A logistic regression model to predict the next rabies virus host-shift event

Abstract

Abstract Rabies virus (RABV) host-shift events (HSEs) are thought to be promoted by viral genomic and ecological factors, but the relative balance of the two is unclear. Using a dataset of 19,170 species pairs that were known or not known to be linked by an HSE, we developed a logistic regression model to explore how biological and ecological characteristics of cross-species infections and their associated RABV variant (RVV) influence HSE risk. The model incorporates relatedness, body temperature, litter size, adult weight, and the broad lineage (bat or canine) of the RABV variant maintained by the reservoir species. Assessed with leave-one-out cross validation, the model identifies known HSEs with 90% accuracy (sensitivity 90%, specificity 82%). The susceptible-reservoir infection on each continent with the highest risk of HSE are coyotes with Canine-associated RVVs (North/Central America, RR = 187.0), culpeos with Canine-associated RVVs (South America, 100.9), dholes with Canine-associated RVVs (Asia, 159.8), arctic foxes with Raccoon Dog RVV (Europe, 115.8), and African wild dogs with Canine-associated RVVs (Africa, 134.8). The results of this model can be used to help predict the next HSE, identify potential cryptic RABV reservoirs, inform contingency actions when a high-risk event is identified, and prepare for importation or incursion events.