such as analyzing spatial accessibility and healthcare resources [75], to reduce the risks they face could positively affect the whole of society. Furthermore, comorbidities are associated with a higher risk of severe cases of COVID-19 that demand specialized clinical care [76], so Brazilians aged over 60 need to be taken into account since they have almost 12 times higher odds of developing multiple chronic diseases than young adults [77].
Government measures to control the transmission of the virus were imposed in advance, less than a week after the first reported COVID-19 case in the state of Pernambuco. Notwithstanding, their mild severity in addition to a growing lack of society support were some of the reasons why the curve of infections did not flatten sooner. Then a 15-day milder version of the lockdown adopted in other countries was implemented in mid-May 2020, which led to a sustained tendency of incidences to fall and stabilize at a lower level. A lockdown is effective in reducing the number of new cases, particularly when it lasts for at least 10 days [78]. Our findings reaffirm what a previous study found about the positive effects of the strict quarantine in Pernambuco, which helped to increase the number of people who adopted and maintained social distancing and to reduce the reproduction rate of the virus [6].
Some factors were considered relevant for the majority of the dates examined regarding the evolution of COVID-19 cases, but were cut in regression analysis from the most recent set of determinants after the quarantine period. That was the case of lottery shops, owned home and literacy, and all of them positively affected the variability of reported cases. Lottery shops, which also act as sub-agencies of a public bank, have been used along with Caixa Econômica Federal bank agencies to make the payment of emergency aid to at least 25% of the Brazilian population since April 2020 [79]. So, the presence of these facilities could have acted as a way for promoting long queues with social distancing not being respected and physical overcrowding between socially vulnerable people [80]. For their part, literate people tend to have more information about the disease, but we cannot assume that they consider this subject as weighty and/or they may not have the resources to follow all the recommendations on isolation. This direct association of literacy diverges from other studies [15], so that our findings seem to give a distinctive characteristic of Recife’s neighborhoods. Finally, ownership of one’s own home does not point up a clear social difference among the hotspots of COVID-19 cases. This determinant was found statistically significant considering data gathered on the first and last dates examined. So it remained relevant in hotspots of cases even though the disease advanced from privileged areas to others that are historically known for under-privileged living conditions [34].
Confirmed cases remained stable or in decline for a while during the period evaluated—even after restrictive measures were relaxed -, which denoted a possible control of the pandemic. Nevertheless, at the end of 2020, Brazilian cities suffered a second wave of the COVID-19 outbreak. This was intensified as new SARS-CoV2 variants with greater transmission power started to spread worldwide [7]. Further studies should verify whether the patterns identified in the first wave of infections in 2020 have been maintained, and they should also consider the pace of vaccination campaigns.
This paper has some limitations since there is significant underreporting of COVID-19 cases and deaths in Brazil due to the limited availability of tests, and the capacity of local surveillance services. Further analysis should include more recent data to represent socio-economic characteristics, in addition to the previous health status of infected people regarding comorbidities. Finer variations in the behavior of the disease could also be captured by exploring spatial units that are even smaller than neighborhoods, such as census tracts. Another limitation refers to the influence of possible confounded variables and other extraneous variables in the model, although there is no consensus in the literature about the best strategy for dealing with them [81]. According to [82], it is equally possible that adding control variables introduces overcontrol and endogenous selection biases, thus creating alternative interpretations rather than ruling them out. Additionally, it is known that it is difficult, if not impossible, to include a comprehensive list of all factors influencing the spread of COVID-19 in a community. Thus, we believe that our findings outweigh these limitations.
Conclusion
This study combined spatial clusters and statistical analysis to evaluate the influence of socio-economic factors and essential services on the spread of COVID-19 (in terms of reported cases and case-fatality) in the city of Recife, Brazil. Our findings reveal that an increased risk of transmissions was associated with children and the elderly, the size of the population, household income, the level of education, and the presence of some facilities that have remained open throughout the pandemic. Moreover, the spatial spread of the disease occurred by moving from well-developed to deprived neighborhoods during the initial stages of the pandemic. What was also found was for there to have been a tendency for there to have been harsh impacts (due to higher case-fatality rates) on socially vulnerable and densely populated communities, specially those with many everyday places that are prone to overcrowding (e.g. bakeries, grocery stores).
Brazil manages a public health system that is widespread in all federative units, even though the country has continental dimensions and faces complex challenges. However, currently, public agencies have been going through management difficulties. In this sense, this study can support strategic decisions to help mitigate the spread of COVID-19 not only in Brazil, but also in other developing and economically emerging countries. Furthermore, in the long term, knowledge produced during the COVID-19 pandemic in this heterogeneous context, regarding local characteristics and spatiotemporal patterns, can be used to structure policies for tackling new epidemics of viral infectious diseases.
Acknowledgments
The authors are grateful to SEPLAG for making the data in their repository available and to Roderick Kay for his professional support in providing language editing services.
References
- 1. World Health Organization. WHO Director-General’s Opening Remarks at the Media Briefing on COVID-19. 2020 [cited 25 March 2022]. Available from: https://www.who.int/director-general/speeches/detail/who-director-general-s-opening-remarks-at-the-media-briefing-on-covid-19—11-march-2020.
- 2. Siddiqi HK, Libby P, Ridker PM. COVID-19 –A vascular disease. Trends Cardiovasc Med. 2021;31: 1–5. pmid:33068723
- 3. Roberts KA, Colley L, Agbaedeng TA, Ellison-Hughes GM, Ross MD. Vascular Manifestations of COVID-19 –Thromboembolism and Microvascular Dysfunction. Front Cardiovasc Med. 2020;7: 215. pmid:33195487
- 4. Adhikari SP, Meng S, Wu YJ, Mao YP, Ye RX, Wang QZ, et al. Epidemiology, causes, clinical manifestation and diagnosis, prevention and control of coronavirus disease (COVID-19) during the early outbreak period: A scoping review. Infect Dis Poverty. 2020;9: 29. pmid:32183901
- 5. Bartsch SM, O’Shea KJ, Wedlock PT, Strych U, Ferguson MC, Bottazzi ME, et al. The Benefits of Vaccinating With the First Available COVID-19 Coronavirus Vaccine. Am J Prev Med. 2021;60: 605–613. pmid:33632650
- 6. Kerr L, Kendall C, da Silva AAM, Aquino EML, Pescarini JM, de Almeida RLF, et al. COVID-19 in Northeast Brazil: achievements and limitations in the responses of the state governments. Cien Saude Colet. 2020;25: 4099–4120. pmid:33027347
- 7. Moore JP, Offit PA. SARS-CoV-2 Vaccines and the Growing Threat of Viral Variants. JAMA. 2021;325: 821. pmid:33507218
- 8. Fatima M, O’keefe KJ, Wei W, Arshad S, Gruebner O. Geospatial Analysis of COVID-19: A Scoping Review. Int J Environ Res Public Heal. 2021;18: 2336. pmid:33673545
- 9. Ma Q, Gao J, Zhang W, Wang L, Li M, Shi J, et al. Spatio-temporal distribution characteristics of COVID-19 in China: a city-level modeling study. BMC Infect Dis. 2021;21: 1–14. pmid:33390160
- 10. Xiong Y, Wang Y, Chen F, Zhu M. Spatial statistics and influencing factors of the COVID-19 epidemic at both prefecture and county levels in Hubei Province, China. Int J Environ Res Public Health. 2020;17: 3903. pmid:32486403
- 11. Sigler T, Mahmuda S, Kimpton A, Loginova J, Wohland P, Charles-Edwards E, et al. The socio-spatial determinants of COVID-19 diffusion: the impact of globalisation, settlement characteristics and population. Global Health. 2021;17: 1–14. pmid:33397377
- 12. Huang Z. Spatiotemporal Evolution Patterns of the COVID-19 Pandemic Using Space-Time Aggregation and Spatial Statistics: A Global Perspective. ISPRS Int J Geo-Information. 2021;10: 519.
- 13. Castro MC, Kim S, Barberia L, Ribeiro AF, Gurzenda S, Ribeiro KB, et al. Spatiotemporal pattern of COVID-19 spread in Brazil. Science. 2021;372: 821–826. pmid:33853971
- 14. Gomes DS, Andrade LA, Ribeiro CJN, Peixoto MVS, Lima SVMA, Duque AM, et al. Risk clusters of COVID-19 transmission in northeastern Brazil: prospective space–time modelling. Epidemiol Infect. 2020;148: E188. pmid:32829732
- 15. De Souza CDF, Machado MF, Do Carmo RF. Human development, social vulnerability and COVID-19 in Brazil: A study of the social determinants of health. Infect Dis Poverty. 2020;9: 124. pmid:32867851
- 16. Ali T, Mortula M, Sadiq R. GIS-based vulnerability analysis of the United States to COVID-19 occurrence. J Risk Res. 2021;24: 416–431.
- 17. Sun Y, Hu X, Xie J. Spatial inequalities of COVID-19 mortality rate in relation to socioeconomic and environmental factors across England. Sci Total Environ. 2021;758: 143595. pmid:33218796
- 18. Oluyomi AO, Gunter SM, Leining LM, Murray KO, Amos C. COVID-19 community incidence and associated neighborhood-level characteristics in Houston, Texas, USA. Int J Environ Res Public Health. 2021;18: 1495. pmid:33557439
- 19. Mansour S, Al Kindi A, Al-Said A, Al-Said A, Atkinson P. Sociodemographic determinants of COVID-19 incidence rates in Oman: Geospatial modelling using multiscale geographically weighted regression (MGWR). Sustain Cities Soc. 2021;65: 102627. pmid:33288993
- 20. Iyanda AE, Adeleke R, Lu Y, Osayomi T, Adaralegbe A, Lasode M, et al. A retrospective cross-national examination of COVID-19 outbreak in 175 countries: a multiscale geographically weighted regression analysis (January 11-June 28, 2020). J Infect Public Health. 2020;13: 1438–1445. pmid:32773211
- 21. Kwok CYT, Wong MS, Chan KL, Kwan MP, Nichol JE, Liu CH, et al. Spatial analysis of the impact of urban geometry and socio-demographic characteristics on COVID-19, a study in Hong Kong. Sci Total Environ. 2021;764: 144455. pmid:33418356
- 22. Maiti A, Zhang Q, Sannigrahi S, Pramanik S, Chakraborti S, Cerda A, et al. Exploring spatiotemporal effects of the driving factors on COVID-19 incidences in the contiguous United States. Sustain Cities Soc. 2021;68: 102784. pmid:33643810
- 23. Weinstein B, da Silva AR, Kouzoukas DE, Bose T, Kim GJ, Correa PA, et al. Precision mapping of COVID-19 vulnerable locales by epidemiological and socioeconomic risk factors, developed using South Korean data. Int J Environ Res Public Health. 2021;18: 604. pmid:33445701
- 24. Wang J, Wu X, Wang R, He D, Li D, Yang L, et al. Review of Associations between Built Environment Characteristics and Severe Acute Respiratory Syndrome Coronavirus 2 Infection Risk. Int J Environ Res Public Heal. 2021;18: 7561. pmid:34300011
- 25. Cuadros DF, Xiao Y, Mukandavire Z, Correa-Agudelo E, Hernández A, Kim H, et al. Spatiotemporal transmission dynamics of the COVID-19 pandemic and its impact on critical healthcare capacity. Health Place. 2020;64: 102404. pmid:32736312
- 26. Lak A, Sharifi A, Badr S, Zali A, Maher A, Mostafavi E, et al. Spatio-temporal patterns of the COVID-19 pandemic, and place-based influential factors at the neighborhood scale in Tehran. Sustain Cities Soc. 2021;72: 103034.
- 27. Zheng A, Wang T, Li X. Spatiotemporal Characteristics and Risk Factors of the COVID-19 Pandemic in New York State: Implication of Future Policies. ISPRS Int J Geo-Information. 2021;10: 627.
- 28. Campos de Lima EE, Gayawan E, Baptista EA, Queiroz BL. Spatial pattern of COVID-19 deaths and infections in small areas of Brazil. PLoS One. 2021;16: e0246808. pmid:33571268
- 29. da Silva CFA, Meira LH, de Oliveira LK, dos Santos PRG, de Oliveira IK. Analysis of spatial correlation between public transportation system users and covid-19 cases: A case study in Recife (PE). Rev Bras Gestão e Desenvolv Reg. 2020;16: 346–359.
- 30. Quinino LR de M, Vasconcellos FHM, Diniz IS, Aguiar LR, Ramos YT de M, De Bastiani F. Aspectos espacial e temporal e fatores associados à interiorização da Covid-19 em Pernambuco, Brasil. Cien Saude Colet. 2021;26: 2171–2182. pmid:34231729
- 31.
IBGE. Cidades e Estados. 2020 [cited 29 March 2022]. Available from: https://www.ibge.gov.br/cidades-e-estados/pe.html.
- 32.
PNUD. IDHM Municípios 2010. [cited 16 March 2022]. Available from: https://www.br.undp.org/content/brazil/pt/home/idh0/rankings/idhm-municipios-2010.html.
- 33.
SIS. Síntese de Indicadores Sociais 2020. In: Síntese de Indicadores Sociais 2020 [Internet]. 2020 [cited 9 May 2021]. Available from: https://www.ibge.gov.br/estatisticas/sociais/saude/9221-sintese-de-indicadores-sociais.html?=&t=%0Adownloads.
- 34. Pereira DV e. S, Mota CM de M. Human Development Index Based on ELECTRE TRI-C Multicriteria Method: An Application in the City of Recife. Soc Indic Res. 2016;125: 19–45.
- 35.
SEPLAG. COVID-19 no Mundo, no Brasil e em Pernambuco. 2020 [cited 27 May 2020]. Available from: https://dados.seplag.pe.gov.br/apps/corona.html#painel.
- 36.
IBGE. Censo 2010. 2016 [cited 29 March 2022]. Available from: https://censo2010.ibge.gov.br/resultados.
- 37.
PHS. Pernambuco contra a Covid-19. 2020 [cited 29 March 2022]. Available from: https://www.pecontracoronavirus.pe.gov.br/.
- 38.
Silverman BW. Density estimation for statistics and data analysis. London: Chapman Hall; 1986.
- 39. Buckingham-Jeffery E, Morbey R, House T, Elliot AJ, Harcourt S, Smith GE. Correcting for day of the week and public holiday effects: improving a national daily syndromic surveillance service for detecting public health threats. BMC Public Health. 2017;17: 1–9. pmid:28049454
- 40. Kulldorff M, Huang L, Konty K. A scan statistic for continuous data based on the normal probability model. Int J Health Geogr. 2009;8: 58. pmid:19843331
- 41. Francis SS, Enders C, Hyde R, Gao X, Wang R, Ma X, et al. Spatial–Temporal Cluster Analysis of Childhood Cancer in California. Epidemiology. 2020;31: 214–223. pmid:31596791
- 42. Moran PAP. The Interpretation of Statistical Maps. J R Stat Soc Ser B. 1948;10: 243–251.
- 43.
Wong DW., Lee J. Statistical Analysis of Geographic Information with ArcView GIS and ArcGIS. Hoboken, USA: John Wiley & Sons, Ltd; 2005.
- 44. Anselin L. Local Indicators of Spatial Association-LISA. Geogr Anal. 2010;27: 93–115.
- 45.
Fischer M., Wang J. Spatial data analysis: models, methods and techniques. Berlin/Heidelberg, Germany: Springer Science & Business Media; 2011.
- 46. Gigerenzer G, Gaissmaier W, Kurz-Milcke E, Schwartz LM, Woloshin S. Helping Doctors and Patients Make Sense of Health Statistics. Psychol Sci Public Interes. 2007;8: 53–96. pmid:26161749
- 47. Siettos CI, Russo L. Mathematical modeling of infectious disease dynamics. Virulence. 2013;4: 295–306. pmid:23552814
- 48.
Hutcheson GD, Sofroniou N. The multivariate social scientist: Introductory statistics using generalized linear models. Newbury Park, USA: Sage; 1999.
- 49. Huang X, Zhou H, Yang X, Zhou W, Huang J, Yuan Y. Spatial Characteristics of Coronavirus Disease 2019 and Their Possible Relationship With Environmental and Meteorological Factors in Hubei Province, China. GeoHealth. 2021;5. pmid:34189364
- 50. Al Kindi KM, Al-Mawali A, Akharusi A, Alshukaili D, Alnasiri N, Al-Awadhi T, et al. Demographic and socioeconomic determinants of COVID-19 across Oman—A geospatial modelling approach. Geospat Health. 2021;16: 985. pmid:34000790
- 51. Simsek S, Dag A, Tiahrt T, Oztekin A. A Bayesian Belief Network-based probabilistic mechanism to determine patient no-show risk categories. Omega. 2021;100: 102296.
- 52. Waal T, Delden A, Scholtus S. Multi‐source Statistics: Basic Situations and Methods. Int Stat Rev. 2020;88: 203–228.
- 53. Vantaggi B. Statistical matching of multiple sources: A look through coherence. Int J Approx Reason. 2008;49: 701–711.
- 54.
Refaeilzadeh P, Tang L, Liu H. Cross-Validation. In: Liu L, Özsu MT, editors. Encyclopedia of Database Systems. Boston, MA: Springer; 2009. pp. 532–538.
- 55. Brunsdon C, Fotheringham AS, Charlton ME. Geographically weighted regression: a method for exploring spatial nonstationarity. Geogr Anal. 1996;28: 281–298.
- 56.
Wheeler DC, Páez A. Geographically Weighted Regression. In: Fischer MM, Getis A, editors. Handbook of Applied Spatial Analysis. Berlin/Heidelberg, Germany: Springer Science & Business Media; 2010. pp. 461–486.
- 57. Wu X, Zhang J. Exploration of spatial-temporal varying impacts on COVID-19 cumulative case in Texas using geographically weighted regression (GWR). Environ Sci Pollut Res Int. 2021;28: 43732–43746. pmid:33837938
- 58. Yao Y, Shi W, Zhang A, Liu Z, Luo S. Examining the diffusion of coronavirus disease 2019 cases in a metropolis: a space syntax approach. Int J Health Geogr. 2021;20: 1–14. pmid:33397338
- 59. Liu C, Liu Z, Guan CH. The impacts of the built environment on the incidence rate of COVID-19: A case study of King County, Washington. Sustain Cities Soc. 2021;74: 103144. pmid:34306992
- 60. Middya AI, Roy S. Geographically varying relationships of COVID-19 mortality with different factors in India. Sci Rep. 2021;11: 7890. pmid:33846443
- 61. Boivin R. On the use of crime rates. Can J Criminol Crim Justice. 2013;55: 263–277.
- 62. Pereira DVS, Mota CMM, Andresen MA. Social Disorganization and Homicide in Recife, Brazil. Int J Offender Ther Comp Criminol. 2017;61: 1570–1592. pmid:26723297
- 63. Verity R, Okell LC, Dorigatti I, Winskill P, Whittaker C, Imai N, et al. Estimates of the severity of coronavirus disease 2019: a model-based analysis. Lancet Infect Dis. 2020;20: 669–677. pmid:32240634
- 64.
Chatterjee S, Hadi AS. Regression Analysis by Example. Hoboken: John Wiley & Sons; 2013.
- 65. Jenks GF, Caspall FC. ERROR ON CHOROPLETHIC MAPS: DEFINITION, MEASUREMENT, REDUCTION. Ann Assoc Am Geogr. 1971;61: 217–244.
- 66. Bermudi PMM, Lorenz C, Aguiar BS de, Failla MA, Barrozo LV, Chiaravalloti-Neto F. Spatiotemporal ecological study of COVID-19 mortality in the city of São Paulo, Brazil: Shifting of the high mortality risk from areas with the best to those with the worst socio-economic conditions. Travel Med Infect Dis. 2021;39: 101945. pmid:33278610
- 67. Bambra C, Riordan R, Ford J, Matthews F. The COVID-19 pandemic and health inequalities. J Epidemiol Community Health. 2020;74: 964–968. pmid:32535550
- 68. Li B, Peng Y, He H, Wang M, Feng T. Built environment and early infection of COVID-19 in urban districts: A case study of Huangzhou. Sustain Cities Soc. 2021;66: 102685. pmid:33520609
- 69. Florindo AA, Turrell G, Garcia LMT, dos Anjos Souza Barbosa JP, Cruz MS, Failla MA, et al. Mix of destinations and sedentary behavior among Brazilian adults: a cross-sectional study. BMC Public Health. 2021;21: 347. pmid:33579233
- 70. de Lima J, Maia ML, Lucas K. Income vs. travel time: Why do the poorest and the richest travel fastest in northeastern Brazil? Transp Res Procedia. 2017;25: 4285–4295.
- 71. Kronbichler A, Kresse D, Yoon S, Hwa Lee K, Effenberger M, Il Shin J. Asymptomatic patients as a source of COVID-19 infections: A systematic review and meta-analysis. 2020;98: 180–186. pmid:32562846
- 72. Kameda K, Barbeitas MM, Caetano R, Löwy I, Oliveira ACD de, Corrêa MCDV, et al. Testing COVID-19 in Brazil: fragmented efforts and challenges to expand diagnostic capacity at the Brazilian Unified National Health System. Cad Saude Publica. 2021;37. pmid:33852667
- 73. Lyu J, Miao T, Dong J, Cao R, Li Y, Chen Q. Reflection on lower rates of COVID-19 in children: Does childhood immunizations offer unexpected protection? Med Hypotheses. 2020;143: 109842. pmid:32425304
- 74. Yu X. Risk Interactions of Coronavirus Infection across Age Groups after the Peak of COVID-19 Epidemic. Int J Environ Res Public Health. 2020;17: 5246. pmid:32708118
- 75. Kang J-Y, Michels A, Lyu F, Wang S, Agbodo N, Freeman VL, et al. Rapidly measuring spatial accessibility of COVID-19 healthcare resources: a case study of Illinois, USA. Int J Health Geogr. 2020;19: 36. pmid:32928236
- 76. Gao YD, Ding M, Dong X, Zhang JJ, Kursat Azkur A, Azkur D, et al. Risk factors for severe and critically ill COVID‐19 patients: a review. Allergy. 2021;76: 428–455. pmid:33185910
- 77. Shi X, Lima SMS, Mota CMM, Lu Y, Stafford RS, Pereira CV. Prevalence of multimorbidity of chronic noncommunicable diseases in brazil: Population-based study. JMIR Public Heal Surveill. 2021;7: e29693. pmid:34842558
- 78. Alfano V, Ercolano S. The Efficacy of Lockdown Against COVID-19: A Cross-Country Panel Analysis. Appl Health Econ Health Policy. 2020;18: 509–517. pmid:32495067
- 79. Garcia MLT, Pandolfi AF, Leal FX, Stocco AF, Borrego AE, Borges RE, et al. The COVID-19 pandemic, emergency aid and social work in Brazil. Qual Soc Work. 2021;20: 356–365. pmid:34253979
- 80.
G1. Agências da Caixa voltam a registrar longas filas em todo o Brasil. 2020 [cited 6 Aug 2020]. Available from: https://g1.globo.com/jornal-nacional/noticia/2020/07/24/agencias-da-caixa-voltam-a-registrar-longas-filas-em-todo-o-brasil.ghtml.
- 81. Atinc G, Simmering MJ, Kroll MJ. Control variable use and reporting in macro and micro management research. Organ Res Methods. 2012;15: 57–74.
- 82. Li M. Uses and abuses of statistical control variables: Ruling out or creating alternative explanations? J Bus Res. 2021;126: 472–488.