Ethics statement

The current study used data from three national survey on prevalence schistosomiasis in schoolchildren. These data are anonymous and available for research purposes by the Brazilian government. Moreover, this study was conducted exclusively with secondary and aggregated data, publicly accessible and in accordance with resolutions of the National Health Council No. 466/2012 [12] and No. 510/2016 [13], exempt from evaluation by the Research Ethics Committee.

Study design

The epidemiological design of the research consists of a prospective study, covering three periods with observational ecological data. The outcome variable was the municipal prevalence of schistosomiasis in schoolchildren from seven to 14 years old.

Studied period and data source

Data were extracted from the three Brazilian surveys of schistosomiasis prevalence, as follows:

1. The National Helminthological Survey of Schoolchildren (IHE) (Inquérito Helmintológico Escolar) by Pellon & Teixeira, conducted from 1947–1953 in two phases [6,7]. The first phase included 11 states considered endemic for the disease, with a sampling plan that addressed locations of more than 1,500 inhabitants in which 440,786 schoolchildren were examined. In the second phase, locations of more than 1,250 inhabitants of five non-endemic states were included, and 174,192 schoolchildren were examined. In both phases, all regions of the country were sampled, except for the North region. In this way, 1,190 locations were surveyed, totaling 614,978 students examined.
2. Survey by the Special Schistosomiasis Control Program (PECE) (Programa Especial de Controle da Esquistosomose) conducted from 1975–1979 [9]. This survey consisted of a non-probabilistic sample of 327 municipalities in 18 states and areas that were disease-free or endemic, in which 447,779 schoolchildren aged 7–14 were examined. This survey took place in municipalities where the program had been implemented by the Ministry of Health and included all municipalities that adhered to PECE. The criteria for inclusion of schools and students were based on the decennial census and an active search in school classes [14,15].
3. INPEG conducted from 2010–2015 [10]. This survey also considered schoolchildren aged 7–14 by applying a cluster sampling plan, with areas categorized in three endemic levels (municipalities in non-endemic, low prevalence, and high prevalence areas) and four categories of population size (fewer than 20,000, between 20,000 and 150,000, between 150,000 and 500,000, and more than 500,000 inhabitants). Thus, samples were drawn from those stratums to determine the analyzed municipalities, elementary schools, and school classes. As a result, 521 municipalities representing all Brazilian states were analyzed. The amount of tests in each municipality ranged from 60% to 100% and in nine states it was higher than planned. In total, 197,564 schoolchildren aged 7–14 were examined.

The broad extension of the Brazilian territory affected the implementation time of each of the three surveys. Therefore, the impossibility of collecting data in just one year led to the need for around five years of data gathering for each survey. Fig 1 describes the surveys, including their respective sampling strategies. S1 Note provides additional details on characteristics of each survey and their specific features.

Download:

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

Fig 1. Descriptive flowchart of the three national surveys on the prevalence of schistosomiasis mansoni in Brazil.

IHE: National Helminthological Survey of Schoolchildren. PECE: Special Schistosomiasis Control Program. INPEG: National Survey of Prevalence of Schistosomiasis and Soil-transmitted helminth infections. DF: Federal District.

https://doi.org/10.1371/journal.pntd.0010804.g001

For obtaining intercensal estimates, data related to the explanatory variables were collected from the 1950, 1960, 1970, 1980, 2000, and 2010 demographic censuses of the Brazilian Institute of Geography and Statistics (Instituto Brasileiro de Geografia e Estatística–IBGE) and from the Institute of Applied Economic Research (Instituto de Pesquisa Econômica Aplicada–IPEADATA) (Table 1).

Download:

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

Table 1. Description of evaluated outcome and explanatory variables, periods, and data source.

https://doi.org/10.1371/journal.pntd.0010804.t001

Inclusion and exclusion criteria

The intense evolution of the political-administrative organization of Brazilian states and municipalities over the decades, reflected in the number of currently existing municipalities, led to the adoption of inclusion and exclusion criteria for this study defined as: (i) sampled municipalities with territorial delimitation compatible with the demographic censuses of each analyzed period; (ii) sampled municipalities and/or municipal districts that, even incorporated to or emancipated from other municipalities or districts during the period of the three surveys (1947–1953, 1975–1979 and 2010–2015), had available legislative and historical information on their establishment or division process; (iii) sampled municipalities that met the criterion of quality of registration. Subsequently, according to the assumed criteria, the municipalities in which it was not possible to detail the evolution of their establishment, fusion, or incorporation, as well as those lacking enough records for the explanatory variables, were excluded. Fig 1 and S2 Note show the complete description of the methodological inclusion and exclusion criteria.

Outcome variable

The outcome variable of the study was the prevalence of infection with Schistosoma mansoni in samples of schoolchildren from 7 to 14 years old per municipality (Table 1).

Independent variables

The independent (explanatory) variables consisted of coverage of water supply and sewerage, and municipalities sociodemographic and socioeconomic variables such as population size, percentages of urbanization and literacy, per capita gross domestic product, and the survey period. These variables were defined considering the factors related to infection as indicated in the literature and the context and availability of data in the information systems for each period. Relevant factors related to the disease, such as family income, coverage of deworming treatment, water treatment for inactivating human schistosome cercariae or chemical molluscicide treatment, malacological surveys, and family or school hygiene practices, could not be included since there are not enough available data from all studied municipalities in the different periods, mainly in the 1950s and 1970s.

The reference year adopted for each of the survey periods were 1950, 1977, and 2013 to facilitate notation. Projection and/or interpolation techniques were used in cases where information about the explanatory variables was not available in the reference year. For 1950, we applied a projection by the population trend method—AiBi projection or Apportionment Method—for the sanitary sewage [16,17]; and a projection by interpolation for the municipal gross domestic product (GDP) per capita using data from 1949 and 1959. For explanatory variables in the 1977 period, estimates were made using linear interpolation techniques. For 2013, interpolation estimates were performed for 2000 and 2010, and then we extrapolated linear and geometric growth for 2011, 2012, and 2013. The projections for the years 1977 and 2013 were adopted because the Brazilian census information is collected every decade, therefore, in a non-annual series [18]. S2 Note provides additional details about the techniques used for each variable, explanations of the use of the AiBi technique and the number of observations involved.

Data analysis

As the national surveys were carried out in different periods and using different strategies, the sampled municipalities were not the same for all periods, resulting in an unbalanced data panel with different municipalities in each sampling period. Based on that, we conducted a prospective study covering three periods with observational ecological data to evaluate trends in the prevalence rates of the infection over time and their associations with economic, health, and social indicators for a total of 1,721 municipalities sampled during the periods represented by the reference years 1950, 1977, and 2013.

Descriptive analyses were performed for the municipal data in each period. In the inferential analyses, multilevel statistical models were fitted to estimate the prevalence of schistosomiasis considering data from the 1,721 sampled municipalities. According to the official territorial division proposed by the IBGE, the Brazilian political-administrative organization is divided into five macro-regions, which include 26 federal units (states) plus the Federal District and 5,570 municipalities. In order to consider this hierarchical characteristic of the data, we applied Generalized Linear Mixed Models (GLMMs) with random effects related to three levels: regions (Level 1,with 5 categories South, Southeast, North, Northeast and Midwest); states (Level 2, representing the 26 federative units plus the Federal District); and municipalities (Level 3, related to the 1,398 different municipalities included in the study with observation in at least one of the three surveys). These three hierarchical levels of data were incorporated into the random intercepts of the GLMMs to allow the joint modeling of data the three periods.

The GLMMs also included fixed effects related to the independent variables previously described. We considered the Poisson and Negative Binomial distributions in the analyses, with and without zero inflation for both cases [19–22]. Thus, we allowed the modeling of different data characteristics, such as over dispersion and zero inflation. Detailed discussion on models definition is provided in S3 Note. The count of the number of positive cases in each municipality was considered the outcome variable and the total number of examined students was used as an offset, responsible for controlling the number of cases per municipality. Logistic regression was used to adjust for the excess of zero.

A backward selection procedure was used to identify the significant fixed effects, considering a 25% significance level for the removal of an explanatory variable. Thus, at each step of the analysis, the explanatory variable with the highest p-value, among those with a p-value>0.25, was removed from the model. After such a procedure, no further variable selection was necessary, as all variables retained were significant at the 5% significance level. The final regression model (Poisson or Negative Binomial distribution with or without zero inflation) was chosen according to the following criteria: (a) lower residual variance; (b) lower values of Akaike (AIC) and Bayesian (BIC) Information Criteria [23].

The software EPI INFO 7.1.1 and Microsoft Office Excel 2010 were used for database construction. Descriptive and inferential analyses were performed in the software R using the statistical package glmmTMB [24].

Descriptive analysis

Table 2 shows the results of the descriptive analysis for the prevalence of schistosomiasis. Despite the large range, the mean prevalence of infection decreased between the three analyzed periods, with 8.3% for reference period 1950 (SD 17.2), 4.8% for 1977 (SD 12.4), and 0.8% for 2013 (SD 3.5). In addition, the median and amplitude of prevalence were 0.2 and 90.9 in 1950; 0.0 and 71.2 in 1977; and 0.0 and 50.0 in 2010–2015. The percentage of municipalities with zero cases of schistosomiasis were 45.4% for 1950, 54.2% for 1977 and 73.7% for 2015.

Download:

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

Table 2. Descriptive statistics on the prevalence of schistosomiasis per 100 students and independent variables per study period in the 1,721 sampled Brazilian municipalities.

https://doi.org/10.1371/journal.pntd.0010804.t002

Table 2 also shows descriptive statistics for the explanatory variables that composed the study. They all showed remarkable increasing values between 1947–1953 and 2010–2015, especially the sanitary variables related to water supply and sewerage coverages. On average, urbanization varied from 25.6% to 68.4% (a 2.6-fold increase); literacy from 38.6% to 84.1% (2.1-fold increase); coverage of water supply network from 6.5% to 71.6% (an 11-fold increase); coverage of sewerage from 2.6% to 31.0% (an 11.9-fold increase); condition of occupancy conditions of households from 54.9% to 76.3% (a 1.4-fold increase); and GDP from 0.91 to 5.77 (BRL) (a 6.3-fold increase). Additionally, for the 41 municipalities common to the three surveys, the percentage decrease in prevalence between the 1947–1953 survey and the 2010–2015 survey ranged from 0.1 percentage point (p.p) to 77.4 p.p., with only three municipalities presenting small positive percentage difference between 0.4 and 0.1 p.p.

Table 3 shows the hierarchical (multilevel) description adopted in the study, detailing the distribution of the number of municipalities according to regions and federative units state in each analyzed period. Regarding the distribution of the studied municipalities along the five geographical regions of the country, 758 (44.0%) are from to the Northeast, 506 (29.4%) from the Southeast, 206 (11.9%) from the South, 153 (8.9%) from the Midwest, and 98 (5.7%) from the North region. The Northeast region had the highest percentages of municipalities in each survey, following by the Southeast. In 1947–1953 (n = 907), the survey included 418 (46.0%) municipalities from the Northeast, 317 (35.0%) from the Southeast, 103 (11.4%) from the South, 69 (7.6%) from the Midwest, and no samples from the North region. For 1975–1979 (n = 293), 114 (38.9%) sampled municipalities belonged to the Northeast region, 73 (24.9%) to the Southeast, 50 (17.1%) to the South, 40 (13.7%) to the Midwest, and 16 (5.3%) to the North. Finally, in 2010–2015 (n = 521), 226 (43.4%) municipalities were in the Northeast, 116 (22.3%) in the Southeast, 82 (15.7%) in the North, 53 (10.2%) in the South, and 44 (8.4%) in the Midwest regions.

Download:

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

Table 3. Hierarchical levels and the distribution of the 1,721 sampled Brazilian municipalities (Level 1) included in the study according to state (Level 2) and region (Level 3) for each period.

https://doi.org/10.1371/journal.pntd.0010804.t003

Statistical models

Because of the larger amount of municipalities with zero cases of schistosomiasis (45.4% for 1947–1953 period; 54.2% for 1974–1979 period and 73.7% for 2010–2015 period), models with and without the adjustment for excess of zeros were employed in order to verify the robustness and consistency of the analyses.

The results between the goodness-of-fit measures for the adjusted models (Poisson and Negative Binomial with and without zero-inflation) can be verified in Table A in S3 Note. The Poisson models presented lower residual variance than Binomial Negative models, being the Poisson zero-inflated specification the model with the lowest AIC and BIC values. Table 4 shows the Rate Ratio (RR) estimates for schistosomiasis infection, and the respective 95% confidence intervals (CI) obtained from the zero-inflated Poisson multilevel regression model.

Download:

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

Table 4. Results from the zero-inflated Poisson multilevel regression model fitted to assess the prevalence of schistosomiasis mansoni in the sampled Brazilian schoolchildren.

https://doi.org/10.1371/journal.pntd.0010804.t004

The explanatory variables that remained in the model of the prevalence of schistosomiasis were the natural logarithm of population size, %Urbanization, %Occupancy condition of the domicile, %Water supply, %Sewerage, and the categorical variable related to the survey period (the 1947–1953 period was used as a reference for the analysis). For the zero-inflation logistic regression, only variable %Urbanization showed statistical significance.

A negative value in the estimate of the effect of a variable indicates that an increase in its value results in a decrease in the prevalence of the infection. This was the case for the variables natural logarithm of population size (-0.148; p-value <0.001), %Urbanization (-0.009; p-value <0.001), % Occupancy condition of the households (-0.014; p-value <0.001), and % Sewerage (-0.004; p-value 0.001) in the modeling of prevalence. Based on the associated RR, the increase of one unit in the numerical value of these variables causes a decrease of 13.8%, 0.9%, 1.4%, and 0.4% in the estimated mean for the prevalence, respectively. We highlight that one unit increase in the natural logarithm scale corresponds to an increase of approximately 2.718 times in the original variable scale. On the other hand, the results showed an inverse effect on the prevalence of infection for the water supply variable, with a positive value for its estimated effect (0.010; p-value <0.001; and RR corresponding to an increase of only 0,1% per one unit increase in the numerical value of the variable). Concerning the categorical variable representing the survey periods, in comparison with period 1947–1953 (taken as reference in the regression model) a positive regression effect was estimated for 1975–1979 (0.301; p-value <0.001) and a negative effect was estimated for 2013 (-1.542; p-value <0.001). Although this result seems to indicate an increase in prevalence from 1947–1953 to 1975–1979, contradicting the descriptive analysis shown in Table 2, it should be noted that the behavior of the other explanatory variables is quite different between those periods. In fact, an analysis of the municipal prevalence estimated by the model provided similar and consistent results with those observed in the data, corroborating the adequacy of the adjusted model (see Table B in S3 Note).

The use of GLMMS allowed the joint modeling of data from all municipalities in the three sampling periods. In order to evaluate the robustness of this approach, we performed a sensitivity analysis involving the data subset composed of the 41 common municipalities between the three sampling periods (see Table C in S3 Note). The zero-inflated Poisson was the best fitted model and composed of the same explanatory variables to explain the prevalence of schistosomiasis as for the multilevel zero-inflated Poisson model presented in Table 4. The estimates of the coefficients and Rate Ratio (RR) are similar. We also performed a statistical analysis comparing the distribution of the municipalities of the three surveys according to the endemicity classification used in the sampling procedure of the third survey (2010–2015). The results indicated that, although there are differences in the form of data collection regarding the selection of municipalities, the three samples are comparable in terms of the endemicity degree criterion used in the 2010–2015 survey.

Limitations

Although the results obtained in this study came from different municipalities and in different periods, consisting of non-serial temporal trend surveys, the analysis of municipalities common to the three surveys supported the other findings (see S3 Note). Other limitations must be considered when interpreting these results. The variables were collected in different periods, and such practice of collecting old census data required a process of harmonization between variables to allow comparisons. Another limitation inherent to census data includes the availability of access-restricted information on public service facilities and not the quality and availability of WaSH services. Further exploration of other data is necessary to understand the positive association between the prevalence of schistosomiasis and the availability of drinking water networks, including the effects of supply interruptions and changes in use based on water quality or behavioral and occupational habits. Finally, we cannot make conclusions on the causality of this association due to a limitation of ecological design.

To the best of our knowledge, this is the first study that used a longitudinal epidemiological design to analyze data from national prevalence surveys covering a large period of many decades. The results showed that the prevalence of schistosomiasis infection in schoolchildren in Brazilian municipalities decreased significantly over the decades. This decrease in prevalence of infection may be associated with environmental factors, urbanization, and housing conditions, which have improved over the decades. It is noteworthy that the association with water supply should be carefully interpreted and focused on other possible factors not evaluated here, confirming the need for a systemic approach. In addition, safe sanitation sewage should be widely provided to the population at the household level and other spheres of life, such as workplaces, health centers and school environments. Other national prevalence surveys and research should be conducted more continuously to monitor the disease prevalence and its determinants over time.