Study area

In this study, the Brazilian Amazon frontier refers to regions where human settlement is still occurring, marked by the construction of communications networks, emergence and growth of towns and cities, human migration, and shaping of a new landscape [16]. Parts of the Amazon are still affected by these activities, with different stages of government policy implementation and private investment, producing extensive heterogeneity and regional disparities.

In an attempt to regionalize the economic frontier in different zones of territorial occupation, the region was divided accordingly [17] to study the relevant temporal and spatial dynamics in the expansion of the economic frontier and to contextualize and discuss the relationship between the spatial distribution of economic activities and the spatial dispersion of malaria in the postfrontier phase. This was especially important because the economic frontier is subject to dynamic changes and can shift or dissolve in the territory. The reconfiguration and transformation of territorial organization in the Brazilian Amazon are now in the postfrontier stage. The postfrontier should also be seen as a recontextualization of the Amazonian geographic space, as new areas are incorporated with contemporary technologies, practices, and institutions that regulate, characterize, and restructure the frontiers [17]. The frontier consolidation sequence is defined [18] as a phase of neoliberal land occupation in postfrontier areas with a high percentage of business ownership and strong international links to agribusinesses. Thus, we divided the range of economic frontier occupation into three major subregions: the consolidated settlement zone, the Central Amazon, and the south and west Hiléia settlement zone (Fig 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Range of economic frontier occupation: cities, highways, and waterways.

https://doi.org/10.1371/journal.pone.0217615.g001

The consolidated settlement zone includes six areas: Belém, southeastern and eastern Pará, Tocantins, Mato Grosso, and Rondônia, all located along the southeastern and northeastern portions of the region. Much of the subregion’s economy is based on logging, cattle ranching, and soybean production. The subregion is highly dynamic, with a dense urban and roadway network consisting of medium-sized cities and state capitals [19].

The Central Amazon includes some municipalities in the states of Acre, Amazonas, Maranhão, Pará, Tocantins, and Rondônia and is located between the Cerrado (savannah) and Amazon biomes. The subregional economy specializes in logging, livestock, soybean production, and mining. It has the second highest population density of the studied areas (more than 16 inhabitants per km²) and a high urbanization rate [19].

The south and west Hiléia settlement zone includes some of the municipalities of the states of Acre, Amapá, Amazonas, Pará, Roraima, and Rondônia and is located in the northern and northwestern portions of the Amazonian Hiléia. The economic structure of the subregion is concentrated on forestry (açaí and Brazil nuts), logging (timber and firewood), fishing, and subsistence agriculture (manioc, bananas, and livestock, mainly water buffalo and beef cattle). The subregion’s urban network is sparsely populated, consisting mainly of small towns and a low population density (less than 1 inhabitant/km²). It contains the Amazon River valley, which is still an important axis for human settlement [19] [20].

Theoretical and conceptual model

A three-tier theoretical and conceptual model (Fig 2) was developed to study associations between the socioenvironmental determinants of malaria at the distal, intermediate, and proximal levels. The conceptual model is based on the social determinants of health (SDH) approach. In light of existing knowledge, the distal level included economic drivers embodie as large projects (highways, dams, hydroelectric power plants, tourism, mining, hunt and fisheries, and large-scale agriculture), per capita GDP and territorial occupation areas [21] [22] [9] [13] [23] [24]. The intermediate level included malaria exposure proxies such as the extraction of açaí and Brazil nuts as well as family farm settlements [25] [26] [24]. The proximal level included individual characteristics that influence malaria risks, such as the total number of men and women with malaria, and types of occupation (agriculture, tourism, mining, vegetal extraction and traveling) [22] [3] [14] [27] [23].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Theoretical and conceptual ecological model for the relationship between socioenvironmental determinants and malaria in the Brazilian Amazon.

https://doi.org/10.1371/journal.pone.0217615.g002

Data collection

Malaria data were obtained from the Malaria Epidemiological Surveillance Information System (SIVEP-Malaria) and retrieved by the Health Surveillance Department of the Brazilian Ministry of Health (SVS/MS). The absolute number of cases diagnosed per year (2003–2013) in each of the 771 municipalities of the Brazilian Amazon were considered, and the municipality where the infection occurred was used as the territorial reference.

Potential factors associated with the occurrence of malaria were extracted from several sources and calculated according to municipality:

1. per capita gross domestic product (GDP): mean aggregate value per capita of goods and services produced in a geographic space during the target year, measured in local currency and at market prices and based on data from the Brazilian Institute of Geography and Statistics (IBGE) [20] [28];
2. Number of malaria-positive individuals related to agriculture, tourism, gold mining, vegetal product extraction, other, mining activities, traveling, and road and dam construction, obtained from the Ministry of Health (SIVEP);
3. Number of men and women with positive test results for each species of Plasmodium analyzed and total malaria rates, obtained from the Ministry of Health (SIVEP);
4. açaí production: the total amount (tons) produced from açaí extraction, obtained from the Automatic Data Retrieval System (SIDRA) of the Brazilian Institute of Geography and Statistics (IBGE) [20] [28];
5. Brazil nut production: the total amount (tons) of Brazil nuts produced from Brazil nut extraction, obtained from the SIDRA system of the Brazilian Institute of Geography and Statistics (IBGE) [20] [28];
6. human settlement: total number of settled families, provided by the regional reports by states of the Brazilian Amazon in the General Information System on Agricultural Settlements of the National Institute of Colonization and Agrarian Reform [29];
7. regionalization: consolidated settlement zone in the Central Amazon (arc of deforestation) and the south and west Hiléia settlement zones.

The study was approved in June 2015 by the Institutional Review Board of the Oswaldo Cruz Foundation (CAAE 44507015.8.0000.5241), case review 1.109.048.

Spatial mapping

The municipal digital grids for the years 2005 and 2013 were used to construct the maps. The municipal grids and the population data with geographical coordinates of the municipal seats were retrieved from the IBGE website.

Health and sociodemographic indicators were georeferenced in the municipal seats and organized in layers in a Geographic Information System environment for the construction of thematic maps.

The annual parasite incidence (API) was interpolated for spatial diffusion maps. API spatial distribution estimates were calculated using the inverse distance weighting (IDW) [30] [31] [32] geostatistical method, an application of the ArcGIS version 10 software (Esri, Redlands, CA, USA), which estimates the value of an indicator in nonsampled areas based on samples from neighboring municipal seats.

A map representing the space-time trend of malaria transmission in the Amazon region was created by interpolating the value of the epidemic year for each municipality. In this study, an epidemic onset was considered the first year in the series from 2003 to 2013 in which the number of cases exceeded the value of 10.

Statistical analysis

A spatial-temporal study on the occurrence of malaria was developed using four specific years of SIVEP data: 2003, 2005, 2010 and 2013. These years were selected to analyze the factors associated with the incidence of malaria in the year preceding an epidemic (2003), in an epidemic year (2005), in the year in which the Millennium Development Goals were defined by the federal government (2010) and in the year after the implementation of the goals (2013). The latter year was used to assess the possible impact of the MDGs on malaria prevention and control.

Since the majority of the malaria cases were caused by Plasmodium Falciparum and Plasmodium vivax, we opted to analyze the determinants of each of these two species as well as the determinants of malaria as a whole. This strategy was used because municipalities simultaneously presented overlapping Plasmodium species. Therefore, the socioenvironmental determinants of the different species were analyzed. The data were disaggregated and then recombined to monitor and analyze the overlapping malaria cases.

Three dependent variables were constructed: 1) total number of individuals who tested positive for P. falciparum; 2) total number of individuals who tested positive for P. vivax; and 3) total number of malaria cases (regardless of Plasmodium species).

The construction of study-dependent variables was based on the Plasmodium species and the various combinations reported in SIVEP. The system differentiates individuals who have been infected by a single protozoan specie or infected by more than one specie, namely, P. falciparum, P. falciparum +P. falciparum gametocytes, P. vivax, P. falciparum + P. vivax, P. vivax + P. falciparum gametocytes, P. falciparum gametocytes, Plasmodium malariae, and P. falciparum + P. malariae.

The analysis was conducted by municipality, which, in Brazil, constitutes a regional political-administrative division and, in the case of the present study, contained SIVEP records for one or more Plasmodium species, i.e., the same municipality could have more than one species of malaria. Previously, an analysis was performed to verify which were the most frequently reported Plasmodium species in the system to construct the dependent variables for the study. The distribution of data showed that in Brazil, P. vivax and P. falciparum accounted for 80.0% of the total malaria reports in SIVEP. Under this scenario, we opted to analyze which of the selected determinants were related to each of the two most common Plasmodium species in Brazil, as well as analyze how these determinants are associated with the different species in the same locality.

Three dependent variables were constructed: 1) the sum of individuals who tested positive for P. falciparum; 2) the sum of individuals whose tested positive for P. vivax; and 3) the total number of cases of malaria (regardless of Plasmodium species). In this last dependent variable, malaria case totals included P. falciparum, P. falciparum + P. falciparum gametocytes, P. vivax, P. falciparum + P. vivax, P. vivax + falciparum gametocytes, P. falciparum gametocytes, P. malariae, and P. falciparum + P. malariae.

An initial descriptive approach using bivariate analysis was followed by an analytical method using multivariate analysis with zero-inflated binomial regression.

In the descriptive approach, the proportion of municipalities with positive malaria results for, positive results for P. falciparum and positive results for P. vivax were calculated by year. In the bivariate analysis, the means with the respective standard deviations of malaria cases by year and type of regionalization were calculated.

The total number of P. falciparum and P. vivax cases according to regionalization of the municipalities was organized in a historical series from 2003 to 2013 as well as by the contextual variables of economic production (açaí extraction, Brazil nuts, and agricultural settlements).

Since the dependent variables (number of P. falciparum, P. vivax, and total malaria cases) were set as nonnegative integer count data with no upper limit, with a large numbers of zeros and large dispersion of values, we decided to apply negative binomial regression models. This model was chosen because it provided an alternative to excessive zero-count data where there is usually data overdispersion [33] [27].

On the basis of the endemic levels of malaria in the resident population of the municipalities of the Legal Amazon, a control diagram was constructed in the following stages: 1) identification of the number of years with malaria cases exceeding 10; 2) identification of the year with the highest occurrence of cases; 3) calculation of the annual average incidences as well as their respective standard deviations; 4) calculation of the endemic upper limit, considering the added mean plus two times the standard deviation; and 5) identification of an endemic outbreak defined by the statistical significance of the Poisson test using the maximum number of cases in the period and the endemic upper limit.

Regional analysis

Concerning the association between contextual variables (per capita gross domestic product (GDP), açaí production, Brazil nut production, and human settlements), the analysis was divided according to the theoretical conceptual model (Fig 1), namely, at the distal, intermediate, and proximal levels, with hierarchical modeling. Before modeling, a nonparametric correlation analysis was conducted to assess potential multicollinearity between independent variables. Contextual variables with correlations greater than 0.60 were excluded from the multivariate analysis, as in the case of hunting and fishing activities.

In the modeling strategy, all variables were introduced directly at the distal level, and only those with p values less than 0.05 were maintained in the model. A similar procedure was used for the intermediate level. For inclusion at the proximal level, only the variables from the distal and/or intermediate levels that were significantly associated with the outcome (p < 0.05) were maintained. This process was applied according to Plasmodium species and total malaria cases by region and year.

Malaria risk according to economic activity

The extractive harvesting of açaí and Brazil nuts is associated with an increased risk of malaria transmission, especially P. falciparum. According to our results, farmers or agricultural workers and travelers showed a nearly two-fold increased risk of P. falciparum infection. Meanwhile, there was a gradual decrease in malaria incidence associated with industrial mining, illegal mining, and road and dam construction, and these activities appeared to act as a protective factor against Plasmodium infection in the respective municipalities. Several international studies have shown an increased risk of malaria transmission in rural areas where people are involved in extractive activities [22] [25] [3] [14] [38] [41] [40] [42] [43] [44] [24]. In the Brazilian Amazon, a survey in 2000–2005 concluded that municipalities where 0.7% of the remaining forests were selectively exploited had a higher risk of malaria (1.72, 95% CI: 1.18–2.51) compared to areas with the highest selective logging rates, which had a lower risk (0.39, 95% CI: 0.23–0.67) (Hahn, Gangnon, Barcellos, Asner, & Patz, 2014). Several studies have also documented malaria transmission in agricultural settlements in the Amazon, emphasizing the importance of differentiated control measures according to the stage of the settlement project (recent or past implementation) and the levels of transmission in each location [22] [25] [3] [14] [38] [41] [40] [42] [43] [44] [24]. More recently, another study found evidence that proximity to forest fringes increases the malaria incidence, especially in locations close to environmental protection areas (EPA) where plant extractive activities usually occur [45].

In the south and west Hiléia settlement zone, the risk and exposure to malaria in açaí extractive or cultivated production areas varies by region. For example, in the Amazonian estuary, the risk increased during the peak harvest season in July and August. From November to May, the possible areas for malaria transmission are those with cultivated açaí in the state of Amazonas (Codajás, Itacoatiara, Careiro, Urucará, Parintins, and Maués) and with extractive açaí activities (Tefé, Coari, Codajás, and Manaquir). In Amapá, the high-risk areas are concentrated in Mazagão, Serra do Navio, Pedra Branca, Santana, and Macapá from August to December. Risk occurs year round in the municipalities of Cruzeiro do Sul, Mâncio Lima, Marechal Thaumaturgo, Porto do Acre, and Rodrigues Alves in the state of Acre, and in the state of Rondônia, the risk is associated with cultivated açaí from January to June in Porto Velho [46].

Açaí and Brazil nut extraction and cultivation are mostly concentrated in rural floodplains [47] [48] that favor the development and permanence of immature anophelines, mainly Anopheles darlingi, the principal malaria vector in the Amazon [49]. Water samples collected from forest fringes were more likely to be positive for An. darling larvae [49].

In the south and west Hiléia settlement and Central Amazon zones, the highest malaria incidence occurred in areas with Brazil nut extraction, located mainly in the states of Acre, Amazonas, and Pará, where the annual harvests occur. In Roraima, the harvest season runs from March to September since meteorological conditions differ from the other Amazonian states. The states of Maranhão, Mato Grosso, and Tocantins also have some areas with Brazil nut extraction. Health services must focus on detection in December or January, except in Roraima where detection is from March to September, due to the environmental conditions and set goals for the containment of epidemics in these extractive production zones [46].

This regional analysis of the Brazilian Amazon showed differences in the spatial distribution of malaria, clearly determined by the methods of territory occupation spatially and over time. P. falciparum and P. vivax cases associated with potential risk factors over time were used to assess malaria trends associated with occupation of the Amazonian frontier. Considering the three main zones of occupation in the Amazon, there was only a marked increase in malaria risk in the south and west Hiléia settlement zone between 2003 and 2013. In 2005, this subregion showed a 12.14 times (95% CI: 7.35–20.05) greater risk for P. vivax infection. In 2013, the increased risk was 17.51 times (95% CI: 9.23–33.21) greater in municipalities with P. falciparum cases. This overview of regions with increased risk of malaria highlights the importance of integrating evidence-based surveillance and disease control by ranges or zones of territorial occupation along the economic frontier.

This study emphasizes the importance of malaria risk according to territorial occupation zones. Each zone, with its individual economic specialization, has areas and individuals who can be affected differently by the dynamics of the health-disease process. This study highlights common characteristics of the localities that help explain how the disease behaves depending on the prevailing environmental conditions. For example, in the south and west Hiléia settlement zone (the new mobile frontier for the exploitation of natural resources), malaria’s spatial dispersion displays marked heterogeneity. In this subregion, malaria is strongly associated with deforestation and subsistence agriculture due to the unplanned development of new agricultural settlements [45].

One hypothesis suggests that the decreasing malaria incidence is associated with new forms of occupation of the economic frontier, with export-oriented market mechanisms. This is paralleled by the frontier workers’ mobility due to the lack of governmental policies for organized territorial settlement. This mobility increases the frontier workers’ vulnerability, with greater risk of exposure to infection. These findings corroborate previous studies emphasizing the role of mobility and area closure (postfrontier rationale) in the spatial dispersion of malaria in the postfrontier stage, probably due to a combination of the following: (i) expanded mechanized soybean monoculture, extensive agriculture, energy infrastructure projects (e.g., the Madeira and Belo Monte hydroelectric dams), and industrial mining [46]; (ii) the establishment of protected areas and reduced deforestation regions [50] [42]; (iii) the introduction of cattle grazing as part of established soybean and corn crop rotations [50], leading to the further concentration of land ownership and thus clearing fields and transferring infection risks to periurban areas [22]; and (iv) the development of local surveillance and health care systems [43].

Mobility, spatial spread, and malaria

An important development in the Brazilian Amazon was reduced interregional migration, suggesting that migration in the region was driven by specific projects associated with recent Federal government policies such as “Brasil em Ação” (Brazil in Action) and the Growth Acceleration Program (PAC) [19] [51]. In other words, migration in the region was determined by specific projects that potentially attracted Brazilians to the states of Amazonas, Acre, Amapá, and southwest of Pará, all of which experienced growing populations [52]. There has also been significant population growth in urban areas, especially regional urban hubs and medium-sized cities [53].

The south and west Hiléia settlement and Central Amazon zones showed a malaria risk in all the years studied. However, there was a decrease in cases due to reduced economic activity in the migratory frontiers in recent years, meaning the Brazilian Amazon had a lower rate of migration to rural areas. Highly extensive agricultural activities (livestock, soybeans, cotton) have led to a reduced labor force and reduced migration to rural areas [54]. Meanwhile, the region showed stable urbanization during the same period, with towns and cities becoming the last option for migrants (a situation further aggravated by poor urban infrastructure, work and employment conditions in the region’s towns and cities) [53].

Public services were effective in the consolidated settlement zone, which had large cities and a dense road network. This zone did not contribute to malaria spread in the region due to the reduced natural habitats for anopheline mosquito reproduction.

The current study corroborates several previous studies on malaria spread and mobility [55] [56] [57], showing that human migration to endemic malaria areas increases the risk of infection by thirty-fold. There are three stages in the spread of frontier malaria [8]: the first stage is marked by a rapid increase in malaria incidence in the early years of agricultural settlement projects. The second stage shows a progressive decline in malaria rates after three years due to improved environmental and socioeconomic conditions in agricultural settlements. Finally, the third stage is stability, with low malaria incidence as a result of urbanization, increasing sense of community, improved socioeconomic conditions, and a reduction in environmental changes. In 2003, the contiguous distribution of malaria occurred in the Central Amazon and south and west Hiléia settlement zones. This contiguous distribution and subsequent reduction were evident in the 2005 epidemic in the south and west Hiléia settlement zone; in 2013, malaria distribution was concentrated in the south and west Hiléia settlement zone, specifically in the states of Amapá, Amazonas, and Roraima [35].

In the current study, municipalities with important agricultural colonization and environmental protection areas for açaí and Brazil nut extraction were associated with the spread of malaria. In these areas with regular and intense human circulation, the proportion of recent immigrants is assumed to be large, as is the proportion of young men in the population. These municipalities are characterized by immigration and substandard living conditions; therefore, these areas are socially vulnerable to malaria and maintain endemicity with high incidence rates [22] [14] [58] [44]. Forest conservation in environmental protection areas can increase the disease prevalence in the vector species [24] [59] [60], while the risk for transmission is reduced in deforested areas of the Brazilian Amazon, which is generally covered by shrubs and secondary vegetation, where forests tend to be replaced by pastures and soybean crops [61].

The observations of the economic activities and social mobility showed that malaria risk was highest in individuals working in vegetal product extraction (açaí and Brazil nuts) and agricultural settlements, the latter specifically in the south and west Hiléia settlement zone. The highest malaria risk for vegetal extraction activities was in young men, corroborating several other studies and reflecting social determination and gender asymmetry in malaria incidence. The literature shows that work near forest areas is mostly done by men and that the mean number of malaria cases per month and per person tends to be higher in men than in women [50].

Importantly, malaria risk can be associated with situations of social stratification caused by differential access to land (i.e., workers’ activity on the economic frontier according to land tenure and transformations in soil use observed over time) and the social mobility of individuals or families. In the Brazilian Amazon, trends in human mobility are not strictly linear. Many people are in a cyclical transition between different stages of social stratification and displacement on the economic frontier, and they can often be categorized differently at different times; for example, a poor migrant with the social status of an illegal miner can become a settler or landless squatter, or a settler can transform from small farmer to medium-sized or large cattle rancher [62].

Circular migration in the Amazon is highly common. It occurs when the land, labor, services, markets and preexisting social structures are marked by the concentration of land and natural resources, fueling the marginalization of people and social inaccessibility, facilitating poor living conditions and a higher likelihood of displacement, social mobility, and the development of adverse health conditions for malaria [54] [63] [64].

Malaria control on the Amazonian frontier

The implementation of the National Malaria Control Program (PNCM) in 2003 led to fluctuations in transmission of the disease, and 2003 and 2005 were marked by periods of major epidemics followed by periods of reduced disease spread. Brazil achieved the Millennium Goal in 2015 with a reduction in malaria cases of more than 75% compared to the number of cases in 2000. In the last 15 years, the country has reported an average of 320,000 annual cases, which is 50% of all malaria cases in the Americas [1].

The PNCM intensified its malaria control activities with an important increase in the network of microscopy laboratories, a free supply of long-lasting insecticide-treated mosquito nets (ITNs), indoor spraying focused in areas with increased risk, and improved community awareness and education [65].

The P. vivax/P. falciparum ratio in Brazil increased from 3.8:1 in 2003 to 4.9:1 in 2013 and 7.8:1 in 2016. The ratio was approximately 1:1 until 1989. The introduction of combination therapies with artemisinin derivatives (ACTs) for the treatment of P. falciparum in 2007 was undoubtedly instrumental in reducing malaria cases in the country [66]. The expanded diagnostic network and inclusion of rapid diagnostic tests (RDTs) contributed to this decline by interrupting the P. falciparum transmission chain, since P. falciparum takes 7 to 10 days after the onset of symptoms to produce infectious gametocytes [39].

In 2003, the Brazilian Ministry of Health implemented an information system for the epidemiological surveillance of malaria (SIVEP-Malaria), which allows disease case monitoring and real-time updating by municipal health systems, although this does not always occur. Although the system is simple, it is not always used by municipal managers to assess the epidemiological situation and issue prompt alerts at the onset of epidemics. A study in 2010 using SIVEP data showed that epidemics occurred in 41.9% of the municipalities in the endemic area, lasted 1–4 months in 58.3% of the municipalities, 5–8 months in 24.3% of the municipalities, and 9–12 months in 17.4% of the municipalities [39]. For control and elimination of the disease, adequate use of this information system can be key for assessing the epidemiological situation, detecting the early onset of epidemics and observing their main determinants to contain outbreaks and make appropriate response decisions.

Regardless of the frontier workers’ mobility, moving from deforested areas to environmental protection areas exposes workers to different risk situations and thus requires close monitoring by malaria control policymakers.

Study limitations

The study had three important limitations. First, it drew mainly on secondary data, with the potential for inaccurate information and omission of relevant variables. More than one positive test may have been performed and reported for the same malaria case (overreporting), and some cases may not have been reported due to an incorrect diagnosis. Second, the annual parasite incidence (API) was adopted as an indicator of malaria transmission risk and was used to classify areas by an estimated disease incidence rate. The API consisted of counting positive tests rather than actual cases of the disease. Third, since the API spatial interpolation only provided an estimate of a value based on other known values, the distribution of malaria cases and low data density in some municipalities may have impacted the quality of the results. Additionally, the resolution of the interpolation grids must be consistent with the resolution of the sample grids. We emphasize that these three limitations are typical of studies conducted on regional geographic scales, demonstrating the difficulty in determining inherent relations between local dynamics and broader regional scales, the results of which may or may not be generalizable to a broader spatial context and may overlook many of the local complexities of malaria transmission.