Microbial Risk Analysis 8 (2018) 22–31

Contents lists available at ScienceDirect

Microbial Risk Analysis

journal homepage: www.elsevier.com/locate/mran

Quantitative microbial risk assessment of repairs of the drinking water T

distribution system
⁎
Mirjam Blokker , Patrick Smeets, Gertjan Medema
KWR Watercycle Research Institute, Groningenhaven 7, 3433 PE Nieuwegein, the Netherlands

A R T I C L E I N F O A B S T R A C T

Keywords: A Quantitative Microbial Risk Assessment model was developed to assess the infection risk due to faecal con-
QMRA tamination events after repairs of drinking water mains. Ingress was modelled per section between isolation
Drinking water distribution valves; transport of pathogens and consumed dose were modelled with a hydraulic network model and stochastic
Repairs drinking water demand patterns. Dose response models were then used to calculate the risk of infection. The
Consumption pattern
sensitivity analysis showed that the contamination concentration is the most important parameter for the in-
gested dose, while the choice of dose response relation highly impacts the resulting infection risk. The time of
day that valves are opened after repairs, releasing the contamination, in combination with the time that drinking
water is withdrawn from the tap influences the amount of contamination consumed versus flushed away through
non-exposure uses such as toilet flushes and showering. A standard QMRA approach that typically neglects the
diurnal consumption pattern may underestimate the risk. The consumption volume is less important. Issuing a
boil water advice and opening only one valve before “releasing” the isolation section are effective mitigation
options to reduce the infection risk per event by 50% to 80%. The statutory Dutch sampling protocol requires E.
coli analysis of a 100 ml sample taken the day after the repair. The modelled probability of detecting a con-
tamination under these circumstances is approximately 25%, because the contamination rapidly leaves the
drinking water distribution system through the customers’ taps, toilets and showers. If a sample is taken 1–4
hours after the repair the modelled probability of detection exceeds 80% when taken at the optimal location.

1. Introduction barrier in the water supply process. This works well as long as the
systems integrity is kept intact. Many events associated with the repair
Contaminations in the drinking water distribution system (DWDS) and maintenance of DWDS can contribute to pathogen contamination,
have led to a significant number of waterborne disease outbreaks. such as replacing a pipe or a negative pressure event when an isolation
Moreover, the proportion of waterborne disease outbreaks associated section of the DWDS is closed off for repairs (Lambertini et al., 2011;
with DWDS problems is increasing (National Research Council, 2006). Karim et al., 2003; LeChevallier et al., 2003). These failures can lead to
In the USA, DWDS deficiencies were associated with approximately loss of physical or hydraulic integrity in DWDS for a few seconds during
30% of outbreaks in community water systems and since 1991 there has a negative pressure event or for 2–6 hours for a water main repair
been an increased proportion of waterborne disease outbreaks asso- (Kirmeyer and Martel, 2001; Besner et al., 2011). This can result in
ciated with contaminants entering the DWDS after treatment short term contamination events that can lead to adverse health im-
(Craun and Calderon, 2001; Craun et al., 2006; Hrudey, 2004). In pacts for customers. The current study doesn't address other important
Europe, 31% of 61 studied outbreaks were caused by events in the risks during distribution such as cross connections to non-potable water
DWDS (Risebro et al., 2007). While outbreaks highlight the failures in or growth of pathogens in biofilms.
the DWDS, they do not tell the whole story because many smaller, and Assessing the potential health impacts from drinking water con-
undetected, contamination events are likely occurring on a more reg- tamination events and the benefit of mitigating effect is no easy task. It
ular basis (van Lieverloo et al., 2006). A risk evaluation for the different requires understanding the fate and transport of contaminants through
types of outbreaks, that vary in severity, needs to be completed the DWDS, the exposure of consumers to contaminated water, and the
(National Research Council, 2006). response of the individuals to the exposure (Davis and Janke, 2009). As
DWDS are the final component of public water supplies and the last sampling the DWDS for faecal contamination indicators typically leads

⁎
Corresponding author.
E-mail address: Mirjam.blokker@kwrwater.nl (M. Blokker).

https://doi.org/10.1016/j.mran.2017.12.002
Received 8 September 2017; Received in revised form 21 December 2017; Accepted 21 December 2017
Available online 23 December 2017
2352-3522/ © 2017 Elsevier B.V. All rights reserved.
M. Blokker et al. Microbial Risk Analysis 8 (2018) 22–31

to a large underestimation of the contamination risk, a modelling ap- of uncooked drinking water, of volume per consumption, of time of
proach is required. The estimation of the potential risk from short-term consumption, and of dose response relations.
contamination events needs to be done by Quantitative Microbial Risk This paper describes the development (Blokker et al., 2014) and
Assessment (QMRA) coupled with a hydraulic network model calcula- sensitivity analysis of a QMRA model for contamination events after
tion. This method was previously used to evaluate the risk from mi- mains repairs in the DWDS combining hydraulic, consumption and dose
crobial intrusion during negative pressure events that led to con- response models. Its goal is to determine the infection risk of various
tamination of distribution systems (Teunis et al., 2010; Yang et al., contamination scenarios of the drinking water after mains repairs at
2011). In these previous studies, the focus was on contamination several locations in the DWDS and at several times of the day, and
through transients and the effect of chlorination. No chlorination is evaluate the effect of mitigation alternatives to limit the infection risk
applied in the Netherlands and with the country being very flat, the of a potential contamination by e.g. not opening all isolation valves
most important route of contamination is not through transients, but after mains repair or issuing a boil water advice. The model also allows
rather through the pipe repair scenario. This study focusses on the to assess the likelihood of detecting a contamination after mains repair
QMRA for pipe repairs. in the DWDS with the standard sampling protocol for drinking water
The Dutch Drinking Water Decree (Drinking Water Decree, 2015) systems.
requires drinking water companies to perform a QMRA for Cryptos-
poridium, Campylobacter, Giardia and enteroviruses on the treatment 2. Methods and materials
works. The DWDS is not being considered in that QMRA. This study
focusses specifically on the DWDS, and uses the same types of patho- 2.1. Hydraulic model
gens as the QMRA for treatment does. Therefore this study focuses on
faecal contamination of drinking water, for which ingestion is the main An EPANET model is used to simulate the distribution of the con-
route of exposure. Opportunistic pathogens such as Legionella, and other taminated water through the DWDS. The model of a part of the town
exposure routes such as inhalation during showering (Schijven et al., Zandvoort was used. It supplies about 1000 homes, 3 larger hotels and
2016) are not included in our study. In the aforementioned studies 3 beach clubs (Blokker et al., 2010a). For this area the demand patterns
(Teunis et al., 2010; Yang et al., 2011; Schijven et al., 2016), the were generated with SIMDEUM (Blokker et al., 2010b). The residence
amount of water consumption was assumed to be known with the daily times in this hydraulic model, with these demand patterns, were vali-
intake volume following a lognormal distribution (which means never dated with a tracer test (Blokker et al., 2010a). The tracer test also
equal to or lower than 0) to be consumed as one volume at one moment showed that plug flow, with limited dispersion, can be assumed to
of the day. However, individuals can ingest water at multiple time adequately represent the flow and mixing conditions. The SIMDEUM
points over one day drawn from the tap at several moments during the demand patterns are further described in Section 2.5. The pattern time
day. Van Abel et al. (2014) have shown that this can impact the cal- step was 5 minutes and the water quality time step was 1 minute. There
culated risk. Since ingestion is the main route of exposure, accurately are 56 valves in the system, dividing the network into 44 sections.
characterizing this parameter is important (Besner et al., 2011; EPANET is used to calculate flows and the spreading of con-
Nygård et al., 2007). A related issue, the issue of timing, was previously taminated water through the DWDS. This helps to recognize the cus-
examined though the use of a hydraulic model combined with nu- tomers who are downstream of a contamination event. The likelihood
merous exposure models to estimate the impact on dose. In their ana- of a contamination passing by and coinciding with a withdrawal at the
lysis, Davis and Janke (2009) looked at a variety of exposure models, tap of water for uncooked consumption is modelled outside of EPANET.
such as ingesting water every hour of the day or during every meal, for The number of people on a demand node is determined by the total
the potential impact on dose. However, this work did not investigate water demand on that node divided by the average water demand of
combining a hydraulic model with an exposure and risk model to look 120 L/person.day (Blokker et al., 2010a).
at the impact of the number of consumption events on the probability of
infection of a pathogen in a community. As the time of consumption 2.2. Overview of the QMRA model steps
may be important, the time of the contamination entering the DWDS
may also be important. This study takes into account best estimates plus The QMRA model calculates the risk of an event. In this study an
variability of time and location of a contamination after repair, of the event is defined as a contamination of an isolation section after repair
severity of the contamination, of water demands and thus flows and with 104 pathogens. Table 1 summarises the QMRA model steps. These
contamination spread through the network, of number of consumptions are described in detail in the next sections. There is a specific section on

Table 1
Overview of model steps.

QMRA model step Description of QMRA model step Quantitative approach Variability in investigated in sensitivity analysis

1. Event Contamination after maintenance of Pathogen is diluted in total Type of pathogen

DWDS (section isolated) isolation section volume Concentration of pathogens due to a contamination entering the
system and to dilution in isolation section
Location in the DWDS (at random or based on pipe material and
diameter dependent break frequency)
2. Distribution Contamination is distributed through Hydraulic model (EPANET) Valves opened (all or only one feeding valve)
DWDS
Valve opening time
3. Consumption Contamination leaves the DWDS through Drinking water demands (SIMDEUM) 2 different sets of demand patterns
customers’ taps
4. Ingestion Contamination is ingested by people Volume × concentration Time of tapping water for drinking
drinking water
Volume of water for drinking
Number of ingestion moments
5. Infection Ingested water infects a person Dose response relation Four different pathogens
Average and worst case
6. Event risk Infection risk per event

23
M. Blokker et al. Microbial Risk Analysis 8 (2018) 22–31

another location in the isolated section that is unpressurised

(Kirmeyer and Martel, 2001). The extent to which these contaminations
occur in practice is unknown. A literature search was performed to
obtain realistic estimation of the frequency of occurrence, the volume of
contaminated water or soil entering the DWDS, and the pathogen
concentration in that contamination.
Sewage generally contains high concentrations of pathogens and
indicator organism E. coli (Karim et al., 2003; Kirmeyer and
Martel, 2001; van Lieverloo et al., 2007; Besner et al., 2008;
Abbaszadegan et al., 2003; Hunt et al., 2010; Yates et al., 1985;
Hoogenboezem et al., 2001; Soller et al., 2010; Sales Ortells, 2015).
Intrusion of 1 l sewage can typically result in 104–105 Campylobacter,
Cryptosporidium and Giardia and 103 enteroviruses in the isolated sec-
tion. Paul et al. (2004) showed that sewer leakage occurs frequently,
Fig. 1. Schematics of pipe repair and consecutive actions. with 31,006 defects in 208 km of sewer pipes. They also showed that
groundwater within several meters of sewers was contaminated. As in
the variability in the model and how the sensitivity analysis and effect Dutch cities sewers are often within 1 m distance of drinking water
of mitigation options are taken into account in the model. The best pipes, this is an important contamination route
estimates for the contamination event, distribution, consumption, in- Human faeces can contain very high levels of pathogens when the
gestion and infection were used in the so called reference scenario for person in question carries an infection, which can occur asymptoma-
the QMRA model to determine the infection risk. This scenario acts as a tically. One milligram of faeces can contain 105 Campylobacter,
reference for the sensitivity analysis where variations on the reference Cryptosporidium and Giardia, and even higher concentrations of viruses,
scenario are used. The reference scenario also acts as a reference for the although enteroviruses were not reported (Westrell et al., 2004).
risk mitigation scenarios where some changes are made with respect to Human pathogenic microorganisms in animal faeces are highly variable
the reference scenario. and depend on the type of animal and its individual health status (like
human faeces). They can reach the same high concentrations of pa-
thogens as in human faeces, but don't contain human pathogenic
2.3. QMRA model step 1: event scenario viruses (Soller et al., 2010).
Pathogens are present in most surface waters due to wastewater
The events that are simulated are contamination events during discharge, agricultural runoff and wildlife, however concentrations
maintenance on the piped network. The maintenance may be planned vary greatly. Concentrations are several orders of magnitude lower than
work or repair of a pipe failure. During the repair (Fig. 1) a section is in sewage (van Lieverloo et al., 2007), and it would require a very large
isolated by closing valves, the section has no water pressure and a intrusion volume to reach the same levels of contamination as 1 l of
contamination may enter the isolated section. Standard procedure in sewage. However, intrusion of large volumes of nearby surface water
the Netherlands is that the isolated section is flushed before it is re- through pipe leaks is feasible when the system isn't pressurized.
leased back into service. Flushing would lead to a contamination being The estimated amount of faecal contamination and the resulting
washed away, or to be diluted in the entire volume of the isolation amount of pathogens in the isolated section for each potential pathway
section. In case customers are connected to the section, one (or more) was highly variable, due to the range of pathogen concentrations found
valves are then opened, to ensure water supply is restored. The next day in faecal contamination sources. Based on these estimates a con-
(protocol requires 12–24 hours after the repair, in practice this is the tamination of 104 Campylobacter in the isolated section were used as a
next day during working hours) a 100 ml sample is taken and tested for realistic reference scenario. This corresponds to a contamination with 1
E. coli and enterococci. Due to uneven distribution of microorganisms in litre of sewage or 1 mg of human or animal faeces. Since the average
water and the small sample volume, a contamination may still be pre- isolation section volume is 3.4 m3 this implies that the average con-
sent then, although undetected. When the sample is negative for faecal centration of micro-organisms in the simulation is 0.3 per 100 ml
indicators, all valves are opened. (Fig. 2).
Several potential pathways of a faecal residue contaminating the Ten years of data (2003–2013) of six Dutch drinking water com-
trench or soil surrounding the DWDS under repair can be identified: panies on E. coli measurements both the regulatory routine samples and
the mains repair samples were analysed. On average 0.5% of all mains
• Water leaking from a sewer may reach the drinking water main,
after a short soil passage. Typically, in urban areas in the
Netherlands sewers are located within 1 m from drinking water
mains.
• Human or animal faecal residues may be present in nearby surface
water that may infiltrate to the groundwater and reach the drinking
water main.
• Animal faecal residues on the surface may either fall directly into
the trench, or flush into the trench with rain water, or pass with rain
water through the soil into the trench.
• Human faecal residues on the hands of personnel may contaminate
the drinking water.
• Animal faecal residues may have been dropped inside the new pipe.
• Human or animal faecal residues may be present on the tools that
are used to repair the main.

The faecal residues, once they are in the trench, can then enter the
Fig. 2. Number of micro-organisms per 100 ml in the simulations.
main under repair, either at the repair point or through a leak at

24
M. Blokker et al. Microbial Risk Analysis 8 (2018) 22–31

Table 2
Failure frequency per pipe material (PVC, Cast Iron and Asbestos Cement) and diameter
(USTORE (Vreeburg et al., 2013) data 2011, collected 3 April 2013).

Diameter (mm) Failure frequency (#/km.year)

PVC CI AC

50 0.017 0 0
100 0.032 0.123 0.099
150 0.025 0.049 0.085
250 0.032 0.043 0.091
400 0.034 0.015 0.044

repair samples contained E. coli, indicating that the operating proce-

dures are effective. Only 0.07% of the regulatory samples contained E. Fig. 3. Closure and opening times of valves after unplanned (Water Company 1, 2012,
coli (Blokker et al., 2016), suggesting that the water is more often 349 events; Water Company 2, 2011, 548 events; Water Company 3, 2012, 147 events)
contaminated after mains repairs. However, this data doesn't show the and planned maintenance (Water Company 1, 2012, 77 events). The dotted light grey line
(unknown) false-negative samples where contamination occurred and corresponds to a uniform probability distribution over the day.

pathogens remained present in the section but no E. coli was detected.

Therefore it was not possible to estimate the likelihood of a con- hour,
tamination event and the involved pathogen concentrations from this 3. The contamination was added to a node in this section with a load
data. The goal of the QMRA model study was then set to quantify the equal to 104 that was diluted in the section volume (Fig. 2 shows the
risk given that a contamination has occurred, regardless of its prob- typical concentrations that were simulated),
ability. 4. On the nodes next to all valves extra demands were added to quickly
The location of the event is related to the probability of main- spread the contamination over the section (to mimic what happens
tenance on the network, which in turn is related to pipe material and during flushing),
diameter (Table 2). The age of the pipes was not distinctive in the 5. After 3 time steps (15 minutes) the valves were opened again and
model network. The relative failure frequency per pipe section in the the extra demand was removed,
model study area was therefore based on the parameters in Table 2, 6. The normal demand situation with a contamination section was
which means the likelihood of a repair is about a factor of three higher then simulated (flows and water quality).
on small (≤250 mm) AC pipes then on small PVC pipes. It was assumed
that planned work on the network follows the same relative pipe ma- In the reference scenario the data for opening valves after un-
terial and pipe diameter dependency as the repair of failures. planned maintenance (Fig. 3, black solid line) was used; in the sensi-
tivity analysis no time dependence was used (random time over the day,
2.4. QMRA model step 2: distribution through the DWDS Fig. 3, dotted light grey line). An alternative scenario in QMRA model
step 2 is to consider a maximum of 107 pathogens being diluted in the
To define the spread of the contamination through the DWDS the section volume. This could occur if 10 l of sewer water with a con-
following was considered: centration of 106 pathogens per litre would enter the DWDS
(Jones, 2001). A sensitivity analysis was done on both the reference
• A search routine in Matlab ®
code is used to determine in which scenario and this 1000-fold higher concentration to determine if the
isolation section the (modelled) pipe failure occurs and thus: uncertainties in the model are affected by the pathogen concentration.
○ which valves are closed to isolate the section;
○ which pipe volume is used to dilute the contamination in;
2.5. QMRA model step 3: water demand by users
○ which valves are opened after the (modelled) repair. Two sce-
narios are considered: (1) open all valves and allow the con-
The demand patterns in the hydraulic model were constructed with
tamination to spread downstream (reference scenario), (2) open
SIMDEUM. SIMDEUM generates random diurnal demand patterns with
only one valve to supply the house connections within the section
a high temporal resolution (1 s) per household connection based on
with water again, but the contamination is restricted to the iso-
statistical information on household composition, type of water using
lated section (mitigation scenario). In the second scenario the
appliances and water using behaviour (Blokker et al., 2010b). The same
valve that in normal operation would experience the largest flow
approach was followed as in an earlier study on residence times in the
is considered to be the upstream valve to be opened. In case
same area as this case study (Blokker et al., 2010a). This earlier study
opening only this valve would mean a downstream section of the
compared measured residence times and modelled residence times with
network would still be cut off from supply, an extra valve would
SIMDEUM demand patterns in the hydraulic model. It was shown that
be opened (in Fig. 1 this would mean opening valve 2 if section C
the SIMDEUM demands lead to realistic flows in the DWDS. Blokker
can only be supplied from section A and B).
• The starting time is determined by a probability function based on
et al.(2010a) give a thorough description of the hydraulic model and its
demand patterns. For the QMRA study some alterations were made; the
maintenance data on valve closure and opening times after repairs
base demand of the hotels was set to a fixed value and the extra demand
of 3 Dutch water companies.
• The simulation time was 3 days.
on the measuring nodes was removed. In this study two sets of SIM-
DEUM demand patterns were applied, where one set is used in the re-
ference scenario and the other set is used in the sensitivity analysis. The
An EPANET model was used to determine the distribution of the
two sets differ as they are the results of two runs of a Monte Carlo
contamination through the network. In the EPANET model
simulation. It can be thought of as model run 1 representing the
drinking water demand on one day, and model run 2 representing the
1. The isolation section under repair was selected (QMRA model step
drinking water demand on the next day. The sensitivity analysis in-
1), and its valves were identified;
vestigates the effect on the infection risk of changes in demand, i.e.
2. The isolation valves were closed at the randomly picked starting
would a contamination scenario have a significant difference in

25
M. Blokker et al. Microbial Risk Analysis 8 (2018) 22–31

infection risk on one day compared to the next day or next week. set proportional to the total drinking water over the day (Fig. 5A),
which is a common method, or proportional to the water use at the
kitchen tap (Fig. 5B) as suggested by Davis and Janke (Davis and
2.6. QMRA model step 4: ingestion of contaminated water
Janke, 2009). It could occur that the total water demand of the SIM-
DEUM at a certain node is zero during the average Dutch kitchen tap
The (contaminated) drinking water leaves the DWDS through the
use, e.g. because no one is at home during lunch. To correct for this, the
customers’ taps. Most of it is flushed in the toilet or used in the shower;
kitchen tap use over the day is multiplied by zero during moments of
only a small portion is ingested. Two model variations were considered,
zero water use at each node, as illustrated in Fig. 5C. Our reference
viz. the drinking water ingestion amount and the drinking water in-
scenario used the method of Fig. 5C; in the sensitivity analysis the
gestion time.
method of Fig. 5A was evaluated.
The ingestion amount for each person in the study area was mod-
Based on the annual nodal demand and a demand of 120 l per
elled as a Poisson distribution on number of “glasses”, or actually the
person per day, the model determines the number of people supplied by
number of times someone would get some water from the tap for
each nonzero demand node in the hydraulic model. For each person the
drinking purposes, and a lognormal distribution of the ingested volume
number of “glasses” during the three days of the simulation is de-
per consumption. The original Dutch data from the publication by
termined from the appropriate probability distribution. And for each
Mons et al. (2007), i.e. without the correction for assuming 250 ml per
glass the volume and the moment of filling the glass are determined
glass, were analysed to determine the Poisson and lognormal para-
from their specific probability distributions. The dose per intake is
meters. A Poisson distribution with an average of 1.43 glasses per day
equal to the volume multiplied by the concentration at the moment of
(10 per week were reported, this leads to 24% of peoples drinking 0
withdrawal and the total exposure is determined by the sum of the dose
glasses per day) in combination with a lognormal distribution with
for all intakes.
μ = −3.19 en σ = 1.485 (mean = variance = 0.1238 l, as from the
consumption study) describes the amount of uncooked tap water con-
sumption quite well (Fig. 4). Since 1998 there was an increase in the 2.7. QMRA model step 5: infection
Netherlands in the consumption of drinking water from the tap
(Schijven et al., 2016; Van Rossum et al., 2011). In our study the A statutory QMRA is required for drinking water produced from
number of glasses per day was increased, in accordance to e.g. the surface water in the Netherlands, which is evaluated with the
Australian and UK consumption data (Mons et al., 2007) while the QMRAspot software (Schijven et al., 2011). The dose response models
amount per glass was kept the same. A 2010 Dutch survey on re- used in QMRAspot were therefore used in the current QMRA study of
sidential water use reported that on average people took drinking water infection risk due to DWDS contaminations. These dose response rela-
from the kitchen tap for uncooked consumption 2.3 times per day tions (DR) consisting of 10,000 parameter (α, β) pairs (Teunis et al.,
(Foekema and van Thiel, 2011). For this study the frequency of con- 2010; Blokker et al., 2016) for the hypergeometric function distribution
sumption was rounded toward 2.5 (Poisson distribution on the number for four pathogens (Campylobacter, Cryptosporidium, Giardia and rota-
of glasses with λ = 2.5) and the same lognormal distribution on the virus as model for enterovirus) were obtained from the authors of
amount per glass was used (with μ = −3.19 and σ = 1.485, Fig. 4). A (Schijven et al., 2011). To decrease the computing time the median
minimum of 0 and a maximum of 4.2 l per person per day was set as a dose response relation was used in the QMRA model.
boundary in accordance with the reported maximum in the 1998 study.
To study the effect of the variation in consumption, a sensitivity ana- 2.8. QMRA model step 6: event risk
lysis was done with the number of glasses fixed (to 1 and to 3) and the
total amount fixed (at 300 ml per day). The number of infected people was calculated per contamination
The important moment in time is the moment a glass or bottle is event. Drinking water companies want to avoid significant risks from
filled, rather than the moment of ingestion. Therefore, the moment of drinking water, even if this event would only occur sporadically and
ingestion as reported in the 2011 food consumption study only affects a few people in the population. The model calculates the
(Schijven et al., 2016; Van Rossum et al., 2011) cannot be used. The infection risk per person per event. For each event all the risks are
time of filling a glass or bottle for each person in the study area can be summed, and this sum is rounded toward the nearest integer number,

Fig. 4. Consumption amount. Top left: the number of glasses with

a Poisson distribution with λ = 1.43 glass (model 1998) and
λ = 2.5 (model 2016). Bottom left: the volume per glass with a
lognormal distribution (μ = −3.19 en σ = 1.485). Right: the
volume per person per day based on number of glasses and vo-
lume per glass, and maximum of 4.2 l; the model of 1998 de-
scribes the measured data of 1998 quite well. An increase in
consumption is assumed for 2016.

26
M. Blokker et al. Microbial Risk Analysis 8 (2018) 22–31

Fig. 5. Examples of probability of filling a glass or bottle for consumption over the day. The solid lines represent the probability distribution (normalised to a maximum of 1); the dotted
lines represent the cumulative probability. The black line illustrates how a random number between 0 and 1 (0.4) is translated to a time of the day. A) Total consumption (10:20); B)
Consumption at the kitchen tap, Dutch average (14:00); C) Consumption at the kitchen tap, with correction for total consumption at a specific node with a SIMDEUM demand of three
homes (17:10).

representing the median number of people that would be infected by analyses on the type of pathogen, the concentration and the ingestion
the event. For all simulated events the result is then graphically pre- amount and times, the same basic hydraulic water quality models were
sented as a cumulative probability distribution. used. For the sensitivity analyses on contamination location, the
opening of valves (which ones and when) and the sets of demand pat-
terns, a new hydraulic (and water quality) simulation was performed.
2.9. Stochastic modelling approach, sensitivity analysis and effect of
The sensitivity analysis was done with the basic concentration a
mitigation options
factor of 104 (Concentration Basic) and 107 (Concentration High). This
effectively means that the dose is so high that the response is equal to 1;
The reference scenario for which the infection risk is determined
i.e. every ingestion of contaminated water would lead to an infection.
involves an ingress of 104 Campylobacter per section volume during a
The results of the sensitivity analysis with the 104 concentration are all
repair and a Monte Carlo simulation from the appropriate probability
shown graphically; results of the sensitivity analysis with the 107 con-
distributions for (1) the contamination location (valve isolation sec-
centration are only shown graphically when a significant effect can be
tion), (2) the times of opening valves, (3) the flows in the DWDS, (4) the
seen.
times for consuming water, (5) the number of consumptions, (6) the
The table also shows possible mitigation options that a water
consumption volume. The infection risk for each person in the study
company has; and how these are modelled. Here some background
area was determined for 200 random events.
information on the mitigation options is given.
Table 3 summarises the values of the model parameters for the re-
One mitigation option is in fact standard procedure, viz. to open
ference scenario and how model input is varied in the sensitivity ana-
only the upstream valve, instead of releasing an entire section by
lysis. The variations in the input parameters in the sensitivity analysis
opening all valves. This will contain the possible contamination in a
have been described in the previous sections.
smaller part of the network where only the connections in the pre-
It should be noted that the concentration in QMRA model step 1 was
viously isolated section receive contaminated water. Note that it is as-
not varied in the hydraulic model. Rather, it was linearly adjusted after
sumed that the closed valves are all functioning well. Another
the model simulation was completed. This means that for the sensitivity

27
M. Blokker et al. Microbial Risk Analysis 8 (2018) 22–31

Table 3
Summary of the reference scenario and adaptations to it. The ‘x’ indicates that the step was not altered in the sensitivity analysis or mitigation.

QMRA model step Model parameter Reference scenario Sensitivity analysis Mitigation options Results in Fig. 6

Event Type of pathogen Campylobacter Giardia, Cryptosporidium, Rotavirus x Fig 6B

Location Pipe material and diameter Random location in the DWDS x Fig 6C (solid grey
dependence (Table 2) line)
Concentration Basic 104 per isolation section volume 10 times more, 10 times less x Fig 6A
Concentration High* 107 per isolation section volume 10 times more, 10 times less x not shown
Distribution Valves that are opened All valves of the isolated section x Only the upstream valve is opened Fig 6F (solid grey
are opened line)
Valve opening time According to reported times Random time (Fig. 3, dotted light Work more often during the night Fig 6C (dash
(Fig. 3, solid black line) grey line) (Fig. 3, dotted light grey line) dotted grey line)
Consumption Demand patterns that are Set of SIMDEUM patterns (day 1) Second set of SIMDEM patterns x Fig 6E
used (day 2)
Ingestion Time of tapping water* Proportional to time of use of Proportional to time of use total x Fig 6D (solid light
kitchen tap (Fig. 5C) demand (Fig. 5A) grey line)
Volume of ingested Lognormal distribution (Fig. 4) Fixed volume (0.3 L/cap.day) x
water
Number of ingestions* Poisson distribution, λ = 2.5 Fixed number: 1 Boil water advice: λ = 0.5 Fig 6D (solid grey
(Sec. 2.6) line)
Fixed number: 3 Fig 6D (dash
dotted line)
Infection Dose response relation Median and max of Respective DR for the other x Fig 6B
Campylobacter pathogens

⁎
The sensitivity analysis for time of tapping water and number of ingestions in combination with a high concentration are shown in Fig 6D (lines at the right).

mitigation option could be to work on repairs, or actually open the limit was set very low, so that typically any contamination would be
valves after finishing the work, during other times of the day, so e.g. detectable when sampled at the right time and location. A Monte Carlo
more often during the night. This mitigation option is evaluated already simulation of sampling was performed and the range of the detection
in the sensitivity analysis. One more mitigation option is to issue a boil likelihood was determined. This example discards the practical limita-
water advice. On average 80% of the people would follow this advice tions on finding E.coli in 100 ml samples due to incomplete mixing and
(Karagiannis et al., 2009). In the model a reduction of 80% of the association of pathogens with particles that may behave differently
number of tappings was used, i.e. the average number of glasses of than single free-floating organisms.
uncooked water is reduced from 2.5 to 0.5, but the amount of water per
glass still follows a lognormal distribution.
3. Results and discussion
Chlorination was not considered as a measure. First of all, the Dutch
drinking water companies do not use a residual disinfectant in the
3.1. Reference scenario
DWDS and do typically not use chlorine after repairs either
(Meerkerk, 2016; Smeets et al., 2009). Secondly, chlorine does not
Fig. 6 shows the results for the reference scenario of ingress of 104
eliminate Cryptosporidium and has limited effect on Giardia. And thirdly,
Campylobacter per section volume during a repair (dashed black line in
with a pathogenic contamination from the trench it is likely that also
all figures). The cumulative probability distribution shows that for
organic material entered the DWDS and this may increase chlorine
Campylobacter in 60% of the events 19 people or less are infected; and
decay, and therefore the effect of chlorine on viruses and Campylobacter
90% of the events infect 31 people or less. The maximum number of
is not known either.
people infected through a single event is 240. Note that the total
A more long term mitigation act would be to design the network
number of people in this area equals 4347.
differently. More branched systems may be inherently safer as valves
are not needed to contain the contamination in a section. By building
shorter sections, the number of connections in a section is lower and the 3.2. Sensitivity analysis
number of infected people can also be kept low. This mitigation option
is not studied here. Fig. 6 shows the results for the sensitivity analysis in comparison to
the reference scenario.
The contamination concentration has a large effect. A factor 10 in
2.10. Detection likelihood concentration, compared to the reference scenario with 104 per section,
leads to a factor of 4–5 in number of people infected (Fig. 6A). For
The simulated water quality at all nodes (result of QMRA model step higher concentrations the effect is much smaller. Comparing 108 per
2) was also used to determine the probability of detecting E. coli after section volume with 107 per section volume shows only an increase
ingress when using the standard Dutch sampling protocol. This protocol with a factor of 1.3 in number of infected people. At higher ingress
dictates that 12–24 hours after the repair a 100 ml sample should be concentrations, the concentration in contaminated water becomes so
taken (Meerkerk, 2016). In practice this means that a sample is taken on high that the risk of infection is primarily determined by the risk of
the next day during working hours. The sample is tested for E. coli as an exposure. Once exposed, the dose is high and the resulting risk of in-
indicator of faecal contamination. fection is also high. This effectively means that a person either ingests
The potential sampling locations are all model nodes within the no pathogens, or so many pathogens that every ingestion of con-
repair section, and the sampling times that were considered were any taminated water would lead to an infection. Since the concentration is
time 12–24 hours after the repair was finished, samples were taken such an important factor, more knowledge about the actually occurring
during office hours. To determine if a different sampling protocol would contamination concentrations would provide a more quantitative esti-
increase the detection likelihood, sampling at more optimal locations in mation of the infection risk. Also, the assumption that the contamina-
the repair section sooner after the repair were considered, i.e. at any tion is diluted and uniformly mixed in the total isolation section volume
time between 1–4 hours after the repair was finished. The detection is an assumption that needs to be evaluated. Flushing the section does

28
M. Blokker et al. Microbial Risk Analysis 8 (2018) 22–31

Fig. 6. Sensitivity analysis of the number of people affected per event (median infection risk). The reference in all cases is from events with a concentration C of Campylobacter, where
events show a preference in failure location (failure probability depends on pipe diameter and pipe material) and in failure time (opening times of valves are taken from Dutch registration
times), where the volume and amount of consumption is variable and the consumption time is related to the water use at the kitchen tap, SIMDEUM demand pattern set 1 was used, and
no boil water advice was issued, nor were any valves closed. A) for different concentrations of Campylobacter. B) for four different pathogens; C) for events without a preference in failure
location (no pipe diameter and pipe material dependence of failure probability) and without a preference in failure time (opening times of valves is not weighted over the day); D) the
amount of consumption fixed to either 1 or 3 glasses and with a preference time correlated with total water demand (Fig. 5A) instead of use at the kitchen tap (Fig. 5C). The left group of
lines are for the scenarios with concentration C., the right group of lines are for the scenarios with a concentration that is 1000 times higher; E) for SIMDEUM demand pattern sets 1 and 2.
F) Effect of mitigation options on the number of people affected per event for opening only one valve and thus creating a single side feed and for issuing a boil water advice.

not remove the contamination as a plug flow but rather mixes and di- distributing the ingestion of water over more than one glass (over the
lutes the contamination in the section. Flushing tests in a lab setup day) leads to a slightly higher risk. The sensitivity analysis at higher
suggest that this actually happens (van Bel, 2016). concentrations does show a clear effect of the number of consumptions.
The type of pathogen has a large effect on the risk because their In that case assuming a single ingestion of the total volume per day
dose response relations differ (Fig. 6B). Campylobacter is the most in- leads to a 1.7 lower number of infections and three ingestion volumes
fectious. Increasing concentrations 103 times did not affect the relative per day leads to a 1.3 higher number of infections compared to the
risk between pathogens. The same concentration for all pathogens was standard assumption (Poisson distributed number of ingestions). This
assumed, however some pathogens are more likely to occur at higher result agrees with a previous study (van Abel et al., 2013). The con-
concentration depending on contamination source and scenario. sumption time over the day did not have an effect at these higher
Therefore the combination of pathogen and its occurrence in the con- concentrations. This means that the number of consumptions is espe-
tamination together largely determine the risk. cially important at higher contamination levels, and the consumed
With respect to the characteristics of maintenance (Fig. 6C) it can be volume per event is not very important. The assumption that the in-
seen that the time of opening valves has a clear effect on the calculated crease in drinking water from 1998 to 2016 in the Netherlands
risk. Not taking into account the time of opening valves underestimates (Section 2.6) is due to the number of glasses, rather than an increase in
the risk with a factor of approximately 1.6 (at the median). Or, alter- volume per glass has not yet been substantiated with measurement
natively, doing maintenance in the night hours may reduce the infec- data, but seems to be a conservative approach. The data from
tion risk. The probability of pipe failure per pipe material and diameter Schijven et al. (2016) suggest that still the same amount of people
has little effect on the calculated frequency distribution of event risks in (around 15%) do not drink from the tap which would suggest a Poisson
this study area when compared to random failure locations. The reason distribution with λ = 0.43 would still fit the 2011 data that they have
may be that the pipe failure frequencies in Table 2 are not very dis- used. Schijven et al. (2016) have used two extra consumption moments
tinctive. The sensitivity analysis on this aspect with the 103 higher of ingestion during tooth brushing. This ingestion route may be an
concentrations showed the same effect as in the reference scenario, as it important one, because the number of ingestions is increased. More
has the same increase in number of infected people. evidence is needed to validate the assumptions of volume (1–5 ml) and
With respect to the consumption characteristics (Fig. 6D) the time of number of ingestions (twice per day).
consumption has a notably clear effect on the calculated risk. Not The effect of the different sets of SIMDEUM demand patterns
taking into account that the time of consumption is connected with the (Fig. 6E) is limited (factor of 1.1 at the median level), only in the
preparation of meals (water use at the kitchen tap) underestimates the maximum number of people per event there is some difference. It
risk. However, the number of consumptions has limited effect; should be mentioned that in the alternative consumption scenario

29
M. Blokker et al. Microbial Risk Analysis 8 (2018) 22–31

(SIMDEUM day2) the event locations and valve opening times are dif- ingested dose, while the type of pathogen highly impacts the resulting
ferent (new set of random event scenario's), which may also cause the infection risk due to the large differences the in dose response re-
difference. The sensitivity analysis on this aspect with the 103 higher lationships. The time of day that valves are opened after repairs, re-
concentrations showed the same effect as in the reference scenario, as it leasing the contamination, in combination with the time that drinking
has the same increase in number of infected people water is withdrawn from the tap influences the amount of contamina-
tion consumed versus flushed away through non-exposure uses. A
3.3. Effect of mitigation options standard QMRA approach that typically neglects the diurnal con-
sumption pattern may underestimate the risk. The consumption volume
Fig. 6 shows the results for the effect of mitigation options in is less important. Also, the effect of pipe specific failure frequencies in
comparison to the reference scenario. Limiting the contamination to a the model study area was limited. Issuing a boil water advice and
small amount of the network by opening only the upstream valve and opening only one valve before “releasing” the isolation section are ef-
issuing a boil water advice are very effective (Fig. 6F). The measure of fective mitigation options to reduce the infection risk per event by
the one sided feed is standard procedure in the Dutch approach 50%–80%. The statutory Dutch sampling protocol requires E. coli
(Meerkerk, 2016). The one sided feed reduces the number of infected analysis of a 100 ml sample taken the day after the repair. The modelled
people with a factor 2 (at low contamination concentrations, Fig. 6F) to probability of detecting a contamination under these circumstances is
4.5 (at high contamination concentrations, not shown). The issuing of a approximately 25%, because the contamination is rapidly flushed out of
boil water advice in the Netherlands is typically only done when there the customers’ taps, toilets and showers. If a sample is taken 1–4 hours
are indications of potential contamination during work such as the after the repair the modelled probability of detection exceeds 80%
proximity of a sewer line and after a positive E. coli sample. The boil when taken at the optimal location.
water advice with a reduction of 80% in uncooked water consumption
(factor 5 in ingestion) reduces the number of infected people with a Acknowledgements
factor 4. The combination of the two mitigation options will even fur-
ther reduce the number of infected people. Equally spreading work over The research was funded by the Dutch drinking water companies
24 hours per day, which means working relatively more often during through the Joint Research Programme.
the night hours (Fig. 6C, the dash dotted grey line means equal like-
lihood of working at all hours of the day) has a positive effect on the Supplementary materials
calculated risk; this mitigation option would reduce the risk with a
factor of approximately 1.6 (at the median). Supplementary material associated with this article can be found, in
the online version, at doi:10.1016/j.mran.2017.12.002.
3.4. Probability of detection
References
The theoretical probability of finding E. coli, assuming the 104 in-
gress load, in the 12–24 hours after the contamination took place is Abbaszadegan, M., Lechevallier, M., Gerba, C., 2003. Occurrence ofViruses in US
limited. The reason is that the contamination is rapidly being flushed Groundwaters. J. Am. Water Works Assoc. 95, 107–120.
Besner, M.-C., Lavoie, J., Morissette, C., Payment, P., Prévost, M., 2008. Effect of water
out of the customers’ taps in the isolated section, most likely through main repairs on water quality. Am. Water Works Assoc. J. 100 (7), 95.
toilet flushing or showering water. In the situation of the standard Besner, M.-C., Prévost, M., Regli, S., 2011. Assessing the public health risk of microbial
sampling protocol there is a 50% probability that the contamination is intrusion events in distribution systems: conceptual model, available data, and
challenges. Water Res. 45 (3), 961–979.
not detected, for the other 50% of the cases the median detection Blokker, E.J.M., Vreeburg, J.H.G., Beverloo, H., Klein Arfman, M., van Dijk, J.C., 2010a. A
likelihood is 40%. On average the detection likelihood is 25.7%. If a bottom-up approach of stochastic demand allocation in water quality modelling.
sample would be taken between 1 and 4 hours after the repair, the Drink. Water Eng. Sci. 3 (1), 43–51.
Blokker, E.J.M., Vreeburg, J.H.G., van Dijk, J.C., 2010b. Simulating residential water
average detection likelihood increases to 61.4%. A further increase can demand with a stochastic end-use model. J. Water Resour. Plan. Manag. 136 (1),
be reached by aiming to take the sample at the location with the highest 19–26.
detection likelihood, this leads to a detection likelihood of 81.1%. Blokker, E.J.M., Smeets, P., Medema, G., 2014. QMRA in the drinking water distribution
system. In: Proceedings of the 16th Conference on Water Distribution System
The detection probability can also be linked to the probability of
Analysis, WDSA. Procedia Engineering.
infection, using the indicator-to-pathogen ratio's used in Blokker, E.J.M., Moerman, A., Smeets, P.W.M.H., 2016. QMRA Van Het Distributienet.
van Lieverloo et al. (2007). In case a 100 ml sample would contain 1 E. KWR, Nieuwegein BTO 2016.017.
coli, this would mean a Campylobacter concentration of a factor 3 to Craun, G.F., Calderon, R.L., 2001. Waterborne disease outbreaks caused by distribution
system deficiencies (PDF). J.-Am. Water Works Assoc. 93 (9), 64–75.
1000 (depending on surface water or sewage as the source of the Craun, M.F., Craun, G.F., Calderon, R.L., Beach, M.J., 2006. Waterborne outbreaks re-
contamination) lower, i.e. 0.01 to 3 Campylobacter per litre. In the ported in the United States. J. Water Health 4 (Suppl 2), 19–30.
QMRA model this would be the same as setting the concentration of the Davis, M.J., Janke, R., 2009. Development of a probabilistic timing model for the in-
gestion of tap water. J. Water Resour. Plan. Manag. 135 (5), 397–405.
reference scenario approximately a factor of 3 to 1000 lower. Fig. 6A Drinking Water Decree, 2015. Drinking Water Decree. [cited 2015 27 November 2015];
shows that a factor of 10 lower leads to a maximum of ca. 20 infected Available from: http://wetten.overheid.nl/BWBR0030111/geldigheidsdatum_27-11-
persons. The QMRA model showed that in order to have the number of 2015.
Foekema, H., van Thiel, L., 2011. Water Use At Home 2010. TNS NIPO, Amsterdam.
infected persons to be less than 1 at the 95 percentile certainty level, a Hoogenboezem, W., H. Ketelaars, G. Medema, G. Rijs, and J. Schijven, Cryptosporidium
concentration of 3.8·10−2 lower than the reference scenario (104 pa- en Giardia: voorkomen in rioolwater, mest en oppervlaktewater met zwem-en
thogens diluted in a section of on average 4 m3) is required. It also drinkwaterfunctie. RlWA/RIVM/RIZA-rapport. ISBN 9036953324, 2001.
Hrudey, E.J., 2004. Safe Drinking Water: Lessons from Recent Outbreaks in Affluent
showed that a serious contamination event with more than 10 people
Nations. IWA publishing.
being infected should be detectable when sampling for E. coli when Hunt, R.J., Borchardt, M.A., Richards, K.D., Spencer, S.K., 2010. Assessment of sewer
within 1–4 hours. source contamination of drinking water wells using tracers and human enteric
viruses. Environ. Sci. Technol. 44 (20), 7956–7963.
Jones, K., 2001. Campylobacters in water, sewage and the environment. J. Appl.
4. Conclusions Microbiol. 90 (S6), 68S–79S.
Karagiannis, I., Schimmer, B., de Roda, H.A., 2009. Compliance with boil water advice
A stochastic Quantitative Microbial Risk Assessment model was following a water contamination incident in the Netherlands in 2007. Euro Surveill.
14 (12), 334–338.
developed to assess the infection risk due to faecal contamination Karim, M.R., Abbaszadegan, M., LeChevallier, M., 2003. Potential for pathogen intrusion
events after repairs of drinking water mains. The results show that the during pressure transients. J.-Am. Water Works Assoc. 95 (5).
contamination concentration is the most important parameter for the Kirmeyer, G.J., Martel, K., 2001. Pathogen Intrusion Into the Distribution System.

30
M. Blokker et al. Microbial Risk Analysis 8 (2018) 22–31

American Water Works Association. human health risks from exposure to recreational waters impacted by human and
Lambertini, E., Spencer, S.K., Kieke Jr, B.A., Loge, F.J., Borchardt, M.A., 2011. Virus non-human sources of faecal contamination. Water Res. 44 (16), 4674–4691.
contamination from operation and maintenance events in small drinking water dis- Teunis, P.F.M., Xu, M., Fleming, K.K., Yang, J., Moe, C.L., LeChevallier, M.W., 2010.
tribution systems. J. Water Health 9 (4). Enteric virus infection risk from intrusion of sewage into a drinking water distribu-
LeChevallier, M., Gullick, R., Karim, M., Friedman, M., Funk, J., 2003. The potential for tion network. Environ. Sci. Technol. 44 (22), 8561–8566.
health risks from intrusion of contaminants into the distribution system from pressure van Abel, N., Blokker, E.J.M., Smeets, P.W.M.H., Meschke, J.S., Medema, G.J., 2013. The
transients. J Water Health 1, 3–14. number of consumption events is an important parameter in QMRA of drinking water
Meerkerk, M.A., 2016. PCD 1-4 Hygienecode Drinkwater. Opslag, Transport En contaminated during short term events. In: Proceedings of the 17th International on
Distributie. KWR, Nieuwegein PCD 1-4 (2016). Health-Related Water Microbiology Symposium Water Micro. Florianópolis, Brazil.
Mons, M.N., van der Wielen, J.M.L., Blokker, E.J.M., Sinclair, M.I., Hulshof, K.F.A.M., Van Abel, N., Blokker, E.J.M., Smeets, P.W.M.H., Meschke, J.S., Medema, G.J., 2014.
Dangendorf, F., Hunter, P.R., Medema, G.J., 2007. Estimation of the consumption of Sensitivity of quantitative microbial risk assessments to assumptions about exposure
cold tap water for microbiological risk assessment: an overview of studies and sta- to multiple consumption events per day. J. Water Health 12 (4), 727–735.
tistical analysis of data. J. Water Health 5 (1), 151–170. van Bel, N., 2016. Effectiviteit Van Spuien En Chloren Op De Verwijdering Van Microbiële
National Research Council, 2006. Drinking Water Distribution systems: Assessing and Verontreinigingen in Het Distributienet - een pilotonderzoek. KWR, Nieuwegein BTO
Reducing Risks. National Academies Press. 2016.075.
Nygård, K., Wahl, E., Krogh, T., Tveit, O.A., Bøhleng, E., Tverdal, A., Aavitsland, P., 2007. van Lieverloo, J.H.M., Blokker, E.J.M., Medema, G., Hambusch, B., Pitchers, R., Stanfield,
Breaks and maintenance work in the water distribution systems and gastrointestinal G., Stanger, M., Agutter, P., Lake, R., Loret, J.F., Soyeux, E., 2006. Contamination
illness: a cohort study. Int. J. Epidemiol. 36 (4), 873–880. During Distribution. KWR, Nieuwegein.
Paul, M., Wolf, L., Fund, K., Held, I., Winter, J., Gallert, C., Hötzl, H., 2004. van Lieverloo, J.H.M., Blokker, E.J.M., Medema, G.J., 2007. Quantitative microbial risk
Microbiological condition of urban groundwater in the vicinity of leaky sewer sys- assesment of distributed drinking water using faecal indicator incidence and con-
tems. Acta Hydrochimica et Hydrobiologica 32 (4–5), 351–360. centrations. J. Water Health 5 (1), 131–149.
Risebro, H.L., Doria, M.F., Andersson, Y., Medema, G., Osborn, K., Schiosser, O., Hunter, Van Rossum, C., H. Fransen, J. Verkaik-Kloosterman, E. Buurma-Rethans and M. Ocké,
P.R., 2007. Fault tree analysis of the causes of waterborne outbreaks. J. Water Dutch National Food Consumption Survey 2007-2010: Diet of children and adults
Health 5. aged 7 to 69 years. RIVM rapport 350050006, 2011.
Sales Ortells, H., 2015. Health Impact Assessment of New Urban Water Concepts. TU Vreeburg, J., Vloerbergh, I., van Thienen, P., de Bont, R., 2013. Shared failure data for
Delft, Delft. strategic asset management. Water Sci. Technol. Water Supply 13 (4), 1154–1160.
Schijven, J.F., Teunis, P.F., Rutjes, S.A., Bouwknegt, M., de Roda Husman, A.M., 2011. Westrell, T., Schönning, C., Stenström, T.-A., Ashbolt, N., 2004. QMRA (quantitative
QMRAspot: a tool for quantitative microbial risk assessment from surface water to microbial risk assessment) and HACCP (hazard analysis and critical control points)
potable water. Water Res. 45 (17), 5564–5576. for management of pathogens in wastewater and sewage sludge treatment and reuse.
Schijven, J., Forêt, J.M., Chardon, J., Teunis, P., Bouwknegt, M., Tangena, B., 2016. Water Sci. Technol. (50), 23–30.
Evaluation of exposure scenarios on intentional microbiological contamination in a Yang, J., LeChevallier, M.W., Teunis, P.F., Xu, M., 2011. Managing risks from virus in-
drinking water distribution network. Water Res. 96, 148–154. trusion into water distribution systems due to pressure transients. J. Water Health 9
Smeets, P., Medema, G., Van Dijk, J., 2009. The Dutch secret: how to provide safe (2), 291–305.
drinking water without chlorine in the Netherlands. Drink. Water Eng. Sci. 2 (1), Yates, M.V., Gerba, C.P., Kelley, L.M., 1985. Virus persistence in groundwater. Appl.
1–14. Environ. Microbiol. 49 (4), 778–781.
Soller, J.A., Schoen, M.E., Bartrand, T., Ravenscroft, J.E., Ashbolt, N.J., 2010. Estimated

31