The use of faecal indicator organisms to manage microbial health risks in recreational waterways not impacted by point sources of sewage: a systematic review of the epidemiological evidence Open Access

The primary goal of managing natural recreational waters is to protect and promote public health. Scrutinising the evidence base for recreational water guidelines is therefore necessary to ensure that health and economic benefits are maximised while potential health risks and economic burdens are minimised (World Health Organization 2021). A key part of determining the suitability of water for primary recreation is based on levels of microbial contamination, which are frequently assessed by quantifying faecal indicator organisms (FIOs), often enterococci or Escherichia coli (EU Bathing Water Directive 2006; National Health & Medical Research Council 2008; U.S. EPA 2012). However, the evidence for their use is based on epidemiological studies conducted in waters impacted by point sources of sewage (U.S. EPA 2012). As both FIOs and pathogens are found in sewage in consistently high concentrations (Aw 2018), there is both epidemiological evidence and a logical basis to support the use of FIOs for managing recreational waters impacted by point sources of sewage.

The evidence for using FIOs to manage recreational waters not impacted by point sources of sewage is less clear. While it is acknowledged that non-point/diffuse contamination may still contain human sewage, the relationships between FIOs and pathogens diminish as the number of individuals contributing to the faecal source decreases, an important caveat noted by Cabelli (1983). Regulatory bodies such as the United States Environmental Protection Agency (U.S. EPA) acknowledge that FIO–illness associations are most robust at sites dominated by point-source sewage pollution. However, their current guidelines state they are protective of public health, regardless of contamination sources (U.S. EPA 2012). This is a point of contention (Fujioka et al. 2015) that has led to the progressive development of technical guidance for developing alternative, site-specific approaches that are risk-based, including Quantitative Microbial Risk Assessment (QMRA) where sources are predominantly non-human or non-faecal (U.S. EPA 2012, 2014, 2024). Despite this guidance, waterway managers largely continue to rely on default FIO-based guidelines due to precedent, practical constraints, and the cost and complexity of implementing site-specific risk assessments.

Supposing FIOs are not appropriate measures of health risk outside of point-source impacted recreational waters, two problematic consequences may arise: (1) Unrealised health and economic benefits resulting from waterway closures when FIOs are elevated, but pathogens (with their attributable health risks) are not (=false-positive), and (2) a failure to protect public health when FIO levels are low in the presence of elevated pathogens (=false-negative) (Abdelzaher et al. 2010).

To date, reviews and meta-analyses of epidemiological studies in recreational waters have been conducted across a range of timeframes (Prüss 1998; King et al. 2014; Fewtrell & Kay 2015), illness types (Yau et al. 2009; Mannocci et al. 2016), fresh or marine water types (Wade et al. 2003; Adhikary et al. 2022), and animal wastes (Dufour et al. 2012). Some reviews have included and commented on a subset of the literature addressing non-point/diffuse sewage-impacted recreational waters, highlighting potential deficiencies in the current FIO approach (Fewtrell & Kay 2015; Arnold et al. 2016). However, there remains a gap in the review space for an in-depth consideration of the global epidemiological evidence for this subset of recreational waters. This is reflected in non-point sources of contamination being listed as a research need in the most recent recreational water guidelines by the World Health Organization (WHO) (2021).

In conducting this systematic review, we sought to answer the following research questions:

• 1. Do bathers have a greater risk of gastrointestinal (GI) or other illnesses compared with non-bathers following exposure to recreational waters that are not impacted by point sources of sewage pollution?

• 2. What is the strength of the evidence linking microbial water quality parameters to reported health outcomes in waters not impacted by point sources of sewage pollution?

We addressed these specific research questions by systematically reviewing the epidemiological literature published globally since 1986 when the U.S. EPA first recommended E. coli and enterococci as recreational water indicators over faecal coliforms.

We conducted a systematic literature review using the Preferred Reporting Items for Systematic Review Recommendations (PRISMA) protocol (Moher et al. 2009). Original research publications were identified by searching the Medline, Scopus, and Web of Science digital databases, while dissertations/theses were identified using the UMI/ProQuest Digital Dissertation Database. Searches were carried out in November 2020 and repeated prior to publishing in July 2024. Search terms included ‘((recreat* OR swim* OR bather OR bathing) AND (water* OR beach*) AND (epidemiolo* OR randomised* OR cohort* OR cross-sectional OR ‘cross sectional’ OR ‘case control’ OR ‘case-control’ OR prospective OR retrospective OR outbreak*) AND (microbial OR bacteria* OR fecal* OR faecal* OR indicator* OR pathogen*))’. Results were screened and imported into an Endnote library, with 50 irrelevant titles in a row used as the cut-off to stop evaluating articles from the databases. Titles and abstracts were screened within the Endnote database, and irrelevant articles were removed. Additional references were identified by forward and reverse citation searches of included studies and previously published recreational water epidemiological reviews and meta-analyses (Prüss 1998; Yau et al. 2009; Dufour et al. 2012; Fewtrell & Kay 2015; Arnold et al. 2016; Mannocci et al. 2016; DeFlorio-Barker et al. 2018; Leonard et al. 2018; Russo et al. 2020; Adhikary et al. 2022) as well as a WHO report providing recommendations for the EU Bathing Water Directive (World Health Organization 2018).

This review considered epidemiological evidence from studies of recreational waters that were not impacted by point sources of human sewage. Point sources of human sewage were defined as:

• Treated or untreated discharges from sewer outfalls, and urban stormwater contaminated by sewage (including discharges caused by failing wastewater pumping stations, sometimes called lift stations).

Contamination sources that were considered included:

• Diffuse urban runoff, point and non-point sources of agricultural runoff, contamination from wildlife and domestic animals, and ‘non-point’ sources of human sewage such as from bathers themselves or leaky infrastructure impacting nearby ground, sand or water.

Studies were eligible for inclusion in this review if they met the following criteria: (1) published peer-reviewed and/or grey literature epidemiological study conducted in natural recreational waters without point sources of human sewage; (2) reported in English; (3) investigated the incidence of illness outcomes in bathers compared with non-bathers, and/or; (4) measured associations between bather health outcomes and microbial water quality parameters.

Studies were excluded if they reported a point source of human sewage directly impacting the recreational site or if inadequate information was presented to determine the sources of contamination. Some studies reported point sources of sewage upstream from the study site; these were included if the authors provided sufficient evidence of their lack of impact, e.g., through previous site investigations or by virtue of hydrodynamic conditions.

Data were extracted from each paper into a topic-specific Excel database that included but was not limited to: basic study information such as year, title, and authors; descriptive information on study design, climate, water type, water quality measures, population type/size, illness symptoms, and analysis methods; and research findings such as bather vs. non-bather illness outcomes and significant relationships with water quality measures. Similarities, differences, and trends of studies were assessed using exploratory summary tables and conditional formatting rules in Excel. This review focused primarily on GI illness, the most investigated and common illness outcome in recreational water epidemiological studies (Leonard et al. 2018). However, we also consider other outcomes such as ear, eye, skin, and respiratory symptoms as these are also commonly associated with recreational exposures (Prüss 1998; Leonard et al. 2018). A statistically significant result was considered if the adjusted 95% confidence interval (95% CI) did not span zero or if a p-value alpha of <0.05 was reported.

Study quality was evaluated using the Office of Health Assessment and Translation (OHAT) Risk of Bias Rating Tool for Human and Animal Studies (OHAT 2015). Two team members collaborated to evaluate each study, applying a checklist of seven to nine questions (depending on the study design type) across seven domains of bias (OHAT 2015). Studies were rated on each of these questions using the OHAT four-point scale (‘definitely low risk’, ‘probably low risk’, ‘probably high risk’, and ‘definitely high risk’). The OHAT protocol was modified to include additional sub-criteria for both ‘exposure’ and ‘outcome’ bias in the detection domain. This modification allowed us to differentiate levels of bias arising from the data collection methods (e.g. self-report or measured) and the content validity of the exposure and outcome metrics (e.g. was the case definition of GI illness credible?). The sum of the risk of bias ratings was used as a crude measure to rank the studies from lowest to highest risk of bias. This was done by assigning points to the four-point scale (from +2 to −2) from ‘definitely low risk of bias’ to ‘definitely high risk of bias’, respectively. We excluded studies with an overall negative rating for this section of the results to ensure the synthesis of the epidemiological findings was based on evidence of sufficient quality. Additional criteria for assessing the risk of bias are in Supplementary material, A.

An overview of the key characteristics of the included studies is presented in Table 1 and described in Sections 3.2.1–3.2.4. Table 2 contains an overview of the target population, comparator group, and exposure and outcome assessment methods used in each study.

Two epidemiological study designs were identified: prospective cohort (13 studies) and randomised control trial (RCT) (one study). One of the prospective cohort designs differed from others; they recruited families from the local region, and participants had repeated exposures recorded over a bathing season instead of a larger cohort with participants reporting on a single exposure event (Calderon et al. 1991). The sample sizes from the study sites of interest ranged from 343 to 12,469 individuals, except for the study with repeated exposures, which had 104 families and 1,310 exposure person-days. Most studies had a single follow-up interview (within 3–19 days), but one study had three follow-up periods, at approximately days 2, 5, and 21. (Dorevitch et al. 2011).

Almost two-thirds of the studies were conducted in the United States (9/14) (Table 1). There was one each from Australia, Greece and New Zealand, and two from South Africa (representing the pilot and main studies from the one location). Most focused on marine waters (10/14), with three in fresh waters and one in an estuarine environment. Six different Köppen climate classifications were represented. The majority represented either a ‘warm’ or ‘hot’ summer Mediterranean (n = 7) or a ‘warm-summer’ humid continental climate (n = 3). All studies conducted their fieldwork (or a significant portion thereof) over their high bathing season, typically summer.

Most studies recruited both adults and children, though several had additional criteria around a minimum child age or explicitly recruited families with a child ten years or younger. The single RCT considered adult bathers only.

All studies had a non-bather comparator group. All but two of the prospective cohort studies used a comparator group whose members were recruited concurrently at the same site(s) as the bathers. The other two cohort studies used either age-adjusted background illness rates as their control or recruited non-bathers from the same residence areas as the bathers. Three studies included those who waded in the water (but only to specific depths and without submerging their head) as controls alongside non-bathers in their analysis.

The RCT by Fleisher et al. (2010) had the most robust approach; supervising and recording the exposure activity of each bather, including the duration, number of head immersions, location, and unusual activities for each individual. However, the 15-min exposure is comparatively low to the durations reported for self-selected bathers, e.g., a mean of 50 min in Papastergiou et al. (2012) and a 50th and 75th percentile of 20 and 120 min in Colford et al. (2005). Dorevitch et al. (2012) interviewed individuals rather than relying on the proxy self-report of measures by an adult on behalf of the family unit as in the rest of the studies.

The most common definition of bathers required head immersion. This was a requirement of participants in the RCT. Of the cohort studies, most (9/14) recorded varying levels of each participant's extent of exposure to recreational water, but five did not differentiate between bathers with complete (head immersion) and partial body immersion. Only three of the cohort studies included and reported on exposure duration explicitly in their analyses. In both cases, duration was dichotomised (above or below 30 or 60 min) to assess the risk of illness for short and long-duration bathers.

The epidemiological evidence assessment presents a more detailed overview of microbial water quality variables (Section 3.4.2).

All studies relied on self-report data for the illness outcome assessment. While no study used a ‘gold standard’ clinical method for assessing outcomes (OHAT 2015), one invited symptomatic participants to provide stool samples for pathogen analysis to compare bather and non-bather populations (Dorevitch et al. 2012) (see Section 3.4.2).

There were 20 unique case definitions for GI illness across the 14 studies reviewed. Most used a tiered approach, with simple GI illness definitions and/or more stringent definitions of GI illness. An example of simple GI case definitions included ‘any report of vomiting, diarrhoea, stomach-ache or nausea’ while an example of the more credible GI case definitions included ‘three or more episodes of diarrhoea within 24 hours.’ Most (11/14) included one or more credible GI case definitions that were referred to as either ‘Highly Credible Gastrointestinal Illness’ (HCGI) or ‘Acute Gastrointestinal Illness’ Of the three studies that did not use a complex definition, only Calderon et al. (1991) attempted to provide a better indication of illness severity by indexing ‘whether or not an individual had stayed at home, remained in bed or sought medical help’. However, no results were reported that utilised these variable categories of severity. All studies included vomiting alone as one definition of GI illness.

The 14 studies were assigned variable risk of bias ratings across the 13 domains of bias assessed (Table 3). There was no single domain with a consistently low risk of bias across all the studies, reflecting the variety of protocols used. Strengths observed across most studies included appropriate comparator groups, robust design/analysis for confounding factors, transparent reporting of attrition, appropriate statistical methods employed, and adequate reporting of outcomes. It is worth noting that the OHAT tool stipulates a rating of ‘Probably Low Risk of Bias’ for studies that do not report on protocol deviations, which may not be informative for discriminating between study qualities (OHAT 2015). Only two studies reported explicitly on this domain (Fleisher et al. 2010; Wade et al. 2010), with the remainder receiving the default ‘Probably Low Risk of Bias’.

Table 3

View Large

Weaknesses in study designs were mainly observed within the detection domain. These included ambiguity in definitions and measures of both exposure and illness outcomes. Exposure content validity refers to how well the measured parameter reflects the primary exposure of interest (i.e. ingestion of pathogens). Most studies measured varying levels of recreational water contact (e.g. body immersion, head immersion, swallowed water). However, few studies included bathing duration, which could have further differentiated likely levels of ingested water. This led to a rating of ‘Probably High Risk of Bias’ for most of the cohort studies due to the potential for misclassification bias. Regarding exposure-outcome measures, only one of the cohort studies explicitly reported using a survey instrument validated through experimental exposure studies (Dorevitch et al. 2011). All other studies received a ‘Probably High Risk of Bias’ due to the potential for detection bias under this criterion. Regarding illness outcome, the use of self-report data for illness outcome measures was common to all studies. Therefore, all studies received a ‘Probably High Risk of Bias’ rating due to the potential for recall, risk perception and misclassification bias (Fleisher & Kay 2006; OHAT 2015). Most (11/14) studies used at least one credible case definition for GI illness, but none conducted clinical sampling for all participants, leading to an overall rating of ‘Probably Low Risk of Bias’.

As expected, the crude rating of the studies showed that the single RCT study by Fleisher et al. (2010) had the lowest risk of bias. Among the other studies, most (8/13) received a positive (‘low-bias’) rating, while five studies received ratings of ≤zero. Consequently, only the nine ‘low-bias’ studies with a positive rating were considered in synthesising the epidemiological evidence.

Most of the low-bias studies were conducted in the US (7/9), in marine waters (8/9) and included large cohorts (>2,000 participants in the prospective cohort studies). These studies included more credible GI case definitions and used appropriate statistical methods with adequate adjustments for confounding variables (9/9). Most explicitly accounted for clustering of illnesses at the household level (5/9), or in the case of Dorevitch et al. (2011), determined that this was not required as <5% of participants enrolled with a family member and individuals rather than family units were interviewed. Additionally, all studies collected data about recent contact with ill persons (including household members), though it was not always apparent if the final models included adjustment for this variable (McBride et al. 1998; Colford et al. 2007).

The details of the low-bias studies tended to be published in either large technical reports, over several publications and/or included supplementary content. Shorter publications that received a higher risk of bias ratings for some domains (such as ‘attrition’ and ‘selective reporting bias’) may have done so for the following two possible reasons: either the condensed reporting format did not lend itself to reporting on all domains or less robust methods were used and not reported transparently. It was not possible to differentiate between these reasons. However, these shorter studies also tended to have a less robust comparator group and poor adjustment for confounders, making the latter reason more likely in some cases.

There was variation in the case definitions of GI illnesses within the low-bias studies. While the studies published in the U.S. have mostly standardised the primary GI case definition, differences were still identified, particularly in the earlier studies (Colford et al. 2007; Fleisher et al. 2010). The two studies conducted outside of the U.S. used case definitions that were notably different from the U.S. studies. Further contributing to this variation was the use of composite case definitions that were often aggregated in the analyses, i.e., for most studies, it is not possible to differentiate the proportions of participants who experienced three or more cases of diarrhoea within 24 h from those who experienced nausea with a self-diagnosed fever or cramps with missed daily activities, etc.

Assessing the impact of variations in case definitions was beyond the scope of this review. To facilitate comparison across studies, we categorised the case definitions used by the study authors into ‘Credible GI’ or ‘Simple GI,’ aligning with how the authors structured their case definition tiers (see Supplementary material, B).

The depth of information characterising contamination sources was variable across the reviewed studies. For example, some studies referenced Microbial Source Tracking (MST) work and/or reported on site-specific drivers of contamination. Others only briefly reported key sources of contamination.

Three marine studies noted predominantly non-human faecal contamination, including birds (Colford et al. 2007), dogs and birds (Fleisher et al. 2010), and the work from McBride et al. (1998) which included two ‘control (minimal impact)’ and two ‘rural (animal waste impact)’ beaches. There was ambiguity around the specific animal faecal sources, however, their introduction referred to the large population of cows and sheep in New Zealand (McBride et al. 1998). Except for the study by McBride et al. (1998), all of the studies included in this review reported known/potential human sources of contamination at the sites of interest. These human sources typically included diffuse urban runoff and suspected leaky infrastructure such as from onsite septic systems or other sewerage infrastructure.

The evidence synthesised here is restricted to the nine epidemiological studies rated as ‘low risk of bias’ (Table 3). An overview of each study and its research objectives are provided in Supplementary material, C. The results are considered from several perspectives, including:

• Rates of illness between bathers and non-bathers

• Associations between microbial water quality and bather illness outcomes

• Associations between non-microbial parameters and bather illness outcomes

The reviewed studies used a range of metrics to measure excess bather illness, including adjusted odds and risk ratios (relative risks), adjusted risk differences and attributable risks. Studies that reported a significant association between exposure level and at least one health outcome (8/9) are summarised in Table 4. The strength of association for most health outcomes typically increased with increasing exposure level suggesting an exposure–response relationship.

Associations between bathing and GI illness were found consistently (7/9 studies, Table 4). Of note is that McBride et al. (1998) did not report statistics on bather vs non-bather health outcomes for the non-point sites of interest (instead pooling data across all seven study sites), so these results were not included in Table 4. However, they did note that bathers' illness risk at the animal-impacted sites was similar to those at the human-impacted sites, which were both significantly higher than at the ‘minimal impact’ control sites. The RCT by Fleisher et al. (2010) reported non-significant results (i.e., the 95% CI p-value was not <0.05), however, their results revealed that bathers reported a shorter time–to-illness onset for every illness outcome compared with non-bathers, consistent with several studies that reported significant relationships. Significant associations were often of borderline statistical significance, had wide confidence intervals, or were mediated by environmental conditions. For example, Yau et al. (2014) found that bathers had a significantly elevated risk of Credible GI only when sewage-contaminated groundwater discharge was high or solar radiation was low.

The risk of GI illness for bathers appears to be highest within only a couple of days following water exposure. Four of the nine studies considered the time-to-illness onset in various ways. Colford et al. (2012) reported a strong peak in the onset of diarrhoea among bathers two days after exposure but did not observe the same for non-bathers. This rapid onset of bather illness is supported by the findings of Dorevitch et al. (2011), who compared the risk of GI after 0–2, 0–4, and 0–5 days and found the odds of GI illness decreased as the length of follow-up increased (though actual odds ratios were not presented). Arnold et al. (2013) specifically included time-to-illness onset in their modelling and found that by day four, the incidence of diarrhoea and Credible GI among bathers returned to non-bather levels.

The impact of bathing duration on reported health outcomes was poorly characterised. Only four of the nine studies reported information about bathing duration. Where the RCT controlled bathing duration at 15 min, the other studies dichotomised this variable at either ≥60 min (Colford et al. 2007; Papastergiou et al. 2012) or ≥30 min (McBride et al. 1998). Two of these studies found a significantly increased risk of Credible GI (McBride et al. 1998) or Simple GI (Papastergiou et al. 2012) associated with longer duration bathing. While Colford et al. (2005) compared long-duration bathers with non-bathers and found a significantly increased risk of diarrhea, this result was similar to the results for all bathers compared with non-bathers, which they interpreted as a small but insignificant increase in the risk from prolonged exposure. Significant findings were also reported for respiratory outcomes by McBride et al. (1998) and eye illness by Papastergiou et al. (2012). However, the results reported by McBride et al. (1998) represent the pooled data of seven differentially contaminated sites, three of which were excluded from this review for having point sources of sewage contamination.

Numerous microbial targets with a range of analytical methods were employed across the included studies. All studies presented water quality results for enterococci by culture methods. Most also included three or more other measures of FIOs such as E. coli, total coliforms, faecal coliforms (by culture, molecular, and most probable number methods), Coliphages (culture) and various microbial source tracking markers for faecal contamination (qPCR). A summary of the measures is presented in Supplementary material, D.

There was low variability and a positive skew in cultured enterococci (CFU/100 mL) within the reviewed studies. Overall, water quality was typically good to excellent, with only two studies conducted in waters that would be considered poor quality. One of these was conducted in marine water (Fleisher et al. 2010) and the other included selected riverine sites within the freshwater study that considered limited contact recreational exposures (Dorevitch et al. 2011). An overview of the reported water quality for enterococci concentrations for each study is presented in Supplementary material, E. Variable measures of the central tendency of enterococci concentrations were reported, including the geometric mean, mean and median which were typically ≤30 CFU/100 mL (7/9 studies). The remaining two studies reported a geometric mean of 70 CFU/100 mL (Dorevitch et al. 2015) and a median of 71 CFU/100 mL (Fleisher et al. 2010).

Few positive associations were found between microbial measures and bather illness rates despite many targets investigated (Table 5). In fact, around half of the significant reported associations were negative, underscoring a lack of a directional relationship. . While five of the nine studies reported significant and positive associations, the results were not consistent. In short, no two studies reported the same combination of health outcome and microbial water quality measure. Furthermore, while enterococci were the most frequently reported FIO to correlate with varying illnesses, both positive and negative associations were found, sometimes within the same study where alternative enumeration methods were used in parallel.

There is little evidence of a consistent dose-response relationship between microbial measures and GI illness. When considering the ‘Credible GI’ case definitions, both positive and negative associations were again reported. Further, Colford et al. (2007) raised caution in interpreting the significant positive associations between Credible GI and ‘per unit increase’ of male-specific coliphage as a low number of participants were exposed to the water when this coliphage was detected.

No evidence was found to indicate that measured pathogens have utility as predictors of health outcomes in waters not impacted by point sources of sewage. Most of the reviewed studies (7/9) included at least one pathogen, most commonly the bacterial pathogen Staphylococcus aureus (5/9), and/or viral pathogens such as adenovirus (4/9), norovirus (3/9) and enterovirus (3/9). Protozoan pathogens such as Cryptosporidium and Giardia were less commonly analysed (2/9) (Supplementary material, D). Pathogen detections were infrequent and not predictive of any health outcomes investigated. They were sometimes mentioned only in brief and were absent from tabulations of health associations, assumedly due to non-detects or a lack of associations with health outcomes. One study invited symptomatic participants (both bathers and non-bathers) to submit stool samples for pathogen analysis (Dorevitch et al. 2012). No specific pathogens or pathogen groups were significantly associated with bathing or the extent of exposure.

It is crucial to consider the extent of microbial contamination when significant associations are reported between bather illness and microbial parameters. Water quality guidelines are designed to be moderately protective of health; thus, waters with fewer exceedances would be expected to have lower rates of GI and other illnesses. The evidence summarised here suggests that this assumption may not apply to waters that are not impacted by point sources of sewage. While the majority (6/9) of studies reported low percentages (<14%) of samples that exceeded guidelines for enterococci, three of these six studies reported significant and positive associations between FIOs and one or more illnesses. In contrast, the study with the most exceedances (35% of samples in Fleisher et al. 2010 exceeded the California single sample water quality standard of 104 Colony Forming Units (CFU)/100 mL) did not report significant associations between microbial water quality and health outcomes (at the p < 0.05 level). While they did report a positive but non-significant association between enterococci and skin symptoms, the authors suggested that this was explained by the significant association between skin symptoms and 24-h antecedent rainfall (which may have flushed contaminants into the water (Fleisher et al. 2010). An overview of the reported water quality for each study is presented in Supplementary material, E.

Several studies reported significant associations between self-reported symptoms and non-microbial parameters (Table 6). Parameters relating to bather-only analyses included bather density, bathing duration, 24-h antecedent rainfall, and water temperature. Yau et al. (2014) found that Credible GI was only associated with FIOs under conditions of high submarine groundwater discharge or low solar radiation. Papastergiou et al. (2012) conducted a prospective cohort study in three marine beaches in Greece that met the EU's ‘excellent’ water quality guidelines. Those participants reporting Credible GI illness were almost two times more likely to have spent >60 min in the water (compared with <60 min) and were two times more likely to have been exposed to the beach with the highest bather density (compared to the lowest bather density beaches). These non-microbial parameters were also associated with Simple GI, respiratory, eye and ear symptoms. Sinigalliano et al. (2010) reported results of the RCT at one marine beach in the US where reported symptoms were associated with environmental parameters. Reported skin symptoms were positively related to 24-h antecedent rainfall, while acute febrile respiratory illness was inversely associated with water temperature.

The purpose of this review was twofold. First, to investigate whether bathers have a greater risk of health outcomes compared with non-bathers, and second, to evaluate the strength of the evidence linking microbial water quality parameters to reported health outcomes for bathers. The evidence summarised included only low-bias epidemiological studies at locations that were not impacted by point sources of human sewage. Notably, the focus on this subset of recreational waters yielded only nine low-bias studies, with all but one conducted in marine waters. In contrast, reviews and meta-analyses that include point sources of sewage often contain dozens of papers, even when narrowed in scope by water type or illness outcome. This presents only a small pool of evidence that regulators can use to inform policy, particularly as it pertains to safe freshwater recreation. In essence, this limits the extent to which patterns in associations (or a lack thereof) can be differentiated confidently from randomness. Nonetheless, these findings provide useful insights which are categorised below.

Bathers had a significantly higher risk of one or more specific illnesses than non-bathers. Further, strengths of association typically increased with increasing levels of exposure for most significant health outcomes, indicating a biological gradient as opposed to systematic sources of bias. However, some of the wide confidence intervals reported suggest that results are subject to a high degree of variability and/or a low degree of precision (Naimi & Whitcomb 2020). This may be explained in part by limitations described under methodological insights below. For example, several studies reported that the highest rates of bather GI illness occurred within a few days following water exposure. However, most studies used follow-up periods >7 days, and typically between 10–14 days. Arnold et al. (2013) speculate that the longer follow-up periods could dilute associations between bathing and infection risk by introducing recall bias and/or illness ‘noise’ from unrelated exposures. In contrast, while shorter follow-up periods may reduce recall bias, they may discount bathing-related illnesses with longer incubation times. The selection of an appropriate follow-up period(s) at a given site should be guided by the incubation times of the likely pathogens involved given identified sources of contamination, rather than what has been used in past studies.

There is insufficient epidemiological evidence to support the use of microbial measures to predict human health risks in these waters. No specific microbial measures of water quality (including pathogens and alternate FIOs) were associated consistently with any illness outcome. Further, the relatively even split of positive and negative associations indicates the potential for either unrealised health and economic benefits during unnecessary waterway closures and/or a failure to protect human health when a waterway closure is warranted. In essence, waterway managers should be aware of the potential to misclassify the microbial safety of a waterway if decisions are based solely on microbial measures.

The health outcomes of bathers sometimes contradicted what would be expected based on the FIO safety benchmarks derived from epidemiological studies with point sources of sewage. For example, bathers at some sites had a higher risk of illness when exposed to compliant water quality (Papastergiou et al. 2012; Arnold et al. 2013), while exposure to waters that exceeded threshold guideline levels of enterococci did not always correlate with a higher risk of illness (Colford et al. 2007; Colford et al. 2012). These contradictions are consistent with previously published reviews that touched on non-point-impacted waters (Fewtrell & Kay 2015; Arnold et al. 2016). The meta-analysis of 13 U.S. studies by Arnold et al. (2016) included a comparison of nine point and four non-point-impacted sites (the latter being included in this review). While not the primary focus of their paper, the authors noted that associations between illness outcomes and microbial measures did not hold for non-point-impacted sites. This is not surprising; the limitations of using bacterial FIOs as surrogates for all types of pathogens (including viral and protozoan) have been discussed extensively in the literature, especially in recreational waters not impacted by point sources of sewage (Boehm et al. 2009; Harwood et al. 2014; Korajkic et al. 2018). Put simply, the FIO paradigm does not account for the fact that the dominant source of FIOs is not always the dominant source of health risk (Schoen & Ashbolt 2010).

The limited variability in microbial concentrations (both within and across studies) presents a challenge to establishing a biological gradient from the epidemiological studies reviewed. As noted by King et al. (2014) when distributions of microbial water quality parameters are positively skewed with low variability, the insufficient contrast in exposure levels reduces the statistical power of detecting significant associations. As stated in some of the reviewed studies, a larger variation in FIO concentrations would improve our ability to elucidate these relationships further (McBride et al. 1998; Arnold et al. 2013). However, including sites with higher concentrations of FIOs in epidemiological studies may prove difficult given that (1) waterway managers should be actively dissuading recreation at such sites, (2) Fewer people may be available for recruitment at such sites, and (3) it would be more difficult to get ethical approval for studies where participants are knowingly being exposed to waters with high levels of microbial contamination.

There was virtually no evidence that measured pathogens caused the health risks observed. This was expected for several interrelated and potentially compounding reasons, including limited pathogen sampling (compared with FIOs) and others summarised in Table 7. Moreover, the single study that attempted to identify causative agents in stool samples from a subset of participants with GI illness found no evidence of a distinct set of pathogens among bathers compared with non-bathers (Dorevitch et al. 2015). The utility of pathogen sampling is likely limited to specific situations and outbreak investigations given the high resource cost.

Stratifying the findings by the various sources of contamination is challenging due to the limited number of studies and the heterogeneity of study protocols and site-specific drivers. This essentially precludes any firm conclusions being drawn viz. health outcomes following exposure to animal vs human sources of faecal contamination. For example, the three studies on marine waters impacted predominantly by animal faecal sources reported an elevated risk of some illnesses for bathers compared with non-bathers, similar to the studies of diffuse human-impacted sites (McBride et al. 1998; Colford et al. 2007; Fleisher et al. 2010). Similarly, reported associations between illnesses and measures of water quality were not consistent. However, the sites impacted by birds (Colford et al. 2007) or dogs and birds (Fleisher et al. 2010) also showed minor signals of human sewage contamination via MST sampling, further confounding a sound comparison. McBride et al. (1998) reported ‘no substantial differences’ for bather illness risk based on the source of faecal material (i.e., human or animal), though odds ratios were aggregated across differentially contaminated sites and the specific animal sources were ambiguous. Interestingly, their short follow-up period of 3–5 days seems incongruent with the incubation period of key protozoan pathogens. For example, Cryptosporidium spp. and Giardia infections can incubate for up to two weeks (Certad et al. 2017). Given they are found in bovine and ovine faeces within New Zealand (Moriarty et al. 2011; Mawly et al. 2015), this raises the question of whether some bathing-related illnesses at the animal-impacted sites were underreported.

A nuanced understanding of site-specific characteristics is needed to better protect bather health. Finding meaningful and consistent associations often relies on the use of standardised protocols that enable direct comparisons and meta-analyses, as in the case of setting standards for ambient air quality (Katsouyanni et al. 1996). However, meta-analysis may not be feasible for this subset of recreational waters as it presupposes consistent underlying associations between select measures of microbial quality and bather health outcomes. The evidence suggests that the sources and drivers of health risks are site-specific in the absence of point sources. Examples from the reviewed studies included bather density (Papastergiou et al. 2011), antecedent rainfall (Fleisher et al. 2010), tidally influenced discharge of sewage-contaminated groundwater (Yau et al. 2014) and whether an intermittent sand berm prevented potentially sewage-contaminated river inflow (Colford et al. 2012). These site-specific drivers are spatially and temporally variable and not generally incorporated into conventional routine monitoring. Waterway management strategies would be better served with the nuance of site- and regional-specific data (Verhougstraete et al. 2020), with a particular focus on catchment/sanitary surveys rather than a reliance on compliance-based monitoring derived from sewage-based epidemiological studies. Indeed, the latest recreational water guidelines by the WHO (2021), state that water quality managers should be aware of potential exceptional circumstances that could affect their waterbodies and what actions to take in response. They further state that monitoring should be a tool to support common sense and practical preventative measures, rather than being the focus of water safety management.

The rigorous design of epidemiological studies is critical to the evidence-based management of recreational waterways. The largest threats to validity that emerged in the Risk of Bias assessment stemmed from the poor characterisation of either exposure or health outcome variables, or both (Table 3). Most of the studies received a ‘probably high risk of bias’ rating for these domains, from their use of non-validated instruments that relied heavily on self-reported information. These methodological issues introduce a range of biases that can distort associations in either direction (OHAT 2015) and challenge any firm conclusions from the evidence. Overall, the methodological limitations and heterogeneity of study protocols are not unique to this subset of studies, with previous reviews reporting similar challenges among a broader range of studies (Leonard et al. 2018; King et al. 2014).

Case definitions of GI illness varied between the studies, challenging direct comparisons. This difference was greatest for the two studies conducted outside of the U.S. Different case definitions can significantly impact the observed incidence of GI illness; Majowicz et al. (2008) demonstrated a 1.5–2.1 times difference in incidence rates between four case definitions of GI illness using intra-country population data. For the reviewed studies, this likely resulted in more or fewer illnesses being reported for similar water quality and exposure levels between studies; an issue also highlighted by the U.S. EPA (2012). The case definitions used to describe Credible GI illness (such as HCGI, AGI) are more likely to be pathogen-mediated compared to those describing Simple GI cases such as nausea alone. However, as no study utilised a ‘gold standard’ clinical method for assessing outcomes for every participant it is not possible to rule out asymptomatic cases, recall bias, or alternative causes for reported symptoms, such as physical irritation from the ingestion of salt water (Colford et al. 2007; Arnold et al. 2013) or alcohol (Preedy 2018).

Emerging tools provide an exciting scope for reducing bias in environmental epidemiology. It is more conceivable than ever that rapid, reliable, and cost-effective methods may soon be available to enhance epidemiological insights. For example, multiplex immunoassays from saliva samples are being increasingly used to examine antibodies as biomarkers of exposure to waterborne pathogens (Wade et al. 2019; Augustine et al. 2021; Egorov et al. 2021). Other examples emerging in the water sector include rapid analytical biosensors (Nnachi et al. 2022), remote sensing (Oliva et al. 2023), and additive manufacturing (including 3D printing) (Priyadarshini Dikshit & Zhang 2020). Further, a review of genetic faecal pollution diagnostics highlights our rapidly expanding understanding of health-related water microbiology (Demeter et al. 2023). Such tools and knowledge will be valuable in improving the accuracy, sensitivity and consistency of measures of exposure, outcome, and susceptibility. These advances could effectively increase the statistical power of epidemiological studies by reducing nondifferential misclassification bias and/or confounding (Pekkanen & Pearce 2001).

Global trends are towards wider programmes of integrated catchment management, utilising molecular tools such as MST markers, and modelling methods such as QMRA (Cho et al. 2016; Federigi et al. 2019). In the United States, the U.S. EPA has facilitated the use of QMRA to set site-specific water quality objectives in selected recreational waters (U.S. EPA 2014, 2024). These tools hold promise for waterway managers, especially with emerging evidence characterising the risk profiles of different contamination sources (Soller et al. 2014; Brown et al. 2017; Boehm & Soller 2020; Ahmed et al. 2024; Burch et al. 2024). However, this is a rapidly developing area of science and significant limitations and knowledge gaps persist. These are highlighted in a publication arising from a QMRA community of practice workshop (Hamilton et al. 2024) and in a review of the utility of MST for recreational water management (Leonard 2022). As an example, Zimmer-Faust et al. (2020) sought to progress the development of site-specific water quality objectives at a beach that was considered a good candidate for QMRA. However, applying highly sensitive (digital PCR) MST methods yielded frequent but low concentrations of human markers, contradicting earlier qPCR-based MST evidence. Despite considerable time and money spent on source characterisation at the site, a QMRA was not conducted because pathogen loading from potential human sources could not be quantified confidently. It seems likely that this would constrain many QMRAs in urban recreational waters impacted by ‘predominantly non-human’ sources, but with low-levels from human sources that are spatially and temporally variable. This is further complicated by factors such as differential environmental decay of microbial targets (WHO 2021) in the context of increasing global temperatures and weather extremes (IPCC 2023). While rigorous source tracking can provide valuable insights into contamination sources, the case study by Zimmer-Faust et al. (2020) raises a crucial cost-benefit question: how can we ensure the safety of recreational waters equitably, considering a finite pool of resources?

Managing recreational waterways to protect public health remains a complex challenge. Ultimately, robust associations that could predict bather illness were not apparent in the studies reviewed, despite evidence of elevated illness risks among bathers. A confluence of evidence (via epidemiology, MST, QMRA, etc.) may produce a more comprehensive understanding of the public health risks and inform the development of effective management strategies that adequately protect public health without precluding other health and economic benefits. However, in the context of competing global challenges (United Nations 2024) and finite budgets, relatively few jurisdictions are likely to afford the costs of developing site-specific water quality objectives. As a result, the default of using evidence derived from point-source sewage-impacted waters to regulate such waterways is likely to continue. Meanwhile, a pragmatic option might be to develop a toolbox of validated, site-specific approaches that utilise MST markers. This would improve the robustness and interpretation of sanitary/catchment surveys under varying meteorological conditions. This risk-based, site-specific strategy could support common sense preventative measures – such as closing waterways for given times following well characterised contamination events – thereby contributing to robust recreational water safety plans as advocated by the WHO (2021).

Principles of robust design (such as those covered in the OHAT or other risk of bias tools) should be considered to strengthen epidemiological evidence and reduce the influence of different types of bias in recreational waters that are not impacted by point sources of human sewage. This includes clear and robust conceptual and operational definitions of the exposure and outcome variables. Site-specific factors should be incorporated explicitly into the design of future studies and reported on to facilitate the comparison of findings in future reviews. To address knowledge gaps, future research should focus on the following areas:

• Studies conducted in fresh or estuarine waters.

• Studies conducted outside of the U.S.

• The consideration of different age groups, including children.

• The inclusion/reporting of the time-to-onset of illnesses.

• Characterisation of the health impacts of bathing following rainfall events and other drivers of contamination, to capture a higher variability in microbial concentrations.

• Better characterisation of waterway exposure, including duration and extent exposure for different waterway uses.

• The consideration of sand or equivalent substrates as a reservoir of pathogens.

• The incorporation of emerging technologies to improve the characterisation of exposure (e.g., optical and microfluidic sensors) and health outcomes variables (e.g., biomarkers including those detected by multiplex immunoassays).

The systematic approach (PRISMA) used for this review ensured that it was comprehensive while reducing some biases associated with narrative reviews. Some subjectivity was required in assessing the level of bias of included studies. However, decision criteria were made transparent where possible to enhance the reproducibility of this work within the bounds of the inclusion and exclusion criteria. Several studies were excluded because their inadequate reporting could not rule out point sources of contamination and/or data were not presented in a manner that could address our research questions. Including grey literature and published reports broadened the scope of studies that could be included; however, excluding studies not published in English may have led to the omission of some relevant works.

Conceptualisation: S.K, A.R, H.S, S.T, W.A, C.V, P.F and M.W. PRISMA evaluation: S.K, A.R. and K.G-H. OHAT Risk of Bias evaluations: S.K, A.R. and K.G-H. Writing of original draft preparation: S.K and A.R. Review: S.K, A.R, H.S, S.T, W.A, C.V, P.F, K.G-H and M.W. Editing: S.K and A.R. Supervision: A.R, H.S, S.T, and W.A. All authors have read and agreed to the published version of the manuscript.

The authors would like to thank Prof. Susan Petterson and Seqwater for their support and constructive feedback.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.