Modeling the spatial distribution of Chagas disease vectors using environmental variables and people´s knowledge
Jaime Hernández, Ignacia Núñez, Antonella Bacigalupo, Pedro E Cattan
DOI: 10.1186/1476-072X-12-29
The distribution of two triatomine species: Triatoma infestans and Mepraia spinolai were modeled across different spatial scales within Chile. Each species is associated with particular niches and their risk of transmitting Chagas disease to humans varies (i.e. depends on the degree of domiciliation). Hernandez et. al. uses ENM to address spatial and temporal issues relevant to domestic transmission control and despite lack of data available, makes a predictive model by extrapolating actual data of the ecological niches to areas that have similar characteristics across different regions of Chile. Regardless of the degree of affiliation with human dwelling (which could omit the whole point of using ENM), relevant macro environmental variables from satellite based imagery and triatomine presence/absence data were used. To make a predictive model, authors used the machine learning algorithm Random Forest to predict the probability of triatomine presence. Random Forest generated model statistics to deliver strong information predictors across 10km, 5km, and 2.5km scales which were used to make the most suitable predictive model. Despite creating a model that ultimately determined degree of overlap between vector species and predicted that T. infestans can persist outside of domestic conditions, the type of data (presence/absence) does not seem like the best choice overall (because there could be false absence). Instead, to create a model where data is already scarce, esp. concerning true absences, it would be safest (esp. in a human disease health risk scenario) to use presence/background based model.