UV treatment and air quality in a pool facility

ACH

Air changes per hour

AM

Arithmetic mean

AR (1)

First order autoregressive

ATD

Automatic thermal desorption

BDCM

Bromodichloromethane

DBCM

Dibromochloromethane

DBP

Disinfection by-products

ML

Maximum likelihood

NCl₃

Trichloramines

OR

Odds ratio (odds of being exposed if you are diseased, divided by the odds of being exposed if you are not diseased)

REML

Restricted maximum likelihood

SD

Standard deviation

TCM

Chloroform

TBM

Bromoform

tTHMs

Sum of the four most common trihalomethanes in swimming facilities (TCM, BDCM, BDCM and TBM)

UV

Ultraviolet

To prevent the growth of hazardous microorganisms, it is mandatory that pool water be disinfected. However, oxidizing biocides, such as chlorine, will react with the contaminants in the water and thereby create a variety of hazardous disinfection by-products (DBPs), some of which are known for their ability to trigger asthma attacks (NCl₃) and for being possibly carcinogenetic to humans (trihalomethanes (THMs), haloacetic acids, and nitrosamines). The types of disinfectant and water may introduce different precursors and DBPs, thereby representing different health effects (Nitter et al. 2017).

In accordance with Norwegian regulations, it is mandatory to control the water concentration of free and combined chlorine continuously if hypochlorite is used to disinfect the pool water (Norwegian Ministry of Health 1996). In most of the approximately 900 Norwegian swimming facilities, the water is treated with UV radiation in addition to hypochlorite. This is mainly due to the UV treatment's well-documented effect of reducing the concentration of combined chlorine in the water and for inactivating microorganisms by absorption (Cassan et al. 2006; World Health Organization 2006; Cossec et al. 2016). Combined chlorine consists of the three DBPs, i.e., mono-, di-, and tri-chloramine (NCl₃), and in previous literature it has been found that all three components can be degraded by UV irradiation (Jing & Blatchley 2009). The most commonly used UV lamp for pool water treatment in Norway is a medium-pressure UV lamp which emits wavelengths between 200 nm and 600 nm.

A significant correlation between NCl₃ and tTHMs in the air, and between tTHMs in the water and air have been found (Cossec et al. 2016). The concentration of THMs in the air is consequently higher compared to those measured in the water (Cossec et al. 2016; Tardif et al. 2016). Although the results are somewhat inconsistent, several researchers have documented different formation potentials of certain DBPs when UV treatment, followed by chlorination, is used in the water treatment system. Some researchers have found that the use of UV treatment in the water treatment system leads to an increase in the concentrations of chloroform (TCM) and bromodichloromethane (BDCM) and other DBPs, in the air and water (Cassan et al. 2006; Cossec et al. 2016). The increase in certain DBPs is explained by the increase in active chlorine and by radicalizing mechanisms initiated by UV irradiation (Cassan et al. 2006). The THMs are not formed directly via exposure to UV treatment, but rather when the water is post-chlorinated after UV treatment. This result is explained by the increased reactivity of organic precursors present in the pool water towards chlorine when the water is treated by UV (Spiliotopoulou et al. 2015). The theoretical effect of UV treatment is found to decrease significantly with increased water turnover time (Soltermann 2015).

Of the combined chloramines, NCl₃ is the component that has received the most attention, as it is suspected of being the main trigger for asthma and other respiratory irritations (Thickett et al. 2002; Jacobs et al. 2007). At 20 °C, NCl₃ is estimated to be 966 and 286 times more volatile than mono- and di-chloramine, respectively (Holzwarth et al. 1984), and when combined chlorine is measured in the air, NCl₃ is the dominant component. In several epidemiological cross-sectional studies, researchers have found an increased odds ratio (OR) amongst swimmers and lifeguards (exposed) compared to non-swimmers (unexposed) for sore throat, chest tightness, and more severe airway irritations (Jacobs et al. 2007; Fornander et al. 2013). NCl₃ is considered the most important cause behind the increased prevalence of symptoms (Hery et al. 1995; Jacobs et al. 2007), although the agreement is somewhat inconsistent (Fornander et al. 2013). The World Health Organization has acknowledged these health effects, and, in 2006, a proposed limit value for exposure to chlorine species, expressed as NCl₃, of 0.5 mg/m³ was published (World Health Organization 2006). No limit value for exposure to NCl₃ in Norwegian indoor swimming pool facilities exists.

No evidence of cancer or increased toxicity has been found in studies in which animals have been exposed to mono- or di-chloramine. The health concern related to exposure to chloramines in humans and animals is increased inhalation toxicity related to exposure to NCl₃ (Hery et al. 1995; Massin et al. 1998). THMs, represented by the four components TCM, BDCM, dibromochloromethane (DBCM), and bromoform (TBM), are characterized by the European Chemical Agency (ECHA) as one of the quantitatively most important groups of DBPs (The European Chemicals Agency 2017). TCM and BDCM are classified as possibly carcinogenetic to humans by the International Agency for Research on Cancer (World Health Organization 2017). Long-term exposure to THMs is associated with increases in the prevalence of bladder cancer and stillbirth (Villanueva et al. 2006; Rivera-Núñez et al. 2018). Both THMs and NCl₃ are extremely volatile and present in higher concentrations above the water surface compared to in the pool water (Jacobs et al. 2007). From animal studies, the liver is found to be the most likely target organ for DBP exposure (Li et al. 2015).

UV treatment is used in the water treatment system to reduce the concentration of combined chlorine in the water. However, it is shown that use of UV treatment does not affect the concentration of NCl₃ considerably in the water when the turnover time is several hours (Soltermann 2015). It is therefore hypothesized that this applies also for the air and that use of UV treatment will increase overall exposure of airways.

The aim of this work is to understand the impact on the concentrations of THMs and NCl₃ in the air when a medium-pressure UV lamp is used in the water circulation system. In addition, a linear mixed effect model will be applied to identify which determinants contribute the most to the observed variability in the two exposure variables and how much of the variability can be attributed to the UV treatment of the water circulation system.

The subject of this study was a poolroom with one therapy pool. To maintain the concentration set points of free and combined chlorine, the pool water is disinfected using sodium hypochlorite (NaOCl) in addition to a medium-pressure UV lamp with an intensity of 75 mJ/cm². When the UV lamp was switched on, it was always at 100% power. The pH value is adjusted using H₂SO₄, and the turnover time for the pool water is approximately 6 hours. The pool water flow rate is approximately 35 m³/h. No water is taken out of the swimming pool, but water lost due to evaporation is replaced by fresh water. All of the water is filtered through sand filters, and about 15–20% is also filtered through charcoal. Backwashing is performed every Thursday morning. The total air and water volumes are approximately 4,300 m³ and 210 m³, respectively. The average number of air changes per hour (ACH) is 10, and between 40% and 60% (4–6 ACH) of the air involved in the ACH is fresh air from outside. The air is supplied from the floor, then flows up along the window façade on one side of the poolroom, and then is extracted via an exhaust grill (4 m²), which is located on one of the walls diagonal to the supply duct and approximately 2 m away from the pool border. The study object is used mostly for arranged water activities, such as swimming for babies and water aerobics, and, from Monday to Friday, most hours are booked from 09.00 to 20.00. The swimming pool is also used for public swimming on the weekends.

In total, 162 and 36 stationary air samples of THMs and NCl₃, respectively, were collected on Tuesdays and Thursdays for 5 weeks. The UV treatment was switched off for 2 weeks, turned on for 2 weeks, and then switched off again for 1 week. In the final week, samples were collected on Tuesday and Thursday from 09.00 to 12.00, while, for the first 4 weeks, samples were collected on Tuesdays and Thursdays from 09.00 to 12.00 and then again from 13.00 to 16.00.

Air samples of THMs and NCl₃ were collected from the poolside (location 1) and exhaust grill (location 2) simultaneously using two stationary test stands. Three air samples of THMs (two from sampling location 1 (0.30 m above the floor and 0.05 m above the water's surface) and one from sampling location 2 (0.30 m above the floor)) were collected simultaneously each hour. Two samples of NCl₃ (one from sample location 1 and one from sample location 2) were collected every third hour. Additional air samples of THMs were also collected in the supply duct on two different days to document the amount of THMs being recycled back into the poolroom with the recirculated air.

To collect information about NCl₃ in the air, a total of 180 L was sampled onto glass fibre filters impregnated with sodium carbonate and diarsenic trioxide. The filters were attached to an air pump (SKC Air sampler) calibrated to deliver 1 L/min for 180 minutes, in accordance with previously published methods. Using this method, information about mono- and dichloramine, as well as other oxidant forms of chloride, is also measured. However, NCl₃ accounts for around 90% (Hery et al. 1995; Massin et al. 1998). The air flow through the filter was verified four times during the sampling period of 180 min. In total, four NCl₃ samples were collected each sampling day. All of the NCl₃ samples were prepared by and sent to the accredited laboratory in the Department of Occupational and Environmental Medicine at Umeå University Hospital, Sweden, for analysis.

The sampling, analysis, and quality assurance for collecting samples of THMs in the air are based on the published methods US EPA TO-17 (United States Environmental Protection Agency 1999) and the ISO 16017 (International Organization for Standardization 2000). The samples were obtained by pulling air onto automatic thermal desorption (ATD) tubes containing 200 mg of Tenax TA 35/60 (Markes Int.) using three low-flow pumps (Markes Int.) with an average flow rate of 40 mL/min for 20 min. The low-flow pumps were calibrated in situ, both before and after each sample. Before and after sampling, the ATD tubes were stored in an airtight container with charcoal. Each ATD tube was sealed using Swagelok caps combined with PTFE ferrules and packed in uncoated aluminium foil to avoid contamination and losses. The analysis setup, as well as calibration methods, are explained by the authors elsewhere (Nitter et al. 2017). The analysis of THMs was performed at the Department of Industrial Economy and Technology Management at Norwegian University of Science and Technology (NTNU).

Information on sample type, number, height and location, water activity and number of bathers were recorded during the sampling days. The concentration of free and combined chlorine, the pH value and the water temperature are automatically measured at intervals of 120 s and the values were collected from the supervisory control and data acquisition system. Control samples to verify the concentration of free and combined chlorine were collected and analysed by the technical staff, three times a day, using the N,N-diethyl-p-phenylenediamine method. The air temperature, relative humidity, and outside temperature were recorded continuously using three EasyLog USB data loggers. One EasyLog was placed outside the window of the pool facility and the two others were attached to the two test stands. The ventilation log was also collected from the supervisory control and data acquisition system after each day of sampling.

Air samples were interpreted and analysed with the data program Statistical Package for the Social Sciences 25.0 (SPSS). Descriptive data, such as the arithmetic mean (AM), standard deviation (SD), and lowest and highest values obtained (min, max), for the different variables are presented in Table 1. Both exposure variables were positively skewed and log-transformed before statistical analysis. A t-test was applied to compare the different chemical–physical parameters measured when the UV treatment was switched off and on.

A linear mixed effect model was applied to identify how much of the variability in the log-transformed tTHMs and NCl₃ could be attributed to the significant determinants of exposure. The advantages of a linear mixed effect model include the ability to account for the correlation between the repeated observations and to take effects that unfold during the experiment into account (Baayen et al. 2008). The concentration patterns for the two sampling locations were identical, meaning that the concentrations measured at location 2 were highly dependent on the concentrations measured at location 1. By default, the subjects are assumed to be independent and therefore the dependency between the two sampling locations was accounted for by taking the average of both sampling locations, leaving only one sample location for the analysis. To account for the correlation between the repeated samples, first order autoregressive (AR (1)) was used for the covariance structure. This structure assumes the correlation function decays exponentially as the intervals between the measurements increase.

One model was built stepwise and separately for each of the two exposure variables (ln tTHMs and ln NCl₃). The determinants of exposure were treated as fixed effects, and only those that improved the fit of the model significantly, as judged by the likelihood ratio test (p > 0.05), were kept in the final model. While the model was built, estimates were made using the method of maximum likelihood (ML). In the final model, the method of restricted maximum likelihood (REML) was used to estimate the variance component, considering that REML yields less biased estimates of the variance components compared to ML.

The AMs, SDs, and the lowest and highest values observed for the different physical–chemical parameters are listed in Table 1. Water temperature, air temperature, pH value, number of bathers, free chlorine, and relative humidity had very low variability. However, with the exception of TBM and the number of bathers, all variables varied significantly between when the UV treatment was switched off and on. No values exceeded the limit values established in the Norwegian regulations for swimming pools and spas (Norwegian Ministry of Health 1996).

Slightly higher concentrations of tTHMs (14%) and NCl₃ (10%) were observed by the poolside (location 1) compared to in the extract channel (location 2), and the contamination level observed in the two sampling locations was statistically significant (p = 0.02). On average, 2% difference in tTHMs was observed between the two heights, 0.05 m (above the water surface) and 0.30 m (above the floor), by the poolside, and this result was statistically insignificant (p = 0.66). Based on this result, only samples collected 0.30 m above the floor at location 1 were used in the linear mixed effect model.

The results show that, when the UV treatment was on, the concentrations of tTHMs, TCM, DBCM, and BDCM increased by 37%, 41%, 51% and 68%, respectively, compared to when the UV treatment was switched off. The concentrations of NCl₃ and TBM decreased by 15% and 12%, respectively.

Air samples of tTHMs collected from the outlet of the supply duct on two different days showed that between 42% (afternoon) and 56% (morning) of the contaminants extracted from the poolroom are recirculated back into the poolroom. Although only two samples were collected from the supply duct during the sampling period, the ventilation log showed that the air changes per hour and the fresh air supply were approximately constant throughout the sampling period.

The Shapiro–Wilk test for normality showed that the log-transformed exposure variables of tTHMs, TCM, DBCM, BDCM, and NCl₃, show no significant deviation from normality. TBM, however, did not seem to follow a normal distribution. In Table 2, the Pearson correlation coefficients between the log-transformed tTHMs and NCl₃ are shown for when the UV treatment was switched off and on. However, the Pearson correlation coefficient is suspected to be sensitive to non-normality and thus was not used to explain the correlations between TBM and NCl₃. As shown in Table 2, the correlations between airborne levels of NCl₃ and tTHMs, BDCM, and TCM were stronger when the UV treatment was switched on compared to when the UV treatment was switched off.

Although the contamination levels observed at the two sampling locations were significantly different from each other, the between sampling location variability was low with respect to both ln tTHMs (0.0057) and ln NCl₃ (0.0034). This suggests that the room is well mixed and that the concentration obtained at sample location 2 is highly dependent on the contamination level obtained at sampling location 1. As shown in Tables 3 and 4, the only factors contributing significantly to the fit of the model, judging by the likelihood ratio test, were bathers and UV treatment.

When UV treatment was added to the ln tTHMs model, the variability was reduced by 29.5%, and the fit of the model was improved significantly (p = 0.000) compared to the model in which no factors were included. As shown in Table 3, approximately 14% of the variability observed in ln tTHMs could be attributed to the bathers in the pool.

When UV treatment was added as determinant in the ln NCl₃ model, this factor did not significantly improve the fit of the model (p = 0.06), and UV treatment as fixed factor was insignificant with a p-value of 0.09. However, when the variable bathers was added to the model first, UV treatment became significant, and both variables together significantly improved the fit of the model, judging by the likelihood ratio test.

Approximately 29% of the variability observed in ln NCl₃ could be attributed to the bathers in the pool. When UV treatment was added to the model, the variability increased by 9%. In general, the observed concentration of NCl₃ varied more when the UV treatment was switched off compared to when the UV treatment was switched on. This pattern was not observed for tTHMs.

As expected, higher concentrations of tTHMs (37%) were measured in the air when the UV treatment was switched on. This result is in accordance with previous studies (Cossec et al. 2016). However, in some studies, the results show that, even though lower concentrations of combined chlorine were measured in the water of pools disinfected using UV treatment and chlorine, no significant effect was observed in the level of NCl₃ in the air between pools with and without UV treatment (Gérardin et al. 2005; Cossec et al. 2016).

In our study, when the UV lamp was switched on, the concentration of combined chlorine in the water decreased by 58% (the ratio of off vs on is 2.35), but the concentration of combined chlorine in the air only decreased by 15.2% (the ratio of off vs on is 1.18). The formation of NCl₃ in the pool water is considered to be a rather fast process (Soltermann 2015), and, due to the high volatility of NCl₃ and a turnover time of 6 hours, this component is likely to transport from the water to the air before the water is treated through the water treatment system.

The correlation between tTHMs and NCl₃ was stronger when the UV treatment was switched on (r = 0.936, p = 0.01) compared to when the UV treatment was switched off (r = 0.472, p = 0.05). This finding suggests that NCl₃ and tTHMs can be used to predict the contamination level of one another when the UV treatment is switched on. The within and between variabilities of sampling locations also suggest that the contamination level in this specific poolroom, with an ACH of 10 and a fresh air supply of between 40% and 60%, is well mixed. Even though the chosen ventilation strategy does provide mixing of the contaminants in the poolroom, the samples collected from the supply channel show that the potential for reducing the contamination level is between 42% and 56% if filters able to adsorb gasses in the air are installed in the return air channel of the air-handling unit.

Approximately 30% and 14% of the variability observed in ln tTHMs could be attributed to UV treatment and bathers, respectively, and the difference in observed concentration of tTHMs when the UV treatment was switched off and on was highly significant (p = 0.003). For the NCl₃ model, most of the variability observed could be attributed to the number of bathers in the pool (approximately 29%). When UV treatment was added to this model, the variability increased by 9%, and UV treatment became a significant determinant only when the variable bathers was added to the model first. Considering that NCl₃ is more volatile and less water soluble in comparison to the four THMs, this fact might explain why NCl₃ is affected more by the bather load than tTHMs.

The main advantage of a linear mixed effect model is to take effects that unfold during the experimental period into account (Baayen et al. 2008). This advantage is especially important for exploratory studies in which the effects of different implements are to be determined and effects are most likely to unfold during the experimental period, such as bather load, chlorine concentration in the water, and outdoor temperature. The importance of a linear mixed effect model is also highlighted by the dependency and large correlation estimated between the repeated samples.

A significant dose-dependent increase in the prevalence of red eyes, itchy eyes, runny nose, loss of voice, and significant decrease in forced expiratory volume have been found in pool workers exposed to pool air (Thickett et al. 2002; Fantuzzi et al. 2012). However, in these studies, the only observed contaminant in the air was NCl₃. These studies are also based on a very limited number of samples, which makes the estimated dose-response relationship highly questionable. In some studies conducted in Sweden, the researchers have also found a relationship between exposure in pool facilities and increased prevalence of respiratory illnesses; however, this was not described by the measured level of NCl₃ in the air (Fornander et al. 2013). It is therefore questioned whether one should focus on a single component knowing that several of the components in pool facilities might interact with each other. The increased prevalence of certain diseases is acknowledged to be caused by long-term exposure in the poolroom. Based on the study design in previous studies, the specific dose-response relationship between exposure to NCl₃ and health effects is somewhat questionable. However, the effort should be expended on reducing the overall exposure level. To reduce the level of combined chlorine in the water, other methods, such as increased turnover time and fresh water supply, reduced concentration of free chlorine or other filtration methods, could be tested and considered, if microbiological and hygienic water quality is maintained.

UV treatment in the water system reduces the concentration of combined chlorine in the water but has only a limited effect on the level of NCl₃ in the air. However, UV treatment also increases the concentration of other DBPs, such as THMs, of which we have little knowledge. Combined chlorine is controlled continuously in the water of Norwegian swimming pools, and hazard controls are implemented immediately if the level becomes unacceptably high. To prevent increased formation of other DBPs, combined chlorine in the water should be controlled using methods besides UV treatment, such as reduction/removal of DBP precursors, increased fresh water supply, increased turnover time, adsorbents in the ventilation return-air channel, or improved bather hygiene. The limit value for combined and free chlorine in the water could also be reduced as long as doing so does not affect the hygienic parameters.

This project is founded and supported by the Departments of Civil and Environmental Engineering at NTNU.