Background
As countries accelerate towards malaria elimination, reduction in malaria transmission is highly variable leading to significant heterogeneities in residual transmission across their territories [
1‐
4]. In such circumstance, a ‘one size fits all’ approach to public policy and intervention may no longer be cost-effective and may slow progress towards the final goal of local elimination of transmission. Understanding both the dynamic nature of these heterogeneities and developing systems to efficiently identify and characterize residual transmission areas is essential to developing locally adapted, sub-national malaria elimination policies and intervention packages.
Brazil has seen long-term trends of nationwide reductions in notified
Plasmodium vivax and
Plasmodium falciparum cases since 2000, and a retreat of malaria into the Amazon Basin in the north of the country [
5]. These epidemiological trends have been dependent on socio-economic development, and sustained effective case management, largely based on intensive passive case detection (PCD) [
3]. In Brazil, treatment is provided publicly and free of charge in government-run Health Units, including hospitals and mobile healthcare workers in both urban and rural communities [
6]. From these Health Units, data on treated cases are digitally recorded in the Malaria Epidemiological Surveillance Information System (SIVEP) database. Containing digitized data on almost 6 million malaria cases with individual level covariates, the Brazilian SIVEP database is one of the most detailed health systems databases ever assembled in a malaria endemic country, covering the entire endemic region and resulting in virtually no underreporting [
7,
8]. The collected data span a crucial period, as many regions of Brazil are nearing and some even achieving local malaria elimination, providing examples of what malaria elimination looks like from a health system perspective.
In addition to the long-term, large-scale trends in malaria epidemiology, there is considerable variation in the numbers of notified cases between regions and over time. Factors contributing to spatio-temporal heterogeneity include deforestation, changes in land use and economic activity, ecological suitability of mosquitoes (most notably of
Anopheles darlingi), climate variations, internal and international migration, and political crises in neighbouring countries [
9‐
11]. Historically, the Brazilian malaria landscape was largely shaped by what has been referred to as ‘frontier malaria’ where new infections were driven by non-immune populations settling for economic and political purposes [
3,
12‐
14]. More recently, although malaria is in an epidemiological transition towards elimination in some regions, it has maintained itself as an endemic disease with sustained transmission over many years in many regions [
2,
3]. Despite this heterogeneity, policies on surveillance, diagnosis, and antimalarial treatment are exactly the same for the whole country. Major guidelines are applied to all states and municipalities, each being responsible for executing actions with little flexibility or customization regarding specific scenarios [
9].
Micro-epidemiological studies provide an in-depth local understanding of malaria transmission, and are crucial for designing stratified locally adapted intervention strategies [
15]. Data from micro-epidemiological settings can be aggregated to construct a macro-epidemiological picture of malaria at varying regional levels, such as within Brazilian states, or across the entire nation of Brazil. Understanding how thousands of such micro-epidemiological patterns combine to produce the macro-epidemiology of malaria in Brazil is a challenging problem.
Data aggregation across multiple levels is a more general problem encountered in many fields. In economics, there are many statistical tools for quantifying the relationship between micro-economic and macro-economic phenomena. An important example is the Gini coefficient, an index which quantifies wealth inequality between individuals on a national level. Recent advances in health systems databases allow for conceptual parallels between economic and epidemiological phenomena to be exploited [
16]. For example, statistical tools originally designed for the analysis of wealth inequality can be repurposed to provide insight into malaria heterogeneity.
Improved understanding of heterogeneity across multiple spatial scales will better inform stratification of malaria risk zones allowing for more efficient targeting of interventions. In Brazil, the National Malaria Control Programme (NMCP) already pursues a targeted approach with vector control interventions being distributed to high risk areas [
10]. In this macro-epidemiological analysis of Brazilian SIVEP-malaria health system data, it is demonstrated how a Gini index applied to malaria cases over different spatial units (state, municipality, and others) describes heterogeneity of malaria transmission in a way that can aid stratification for improved malaria control. This could be a necessary first step in order to move the national strategy from control to malaria elimination.
Discussion
The majority of cases of malaria in Brazil occur in the Amazon basin. Although there is malaria transmission in some sylvatic areas along the coast of Brazil, most transmission events notified in areas other than the Amazon basin originate from the states in the Northern region of the country. Therefore, any efforts to eliminate malaria depend on the incidence in the Amazon region.
Stratification of at-risk populations at lower administrative levels reveals increasing heterogeneity in the distribution of malaria cases throughout the Brazilian Amazon. This is consistent with malaria having a fractal nature within the levels considered, with evermore variation observed at lower spatial resolution [
23]. A limitation of our analysis is that the lowest level with accurate data on cases and denominator populations was the municipality. Although inference was possible for health units within municipalities, this required estimation of catchment populations. However, two lines of evidence indicate that additional heterogeneity would be observed with finer population stratification at lower spatial scales. Firstly, demographic stratification by age and gender reveals substantial heterogeneity. Secondly, micro-epidemiological analyses of data from cross-sectional and longitudinal cohorts in Brazils with geo-located households reveals considerable spatial clustering [
24].
Borrowing from econometrics, the Gini coefficient was utilized to demonstrate that heterogeneity in the distribution of malaria cases is increasing over time (Fig.
4a). In particular, this is closely associated with decreasing incidence of malaria cases (Fig.
4b). These results suggest that the spatial distribution of malaria cases across populations was dependent on changing transmission intensity and the size of the population at risk. This pattern of increasing heterogeneity with declining transmission has also been demonstrated on a micro-epidemiological level, with a higher clustering of infectious mosquitoes in lower transmission settings [
16].
Stratification of cases by age and gender reveals patterns consistent with two key archetypes of malaria transmission settings in Brazil: peri-domestic and occupational exposure. Peri-domestic transmission is characterized by exposure to infectious mosquitoes in and around peoples’ households. In these settings, age and gender stratification typically reveal cases across all age groups, in both males and females. In particular, the distribution of cases represents the underlying demography of the population. Municipalities with high API are often, but not always, characterized by peri-domestic transmission, as cases are notified from all segments of the population. In agreement with data from sub-Saharan Africa [
25], a peak of cases in children is observed, with this peak shifting towards younger ages in higher transmission settings due to the acquisition of clinical immunity (Additional file
1: Fig. S4) [
26]. This allows existing routinely collected data to help target interventions for malaria control. Many peri-domestic settings with cases notified across all age groups in males and females would benefit from targeted vector control, as there is evidence to suggest that transmission is occurring in and around households [
10,
27].
Occupational transmission is characterized by exposure to infectious mosquitoes in activities such as mining, agriculture, forestry, and construction [
3,
28]. Age and gender stratification typically reveal a concentration of cases in working age males. Occupational exposure settings often have low malaria transmission when assessed across an entire population; however, this may obscure intense and stable transmission in core risk groups. Vector control targeted at households is unlikely to be effective in this settings [
29]. Classification of malaria exposure into these two archetypes underestimates the diversity of malaria transmission systems. In particular, it may not reflect differences between urban and rural transmission, or indoor and outdoor based transmission systems.
Systematic analyses of the breakdown of cases across the range of transmission intensities reveals important characteristics of the pathway to elimination. Namely, a higher proportion of cases are seen in males, in adults, due to importation, and due to
P. vivax. An analysis of data on notified measles cases argued that there is a canonical path to measles elimination, with all regions going through similar epidemiological transitions as they progress towards measles elimination [
30]. Although distinct epidemiological phenomena are clearly observed on the path to elimination, the data presented here suggest that there is no canonical path to malaria elimination. Take the example of Tocantins state; although formal elimination has not been declared, the low case numbers and high proportion of importation is consistent with many municipalities having eliminated local malaria transmission in the past two decades. If progress continues over the next two decades, will high transmission municipalities in Acre follow the same trajectory as Tocantins? As Acre is a peri-domestic transmission setting, and Tocantins is an occupational transmission setting, then these regions will not follow the same epidemiological transitions on the way to elimination. More generally, the diversity of malaria transmission systems, suggests that there is no canonical path to malaria elimination, and that each region will need to follow its own path.
The recent history of malaria control in Brazil has mostly been a successful one, with a nationwide reduction of cases, and many municipalities reducing the numbers of notified cases to levels consistent with the elimination of local transmission. This progress has been built on Brazil’s strong case management, with all cases being diagnosed by thick blood smear or rapid diagnostic tests, and treated free of charge at local Health Units. Notably, Brazil is one of few countries that routinely provide radical cure for
P. vivax via chloroquine, and a total dose regimen of 0.5 mg/kg/day primaquine for 7 days, only excluding children under 6 months and pregnant women. Despite legitimate concerns about this regimen’s adherence and efficacy [
31‐
34], it is playing a key role in the effort for
P. vivax elimination. The recent approval of tafenoquine for
P. vivax case management provides important opportunities for malaria control in Brazil. With only a single dose, tafenoquine has been demonstrated to have comparable efficacy to primaquine, although testing for G6PD deficiency will be mandatory due to the risks of treatment-induced haemolysis [
35,
36]. The Brazilian NMCP will need to make decisions on whether tafenoquine is implemented universally through a “one-size-fits-all” strategy, or whether roll-out should be targeted at specific regions or risk groups.
The patterns of spatio-temporal heterogeneity demonstrated here have important implications for NMCPs pursuing malaria elimination. The increase in heterogeneity as transmission declines causes the remaining malaria cases to be concentrated in smaller regions or populations, potentially allowing more efficient and cost-effective targeting. For example, rather than implementing standardized interventions across the entire endemic region, targeted policies could be directed at 4.5% of the population or ~ 600 health units, where 80% of cases are known to be reported. However, these remaining pockets sometimes referred to as hotspots can have intense and stable malaria transmission. Sub-microscopic infections undetectable at local clinics may further mask heterogeneous patterns. In such hard-to-eliminate settings, it is crucial to sustain interventions for a long duration and employ multiple strategies that seek to improve in combination radical cure adherence (e.g. single-dose tafenoquine), bed net coverage in remote areas, and diagnostic tools to identify asymptomatic infections.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.