Spatiotemporal Fluctuations and Triggers of Ebola Virus Spillover

Abstract

Because the natural reservoir of Ebola virus remains unclear and disease outbreaks in humans have occurred only sporadically over a large region, forecasting when and where Ebola spillovers are most likely to occur constitutes a continuing and urgent public health challenge. We developed a statistical modeling approach that associates 37 human or great ape Ebola spillovers since 1982 with spatiotemporally dynamic covariates including vegetative cover, human population size, and absolute and relative rainfall over 3 decades across sub-Saharan Africa. Our model (area under the curve 0.80 on test data) shows that spillover intensity is highest during transitions between wet and dry seasons; overall, high seasonal intensity occurs over much of tropical Africa; and spillover intensity is greatest at high (>1,000/km2) and very low (<100/km2) human population densities compared with intermediate levels. These results suggest strong seasonality in Ebola spillover from wild reservoirs and indicate particular times and regions for targeted surveillance.

Plant Phenology Supports the Multi-emergence Hypothesis for Ebola Spillover Events

“…EMXT, the highest monthly daily maximum temperature, was a significant predictor for the number of Ebola spillover events in humans and animals. These results align to previous studies that have found a climatic dimension for Ebola spillover events (Tucker et al 2002; Pigott et al 2014; Schmidt et al 2017). In our study, however, we could additionally show that plant phenology variables [including the anomaly of Normalized Difference Vegetation Index (NDVI) between July and December, the proportion of population fruiting in Kibale National Park, Uganda, and the flowering anomalies in Lope, Gabon] informed neural network models with a superior fit to the data than when climate or climate in conjunction with phenology variables were used as inputs.…”

“…In our study, however, we could additionally show that plant phenology variables [including the anomaly of Normalized Difference Vegetation Index (NDVI) between July and December, the proportion of population fruiting in Kibale National Park, Uganda, and the flowering anomalies in Lope, Gabon] informed neural network models with a superior fit to the data than when climate or climate in conjunction with phenology variables were used as inputs. Previous studies using Normalized Difference Vegetation Index or Enhanced Vegetation Index have not yet considered their predictive power for Ebola spillover events independently from climate variables (see Schmidt et al 2017). …”

“…This seasonal association between Ebola spillover events and a multitude of phenology variables describing seasonality of flowering and fruiting was not previously reported and corroborates the newly found close association between inter-annual Ebola spillover events and inter-annual variation in vegetation and phenology variables. The seasonality in spillover events is thought to be associated with transitions between wet-to-dry and dry-to-wet conditions (Tucker et al 2002; Pinzon et al 2004; Groseth et al 2007; Altizer et al 2013; Schmidt et al 2017), and thus may be associated with spatiotemporal fluctuations of the African monsoon (Cornforth 2013). While the September spike in human + other animal spillover events was associated with such a transition of climatic variables, the December spike was not.…”

“…Evapotranspiration and Enhanced Vegetation Index are predictive of the spatial environmental niche of Ebola virus (i.e. being located in the African tropical rainforest biome), and monthly rainfall and rainfall anomaly can predict Ebola virus spillovers (Pinzon et al 2004; Schmidt et al 2017). …”

Confronting models with data: the challenges of estimating disease spillover

“…By phenomenological, we mean approaches that relate the observed spillover events, z, as the dependent variable to some linear or nonlinear function of predictor variables, X, and coefficients b (e.g. [5,8,9]). Alternatively, one could combine information on host species density, prevalence, contact rate and infection rates to predict the number of spillovers,ẑ, [7] even when observed data on spillovers are rare or non-existent.…”