Climate change projections of West Nile virus infections in Europe: implications for blood safety practices

Background West Nile virus (WNV) is transmitted by mosquitoes in both urban as well as in rural environments and can be pathogenic in birds, horses and humans. Extrinsic factors such as temperature and land use are determinants of WNV outbreaks in Europe, along with intrinsic factors of the vector and virus. Methods With a multivariate model for WNV transmission we computed the probability of WNV infection in 2014, with July 2014 temperature anomalies. We applied the July temperature anomalies under the balanced A1B climate change scenario (mix of all energy sources, fossil and non-fossil) for 2025 and 2050 to model and project the risk of WNV infection in the future. Since asymptomatic infections are common in humans (which can result in the contamination of the donated blood) we estimated the predictive prevalence of WNV infections in the blood donor population. Results External validation of the probability model with 2014 cases indicated good prediction, based on an Area Under Curve (AUC) of 0.871 (SD = 0.032), on the Receiver Operating Characteristic Curve (ROC). The climate change projections for 2025 reveal a higher probability of WNV infection particularly at the edges of the current transmission areas (for example in Eastern Croatia, Northeastern and Northwestern Turkey) and an even further expansion in 2050. The prevalence of infection in (blood donor) populations in the outbreak-affected districts is expected to expand in the future. Conclusions Predictive modelling of environmental and climatic drivers of WNV can be a valuable tool for public health practice. It can help delineate districts at risk for future transmission. These areas can be subjected to integrated disease and vector surveillance, outreach to the public and health care providers, implementation of personal protective measures, screening of blood donors, and vector abatement activities. Electronic supplementary material The online version of this article (doi:10.1186/s12940-016-0105-4) contains supplementary material, which is available to authorized users.


Introduction
"experience only with flu-like symptoms" -remove word "with" "and Ukraine, but since 1996 has caused..." → change to "and Ukraine. However, since 1996 WNV has caused..." Before the discussion about outbreaks in rural and urban areas, it would be useful to briefly explain the transmission cycle and different hosts (e.g. birds, horses, humans...) "Two types of land-use have been associated with outbreaks in Europe, namely rural and urban areas" → change to "WNV outbreaks occurs in both rural and urban areas" Last paragraph of introduction: -explain that the different predictors were tested in a statistical model. -explain what the A1B climate change scenario is here, and place it in context with the range of different scenarios available.

Methods
Change sub title to "Temperature data" Are the monthly anomalies for mean temperature? (or max or min?) When describing WNV epidemiology, remind the reader when the WNV transmission season in Europe takes place.
Is the sentence about temperature data at the end of the WNV epi section needed?
Environmental variables: explain the motivation for using NDVI/MNDWI values. How could they help to predict WNV transmission?
Multivariate models: Although the reference to the article describing the model is included, it would be useful to include here some more details about the model, e.g. the probability distribution assumed (Gaussian? Binomial) and how the response variable was constructed (e.g. frequency?) "the occurrence of a WND outbreak of the previous year" On page 7, should WND read WNV? Change "of" to "in" "logistic regression models' coefficients" → change to "parameter estimates" "λ: Weighted average of the number of infected districts amongst the neighbourhood the previous year" -Does the neighbourhood refer to the neighbouring NUTS3 regions?
"July temperature anomalies were entered for each year between 2015 and 2050 in order to compute the probability of WNV infections per district and per year which is one of the parameters in the model (infection the previous year) [10]." -Is this parameter λ? If so, include in text.
Provide some more details about the EUFRAT tool.

Results
How was a "correct prediction" defined? Did you use a threshold probability to decide if the model correctly predicted an outbreak or not?

Figures
In Fig 6 it is not clear what part A and B are showing. It would be good to explain in the caption. The authors might consider using a different colour scale to show old and new districts (so as not to confuse with the probability maps).
In Fig 7 it would be good to explain the difference between the prevalence estimates and the probability maps, and if relevant, consider using a different colour to show prevalence.

Background:
West Nile virus (WNV) is transmitted by mosquitoes in both urban as well as in rural environments and can be pathogenic in birds, horses and humans. Extrinsic factors such as temperature and land use are effective determinants of WNV outbreaks in Europe, along with intrinsic factors of the vector and virus.

Methods:
With a multivariate model for WNV transmission we computed the probability of WNV infection in 2014, with July 2014 temperature anomalies. We applied the July temperature anomalies under the A1B climate change scenario for 2025 and 2050 to model and project the risk of WNV infection in the future. Since asymptomatic infections are common in humans (which can result in the contamination of the donated blood) we estimated the predictive prevalence of WNV infections in the blood donor population.

Results:
External validation of the probability model with 2014 cases indicated good prediction, based on an Area Under Curve (AUC) of 0.871 (SD=0.032), on the Receiver Operator Characteristic Curve (ROC). The climate change projections for 2025 reveal a higher probability of WNV infection particularly at the edges of the current transmission areas (for example in Eastern Croatia, NorthEastern Greece and Northwestern Turkey) and an even further expansion in 2050. The prevalence of infection in (blood donor) populations in the outbreak-affected districts will also expand in the future.

Discussion:
Predictive modelling of environmental and climatic drivers of WNV can be a valuable tool for public health practice. It can help delineate districts at risk for future transmission. These areas can be subjected to integrated disease and vector surveillance, outreach to the public and health care providers, implementation of personal protective measures, screening of blood donors, and vector abatement activities.

Introduction
West Nile virus (WNV) is responsible for the largest outbreaks of fatal neuroinvasive disease in the world [1]. WNV infections occur predominantly through mosquito bites but also through blood transfusion or organ, tissue and cell-transplantations. Most human infections are asymptomatic and mild cases experience only with flu-like symptoms; more severe cases present with signs of encephalitis, meningo-encephalitis or meningitis. Globally, WNV is the most widespread arthropod-borne virus, an enveloped, single-strand RNA virus of the genus Flavivirus in the family of Flaviviridae [2][3][4]. It circulates in Africa, Americas, Asia, Europe, and Australia, where it is thought have been introduced from the Middle East [4].
In Europe, Middle East and Africa, WNV has been responsible for sporadic outbreaks in the 1950s in Israel, in the 1960s in Russia and France, in the 1970s in Belarus, South Africa, and Ukraine, but since 1996 has caused more recurrent outbreaks in Europe and northern Africa [5][6][7][8]. In 2010, large outbreaks in humans occurred in Southeastern and Eastern Europe. The European outbreaks occurred in Russia, the Czech Republic, Hungary, Romania, Turkey, Greece, Italy, France, Spain and Portugal [9]. Since 2010, there have been annual outbreaks in Southeastern and Eastern Europe, suggesting an endemic transmission cycle and thus a resurgent public health problem [10]. Two types of land-use have been associated with outbreaks in Europe, namely rural and urban areas. They respectively (and independently) contribute to high concentrations of hosts with competent mosquito vectors that support intense local avian transmission. Rural areas with estuaries, wetlands or marshes attract migratory birds for breeding, nesting, and rearing their young. These bird habitats also attract bird-feeding mosquitoes where congregating bird populations can get infected with WNV. In fact, a number of human outbreaks have originated in European estuaries such as in the Danube delta in Romania, in the Volga delta in Russia, and in the Rhone delta in France [11]. Urbanized areas can also attract large bird and mosquito populations where humans can get exposed [12] and local environmental conditions can increase the potential for mosquito breeding in urban settings [13]. For example, in 1996, a large outbreak occurred in the city of Bucharest, Romania that affected 4% of the population [7]. Over 800 patients were hospitalized during another large WNV outbreak that occurred in 1999 in Volgograd City on the west bank of the Great Volga River [14] and in 2010 WNV infections were documented for the first time in humans in the Greek city of Thessaloniki [15,16].
Certain metropolitan areas can sustain a high breeding density of birds; for example, European Starlings thrive on urban lawns or parks where they can feed, or gulls proliferate near open water [17]. Birds from these types of urban environments harbour viruses with higher genetic diversity than birds from residential areas, indicating that anthropogenic factors associated with urbanization play an important role in arboviral transmission and evolution [18]. In the United States, WNF infection rates in crows and humans are higher in more urbanized environments that are less forested [19,20], while in Europe, WNV transmission can occur both in rural and urban areas.
Ambient temperature is another important environmental determinant in the transmission of WNV as it has a direct impact on mosquito survival, developmental rates of immature stages, growth rates of vector populations and decrease in the interval between blood meals [21][22][23][24]. Moreover, temperature also affects the extrinsic incubation period (the number of days from ingestion to transmission) by influencing the viral replication rates and thus the transmission of WNV [25,26]. In a modelling study elevated air temperature was the strongest predictor of increased infection in mosquito vectors [24]. As for Europe, it was found that the unprecedented upsurge in the number of human WNF cases in summer 2010 was accompanied by extremely hot spells in the region [21]. Moreover, recent research analysed the status of infection by WNV in Europe and its neighbouring countries in relation to environmental and climatic risk parameters. The anomalies of temperature in July were identified as one of the main risk factors [10]. During recent decades, parts of Europe have warmed up more than the global average. Additionally, more frequent and more intense hot extremes have occurred. This trend is expected to continue while predictions suggest a further temperature increase (between 1.0°C and 5.5°C) by the end of the century [27].
Increases in ambient temperature due to climate change are therefore projected to impact WNV transmission in Europe and its neighbouring countries [28][29][30][31][32].
In order to examine these land-use and climatic variables as predictors of the probability of WNV infection [4] we tested the contribution of remotely sensed temperature, the state of vegetation and water bodies, and bird migratory routes. We also project the WNV risk in Europe into 2025 and 2050, with July temperature projections under the A1B climate change scenario. Insights from these analyses can also be used to assess the current and future WNV risk to the safety of blood supply in the region [33]. Thus, we provide a quantification of the risk of WNV transmission through blood transfusion by estimating the prevalence of infection in (donor) populations of an area affected by a WNV outbreak, as well as the infection risk of a blood donor that visited an outbreak affected area. In the longrun, the environmental determinants identified in this model lend themselves for an integration of environmental monitoring in public health surveillance systems of human cases, serological surveillance of domestic and wild avifauna, and entomological surveillance [4,34,35].

Temperature
Monthly anomalies of July 2014 temperature at the locations of WNF outbreaks reported in humans were computed and predicted surface temperatures and anomalies were extracted for 2015-2050 from NCAR climate change scenarios for the gridded region 30°N-60°N and 10°W-55°E [36]. The A1B Scenario was chosen since it is a balanced scenario; its main characteristics include: low population growth, very high GDP growth, very high energy use, low-medium land use changes, medium resource (mainly oil and gas) availability, rapid pace and direction of technological change favouring balanced development. The chosen model output is the Ensemble Average which is the mean state of the climate among all model runs. Global climate simulations were produced at NCAR by the Community Climate System Model (CCSM3) for the 4th Assessment report of the IPCC [37].
Climate simulations from the CCSM3 are generated on a Gaussian grid, where each grid point can be uniquely accessed by one-dimensional latitude and longitude arrays (i.e. the coordinates are orthogonal). In the CCSM3 model output, the longitudes are equally spaced at 1.40625º, while the latitudes vary in spacing from 1.389º to 1.400767º. Therefore, the approximate spatial resolution of the global climate projections is 155 km. Because of the irregular grid in the CCSM model, this portal distributes data in a point shape file format, where each point represents a centroid of a corresponding CCSM grid cell. Anomalies are relative to the 20th Century Experiment 1980-1999, based on NCAR methodology [37].

WNV epidemiology
The methods for the WNV epidemiological model in Europe have been described previously [10]. Briefly, the epidemiologic data of human West Nile cases were obtained from the West Nile fever surveillance conducted at ECDC during the transmission season in Europe [38]. Population data for the European Union were based on the nomenclature of territorial units for statistics classification (NUTS) at level 3 with population estimate of 2010 [39].
Global Administrative Unit Layers (GAUL) were used for regions outside of the European Union, and project and population estimates for the year 2010 were derived from the Gridded Population of the World (GPW) dataset [40]. The gridded data of monthly mean of air temperature for the region between 30°N-60°N and 10°W-55°E was obtained from the NOAA NCEP-NCAR database [41].

Multivariate models
We used multivariate logistic regression models to test the probability of an 'infected district' as the response variable, and as explanatory variables the population, the presence of wetlands, the presence of birds' migratory routes, the anomalies of temperature, NDVI and MNDWI [10]. We also tested as explanatory variable: the occurrence of a WND outbreak of the previous year, considering that WNV could persist locally through survival in overwintering mosquitoes or infected birds. A bootstrap procedure (1,000 replicates) was applied to estimate the 95% confidence interval (95% CI) of the logistic regression models' coefficients, selecting randomly each time from the original set of 1113 districts 90% of infected districts between 2002 and 2011 (n = 98) and 90% of non-infected districts (n = 903). The final model used in the current analysis is described in Table 1 The average, standard deviation, and anomalies of the 2014 temperature was computed and applied to the model described above [10]. The results were compared with the actual occurrence of WNV in 2014. Validity of the model was assessed based on the ability of the model to distinguish between districts with and without WNV, using sensitivity, specificity, and Area Under Curve (AUC) of the Receiver Operator Characteristic Curve (ROC).

Climate change projections
The model was used to predict the probability of WNV infection by applying the projected July temperatures for 2025 and 2050. We extracted gridded temperature projections for the A1B Scenario, which is a balanced climate change scenario [37]. WNV infections are a significant concern to the safety of the blood supply, because the blood donated by asymptomatic carriers might inadvertently contaminate the blood supply [33]. Therefore, we calculated the prevalence of infection in (donor) populations in the outbreak-affected areas. We mapped the donor population infectivity (figure 7) and found an extended area of elevated WNV infection hazard for the safety of the blood supply in 2025 compared to 2014.

Projections of WNV risk
The Blood supply safety The arrival and dispersal of tropical pathogens to Europe and its neighbouring countries commonly associated with warmer temperatures pose a threat to the supply of safe blood products, particularly if they are unknown or without diagnostic tests [33,43]. Thus, emerging infectious diseases will continue to pose a threat to transfusion safety on European and international levels [44][45][46]. Specifically, a progressive expansion of areas with an elevated probability for WNV infections will increase the threat to the safety of blood transfusion. In geographically larger areas affected by WNV, a higher number of blood donors will be exposed to infection for a longer time period (if the duration of the annual mosquito activity season will be prolonged). Our climate change projections of the predicted probability of WNV infection in Europe have far reaching implications for public health in the future, because the findings in this paper can contribute to WNV preparedness activities. Besides the cases of primary WNV infections, secondary infections through contaminated blood products are of increasing concern to threaten the safety of the blood supply [33]. The asymptomatic blood-borne phase of a WNV infection increases the potential for transmission by transfusion, even if it is relatively short, compared to Hepatitis B virus or HIV [47]. Moreover, WNV has the ability to survive and persist in collected blood and stored blood components and subsequently cause an infection through the intravenous application. Thus, we calculated the prevalence of infection in the donor populations in the outbreak-affected areas (Figure 7). However, this instantaneous estimate may underestimate the true prevalence of infection if the timing of the WNV epidemic is at the peak of the epidemic curve. Nevertheless, the map reveals considerable vulnerabilities in Southeastern Europe when it comes to the safety of the blood supply. These insights can help transfusion service and clinical staff identify, manage and plan for transfusiontransmitted infections. The overall management of blood safety should be addressed at the institutional level, specifically at regulatory agencies or professional organizations [47]. To preserve the number of eligible donors and an adequate blood supply, authorities will inevitably reassess risk reduction interventions [33]. These include deferral strategies [48,49], screening strategies and triggers [50][51][52], and also pathogen reduction technologies [33,[53][54][55]. Moreover, it might be necessary to distribute blood components to outbreak areas from unaffected areas in order to prevent intermittent shortfalls in the blood supply [56]. Such coordination requires supra-national inventories of blood products that can be dispatched upon demand. Therefore, the projected temperature change with elevated probability for WNV infections and possible increased prevalence of WNV infected blood donors should be taken into account in developing preparedness plans for the WNV safety of the blood supply. This could substantially increase the costs of blood transfusion therapy.
The projections presented here may therefore offer an insight into future developments in the risk of transfusion from WNV infections; it also provides an opportunity to timely define the optimal strategy of blood safety in the face of limited available resources.

Conclusion
In several countries of Southeastern Europe, WNV transmission is now established. Our predictive model suggests further WNV dispersal in the coming years to adjacent districts.
Monitoring and modelling climatic and environmental conditions permissive for the interaction of migratory birds, resident birds, competent mosquito vectors and humans can help delineate districts at risk of transmission. Areas at risk for current and future WNV transmission can be targeted for integrated surveillance, vector control measures, outreach to the public and health care sector, strengthened laboratory capacity for reliable WNV diagnosis, and systematic screening of blood donors [57]. These activities call for intersectorial collaboration to tackle the challenges of WNV transmission.    Note: A probable case is any person meeting the clinical criteria AND with at least one of the following two: -an epidemiological link; -a laboratory test for a probable case. A confirmed case is any person meeting laboratory criteria for case confirmation.