Role of filtration in managing the risk from Cryptosporidium in commercial swimming pools – a review

Most commercial swimming pools use pressurised filters, typically containing sand media, to remove suspended solids as part of the water treatment process designed to keep water attractive, clean and safe. The accidental release of faecal material by bathers presents a poorly quantified risk to the safety of swimmers using the pool. The water treatment process usually includes a combination of maintaining a residual concentration of an appropriate biocide in the pool together with filtration to physically remove particles, including microbial pathogens, from the water. However, there is uncertainty about the effectiveness of treatment processes in removing all pathogens, and there has been growing concern about the number of reported outbreaks of the gastrointestinal disease cryptosporidiosis, caused by the chlorine-resistant protozoan parasite Cryptosporidium. A number of interacting issues influence the effectiveness of filtration for the removal of Cryptosporidium oocysts from swimming pools. This review explains the mechanisms by which filters remove particles of different sizes (including oocyst-sized particles, typically 4–6 μm), factors that affect the efficiency of particle removal (such as filtration velocity), current recommended management practices, and identifies further work to support the development of a risk-based management approach for the management of waterborne disease outbreaks from swimming pools. This is an Open Access article distributed under the terms of the Creative Commons Attribution Licence (CC BY 4.0), which permits copying, adaptation and redistribution, provided the original work is properly cited (http://creativecommons.org/licenses/by/4.0/). doi: 10.2166/wh.2019.270 ://iwaponline.com/jwh/article-pdf/17/3/357/639143/jwh0170357.pdf Martin Wood (corresponding author) Lester Simmonds Pool Sentry Ltd, Dale Cottage, Stanton Dale, Ashbourne DE6 2BX, UK E-mail: martin@poolsentry.co.uk Jitka MacAdam Francis Hassard Peter Jarvis Cranfield University, Cranfield, Bedfordshire MK43 0AL, UK Rachel M. Chalmers Cryptosporidium Reference Unit, Public Health Wales, Microbiology and Health Protection, Singleton Hospital, Swansea, SA2 8QA, UK This article has been made Open Access thanks to the generous support of a global network of libraries as part of the Knowledge Unlatched Select initiative.


INTRODUCTION
The source water which is used in swimming pools is usually of drinking water quality as it first enters the pool; thereafter it is sullied by bathers, either overtly as accidental releases of urine and faecal material, or more subtly due to ineffective showering practices, which can adversely affect the overall pool water quality (Ryan et al. ). This presents a risk to the health and safety of swimmers using the pool, which is mitigated through the implementation of pool water treatment processes that aim to return and continually maintain the water to a quality standard that is acceptable for bathing use. There has been growing concern about the increasing numbers of reported outbreaks of the gastrointestinal disease cryptosporidiosis, caused by the protozoan parasite Cryptosporidium. In 1988, the first reported outbreaks of cryptosporidiosis linked to a swimming pool occurred in the USA (Sorvillo et al. ) and in the UK (Joce et al. ). Between 1992 and 2011, there were 85 outbreaks of cryptosporidiosis linked to swimming pools in England and Wales (Chalmers ). However, this is likely to be an underestimate because the cryptosporidiosis outbreaks may not be identified or reported (Ryan et al. ).
The recirculating water treatment process in swimming pools typically includes a combination of filtration to physically remove particles from the water and disinfection by maintaining a residual concentration of an appropriate biocide, e.g. chlorine, in the pool water to kill or inhibit the growth of microorganisms. The pressurised filters remove suspended solids to maintain water clarity acceptable to both lifeguards and bathers, as well as microbiological safety of pool water. Pool filters normally contain sand, with typical particle size in the 600-800 μm diameter range, though alternatives are used, e.g. crushed recycled glass (Rutledge & Gagnon ). Some pools will have non-residual secondary disinfection in place such as ultraviolet (UV) light treatment or ozone dosing.
Most swimming pool treatment processes are not specifically designed and optimised for the removal of Cryptosporidium-sized particles; the main objective for filtration of pool water has traditionally been to clarify the water and to ensure that lifeguards can see across the whole of the bottom of the pool. Furthermore, due to the extended persistence and infectivity of the Cryptosporidium oocysts, it is thought that swimming pools could be a sink of Cryptosporidium and act as a vector for disease outbreaks.
Cryptosporidium is a single-celled protozoan parasite that infects the gut and can cause gastroenteritis, characterised by watery diarrhoea, abdominal pain, nausea and vomiting, and low-grade fever. It is transmitted between hosts (humans or animals) by the oocyst stage shed in faeces. The oocysts of human-infective species are typically 4-6 μm in diameter and are largely resistant to disinfection by conventional pool water biocides such as free chlorine (Chalmers & Davies ); the C t value (concentration of disinfectant multiplied by exposure time) for a 3 log 10 reduction in oocyst viability reported for free chlorine at pH 7.5 and at 25 C is up to 15,300 mg/L min (Shields et al. ). This corresponds to a disinfection time of 10.6 days in pool water containing 1 mg L À1 free chlorine (Chalmers et al. ) and confirms that normal free chlorine dosage provides no practical residual disinfection for Cryptosporidium oocysts in a swimming pool.
A single accidental faecal release (AFR) in a 450 m 3 municipal pool, if well-mixed, could result in an average concentration of around 20,000 oocysts L À1 (20 mL À1 ) (Gregory ). Dufour et al. () estimated that the average amount of water swallowed during a 45 min pool session was 37 mL for non-adults and 16 mL for adults. If a pool user swallowed only 10 mL water containing 20 oocysts mL À1 , this would lead to ingestion of 200 oocysts. This is well above the reported infective dose (<100 oocysts) (Ryan et al. ), and dose-response modelling has shown a chance of infection from ingestion of just a single oocyst (Messner et al. ). This could, therefore, pose a major risk to the public health of swimmers using a pool.
In most swimming pools, the principal protection against Cryptosporidium is removal through filtration (WHO Data from the drinking water treatment industry indicate that, in combination with effective coagulation/ flocculation and sedimentation, sand filters can achieve a 1.5-3 log 10 removal of oocysts (LeChevallier et al. ; Gregory ; Betancourt & Rose ). This equates to 97.2% and 99.9% removal of oocysts from water in a single pass through the filter (hereafter referred to as the filter efficiency, E). However, less is known about the removal of oocysts from swimming pool water, though it is suggested that swimming pool filtration systems may be less effective than drinking water treatment due to the frequent absence of effective coagulation/flocculation and sedimentation and often sub-optimal backwashing procedures. Furthermore, the maximum recommended water velocity through the filter in public pools with medium-rate filters is 25 m h À1 (PWTAG b), though this may be exceeded in practice. These are substantially greater than velocities of no more than 10 m h À1 used in drinking water treatment (Gregory ). Studies on removal of oocysts by pool filters have been restricted mainly to the use of surrogates, e.g. polystyrene microspheres in pilot-scale studies (e.g. Croll et al. ; Lu & Amburgey ) and in one case measurement of oocyst-sized particles in an operational pool (e.g.

Stauder & Rodelsperger ).
The removal of oocysts from water through filter retention is a complex process in packed-bed sand filters. This includes factors that affect the delivery of oocysts from the pool to the filter (e.g. the location and number of filter inlets and outlets, and how this influences the mixing characteristics of the pool) through to processes within the filter itself. The aim of this paper is to focus on the latter and to provide a review of the current knowledge of the removal of Cryptosporidium oocysts by filters in commercial swimming pools and to identify the major risks and the gaps in our current understanding. The review considers the mechanisms by which filters remove particles of different sizes (including oocyst-sized particles), factors that affect the efficiency of particle removal (such as filtration velocity and use of coagulant), current recommended management practices, and suggestions for further work to support the development of a risk-based management approach. The main factors to be considered are listed in Table 1.

HOW DO SAND FILTERS REMOVE PARTICULATE MATERIAL?
A swimming pool sand filter bed consists of packed solid particles, which, in the case of 16/30 sand (sand which passes through a No. 16 sieve but is retained by a No. 30 sieve), range in size from 0.6 mm to 1.2 mm (600-1,200 μm), with the majority normally being in the range 600-800 μm. The spaces between the packed sand particles (the porosity) make up 35-50% of the total volume occupied by the particles depending on how rounded or irregular the shape of the grains. If the particles are assumed to be spherical, then the effective diameter of the spaces between the particles is equivalent to one-seventh of the diameter of the sand grain (Huisman & Wood ). For 16/30 sand, the smallest pore size will, therefore, be about 0.1 mm (100 μm), so particles smaller than this (about the smallest size that can be resolved by the naked eye) will not be removed by a simple straining mechanism, but will move into the body of the media bed rather than being retained at the surface. Cryptosporidium oocysts (4-6 μm in size) will, therefore, not be retained simply by the sand particles acting as a screen. However, as larger particles become trapped within the pores, the space sizes are reduced further, or where localised restrictions occur due to irregular-shaped sand particles coming into close contact, increased straining could occur which may trap some oocysts. Straining is not, therefore, likely to be the major mechanism for removal of Cryptosporidium oocysts from pool water, unless the oocysts are present within a much larger floc which could occur where there is effective coagulation/flocculation or if the oocysts are attached to faecal material.
As water travels through the pores between the sand particles, it will pass by the extensive surfaces of the sand grains. For example, 1 m 3 volume of 0.6 mm diameter spheres will have an estimated total surface area of 6,252 m 2 (Huisman & Wood ). Particles which are too small to be screened could be retained by the filter media as a result of weak intermolecular binding forces that come into play if the particles can get very close (i.e. within nanometres) to the surface of the sand grains. This surface adsorption is likely to be the principal mechanism for the removal of Cryptosporidium oocysts.
One approach to quantifying the effectiveness of particle removal by filters is to compare the measurements of the particle contents of the influent and effluent. The simplest index of the effectiveness of filtration is the filter efficiency (E), defined as the fraction (often expressed as a percentage) of particles removed from water as it passes through the filter: where C 0 and C L are the influent and effluent solid concentrations (or the particle counts, or turbidity, depending on the measurements made).
However, E is not just a measure of the effectiveness of the media in removing particles, as this depends also on the depth of the media bed. The impact of the depth of the bed can be accounted for by considering a sand filter as a deep packed bed comprising layers of single collectors (sand grains), where each layer removes a fixed proportion of the particles suspended in the water approaching the layer.
This gives rise to an exponential decrease in the particle content of the water as it moves down through the filter: where L is the filter depth in m and λ is the filter coefficient in units of m À1 . 1/λ is known as the characteristic length of the filter (in m) which is sometimes used as a measure of the intrinsic effectiveness of the media in removing particles (Lawler & Nason ). Equation (2) can be used to derive an empirical value for the filter coefficient based on the measurement of the particle content of the influent and effluent. For example, if a media bed 0.8 m deep removes 70% of particles, then the value of λ will be 1.50.
This value will vary during the backwash cycle, dependent on the degree to which the filter media are loaded with finer material removed from the water (Amburgey ), and may also change if the filter media degrade in some way, e.g. if balling or channelling occurs (PWTAG b).
However, it is an oversimplification of a real filter because the particle removal capability of the layers will not be the same throughout the filter once the upper parts of the filter become loaded.
There is a wide range of factors that affect the filter coefficient (Tufenkji et al. ) and it is, therefore, unrealistic to expect to model the complexity of a real swimming pool filter from first principles. However, attempts to produce a mechanistic model of particle removal provide a valuable insight into some of the key processes (Ncube et al. a).
A starting point is to model the removal of particles from water as it approaches and passes a single spherical collector (Yao et al. ). Only a small proportion of the particles approaching a collector will get close enough to the collector surface for attachment to be possible (this dimensionless fraction being termed the transport coefficient, η), and only a proportion of those particles that make contact with the collector surface will attach (this proportion being the attachment coefficient, α). The overall proportion of particles approaching the collector that actually attach is, therefore, the product of these two terms (ηα).
The model of Tufenkji & Elimelech () scales up the single collector model of particle removal to predict particle removal by a filter bed. This model relates the filter coefficient to the media geometry (the collector diameter, d m and the filter bed porosity, ϵ) and also to the single collector transport and attachment coefficients discussed by Yao et al.
Particles attach to the collector surface by van der Waals forces, which is a universal but short-range phenomenon which holds particles at the surface once contact has been made. This is an important mechanism for removal of microscopic particles from pool water (Huisman & Wood ; Tufenkji & Elimelech ). In this context, van der Waals forces are the result of the sum of all the individual intermolecular forces between the two interacting surfaces.
These forces will only be effective over very short distances (nanometres); hence particles, to all intents and purposes, have to be in contact with the surface of the media before adsorption can occur.
The starting point to modelling the transport coefficient is to recognise that contact between sand grain surfaces and suspended particles is compromised because the water flowlines typically divert around the edges of the collectors (sand grains). If suspended particles are to make close contact with the collector surface, then they need to divert (break out) from the water flowlines. There are three principal mechanisms that can achieve this (  filter efficiency (E) and particle diameter, with a local minimum that modelling studies predict to be in the 1-5 μm range (Lawler & Nason ). It is important to note that with respect to Cryptosporidium oocyst removal by filtration, it is the particles within the 4-6 μm size range that have the least chance of making sufficiently close contact with sand surfaces to enable their adsorption. Particles of this size tend to be too small for sedimentation/impaction to be effective and too large for diffusion to be effective.
Together, these factors render the removal of Cryptosporidium oocysts from pool water by filtration alone a challenge.
Though the mechanistic modelling studies above give an insight into the issues affecting the effectiveness of swimming pool filters, there will be need for at least a semi-empirical approach to assessing quantitatively the effectiveness of swimming pool filtration. Such assessments would be based on measurements of particle removals, which might be achieved by monitoring the changes in either turbidity or else be monitoring the changes in the number of particles present within the particle size band of interest as water passes through a filter.

WHAT DO TURBIDITY MEASUREMENTS AND PARTICLE COUNTING TELL US?
Turbidity is caused by particles in suspension, primarily  were dominated by the smaller-sized 1-10 μm particles (89% of total particle counts). Removal efficiencies (number of particles removed by the filter relative to the number of particles entering the filter) for the 1-10 μm size class were >98%.
The maximum allowable turbidity for drinking water is 4 NTU (DWI ), although municipal supplies should normally achieve <0.5 NTU prior to disinfection (WHO ); for pool water, the turbidity should be <0.5 NTU (PWTAG b), which is almost an order of magnitude lower than the limit of detection by the naked eye in relatively shallow water (WHO ). So, while visual assessments of water clarity are useful for assessing gross failures of the filtration system, they cannot be relied upon for detecting changes in turbidity that could be critical for microbiological risk to pool users. The advantages of particle counting compared to the measurement of bulk turbidity in relation to the risk from Cryptosporidium oocyst in pools require further investigation.
One important difference between swimming pool filtration and industrial/potable water treatment is that the pool water is constantly recycled through the filtration system. Pool water is also normally cleaner than that treated for drinking water and is subject to large fluctuation in particle content and turbidity during diurnal cycles due to variations in bathing load. During prolonged periods where there are no bathers (and so very little particle input to the pool), such as at night or during a closure period to clean up after an AFR, most of the water will pass through the filter several times with little further particle input. One study reported turbidity of <0.1 NTU during nighttime monitoring of a very busy outdoor paddling pool (Stauder & Rodelsperger ). However, there is a dearth of data on how filter removal efficiencies change with variation in bathing load during opening hours and overnight and during prolonged closure periods. It is thought that the recovery which is believed to occur overnight is important in the cleanup of pool water following faecal contamination events, but this requires further investigation. For example, the particle content of filtrate will be contributed to by particles that are detaching from a filter as well by particles that are simply passing through the filter without attaching.
As the detachment process will continue throughout the day and night, this might result in reduced apparent filter efficiencies at night when the water being delivered to the filter is very clean. Further detailed work is needed to understand the relationship between the particle content of water and particle removal by packed-bed sand filters.

WHAT IS THE EFFECT OF FLOW RATE ON FILTRATION?
The flow rate will have no impact on the screening proper- Microsphere removal was 90% at 30 m h À1 , but only 50% at the faster flow rate of 37 m h À1 . This suggests that the sensitivity of Cryptosporidium oocyst removal to filtration velocity can be much greater than would be predicted by Gregory ().

WHAT IS THE EFFECT OF FILTER RIPENING AND BACKWASHING?
Cleaning of a filter is achieved by backwashing; this involves reversing the flow through the media bed, fluidising the filter material and passing the backwash water to waste (WHO ). This is normally triggered by one or more of the following: the maximum allowable turbidity value has been exceeded (particularly if the filtrate is being monitored and elevated turbidity is indicative of filter breakthrough occurring), or a specified period of time has elapsed since the previous backwash, or a specified differential is observed between the water pressures at the inlet and outlet of the Immediately following backwashing of a filter, a ripening sequence is observed during which the quality of the filter effluent initially deteriorates, then recovers (Amburgey ). In general, there will be an initial flush of particles in the filtrate immediately following a backwash, which can be diverted to waste in a rinse procedure (PWTAG b), though many existing pool filtration systems do not have the pipework required for rinsing. This is followed by a more gradual improvement in particle removal as the filter ripens. Once the filter is ripened, there will be a period of near-optimum filtration. As the filter becomes loaded with trapped particles, there will be a further phase of decreased removal and a point is reached where a surge of particles breaks through the filter (Amirtharajah ). The duration and extent of each of these stages depend on many factors such as the filter media, filter loading rate, backwash flow rate and duration (Amburgey ).
There has been extensive research into the factors affecting filter ripening and in developing backwashing techniques that promote ripening and optimise the amounts of water used in the backwashing and rinsing procedure (e.g. Amburgey ). However, this work has been directed primarily at the drinking water industry where the source water being filtered is normally dirtier than in swimming pools, and backwashing is carried out more frequently than is the case for swimming pools. In industrial applications of sand filters, the backwash is normally optimised due to high operational costs of backwashing. This is not generally the case in swimming pool operations where the backwash process is often sub-optimal in terms of factors such as the timing, water velocities used, and whether or not the procedures include air-scouring or rinsing.

WHAT IS THE EFFECT OF DIFFERENT FILTER MEDIA?
There are a number of alternative media available for use in swimming pool filtration, with a trend for traditional 16/30 sand being replaced by a range of artificial glass media and non-glass media such as extruded high-density polyethylene (e.g. OC-1 media). These media can differ widely in a number of characteristics that will affect the filter performance in terms of removal of oocyst-sized particles. These include the porosity (which will affect the average water velocity), the sizes of water channels which will affect the water velocity profiles adjacent to collector surfaces (Bradford et al. ), and the micro-topography of the interceptor surfaces will affect the strength of short-range forces.
Surfaces with irregularities will tend to provide greater opportunities for contact to occur ( Jin et al. ), and collector shape irregularity may also provide opportunities for straining of oocyst-sized particles at pore necks (Tufenkji et al. ).
This raises the question of the extent to which the smoothness of the filter media affects its performance in terms of removal of Cryptosporidium oocysts. For example, there is evidence that sand grains are more effective than artificial glass in terms of the removal of Cryptosporidium-

WHAT IS THE MINIMUM DEPTH OF MEDIA BED REQUIRED?
Lawler & Nason () developed an approach to designing filters which combined Equations (1)-(3) above to determine the minimum depth of media bed required to achieve satisfactory filtration. The starting assumption was to use the Tufenkji & Elimelech () model to predict the particle size giving the lowest removal efficiency (i.e. the smallest value for η, for the filter in question, based on the filtration velocity, temperature and grain size, which in their case was a particle size of 1.5 μm). Lawler & Nason () then identified eight wastewater/potable water treatment filters known to give good filtration and estimated in each case the value of λ (Equation (3)), and hence, the filtration efficiency (combining Equations (1) and (2)) predicted for the particle size with least efficient removal.
The calculated filtration efficiencies for these filters known to perform well were found to fall within a narrow band around 25% for the 1. • This is for a clean filter bed, and the filtration efficiency is likely to improve as the filter ripens. However, in the context of swimming pool filters that are filtering very clean water, it is unclear how long it takes for filters to ripen to the point where there is more effective removal of 4-6 μm size particles.
• This assumes that there is an appropriate addition of coagulants to ensure particle destabilisation (and hence an attachment coefficient α ≈ 1). If electrostatic repulsion is preventing attachment, then the filter efficiency will be much lower. On the other hand, if the Cryptosporidium oocysts are coagulating into larger flocs before arriving at the sand bed, then filtration efficiencies will be higher.

INFLUENCE OF POOL WATER CIRCULATION ON RISK MANAGEMENT
The conventional indicator used to assess whether water circulation is adequate to maintain good water quality is turnover time. This is described as For example, PWTAG (a) recommends that a 25 m leisure pool should have a turnover time of no more than 3 h. This does not mean that in a single 3 h period, all of the water in the pool will have passed through the filtration system. Even if the water in the pool is perfectly mixed at all times, theory indicates that only 63% of the water in the pool will pass through the filtration system in one turnover time (Gage et al. ). In other words, 37% of the water in the pool tank will not be filtered in each turnover.
The current guideline for cleaning up a pool after a suspected Cryptosporidium contamination is that the pool is closed and the water allowed to circulate for six turnovers, after which the filters are backwashed and the pool can then be returned to use if the pool operator is confident in their backwashing procedure (PWTAG a). If the pool is perfectly mixed, then after each turnover, 37% of the water in the pool at the start of each turnover period will remain unfiltered at the end of each turnover; so, after six turnovers, the amount of untreated water remaining in the pool will be equivalent to 0.25% of the water originally present in the pool.
If we also take into account the efficiency of the filters, are based on the understanding that effective pool filters will remove 90% of oocysts in a single passage of water through the filter. If so, then a well-mixed pool will lose 63% × 90% ¼ 56.7% of oocysts from the pool in each turnover (in other words, 43.3% of oocysts in the pool tank at the start of a turnover period will remain in the pool after each turnover). As a result, after six turnovers, the number of oocysts remaining in the pool will be 0.66% of those present in the pool at the start of the six-turnover period (i.e. >2 log reduction).
However, even pools designed specifically to have rapid mixing can have poorly mixed dead zones (Lewis et al. ; Chalmers et al. ), and there is, therefore, considerable uncertainty about the proportion of pool water that will pass through the filtration and circulation system in any one turnover time. For any particular pool, the proportion of filtered and unfiltered water at a particular point in time is unknown. For example, if the impact of poorly mixed dead zones, and any short-circuiting of flows between inlets and outlets, was to increase the percentage of water that is untreated per turnover from 37% to 47%, then the number of turnovers would have to increase to eight to achieve the same result as a pool assumed to be perfectly mixed (Table 2).
A further complication is that after an AFR, any oocysts present in the faecal material will not be uniformly distribu- Furthermore, the filtration system might be less effective than is presumed. For example, if the filters are only removing 50% of oocysts in a single pass of water rather than 90% (as in the case reported by Lu & Amburgey ), then this Required to achieve same removal as a best-case scenario, i.e. 63% of water treated per turnover and filters operating at 90% efficiency. c Based on the initial contamination of 20,000 oocysts L À1 estimated for a typical AFR into a 450 m 3 pool (Gregory 2002), equivalent to 740/37 mL, using the value of 37 mL as the average volume of water swallowed by non-adults during a 45 min pool session of swimming (Dufour et al. 2006).
would increase the number of turnovers required to achieve similar removal of oocysts from six to 13 turnovers (Table 2).
A combination of poor mixing (47% of water untreated per turnover) and poor filtration (50% oocyst removal efficiency) would increase the number of turnovers required to achieve a similar removal of oocysts from six to 16 (Table 2).
The first data row in Table 2

KNOWLEDGE GAPS AND CONCLUSIONS
This review of the processes by which filters remove particles of different sizes from pool water has highlighted a number of areas where significant gaps exist in our understanding of the factors controlling the risk to pool users from particulate material, and specifically for Cryptosporidium oocysts. Though we have used the removal of oocysts following an AFR to illustrate conclusions, the discussion about the factors affecting filtration apply equally to maintaining water in good condition during normal operation. We have shown that for a 'best-case scenario' pool, the UK guidelines on dealing with faecal contamination (PWTAG ) will significantly reduce the numbers of oocysts in the pool and minimise the risk of infection while still providing operators with a realistic course of action. However, if there is less than optimal filtration, the absence of coagulation in pool water treatment or a poorly mixed pool current guidelines may fail to mitigate the risk.
In view of the fact that commercial swimming pools do not lend themselves to experimentation, advances in our understanding are most likely to be gained from a combination of modelling and in situ measurement.
Mass balance models such as that used by Stauder & Rodelsperger () can underpin our understanding of the behaviour of particulate material (e.g. Cryptosporidium oocysts) in pool plant systems and, when combined with the measurement of turbidity and particle counting, offer opportunities for the development of tools which integrate our understanding of both the hydraulics of the pool and the filtration efficiency to assist in assessing risk, e.g. from Cryptosporidium oocysts in pool water.
A number of specific areas that require further investigation have been identified in this review: 1. Validation of existing models with full-scale swimming pool studies will also enable quantification of filtration effectiveness in relation to different particle size fractions. Full-scale studies on operational pools are also necessary to investigate factors affecting the filtration, including the use of coagulants and filter aids, as well as the employment of different filtration media types.
2. The understanding of the removals of different particle size fractions needs to be improved via a detailed study of the nature and behaviour of these particles, together with better understanding of the value and limitations of turbidity measurements with respect to the removal of Cryptosporidium oocyst-sized particles.
3. Impact of circulation rate, particle loading, backwash frequency, backwash flow rate and time following backwash on the filtration of different size particles requires a detailed investigation that should include continuous monitoring of filter performance through a backwash cycle at a range of pool sites.
4. The effect of the many factors that affect the delivery of oocysts/turbidity from the pool to the filtration system needs to be quantified. These include: • the location and number of filter inlets and outlets, and how these impact on the mixing characteristics of the pool; • moveable floors; • the ratio of sump flow to surface draw-off; • bathing load and distribution within the pool; • the likely input of particulate material from bathers, and how this is affected by factors such as age, pre-swim hygiene, and whether pools are indoor or outdoor.
Once this information starts to become available, it then becomes possible to develop a risk-based approach to managing swimming pool water, particularly for the management of waterborne disease outbreaks, along the lines proposed for drinking water (e.g. Havelaar ; Petterson & Ashbolt ). For example, HACCP (Hazard Analysis and Critical Control Points) provides a framework for integrating current scientific knowledge of microbiological hazards into a quality management system based on monitoring of critical control points in the water treatment process and has been applied to the production of drinking water in Belgium (Dewettinck et al. ). Such an approach would enable the outputs of the research identified in this review to be used in ways that will be of direct benefit and reassure operators, regulators and users of the safety of commercial swimming pools.