^{1}

^{2}

^{3}

^{*}

^{3}

^{3}

^{2}

^{3}

^{3}

^{1}

^{1}

^{4}

^{3}

^{5}

^{3}

^{6}

^{1}

Conceived and designed the experiments: RG VK LW ML NB EC CB JC. Performed the experiments: VK ML NB LC VP EC AS FS. Analyzed the data: LW ML. Contributed reagents/materials/analysis tools: RG EC CB JC. Wrote the paper: JM LW ML NB CB.

Millions of people are infected with

We performed a serological survey in children 2–18 years old living in a peri-urban community of Arequipa, Peru, and linked the results to entomologic, spatial and census data gathered during a vector control campaign. 23 of 433 (5.3% [95% CI 3.4–7.9]) children were confirmed seropositive for

We found evidence of spatially-focal vector-borne

Chagas disease kills more people than any other parasitic disease in the Americas. The disease is caused by

An estimated 11 million people are currently infected with the causative agent of Chagas disease,

Anti-trypanosomal drug therapy can cure 60% or more of infected children aged < 13 years

Vector-borne transmission of

The aim of the study was to develop targeted screening strategies to detect

Arequipa is located at an elevation of 2300 meters in southern Peru. Arequipa's climate is arid most of the year, though there is rainfall between the months of January and March. Santa Maria de Guadalupe and Alto Guadalupe (hereafter referred to together as Guadalupe) are two of hundreds of communities located on hillsides on the outskirts of Arequipa (16.44°S, 71.59°W) and have been described previously ^{2}). Three hundred and seventy-four of the 397 households were sprayed with deltamethrin powder (Bayer K-othrina, Lima, Peru) suspended in water at an intended rate of 25 mg/m^{2} by the Arequipa Regional Office of the Ministry of Health in November and December of 2004. Twenty-three households either were closed or refused insecticide treatment. At the time of insecticide application, 194 (52.0%) households were found to be infested with

During the course of the first insecticide application to households in Guadalupe, 2 trained triatomine collectors systematically searched each room of the human dwelling, each animal enclosure, and the remaining peridomestic area for a total of 1 person-hour. Triatomines captured from each site of collection were stored separately on ice packs and taken to the University of San Agustin where they were counted by site, stage, and sex (for adults). A sample of 10 live and moribund adult and 5^{th} instar triatomines from each site of collection were examined microscopically for

Serologic testing was carried out between August and October of 2005. All children < = 18 years old were invited to participate in the study. Trained study staff explained the study to children and their parents or guardians in schools and over the course of several meetings in the community of Guadalupe. Participants 18 years of age provided informed consent. The parents or legal guardian of all participants under 18 years of age provided informed consent and each participant 7 years or older provided informed assent. The consent form was read aloud to all illiterate parents and participants, and in these cases consent or assent was indicated with the person's fingerprint rather than a signature.

After informed consent, 5 ml of venous blood was drawn by a trained research nurse from children over 5 years old; 3 ml was drawn from younger children. Blood was kept on ice and separated on the day of collection by centrifugation. Aliquots of sera were stored at −20°C until testing. Sera were tested by commercial ELISA (Chagatek, Biomerieux, formerly produced by Organon Teknika). All positive sera, and 10% of negative sera, were tested by immunofluorescence assay (IFA) at the Centers for Disease Control and Prevention. A specimen was considered positive by ELISA if absorbance was at least 0.100 greater than the average absorbance of three negative controls, following the manufacturer's indications. IFA titres of 1∶32 or higher were considered positive (F. Steurer, Division of Parasitic Diseases, CDC, Atlanta, GA). Children whose blood was positive by both ELISA and IFA were considered seropositive

To describe the spatial distribution of vectors, parasite-infected vectors, and confirmed seropositive children we calculated the spatially smoothed density of households containing each across the study area. We used a kernel smoothing function with a Gaussian kernel of a bandwidth of 25 meters

After examining the spatial patterns of vector infestation, parasite-infected vectors and seropositive children, we evaluated the association between the presence of a seropositive child and local covariates measured in each household. All data on local covariates were collected during the insecticide spray campaign. We divided these covariates into groups based on the amount of effort needed to collect each type of data. Census data, such as age and the presence of domestic animals, required the administration of a questionnaire. Routine spray data were those data collected by the Ministry of Health during spray campaigns, such as the presence or absence of vectors in the domestic and in the peridomestic area. Timed vector search data required a systematic timed entomologic search, and consisted of estimates of vector densities in the domestic and peridomestic areas. Microscopic examination data comprised the presence of

In univariate analyses, associations between confirmed seropositivity and binary covariates were evaluated with the χ^{2} test. For continuous covariates a non-parameteric receiver operating characteristic (ROC) curve was plotted and the area under the curve (AUC) estimated by the trapezoid rule of integration. Larger areas (greater than 50%) indicate positive association between the covariate and confirmed diagnosis; smaller areas (less than 50%) indicate negative association. The area under the non-parametric ROC curve is equivalent to the Wilcoxon rank sum statistic (also known as the Mann-Whitney U statistic), and p-values for associations between continuous covariates and confirmed seropositivity were estimated by the Wilcoxon rank sum test

We used Bayesian hierarchical mixed modeling techniques to estimate the effects of multiple covariates on the probability of confirmed seropositivity controlling for spatial autocorrelation of the outcome variable. The mixed model conditions inferences and predictions on an unknown underlying risk of each child (a random effect)

Bayesian hierarchical models have been used for many years in epidemiology, and especially spatial epidemiology (reviewed in Boyd 2005 ^{6}. We specified a Gaussian distribution for the random effects, and set a uniform prior distribution for the inverse of the standard deviation of the Gaussian distribution. Overly wide priors for the inverse standard deviation of the random effects led to numerical overflow errors; a uniform prior with a range from 0 to 15 was broad enough to avoid truncating the posterior estimates and narrow enough to avoid overflow errors. Models were fit in WinBugs 14.1; code is available upon request.

We used the fit Bayesian models to rank the children in the community based on their age and local covariates, but without including any information about the spatial component of risk, which is unknown prior to testing. Using the ranking from the fit models we plotted non-parametric ROC curves and calculated the area under these curves. We then considered two-step case detection strategies that reincorporated spatial information. In two-step strategies the fit multivariate models are used to rank the risk of seropositivity in children based solely on their age and local covariates collected from their households during the spray campaign. In the first step of screening a proportion of the highest-risk children are tested. The results of the preliminary screening are used to identify children living within a given distance of seropositive children, and in the second step of screening these children are tested (ring screening). We considered ring screening radii of 10, 20, 30, 40, 50, 60 and 70 meters. For each radius we plotted an ROC curve for the two-step screening strategy and calculated the area under the curve. We calculated the sensitivity (number of cases detected/total number of cases) and specificity (number of non-cases detected/total number of non-cases) for all potential cutoff points of the ROC curves. We report the percent of non-cases (1-specificity) that must be tested to identify >80% of cases for each fit model. ROC analyses were programmed in the R statistical environment (

Specimens were tested for a total of 433 children. Of these, 26 (6.0% [95% CI 3.8–8.4]) were positive for antibodies to

Of census data variables, age and the presence of animals, almost exclusively dogs and cats, sleeping in the domestic area of the household were significantly associated with confirmed

Data Source and Covariates | Mean (range) or number of children (%) (N = 398) | Odds ratio or Area under ROC curve | p-value |

Age | 10.9 (1–18) | AUC = 64.4% [54.3–74.5] | 0.020 |

Presence of animals sleeping inside house | 80 (20.1%) | OR = 2.75 [1.01–7.12] | 0.019 |

Domestic infestation | 216 (54.3%) | OR = 2.51 [0.02–7.92] | 0.051 |

Peri-domestic infestation | 125 (31.4%) | OR = 2.10 [0.81–5.36] | 0.081 |

Estimated domestic vector density | 7.1 (0–200) | AUC = 63.2% [51.4–75.0] | 0.025 |

Estimated peri-domestic vector density | 9.7 (0–616) | AUC = 58.6% [46.90–70.3] | 0.092 |

Domestic | 61 (15.3%) | OR = 2.60 [0.86–7.05] | 0.038 |

Peri-domestic | 42 (10.6%) | OR = 4.25 [1.37–11.79] | 0.001 |

The difference in Ripley's K functions for households with and without triatomines never exceeded the 99% tolerance limits, suggesting no significant spatial clustering of vector infestation (

After controlling for spatial effects, the risk of

Data Source | Model | Formula | Estimated Odds Ratio [2.5%, 97.5% quantiles] |

Census alone | A | α+β_{age}*X_{a}+φ_{j} | e^{βage} = 1.20 [1.04–1.43] |

α+β_{age}*X_{a}+β_{an}*X_{an}+φ_{j} | e^{βage} = 1.21 [1.04–1.45] | ||

e^{βan} = 2.41 [0.25–17.92] | |||

Census & routine spray | B | α+β_{age}*X_{a}+β_{vp}*X_{vp}+φ_{i} | e^{βage} = 1.21 [1.03–1.45] |

e^{βvp} = 8.40 [0.93–185.68] | |||

Census & timed vector search | C | α+β_{age}*X_{a}+β_{vd}*X_{vd}+φ_{j} | e^{βage} = 1.22 [1.04–1.47] |

e^{βvd} = 1.04 [1.01–1.09] | |||

Census, timed vector search & microscopic observation | D | α+β_{1}*X_{a}+β_{tc*} X_{tc}+φ_{i} | e^{βage} = 1.21 [1.04–1.44] |

e^{βtc} = 2.91 [0.32–35.84] | |||

α+β_{age}*X_{a}+β_{2}*X_{vd}+β_{tc*} X_{tc}+φ_{j} | e^{βage} = 1.22 [1.04–1.46] | ||

e^{βvd} = 1.04 [1.01–1.09] | |||

e^{βtc} = 2.01 [0.17–24.75] |

β's are coefficients of the fit models, X's are observations for each child. Subscripts describe the variables age = age of child, an = animal sleeping inside the domestic area, vp = Domestic vectors present, vd = Domestic vector density, tc = Domestic _{j }is the spatial random effect assigned to each household (denoted by the subscript j).

The area under the curve for the predictive model based on age alone was 0.64 (

Model A includes information on age only, Model B includes data on age and vector presence, Model C includes age and domestic vector density, and Model D includes information on age, domestic vector density, and the presence of domestic

Covariates | Model | Area Under the ROC Curve | ||||||||

No Ring Case Detection | Radius of Ring Case Detection (Meters) | |||||||||

0 | 10 | 20 | 30 | 40 | 50 | 60 | 70 | |||

Age | A | 0.64 | 0.75 | 0.78 | 0.80 | 0.80 | 0.81 | 0.78 | 0.81 | 0.79 |

Age & domestic vector presence | B | 0.68 | 0.75 | 0.85 | 0.85 | 0.84 | 0.84 | 0.81 | 0.81 | 0.80 |

Age & domestic vector density | C | 0.71 | 0.80 | 0.84 | 0.85 | 0.84 | 0.83 | 0.82 | 0.81 | 0.79 |

Age, domestic vector density & domestic | D | 0.72 | 0.80 | 0.85 | 0.85 | 0.83 | 0.83 | 0.82 | 0.80 | 0.78 |

One of the two-step screening strategies with an area under the ROC curve of 85% would begin by ranking children based on their age and the number of vectors captured within their houses. In the first round of screening 15% of the highest-ranked children would be tested, and the households of all seropositive children identified. In the second round of screening all children living within 20 meters of households with seropositive children would be tested. In total 23% of the population would be screened and 19/23 (83%) of seropositive children diagnosed.

Chagas disease transmission cycles have become established in communities on the outskirts of the city of Arequipa, Peru. A vector control campaign is currently disrupting transmission of

We observed spatial evidence that transmission of

Traditional epidemiologic methods are not well suited for epidemics when infection is clustered in space

Although no published study has described risk factors for

We were not able to evaluate the sero-status of participant's mothers to consider the potential that children were infected congenitally

Knowledge of local risk factors

Economic analysis is needed to optimize targeted screening strategies for Arequipa given the real costs of gathering data and serologic testing. Measuring vector density by the person-hour method requires that an additional person accompany each spray worker at the time of application of insecticide. Alternative methods for estimating vector density, such as leaving collection bags with homeowners following spraying

We are limited in terms of our ability to extrapolate the findings of our study to other areas. If, as we suggest, transmission is epidemic in peri-urban Arequipa, the results of our predictive models might be sensitive to the precise timing of insecticide application. Had Guadalupe been sprayed one year later not only might the prevalence of infection have been higher, but the associations between covariates and infection might have been quantitatively different. In an analogous analysis, Struchiner et al. demonstrate how estimates of the effect of a vaccine against malaria would change over the course of a malaria epidemic

In conclusion, our results suggest that peri-urban communities in Arequipa may be in the midst of an epidemic of vector-borne

Translation of abstract into Spanish

(0.02 MB DOC)

Click here for additional data file.

We would like to thank the community of Guadalupe for their participation and hospitality during this study. We especially thank Victor Quispe, Jose Ylla, the spray brigade, nurses and bug collectors for all their help. We also thank Robert Wirtz, Gregory Martin, Jeffrey Stancil, Ellen Dotson, Gena Lawrence, Cesar Bocangel, Fredy Delgado, Jenica Pastor, and Fernando Malaga for their assistance and support.

The authors have declared that no competing interests exist.

MZL is supported by a Howard Hughes Pre-doctoral fellowship. NMB was supported by a Fogarty/Ellison fellowship during this study and National Institutes of Health training grant 5T35AI007646-03. VK was supported by a grant from the International Society for Infectious Diseases. Additional support came from National Institutes of Health grants U19-AI-33061 and RO1-AI047498.The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.