Once duplicates were removed, a total of 1907 titles were obtained using the initial search terms. Following screening of titles, results were limited to 457 articles. After screening abstracts for relevance, 79 full-text articles were read full text, of which 57 were excluded. Two articles were included after checking the reference lists of the already selected articles. In total, 24 analytical research articles, in which the association between extreme precipitation or air temperature and waterborne infections had been assessed, were included in the literature review (Figure 1).
Studies of drinking water-related waterborne infections, geographical location and data sources
Articles using outbreaks as study units (n = 4)
Four studies used drinking water related waterborne outbreaks as study units [11-14]. Two articles presented studies that were performed using data from North America (Canada and United States) [11,14] while one used data from Europe (England and Wales) [13]. One study included data from several continents [12]. There were different data sources used to obtain outbreak data, including surveillance data, publicly available databases, previous published compilations and unpublished reports. The four studies assessed the association between outbreaks and precipitation. Two of them also studied the relationship with temperature. Meteorological data under study were obtained from records available at international organizations or from readings from the relevant weather stations.
Articles using cases of infection as study units (n = 20)
The remaining 20 articles used cases of infection as study units [15-34]. Most of the articles (n = 7) were performed in Asia (Bangladesh, Indonesia, Vietnam, India, Taiwan and China) [23,24,27,29-31,33]. Four were performed in North America (United States and Canada) [16,21,25,26], four in Oceania (Australia, New Zealand and Pacific Islands) [17,20,28,34], two in Europe (Sweden; and England and Wales) [18,22], one in central America (Haiti) [15], and one in Africa (Lusaka) [19]. One article used data from more than one continent, Asia and Central America [32].
The most common approach was to use cases of gastrointestinal infections without specifying the type of microorganism (n = 6). Among those studies focusing on specific microorganisms, cholera was most frequently studied (n = 6), followed by campylobacteriosis (n = 3) and typhoid fever (n = 3). Other infections, such as shigellosis, cryptosporidiosis, giardiasis, hepatitis A and paratyphoid fever, were also studied.
Cases of infection were obtained from several sources, including surveillance data, clinical records and registries, governmental reports and nurse advice telephone lines. All studies assessed the association between cases of infection and precipitation, while eleven of them also examined the relationship with temperature. The meteorological data under study were obtained from records available at international organizations, satellite sensors, gauge estimates, interviews or from local weather stations.
Definition extreme precipitation or temperature, covariates and statistical analysis
The definition of extreme weather events varied across the studies. There were different ways of categorizing meteorological variables, according to the amount or range of precipitation (i.e. groups including different categories; accumulated; smoothed using a certain number of days moving average; dichotomous, above and below a threshold; total in a given period; exceeded the upper limit of a given reference range). Only seven articles presented analyses stratified by water source or type of water supply, aiming to disentangle differences in the association with the occurrence of waterborne infections.
Analysis using Poisson regression or other types of count model regression was the most commonly adopted method to investigate whether variation in disease occurrence could be partly explained by changes in variables related to extreme weather events. Count model regression was used in eleven studies, one with outbreaks [12] and ten with cases of infections [15-17,20,22,25,27-30]. In some cases, the Poisson regression model was adjusted to account for: a) overdispersion, either by estimating an additional dispersion parameter using quasi-Poisson regression models [15,30] or more formally by using negative binomial regression models [28], b) excess zero counts in the observations, by using Zero-inflated Poisson regression models [12,16]. Time series data are prone to be influenced by seasonal and long-term variations, which may mask the short-term association between disease and extreme weather events. Seasonal trend decomposition was conducted in different ways, such as by adding trend and seasonal components into the Poisson regression [17], or by using Fourier terms [20,25,27]. In some studies, temporal correlations were handled by using generalized additive models (GAM) with time and sometimes other variables related to weather were added as smoother variables [16,29]. Delayed effects and a time varying relationship between the exposure and outcome variables were considered using generalized additive mixed models (GAMM) [29] or nonlinear distributed lag functions [15,22]. Case-crossover analysis was most frequently used when the study units were outbreaks [11,13]. It was also used in two studies using cases of infections [15,25]. In this analysis, the weather exposure at the location of an outbreak was compared with the exposures at the same location and same time of the year during control periods without an outbreak through use of conditional logistic regression. The method controls for time-invariant seasonal and geographic differences by design, although it assumes that neither exposure nor confounders change in a systematic way over the course of the study.
Findings of the studies
All four publications studying outbreaks found an association between precipitation and waterborne disease. Three found a positive association with extremes of precipitation [11,13,14], and one found an inverse association between waterborne outbreaks and average precipitation [12]. Among the two studies that assessed the association with temperature, one found a significant positive association [11]. Of the twenty articles using cases of waterborne infection as study units, amount of precipitation was found to have a positive association with infection in nine of them [15,16,19,22,24,26,28,29,32]. Two studies found a positive association in both extremes of precipitation (low and high) [20,27] and six did not find an association [17,18,21,25,31,34]. In three studies, statistically significant results were heterogeneous depending on the diseases or geographical regions they were assessing [23,30,33]. Regarding temperature, seven studies found a direct association between infections and temperature [17,18,20,24,25,27,32] and four did not find an statistical association [16,31,33,34]. In one study, statistically results depended on the disease that was being studied [28].