Water recreation and illness severity

Over the past 60 years epidemiologic studies have evaluated risk due to water recreation on US surface waters by comparing rates of gastrointestinal (GI) illness occurrence among recreators as a function of water quality and water exposure (Stevenson 1953; Cabelli 1983; Dufour 1984; Wade et al. 2008). Escherichia coli (E. coli) and enterococci (‘indicator bacteria’) have been identified as predictors of the occurrence of GI illness among water recreators (Cabelli 1983; Dufour 1984). In those epidemiologic studies, GI illness was defined as the presence or absence of a set of symptoms, though the specific elements of that set varied among studies. Relationships between water quality and the risk of GI illness occurrence observed in many of these studies have been used to inform water quality criteria recommendations (US EPA 1986, 2012).

As a result of this binary classification system of illness, potentially important distinctions in disease severity are lost. A definition of GI illness used in recent epidemiology studies of recreational water exposure (Wade et al. 2006, 2008, 2010; Dorevitch et al. 2012) was diarrhea (three or more loose stools in 24 hours), or vomiting, or nausea with stomachache, or nausea that interferes with daily activities, or stomachache that interferes with daily activities. Consider three individuals, all of whom developed GI symptoms as a result of exposure to recreational waters. Case A experienced two loose stools over a 24-hour period, took over-the-counter (OTC) medication for her symptoms, and missed 1 day of work due to symptoms. Case B had three loose stools over a 24-hour period but did not require medical treatment, and did not miss work. Case C had several bouts of vomiting and loose stools per day for 3 days, became dehydrated and received intravenous fluids in an emergency department (ED). Even though cases A and B seem similar to one another, A would be considered to not have GI illness, while case B would. Furthermore, the substantial difference in symptom severity and resulting medical treatment between cases B and C would not be captured by the binary classification system.

Presently, no ‘gold standard’ definition exists for GI symptoms resulting from waterborne (or any other) exposures. Evaluating measures of symptom severity on an ordinal scale, rather than considering illness occurrence as a dichotomous definition, may lead to better understanding of risks, predictors, and consequences of recreational waterborne illness. We utilized data from two epidemiologic studies of water recreation to characterize the severity of symptoms among water recreators, highlight differences between a binary and an ordinal classification system, and evaluate the relationship between symptom severity and water quality or water exposure. Limited data support an association between a higher dose of infectious pathogens and greater symptom severity among human volunteers (Glynn et al. 1995). Thus, a range of fecal indicator concentrations in surface water might be associated with a continuum of symptom severity among water recreators.

The National Epidemiological and Environmental Assessment of Recreational Water (NEEAR) study evaluated health risks (including GI, respiratory, ear, eye, or skin symptoms) of swimming and wading at marine and freshwater beaches impacted by human fecal pollution in Alabama, Indiana, Michigan, Mississippi, Ohio, and Rhode Island. During 2003–2007, beachgoers with and without beach water contact were recruited. Recruiting, survey administration, and corresponding water sampling methods have been described previously (Wade et al. 2008, 2010).

The Chicago Health Environmental Exposure and Recreation Study (CHEERS) evaluated health risks of incidental-contact water recreation (boating, canoeing, fishing, kayaking, and rowing). During 2007–2009, participants were recruited from 39 locations in the Chicago, Illinois area, including Lake Michigan, small inland lakes, and the Chicago River system, which, at the time of this study, was predominantly treated wastewater effluent (Rijal et al. 2011). An unexposed group, people engaged in outdoor recreation that did not involve water, was recruited as well. Study protocols, questionnaires and the general design were based on NEEAR. Survey and water quality assessment methods for CHEERS have been described previously (Dorevitch et al. 2012, 2015).

Some studies have found that the strongest association between GI illness and water recreation was observed within 0–3 days following recreational activity (Dorevitch et al. 2012; Arnold et al. 2013). In the current analysis, acute gastrointestinal illness (AGI), was defined as illness, meeting the aforementioned case definition used in previous studies (Wade et al. 2006, 2008, 2010; Dorevitch et al. 2012), within 0–3 days of recreation. Severity was assessed among those with AGI, as well as those who had incident symptoms within 0–3 days of water recreation but did not meet the AGI case definition.

CHEERS and NEEAR study protocols were not identical. In NEEAR, one adult member per household provided information during a single telephone interview 10–12 days following the recreation event about all participating family members (Wade et al. 2006, 2008, 2010). In CHEERS, participants provided follow-up information about themselves, except in the case of children <8 years old whose information was provided by parents (children of 8–17 years were allowed to answer on their own, but were assisted by parents). Participants were interviewed via telephone on approximately days 2, 5, and 21 following the recreation event (Dorevitch et al. 2012). In CHEERS, follow-up telephone interviews captured information about the number of loose stools per individual in a 24-hour period, however, in NEEAR, the number of loose stools was not ascertained, only whether that number was at least three. Therefore, among CHEERS participants, reports of less than three loose stools were excluded from the definition of GI symptoms (NEEAR definition) to improve comparability of information between the studies. Additionally, all symptoms reported (including those with less than three loose stools) (CHEERS definition) are also included in the analysis, to compare the two definitions (NEEAR versus CHEERS definition).

Severity metrics, or scores, were based on information collected in NEEAR and CHEERS. Severity was assessed for all participants that developed GI symptoms within 0–3 days of water recreation, regardless of whether they met the case definition for AGI. Those that did not develop symptoms were assigned a severity score of zero. First, the duration of GI symptoms was assessed (Supplemental material, available with the online version of this paper). Since CHEERS had a longer follow-up, symptoms lasting longer than 10–12 days were truncated at 12 days, to make the results comparable to those of NEEAR. The metric ‘symptom-days (SDs)’ was created by summing the number of symptoms and the duration (in days) of each symptom. Two SDs metrics were developed: for only GI symptoms (diarrhea, vomiting, nausea, or stomachache), and for total symptoms (all GI, respiratory, ear, eye, or skin symptoms). It has been suggested that more attention should be placed on more severe symptoms, such as vomiting and diarrhea (Freedman et al. 2010). Therefore, in calculating SDs, we weighted the duration of vomiting and diarrhea by a factor of two; duration of nausea and stomachache were weighted by a factor of one. For example, a person can have a duration of GI symptoms equivalent to two, but have nausea for 2 days and vomiting for 1 day. Therefore, their GI SDs would be equivalent to four (2 days of nausea + (1 day of vomiting × 2)).

Severity was also assessed according to ‘responses to symptoms’. We assumed that symptom severity influences the decision to take medication, to visit a healthcare provider (HCP) or to stay home from work or school (Stratmann 1999; Scallan et al. 2006). During telephone follow-up for both NEEAR and CHEERS, participants were asked if they stayed home from work or school, took OTC medication, took prescription medication, contacted a HCP, or were admitted to an ED or hospital. Previous analyses of foodborne illness have considered illnesses that result in contact with a HCP to be more severe than illnesses that require no contact (Hoffmann et al. 2012). Therefore, participants who took prescription medication, contacted a HCP, lost time from work/school or daily activities, or went to an ED or hospital had a major response to symptoms, while all others with symptoms were considered to have a minor or no response.

Severity metrics were evaluated using χ² and the Kruskal Wallis test, where appropriate, to determine if any significant differences in severity existed among those with AGI compared to those with symptoms only.

Water exposure was evaluated by post-recreation interviews. NEEAR participants who were only wet below their waist were considered waders, while those wet above the waist were considered swimmers (Wade et al. 2008). The degree of water exposure among NEEAR water recreators was defined on an ordinal scale: (1) swimmers who swallowed water, (2) swimmers who did not swallow water, and (3) waders (reference group). For CHEERS water recreators, the three ordinal categories were: (1) those who swallowed water; (2) those who did not swallow water but got their face, torso, or hands wet; and (3) those who reported no water exposure (reference group). NEEAR participants who reported no water contact and CHEERS participants who did not participate in water recreation were excluded from all analyses, which focus on water quality and water exposure as a predictor of symptom severity, as they had no water exposure.

Exposure to increased levels of fecal contamination in recreational water was hypothesized to be related to symptom severity because it could increase the quantity of pathogens ingested by the recreators. Dose of pathogens has been demonstrated to be related to the occurrence of GI illness in an outbreak at an Oregon lake (Keene et al. 1994) and has been suggested to be directly related to severity of symptoms in outbreaks of salmonellosis (Taylor et al. 1984; Glynn & Bradley 1992; Mertens et al. 2013) and studies related to hepatitis E (Purcell & Emerson 2008).

Surface water was sampled during water recreation. Among NEEAR water recreators we assessed severity in relation to enterococci by both a culture-based method (US EPA Method 1600) and by quantitative polymerase chain reaction (qPCR) (comparable to US EPA Method 1611). The qPCR method measures enterococci DNA, regardless of whether bacteria are viable and culturable. Among CHEERS water recreators we assessed severity in relation to concentrations of viable and culturable enterococci (US EPA Method 1600) and F+ coliphage (US EPA Method 1602). Male-specific (F+) coliphages, viruses that infect E. coli with pili, are thought to have similar fate and transport as viral pathogens in surface water (Colford et al. 2007; Bosch 2010). Daily mean water quality measurements from both studies were utilized in this analysis. Due to differences in water quality in both studies, quartiles and dichotomization at the 75th percentiles were calculated separately.

We considered the duration of GI symptoms and SDs to be count outcomes, and assessed the relationship between these severity metrics and water exposure and fecal indicators using zero-inflated negative binomial (ZINB) regression, since many participants did not develop symptoms and were considered to have no severity. Zero-inflated models utilize two separate processes for fitting models: a logistic process to evaluate zeros versus non-zeros (symptoms vs. no symptoms), and a negative binomial process to evaluate the non-zero values (extent of severity among those with symptoms) (Hilbe 2011). Exponentiating beta values from negative binomial regression of event occurrence data yields the incidence rate ratio. Because incidence rate ratio could be misconstrued as the ratio of incidence rates, rather than the ratio of symptom severity, we refer to this as the duration ratio or the SDs ratio, since it represents the duration or number of SDs compared to the reference group. Responses to symptoms (use of medication, missing work, etc.) were assessed as a binary outcome, using logistic regression. Variables that were identified in prior analyses of NEEAR and CHEERS as potential confounders of associations between water exposure and GI illness (Wade et al. 2006, 2008, 2010; Dorevitch et al. 2012) were considered in multivariate regression models between water exposure and symptom severity (Supplementary Figure S2, Supplementary Table S1, available with the online version of this paper). A change-in-estimate analysis, using a list of potential confounders, identified confounders that resulted in at least a 5% change in the final measure of association between water exposure and symptom severity, were retained in the final model (Lee 2014). Age and comorbidities were thought to potentially modify the association between water exposure and severity, because those at the extremes of age or with comorbid conditions have been found to develop more severe symptoms after exposure to enteric microorganisms (Szilagyi et al. 1985; Glynn & Bradley 1992; Karaoglu et al. 2004; Lund & O'Brien 2011; Rocourt 2014). Separate models with multiplicative interactive terms between each of these factors and water quality were evaluated. Stratum-specific measures of association were assessed for interaction p-values <0.20. All models using NEEAR data controlled for beach as a fixed effect, and household cluster was also accounted for in logistic models. No clustering was evident in CHEERS. All data were analyzed using SAS^® 9.2 (SAS Institute Inc., Cary, NC).

The NEEAR data set consisted of 27,276 participants of whom 17,571 were water recreators, who engaged in telephone follow-up (Table 1). CHEERS had 11,297 participants of whom 7,710 were water recreators, who engaged in telephone follow-up (Table 2). In general water recreators tended to be younger compared to non-water recreators. In both studies, a greater fraction of water recreators who swallowed water developed GI symptoms within 0–3 days of recreation compared to those with less water exposure or to non-water recreators. Among CHEERS participants, those with GI symptoms were defined according to the data available in CHEERS and according to the data available in NEEAR (Table 1). When including those with less than three loose stools in a 24-hour period (CHEERS definition), there were more participants in CHEERS which could be considered to have GI symptoms.

In NEEAR, daily geometric mean enterococci measured by culture ranged from 0.3 colony forming units (CFU)/100 mL to 1,042.1 CFU/100 mL, with the 75th percentile of 43.0 CFU/100 mL. Enterococci measured using qPCR ranged from 5.8 calibrator cell equivalents (CCE)/100 mL to 1,442.7 CCE/100 mL with a 75th percentile of 169.2 CCE/100 mL. In CHEERS, water quality varied greatly according to the type of surface water being assessed. The Chicago River system was one of the locations for participant enrollment and water quality monitoring. At the time of the study, the majority of the flow through the Chicago River system was secondary treated but non-disinfected wastewater effluent (Rijal et al. 2011), resulting in much higher concentrations of both F+ coliphage and enterococci by culture compared to waterbodies such as Lake Michigan and inland lakes (Dorevitch et al. 2015). Water quality measurements were dichotomized at the 75th percentile for all combined water locations. The concentration of F+ coliphage ranged from 1.0 plaque forming units (PFU)/100 mL to 834.0 PFU/100 mL, with the 75th percentile at 16.1 PFU/100 mL. The concentration of enterococci by culture in CHEERS ranged from 0.1 to 18,318.1 CFU/100 mL, with the 75th percentile at approximately 463.8 CFU/100 mL.

No statistically significant duration ratios and SDs ratios were observed among NEEAR water recreators swallowing water and swimming compared to waders for all semi-continuous severity outcomes using the ZINB model (Table 4). However, enterococci by qPCR appeared to be an important predictor of duration of GI symptoms (duration ratio = 1.54 (95% CI: 1.04, 2.28)) and GI SDs (SDs ratio = 1.42 (95% CI: 1.10, 1.84)) in the ZINB models (Table 4). Results of logistic regression showed that major responses to symptoms (hospitalization, ED visit, HCP visit or phone call, prescription medication use, or missing work or missing daily activities, vs. only taking OTC medications or having no response to symptoms) were more likely among those with greater water exposure. Greater water exposure among NEEAR participants (swimming with swallowing water vs. wading) was associated with increased odds of reporting major responses to symptoms (OR = 1.57 (95% CI: 1.27, 1.95)). However, microbial measures of water quality concentrations were not associated with a major response to symptoms among NEEAR participants. No differences were observed for any outcome when assessing water quality in quartiles, rather than a dichotomous outcome among NEEAR participants (Supplementary Table S2, available with the online version of this paper).

Similar to NEEAR, no statistically significant associations were observed among CHEERS water recreators swallowing water and having body wetness compared to water recreators with no water contact for all semi-continuous severity outcomes using the ZINB model (Table 5). In addition, semi-continuous measures of severity were not associated with exposures to fecal indicator microbes, when dichotomized at the 75th percentile or when assessed as quartiles (Supplementary Table S3, available with the online version of this paper). However, greater water exposure among CHEERS participants (swallowing water or body wetness vs. water recreators with no water contact) was associated with an increased odds of reporting major responses to symptoms (OR = 1.50 (95% CI: 0.98, 2.30) and OR = 1.29 (95% CI: 1.06, 1.57), respectively). However, when dichotomized at the 75th percentile, concentrations of fecal microbes were not associated with major responses to symptoms among CHEERS participants. When assessing the relationship between quartiles of F+ coliphage and major responses to illness, we observed a significant association between quartile 2 compared to quartile 1. However, this association was not observed in comparisons of response to illness among participants who recreated in water with F+ coliphage concentrations above the median (Supplementary Table S3).

Model details can be found in Supplementary Tables S4 and S5 (available with the online version of this paper). Additional ZINB models using unexposed non-water recreators as a reference group can be found in Supplementary Tables S6 and S7 (available with the online version of this paper). Overall, there were no major impacts on the results when non-water recreators were considered as the reference group.

Results describing effect modification by comorbidities are shown in Table 6. No significant differences across strata (p < 0.20) were noted when considering age as a modifying factor. In CHEERS, the average number of daily bowel movements at baseline was found to modify the association between major versus minor responses to symptoms and water exposure. Overall, those with two or more daily bowel movements at baseline had greater odds of severe symptoms associated with greater recreational water exposure compared to those with only one bowel movement per day at baseline. In NEEAR, having a previous GI condition or having any comorbid condition (previous GI condition, asthma, or skin condition) modified the association between exposure and symptom severity (among all study participants, including those with no symptoms, GI symptoms only, and AGI). In general, the presence of comorbidities resulted in a stronger association with severity.

This study used a more detailed approach to characterize illness associated with water recreation by incorporating indicators of severity. Rather than considering illness to be a simple, dichotomous outcome, we made use of a continuum of symptom severity data. We observed a wide range of symptom severity among swimmers, waders, and incidental-contact water recreators, including those who did not meet the case definition of AGI. In general, most individuals who develop symptoms have mild and self-limiting symptoms; however, some individuals develop more severe symptoms that require use of the healthcare system and lost productivity. Indications of severity were common among those who had GI symptoms, but not necessarily AGI. Overall, qPCR measures of enterococci were associated with increased severity of symptoms among NEEAR participants (Table 4). The degree of water contact was associated with increased odds of reporting a major response to symptoms among both NEEAR and CHEERS participants (Tables 4 and 5).

To evaluate the semi-continuous severity outcomes, ZINB models were used in this analysis in order to take into account the excess zeros provided by participants that did not develop symptoms. Traditional negative binomial models were also evaluated, but were found to have poorer model fit according to the Akaike information criterion (AIC). Details of these models, are found in the Supplementary Tables S4–S7. Additionally, both Poisson and zero-inflated Poisson models were also considered, but resulted in poor model fit (not shown).

The results of this analysis suggest that increased exposure to fecal indicators may be associated with more severe outcomes among swimmers in NEEAR, but not among incidental-contact recreators in CHEERS. We did however observe increased odds of major responses to symptoms when evaluating quartiles of F+ coliphage, but this was only limited to one quartile comparison (quartile 2 versus quartile 1) (Supplementary Table S3). This is consistent with the finding that the occurrence of GI illness was associated with water quality in NEEAR (Wade et al. 2006, 2008, 2010) but not in CHEERS (Dorevitch et al. 2015). It is possible that the greater water exposure experienced by swimmers (Dufour et al. 2006) in NEEAR compared to incidental-contact water recreators (Dorevitch et al. 2011) in CHEERS results in greater exposure and potential dose. Previous waterborne studies (Keene et al. 1994) and other outbreaks (Taylor et al. 1984; Glynn & Bradley 1992; Mertens et al. 2013) have indicated that a greater dose of microorganisms can be associated with greater severity of symptoms.

The majority of recent US studies have relied on a common case definition to describe the occurrence of GI illness among water recreators, but a few alternatives to describing the occurrence of illness have been used. Some studies have used a definition of the occurrence of diarrhea (Colford et al. 2007; Heaney et al. 2012; Arnold et al. 2013), while others have utilized alternate more or less stringent definitions of illness (Colford et al. 2007). Moving beyond the use of a yes/no case definition may have potential to generate more complete information regarding illness among water recreators in future studies.

In contrast to the dichotomous illness classification used in prior studies, we observed that on either side of the dividing line between well and ill, a range of severity was observed. While those who met the case definition for AGI had on average higher durations of GI symptoms and more SDs, those not meeting the case definition also had appreciable severity of symptoms. Furthermore, considerable variability in severity was observed among those who met the case definition (Table 3).

The collection and analysis of responses to symptoms or use of the healthcare system may be more important for understanding the overall health impacts of exposure to fecally-contaminated recreational waters. For example, an individual may take an antiemetic to relieve symptoms of vomiting and nausea, or an anti-diarrheal agent to relieve symptoms of diarrhea. If these medications are effective in preventing vomiting and no more than two loose stools per 24 hours, the individuals, who clearly were symptomatic, would not be considered ‘cases’. Thus, the use of the semi-continuous measures and responses to symptoms we assessed may provide more detailed and nuanced description of the short term health consequences of water recreation.

The current study had several strengths. First, the severity of symptoms was evaluated among two large cohort studies of water recreation with similar protocols. The studies were set in several US locations impacted by human fecal pollution and evaluated exposures in urban (Dorevitch et al. 2012) and less-urban environments (Wade et al. 2006, 2008, 2010). This analysis also evaluated severity using several severity metrics relating to different types of water recreation and varying degrees of water exposure within those types of recreation.

The results of this analysis should be considered in light of several important limitations. First, the ascertainment of severity relied on self-report. Furthermore, several metrics, including the duration of symptoms and SDs, relied on self-reported start and end dates of symptoms. The ability for participants to recall specific information regarding the date their symptoms began may be problematic, and could impact the accuracy of the duration estimates. However, we do not think that bias was introduced, in that swimmers, waders, and incidental-contact recreators, with varying exposure to microbes, would likely have comparable patterns in self-reporting symptom onset and resolution dates. Also, while calculating SDs in this analysis, we chose to weight more severe symptoms, such as vomiting and diarrhea, by a factor of two, while leaving less severe symptoms (nausea and stomachache) weighted by a factor of one. While highlighting the severity of individual symptoms (diarrhea, vomiting) has been suggested (Freedman et al. 2010), objectively derived weighting values have yet to be described. For that reason, we have no basis for assessing the accuracy of the weighting factors that we applied. In addition, the two epidemiology studies took place on waters impacted by fecal pollution. Therefore, the results of the analyses of water exposure or microbe density as a predictor of severity may not be generalizable to the users of waters not impacted by point sources of fecal pollution. In addition, we chose to evaluate GI symptoms developing within 3 days of recreation, as this time frame to development of symptoms has been previously shown to be the most strongly associated with water recreation (Dorevitch et al. 2012; Arnold et al. 2013). However, we could be underestimating overall severity by not including symptoms corresponding to pathogens or viruses with longer incubation periods.

Additionally, differences in telephone follow-up may account for the numbers who reported GI symptoms, but not AGI among CHEERS versus NEEAR recreators (Table 3). It is possible that in NEEAR, the individual responding for the entire household may not have been fully aware of all minor symptoms (stomachache, a half a day of nausea) experienced by the entire household, which may have been recognized by individual self-report in CHEERS. However, it is unclear if proxy reports would over- or underestimate these measures.

Others have found that outside of the context of water recreation, socioeconomic status can impact decisions to seek medical care (Adler et al. 1993; Stratmann 1999; Anderson 2007; Blackwell et al. 2009). We found that among CHEERS participants, median household income in the participants’ home zip code was significantly higher (p < 0.01) among water recreators compared to non-water recreators. Furthermore, those who lived in zip codes with higher incomes were less likely to report indicators of severity in both NEEAR and CHEERS (Supplementary Tables S8 and S9, available with the online version of this paper). Potential socioeconomic differences should be further examined among water recreators compared to non-water recreators in future studies.

The severity of symptoms may be useful as a health outcome measure in future studies of water recreation. Evidence from this analysis indicates that the definition of illness used in recent studies results in some individuals with relatively low severity being included as cases of illness while some individuals with high severity are excluded. By examining the relationship between severity and water exposures, optimized definitions can be identified based on the resulting differences in symptom severity between water recreators and non-water recreators. Modeling environmental exposures as predictors of health outcomes might be improved if the burden of disease, which takes into account both the incidence and severity of symptoms, were studied as a health endpoint rather than simply looking at incidence.

For the CHEERS study: This research was funded by the Metropolitan Water Reclamation District of Greater Chicago. A peer review panel for this study was supported by the Water Environment Research Foundation. Support for analyses of water quality measures as environmental public health indicators was provided by US Environmental Protection Agency (US EPA) STAR agreement RD-83478901. The contents of this work on CHEERS are solely the responsibility of the grantee and do not necessarily represent the official views of the EPA. Further, the EPA does not endorse the purchase of any commercial products or services mentioned herein.

The views expressed in this manuscript are those of the individual authors and do not necessarily reflect the views and policies of the US EPA. Mention of trade names or commercial products does not constitute endorsement or recommendation for use.

In the past 3 years Dr Dorevitch has received grant support from the US EPA, US CDC, the National Association of Clean Water Agencies (NACWA), and the NSF. He has reviewed draft US EPA water quality documents as a consultant to the Eastern Research Group. No other potential competing interests are declared.