Heavy weather events, water quality and gastroenteritis in Norway

Climate change will lead to more extreme weather events in Europe. In Norway, little is known about how this will affect drinking water quality and population's health due to waterborne diseases. The aim of our work was to generate new knowledge on the effect of extreme weather conditions and climate change on drinking water and waterborne disease. In this respect we studied the relationship between temperature, precipitation and runoff events, raw and treated water quality, and gastroenteritis consultations in Norway in 2006–2014 to anticipate the risk with changing climate conditions. The main findings are positive associations between extreme weather events and raw water quality, but only few with treated drinking water. Increase in maximum temperature was associated with an increase in risk of disease among all ages and 15–64 years olds for the whole year. Heavy rain and high runoff were associated with a decrease in risk of gastroenteritis for different age groups and time periods throughout the year. No evidence was found that increase in precipitation and runoff trigger increased gastroenteritis outbreaks. Large waterworks in Norway currently seem to manage extreme weather events in preventing waterborne disease. However, with more extreme weather in the future, this may change. Therefore, modelling future climate scenarios is necessary to assess the need for improved water treatment capacity in a future climate.


Introduction
Climate change is predicted to lead to more frequent extreme weather events in the European region [1]. In the Nordic region the detected temperature increase is larger than in central and southern Europe, and this is expected to continue. This region has also experienced an increase in annual average precipitation that is expected to continue as opposed to central and southern Europe, where no change or even a decrease can be expected. The rate of change towards the end of the century is highly dependent on the choice of emission scenario. With the high emission scenario (RCP8.5) the annual temperature in Norway may increase by 4.5 • C and the annual precipitation may rise by 18% by the end of this century. Scenarios that are more moderate show less response with for example a median temperature increase of 2.7 • C for RCP4.5. Heavy rainfall and subsequent pluvial floods will be more intense and frequent also in Norway, but again the degree of change will depend on the rate of climate gas emissions [2].
Evidence is accumulating on the links between extreme weather events and drinking water quality [3][4][5]. Studies have shown that concentrations of coliform bacteria, E. coli, enterococci, and enteroviruses increase in environmental samples following rainfall. During extreme rainfall and runoff events, other water quality indicators, such as the turbidity, pH, and nitrate values are also influenced. Heavy rainfalls can rapidly elevate the total organic carbon and particles in surface water affecting the transport of microorganisms [3][4][5]and increasing the risk and burden of waterborne disease. Surface water is particularly vulnerable to changes in temperature, heavy rainfall and floods. In general, Nordic countries get most of their drinking water from groundwater, but as 90% of Norwegian drinking water is obtained from surface water [5,6], the water supply could be particularly impacted in case of increased extreme precipitation and floods.
In Norway, little is known about how extreme weather events affect drinking water quality and health. Our aim was to enhance the current understanding of the association between extreme weather events, water quality and waterborne disease by using epidemiological, hydrological, and meteorological data. And, we have studied to what extent extreme weather events during the past years have influenced the quality of raw and treated drinking water, and the subsequent risk of waterborne disease.

Water quality data
We used the National waterworks registry (VREG) to select the 26 largest waterworks in Norway for the study (with a total of 37 water treatment plants): including 23 surface water and three groundwater works (Fig. A). VREG include all Norwegian waterworks supplying more than 50 persons. We choose the biggest waterworks, serving approx. 60% of the Norwegian population, because the smaller waterworks did not have sufficient data to be included in the statistical analysis for associations with climate parameters. Water treatment at the included waterworks, except for the three groundwater works, include at least coagulation, UV disinfection and chlorination. The groundwater works had protected aquifers and disinfection (in stand-by). Water quality data for raw water (source water) and treated water (clean/distributed) on the following parameters were collected: E. coli, intestinal enterococci, coliform bacteria, heterotrophic plate counts, turbidity, colour, pH and conductivity, for the period 2006-2014 (only available data when implementing the study). The indicator bacteria: E. coli, intestinal enterococci, coliform bacteria, and heterotrophic plate counts, are standard part of water quality surveillance in Norway following the Drinking Water Directive 98/83/EC [7].
We requested accredited water quality data from routine monitoring scheme, data from internal control and data from on-line raw and treated water monitoring from the selected waterworks.

Gastroenteritis consultations
We extracted data on gastroenteritis consultations in Norway during the period 2006-2014 from the Norwegian Syndromic Surveillance System (NorSySS) [8] operated by the Norwegian Institute of Public Health (NIPH). NorSySS monitors how many infectious diseases are reported during consultations with general practitioners (GPs) and outof-hours primary care facilitiesNorSySS contains data about the patient's age group, gender and the municipality of residence, the municipality where the consultation was performed, date and diagnosis code in the International Classification of Primary Care system (ICPC-2). Gastrointestinal infections are a collective term for the diagnosis codes (ICPC-2) D11-Diarrhoea, D70-Gastrointestinal infection and D73-Gastroenteritis presumed infection. The Fig. B shows somewhat increased number of consultations with these diagnostic codes since 2006, but remaining stable over recent years, with over 200,000 consultations per year. [8]. NorSySS does not provide the exact number of infected people since some people will contact their GP several times with the same diagnosis and others may not consult their GP at all. NorSySS uses Quasi-Poisson Regression and the Farrington method to estimate the expected number of cases during a non-outbreak period. We then applied this algorithm to each municipality using weekly data and defined an outbreak as a week when the observed value was higher than two standard deviations from the expected non-outbreak baseline.

Meteorological and hydrological data
The Norwegian Meteorological Institute (MET) and the Norwegian Water Resources and Energy Directorate (NVE) provided gridded data for runoff (m 3 /1000m 2 ), accumulated daily precipitation, daily mean temperature, and observed daily maximum temperature. Precipitation at daily mean air temperatures above 0.5 • C was defined as rain. Precipitation is measured at around 400 locations in Norway, with slight technique variations. At older stations, meteorologists measure manually, with every monitoring representing the past 24 h. At newer stations, measurements are automatic and conducted every hour based on accumulation in a storage. All meteorological and hydrological observations are freely available through the web portal seklima.met.no and sildre.nve.no. To produce the gridded maps (1 km horizontal resolution), accumulated daily precipitation and daily mean temperature observations from all available measurement stations were used [9]. Gridded runoff was estimated using a hydrological rainfall-runoff model with the gridded temperature and precipitation as forcing data [10,11]. The gridded datasets of runoff, daily precipitation and daily mean temperature data are available at Senorge [12].

Data analysis
We conducted three analyses on the relationship between precipitation, runoff, water quality and gastroenteritis.

Weather events (exposure) and raw/treated water quality (outcome)
Each of the meteorological and hydrological variables (exposures) were collapsed to the average of the four weeks preceding the date of the outcome variable. Water quality variables (outcomes) were collapsed to the weekly average. The observed maximum temperature was taken from the nearest meteorological station to the water source intake. For the gridded precipitation and daily mean temperature, we used the grid cell covering the intake point of the waterworks, while the gridded runoff data were calculated for the catchment area of the waterworks. The water quality data variables were subsequently merged with the meteorological and hydrological variables, by time-point and geographical location. All outcomes were transformed using log (val-ue+1) to reduce skewness [13]. Online continuous water monitoring data had to be excluded due to unreliable sensors. For each exposure and outcome combination, we ran mixed effects linear regression models with random intercepts [14] for water work observation point and fixed effects for: month (dummy variables), internal vs external data source, and water source (surface water vs ground water vs river). We performed these procedures on all data, and then stratified by season (winter: Dec-Feb, spring: Mar-May, summer: Jun-Aug, autumn: Sep-Nov). Interaction models were used to identify if the exposure coefficients differed significantly between seasons.
We analysed effects on treated water (outcome variable) for colour and turbidity (exposure). Since the parameters, coliform bacteria, E. coli and intestinal enterococci mostly were reported as zero in the water quality routine monitoring data set, we were unable to run valid regression models. Bonferroni correction was applied to account for multiple testing [15].

Raw/treated water quality (exposure) and gastroenteritis consultations (outcome)
Water quality variables (exposures) were collapsed to the average of the preceding four weeks for each municipality. The municipality averaged water quality variables were merged with the weekly outbreak (outcome) data variables, by time-point and geographical location. For each exposure, outcome, and age combination, we ran mixed effects linear regression models obtaining percentage point increase likelihood of an outbreak per unit exposure increase with random intercepts for waterworks observation point and fixed effects for month (dummy variables). We performed these procedures in all data, and then stratified by season (winter: Dec-Feb, spring: Mar-May, summer: Jun-Aug, autumn: Sep-Nov) and age (all ages, 0-4, 5-14, 15-64, 65+). We used Interaction models [16] to identify if the exposure coefficients differed significantly between seasons and/or ages. Bonferroni correction was applied to account for multiple testing [15].

Weather (exposure) and gastroenteritis consultations (outcome)
For each combination of these exposure variables: runoff from municipal average using gridded data, rain and temperature from weather station located nearest to municipal centre; we transformed their daily values into extreme vs not-extreme using the 95th percentile (municipal specific). Then we summed these variables over a rolling four weeks period, generating variables representing the number of extreme days in a 28 days period. We subsequently analysed the exposure variables as continuous (number of extreme days in 28 days period). The municipality exposure variables were subsequently merged with the outbreak (outcome) data variables, by time-point and geographical location.
For each exposure, outcome, and age combination, we ran mixed effects linear regression models obtaining percentage point increase likelihood of an outbreak per unit exposure increase with random intercepts [14] for waterworks observation point and fixed effects for month (dummy variables). We performed these procedures in all data, and then stratified by season (winter: Dec-Feb, spring: Mar-May, summer: Jun-Aug, autumn: Sep-Nov) and age category (0-4, 5-14, 15-64, 65+). Interaction models [16] were used to identify if the exposure coefficients differed significantly between seasons and ages. Bonferroni correction [15] was applied to account for multiple testing.

Weather and raw/treated water quality
In raw water, the interaction of the exposure with season was statistically significant for all exposures and all outcomes except for the combination of colour and temperature (Fig. C). In treated water, the interaction of the exposure with season was statistically significant for all exposures and all outcomes (Fig. C).
In raw water, increased rain and runoff were associated with increased levels of coliform bacteria, colour, E. coli, intestinal enterococci, and turbidity throughout the entire year with p-value <0.001 (Table A.1). When stratifying by season, results for rain and runoff varied, but generally acted in unison (Fig. C). In raw water, increased maximum temperature was not significantly associated with increased or decreased levels of any of the outcomes throughout the entire year, however, was associated with an increase in coliform bacteria, E. coli, intestinal enterococci, and turbidity in winter, and turbidity in spring (Fig. C). The main water sources in Norway are lakes, and in the last years, the period with ice-cover has decreased, probably due to changing climate. The winters are milder and are more often associated with heavy rainfall and air temperatures above zero o C. The ice cover is normally a natural protection against microbes and particles flushing into the water source and prevents from an increase in microbial content and turbidity. In treated water, increased rain and runoff were associated with increased colour throughout the entire year. Increased runoff was associated with increased colour in summer and decreased colour in winter. No other associations were found (Fig. C

Raw/treated water quality and gastroenteritis consultations ("outbreaks")
An association was found between colour in treated water and increased risk of outbreaks among 0-4-year-old children with p-value <0.001 (Table B.1) in spring. No other significant associations between raw (Table B.2) or treated water (Table B.1) quality and outbreaks were found.

Weather and gastroenteritis consultations
Correlations between all ages and exposure were statistically significant (Fig. D, Table C). Interactions between seasons and exposure were statistically significant for all exposures and outcomes except for temperature that was a priori not tested, because extreme temperature events occurred primarily in summer and never in winter, which caused issues with model-fitting (Fig. D). We found that an increase in maximum temperature was associated with an increase in risk of outbreaks among all ages and 15-64 years olds for the whole year (Fig. D). We also found that an increase in rain and runoff were associated with a decrease in risk of outbreaks for different age groups and time periods throughout the year (Fig. D).

Discussion
Our results reveal a positive correlation between heavy rainfall and high runoff, and raw water quality parameters. There is large evidence in the literature about how raw water parameters get influenced by extreme weather events [4,17] which correlates well with our results. However, we did not find positive associations between increased rainfall and runoff and increase in gastroenteritis consultations (outbreaks). This is in line with our findings demonstrating no clear association between increased rainfall, runoff and treated drinking water. This suggests that the larger water works have treatment capacity that can manage changes in raw water quality due to heavy rainfall and runoff in today's climate. Interestingly, increased rain and runoff were associated with fewer consultations for gastroenteritis, while higher temperatures were associated with a higher number of more consultations. This may be linked to known risk-factors associated with good weather, such as barbecuing [18,19] and swimming in recreational water [20,21]. Regarding the small water bodies in Norway, they have been less subject to inspection and requirements for robust drinking water treatment compared to the large water works comprising of water bodies vulnerable to contamination from activities such as agriculture and surface run-off. In general, there is a requirement to protect the water sources from contamination from farms, however incidents with overflow especially with extreme weather events may be a risk. Waterborne outbreaks occur yearly, mainly caused by Campylobacter and norovirus where the source have been suspected to be birds or human faecal contamination. More rare cases have been linked to EHEC or Cryptosporidium. In case of outbreaks, a close collaboration between public health and food safety authorities is initiated for outbreak management. Recent updates in the Norwegian Drinking Water Regulation have strengthen the reporting of small-scale drinking water supply systems to the Norwegian Food Safety Authority for closer surveillance [22].
The relationship between extreme weather events and the subsequent risk of gastroenteritis due to contaminated drinking water is a complex issue [23]. In 2015, we published a review where we included analytical research studies analysing associations between extreme precipitation or temperature and waterborne disease [24]. Studiesfi ndings were heterogeneous. While most of them identified a positive association between increased precipitation, temperature and infection, others did not. Geographical region, season or water supply profile could play a role in this heterogeneity. Taking these findings into account, we have stratified our analysis by municipality, age, and season. However, we have not found any substantial differences suggesting, again, that effective water treatment procedures could be the reason for this.
Methodological issues, such as the definition of the outcome "waterborne disease", could also influence research results looking into extreme weather-waterborne disease relationship. For instance, among the articles included in the above mentioned review, different types of study units were used to define "waterborne disease": waterborne outbreaks [25,26], specific waterborne infections trends such as campylobacter infections or cholera [27], or number of gastroenteritis consultations in a health care facility [28]. All publications studying waterborne outbreaks (n = 4) found an association between precipitation and waterborne disease, while findings in those using single cases of infection or consultations were more heterogeneous (n = 20),. In 2016, a previous collaboration between NIPH, MET and NVE together with national public health institutes from three additional Nordic countries and the European Centre for Disease Prevention and Control examined the association between heavy precipitation events and waterborne outbreaks between 1992 and 2012 using waterborne outbreaks as study units [29]. We did find a positive association between increased precipitation during the preceding week and the occurrence of an outbreak, specifically involving single household supplies. Outbreaks associated with waterworks did not present a significant correlation with heavy precipitation events, which is in line with the results of our current study. The conclusions from the previous project were somehow limited as underreporting is an inherent problem in surveillance systems and notified outbreaks are the tip of the iceberg of the real burden of waterborne disease. Even with the use of more sensitive registry based on syndromic surveillance, NorSySS, we found no evidence for an association between extreme weather events and gastroenteritis consultations. A potential limitation of this study, could be that gastroenteritis consultations included in NorSySS are an aggregated large group of consultation from which we cannot disentangle those that are waterborne, which may dilute the potential relationship we are assessing.
For the first analysis in this study, each of the meteorological, hydrological, and water quality variables were collapsed to averages for each location, before the associations between weather, hydrology and water quality indicators were estimated. As contaminations of the drinking water are expected to occur after heavy rainfall events, using "number of days in a week where the meteorological and hydrological variable was above the 95th percentile" was considered. However, weekly averages provided very similar results. The associations found between water quality and weekly average rainfall therefore also indicate associations between water quality and a high frequency of days with extreme precipitation.
The spatial pattern of rainfall is highly variable due to complex relationships between climate regimes, seasonality, and topography. To capture this variability, it demands a high-resolution network of measurement stations. Levy et al. [30] found that spatial incompatibility between exposed populations and rain gauges in their study resulted in a reduction of about 50% in the association between extreme rainfall and diarrheal disease incidence. Although the observational network in Norway used in the gridded datasets consists of several hundred weather stations and gives a good representation of the regional climate, there is uncertainty in the estimates for the exposure location. As the gridded climate data is used as input to a conceptual rainfall-runoff model to estimate runoff, this uncertainty is also embedded in the runoff data. In our study, we used daily accumulated precipitation and daily mean temperature as a basis for the calculations. Higher time resolutions would have better captured extreme rainfall events that occur over a short duration, but only a limited number of such datasets exist. However, as we aggregate the data to a weekly time resolution, this is not believed to have a major impact on the results. The lack of exact weather representations at the exposure locations, in combination with limited access to data about the exact locations of the intake points for the waterworks, due to security reasons, may have led to a decrease in potential associations between heavy rainfall events and drinking water quality.

Conclusions
Our results illustrate that larger water works in Norway seem to cope with those extreme weather events that we have experienced so far. However, we have not been able to assess small waterworks and single households in this study as they do not collect enough data on raw or treated water quality to be able to do an analysis of the association.
Aging drinking water treatment, distribution systems and sewage systems will be particularly vulnerable to flooding, leading to potential deterioration in the quality of drinking water. Therefore, we also perform work to assess how future climate scenarios will challenge raw water quality and the water treatment capacity in a follow up study that is a part of the same research project. Using the data from this study and future climate scenarios combined with modelling of treatment effect and QMRA [31] to assess probability of disease is needed to address and prepare for a changing climate with more extreme weather events and potential need for increased treatment capacity. This will be central to minimize the risk and burden of waterborne disease in the future.
Based on our results, water safety planners, in areas using surface water as a source for drinking water production, should consider including coagulation and UV radiation as treatment, as it appears as robust for changes in water quality due to extreme weather events. However, treated water also may be contaminated during distribution to the consumers. Therefore, effects of extreme weather events associated with drinking water distribution systems and waterborne disease needs to be further explored. The One Health link in this study could be a marker for future studies, in the facilitation of human health, veterinary and environmental aspects to shed light of an emerging health issue concerning climate change, water and health.

Funding
This work was supported by the Norwegian Research Council [No. 244147/E10].

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.