Spatiotemporal patterns of AIDS incidence among adults in São Paulo, Brazil

Introduction: AIDS remains a major public health concern in Brazil. Methods: This study investigated spatiotemporal patterns of reported AIDS cases among adults in São Paulo between 2000 and 2016, and their associations with human development index. Results: In the early 20th century, the more developed administrative districts (ADs) indicated higher AIDS incidences among men. From the 2010s, ADs with lower development indicate higher rates of the disease among women. Conclusions: The results are useful to support the planning of actions aimed at controlling the incidence and transmission of AIDS in certain areas, based on diversification by gender and risk populations.

The dynamics of the AIDS epidemic in large populations are highly complex and can result in several regional subepidemics defined according to different social interactions. In this context, spatial studies have been applied to understand the characteristics of the dynamics of AIDS in Brazil 1,2 using geoprocessing methods integrated with spatial statistical models 3 . Spatial epidemiology statistical methods provide resources to explore the spatial location of events, facilitated by the use of geographic information systems, allowing identification, location, and visualization of the occurrence of phenomenon of interest. Spatial studies on this occurrence in a given population (e.g., [1][2][3][4] ) allow the understanding of its spatial distribution and constitute an effective tool to optimize public health strategies by allocating resources according to each region's needs 4 .
The objective of this article is to study the spatial and temporal distribution of notified AIDS cases among adults (18 years old or more) in São Paulo, between 2000 and 2016, and their associations with the Human Development Index (HDI) and its three components: education, income, and life expectancy. Open-access data from the Brazilian HealthCare Computer System (DATASUS) regarding the number of new cases of AIDS recorded from 2000 to 2016 in each of São Paulo's administrative districts (AD) were gathered. São Paulo is the most populated city in South America with more than 12 million inhabitants. The city is divided into 96 ADs based on geographical and administrative features. Data on HDI (year 2000) were obtained from the Atlas of Human Development in Brazil.
Let Y itk be the number of incident cases of AIDS observed in the i-th AD (i = 1,...,96) in the year t (t = 1 if 2000, t = 2 if 2001, ..., t = 17 if 2016), considering individuals by gender k (k = 1 if males and k =2 if females). The statistical model considers that Y itk is a random variable that follows a Poisson distribution with mean value given by N itk θ itk , where N itk is the number of inhabitants of the i-th AD in the year t and gender k, and θ itk is an unknown parameter to be estimated relative to the incidence rate for each AD, year, and gender. Two different models were fitted.
Model 1: This model was defined in order to obtain spatially smoothed incidence rates of the disease, where ln(θ itk ) =α k + λ ki + W itk . In this formulation, α k is an intercept for each gender (k = 1, 2) following a flat prior distribution and λ ki (k = 1, 2, i = 1,..., 96) are random effects capturing the possible spatial correlations among the AIDS incidences, considering the effects of neighboring fields. A bivariate intrinsic conditional autoregressive (CAR) normal structure was assigned to λ ki , where  the neighborhood was defined by the adjacency criterion 5 . Finally, W itk are longitudinal effects for the incidence rates, following multivariate normal distributions with a mean vector of zero and variance matrix with components described in Branscum et al. 6 . Model 2: This model included a covariate denoted by X based on the function ln(θ itk ) =α kt + λ ki + β kt (x i -m), where α kt are constant intercepts for each gender and year, λ ki are spatial random effects defined in Model 1 and parameters β kt measure the effect of covariate X on the incidence rate for each gender and year. Further, x are observations of X and m is the mean of these observations. It is not possible to fit multiple models to the data, simultaneously including all HDI components, due to high collinearity among these variables. Flat prior distributions were assigned to α kt and normal prior distributions with zero mean and large variance were assigned to parameters β kt .
Posterior distributions for the parameters of interest were simulated by using the Markov Chain Monte Carlo (MCMC) method in the OpenBUGS software 7,8 . The MCMC procedure was run for 60,000 iterations after a burn-in of a burn-in of 6,000 iterations. The 95% credible intervals (95% CI) were obtained by taking the mean and 2.5% and 97.5% quantiles of the MCMC posterior sample of the respective parameters. Credible intervals are the Bayesian analogue of frequentist confidence intervals. For the fit of Model 2, whenever the limits of CI for parameter β kt do not include zero, there is a significant effect of covariate X on the AIDS incidence for individuals of gender k at time t.
This research was approved by the Research Ethics Committee of Hospital das Clínicas, Ribeirão Preto Medical School, University of São Paulo (protocol number 565.914). Figure 1 describes the spatial distribution of HDI and its three components. The highest scores are concentrated in the central region, showing areas of greater wealth, decreasing as they move away from the center toward the peripheries. The fit of Model 1 allowed for the smoothing of crude rates regarding gender, according to the spatiotemporal pattern. Figure 2 indicates the spatial distribution of these smoothed rates of AIDS incidences for men and women, considering the years 2000, 2008, 2013, and 2016. Maps for the other years are not shown because of the large quantity of data. Among men, the central ADs tend to have highest incidence rates per 100,000 inhabitants; however, these rates tend to decline over the years. Considering the male population, in 2000, São Paulo had a median smoothed AIDS incidence of 37.98 cases per 100,000 inhabitants among the 96 ADs (interquartile range The maps also indicate that, among women, the highest AIDS incidence rates were found in some central areas of the municipality at the beginning of the study period; however, the maps tend to indicate a more heterogeneous spatial distribution of the incidence rates as we approach the end of this period. In 2000, São Paulo's ADs had a median smoothed incidence of 17.58 cases per 100,000 inhabitants among women (interquartile range [IQR] 14.47 to 25.35). This incidence decreased to 13.27 (IQR 10.93 to 19.14) and 2.14 (IQR 1.76 to 3.09) in 2008 and 2016, respectively. Figure 2 also indicates maps of the male to female AIDS incidence ratio in São Paulo, also estimated from the fit of Model 1. The maps suggest an increase in this ratio over the study period. In 2000, the median for the male to female incidence ratio was 2.0 (IQR 1.8 to 2.9) similar to 2.1 (IQR 1.8 to 3.0) in 2008. However, in 2013 and 2016 the median increased to 2.9 (IQR 2.5 to 4.1) and 3.9 (IQR 3.4 to 5.5), respectively. In general, the highest male to female incidence ratios were observed in the central ADs. However, at the end of the study period, the maps suggested a more homogeneous spatial distribution of this ratio. Figure 3 indicates 95% CI for the estimates of parameter β kt in Model 2 regarding the effect of each HDI component on AIDS incidence for each gender and year. Intervals containing zero indicate no evidence of significant effect. A positive association         between HDI and its three components and AIDS incidence (values for β kt higher than zero) was observed among men, considering the initial years of the period. This relationship was not observed at the end of this period. From these results and by comparing the maps shown in Figures 1 and 2, we have evidence that in the early 20th century the new AIDS cases among men were more likely to occur in more developed regions of São Paulo; however, this trend began to diminish in the second decade of this century. Figure 3 suggests a different association pattern between HDI components and AIDS incidence among women. A lack of association is visible at the beginning of the century (CIs for β kt include zero); however, the estimates for the association parameters become negative and significant at the end of this period. These results suggest that, from the second decade of this century, AIDS incidences among women decreased in the more developed regions of the municipality.

5/6
This study proves that the geographical distribution of ADs in São Paulo indicates deep social and economic differences, with central areas being more developed due to the concentration of higher income people with greater purchasing power, whereas lower income people with less purchasing power are concentrated in the peripheral areas. Therefore, our results suggest that in the early 20th century, the more developed ADs indicated higher AIDS incidences among men. However, from the second decade of the century, ADs with lower economic classes indicate higher rates of the disease among women. In part, this phenomenon is already well known across Brazil, mainly in the country's major cities, and is referred to as impoverishment and feminization of the AIDS epidemic. This is translated into a dissemination of the disease among the poor and female population 9 , proving the HIV-carriers' behavioral aspects and changes in the predominantly homosexual pattern of disease transmission between men, which now includes women, resulting in another phenomenon, the so-called heterosexualization of AIDS 10 . This phenomenon suggests that women from lower socioeconomic classes are more susceptible to AIDS due to their partner's behavioral aspects, such as unprotected extramarital sex, non-use of condoms during sex with the wife, and use of injectable drugs 11 . In sum, it is known that the AIDS history in Brazil indicates that the disease had initially reached upper socioeconomic groups and homosexual individuals; however, now it is reaching people from all socioeconomic classes regardless of their sexual behavior 2 .
This study was performed in a metropolis with great socioeconomic diversity, harboring people from various parts of the world whose behaviors and cultural habits are typical of those observed in other major cities worldwide. In this way, the distribution pattern of the disease does not differ from those reported by other studies performed in other metropolises, including some of China's and North-America's most populated cities 12,13 . The results for these similarities are no secret since they are undoubtedly due to the poor life conditions such as low income and education level, urban agglomeration, promiscuity, and other risk behaviors, among other factors, all being currently observed in the peripheral regions of major cities 14 . However, similar events have also been observed in smaller cities and rural towns, pointing to the emergence of other phenomena, namely, interiorization and ruralization of the disease 15 .
Potential limitations of this study are related to information type and study design. Although our data have come from health information systems, they do not precisely describe the actual situation as there are issues regarding how the AIDS cases are notified and related recorded information, despite the mandatory notification of the disease. Because this is an ecological study, it is not possible to generalize the results to an individual information level. The use of synthetic indicators also restricts any study extrapolation given the methodological and conceptual limitations, including inadequate use of indicators when confronted with health information.
Our results are useful to support the planning of public health policies and actions aimed at controlling the incidences and transmission of AIDS in certain areas on the basis of gender and risk populations diversification in São Paulo, Brazil. We also emphasize the need for personalizing access to healthcare services by administrative district and gender in order to propose a healthcare strategy for men and women in each area by establishing local support teams with particular characteristics, as demonstrated in this study.