Predicting suitability of regions for Zika outbreaks with zero-inflated models trained using climate data

Abstract

AbstractIn recent years, Zika spread through the Americas. This virus has been linked to Guillain-Barré syndrome, which can lead to paralysis, and microcephaly, a severe birth defect. Zika is primarily transmitted by Aedes (Ae.) aegypti, a mosquito whose geographic range has expanded and is anticipated to continue shifting as the climate changes.We used statistical models to predict regional suitability for autochthonous Zika transmission using climatic variables. By suitability for Zika, we mean the potential for an outbreak to occur based on the climate’s habitability for Ae. aegypti. We trained zero-inflated Poisson (ZIP) and zero-inflated negative binomial (ZINB) regression models to predict Zika outbreak suitability using 20 subsets of climate variables for 102 regions. Variable subsets were selected for the final models based on importance to Ae. aegypti survival and their performance in aiding prediction of Zika-suitable regions. We determined the two best models to both be ZINB models. The best model’s regressors were winter mean temperature, yearly minimum temperature, and population, and the second-best model’s regressors were winter mean temperature and population.These two models were then run on bias-corrected climate projections to predict future climate suitability for Zika, and they generated reasonable predictions. The predictions find that most of the sampled regions are expected to become more suitable for Zika outbreaks. The regions with the greatest risk have increasingly mild winters and high human populations. These predictions are based on the most extreme scenario for climate change, which we are currently on track for.Author Summary:In recent years, Zika spread through the Americas. This virus has been linked to Guillain-Barré syndrome, which can lead to paralysis, and microcephaly, a severe birth defect. Zika is primarily transmitted by Aedes (Ae.) aegypti, a mosquito whose geographic range has expanded and is anticipated to continue shifting as the climate changes. We used statistical models to predict regional suitability for locally-acquired Zika cases using climatic variables. By suitability for Zika, we mean the potential for an outbreak to occur based on the climate’s habitability for Ae. aegypti. We trained statistical models to predict Zika outbreak suitability using 20 subsets of climate variables for 102 regions. Variable subsets were selected for the final two models based on importance to Ae. aegypti survival and their performance in aiding prediction of Zika-suitable regions. These two models were then run on climate projections to predict future climate suitability for Zika, and they generated reasonable predictions. The predictions find that most of the sampled regions are expected to become more suitable for Zika outbreaks. The regions with the greatest risk have high human populations and increasingly mild winters.