Zoonoses are infectious human infectious diseases acquired from animals. Most emerging human pathogens are zoonoses. The ecology of animal reservoir species and animal-human interactions are therefore important determinants of the frequency and location of spillover events. Our research aims to develop models and data-driven methods for prediction, mapping, and forecasting. Filoviruses, including Ebola virus and Marburg virus, are exemplary zoonoses that frequently spillover into human and non-human primate populations and occasionally cause epidemics/epizootics. Anecdotal evidence suggests that spillover events and ensuing transmission are intensified by changing environmental conditions. But, discerning environmental triggers of these rare events from the vast constellation of environments realized in sub-Saharan Africa during the time since Ebola's original description is an exceedingly challenging problem for statistical learning. We developed a new method for space-time risk mapping over spillover intensity by combining ecological knowledge, environmental and weather data, and techniques from statistical learning (namely, bootstrap aggregation of weakly regularized, low-dimensional regression equations).
Results/Conclusions
Tested on data withheld from model development, our model predicted historic Ebola spillover events with remarkable accuracy (AUC = 0.80). By contrast, performance on training data was not much greater (AUC = 0.83), indicating the high generalizability of this approach. Inspection of model sensitivity to data inputs shows that spillover intensity varies seasonally over much of tropical Africa and is greatest at high (>1,000/km2) and very low (<100/km2) human population densities compared with intermediate levels. Predicted spillover intensity was generally, but not always, lowest in dry months (rainfall <50 mm). It is hypothesizes that spillover risk might be affected by the concentration of human population, which governs the number of opportunities for infection to be transmitted from animals to humans, but we found statistically the effect of human population on Ebola spillover intensity to be much smaller than climatic or seasonal effects. Taken together, these results indicate that there is strong seasonality in Ebola spillover from wild reservoirs and indicate particular times and regions for targeted surveillance. It is envisioned that the model could form the basis for an automated information system that quantifies spillover risk.