henhaobio
Sign in

One HealthEcologyPeer-reviewed journal article

Modeling spatio-temporal annual changes in the probability of human tick-borne encephalitis (TBE) occurrence across Europe

Francesca DagostiniD, Diana ErazoiD, Giovanni MariniiD, Daniele Da ReiD, Valentina TagliapietraiD, Mária AvdičováiD, Tatjana Avšič‐ŽupanciD, Timothée DubiD, Nahuel FioritoiD, Nataša KnapiD, Céline M GossneriD, Jana KerlikiD, Henna MäkeläiD, Mateusz MarkowicziD, Roya OlyazadehiD, Lukas RichteriD, William WintiD, Maria Grazia Zuccali, Milda Žygutienė, Simon DellicouriD, Annapaola P. RizzoliiD

PDFDOIPreprintData
In-depth full-text digest available · Jump to digest ↓

Abstract

IntroductionTick-borne encephalitis (TBE), caused by tick-borne encephalitis virus (TBEV), is a zoonotic disease that can cause severe neurological symptoms. Given the increasing number of reported human TBE cases in Europe, a spatio-temporal predictive model to infer the year-to-year probability of human TBE occurrence across Europe at the regional and municipal administrative levels was developed. MethodsThe distribution of human TBE cases at the regional (NUTS-3) level during the period 2017-2022, was derived by using data provided by the European surveillance system (TESSy, ECDC), and at the municipal level by using data from Austria, Finland, Italy, Lithuania, and Slovakia. The probability of presence of human TBE cases at the regional and municipal levels for the period 2017-2024 was modeled with a boosted regression trees model, including covariates that affect both the natural hazard of virus circulation and human exposure to tick bites. ResultsAreas 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. Our results also 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, offering a spatio-temporal predictive framework for the assessment of areas where human TBE infection are most likely to occur. The model showed good predictive performance, with a mean AUC of 0.85 (SD = 0.02), sensitivity of 0.82, and specificity of 0.80 at the regional level, and a mean AUC of 0.82 (SD = 0.03), sensitivity of 0.80, and specificity of 0.69 at the municipal level. DiscussionWith ongoing climate and land use changes, the number of human TBE cases is likely to increase and expand into new areas, as trends are already indicating. This underscores the need for predictive models that can help prioritize intervention efforts. The approach adopted, by leveraging lagged covaries, enables timely one-year-ahead predictions, thus supporting surveillance, prevention, and control of human TBE infections by public health authorities. StatementsO_ST_ABSEthical statementC_ST_ABSEthical approval was not needed. Funding statementThis project has received funding from the European Unions Horizon 2020 research and innovation programme under grant agreement No 874850 and is catalogued as MOOD 081. The contents of this publication are the sole responsibility of the authors and dont necessarily reflect the views of the European Commission. Conflict of interestNone. Authors contributionsFrancesca Dagostin: Conceptualization, Methodology, Data Curation, Formal Analysis, Writing - Original Draft. Diana Erazo: Conceptualization, Methodology, Writing - Review & Editing. Giovanni Marini: Conceptualization, Methodology, Writing - Review & Editing. Daniele Da Re: Conceptualization, Methodology, Writing - Review & Editing. Valentina Tagliapietra: Conceptualization, Methodology, Writing - Review & Editing. Maria Avdicova: Resources, Writing - Review & Editing. Tatjana Av[s]i[c] - [Z]upanc: Resources, Writing - Review & Editing. Timothee Dub: Conceptualization, Resources, Writing - Review & Editing. Nahuel Fiorito: Resources, Writing - Review & Editing. Nata[s]a Knap: Resources, Writing - Review & Editing. Celine M. Gossner: Resources, Writing - Review & Editing. Jana Kerlik: Resources, Writing - Review & Editing. Henna Makela: Resources, Writing - Review & Editing. Mateusz Markowicz: Resources,Writing - Review & Editing. Roya Olyazadeh: Resources, Writing - Review & Editing. Lukas Richter: Resources, Writing - Review & Editing. William Wint: Resources, Writing - Review & Editing. Maria Grazia Zuccali: Resources, Writing - Review & Editing. Milda [Z]ygutiene: Resources, Writing - Review & Editing. Simon Dellicour: Methodology, Writing - Review & Editing. Annapaola Rizzoli: Conceptualization, Methodology, Writing - Review & Editing. Data availabilityThe data that support the findings of this study were provided by ECDC, Azienda Provinciale per i Servizi Sanitari Provincia Autonoma di Trento (APSS), Unita Locale Socio Sanitaria Dolomiti (ULSS N.1 Dolomiti), Public Health Authority of the Slovak Republic, Austrian Agency for Health and Food Safety (AGES), Finnish Institute for Health and Welfare (THL), National Public Health Center under the Ministry of Health (Lithuania) and University of Ljubljana. Restrictions apply to the availability of these data, which were used under license for the current study, and so are not publicly available. The interactive risk maps can be explored in detail at https://mood-platform.avia-gis.com. DisclaimerThe views and opinions of the authors expressed herein do not necessarily state or reflect those of ECDC. The accuracy of the authors statistical analysis and the findings they report are not the responsibility of ECDC. ECDC is not responsible for conclusions or opinions drawn from the data provided. ECDC is not responsible for the correctness of the data and for data management, data merging and data collation after provision of the data. ECDC shall not be held liable for improper or incorrect use of the data.

In-depth summary

Claude Opus 5.5 · max effort · 8 Nov 2026 · from the full text of version 3 (2 Dec 2024)

Boosted regression trees predict, one year ahead, where human tick-borne encephalitis (TBE) cases are likely in Europe (AUC 0.85 regional, 0.82 municipal), with probability rising in central-western Europe.

Reported TBE cases in Europe rose from 2412 in 2012 to 3514 in 2022, yet prior European models mostly mapped tick vectors or virus suitability in ticks and hosts, human-case work covered limited areas or single countries (Finland, Germany, Sweden), and continent-wide maps of notification rates are usually at country resolution. This work predicts human case occurrence across Europe at regional and municipal level one year ahead, with human-exposure predictors the authors say earlier models generally lacked.

Key results

  1. ShownTrained on 2017-2021 human TBE presence/absence in NUTS-3 (Nomenclature of Territorial Units for Statistics, level 3) regions, the regional model performed comparably in spatial cross-validation and on 2022 notifications held out from training.

    Leave-location-out spatial block cross-validation, 10 replicates: mean AUC 0.85 (SD 0.02), sensitivity 0.82, specificity 0.80, prevalence-pseudoabsence-calibrated Sørensen's index (SIppc) 0.71. External test on 2022 NUTS-3 records: AUC 0.87, sensitivity 0.81, specificity 0.78, SIppc 0.73. · Predictions of the probability of human TBE…; Boosted regression trees modelling framework; Abstract

  2. SuggestedPer-area regressions of predicted probability on year (2017-2024) indicate significant increases across central-western Europe, with newly suitable areas to the north-west and possibly south-west, and decreases in parts of Eastern Europe.

    Generalised linear models (prediction ~ year) per NUTS-3 region and municipality; coefficients with p < 0.05 mapped (Figure 5). Largest increases in western Germany, southern Belgium, eastern France and Switzerland; new suitable areas include Belgium, France and the United Kingdom. No numeric slopes are given in the text. Observed positive NUTS-3 regions: 255 in 2017, 279 in 2022. · Short-term variability of predicted probability…; Boosted regression trees modelling framework; Figure legends; Results

  3. ShownThe highest predicted probabilities of human TBE presence lie mainly in central-eastern Europe, the Baltic states and along Nordic coasts as far as the Bothnian Bay, with patchy local heterogeneity at municipal resolution.

    Based on annual 2017-2024 probability maps at regional (Figure 3) and municipal (Figure 4) level; the text gives no per-area values. The text describes observed cases extending west to the Ain department of France, south to northern Italy, east to the Balkans and north to Norway and Finland; the study dataset (Figure 2) has 1,658 NUTS-3 presences across 15 countries. · Abstract; Predictions of the probability of human TBE…; Results; Discussion

  4. ShownApplied to municipality-averaged covariates, the regional model gave finer maps with similar AUC but lower specificity, judged against finer-scale case data from five countries.

    External validation on 2017-2022 presence/absence data from Austrian districts and municipalities in Finland, northern Italy, Lithuania and Slovakia: mean AUC 0.82 (SD 0.03), sensitivity 0.80, specificity 0.69 (regional 0.80), Sørensen's index 0.56 (regional 0.71). · Predictions of the probability of human TBE…; Abstract; Boosted regression trees modelling framework; Collection of epidemiological data

  5. ShownHabitat suitability for key hosts (rodents and deer) carries the largest relative influence, followed by forest road density and driest-quarter precipitation; tick vector distribution layers did not improve the model and were dropped.

    Relative influence: host presence probability 39.44, forest road and path density 21.39, driest-quarter precipitation 20.19, forest proportion 14.85, population density 2.37 (all positive associations), autumnal cooling rate 1.77 (steeper cooling, higher suitability). Removing the host layer lowered performance (Supplementary Table S4); Ixodes ricinus and I. persulcatus layers had no significant effect. · Fitted functions and relative importance of…; Discussion; Hazard and exposure covariates

Methods, data and reproducibility

Boosted regression trees (leave-location-out spatial cross-validation) were fitted to annual presence/absence of human TBE cases in NUTS-3 regions (2017-2021) using hazard covariates (forest cover, one-year-lagged climate, host suitability) and exposure covariates (forest road density, population density), then applied to municipal covariates; per-area regressions of predictions on year mapped trends.

  • Confirmed and probable human TBE cases from The European Surveillance System (TESSy) of the European Centre for Disease Prevention and Control (ECDC), with place of infection at NUTS-3, 2017-2022 · human · n = 13 countries (at least 75% of cases located at NUTS-3)
  • Austrian (Austrian Agency for Health and Food Safety) and Slovenian (University of Ljubljana) regional data; Spain and Ireland included as countries with no autochthonous cases · human
  • Combined NUTS-3 presence/absence set (presence = at least one confirmed case in a year): 2017-2021 for training, 2022 held out for external validation · human · n = 1,658 presences; 4,096 case-free regions (15 countries plus Spain, Ireland)
  • Finer-scale case data, 2017-2022: Austrian districts and municipalities of Finland, Slovakia, Lithuania and northern Italy (Trento and Belluno health agencies); used for external validation · human · n = five countries
  • Weighted probability of presence of critical host species (rodents and deer), previously produced with random forest and boosted regression tree spatial models; held constant over time · rodents and deer
  • Exposure covariates: density of roads and paths in forests from OpenStreetMap (accessibility proxy) and human population density from WorldPop
  • Hazard covariates: forest cover (%), and total precipitation of the driest quarter and autumnal cooling rate, each computed per year with a one-year lag (2016-2023)
Methods & tools
R 4.1.2, gbm (R package), mlr3 (R framework), blockCV (R package), TESSy (ECDC), OpenStreetMap, WorldPop
Reagents & environment
R Statistical Software v.4.1.2 with the R packages gbm, mlr3 and blockCV; package versions, containers and requirement files are not stated.
Reproducibility gaps
Case data from ECDC and national/local health agencies were used under licence and are not publicly available; no code repository is given; the only link is to interactive risk maps; covariate sources and detailed metrics are in Supplementary Material.
Limitations (6)
  • Host occurrence enters as a static habitat-suitability layer for rodents and deer, so annual and interannual changes in host density are not represented, although the authors report that removing this layer worsened performance (Table S4). · stated by the authors
  • The UK, Belgium and the Netherlands, where TBE is emerging, were absent from TESSy or lacked NUTS-3 locations and so contributed no training data; the authors note that adding such data would improve predictions there. · stated by the authors
  • The outcome is notified human cases: 'absence' means no reported case rather than no virus circulation, tick-borne and food-borne infections are not distinguished, and vaccination coverage could not be included for lack of data in all countries. · stated by the authors
  • Municipal maps come from applying the regional model to municipal covariates, since municipal data were too sparse to train on, and were validated only with data from five countries (Austria, Finland, Italy, Lithuania, Slovakia), where specificity fell to 0.69. · stated by the authors
  • Only the autumnal cooling rate and driest-quarter precipitation are described as time-varying, so the 2017-2024 trend maps track lagged climate variation in model output, including 2023-2024 predictions with no case data to check them, rather than observed incidence. · noted in this summary
  • No comparison with other algorithms or with a simple baseline such as previous-year case presence is reported, so the gain over carrying forward known case locations is not quantified. · noted in this summary

Read this if you need annual sub-national predictions of where human TBE cases are likely in Europe, or a template for spatially cross-validated boosted regression tree models built on surveillance notifications and lagged covariates.

Citations
12
FWCI
50.85
Field percentile
top 1%

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISHolle '26Mahon '20Hussain '26Métras '25Ito '26Hussain '26
62% semantic matchEcology
Valén Holle, Raphaëlle Klitting, Nadja Kabisch+1 more·bioRxiv·Jun 10, 2026