From data to action: a machine learning model to support tick-borne encephalitis surveillance and prevention in Europe

Background Tick borne encephalitis (TBE) is a severe zoonotic neurological infection caused by the TBE virus (member of the Flaviriridae family) and it is one of the most important tick-borne viral diseases in Europe and Asia. The infection is mostly acquired after a tick bite, but alimentary infection is also possible. Despite the availability of a vaccine, TBE incidence is increasing with the appearance of new foci of virus circulation in new endemic areas. The increase in TBE cases across Europe - from 2412 in 2012 to 3514 in 2022, has highlighted the need for predictive tools capable to identify areas where human TBE infections are likely to occur. In response, this study presents a novel spatio-temporal modelling framework that provides annual predictions of the occurrence of human TBE infections across Europe, at both regional and municipal levels. Methods We used data on confirmed and probable TBE cases provided by the European Surveillance System (TESSy, ECDC) to infer the distribution of TBE human cases at the regional (NUTS3) level during the period 2017-2022. We trained the model on data from countries with sufficient reporting, i.e., that provided the location of infection at the NUTS-3 level for at least 75% of cases notified during the selected period. To account for the natural hazard of viral circulation, we included variables related to temperature (derived from satellite images acquired by the Moderate Resolution Imaging Spectroradiometer (MODIS) and supplied by NASA with a resolution of 5.6 km), precipitation (derived from the ECMWF ERA5-Land dataset at 30 arc seconds resolution), land cover (extracted from the 2018 Corine Land Cover (CLC) data inventory (class “3.1”) with a resolution of 0.25x0.25 km) and ticks’ hosts presence (originally produced using random forest and boosted regression trees approaches). We also used indexes based on recorded intensities of human outdoor activity in forests (based on the OpenStreetMap database) and population density (obtained from WorldPop) as proxies of human exposure to tick bites. We identified the yearly probability of TBE occurrence using a spatio-temporal boosted regression tree modeling framework. Results Our results highlight a statistically significant rising trend in the probability of human TBE infections not only in north-western, but also in south-western European countries. Areas with the highest probability of human TBE infections are primarily located in central-eastern Europe, the Baltic states, and along the coastline of Nordic countries up to the Bothnian Bay. Such areas are characterised by the presence of key tick host species, forested areas, intense human recreational activity in forests, steep drops in late summer temperatures and high precipitation amounts during the driest months. The model showed good predictive performance, with a mean AUC of 0.85, sensitivity of 0.82, and specificity of 0.80 at the regional level, and a mean AUC of 0.82, sensitivity of 0.80, and specificity of 0.69 at the municipal level. Discussion With ongoing climate and land use changes, the burden of human TBE infections on European public health is likely to increase, as trends are already indicating. This underscores the need for predictive models that can help prioritize intervention efforts. Hence, the development of a modeling framework that predicts the probability of human TBE infections at the finest administrative scale based on easily accessible covariates, represents a step forward towards comprehensive TBE risk estimation in Europe