Burden of recreational water illness due to exposure to cyanobacteria and their toxins in freshwater beaches in Canada: protocol of a prospective cohort study

Study setting

This study will take place at four targeted freshwater beaches in Ontario, Manitoba and Nova Scotia (figure 1). These provinces have seen significant increases in reported cyanobacterial blooms in recent years.3–5 Locations were selected in each province that have a history of cyanobacterial blooms and are at risk of increased cyanobacterial blooms in the future. They are also popular destinations for various recreational water activities (eg, swimming, boating and kayaking) during the summer season. In Manitoba, the study will take place at Grand Beach West and East, located on the southeastern shore of Lake Winnipeg. In Ontario, the study beaches include Colchester Beach in Windsor-Essex Region, located on the north shore of Lake Erie, and Kinsmen Beach in Port Perry, located on the southwestern shore of Lake Scugog in Durham Region. In Nova Scotia, the study will take place at Shubenacadie Canal, a waterway connecting two lakes in Halifax. The diversity of these locations and water bodies provides a cross-section of exposure scenarios that will provide important insights for other, similar Canadian freshwater settings.

Figure 1

Locations of the four targeted beach sites for this prospective cohort study, to be conducted in the summers of 2024 and 2025, Canada.

Study design and population

This study will follow a prospective cohort study design, adapted from prior studies and the methodology used for the U.S. National Epidemiological and Environmental Assessment of Recreational water study.17 18 21–23 Throughout the summers of 2024 and 2025, trained data collectors will visit and recruit recreational water users at each beach. Recreational water users will be surveyed at two sites per year. Recruitment will be conducted approximately two times per week over a 12-week period at each beach, for 24 total recruitment days per beach. The population of interest includes recreational water users of any age, and any pet dogs also present at the beach. Eligibility for participation in the study will include (1) ability to provide informed consent; (2) ability to complete the surveys in English or French; and (3) must not have participated in the study in the past 21 days. Given that the outcomes of interest are acute and self-limiting, individuals will be able to participate more than once following a washout period of 21 days.7 8

Participant enrolment and survey process

Data collectors will set up a study-branded table near the main beach access point and approach recreational water users throughout each recruitment day, prioritising families with children and those with a dog present when possible. Household members will be recruited and surveyed together, with parents or guardians providing informed consent for themselves and children that do not have decision-making capacity to provide informed consent.24 Assent will be sought from children and youth. Eligible and interested participants will be asked to visit the study table at the end of their beach visit to complete the survey and receive an incentive. A $C10 gift card will be provided to all participating households.

The survey will be completed by one member of each household. It will contain two sequential components: (1) beach survey and (2) follow-up survey. The beach survey will be conducted at the end of participants’ beach visit and assess sociodemographic characteristics, chronic health conditions, baseline illnesses and beach exposures and activities conducted (eg, types of water contact, amount of time in the water). The beach surveys will be completed using a web-based survey platform (SimpleSurvey, Quebec, Canada) on study tablet devices. The follow-up survey will be conducted 3 days after the beach survey and determine possible acute illness outcomes experienced by participants, including dogs, since their beach visit.8 Participants will receive an email or text-message prompt with a link to complete the follow-up survey online. Telephone follow-up will also be an option for participants who prefer that method. A 3-day follow-up was selected based on the time to illness onsets reported in harmful algal bloom outbreaks in the USA.8 Surveys will be available for completion in English or French (see English version in online supplemental materials). The questionnaires were reviewed for clarity using a cognitive interviewing approach with 10 individuals.25 The questionnaires are also adapted from versions that were pilot tested in Toronto in 2022.26

Anticipated sample size and precision

Data collectors will approach and aim to recruit as many recreational water users as possible each recruitment day. Based on data from previous and ongoing recreational water illness cohort studies conducted in Toronto and prior U.S. cohort studies,21 26–28 anticipated recruitment is ~15 households per day at the study beaches, with ~1.7 individual participants per household. With 24 recruitment days per beach, this corresponds to a total anticipated sample size of ~2500 recreational water users (~1500 households), or ~625 participants per beach. Assuming approximately 40% of households has a dog,29 the aim is to include data on ~600 pet dogs.

A Bayesian precision analysis was conducted using simulated data with expected characteristics of cyanobacteria levels, water contact levels and acute gastrointestinal illness incidence based on results reported in prior studies.17 19 20 26 The simulations assumed a conservative attrition proportion of 30%, a negative binomial distribution of cyanobacterial cell counts (mean=5000 cells/mL, dispersion parameter=0.5), and an increasing incidence of illness for each quartile increase in cyanobacterial cell counts. Under these conditions, and across 500 simulations, this study is estimated to have an average posterior probability of at least 0.87 to detect a positive association with cyanobacterial levels among participants that report full body immersion in beach water, with an average 95% credible interval width of 0.45 on the log odds scale. Full details of the simulation conditions and parameters are provided in the online supplemental materials.

Exposures and outcomes of interest

The exposure of interest is the cyanobacteria levels in beach water among participants who report water contact in the beach survey. Water contact will be defined across three levels of exposure: (1) no contact (control group); (2) minimal contact; and (3) full body immersion.27 28 30 31 Minimal contact is defined as water contact that does not result in full body immersion (eg, wading below one’s waist, boating, kayaking and canoeing), while full-body immersion is defined as entering above one’s waist (eg, swimming). The primary measure of cyanobacteria exposure will be cell counts (cells/mL) as determined from microscopy.6 32 33 Two water samples will be taken from each beach on each sample day and tested for cyanobacterial cell counts, with the average value used as the primary exposure measure. Water samples will also be tested for the presence of total microcystins using a two-tiered approach: first the samples will be screened for presence/absence of the toxins in the field on each recruitment day using rapid immunochromatographic strip tests.32 34 If microcystin presence is detected, an additional water sample will be taken and sent for analysis to quantify toxin levels using an indirect ELISA test.32–34

Handheld fluorometers will be used to measure and evaluate pigments as potential indicators of cyanobacterial blooms. Specifically, the fluorometers will quantify the levels of chlorophyll-a and phycocyanin in the beach water during each recruitment day.6 32 Additionally, environmental factors will be recorded from the nearest Environment and Climate Change Canada weather stations and water buoys that could also be used to predict possible cyanobacterial presence, including air and water temperature, 48-hour rainfall, wind speed and direction, and water levels.35

Five categories of acute, self-reported illness outcomes in human participants will be measured that reflect a range of possible health effects from contact with cyanobacterial blooms and cyanotoxins: gastrointestinal illness, respiratory illness, skin infections, eye infections and generalised symptoms.6 8 15 Gastrointestinal illness is defined as one or more of (1) diarrhoea (≥3 loose stools in 24 hours); (2) vomiting; (3) nausea with stomach cramps; or (4) nausea or stomach cramps that interfere with regular daily activities (eg, missed work or school).23 27 36 Other outcomes are defined as follows: acute respiratory illness (sore throat, runny or congested nose, difficulty breathing, or cough); skin infection (rash or itchy skin); eye infection or irritation; and generalised illness (headache, fever, fatigue, dizziness, loss of appetite, chills, muscle pain or general pain).8 15 23 36 37 For dogs, illness outcomes will include gastrointestinal illness (diarrhoea or vomiting); skin infection (rash or itching); generalised illness (lethargy, loss of appetite or weakness); and neurological illness (seizures/convulsions, loss of balance, behaviour change or paralysis).7 8 38 Data on various measures of illness severity (eg, medical or veterinary consultations, hospital visits, medication use and hospitalisation) will also be collected.

Confounding variables

Several preidentified confounding variables will be measured that could be related to the exposure and outcomes of interest. These variables will be adjusted in analyses to determine unbiased causal estimates. Two directed acyclic graphs (DAG) have been constructed to determine the minimal sufficient adjustment sets for the exposure outcome of interest for both the human and dog participants.39 40 The human participant DAG is shown in figure 2, with the DAG for dog participants shown in the online supplemental materials. The human participant-level confounders that require adjustment include age, gender, ethnoracial identity, education level, pre-existing or chronic conditions, other beach exposures before or after the beach survey, sand contact and consumption of food on the beach. Additionally, the DAGs include adjustment for beach location, year and beach water faecal indicator bacteria (Escherichia coli) levels. The latter are routinely collected by public health and environmental authorities at the study sites, and analysed using culture-based membrane filtration or comparable methods.41 42 They are typically reported as geometric mean concentrations of multiple samples per beach in CFU/mL. For dogs, age, sex, neuter/spay status, pre-existing or chronic conditions, beach sand or debris contact and other water contact before or after the beach survey will be adjusted as confounders along with owner socioeconomic status variables, beach location, year and water E. coli levels.

Figure 2

Directed acyclic graph of the minimal adjustment set of confounding variables for the relationship between cyanobacteria levels among human participants who report any water contact and acute illness outcomes. The water contact and cyanobacteria level variables are the exposures of interest, while all variables with white circles require adjustment in the analysis to determine unbiased estimates of effect.

Data analysis plan

A series of multilevel logistic regression models will be constructed to determine the relationship between cyanobacteria measures and cyanotoxin levels with the level of water contact and the incidence of each acute health outcome among human participants and pet dogs. The models will be constructed under a Bayesian framework, which accounts for uncertainty in all parameter estimates through prior and posterior distributions.43–45 Each of the illness outcome categories will be assessed as a response variable in a separate model. The models will account for the multilevel structure of participants being clustered within households and households clustered by recruitment date; these variables will be modelled as varying intercept effects.43 45 Additionally, beach location will be included as a varying intercept or varying slope. All models will include as covariates the minimal adjustment set of confounding variables as described above.

The primary exposure variables of interest will be the interaction between the cyanobacteria levels and the level of water contact variable. Separate models will be constructed for each outcome of interest to evaluate and compare associations with different measures of cyanobacteria exposure, including cell counts, total microcystin levels, chlorophyll-a and phycocyanin. The association between cyanobacteria measures and illness outcomes among participants that report no water contact (ie, control participants) will serve as a negative control analysis.46 47 This analysis will identify possible residual confounding or differential outcome reporting bias, as the self-reported health outcomes are unlikely to be associated with cyanobacterial measures among participants that do have any level of water contact.27 47

Differences in illness risks by participant gender and age will be examined by evaluating their influence as index variables or effect modifiers (ie, interaction terms) with the level of water contact. Prior research indicates that boys are more likely to swallow beach water and spend more time in the water than girls.26 48 The following recommended categories will be used for the gender variable: boy/man, girl/woman, transgender and gender fluid.49 Given that children are at highest risk of recreational water illness, risk differences will be examined using the following age groups (0–4, 5–9, 10–14, 15–19 and 20+).22 27 31 48 50 51 For dog models, dog age and sex will be modelled similarly.

Weakly informative prior probability distributions (priors) for model parameters will be determined to ensure that the model is sceptical of highly unlikely and improbably parameter values.43 44 The appropriateness of priors will be assessed via prior predictive checking, and their impact on each model will be assessed via sensitivity analysis.52 Causal effects will be determined by examining and contrasting posterior probability distributions of parameters and their summary measures (mean and 95% credible intervals).43–45 Alternative models and parameter specifications will be evaluated and compared using leave-one-out cross validation, while fit and convergence of models will be evaluated by examining sampling chains, R-hat values and posterior prediction checks.52 53 Marginal effects plots will also be created to visualise the effects of each predictor of interest, as well as interactions, on the predicted probability of each outcome.54 Marginal effects will be calculated to examine the relationship at different cyanobacteria exposure cut-points, including federal health threshold values and quartiles.54 All analyses will be conducted in R statistical software using the RStudio interface.55 56 Bayesian regression analyses will be conducted using the ‘brms’ package, which implements models using the probabilistic programming platform Stan and estimates parameters using Hamiltonian Monte Carlo sampling.45 57 58

Sensitivity analyses

A series of sensitivity analyses will be conducted using alternative specifications of the exposure and outcome variables. This will include separately evaluating the level of water contact as yes/no binary variables of different definitions (any water contact, body immersion or swallowing water). Different specifications of the outcomes will also be assessed, including any illnesses reported and specific severity indicators (eg, medical or veterinary consultations). The influence of various environmental and weather variables on the exposure-outcome relationships will be examined in separate models. Separate models will also be evaluated for each exposure-outcome relationship to evaluate the influence and impact of different prior specifications for model parameters.

Study limitations and mitigation strategies

One possible limitation is low recruitment. However, several previous beach cohort studies with a similar methodology have achieved very high household participation rates (~65% to 90%),23 27 28 36 59 suggesting recruitment should be feasible. Attrition for the follow-up survey is another possible concern, but numerous measures have been taken to mitigate this issue, including offering a gift card incentive to all households, sending repeated reminders by multiple modes (email and text messages) and having a shorter follow-up period (3 days) than prior studies that focused on faecal contamination as the exposure of interest. The shorter follow-up time is possible in this study given the much shorter time-to-illness onset for cyanobacteria and toxin exposure compared with incubation periods of pathogens that cause other types of recreational water illness.8 38 Prior beach cohort studies have also found that those who were lost to follow-up had similar characteristics to those who completed all surveys,21 22 28 37 suggesting that anticipated attrition should not affect causal estimates and inferences. Logistical concerns could include poor weather days (eg, heavy rainfall) which would lower recruitment; the planned approach will be flexible and adaptable to such conditions, including rescheduling as necessary and targeting the busiest days (eg, weekends and holidays).

Recall bias could be a concern for follow-up measures. However, prior beach cohort studies have found that associations were robust to this possible bias,22 23 28 36 and as noted above, this study includes a shorter follow-up time (3 days) than these prior studies. The study illness outcomes are non-specific, symptom-based and self-reported, which could introduce bias. However, this is the most pragmatic approach to assess a wide range of possible outcomes that could be associated with cyanobacteria and cyanotoxin exposure.8 Another potential limitation is that detection of cyanobacterial cell counts or total microcystin levels above federal health threshold values for recreational water (50 000 cells/mL and 10 µg/L, respectively) could trigger local public health advisories or closures at the affected beaches, and would likely be visibly noticeable to users, which would lower participant exposures during the likely highest risk period.6 20 60 However, this study will still be valuable to assess possible risks of illness during the period leading up to active cyanobacterial blooms, including lower-to-moderate and background levels of exposure.