Background
Brazil is a high-burden country for tuberculosis (TB) and human immunodeficiency virus (HIV)-TB co-infection [
1] and also characterised by high levels of inequality in social and health indicators [
2‐
4]. The twin slogans of ‘Leave no one behind’ and ‘Everybody counts’ adopted for World Tuberculosis Day and World AIDS Day, respectively, in 2017, emphasise the importance of reducing inequality to end these leading epidemics [
5]. TB and HIV inequalities may manifest in geographic patterns because the underlying risk factors for TB and HIV infection and death, such as poverty, incarceration, undernutrition, crowding and poor access to health services, vary across geographic areas and through time [
6‐
10]. Additionally, the disease mechanisms of transmission between persons in close contact can lead to geographic clusters of disease burden [
11‐
13]. Brazil has invested in massive social programmes to improve health and equality, such as the Family Health programme of free community-based healthcare, the Bolsa Familia programme of cash transfers conditional on education and health behaviours [
14], and universal eligibility for free TB care and free antiretroviral therapy for HIV infection since its discovery in 1996 [
15]. The national strategy to end TB in Brazil prescribes TB control strategies based on local epidemiology; fine-scale mapping of TB and HIV burden can provide information to prioritise additional programmatic investments toward improving health [
16].
Prior investigations of the spatial distribution of TB and HIV burden in Brazil varied in their scope and level of geographic detail, but few achieved coverage of the entire nation at fine spatial resolution or for long time series. The Global Burden of Disease (GBD) study collaborators modelled mortality due to an exhaustive set of causes, including HIV and TB, at the state level for 1990 through to 2015 [
4]. Other investigations modelled mortality or case notifications at finer spatial scale for portions of the country [
17‐
20]. Harling et al. [
21] completed a nationwide municipal-level analysis of case notifications in Brazil of a shorter time series, 2002 to 2009. Outside of Brazil, there are few national-level spatial modelling studies of TB incidence and, to our knowledge, no nationally comprehensive spatial models of TB mortality at fine spatial scale [
22‐
24]. There are broader spatial modelling efforts for HIV, corresponding to the greater availability of spatially resolved data sources for HIV than for TB in high-burden countries [
25,
26].
There are methodologic challenges associated with spatial modelling of TB and HIV mortality which are addressed by this analysis. First, despite being leading infectious causes of death globally, TB and HIV death counts are low in small areas, leading to instability in case counts and difficulty in separating true differences in risk from stochastic noise for individual geographic areas. A modelling approach that draws strength from neighbouring groups across space and time could stabilise these estimates. Second, TB and HIV deaths may be misclassified due to failure to recognise the cause of death as HIV or TB or stigma associated with reporting these conditions [
27‐
29]. Furthermore, the International Statistical Classification of Diseases (ICD) convention is for TB deaths in persons living with HIV infection (PLHIV) to be assigned to HIV as the underlying cause, which can hide the contribution of TB to these deaths if only a single cause of death is reported in vital registration [
30]. In this study, we address these challenges by utilising comprehensive cause of death assignment and small area estimation to conduct a nationwide analysis of TB and HIV mortality at fine geographic scale. We also estimate the TB case fatality ratio, defined as the proportion of persons with TB who die of TB, a key metric in the World Health Organization (WHO) End TB Strategy [
1]. HIV case fatality ratios are not estimated due to a lack of data to inform HIV incidence.
Discussion
Despite marked progress nationally in reducing deaths due to TB and concentrated gains for HIV, substantial inequality in TB and HIV burden are apparent at each geographic level of analysis. Trends in within-state variation for TB were driven by faster mortality reductions in the lowest-burden municipalities relative to slower improvement in the highest-burden areas, the majority of which remained in the highest-burden decile at the end of the 15-year interval. HIV mortality declines in highly populated, high-burden areas drove national-level declines, but the majority of municipalities demonstrated an increase in HIV mortality rate during this period, which was also observed in prior studies [
38]. Evaluation of the municipalities with the greatest mortality improvements may identify successful strategies that could be extended to areas experiencing increases or slower declines.
Mortality estimates disaggregated by sex revealed differences in TB and HIV burden and geographic distribution. Consistent with known TB and HIV epidemiology, we found a greater burden of TB and HIV mortality in males than in females, but also somewhat different spatial patterns by sex [
28,
39]. Incarceration is a known risk factor for TB infection, with prisoners (pessoas privadas de liberdade) in Brazil having an estimated TB notification rate more than 30 times that of the non-incarcerated population [
40]. HIV prevalence is also higher in Brazilian inmates than in the non-incarcerated population [
41‐
43]. Men comprise more than 90% of the Brazilian prison population. Municipalities with large prison populations, such as several in Sao Paulo state, stand out in the maps showing results for males as having a higher TB incidence and HIV mortality than neighbouring municipalities. In contrast, municipalities where females were at greatest risk for HIV and TB mortality were concentrated along the national border areas and in the interior Amazon.
National-level case fatality ratios for TB improved over the period of this analysis. However, broader efforts are also needed, as only half of municipalities achieved a WHO End TB Strategy case fatality ratio target of < 10% among females and just over one-third of municipalities achieved it among males in the final period of the analysis between 2011 and 2014. Nearly twice as many municipalities achieved the WHO target for females than for males, indicating a critical need for TB treatment completion strategies that successfully engage men. Underreporting of TB case notifications could bias these estimates downward; however, this effect is likely to be small in later years, when reporting completeness is estimated at more than 90% [
1]. There was less apparent geographic patterning for case fatality than for TB or HIV mortality, but the burden of higher case fatality ratios appeared to shift from coastal areas to more inland areas over the analysis period.
While increased funds are required to maintain gains and further improve health and equality, congress approved Constitutional Amendment 95 in December 2016, restricting funds allocated to the health sector and providing no real increase in health funding for the next 20 years [
44]. This austerity has also extended to other sectors impacting health and wellbeing, including education and public utilities such as sanitation. These policies could stall the important progress made in Brazil over the period of this study.
This work extends previous efforts to model subnational TB and HIV burden by generating estimates that are both nationally comprehensive and fine-scale. It supports calls to collect and analyse TB and HIV data with high spatial resolution in order to inform interventions that are most appropriate to the transmission dynamics in particular settings [
45]. Knowledge of the local variation in TB and HIV burden can inform programmatic interventions to improve health outcomes [
16]. TB interventions, such as active case finding and mobile testing units, can be resource-intensive and are utilised most effectively when prioritised to high-burden areas [
46]. Subnational differences in HIV burden have also been used to develop locally tailored strategies for HIV prevention and elimination [
25,
47‐
49]. However, the benefits of highly geographically resolved disease burden estimation should be weighed against the risk of potentially identifying individuals if analyses of exceptionally rare outcomes are carried out over very small areas.
Limitations
There are several limitations to this analysis. While adult mortality data in Brazil are assessed to be complete for the period of this analysis, child mortality data are estimated to be < 95% complete in the vital registration system [
50]. Other national-level analyses have included additional data sources at a different spatial resolution such as household surveys [
51]. Due to the complexity of integrating different data types, only vital registration data were included in this analysis. However, calibrating these estimates to GBD, which includes survey data in all-cause mortality estimation, reduces the undercounting of deaths. Deaths in children under the age of 15 constitute a small proportion of TB (1.6%) and HIV (7%) deaths in Brazil during this period, so the spatial effect of this difference in data sources is not expected to be large. Similarly, TB cases may be under-ascertained in the case notification system. While the overall completeness of TB case notification is estimated currently to be greater than 90%, completeness of reporting may vary spatially [
1]. Future work may assess whether factors such as treatment-seeking behaviour and reporting completeness can be used to improve modelling of TB incidence from case notifications.
HIV and TB are under-ascertained as causes of death, and TB is under-ascertained as a contributing cause of death among persons with HIV infection [
52]. The GBD mortality redistribution method attempts to correct for these biases. Other correction methods include linkage analysis of HIV and TB surveillance systems [
27], or linkage of diagnoses made at health facility encounters with information recorded on death certificates. These could be pursued as additional methods to improve ascertainment of TB and HIV deaths.
Future directions
There are several additional future directions for this work. First, while the causes of TB mortality and TB incidence are further broken down in GBD analyses into drug-susceptible TB, multidrug-resistant TB and extensively drug-resistant TB, data were not available at the geographic scale of this study to inform analysis by drug resistance categories. Additional geographic detail in data sources would facilitate analysis by drug resistance categories. Second, climatologic variables were included in the TB models as an exploratory analysis due to postulated relationships between air temperature, wind speed and TB transmission [
22,
53]. Relationships with these factors may be tested in future spatial models in order to potentially improve estimation of TB burden in areas with minimal health surveillance data. Third, a similar small area estimation approach could be used to estimate all-cause and cause-specific mortality due to other causes at the municipal level in Brazil. Finally, this small area estimation approach to spatial mapping of HIV and TB mortality could be extended to other nations with well-functioning vital registration systems.
Conclusion
Mortality due to TB and HIV exhibited nearly as much relative variation within Brazilian states as within the nation as a whole. This demonstrates the role for increasing geographic detail in burden estimation to guide precision public health responses. Fewer than half of municipalities met the WHO End TB Strategy target for a case fatality rate of < 10%, indicating priority areas for improvement in order to achieve international targets and improve health equity.
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (
http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (
http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.