Exploring the relationships between small non-community drinking water system characteristics and water system performance in Ontario, Canada

MOE

Ministry of the Environment

MOHLTC

Ministry of Health and Long-Term Care

OR

Odds ratio

PHI

Public health inspector

SDWS

Small non-community drinking water systems

UV

Ultraviolet

Small non-community drinking water systems (SDWSs) provide water to an estimated 15% of the population in Canada (Moffatt & Struck 2011). SDWSs are defined as non-municipal and non-community water systems used by a business or premise to provide drinking water to the public, such as water systems serving rural restaurants, hotels, and campgrounds. In Ontario, responsibility for drinking water regulation is shared among the Ontario Ministry of the Environment (MOE), the Ministry of Health and Long-Term Care (MOHLTC), and local public health agencies (Table 1). SDWSs under the oversight of the MOHLTC are the focus of this study.

A systematic review of SDWSs in Canada and the United States found that 293 outbreaks occurred from 1970–2014, resulting in 41,862 illnesses and three deaths (Pons et al. 2015), and a report by the National Research Council (1997) found water systems in the United States serving fewer than 500 people exceeded microbial and chemical water standards twice as often as larger water systems. SDWSs are required to provide the same water quality as larger systems, but face added challenges (Ontario Drinking Water Advisory Council 2005; United States Environmental Protection Agency 2006; British Columbia Ministry of Health 2013). These include greater difficulty in accessing laboratory services for water sampling in a timely manner (British Columbia Ministry of Health 2013), lack of resources and funding support for improvements to infrastructure (Rupp 2001; Butterfield & Camper 2004; United States Environmental Protection Agency 2006; British Columbia Ministry of Health 2013), inadequate treatment methods (Rupp 2001; Butterfield & Camper 2004; British Columbia Ministry of Health 2013), and poor access to operator training (Boag et al. 2010; Kot et al. 2011; British Columbia Ministry of Health 2013).

Previous studies have been conducted to investigate the characteristics of public water supplies in Ontario (Odoi et al. 2003), and a number of qualitative studies have examined the challenges faced by small community drinking water systems in Canada (Jalba & Hrudey 2006; Boag et al. 2010; Kot et al. 2011); however, there is a need to assess characteristics of SDWSs in Ontario to increase our knowledge and understanding of these systems and to identify further opportunities for interventions to improve the water supply, such as promoting training for targeted high risk SDWSs. In Ontario, training for all small water operators is available through multiple providers including the Ontario MOE and Climate Change and the Walkerton Clean Water Centre (as of the date of this publication) (Walkerton Clean Water Centre 2016). The objectives of this work were to investigate the relationships between characteristics of SDWSs in Ontario, and the performance of these systems with respect to whether an adverse Escherichia coli test had been reported, and whether the system had been rated as a high or medium risk (compared to low risk), according to criteria developed by the MOHLTC.

We obtained de-identified Ontario-wide SDWS data from the MOHLTC on system characteristics and risk assessment ratings. In Ontario, SDWSs are regulated under Ontario Regulation 319/08 Small Drinking Water Systems (Government of Ontario 2013), which outlines the requirements for water treatment, frequency of microbial sampling and testing, and the responsibilities of the owner and operator. Under MOHLTC oversight, which began in 2009, public health inspectors (PHIs) conduct a regular site-specific risk assessment of SDWSs to evaluate the risk of microbial water contamination. The PHI interviews the SDWS operator/owner to determine their competency as a water operator, reviews historical water sample test results (which includes a test for E. coli and total coliforms), and completes a risk assessment instrument (questionnaire) which focuses on the water source, treatment, and distribution system criteria (MOHLTC 2008). The risk assessment tool provides one grade for water source and treatment and a second grade for distribution system criteria. The two grades are combined using a matrix to rank the overall risk level of the water system as high (significant level of risk), medium (moderate level of risk) or low (negligible level of risk) (Table 2). Subsequently, re-assessments are conducted at intervals of two years for high risk systems and four years for low and medium risk systems.

Data obtained from the MOHLTC were collected by PHIs, between January, 2009 and December, 2013, during routine inspections and risk assessments of the SDWSs across Ontario. The database is updated each time a SDWS is inspected, so the data reflected only the most recent inspection and assessment for each system. This data set comprised information on 7,730 SDWSs from 35 of the 36 provincial health units, including the postal code of the system, the risk rating (high, medium or low), any positive E. coli or total coliform test results in the 12 months prior to the last inspection, the primary water source (ground water, surface water, or other), water treatment method (disinfection, filtration, both filtration and disinfection, or no treatment), premise type (provincial park, commercial facility, community centre, club/golf course, hotel/motel/bed and breakfast [B&B], restaurant, trailer park/rental cabins, place of worship, park/campground, other), whether they operated seasonally, if the water operator had taken formal operator training (defined as having completed any basic training or certificate course), and the number of people served by the water supply (however, this latter variable was not used in the analysis as the data were not collected in a consistent manner). Some data were not available for all variables, so analyses were conducted with smaller sample sizes, as noted.

A Pearson chi-square test was used to assess the association between water system characteristics (including: presence of water treatment, water source, having a seasonal water system) and operator training and the outcomes: having a positive E. coli test result in the prior 12 months; and risk rating. Unadjusted odds ratios (ORs) were calculated using Fisher's exact test to assess the relationship between selected sub-categories of dichotomous variables, with statistical significance (P < 0.05). All statistical analyses were conducted with STATA version 12.0 (StataCorp 2011).

Two logistic regression models were developed to explore the relationships between SDWS characteristics and performance. The first model investigated whether water system characteristics were associated with testing positive for E. coli in the 12 months prior to the last inspection. This model included the SDWS postal code as a random effect to take into consideration the effect of location on a positive test result. The second model used crossed random-effects logistic regression (which took into consideration the effect of postal code and health unit in the model) to assess whether water system characteristics were associated with the outcome of the SDWS risk assessment rating (high or medium compared to low risk classification of the water system). High and medium risk categories were combined because few systems (7%; 504/7,196) were classified as high risk. A crossed random effects approach was used for this model because the data set did not have a strictly hierarchal data structure (Raudenbush 1993; Rasbash & Goldstein 1994). That is, SDWSs were cross-classified by local public health agency jurisdiction and postal code (since postal codes do not follow health agency geographic areas, and a single postal code may be located in more than one health agency). The crossed random-effects model included three levels of variation: water system (level 1), postal code (level 2), and local public health agency (level 3).

For the E. coli model, seven predictor variables were assessed for significance in univariable analyses: completion of formal operator training defined as any previous course-based training (yes/no); whether the water system operated year round (yes/no); the water source type; premise type; water treatment type; regional location of health agency jurisdiction (North West, North East, South West, Central West, Central East, Eastern); risk level (high, medium, low); and water source and treatment type categorized into one variable (ground water with or without treatment, and surface water with or without treatment).

For the risk assessment rating model, three predictor variables were assessed for significance in univariable analyses: whether the water system operated year round (yes/no); premise type (same categories as above); and having a positive E. coli test result in the 12 months prior to the last inspection (yes/no). The variables operator training, water source and treatment type were weighted variables in the MOHLTC risk assessment tool, used to assign the risk level to the water supply, and as such were excluded from this analysis. While water quality test results are not explicitly included in this risk assessment tool used, the inspector may manually increase the risk rating when there is an adverse water test. Regional location was not assessed in this model because local public health agency was modelled as a random effect.

Collinearity between variables was assessed using the Spearman correlation coefficient, and if two variables were highly correlated (coefficient greater than |0.8|), only the more biologically plausible one was included in the analysis. The predictors were screened in univariable random-effects logistic regression models and were initially included in the multivariable model if P ≤ 0.20. A manual backwards-selection process was used to build the final model, and statistical significance was set at P ≤ 0.05. All variables were re-evaluated for significance using a likelihood ratio test and assessed for confounding (changes > 20% in the remaining variable coefficients). Two-way interactions were investigated between all covariables in the final model. Best linear unbiased predictors, and Pearson and deviance residuals, were used to assess model fit and identify outliers (Dohoo et al. 2009).

The data set obtained from the MOHLTC included 7,730 SDWSs; 400 (5.1%) of these were removed because they were not given a risk rating and another 134 (1.7%) because an associated postal code could not be found. Therefore, a total of 7,196 SDWSs were included in data analysis.

A surface water supply had higher odds of using treatment when compared to ground water supplies (OR = 10.0; 95% CI 7.1–12.5; P < 0.001). Data pertaining to water system characteristics stratified by premise type are shown in Table 3. The highest numbers of premises with no water system treatment were categorized as ‘place of worship’ (49%; 363/732) or ‘other’ premises (41%; 313/766). The ‘other’ premise type included tourist centres, fire halls, migrant housing, not specified, and roadside garages.

Approximately two-thirds (66%, 4,630/6,985) of water operators reported that they had not received formal drinking water operator training. Almost 66% (3,802/5,763) of system operators utilizing ground water, 69% (644/927) of system operators utilizing surface water, and 62% (184/295) of ‘other’ water system operators, respectively, had not taken formal operator training. Operators of surface water systems were more likely to have reported completing a formal training course than operators of ground water systems (OR = 1.17; 95% CI 1.01–1.36; P = 0.039). The premise types with the two highest proportions of trained operators were provincial parks (75%; 94/126) and community centres (56%; 402/717), while the premise types with the two lowest proportions of trained operators were hotels/motels/B&Bs (20%; 263/1,285) and restaurants (22%; 212/984).

A total of 45% (3,134/6,970) of SDWSs were seasonal water systems. Seasonal water systems had a higher odds of utilizing surface water (81%, 751/927) than ground water (40%, 2,319/5,759) (OR = 0.16; 95% CI 0.13–0.19; P < 0.001). Approximately 32% (1,009/3,134) of operators that oversaw seasonal premises had formal operator training, compared to 35% (1339/3836) of operators with year-round premises; this difference was statistically significant (OR = 0.89; 95% CI 0.81–0.97; P = 0.017). Trailer parks/rental cabins (89%; 692/778) and park/campgrounds (88%; 600/684) were the premise types most commonly reported as seasonal systems.

Water system characteristics stratified by risk assessment rating category are shown in Table 4. Seventy-one percent (368/516) of high risk water systems had no treatment, 13% (162/1214) of medium risk systems had no treatment and 37% (2027/5466) of low risk systems had no treatment. Water systems with no treatment had a higher odds of being ranked as high risk compared to medium risk systems (OR = 16.10; 95% CI 12.47–20.86; P < 0.001), and low risk systems (OR = 4.20; 95% CI 3.45–5.17; P < 0.001). Only 15% (78/511) of high risk water system operators reported participating in formal training, which is low compared to medium and low risk systems (26%; 315/1210; OR = 0.51; 95% CI 0.39–0.67; P < 0.001 and 37%; 1962/5265; OR = 0.3; 95% CI 0.24–0.40; P < 0.001, respectively). A significantly greater proportion of systems in the high risk category were seasonal (58%; 299/513), compared to medium and low risk categories (52%; 629/1211; OR = 1.29; 95% CI 1.04–1.59; P = 0.017 and 42%; 2210/5259; OR = 1.92; 95% CI 1.59–2.32; P < 0.001, respectively).

Summary statistics and univariable associations between having a positive E. coli test result in the 12 months prior to the last inspection and water system characteristics are shown in Table 5. There was no correlation between predictor variables.

Results from the final regression model of having a positive E. coli test result are shown in Table 6. Postal code contributed <0.001% of the total variation in this model. Relative to SDWSs using ground water with no treatment, the odds of having a positive E. coli test result in the previous 12 months was 2.00 times greater for SDWSs using ground water with treatment (P= 0.005), 4.33 times greater for SDWSs using surface water with no treatment (P= 0.052), and 1.97 times greater for SDWSs using surface water with treatment (P= 0.036). None of the other variables were significant in the final model. No issues were identified when assessing model fit and outliers.

Results from the final regression model with the outcome being the risk level of the drinking water system (high or medium vs. low) are shown in Table 7. High correlation was not observed among predictor variables and no issues were identified when assessing model fit and outliers. Local public health agency contributed 21% of the total variation in the model, while postal code contributed 13% of the total variation. Relative to SDWSs that operated year round, the odds of having a water system rated high or medium risk compared to low risk was 1.36 times greater if the water system operated seasonally. Relative to SDWSs that did not have an adverse E. coli test result in the 12 months prior to the last inspection of the water system, the odds of having a water system rated as high or medium risk was 1.66 times greater if systems reported an adverse E. coli test result. Relative to SDWSs associated with a community centre, the odds of having a water system rated high or medium risk was greater for commercial facilities, hotels/motels/B&Bs, parks/campgrounds, restaurants, and trailer parks/rental cabins.

In Ontario, water sampling frequency for SDWSs varies and may be required once every week to once every three months; frequency is determined by the PHI and is based on a number of factors including the history of water sampling results, whether water treatment is used, and if source protection is in place (MOHLTC 2008). Samples that are positive for E. coli are required to be reported to the PHI by both the operator and the laboratory conducting the analysis. The PHI then responds by issuing a boil water advisory, establishing corrective action and resampling the water supply (Government of Ontario 2013). The presence of enteric bacteria such as generic E. coli in drinking water is an indication of faecal contamination (Reynolds et al. 2008). A test for E. coli can also provide a measure of the efficacy of water treatment procedures in place (Payment & Pintar 2006). The use of water treatment to remove or otherwise inactivate microbial pathogens in drinking water has a central role in reducing the incidence of waterborne diseases worldwide and is considered one of the most successful interventions for improving public health (Ngwenya et al. 2013). We found that water systems with any treatment were more likely to have a positive E. coli test result than a ground water system with no treatment, indicating that treatment systems may not be performing as intended in SDWSs in Ontario. We hypothesize that the treatment systems in place were either inadequate for their needs or the treatment systems were not properly maintained and monitored. Conversely, water supplies with traditionally good quality water and no treatment had fewer positive water samples. Previous studies (Rupp 2001; Butterfield & Camper 2004; British Columbia Ministry of Health 2013) noted that operators of SDWSs have trouble identifying appropriate treatment systems for their particular water supply. Ultraviolet light (UV) disinfection is the most common treatment system used by SDWSs; 67% of operators in Ontario reported using UV disinfection for their water systems (Pons et al. 2014). UV disinfection can be effective for eliminating microbiological contamination, but effectiveness is dependent on the chemical, physical, and microbiological qualities of the incoming water (Health Canada 2008). A pre-filter is often needed to reduce turbidity and UV disinfection alone has no residual disinfectant properties that protect the water supply as it travels through the distribution system. Establishing a temporal relationship between positive water samples and treatment in a cross-sectional study such as this is a challenge, since we did not have date of treatment initiation relative to test results. There is a need to further assess whether the relationship between water supply type and the likelihood of an E. coli positive water sample is related to the specific water treatment used, and if there is a need to train operators to ensure they fully understand the strengths and limitations of various treatment systems and can ensure their proper operation.

We found that 66% of SDWS operators had not received formal training, and that the proportion of trained operators varied by premise type, with operators of community centres and provincial parks having the highest rates of formal training, and restaurant owners and hotel/motel/B&Bs having the lowest rates of formally trained operators. While operator training was not a significant variable in the E. coli model in our study, other research has shown that operators play an important role in the safety of the water supply and are needed to maintain and operate treatment systems as well as respond to adverse events (Jalba & Hrudey 2006; Bowman et al. 2009). Our results have valuable implications for targeting training efforts to specific groups of operators, emphasizing the utility of customizing training to the needs of operators responsible for different premise types.

Seasonal SDWSs were more likely to rely on surface water sources than ground water, reported fewer trained operators than year round systems (with differences being small but significant), and were more likely to be rated high or medium risk than low risk. Seasonal water systems often face increased demands for water during specific periods and may have many visitors during short periods of time (Craun & Gunn 1979). Surface water sources are at an increased risk of contaminants entering the water supply through run-off and agricultural and wildlife contamination (Reynolds et al. 2008). Systems that provide water on a seasonal basis should have the same level of ongoing monitoring and care, including water operator knowledge, awareness and diligence in maintaining the supply, as year-round systems. It is important that health agencies provide targeted messaging at the beginning of the season to these operators to raise their awareness of these issues.

The regression model examining water system characteristics by the assigned risk level (high or medium vs. low) showed the odds of being categorized as high or medium risk relative to low risk was greater if the water supply was seasonal, the system had an adverse E. coli test result in the prior 12 months, and if the system was associated with certain premises. The risk rating for a system is given based on the water source, treatment and distribution system. Having an adverse E. coli test result does not automatically increase the risk rating of a water system; however, the PHI may decide to change the risk rating accordingly at their discretion. The findings of this model are valuable in targeting water safety information and training efforts to the operator that will improve safety measures in those systems that tend to have a higher risk rating.

Health unit and postal code variables were modelled as random effects in order to take into account the multi-level structure of the data (Dohoo et al. 2009). The observed health unit effect on the risk assessment model was notable (21%), indicating there is some variation in risk assessment ratings between health unit, likely as a result of differences between PHIs who assign the ratings. These differences could contribute to misclassification bias since risk rankings are assigned by numerous inspectors who may categorize systems differently. This result suggests a need to provide more standardization in the use of the risk assessment instrument across the province; this could be achieved through evaluation of the instrument for reliability and validity, and via additional training in the use of the risk assessment tool for PHIs. There was also a smaller geographical effect with postal code contributing 13% of the variation in this model. Ground water quality is influenced by the hydrogeology of an area, and some areas of Ontario are inherently more susceptible to surface water intrusion with subsequent ground water contamination (Hynds et al. 2014).

This analysis would have been strengthened with further information on the type of water treatment used, a standard measure for the number of people using or having access to the water supply, the date of the adverse test result, having multiple test results and dates, and the months during which seasonal systems operate; unfortunately, such data were unavailable. This information should be considered for collection by PHIs in their future inspections in order to facilitate greater understanding of these factors in the operation and safety of the water supply. Furthermore, monitoring and analysing repeated water quality test results would add more in-depth information on the quality of the water supply. Frequency of testing could impact the likelihood of having a positive test result and was not taken into consideration in this study due to the unavailability of these data.

This research has provided a detailed description of the characteristics of SDWSs in Ontario, Canada. We found that the number of formally-trained water operators was low, as was the number of water systems that used water treatment. Additionally, we found that water treatment, as it was used in SDWSs in Ontario, may not be sufficient to reduce the risk of positive E. coli test results. Furthermore, SDWSs that have had a positive E. coli water result were more likely to be from specific premise types, and those that operated seasonally were more likely to be ranked as high or medium risk than low risk. Hence, this research has highlighted opportunities for targeting water operator training to specific groups of SDWS operators based on the premise type, water source and treatment level in order to increase water protection measures, and future data collection needs in these water systems by PHIs. Further research is needed to fully understand the relationship between adverse E. coli water tests and water treatment, including examining types of water treatment, surrounding land uses, and the distribution system.

The authors declare that they have no competing interests.

The authors thank David Pearl for assistance with the data analysis for this project.