Integrating evidence, models and maps to enhance Chagas disease vector surveillance

Abstract

Background: Until recently, the Chagas disease vector, Triatoma infestans, was widespread in Arequipa, Perú, but as a result of a decades-long campaign in which over 70,000 houses were treated with insecticides, infestation prevalence is now greatly reduced. To monitor for T. infestans resurgence, the city is currently in a surveillance phase in which a sample of houses is selected for inspection each year. Despite extensive data from the control campaign that could be used to inform surveillance, the selection of houses to inspect is often carried out haphazardly or by convenience. Therefore, we asked, how can we enhance efforts toward preventing T. infestans resurgence by creating the opportunity for vector surveillance to be informed by data?

Methodology/principal findings: To this end, we developed a mobile app that provides vector infestation risk maps generated with data from the control campaign run in a predictive model. The app is intended to enhance vector surveillance activities by giving inspectors the opportunity to incorporate the infestation risk information into their surveillance activities, but it does not dictate which houses to surveil. Therefore, a critical question becomes, will inspectors use the risk information? To answer this question, we ran a pilot study in which we compared surveillance using the app to the current practice (paper maps). We hypothesized that inspectors would use the risk information provided by the app, as measured by the frequency of higher risk houses visited, and qualitative analyses of inspector movement patterns in the field. We also compared the efficiency of both mediums to identify factors that might discourage risk information use. Over the course of ten days (five with each medium), 1,081 houses were visited using the paper maps, of which 366 (34%) were inspected, while 1,038 houses were visited using the app, with 401 (39%) inspected. Five out of eight inspectors (62.5%) visited more higher risk houses when using the app (Fisher's exact test, p < 0.001). Among all inspectors, there was an upward shift in proportional visits to higher risk houses when using the app (Mantel-Haenszel test, common odds ratio (OR) = 2.42, 95% CI 2.00-2.92), and in a second analysis using generalized linear mixed models, app use increased the odds of visiting a higher risk house 2.73-fold (95% CI 2.24-3.32), suggesting that the risk information provided by the app was used by most inspectors. Qualitative analyses of inspector movement revealed indications of risk information use in seven out of eight (87.5%) inspectors. There was no difference between the app and paper maps in the number of houses visited (paired t-test, p = 0.67) or inspected (p = 0.17), suggesting that app use did not reduce surveillance efficiency.

Conclusions/significance: Without staying vigilant to remaining and re-emerging vector foci following a vector control campaign, disease transmission eventually returns and progress achieved is reversed. Our results suggest that, when provided the opportunity, most inspectors will use risk information to direct their surveillance activities, at least over the short term. The study is an initial, but key, step toward evidence-based vector surveillance.

Fig 1. T. infestans infestation risk map zoomed out (left) and zoomed in (right).

Houses are represented by dots that are colored according to their risk of infestation as estimated by the model. Legend translates to (from top to bottom): ‘Risk of infestation- Lowest; Low; Medium; High; Highest.’ Note that the images display infestation data overlay and app functions only; cartographic details (i.e., roads, parks, etc) have been removed for this publication. Images of the app with cartographic detail derived from data available at the OpenStreetMap project (openstreetmap.org) for the municipality of Arequipa, Perú, and served by MapBox (mapbox.com) are available in the VectorPoint repository, https://github.com/chirimacha/VectorPoint.

Fig 2. Part of the data collection form found in VectorPoint.

Spanish text translates to, from top to bottom: ‘Date’; ‘Property characteristics: Regular house’; ‘State of the inspection: inspection’; ‘Inspection area: Triatoma infestans (‘chiris’), signs of the bugs (‘rastros’), inside the house (‘intra’), yard and patio (‘peri’)’; ‘How many people live on the property?’; ‘What animals are there?: Dogs, Cats, Poultry, Guinea Pigs, Rabbits.’

Fig 3. Map of a locality as represented in the model throughout the four time periods (T1-T4).

Dots represent houses. T1 represents the time period of the ‘attack phase’ of the control campaign, which occurred between January 11th, 1997 and January 6th, 2014, depending on the district. In T1, grey dots represent houses that participated in the ‘attack phase’ of the control campaign but were not found to be infested with Triatoma infestans, and black dots represent houses that participated and were found to be infested with T. infestans. Time periods T2-T4 represent the time period of the surveillance phase (T2: January 7th, 2014—January 6th, 2016; T3: January 7th, 2016—January 4th, 2018; T4: the current calendar year, currently set to (at the time of this publication) January 5th, 2018—present). In T2-4, grey dots represent houses that were inspected and not found to be infested with T. infestans during the surveillance phase.

Fig 4. VectorPoint workflow diagram.

From left, risk estimates generated by the model are first sent to an SQL database, and then sent to the Shiny platform and server for visualization in the risk map. From right, data collected by the app are sent to the Shiny server and then to the SQL database. The model then pulls the new data from the SQL database the next time it is run. All data are TLS encrypted.

Fig 5. Example of a paper map used in vector surveillance under the current practice.

The numbers displayed on each block are the last three digits of the house code, with arrows indicating their ascending order. For example, a block with ‘1 → 12’ indicates that houses with unicodes ending in one through 12 are on that block. Street names and other identifying information have been removed.

Fig 6. Visit outcome distributions.

Inspections shown in blue, visit outcomes making up the ‘other’ category shown in shades of grey. Letters A-H refer to the corresponding inspector. *Significantly higher proportion of houses inspected when using the app (Fisher’s exact test, p<0.05).

Fig 7. Infestation risk distribution of all houses visited for each inspector.

Superscripts in the x axis text indicate arm order (i.e, which medium was used in the first week of the study, and which was used in the second week); values < 4% are not labeled due to space constraints; *p < 0.001, with more higher risk houses visited using the app; ^p < 0.001, with more higher risk houses visited using paper maps.