“Sickenin’ in the rain” – increased risk of gastrointestinal and respiratory infections after urban pluvial flooding in a population-based cross-sectional study in the Netherlands

To collect data on self-reported AGE and ARI after urban pluvial flooding, a retrospective cross-sectional survey was conducted during the summer of 2015 using the methodology described in De Man, et al. [3]. In total, 3382 households that were surrounded by floodwater after extreme rainfall events (extreme rainfall events are rainfall events for which streets are flooded) in 60 locations in the Netherlands received a self-administered questionnaire by regular post. The questionnaire was send 2 weeks after the start of the flooding event and participants were asked to answer questions considering a recall period of 2 weeks, i.e. from the start of the flooding until reception of the questionnaire. Flood locations were identified by monitoring possible flood events on the internet, such as social media and press releases, which generally receive much media attention in the Netherlands. The national website of the Dutch fire brigade was the main source of information to determine which households were surrounded by floodwater (http://​www.​112meldingen.​nl/​). This website provides the causes of a given alert, which the Dutch fire brigade receives from the citizens themselves, corroborates on-site, and disseminates to the public, specifying the affected areas (postal codes). With this information, the addresses of the affected areas to which the questionnaires were sent were derived from publicly available mapping tools like googlemaps.​com [3].

The questionnaire was developed by adapting the questionnaire used in the study of De Man, et al. [3] to collect epidemiologically relevant information for each household member (individual participant). Information was collected on basic demographic characteristics (i.e. age, sex, number of household members). Furthermore, information was gathered about the occurrence of gastrointestinal and respiratory complaints during the 2 weeks before questionnaire completion, underlying chronic diseases, type of exposure (i.e. contact) to floodwater, and type of activity leading to contact with floodwater, duration of flooding in minutes and the magnitude of exposure (height of floodwater in cm). Participants reporting one or more types of contact or types of activities leading to contact with floodwater are hereafter referred to as being exposed to floodwater.

Questions could be answered for up to five household members (most Dutch households are composed of ≤ 4 people according to Netherlands Statistics, https://​www.​cbs.​nl/​, [7]), and the participant was asked to report information on all household members.

This study involved collection and analysis of fully anonymized data, so no ethical approval was necessary according to Dutch regulations. [8, 9] People gave consent upon anonymously completing and returning the questionnaire. Questionnaires were received in de-identified form, containing only data on postal code, sex, and age of the participants. Therefore, names and addresses could not be linked to the questionnaire responses, guaranteeing anonymity of the respondents. Participants were informed that the data they provided were to be analyzed for scientific purposes, and that by returning the questionnaires, they gave consent to do so. No written consent was therefore necessary. Parents were asked to complete the questionnaire and give consent on behalf of their children (family members < 16 years old).

Based on the symptoms reported in the questionnaire, participants were defined as cases of AGE or ARI when experiencing the following complaints in the 2 weeks before completing the questionnaire, and as non-cases otherwise:

Generalized estimating equations (GEEs) were used to compare AGE and ARI in cases with the non-cases, with regard to their type of contact and type of activity leading to contact with floodwater. As our study design led to clustering of data at the household-level, we corrected for dependence of observations deriving from individuals living in the same household [12]. An exchangeable working correlation structure was chosen for the GEE models, meaning that the correlation between individuals within a cluster was the same for all clusters. An overall model was built for all age categories (categorical: 0-5, 6-15, 16-25, 26-45, 46-65 and > 65 years), as well as separate models for adults (categorical: 16-25, 26-45, 46-65 and > 65 years) and for children (< 16 years, continuous), which were further subdivided in a model for type of exposure and a model for type of activity leading to exposure.

A manual backward-forward selection procedure was used to retain only those variables with p ≤ 0.05. Collinear variables were selected based on improved model fit as revealed by the Quasi-likelihood under the Independence model Criterion (QIC) statistic [13]. Biologically plausible interactions between independent, correlated variables were also assessed. Interactions were first tested in the univariate analyses (only interaction terms were included, confounders and other possible risk factors were added later in the multivariate analyses), and those with a p-value ≤ 0.20 were included in the multivariable model. We present the adjusted odds ratios (aOR) and 95% confidence intervals from the final multivariable models. For results from the univariate analyses, see Additional file 1: Table S1 and Additional file 2: Table S2.

We used R (version 3.4.3) [14] and the gee package (version 4.8) [15] for data analyses.

In total, 3382 households were invited and 699 households (response 21%; 1,656 individual participants) completed the questionnaire after three extreme rainfall events caused urban pluvial flooding on respectively July 5th, July 27th and August 15th, 2015, in 60 municipalities in the Netherlands (Fig. 1). On average, the participants returned the questionnaire 9 days after sending (median = 6 days).

Twenty-one individual participants (1.3%) were excluded, because information reported on exposure was contradictory. Information about complaints and exposure was incomplete for 477 individual participants (28.8%). Therefore, they were also eliminated from the analysis (see Additional file 3: Figure S1). The remaining 1,158 individual participants (582 households) represented the study population of which we had information for both cases and non-cases (Fig. 2). There were no pregnant participants, who also reported vomiting during the recall period (see Additional file 3: Figure S1). Table 1 gives an overview of the descriptive statistics. 277 participants (24%) have an underlying chronic disease, 690 participants (60%) were exposed to floodwater (i.e. contact) and 966 participants (83%) performed a type of activity leading to contact with floodwater.

In the study population, 40% were women, 37% were male, and of 23% of the participants the sex was unknown. The median age was 47 years (25-75% percentile P_25-75 27 – 61). Compared to the Dutch general population in 2015 (16,900,726 inhabitants [16]), the sample contained slightly more women (52% vs. 51%) and was slightly older (average age 43 vs. 41 years). Most households (281;42%) reported two individual participants and 28 households (4%) reported five individual participants (Fig. 3).

During the 2-week period after flooding, 75 (12.3%) participants in 51 households reported AGE, and 128 (21.1%) participants in 114 households reported ARI (Table 2), after being exposed. Table 2 also shows the results of the univariate and multivariate analyses. In the multivariate analyses, exposure to floodwater was significantly associated with increased odds of AGE (aOR 4.2, 95% CI 2.1 – 8.4). Furthermore, exposure to floodwater was associated with increased odds of reporting ARI (aOR 3.3, 95% CI 2.0 – 5.4). The analyses on the association between floodwater exposure and AGE in children < 16 years could not be performed given the small number of outcome events. Indeed, there were no children without contact with floodwater that also had AGE. Furthermore, Table 2 shows that there was no significant association between floodwater exposure and ARI in children (OR 1.6, 95% CI 0.6 – 4.0) in the univariate analysis. In adults, floodwater exposure was associated with both AGE and ARI (aOR 3.6, 95% CI 1.8 – 7.3, and aOR 4.0, 95% CI 2.3 – 6.8, respectively).

Results from the final multivariable models for AGE and ARI are presented in Tables 3 and 4 respectively. For results from the univariate models, see Additional file 1: Table S1 and Additional file 2: Table S2.

In the models for type of activity and AGE or ARI, an interaction term was included between contact with floodwater and cleaning floodwater inside or outside the house. The strata included in this interaction term are: no contact with floodwater at all (reference category), contact with floodwater but no involvement in post-flooding cleaning operations, contact with floodwater and involvement in indoor cleaning operations, contact with floodwater and involvement in outdoor cleaning operations, and contact with floodwater and involvement in both indoor and outdoor cleaning operations. The rationale is that people with the least exposure to floodwater have the lowest chance to get AGE or ARI (water contact without cleaning), while people with the largest exposure to floodwater (water contact with cleaning inside and outside the house) have the highest chance to get AGE or ARI.

The average height of the water (magnitude of exposure) in the streets was 19 cm (median = 15 cm, height _min = 2 cm, height _max = 150 cm). The average height of the water after toilet overflow was 5 cm (median = 2 cm, height _min = 1 cm, height _max = 30 cm). Additional univariate analyses showed that the magnitude of exposure in the streets was a risk factor for both AGE (OR 1.0; 95% CI 1.0 – 1.1) and ARI (toilet overflow, OR 1.1; 95% CI 1.0 – 1.2).