Background

Neglected tropical diseases (NTDs) disproportionately affect vulnerable resource-deprived populations from various ethnic groups often inhabiting rural areas [1]. These diseases are both consequence and causes of poverty, giving rise to “poverty traps” [2]. The global budget to control NTDs has historically been restricted. As part of the Millennium Development Goals, a renewed plan to eliminate NTDs has been launched, and Chagas disease is among the priorities for the Americas [3]. The Gran Chaco ecoregion, which mainly spreads over sections of Bolivia, Paraguay and Argentina, is a hotspot for several NTDs, including Chagas disease [4].

Trypanosoma cruzi, the etiological agent of Chagas disease, is transmitted by several routes: triatomine insects, which is the most important route in endemic regions [5]; transplacental or vertical route, the main source of new cases in non-endemic areas [6]; ingestion of food or beverages contaminated with infected triatomine feces [7], never reported in Argentina so far [8]; and blood transfusion and organ transplant, which have been blocked in the Americas by strict donor control [9]. Prevention of vector-borne transmission has mostly relied on insecticide spraying of houses since the 1950s [10,11,12]. Several regional initiatives have been launched to interrupt domestic transmission of T. cruzi. The degree of success has been heterogeneous throughout the continent, though a strong reduction in the geographic distribution of the main domiciliated vectors has been achieved [13]. However, insecticide resistance and rapid house reinfestation from untreated areas or houses have challenged control efforts in the Bolivian and Argentine Chaco regions[14].

A key step to diminish the burden of Chagas disease is to detect and treat T. cruzi-infected people, especially those < 18 years of age, who are eligible for treatment according to the current consensus [15]. Although treatment of chronic adults to prevent disease progression is still controversial, the new guidelines recommend treating women of childbearing age to block transplacental transmission and reduce the probability of acting as sources of T. cruzi to the vectors [16]. However, access to serodiagnosis and treatment has been historically poor, especially for the affected rural populations [17,18,19]. In remote rural areas, where access to the health system is scarce, strategies relying on house-to-house visits to perform blood collection for serodiagnosis were able to achieve high coverage proportions [17, 20,21,22].

Several studies have identified risk factors of human infection with T. cruzi at different levels: individual, household, social group or community, and landscape or region [23,24,25,26,27,28,29,30,31]. In general, duration of exposure, the presence and abundance of infected triatomines in domiciles, building materials of walls and roof, and presence and number of infected dogs and cats and of other infected human residents in the household (i.e. infected co-inhabitants) were associated with human T. cruzi infection in several scenarios, including the Gran Chaco region, where domestic transmission is mediated by Triatoma infestans (see [32] for a review). Risk factor analysis has been used to construct indices with the purpose of identifying individuals with higher infection risk. Some studies have relied on a few selected risk factors, such as infected vector densities, evidence of domestic infestation, self-reported inhabiting an infested house, building materials and T. cruzi-infected co-inhabitants, mother or father [33, 34]. Others used complex modeling with multiple sources of information and spatial analysis to identify households with higher risk of harboring an infected child [35]. To our knowledge, a simple-to-use, on-field risk index aimed to identify households with infected children, rather than infected individuals, has not yet been developed.

This work is part of a long-term research program on the eco-epidemiology and control of Chagas disease launched in Pampa del Indio municipality (Argentine Chaco region) in 2007 [22, 36]. Two program goals were to eliminate T. infestans from all rural houses and to interrupt vector-borne transmission to humans by combining sustained vector control and surveillance with community participation, serodiagnosis and etiological treatment. The rural section of the municipality was divided into four areas for operational purposes (arbitrarily named Areas I–IV, Fig. 1). High levels of house infestation with T. infestans (range across areas: 14.4–41.4%) were recorded at baseline; a strong decline was achieved within 2 years of community-wide insecticide spraying and surveillance [22, 37,38,39]. Trypanosoma cruzi infection in T. infestans, dogs and cats was higher in Qom (the main ethnic group) households than in creole ones [36, 40], though infestation and human infection were not associated with household ethnicity when adjusted for other factors [41, 42]. The potential burden of Chagas disease was high as evinced by the levels of human seropositivity for T. cruzi recorded in extensive serosurveys covering Areas I and III [41]. Vector-borne transmission to humans was virtually interrupted shortly after the onset of the program, with no human incident cases detected among stable residents by 2016, and infection in only one child was attributed to vertical transmission [43].

Fig. 1
figure 1

a Map of the study houses and communities from Areas II and IV shown as colored-shaded polygons. b Location of Pampa del Indio Municipality (star) within Chaco Province and the Gran Chaco ecoregion. CM, Colonia Mixta; CQ, Campo Cacique; CY, Campo Alemany; EP, Ex-Parque; H2, La Herradura; LA, Cancha Larga; LC, Lote Cuatro; LM, Las Muñecas; ME, Campo Medina; NU, Campo Nuevo; OM, Pampa Ombú; TCZ, Tacuruzal

In this study, we aimed to (i) determine the seroprevalence of human infection with T. cruzi in a well-defined rural population composed of creole and Qom people inhabiting Areas II and IV; (ii) identify risk factors associated with human infection; (iii) develop two simple risk indices to identify households harboring T. cruzi-infected children under 18 years of age; and (iv) test for spatial aggregation of houses harboring infected children. Given the strong decline of domestic infestation [22, 37,38,39] and the interruption of vector-borne transmission to humans [43], we hypothesized that children born after the onset of sustained vector control efforts would be seronegative for T. cruzi. In residents born before interventions, infection would be positively associated with the duration of exposure to triatomines and domicile infestation, belonging to the Qom ethnic group and the presence of T. cruzi-infected co-inhabitants. Infection is expected to be negatively related to the goat-equivalent index (a weighted average of livestock owned by the household in terms of goat biomass) and household educational level, which are surrogates of better socio-economic conditions [39, 44]. Households harboring infected children were expected to be spatially clustered because of the spatial aggregation of human infection [35, 43] and house infestation with T. infestans pre- and post-spraying [22, 37, 42, 45]. We also described socioeconomic characteristics of Area IV; those for Area II were reported elsewhere [37].

Methods

Study area

This study was conducted in the municipality of Pampa del Indio (26°2′0″S, 59°55′0″W), Libertador General San Martín Department, Chaco Province, Argentina. Pampa del Indio is located in the transition zone between the dry and humid Chaco (Fig. 1). According to the national census in 2010, 15,287 people lived in 3862 houses [46]. Fieldwork took place in Areas II and IV, which comprised 409 and 250 inhabited houses as of 2016, respectively, in the southern part of the municipality. The study area also comprised two small villages in the northern section (Campo Cacique, CQ, and La Herradura, H2), which had been intermittently inhabited because of occasional flooding; these houses were sprayed with insecticide along with those in Area IV.

Field interventions included a house-to-house survey and a community-wide insecticide spraying of the rural area of the municipality, which started at Area I in late 2007. These interventions were gradually scaled up to the rest of the municipality until achieving full coverage of rural sections in 2010. House infestation was periodically monitored thereafter. A full description and analyses of the effects of the interventions on house infestation, triatomine abundance and T. cruzi infection were described elsewhere [22, 36]. Briefly, in 2008, a systematic cross-sectional survey of house infestation was carried out in one of every three houses of Area II. Immediately after the baseline survey, house spraying with pyrethroid insecticide was implemented area-wide. In Area IV, the baseline vector survey was conducted in November 2009, May and November 2010 and was followed by community-wide insecticide spraying. Any house found infested with T. infestans during the surveillance phase was selectively sprayed with insecticides. In 2016 (i.e. 7–9 years after program onset), house infestation with T. infestans dropped below 1% [22].

Study design

This study complies with the STROBE checklist (Additional file 1). Several cross-sectional serosurveys aiming at full coverage of rural residents were carried out in Areas II and IV over 2014–2017. We additionally collected socio-demographic information during house infestation surveys for risk factor analysis of human infection. A timeline of key activities is shown in Fig. 2.

Fig. 2
figure 2

Timeline of key activities conducted at the study area. Pink and yellow boxes are vector surveys for Area II and IV, respectively. Purple boxes are the serosurveys at both areas

Vector and socio-demographic surveys

We conducted several house-to-house surveys to georreference each house, assess house infestation status by timed manual searches with a dislodging aerosol performed by vector control personnel [42] and record household environmental and sociodemographic features during 2008–2018 in Areas II and IV (Fig. 2). A house-to-house census of all rural households and residents was conducted in 2011 and 2014 for Area II and in 2015, 2016 and 2018 for Area IV. We recorded the full name of each household resident, age, relationship with the household head, permanent or occasional residence, alternative place of residence and whether they had previously been tested for T. cruzi infection and the outcome if known. We also scored each house as suitable for triatomine colonization or not based on the building materials of walls (mainly mud, thatch, sticks, brick with mud joints) or roofs (tarred-cardboard sheets, sticks or thatch, among others) and the presence of wall cracks as reported elsewhere [37]. Additionally, we recorded ethnic background for each household based on self-perception; whether they spoke Qomlataq language; participated in traditional Qom organizations or lived in ancestral indigenous territories. When in doubt, we checked with local health care personnel of Qom descent.

Serosurvey

Five serosurveys were conducted in Areas II and IV in cooperation with the local hospital personnel during 2014–2017 (Fig. 2). Although all residents from Areas II and IV were eligible for serodiagnosis, we especially targeted children < 18 years of age. The exclusion criteria were: (i) children < 9 months of age since maternal antibodies to T. cruzi may still be detectable; (ii) children < 18 years of age without a parent or guardian available to provide written consent and (iii) adults unable to give consent.

We organized 10 informative meetings at different locations (i.e. local primary schools, healthcare posts, churches) to maximize participation across the study area. Invitations were sent through written notes to parents of schooled children, local radio interviews and primary healthcare agents. When necessary, local healthcare agents of Qom descent explained the information in Qomlataq language. After each meeting, a day and place were set to perform blood sample collection. At least one session was programed at each school or primary healthcare post in 2014 and 2015. House-to-house venipunctures were carried out in 2016 and 2017 to further increase diagnostic coverage. We asked all participants their full names, gender, ethnicity, mother’s name, maternal serostatus for T. cruzi (if known), actual and past residence place, name of the household head, if they recalled having inhabited an infested house and having received a blood transfusion, results of previous serodiagnosis for T. cruzi infection (if any) and whether the individuals that reported seropositivity have received etiological treatment for T. cruzi infection.

Prior to this study, 2064 people from Area I and III were serodiagnosed. The composite prevalence was 27.9% (95% CI = 26.0–29.9%). We calculated that the minimum sample size required to detect a prevalence of 27.9% with an error of 10%, including the finite population correction, was 757 inhabitants from Areas II and IV [47].

Serodiagnosis

Up to 5 ml of blood was extracted by venipuncture from all participants aged ≥ 2 years. A retractile lancet was used for younger children, and the maximum blood volume obtained was 0.5 ml. Each serum sample was tested by at least two enzyme-linked immunosorbent assays (ELISA); one included T. cruzi epimastigote lysate (Chagatest, Wiener) and the other recombinant antigens (ELISA Rec V3.0, Wiener). Each sample was tested in duplicate by each ELISA test as previously reported [17]. A sample was considered seropositive for T. cruzi when it was reactive for two tests. Discordant samples were tested by indirect hemagglutination test (IHAT), ELISA and immunofluorescence antibody test (IFAT) at the National Institute of Parasitology “Dr. Mario Fatala Chabén” (ANLIS-Malbrán, Buenos Aires, Argentina) for final outcome.

Following national guidelines, all T. cruzi-seropositive people aged ≤ 18 years were referred for etiological treatment to the local hospital as reported elsewhere [17]. The local physicians also considered T. cruzi-seropositive people between 18 and 21 years of age and other seropositive adults who requested to be treated eligible for etiological treatment.

Data analysis

Human infection data were merged with the georeferenced database that included house infestation and socioeconomic variables; the latter was available for Area II [37] and was compiled for Area IV in this study. Since the dates of intervention onset differed among communities (i.e. started in August 2008 for Area II and in November 2009 for CM, CY and LM, May 2010 for EP and November 2010 for CQ, H2 and TCZ), we estimated the time span (in years) each tested person was potentially exposed to triatomines by subtracting the time elapsed since the first insecticide spraying campaign at the individual’s community of residence to the age at examination [43]. We assumed that triatomine exposure occurring after control intervention was negligible compared with exposures before interventions based on the very low prevalence of domestic infestation after community-wide insecticide spraying [22, 36]. Therefore, all children born after the onset of interventions were assigned 0 years of exposure to T. infestans. To minimize the occurrence of missing values, whenever the maternal serostatus for T. cruzi infection was not reported we used the outcomes from our serosurveys; 318 participants reported maternal serostatus while we obtained serological information for 911 people.

We used Fisher’s exact tests and t-tests to compare proportions and means, respectively, and employed 95% Agresti-Coull confidence intervals or Wilson confidence intervals when the sample size was ≤ 40 (95% CI) [48]. Agreement between self-reported and observed serodiagnosis of T. cruzi infection was assessed using the kappa index.

Risk factors

Generalized linear mixed-effects models (GLMMs) with a logit link function were used to evaluate the relationship between human infection and a priori selected predictors. We excluded 148 serodiagnosed people who had missing data in at least one variable from analysis. Analyses were performed with two different datasets to evaluate whether the risk factors differed between the complete population and children. Dataset 1 included all serodiagnosed inhabitants from Areas II and IV without missing data (N = 1189); dataset 2 (a subset from the first one) only included residents ≤ 18 years of age at baseline who had been born before the intervention program (N = 492). Based on previous evidence and preliminary analyses, the selected predictors were: duration of exposure to triatomines (in years, range: 0–78); household ethnicity (creole or Qom); area of residence (II or IV); the interaction between duration of exposure and ethnicity; self-reported inhabiting an infested house: whether the person (parents or guardians) recalled having ever inhabited an infested house; whether the mother was seropositive for T. cruzi infection (combining reported and tested serostatus as described above); number of T. cruzi-infected co-inhabitants detected by laboratory serodiagnosis performed in the current study (a three-level categorical variable: 0, 1–2 and 3 or more T. cruzi-infected co-inhabitants), whether the domicile had suitable walls for triatomines (yes or no); and time of residence at the current house up to 2016 (a three-level categorical variable: < 3 years, between 3–10 years and > 10 years). The household ID was included as a random factor. As maternal serostatus and the number of infected co-inhabitants were highly positively correlated (P < 0.001), for the first dataset, we included the number of infected co-inhabitants (which had no missing values), whereas for the second dataset, we included maternal serostatus.

Additional socio-demographic variables were not included in the multiple regression models because they exhibited high frequency of missing data or high correlation between variables. The goat-equivalent index quantifies the total number of livestock and poultry owned by the household in terms of goat biomass and is a household-level surrogate of wealth [45]. The goat-equivalent index was scaled by the biomass of 10 goats. The household educational level was computed as the mean number of schooling years attained by all house residents > 15 years old. Households with three or more occupants per room were under critical overcrowding [46]. The goat-equivalent index, educational level, critical overcrowding and time of residence were correlated; therefore, we only included time of residence in the models since it had fewer missing values. All numerical variables were centered.

Model 1 (dataset 1) was:

T. cruzi serostatus ~ duration of exposure + ethnicity + area + ethnicity \(\times\) area + self-reported inhabiting an infested house + number of infected co-inhabitants + time of residence + suitable walls for triatomines ǁ (1 + Household ID).

Model 2 (dataset 2) was:

T. cruzi serostatus ~ duration of exposure + ethnicity + area + ethnicity \(\times\) area + self-reported inhabiting an infested house + mother serostatus + time of residence + suitable walls for triatomines ǁ (1 + Household ID).

We used a multi-model inference approach to account for the uncertainty of model selection. Akaike’s information criterion (AIC) was used to identify the top-ranking models, i.e. the ones with a difference between the models AIC and the lowest AIC-scored model (ΔAIC) ≤ 2 AIC. The goodness of fit was assessed with the Hosmer-Lemeshow test and the area under the curve (AUC) of the receiver-operating characteristic (ROC) curve. The optimal threshold considered the value that minimized the total misclassification (i.e. false positives + false negatives). The H-index was used to assess classification performance penalized by the relative severity of one type of error over another: misclassifying a seropositive person as seronegative (false negative) was twice as costly as misclassifying a seronegative person as seropositive (false positive, i.e. severity ratio 0.5). All analyses were run in R 4.2.2 [49] using the packages “lme4” [50], "MuMIn” [51], "ResourceSelection" [52], “car" [53] and “hmeasure” [54].

Additionally, we constructed a simpler model to assess the association of T. cruzi infection and ethnicity, area and their interaction, accounting for exposure time, and the household ID. The data set for this analysis included all the participants (n = 1337) since all had record of age, ethnicity and house of residence (i.e. no lost cases due to missing data). This model was: T. cruzi serostatus ~ exposure time + ethnicity + area of residence + ethnicity x area of residence + (1|Household).

Catalytic model

The mean annual force of infection (FOI) was estimated retrospectively using a catalytic model with a recovery rate set to 0 [55]. The incidence of infection is assumed to be constant over time and age; therefore, children born after the onset of vector control efforts were excluded. The model also assumes that no lag occurred between human infection and infectiousness, individual hosts are homogeneously exposed, and the observed prevalence is at equilibrium. The model was FOI = -ln(1-pa)/a, where “a” is the midpoint of the age class, and “pa” is the seropositive proportion for that class. For this analysis we used 13 age classes and estimated the average FOI for the overall population and for the four categories of ethnicity and area of residence. Additionally, we calculated an average FOI for children ≤ 18 years old at the onset of interventions with nine age classes.

Index for risk stratification

We developed two simple indices to identify households with at least one seropositive resident ≤ 18 years of age, while minimizing the potential loss of households with infected children. We selected potential variables for the indices based on the outcomes of the risk factor analysis for individuals described above, the univariate and GLMM. For the index development variables, we considered that they were significantly associated with individual child infection for T. cruzi, or had high RI in the GLMM (RI ≥ 0.5), could be re-categorized to binary variables and could be imputed at the household level (excluding exposure time for example). Then, we performed univariate analyses (Chi-square test) with these variables. The eligible variables were: (i) self-reported inhabiting an infested house ever in their lifetime; (ii) suitable walls for triatomine colonization in domiciles; (iii) whether the mother of resident children was T. cruzi-seropositive; (iv) recent residence at the current house (< 3 years); (v) co-inhabiting with a T. cruzi-seropositive adult person (i.e. excluding infected co-inhabitants < 18 years old). Since access to serodiagnosis is limited in some Chagas disease-endemic regions, one index contained variables obtained from serodiagnosis, whereas the other did not to account for this limitation.

Spatial analysis

To test whether the household number of T. cruzi-seropositive children ≤ 18 years of age at baseline was randomly distributed, we performed a global spatial analysis using the weighted K-function [56]. To evaluate the occurrence of aggregation of houses harboring seropositive children (i.e. hotspots), we performed local spatial analysis using the Getis Gi*(d) statistic [57]. Spatial analyses were implemented using the “spatstat” and “spdep” packages in R [58, 59]. The spatial analysis included 472 of 484 households with children ≤ 18-years old. We excluded six houses from CQ (a village with no spatial contiguity with the rest of the study area) and six houses that lacked georeferenced data.

Results

Population census

We registered 2793 inhabitants in both areas at the last census in 2018. Area II was the most populated one, with 1866 inhabitants residing in 380 houses, while we registered 927 people in 250 houses of Area IV (Additional file 2). The population was biased to males in both study areas (Chi-square test, χ2 = 31.34, df = 1, P < 0.001, male:female = 1.2 for Area II; χ2 = 26.11, df = 1, P < 0.001, male:female = 1.3 for Area IV). Mean age was not significantly different between areas (mean ± SD: 27.9 ± 21.4 and 27.2 ± 19.9 for Areas IV and II, respectively, t-test, t(1423) = 0.80, P = 0.43). However, Area II had a significantly higher percentage of people < 18 years old (42.2%) than Area IV (38.0%) (Fisherʼs exact test, P = 0.03, OR = 1.19, 95% CI = 1.01–1.41, Additional file 3). In both areas, the mean age of creoles was higher than that of Qom residents (t-test, t(573) = 8.80, P < 0.001 for Area II, t(760) = 5.61, P < 0.001 for Area IV, Additional file 3).

Socioeconomic and demographic features of Area IV

We enumerated 250 inhabited houses in the 503 km2 of Area IV in 2016. Population and house densities were 1.8 residents and 0.5 houses per km2, three times lower than in Area II (5.3 and 1.5, respectively). Area IV encompassed a significantly lower percentage of Qom households (34.4%) than Area II (70.9%) (Fisher’s exact test, P < 0.001, OR = 0.22, 95% CI = 0.15–0.31) (Additional file 2). For most of the socio-demographic variables analyzed in both areas, we observed more differences between ethnic groups than between areas, with a few exceptions (Additional file 3). Qom people in Area IV reported the shortest mean residence time. Qom people in Area II attained the lowest household educational level. The use of domestic insecticides was more frequent in both creole and Qom households from Area IV than Area II (Fisher’s exact test, P < 0.001, OR = 2.6, 95% CI = 1.5–4.7 for creole, P < 0.001, OR = 5.2, 95% CI = 1.9–16.3 for Qom). The median goat-equivalent index for Qom households was similar between areas (1.0, Q1-Q3 = 0.2–7.1 for Area II and 0.8, Q1-Q3 = 0.2–3.5 for Area IV). Creoles from Area IV (9.7, Q1-Q3 = 0.7–65.7, Additional file 2) reported a lower median goat-equivalent index than creoles from Area II (26.8, Q1-Q3 = 1.6–94.1), which is the lowest attained by creole households across the municipality (Kruskal-Wallis, χ2 = 14.64, df = 3, P = 0.002) [41, 45].

Survey coverage

In total, 1446 people were examined for antibodies to T. cruzi. Of them, 1337 (93%) resided in the study area; the remaining were mostly from other rural or (peri)urban areas of Pampa del Indio municipality. Diagnostic coverage (overall, 47.9%) was significantly higher for women (52.8%, 95% CI = 50.1–55.7) than for men (43.8%, 95% CI = 41.4–46.3, Fisher’s exact test, P < 0.001, OR = 1.4, 95% CI = 1.2–1.7), implying an enriched sample of women compared with the census (Fisher’s exact test, P = 0.005, OR = 1.21, 95% CI = 1.06–1.38; Additional file 3). Diagnostic coverage for children ≤ 18 years of age was 55.4% (95% CI = 52.5–58.2); they were overrepresented in reference to the census (Fisher’s exact test, P < 0.001, OR = 1.3, 95% CI = 1.1–1.5). For women of childbearing age, the coverage was 56.5% (95% CI = 51.9–61.0). The lowest coverage was observed in children born after the onset of interventions (31.4%, 95% CI = 27.0–36.1), adults > 60 years of age (30.8%, 95% CI = 23.8–38.8) and young men aged 15–25 years (41.8%, 95% CI = 35.3–48.6). Overall, the examined population was enriched for Qom people (Fisher’s exact test, P < 0.001, OR = 1.4, 95% CI = 1.2–1.6) and was younger than the censused population (t-test, t(2852) = 2.14, P = 0.03; Additional file 3). A total of 369 (53.7%) households had at least one participant in the serosurveys whereas 280 (57.9%) households had at least one person aged ≤ 18 years tested for T. cruzi antibodies.

Self-reported and serodiagnosed T. cruzi infection or infestation status

A small fraction (13.2%) of the 1337 participants reported having a previous serodiagnosis for T. cruzi infection. Reported serostatus agreed with our serodiagnosis results in 82.4% of cases, i.e. a moderate agreement (κ = 0.53, 95% CI = 0.40–0.65). There were 31 incongruent cases. Eight of the 24 people who reported being seropositive but tested seronegative were children of whom we have records of having been treated for T. cruzi infection after testing seropositive; these eight cases probably indicated chemotherapeutic success rather than incongruent serodiagnostic reports. Additionally, 15 participants reported being seronegative but tested seropositive. We were not able to corroborate these diagnoses (no access to when, where or which serological tests were employed). Therefore, incidence of infection cannot be ruled out. We compared the reported and observed maternal serostatus for 186 participants and found a large fraction (84.0%) of congruent records, with a substantial agreement (κ = 0.67, 95% CI = 0.57–0.78). In total, 877 (65.6%) participants self-reported the presence of T. infestans at their residence in previous years; 362 (27.1%) people inhabited a house where T. infestans had been captured during the intervention program, and 76.8% reported exposure to triatomines.

Seroprevalence of T. cruzi infection

The global seroprevalence of T. cruzi infection was 24.8% (95% CI = 22.6–27.2, n = 1337), and the age-adjusted prevalence was slightly higher (28.3%; 95% CI = 26.6–30.1). For people born pre-intervention, infection increased with age from 7% in children < 5 years to 39% in those aged 16–20 years, and thereafter fluctuated around 40% (Fig. 3a). Since there were 513 children ≤ 18 years of age without serodiagnosis and the observed seroprevalence of T. cruzi for this age group was 13.9%, we estimated that, on average, 71 (95% CI = 57–88) seropositive children probably remained without access to serodiagnosis and etiological treatment. Creoles from Area IV exhibited the highest seroprevalence (31.6%; 95% CI = 25.5–38.5); it was significantly higher than that of creoles (20.7%; 95% CI = 15.3–27.3) and Qom from Area II (24.1%; 95% CI = 21.1–27.1) but was similar to the seroprevalence of Qom people from Area IV (27.5%; 95% CI = 18.4–30.6; Table 1).

Fig. 3
figure 3

Age-specific seroprevalence of Trypanosoma cruzi infection in residents from Areas II and IV born before the onset of interventions, observed (dots) and 95% CI (whiskers) and predicted according to an irreversible catalytic model seroprevalence (dashed line and dark envelope). a For all ages and b for children ≤ 18 years old at the onset of interventions

Table 1 Seroprevalence of Trypanosoma cruzi infection, odds ratio and estimated force of infection (FOI) according to household ethnicity and area of residence, Pampa del Indio, Chaco

The estimated mean annual force of infection (FOI) was 1.23 per 100 person-years (95% CI: 0.88–1.66; n = 13 age groups) and for the children ≤ 18 years old at the onset of interventions was 2.52 per 100 person-years (95% CI: 2.21–2. 84; n = 9 age groups). For the catalytic model including all ages, the expected proportion of infected people differed from the observed ones in the age classes < 10 years and 16–25 years (Fig. 3a), while the expected values according to the catalytic model for children ≤ 18 years of age did not differ from the observed seroprevalence values (Fig. 3b). Creoles from Area IV and Qom people from Area II experienced a two-fold higher force of infection (1.66, 95% CI: 1.13–2.35; 1.37, 95% CI: 0.94–1.90, respectively) than creoles from Area II, who experienced the lowest force of infection (0.74, 95% CI: 0.30–1.23). Qom people inhabiting Area IV experienced an intermediate FOI (1.16, 95% CI: 0.81–1.58, Table 1).

Multiple logistic regression analysis of human infection revealed statistically significant effects of exposure time (OR = 1.05, 95% CI = 1.04–1.06, P < 0.001), operational area (Area IV OR = 2.6, 95% CI = 1.29–5.25, P = 0.008) and household ethnicity (Qom OR = 2.30, 95% CI = 1.27–4.18, P = 0.006) and a significant interaction term between operational area and household ethnicity (Area IV x Qom OR = 0.34, 95% CI = 0.14–0.83, P = 0.02). The adjusted OR of human infection for creoles from Area IV and Qom from Area II were 2.6 and 2.3 times greater than the creoles from Area II, respectively (Table 1).

We detected one (0.8%) T. cruzi-seropositive child out of 122 children born after the onset of interventions (Table 2). Her mother was also seropositive for T. cruzi and reported no previous etiological treatment. No triatomine was found at their house in any of the four surveys conducted during the follow-up period. Considering that most children born after the onset of interventions were apparently not exposed to infected T. infestans at their houses (Table 2) and 49 children born after intervention onset had an untreated T. cruzi-seropositive mother, the probability of vertical transmission was 2.0% (1 of 49, 95% CI = 0.04–10.7).

Table 2 Seroprevalence of Trypanosoma cruzi infection in children born after the onset of interventions, exposure to Triatoma infestans as determined by timed-manual collections over 2008–2016 and maternal seropositivity for T. cruzi, Areas II and IV, Pampa del Indio, Chaco, Argentina

Risk factors

In the first dataset, univariate analyses showed several variables that were positive and significantly associated with human T. cruzi infection: exposure time, being born before program onset, the number of seropositive co-inhabitants, maternal seropositivity, self-reported blood transfusion, domestic abundance of T. infestans at baseline, self-reported inhabiting an infested house, lower household educational climate, shorter time of residence, presence of suitable walls for triatomines and location of past residence (Additional file 4).

Using a GLMM approach, we identified seven top models for the first dataset (Model 1) and retrieved five top models for the second dataset (Model 2). In Model 1, the variables with higher relative importance (RI) were duration of exposure to triatomines, the number of infected co-inhabitants, time of residence at the current house and suitable walls for triatomines (Table 3). The risk of infection increased by 4% for each year of exposure; was five times higher for individuals with three or more infected co-inhabitants than for individuals with none; was higher for people living in houses with suitable walls for triatomines. People residing in their current house for > 3 years experienced a lower infection risk compared to more recent residents. Self-reported inhabiting an infested house (P = 0.15), ethnicity (P = 0.13) and area of residence (P = 0.15) were not significantly associated with human infection when adjusted for other predictors. The interaction between ethnicity and area was significant but had a low RI (0.35). The AUC of Model 1 was 0.77 and the H-index 0.21; the sensitivity of the averaged model was 72.0%, specificity 67.9% and explained deviance was 22.3%. The model exhibited a poor fit to the data according to the Hosmer-Lemeshow test (χ2 = 31.0, df = 8, P < 0.001). This is probably due to the high number of observations (> 1000), lack of saturation of the model or a non-linear relationship between variables.

Table 3 Generalized linear mixed model (logit link function) of human seropositivity for Trypanosoma cruzi infection for the total population (Model 1, n = 1189) and for children aged ≤ 18 years born before control interventions (Model 2, n = 492) from Areas II and IV, Pampa del Indio, Chaco. RI relative importance, OR Adjusted odds ratio; 95% CI 95% confidence interval

In Model 2, the variables with higher RI were duration of exposure to triatomines, maternal seropositivity to T. cruzi, time of residence and the presence of suitable walls for triatomines. Infection risk was three times higher for children born to a seropositive mother. A longer residence time at the same house was negatively associated with infection, as in Model 1. Duration of exposure to triatomines was positively associated with children infection. We observed a two-fold higher risk of infection for children living in houses with suitable walls for triatomines than for children living in houses without them, with marginal significance (P = 0.1). Ethnicity, area, their interaction and self-reported inhabiting an infested house were either not statistically significant or had low RI (Table 3). The random factor explained 32.8% of the deviance. The AUC was 0.96 and the H-index 0.68. The sensitivity of the averaged model was 91.3% and specificity 88.8%; the explained deviance was 60.8%, and the Hosmer-Lemeshow test was statistically significant (χ2 = 22.7, df = 8, P = 0.004).

Identification of households with infected children

Following the risk factor analysis at individual level, we seek to identify the households harboring infected children. Based on the outcome of the univariate and GLMM risk factor analysis, the variables that significantly discriminated the 74 households with infected children were: self-reported residence in an infested house (Chi-square test, χ2 = 13.4, df = 1, P < 0.001), presence of suitable walls for triatomines (Chi-square test, χ2 = 4.1, df = 1, P = 0.04), maternal seropositivity (Chi-square test, χ2 = 25.5, df = 1, P < 0.001) and co-inhabiting with at least one seropositive adult person (Chi-square test, χ2 = 9.3, df = 1, P = 0.002) (Table 4).

Table 4 Variables evaluated for the development of risk indices to identify households with Trypanosoma cruzi-infected children ≤ 18 years old

A complete dataset to develop the indices was available for 280 households harboring children ≤ 18 years of age. For Index 1, we selected two variables with high sensitivity: self-reported residence in an infested house and the presence of suitable walls for triatomines. These variables identified 88% of the households (247/280) as having a high risk of harboring an infected child. Sensitivity and specificity were 97.3% and 15.0%, respectively (Table 4). The positive and negative predictive values were 29.1% and 93.9%, respectively. Index 2, which included self-reported residence in an infested house and maternal serostatus as predictors, identified 81% of the households as of high risk (227/280). Sensitivity was 98.6% and specificity 25.2%. The positive and negative predictive values were 32.2% and 98.1%, respectively.

Two households were false-negative cases for Index 1. In one household that reported no exposure to triatomines, we had collected domestic T. infestans at baseline. In the second household, the mother was seropositive, and this case was correctly classified by Index 2. Only one household with an infected child was incorrectly classified by Index 2. This household harbored two families: one had seronegative children and negative maternal serostatus, while the second family had a seropositive child but exposure to triatomines and maternal serostatus were unknown. As the household was classified according to the attributes of the first family, this led to a false-negative outcome.

In total, 204 households were excluded from previous analysis since no child ≤ 18 years of age had been tested serologically. Using the information available for Index 1 and 2 for those households excluded, we estimated that 91 and 37 households were at high risk of harboring infected children, respectively. According to the positive predictive values of each index, 26 and 11 households would harbor at least one infected child lost for diagnosis and treatment, respectively.

Spatial analysis

Using global spatial analysis, we found that houses harboring infected children ≤ 18 years of age at baseline displayed no significant clustering for all study radii (14–7400 m, Additional file 5). Local spatial analysis of the number of infected children ≤ 18 years of age at baseline identified three significant hotspots encompassing 54 households within 100–4100 m radius. Most (80%) of these households occurred in a densely populated section of Lote Cuatro village in Area II (Fig. 4a). Two other hotspots were identified: one at the southwest corner of the municipality and the other close to urban Pampa del Indio. The three hotspots of houses harboring infected children overlapped with or were close to previously detected hotspots of T. infestans abundance at three interventions periods (Fig. 4b-d) [22].

Fig. 4
figure 4

a Spatial distribution of households harboring Trypanosoma cruzi-seropositive children ≤ 18 years of age at baseline, Area II and IV, Pampa del Indio, Chaco. Yellow circles indicate local hotspots. Hotspots of households harboring T. cruzi-seropositive children and triatomine house abundance at b baseline, c early period of intervention and d late period of intervention as defined in [22]

Discussion

The global age-adjusted prevalence of infection with T. cruzi of Areas II and IV inhabitants remained high at 28.7%, 6–9 years after the onset of sustained vector control interventions in Pampa del Indio. Infection was positively and significantly associated with duration of exposure to triatomines, the number of seropositive co-inhabitants, maternal serostatus, presence of suitable walls for triatomines and recent residence at the current house. The interaction between ethnicity and operational area had weak effects when adjusted for other predictors in the full models. The sustained vector control actions apparently prevented the occurrence of new infections since no tested child born after program onset was seropositive, except one who may have acquired T. cruzi from the seropositive mother as recorded elsewhere [43]. The two risk indices showed high sensitivity (97.3–98.6%) at the expense of low specificity (15.0–25.2%), and had nearly perfect negative predictive values (93.9–98.1%) when applied to the identification of households harboring infected children ≤ 18 years of age. Additionally, we identified three hotspots of houses harboring infected children that overlapped with hotspots of triatomine abundance at baseline and persistent house infestation [22, 37].

The global human seroprevalence of T. cruzi in Areas II and IV was lower than in Areas I (39.8%) and similar to that registered in Area III (29.0%) [41]. These studies were conducted 3 and 4–6 years after the onset of the interventions, respectively [41], whereas the current study was conducted 6–9 years after sustained vector control actions. Operational areas also differed in the observed baseline household T. infestans-infestation prevalence and abundance [22]. Other locations of the Gran Chaco with active vector-borne transmission and lack of sustained vector control showed higher prevalence of infection [24, 60,61,62,63,64,65,66,67,68]. Similarly, the pre-intervention force of infection (FOI = 1.23) was lower than in other locations in the Gran Chaco without vector control [31, 41, 69, 70]. Such force of infection implies that every year nearly 34 new human infections would have occurred in the study area if pre-intervention transmission conditions had remained unchanged.

Of the 122 children born after program onset and tested for antibodies to T. cruzi, the only seropositive child likely contracted the infection from her T. cruzi-seropositive mother who reported no history of etiological treatment. Similarly, a district-wide serosurvey across Pampa del Indio municipality in 2016 revealed no T. cruzi-seropositive child born after the onset of the intervention program except one attributed to transplacental transmission, and no incident cases were detected among residents who had been seronegative at baseline [43]. These serosurvey outcomes provide further support for the interruption of vector-borne transmission to humans across rural areas of Pampa del Indio [43].

The generalized linear mixed model identified several risk factors associated with children or all-ages T. cruzi infection. Duration of exposure to T. infestans was positively and strongly associated with T. cruzi infection, as reported by several studies [24, 41, 62]. The number of infected co-inhabitants had high RI in Model 1 and was also positively associated with human infection. This variable implies vector-borne transmission risk in two senses: it indicates the number of local parasite sources for triatomines and pinpoints the domicile as a high-risk place for vector-borne transmission in the past. Nevertheless, the number of infected co-inhabitants may not reflect the current risk status of a household since domestic infestation with T. infestans had been virtually suppressed after community-wide insecticide spraying. Residential stability, measured as duration of residence in a given house, is especially important for a population with frequent mobility as in this study. In the past, we detected that moving the household location was a shared feature among some Qom families (formerly semi-nomadic), and we called them “movers” [44]. Movers represented a fraction of the population exhibiting the highest vulnerability, which had a higher risk of residing in an infested house. They accounted for 14% of the population in Area III in a 3-year period. To address the occurrence of “movers” (i.e. detect transient changes in residence location) an ad hoc study design was conducted in Area III, which was not implemented in Area II and IV. Therefore, we could not classify the study households as “movers” or “non-movers.” We addressed this limitation by including residence time in the current house and the place of previous residence in univariate analyses. The relationship between domiciliary infestation and infection of children was modified by household mobility in Area III [44]; in non-infested houses, the infection risk in children was higher among movers than among households with stable residency, while the reverse was observed in infested houses [44]. In our dataset, people with previous residence in urban Pampa del Indio exhibited a lower risk of infection than people with stable residence in the study area. In contrast, people coming from other rural areas of Pampa del Indio municipality exhibited a higher risk of infection. People with previous residence outside Pampa del Indio showed a similar prevalence to stable residents, and no significant difference in the observed risk of infection was detected (Additional Information Table S4). These findings reveal the importance of incorporating variables that consider the history of exposure rather than a snapshot of present conditions, particularly in intervened scenarios like in this study in risk factor analysis.

The strong positive association between entomological indices and human T. cruzi infection has been verified across epidemiological settings [29, 30, 35, 41, 29,70,30]. In this study, the sole entomological variable that considered the individual’s lifetime exposure was self-reported inhabiting an infested house. This variable was critical in the case of residential infestations elsewhere [72,73,74], had the best performance among entomological predictors in the univariate analysis and was therefore included in the risk indices despite having a moderate RI in the GLMM. However, in the current study, the occurrence of triatomines by timed manual searches during baseline and surveillance period was not statistically associated with human infection, in contrast to previous reports [71, 75, 76]. As mentioned above, this observation is likely related to the fact that infestation data were available for the previous 6–9 years, whereas human infection probably occurred before that period, and that during the surveillance phase infestation mainly occurred in peridomestic sites associated with chickens [22]. In addition, the high mobility rate inside the municipality probably also affected the association between entomological indices and human infection as observed in Area III [44]. Immigration of T. cruzi-infected people could add to this discrepancy. However, only 9.2% of the tested individuals reportedly lived outside the municipality, including presumably non-infested urban areas, suggesting this potential source of T. cruzi-infected people was negligible in this population.

Indigenous people usually experienced higher risks of infection for diverse NTDs [77]. Several indigenous groups exhibited a higher prevalence of T. cruzi infection than creoles inhabiting the same area [40, 61, 65, 78, 79]. In this study, we observed strong heterogeneities between and within ethnic groups. Ethnic background was correlated with some socioeconomic variables such as the goat equivalent index, educational climate, time of residence, inhabiting a house with suitable walls for triatomines and critical overcrowding. Qom households exhibited worse socioeconomic conditions than creoles. However, creoles from Area IV exhibited the highest seroprevalence and force of infection found in this study, whereas creoles from Area II displayed the lowest levels and Qom people from either area had intermediate ones. This may be the first report of creoles showing higher prevalence of T. cruzi infection than an indigenous group residing in the same rural area. Creoles from Area IV inhabited remote rural sections with difficult accessibility (TCZ, CQ and EP) and had the lowest goat-equivalent indices of all creole households across the municipality [37, 42, 45]. Nevertheless, on average across the four areas, creole households exhibited higher goat-equivalent indices than Qom ones. These results point to a complex set of risk determinants, as in Area III [44], which escapes from the oversimplifying notion of equaling household ethnicity to infection risk as supported by the low RIs retrieved for the simple effects of ethnicity, area and their interaction when adjusted for other variables.

Regarding non-vectorial transmission routes, maternal serostatus displayed high RI for T. cruzi infection among children born before program onset. Maternal serostatus informs about the possibility of transplacental transmission and the number of infected household residents. According to a recent report on the situation of Chagas disease in the Americas [80], transplacental transmission has gained relative relevance in the context of more widespread vector control, in accordance with our findings among children born after program onset. Our preliminary estimate of the probability of vertical transmission (2%) is similar to the ones recorded in other regions [81].]. Oral transmission appears to be negligible in our study area and elsewhere in Argentina, where no oral outbreak of human T. cruzi infection has been reported so far [8].

The developed risk indices exhibited promising results identifying 93.9–98.1% of households with T. cruzi-seropositive children based on three questions easy to deliver in the field. They also have some limitations; one is the lack of validation with an external data set. Therefore, the reported sensitivity and specificity must be considered cautiously. The inclusion of “self-reported inhabiting an infested house” increased the frequency of false-positive cases. The low specificity implies more extended field operations. As the number of children inhabiting infested houses is expected to drop with control intervention progress, the incorrectly identified households will also decline. The presence of cracked walls (suitable walls for triatomines), easy to canvass during house-to-house inspections for triatomines has been routinely registered by vector control programs at least in Argentina. As a proxy of triatomine infestation risk [42], the inclusion of suitable walls for triatomines in other endemic regions may need to be tailored to the local triatomine species. Another limitation of the second index is its dependence on serological data. However, this constraint may be negligible in areas where serological data are accessible; for example, they could be obtained from the mandatory Chagas disease serodiagnosis of pregnant women in Argentina and in other countries with similar legislation [82].

Different approaches have been proposed for identifying infected children in endemic areas. Researchers were able to identify individuals with increased risk of infection by using indices based on the number of infected co-inhabitants or entomological data [33, 34]. Recent studies aimed to identify households with infected children through extensive data analysis though this approach may have little applicability in routine operations [28]. Here, our aim was to keep the indices simple to use, with the selected questions easy to answer by residents. For the index development, we chose the stratification at the household rather than person level since it is simpler during fieldwork to search a house than an individual. Household clustering of human infection further supports this approach [35].

We identified three local hotspots of infection of children close to hotspots of triatomine abundance [22, 36, 37]. One was associated with houses not inspected at baseline and with the occurrence of pyrethroid-resistant T. infestans [22, 37]. The observed spatial overlap between hotspots supports that children’s infections were mainly vector-borne and occurred before the onset of control efforts. A study conducted by Levy et al. [35] suggested the combined use of spatial analysis and risk indices to identify households harboring infected children in a low-prevalence urban setting, thus saving labor-related costs at the expense of other costs related to data analysis. Further studies are needed to validate these approaches.

The data available for our analyses has some limitations. We were unable to include some variables of interest, such as baseline house infestation and triatomine infection status [30, 35, 41, 70, 75], because of the systematic sampling of houses at baseline in Area II [37]. Self-reported lifetime residence in an infested house [71] is expected to suffer from recall bias. Maternal serostatus was another important variable with missing data for older people, as was the frequency of infected adult people; hence, it could not be included in the regression analysis of the complete dataset.

Our serosurvey achieved a fairly high diagnostic coverage (~ 50%) of the study population, especially because only 13% of the individuals reported had a previous serodiagnosis, showing the feasibility of achieving high coverage in rural endemic zones [17]. There is a chance for non-participation bias. This might happen if the non-participating fraction of people does not represent a random sample of the population or if the reason for non-participating was related to a previous known serostatus for T. cruzi. This is unlikely since < 14% of the participants reportedly had a previous serodiagnosis. An additional source of bias could happen if a whole age class was excluded from the analysis. We addressed this potential issue by conducting house-to-house visits to reach all rural inhabitants. As the coverage of children under 18 years of age was 55.4%, we estimated that 513 children may have remained without access to serodiagnosis, and 71 (95% CI = 57–88) likely were seropositive for T. cruzi. One of the strengths of this study was the consistent use of serological tests and vector detection methods across areas and periods for almost 10 years, which allowed direct comparisons across time and space [17, 38, 39, 43, 83].

The proposed indices may be suitable for other endemic situations in which the main triatomine vector is domiciliated and the local inhabitants report exposure to triatomines (i.e. they need to recognize the vector species). Given that the indices rely on self-reported exposure to triatomines, migration background is not expected to alter index validity. Likewise, these indices are independent of vector control history. However, in the scenario of massive exposure to triatomines, the discriminatory power of the indices will drop abruptly since almost all households will be marked as risky. Although the serodiagnosis of T. cruzi infection at the age of 6 years is guaranteed by a National Law, it has not been implemented across all rural endemic areas. Therefore, developed indices may assist public health officials working with limited resources to prioritize households with higher odds of harboring infected children.

The seroprevalence of children born after the onset of interventions and house infestation data for Areas II and IV of Pampa del Indio support the interruption of vector-borne transmission of T. cruzi to humans 10 years after program onset [22, 36, 43]. This status offers room to address new challenges such as detection of transplacental transmission and ensuring a wider access to serodiagnosis and treatment of rural residents, especially of seropositive women of childbearing age. The risk indices here presented may help to scale up such interventions in a cost-effective fashion.

Conclusions

Nearly a quarter of the study population was seropositive for T. cruzi infection 6–9 years after the onset of sustained vector control interventions in a highly endemic municipality of the Argentine Chaco. We revealed strong heterogeneities in the infection risk between and within ethnic groups inhabiting two neighboring rural areas. Creole households from Area IV exhibited a higher risk of infection than Qom households of either area. Two simple risk indices identified nearly all the households harboring T. cruzi-seropositive children at the expense of low specificity. These indices may be easily implemented and scaled up across endemic areas to improve the access of rural residents to diagnosis and treatment of T. cruzi infection in pursuit of Sustainable Development Goals and disease prevention. Vector control actions should be sustained over time to prevent renewed vector-borne transmission.