Chagas disease has historically occurred in poor rural settings of Latin America [1], [2], [3]. In the rural areas of the Department of Arequipa in southern Peru, reports of Chagas disease and its vectors date back to the early 20^th century [4], [5], [6], where the disease has persisted in an endemic state [7]. However, recent case reports [8] and epidemiologic studies [9], [10] have documented emerging vectorial transmission of Trypanosoma cruzi, the etiologic agent of Chagas disease, in communities of the urban capital of Arequipa. As T. cruzi and Triatoma infestans, the sole insect vector in this setting, spread through the city of Arequipa (pop. 864,250) [11], [12], several hundred thousand people are at risk of infection.

Most individuals have mild or no symptoms during the acute phase of Chagas disease, and pass into the chronic phase without having the infection detected. In the chronic phase, most infected individuals show no signs or symptoms and are considered to have the indeterminate form of the disease. An estimated 20–30% of infected individuals will later develop the cardiac or digestive forms of chronic Chagas disease. Advanced cardiac or digestive disease cannot be reversed and can be fatal [13], [14], [15]. However, for patients with the indeterminate form or early cardiomyopathy, antitrypanosomal treatment is reported to decrease the probability of disease progression [15].

It is important to detect individuals with chronic T. cruzi infection because they need clinical attention, may be candidates for treatment, and pose risk for further transmission – either vectorial, congenitally or through blood donation [16], [17], [18]. Diagnostics for Chagas disease in southern Peru are expensive relative to local income and the limited government budget for vector-borne disease control [19]. Alarmed by the urban encroachment of T. infestans, the Arequipa regional Ministry of Health (MOH) recommends universal testing of all residents of communities with T. cruzi-infected T. infestans. However, such wide-scale testing is too costly for practical implementation. Population screening is further challenged by the very low sensitivity of field-applicable rapid tests in Arequipa, and screening must employ high sensitivity conventional tests such as enzyme-linked immunosorbent assays [20]. Given the large population of Arequipa city and limited health resources, targeted interventions are the only viable option to screen for chronic T. cruzi infection in this population.

Although mass vector-control campaigns have had success in Latin America [21], [22], numerous authors call for improved screening strategies and expansion of treatment for persons with Chagas disease [19], [23], [24], stressing the importance of cost-efficiency, sustainability, and integration of sectors [22], [25]. Several cost-effectiveness studies of Chagas disease interventions have focused on devising optimal insecticide application and blood donor testing schemes [25], [26], [27], [28], but few have examined strategies for human serologic testing in endemic or epidemic areas. A study by Mott et al. [29] indicated the potential for targeted screening around detected positive children under 5 years of age in a rural area of Brazil. Gurtler et al. [19], [30] highlight the potential efficiency of linking serologic screening to vector control campaigns. It remains unclear how to implement cost-effective, targeted screening programs across a large and diverse urban environment with dynamic vector infestation.

In one periurban community of Arequipa that was heavily infested with T. infestans, 5.3% of children had T. cruzi infection [31]. Age-prevalence curves of infection [10] and the spatial distribution of cases suggested that the disease was in an epidemic phase in this community [31]. A retrospective analysis of household vector data carefully collected during an insecticide-application campaign prior to serologic testing, indicated that a two-step targeted screening intervention based on household entomologic risk factors and ring testing around identified cases would have captured 83% of infected children, while minimizing the testing of negative children [31]. In the present operational research study, we test 4 targeted screening strategies based on similar household entomologic data, and one adaptive spatial strategy, in 18 periurban communities of Arequipa, Peru. We compare the performance and cost of each, and evaluate the patterns of infection they reveal. Our objective was to test the operational feasibility of routinely collected data from vector control campaigns for human Chagas disease detection on a larger urban scale.

At the time of insecticide application in our 18 study communities, 1980 of 7739 (25.6%, 95% CI: 24.6–26.6) sprayed households were found to be infested with triatomine vectors, with technicians capturing between 1–301 T. infestans per household. The 90^th percentile cut-off between the high-density and low-density vector houses was determined separately for each community and ranged from 11–85 T. infestans captured per household. Out of the 1980 households in which vectors were detected, eighty-one (4.1%) had vectors carrying T. cruzi. In total, 923 people from 249 households participated in the serological testing across the 18 communities in 3 districts. There were no statistically significant differences in demographics by district (data not shown). Participants had a mean age of 34 years (range: 1–94), and 78.9% were adults of age 18 or over. Of the total participants, 21 (2.28%) tested positive according to ELISA, and were considered T. cruzi infected for this analysis. Nineteen of these also had positive results by IFA, while 2 did not. All 21 infected individuals were over the age of 18.

Table 1 displays the results across strategies. Neither demographic characteristics nor infection prevalence differed significantly among participants recruited through the four alternative strategies. Although not statistically significant, the highest prevalence was observed among those screened in houses with no infestation detected (3.07%). However, we encountered high refusal rates during participant recruitment from the houses with no infestation detected. Of 97 households approached, only 59 (61%) participated. By contrast, 98% of households with T. cruzi-infected vectors detected participated in the study. In addition, the within household participation rate was lower for households in which no infestation was detected versus households in the infected vector group, 55% versus 70% participation, respectively. These data suggest a self-selection bias in participation among people living in houses with no infestation detected. Potential reasons for this are explored in the discussion.

The 923 participants had lived in a mean of 2.5 locations (range:1–15). Mean residence in a periurban location was 26.97 years (CI 25.98, 27.97), while mean residence in an urban location was 3.61 years (CI 3.11, 4.12), and 3.39 years (CI 2.91, 3.88) in a rural location. The T. cruzi infected human cases we detected had lived longer, on average, in rural places than non-cases (mean 7.20 years vs. 3.31, respectively; p = 0.0184).

The results of univariate analysis on T. cruzi infection status are shown in Table 2. The probability of infection with T. cruzi increased by 2% per year of age (p = 0.02), by 2% per year lived in a periurban location (p = 0.185), by 4% per year lived in a rural location (p = 0.04), and by 4% per year lived in a place with triatomine bugs (p = 0.008). Multivariate models with combinations of these variables did not fit the data better than the univariate models (data not shown). Given inherent dependence of all four of the above variables on calendar time, it is likely that all are describing a similar experience of risk.

Table 1 also displays the results of the secondary case detection by 15 meter radii. The adaptive ring sampling of houses within 15 meters of the houses of the 17 index cases yielded 4 additional cases out of 158 individuals tested. Although employed in seven different communities, all four of the secondary detected cases lived in the same community, Simón Bolívar. Of the 10 total cases found in Simón Bolívar, 7 had lived only in urban districts of Arequipa, and 5 only in Simón Bolívar itself, suggesting a local, urban site of transmission.

The estimated costs for each study activity are shown in Table 3. The fixed overhead cost per participant was $7.14, including field materials, telecommunications, transportation, data management, and diagnostic testing, and this was standard across all strategies. However, the recruitment cost per participant varied greatly across strategies, being a function of both the sampling framework and differential participation rates for each. For the households with infected-vectors detected, our field team made 1.32 household visits per participant recruited, versus 2.63 visits to houses in which no infestation was detected, reflecting the higher refusal rate of the latter. Each household visit required an average of 1.54 person-hours, at which rate, the testing in houses where no infestation was detected was the most costly strategy per participant ($14.62 USD), and the infected-vector houses strategy the least costly per participant ($10.91 USD).

Chagas disease is a growing problem in Arequipa, Peru. A previous study in the city uncovered micro-epidemics of transmission associated with high density of vectors, suggesting that targeting screening based on entomologic information could be an efficient means of detecting T. cruzi infected individuals [31]. In this prospective field trial, we uncovered important differences in the association between vector-parasite distribution and human Chagas disease infection, which merit consideration in the design of screening programs for Chagas in Arequipa and other urban settings.

There are numerous reasons why the vector-based targeted screening strategy designed for a previously studied community (Guadalupe), yielded fewer cases when applied elsewhere in the city. In Guadalupe, parasite infections in both humans and vectors were clustered, and the age-prevalence curves suggested established epidemics of T. cruzi transmission [10], [31]. The present study sites, much closer to the city center, also contained clusters of parasite-infected vectors, but these consisted of fewer households. These smaller clusters are likely indicative of a relatively short history of vectorial transmission, leading to few locally acquired cases.

In addition to epidemiology and ecology, social and demographic phenomena may also affect patterns of human Chagas disease [35], [36]. Frequent migratory movement between rural and metropolitan Arequipa may bring parasite into the city [36], [37]; the cases of infection we detected were most associated with a history of triatomine exposure and residence in rural areas, where Chagas disease is historically endemic [4], [5], [6]. It is possible that the few clusters of parasite are due to introductions from outside that did not manage to spread beyond a handful of households in the new urban environment. That periurban time of residence did not significantly contribute to overall risk of infection in this study is another indication of minimal vectorial disease transmission in the study communities.

Interestingly, one of the 18 communities studied, Simón Bolívar, did display patterns of infection reminiscent of the micro-epidemic hotspots observed in Guadalupe. In Simón Bolivar, the 4 secondary cases detected within 15 meters of index cases may suggest significant rates of local vector-borne transmission at the time of insecticide application. The dissimilar migration history of the cases found in Simón Bolívar allow us to rule out that this clustering may be due to a group of infected migrants from the same sending community settling on the same city block, a phenomenon that has occurred in periurban settlements of Arequipa [37], [38]. Importantly, there was no obvious a priori evidence in Simón Bolívar to expect transmission to differ from the other 17 communities considered. This study adds to growing evidence of an uneven distribution of T. cruzi infection in the city of Arequipa [9], [10], not uncommon to this parasitic disease [39]. Other similar pockets of vector-borne transmission may exist in the city; the implementation of small pilot studies in infested areas followed by spatial adaptive sampling around human cases can help uncover them and determine the appropriate screening strategy for each setting. Where urban vectorial transmission is present, this adaptive strategy identifies secondary cases efficiently. When employed in an area without vectorial transmission, adaptive sampling should return few or no secondary cases, such that additional screening is curtailed and expenditures capped. Although harder to obtain, longitudinal entomological data, as opposed to the cross-sectional data utilized here, may be informative in locating these mini-epidemics. We are currently exploring community-based recognition and alert systems as a promising mechanism for obtaining this longitudinal vector data.

In addition, our disease screening activities took place between 5 months and 4 years after household insecticide application and collection of the entomologic data. It is not clear what the effect of this time lapse may be on the association between vector data and human infections detected. The strategies may work much better if these delays were eliminated. However, considering that exposure risk is greatly reduced by the elimination of vectors, we can reasonably expect that there were little new infections prior to our testing.

Another possible cause of error is the inability of migration histories to capture participants' short visits to endemic areas. While a potentially important source of exposure, there are methodological and recall challenges to documenting travel history at such a fine level. We found that screening based on personal risk assessment and residence-exposure history (ie, time lived in rural and/or infested areas) is advantageous for capturing high-risk individuals. Studies of blood donors in Canada and the US have found promising usefulness and operational feasibility of residence history questionnaires [26], [27], [40].

Our study consisted of operational research in a large heterogeneous city. In the wake of vector control campaigns, the target population was well-aware of their risk of Chagas disease. Residents of houses where no infestation was detected could have been at risk of vectorial transmission due to proximity to infested houses. However, during our fieldwork, it became clear that those who participated in this optional Chagas disease screening did so because they believed themselves to be at risk due to prior exposure. Participants from houses with no infestation detected often expressed concern about a previous infestation in their current or prior homes. In contrast, those who refused often reported being uninterested in testing because they did not consider themselves at risk, or because in their memory they had no contact with a vector. As a result, this study suffered from a self-selection bias according to perceived risk among participants living in vector-free houses at the time of the spray campaign. This bias likely caused an overestimate of the Chagas disease prevalence in houses with no infestation detected. Although not directly comparable, our observed overall prevalence of 2.28% is much higher than the 0.73% reported in a 2004–05 cross-sectional screening of pregnant women in Arequipa, which included populations from communities also studied here [41]. In contrast, we believe the prevalence estimate of 2.60% among the infected vector group to be quite accurate for persons living in this risk category due to the 98% participation rate.

Factors affecting participation need to be taken into consideration to design an economically optimal algorithm for targeted screening of Chagas disease. The high refusal rate among houses with no infestation detected required double the number of household visits per participant recruited than for the infected-vector group, making it the most expensive strategy in our cost calculation. If similar entomologic data from vector control campaigns are used to guide a human Chagas disease case-finding strategy, minimizing costs for fieldwork while still detecting cases would be optimal. Targeting screening according to the presence of T. cruzi-infected T. infestans was less expensive and similarly effective compared to the other strategies. Further, coupling screening to ongoing vector control campaigns can improve participation [19] with the added advantage of eliminating the time lapse between entomologic data and human testing that we experienced in some communities of this study.

In large cities like Arequipa, where T. cruzi exposure is highly heterogeneous, a targeted screening program is necessary for the prompt diagnosis of indeterminate Chagas disease, the prevention of future transmission and the maximization of limited resources [19]. A greater knowledge of the patterns of infection can be obtained by pilot studies, and would improve the design of screening strategies. Rural, urban and periurban places have different ecologies, and it is reasonable to expect different epidemiologies of Chagas disease in these settings [42]. The flexibility to adapt to the epidemiology that emerges during pilot screenings is key to an efficient case detection intervention. Finally, this data can help develop a brief residence history questionnaire for referring to diagnostic testing those at greatest risk of Chagas disease. Self-assessment of risk and triatomine exposure time is a potentially useful tool for screening; volunteer screening programs at local health posts or fairs may be very effective ways to capture infections in heterogeneous periurban communities subsequent to informative vector control campaigns.