Environmental Suitability of Vibrio Infections in a Warming Climate: An Early Warning System

The ECDC Vibrio Map Viewer (https://e3geoportal.ecdc.europa.eu/SitePages/Vibrio%20Map%20Viewer.aspx) displays coastal waters with environmental conditions that are suitable for Vibrio spp. growth internationally (Figure 1). It is based on a real-time model that uses daily updated remotely sensed SST and sea surface salinity (SSS) of coastal waters (see below) as inputs to map areas of high suitability for Vibrio spp. that are pathogenic to humans (Copernicus Marine Environment Monitoring Service 2016; NOAA 2016). SST and SSS are two key environmental factors that influence the number of infections, based on a model developed by Baker-Austin et al. (2012). For the Baltic Sea, SSS demarcates the regions suitable for Vibrio infections (Copernicus Marine Environment Monitoring Service 2016) and SST serves as a risk predictor (NOAA 2016). Salinity in coastal waters is strongly modified by rainfall and, in turn, by river flow; the model uses a threshold of 26 practical salinity units (PSU) for SSS and $18°\mathrm{C}$ for SST. The nominal spatial resolution of the output is $5\mathrm{km}$. The daily suitability index ranges from zero to a maximum that is determined by the highest SST value. Thus, the output detects coastal areas with environmental conditions suitable for Vibrio species that can cause infections in humans. These fields, which are estimated on a daily basis by the National Oceanic and Atmospheric Administration’s (NOAA) Atlantic OceanWatch node at the Atlantic Oceanographic and Meteorological Laboratory (AOML) in Miami, Florida, are integrated within the ECDC Vibrio Map Viewer, which is the point of access in the Baltic region.

In the Baltic Sea, low-salinity areas delineate the areas suitable for the occurrence of Vibrio infections, whereas SST serves as a risk predictor (Baker-Austin et al. 2012); however, the influence of SST and SSS on the environmental suitability for Vibrio growth can be extrapolated to other regions of the world to obtain global risk estimates. The ECDC Vibrio Map Viewer was designed to delineate retrospective, current, and short-term forecasts of environmental suitability at a global scale, which requires obtaining reliable SST and SSS, especially in coastal regions where human exposure is more likely to occur (Figure 1). The global model data inputs are SST fields from remote sensing and models, as well as SSS from models, in situ data, and climatological data. The estimates for SST were obtained from a number of sources:

SSS were obtained from the Copernicus Marine Environment Monitoring Service (2016). For retrospective studies, NOAA’s Optimum Interpolation (OI) SST V2 data set provided satellite and model-interpolated daily analysis of SST in a consistent methodology back to September 1981. For the Swedish coastal counties, mean SST were spatially aggregated per county per week for the years of analysis (2006–2014) to generate time-series data sets for each coastal county.

Climate change projections of SST were derived from a Coupled Model Intercomparison Project Phase 5 (CMIP5) model ensemble (r1i1p1) for the Swedish coastline aggregated by county. Time series per month for each county from 2005 through 2100 were derived. Model output was obtained for emission scenarios RCP 4.5 and RCP 8.5, representing a possible range of radiative forcing values in the year 2100 relative to preindustrial values ($+4.5$, and $+8.5\mathrm{W}/\mathrm{m}2$, respectively).

Infections caused by Vibrio cholerae (other than serotypes O1 or O139 and Vibrio cholerae serotype O1 or O139, which are nontoxigenic) are notifiable according to the Swedish Communicable Diseases Act (Swedish Code of Statutes 2004) and include V. parahaemolyticus, V. vulnificus, and V. alginolyticus. Cases are reported to the mandatory notification system at the county medical office and at the Swedish Public Health Agency. We obtained a listing of all Vibrio infections from 2006 through 2014 with clinical and laboratory confirmation from the Swedish Public Health Agency (Folkhelsomyndigheten 2016). The listing included information on county, statistical date and onset of disease, type of infection, Vibrio species, serotype, transmission pathway, sex, and age group of each case. For reasons including consistency in reporting and data completeness, we used data for the period 2006 through 2014 for our analysis. A total of 117 cases were reported for the period from June 2006 through October 2014, of which 111 occurred in coastal counties with a possible link to SST. Thus, being in close proximity to the Baltic provides the opportunity for exposure to coastal water both for case and control times. However, 30 of these cases had no precise place of infection and 25 cases had no date of onset of disease, and these cases were not included in the analysis.

The variables of the 56 Vibrio cases for 2006–2014 were subjected to descriptive statistics and frequency analysis. Because changes in SST occur intermittently, have a short induction time and a transient effect (Vibriosis), a case-crossover study design was chosen to assess the association between SST and Vibrio infections. The SST exposure status (mean SST, spatially aggregated per county and per week) of the Vibrio infection at the time of the Vibriosis onset was compared with the distribution of the SST exposure status for that same Vibriosis case in earlier/later periods. This approach assumes that neither exposure nor confounders change over the study period in a systematic way. Thus, a time-stratified approach at the individual level was used for control days to contrast with the events. An advantage of using such a time-stratified case-crossover design is the automatic adjustment for individual non-time varying factors; these can risk introducing confounding bias in epidemiological studies if not adjusted for. Further, the time-stratified approach used control events before and after the event date for each individual Vibrio infection in the same area. We used 2, 4, and 6 wk as the time window between event data and the control days, both before and after the event. This adjusts for unknown temporal confounders and controls for seasonal influences not related to the seasonality of SST as the primary exposure variable.

The weekdays of the dates of the weekly county means of the SSTs were restricted to Mondays, but the date of infection was for any date. Thus, in order to match the date of infection with its corresponding SST, Tuesday to Thursday were referred to the preceding Monday, whereas Friday to Sunday were referred to the following Monday. For analysis, a data set with the event itself and control events 2, 4, and 6 wk before and after the event was created. A time series with SST county means from 1 to 8 wk before the event and the controls was added.

We used a conditional logistic regression model to ascertain a relationship between SST and Vibrio infection and to derive an exposure–response curve for the relationship between the odds ratio of Vibrio infection and SST. We refer to the odds ratio analogously to relative risk in this study due to the low probability of disease events. We studied the relationship between Vibrio infections and SST using natural cubic splines (4 degrees of freedom) and for different lead times of exposure up to 4 wk before disease occurrence. We identified a piecewise linear model with a knot of SST at $16°\mathrm{C}$ for the final model.

We used the computed case-crossover exposure–response relationship to project how the seasonal window of transmission would change in each of the counties. We used projections of SST data from a global circulation model from CMIP5 for each month in the time period from 2006 through 2099 for each county. Months with elevated risk were categorized as potential transmission months and aggregated as average per decades. The annual maximum elevated risk month was averaged to a change of transmission intensity per decade. Relative risk estimates are presented using the year 2016 as the baseline and describe changes due to SST from there onward.

We used also CMIP5 sea surface temperature projections for the RCP projections to illustrate differences in the projected SST between RCP 8.5 and RCP 4.5 for August 2050. We computed the surface area [in kilometers squared ${(\mathrm{km}}^2)$] of the Baltic Sea that is environmentally suitable for Vibrio growth for RCP 4.5 and RCP 8.5, from 2010 through 2060, by month.

The ECDC Vibrio Map Viewer displays environmental suitability for Vibrio infections based on SST and SSS (Copernicus Marine Environment Monitoring Service 2016; NOAA 2016). However, Vibrio ecology and growth also depend on a number of other variables including marine nutrient concentrations, river discharge, and algae blooms (Boer et al. 2013; Johnson et al. 2012; Julie et al. 2010). For example, long-distance atmospheric deposition and aerosols such as Saharan dust nutrients can promote Vibrio bloom formation in marine surface waters (Ansmann et al. 2003; Westrich et al. 2016). Moreover, individual Vibrio species display different responses in relation to SST and SSS (Boer et al. 2013; Johnson et al. 2012; Julie et al. 2010). Thus, the environmental suitability shown by the ECDC Vibrio Map Viewer represents an approximation of the actual suitability and local variation might apply. In addition, many Vibrio infections are influenced by other factors, such as immunity, travel, and gastrointestinal disease, in addition to coastal water exposure. Currently, the Swedish Public Health Agency recommends that people avoid swimming if they have a significant or open wound and the SST is $20°\mathrm{C}$ or higher.

Our analysis was based on Swedish data because Vibrio infections became reportable in Sweden in 2004. In many other Baltic countries, Vibrio infections are not reportable and therefore, little information is available to assess the risk in those countries. Regrettably, there was not a training data set and a testing data set to validate the exposure–response relationship of Vibrio infections in response to SST. However, our findings are consistent with the documented number and distribution of Vibrio infections clustered around the Baltic Sea area associated with the temporal and spatial peaks in SST (Baker-Austin et al. 2012).

Mortality and morbidity due to Vibrio infections continue to occur in the Baltic Sea area. Moreover, we show that the environmental suitability of Vibrio growth in the Baltic Sea will expand in a warming climate. However, in Europe, there is almost a complete lack of information regarding the persistence/abundance of Vibrio in the environment and the number of human cases. Reporting of Vibrio infections is not mandatory in the European Union, and many laboratories test only for Vibrio infections in patients with diarrhea when they are returning from a foreign holiday (to rule out Vibrio cholerae). The strength of this study lies in the fact that most of the infections were nongastrointestinal and therefore not subject to this selection bias. Thus, in the absence of mandatory notification data on Vibrio infections in Europe, the ECDC Vibrio Map Viewer can forecast the environmental suitability of coastal waters for Vibrio spp. using remotely sensed SST and SSS. These forecasts and potential alerts are currently disseminated by ECDC to public health decision makers, along with different response options for their consideration, through the CDTR: Public access to a beach should be temporarily denied for public safety purposes, warnings should be issued when the environmental suitability of Vibrio infections is imminent, or alerts should be issued to notify health care providers and at-risk individuals such as the immunocompromised. Through this cascade of steps—risk assessment, monitoring of environmental suitability and alert detection, dissemination and communication, and response—the ECDC Vibrio Map Viewer constitutes an important link in an early warning system for Vibrio infections.