Disease burden due to gastrointestinal infections among people living along the major wastewater system in Hanoi, Vietnam

Background: Despite recent improvements of wastewater treatment capacities in urban areas of Hanoi, Vietnam, microbial pollution is still considerable. There is a paucity of burden estimates due to gastrointestinal infection in people living along the wastewater system, and among people who are in direct contact with the wastewater, such as farmers using wastewater in agriculture and aquaculture. Methods: A quantitative microbial risk assessment (QMRA) was pursued focussing on four population groups characterised by different levels of exposure to wastewater: (i) workers maintaining the wastewater conveyance and treatment systems; (ii) urban farmers using wastewater from To Lich River; (iii) community members in urban areas exposed to ﬂooding events in the districts of Hoang Mai and Thanh Tri; and (iv) peri-urban farmers in Thanh Tri district, where Red River water is used for agriculture and aquaculture. The QMRA was developed on the basis of measured concentration of Escherichia coli and Salmonella spp. and Ascaris spp. eggs in water samples. Published ratios between measured organisms and pathogenic strains of norovirus, rotavirus, Campylobacter spp . , pathogenic E. coli , pathogenic Salmonella spp., Cryptosporidium spp. and Ascaris lumbricoides were employed to estimate annual risk of gastroin- testinal infection and disease burden. Results: The QMRA estimated a disease burden of 0.011 disability-adjusted life years (DALYs) per person per year in urban farmers, 0.006 DALYs for sanitation workers, 0.0 0 05 DALYs for urban communities at risk of ﬂooding events and 0.0 0 04 DALYs for peri-urban farmers. Urban farmers had considerably higher incidence estimates for gastrointestinal disease episodes per year (2.0) compared to the other exposure groups ( ≤ 1.0). Conclusions: Urban farmers using wastewater from To Lich River have a high gastrointestinal disease burden, which is about 100 times larger than the health-based targets for wastewater use set by the World Health Organization. These ﬁndings are of direct public health relevance and call for upgrading Hanoi’s wastewater system to reduce microbial contamination. Finally, this study presents a ﬁrst example on how to link QMRA to a sanitation safety planning (SSP) approach in an Asian context and its ﬁndings are interesting in the frame of Sustainable Development Goal (SDGs) #6. © 2016 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license. ( http://creativecommons.org/licenses/by/4.0/ )


Introduction
In many urban centres in low-and middle-income countries (LMICs) large quantities of untreated or partially treated wastewater are discharged into the environment ( Evans et al., 2012;Drech-Considerably larger volumes of only partially treated or untreated wastewater are used in agriculture (e.g. through diversion of water from wastewater receiving surface water bodies) ( Fuhrimann et al., 2014( Fuhrimann et al., , 2015WHO, 2015 ). In Hanoi, for example, an estimated 660,0 0 0 farmers use wastewater for vegetable and rice farming and aquaculture on a surface area of approximately 44,0 0 0 ha ( Raschid-Sally and Jayakody, 2008 ).
While the use of wastewater is an important livelihood strategy for poor urban farmers, there are important public health concerns ( Winkler et al., 2016 ). In Southeast Asian cities, such as Hanoi, large quantities of insufficiently treated effluents contaminate surface waters and soils, resulting in high concentrations of pathogenic organisms and toxic chemicals ( Kuroda et al., 2015;Fuhrimann et al., 2016a ). This is explained by the fact that most of the 6.7 million people living in Hanoi rely on flush toilets directly connected to septic tanks from where the partially treated effluents are discharged into a complex network of drainage channels, which additionally receive effluents from industries ( Fuhrimann et al., 2016b ). Of note, faecal sludge from septic tanks is often informally discarded into the environment; in many cases directly into agricultural fields or ponds used for aquaculture ( Bassan et al., 2015 ). In response, wastewater flows have been largely controlled by channelization of the main urban rivers, water gates and sedimentation pond systems to prevent flooding and to treat wastewater before use in agriculture and aquaculture ( Nguyen and Parkinson, 2005;World Bank, 2013 ).
In a recent water quality assessment we found that, despite effort s to improve Hanoi's wastewater conveyance and treatment systems, water deriving from the wastewater channels being used in agricultural fields for irrigation is heavily contaminated with total coliforms (TC), Escherichia coli and Salmonella spp. ( Fuhrimann et al., 2016a ). Observed values were up to 110-fold above Vietnamese discharge limits for restricted agriculture and up to 260fold above the World Health Organization (WHO)'s tolerable safety limits for unrestricted agriculture ( Fuhrimann et al., 2016a ). Additionally, a cross-sectional epidemiological survey revealed high prevalence of intestinal parasite infections in peri-urban and urban farmers (up to 30%), general communities (up to 10%) and workers maintaining the wastewater channels (10%) ( Fuhrimann et al., 2016b ). In the aforementioned study, the prevalence of selfreported diarrhoea episodes (recall period: 2 weeks) in adults in peri-urban communities, urban farmers and sanitation workers was 8-12%. These observations suggest that treatment efficacy of Hanoi's wastewater management system is insufficient in preventing microbial contamination. Consequently, people exposed to wastewater are at risk of gastrointestinal infection. However, prior research does not provide an estimate of the magnitude of the disease burden caused by specific pathogenic organisms ( Katukiza et al., 2013;Machdar et al., 2013 ). The 2010 Global Burden of Disease (GBD) study estimated the burden caused by diarrhoeal diseases at 0.002 disability-adjusted life years (DALYs) per person per year (pppy) for an average Vietnamese, which is considerably higher than the WHO's health-based target for the exposure to wastewater (a tolerable additional burden between 10 −6 and 0.0 0 01 DALYs pppy is suggested) ( WHO, 2006;Institute for Health Metrics and Evaluation, 2015 ). Context-specific disease burden estimates are, however, relevant to compare the impact of individual pathogens for different population groups, and to govern control measures in the wastewater system ( Drechsel and Seidu, 2011 ).
Here we present a quantitative microbial risk assessment (QMRA), which is linked a sanitation safety planning (SSP) approach ( WHO, 2015 ) with three specific objectives: (i) to estimate the disease burden due to gastroenteritis resulting from exposure to water-borne pathogens along the major wastewater system in Hanoi ( Fuhrimann et al., 2016a ); (ii) to validate model estimates with findings obtained from a cross-sectional epidemiological survey ( Fuhrimann et al., 2016b( Fuhrimann et al., , 2016c; and (iii) to compare disease burden estimates with national and international standards and estimates ( Institute for Health Metrics and Evaluation, 2015;Fuhrimann et al., 2016d ).

Ethical considerations
The study protocol was approved by the institutional research commission of the Swiss Tropical and Public Health Institute (Swiss TPH; Basel, Switzerland; reference no. FK #106). Ethical approval was obtained from the ethics committee in Basel, Switzerland (EKBB; reference no. 137/13) and the Hanoi School of Public Health (HSPH; Hanoi, Vietnam; reference no. 010/2014/YTCC -HD3). This study is registered with the clinical trial registry ISRCTN (identifier: ISRCTN13601686).

Study area
Hanoi is located in the north of Vietnam, situated at the Red River delta (geographical coordinates: 21 °01 42.5 N latitude and 105 °51 15.0 E longitude). The climate is sub-tropical with the main rainy season occurring from April to September and a yearround humidity of 80-90% ( Climate-Data.org, 2015 ). The wastewater flows through the city, from north to south, along a topographic gradient of 20 m to 5 m above mean sea level in three main rivers: To Lich, Nhue and Red ( Nguyen and Parkinson, 2005;Kuroda et al., 2015 ) ( Fig. 1 ).
Our study focussed on three specific parts of Hanoi's wastewater system: (i) the To Lich River, which receives most of Hanoi's storm-and wastewater and the Yen Son treatment plant (operated by the Hanoi Sewerage and Drainage Company (HSDC)); (ii) two typical urban communes (Bang B and Tam Hiep) where wastewater from To Lich River is used for agriculture and aquaculture; and (iii) the peri-urban commune Duyen Ha where Red River water (considered as clean) is used in agriculture and aquaculture ( Fuhrimann et al., 2016a( Fuhrimann et al., , 2016b.

Hazard identification
The hazards considered for the QMRA are seven pathogenic organisms giving rise to gastroenteritis in Vietnam and elsewhere: two types of viruses (norovirus and rotavirus), three different bacteria ( Campylobacter spp., Salmonella spp. and E. coli ), one intestinal protozoon ( Cryptosporidium spp.) and one species of soil-transmitted helminths ( Ascaris lumbricoides ) ( Becker et al., 2013;Katukiza et al., 2013;Barker et al., 2014;Gibney et al., 2014 ;Fuhrimann et al., 2016d ). These pathogens are characterised by faecal-oral transmission, can persist for weeks or months in the environment and are difficult to inactivate with conventional wastewater treatment processes such as those applied in Hanoi ( Kuroda et al., 2015;Fuhrimann et al., 2016a ). Our selection is further justified on the following grounds: • Rotavirus is one of the leading cause of childhood diarrhoea ( Bodhidatta et al., 2007 ). • Norovirus is the major cause of diarrhoeal disease in adults and its secondary attack rate is known to be high causing epidemic situation ( Barker, 2014;. • Campylobacter spp. are zoonotic bacteria that cause campylobacteriosis, with Campylobacter jejuni being a common cause of diarrhoea in LMICs ( Kaakoush et al., 2015 ). • E. coli belongs to the normal gastrointestinal microflora of warm-blooded animals and humans, whilst different types of E. coli (e.g. enterotoxigenic E. coli , enteroinvasive E. coli and enterohemorrhaghic E. coli ) have been associated with diarrhoeal disease or even worse public health impact such as hemolytic uremic syndrome (HUS) ( Okeke, 2009 ). As a simplification for the current QMRA, we refer to pathogenic E. coli that sums up all diarrhoea causing E. coli , whilst using the dose-response relation for probability of illness as derived for E coli O157:H7. • Salmonella spp. have more than 2,500 sero-groups; yet, of concern for human health are only S. typhi and S. para-typhi A, B and C, and the enteric Salmonella strains ( Kariuki et al., 2015 ). As a simplification for the current QMRA, we refer to pathogenic Salmonella spp., which includes the diarrhoea causing Salmonella spp. S. typhi, S. para-typhi A, B and C, S. Enteritidis and S. Typhimurium. • Cryptosporidium spp. is a zoonotic intestinal protozoon that can result in severe health implication in children and immunocompromised individuals (e.g. HIV-positive people). • A. lumbricoides is the most widespread soil-transmitted helminth and is highly endemic in Vietnam. Parasite eggs are known to persist in the environment longer than any of the other soil-transmitted helminth species ( Brooker et al., 2009 ).

Exposure assessment
The exposure scenarios for the QMRA are based on information derived from a cross-sectional survey, consisting of a questionnaire survey and examination of stool samples from 618 people in the districts of Hoang Mai and Thanh Tri ( Fuhrimann et al., 2016b ). The results of the cross-sectional survey were used to obtain mode and frequency of exposure to wastewater and compare prevalence of parasitic infections and self-reported diarrhoea (recall period: 2 weeks) with estimates taking from the QMRA. As the focus of the model was to determine direct contact to contaminated water (considering all modes of pathogen transmission where contaminated water can be swallowed in the respective exposure scenario), the QMRA only included accidental ingestion of contaminated water as an exposure pathway. Other potential exposure pathways were excluded, such as: (i) ingestion of contaminated soil, dermal contact, inhalation and drinking of potentially contaminated water (lack of data); (ii) consumption of contaminated food crops (no data on degree of contamination of food crops); and (iii) exposure to contaminated water used for bathing or washing clothes (uncommon local practice).
Four exposure scenarios along the three selected study areas (To Lich River, wastewater use and Red River) were developed and assumptions about exposure groups, number of people exposed, exposure frequency and volume of ingested water were made ( Figs. 1 and 2 ).
Exposure to water from To Lich River: • Scenario 1 ( S flooding ): Urban communities (all age groups) in Bang B and Tam Hiep living in close proximity to the To Lich River are prone to flooding events. For instance, 5% of the people living in these communities reported to be at risk of flooding events (i.e. 795 out of 15,900 people) ( Fuhrimann et al., 2016b ). According to HSDC, four flooding events occurred during the rainy season in 2013; the year before our epidemiological survey ( Fuhrimann et al., 2016b ). During a flooding event, ingestion of water due to unintentional immersion is assumed to range between 10 and 30 mL ( McBride et al., 2013 ). • Scenario 2 ( S working ): There are 450 registered workers employed by the HSDC who are in charge of the maintenance of the To Lich River and the operation of the wastewater treatment plants (Yen Son). On average, a worker is on duty 322 days per year. Most of the workers wear gloves (91%), which we considered as a proxy for the level of awareness and preparedness to avoid accidental ingestion of contaminated water ( WHO, 2006;Fuhrimann et al., 2016d ). It was assumed that the worst case scenario (an accidental ingestion of 10-50 mL of wastewater per working day) was reduced to ingestion of 0.9-4.5 mL per working day ( WHO, 2006;Labite et al., 2010;Mara and Bos, 2010 ).
Exposure to water use from To Lich River in agriculture fields of urban farming communes in Hanoi: • Scenario 3 ( S farming U ): Urban farmers living in Bang B village or Tam Hiep commune using wastewater from To Lich River were selected for the model. One third of the community (an estimated 5,300) are involved in urban farming (mainly rice, morning glory, neptunia and watercress) or aquaculture. Thus, the likelihood of accidental ingestion of water is considerable. On average, farmers in Bang B village and Tam Hiep commune reported to work 338 days per year in flooded agricultural fields and they are in contact with irrigation water on a daily basis. Three out of four workers (74%) wear gloves, which we considered as a proxy for the level of awareness and preparedness to avoid accidental ingestion of contaminated water ( WHO, 2006;Fuhrimann et al., 2016d ). It was assumed that the worst case scenario (an accidental ingestion of 10-50 mL of wastewater per working day) was reduced to ingestions between 2.6 mL and 13 mL per working day (  Exposure to water used from Red River in agriculture fields of peri-urban farming communes in Hanoi: • Scenario 4 ( S farming PU ): A typical peri-urban farming community living in Duyen Ha commune, 10 km away from the outskirts of Hanoi. Farmers using the irrigation water from Red River, wells or local drains, which are not contaminated with the city's wastewater but contaminated with household effluents. About 38% of the people work in agriculture (i.e. 580 urban famers). On average, farmers reported to work 338 days per year, though, fields are irrigated only every second day. 82% of the workers wear gloves, which we considered as a proxy for the level of awareness and preparedness to avoid accidental ingestion of contaminated water ( WHO, 2006;Fuhrimann et al., 2016d ). It was assumed that the worst case scenario (an accidental ingestion of 10-50 mL of wastewater per working day) was reduced to ingestions between 1.8 mL and 9 mL per working day ( WHO, 2006;Labite et al., 2010;Mara and Bos, 2010 ).

Measurements of pathogens along the wastewater system
Water quality was tested for E. coli , TC, Salmonella spp. and helminth eggs between April and June 2014 ( Fuhrimann et al., 2016a ). The ratio between measured pathogens ( p path ) and E. coli is assumed to vary between 10 −6 and 10 −5 (rotavirus, norovirus and Campylobacter spp.) or between 10 −7 and 10 −6 ( Cryptosporidium spp.). The ratio between pathogenic and non-pathogenic strains of E. coli and Salmonella spp. ( p path ) is assumed to vary between 7.6 × 10 −4 and 1 × 10 −2 ( Shere et al., 2002;WHO, 2006;Soller et al., 2010;Barker et al., 2014;Hynds et al., 2014 ). For Ascaris spp. it is assumed that each egg detected represents an A. lumbricoides ( p path = 1, not considering the occurrence of other helminth species such as, for example, A. suum ) ( Mara and Sleigh, 2010 ).

QMRA structure, implementation and analysis
Our QMRA approach has been described elsewhere ( Fuhrimann et al., 2016d ). We adhered to the WHO 2006 guidelines and the improved Karavarsamis-Hamilton method to determine annual disease and infection risks ( WHO 2006;Karavarsamis and Hamilton 2010;. As summarised in Table 1 , spatial and temporal variability of the number of colony forming unit (CFU) E. coli and Salmonella spp., we fitted normal distributions to the log-transformed enumeration data on concentration in the water ( C water ) ( Fuhrimann et al., 2016a ). A maximum likelihood estimation (MLE) method was used, allowing inclusion of censored data and accounting for the abundance, while considering the measured prevalence of the indicator bacteria in the water along the four systems ( Lorimer and Kiermeier, 2007 ), in Excel 2013 (Microsoft Corporation, Redmond; WA, USA). As a result, the data fitting provided estimates for the true prevalence of contaminated water samples, and the distribution of concentrations in these contaminated samples. For Ascaris spp. eggs, this approach is not possible as only 17 out of 216 samples were positive. Hence, with p = 0.08 a positive count is expected, between 1 and 100 eggs/L, which is included with uniform distribution on a log scale ( Fuhrimann et al., 2016a ). Project evaluation and review techniques (PERT) distributions are fitted to minimum, most likely and maximum ratio of pathogen concentration per E. coli ( p path ) for rotavirus, norovirus, Campylobacter spp. and Cryptosporidium spp. Uniform distribution were fitted to E. coli and Salmonella spp. ratio of pathogen concentration per measured E. coli and Salmonella spp. ( WHO, 2006;Katukiza et al., 2013 ). This is implemented in the model by assuming that a fraction p path of the ingested volumes of water consists of a pathogenic strain of the bacterial species. PERT distributions are also fitted to assumed minimum, most likely and maximum ingestion rates (volume ( V ) in mL water) per exposure event. In a Monte Carlo simulation, values are sampled for these three variables and the ingested amount of pathogens (dose; d ) is calculated as: The variation in C water is implemented as variability per exposure event, the variation in p path and V is implemented as variability per person (i.e. for practical reasons it had the same value for all exposure events for one person in one iteration of the Monte Carlo simulation). As ingested bacteria are discrete units, assumed to be homogeneously distributed in the water, ingested doses are assumed to be Poisson distributed (d ∼ Poisson ( d ) as e.g. in Nauta et al. (2012 ).
Doses ( d ) are used as input in the dose-response relations to obtain the probability of illness P ill (d) ( Eqs. 1 to 6 ). Monte Carlo simulations are performed for 10 0,0 0 0 iterations using @Risk, version 6 (Palisade Corporation; Newfield, NY, USA), where one iteration simulates all the n exposure events ( n different doses d ) and associated P ill (d) of one person in a year. The expected frequency of illness for a person per year (which, in our approach, might be above one) can be calculated as the sum of the n values of P ill (d) obtained (without considering immunity) ( Haas et al., 2014 ) . Model outputs are presented as number of cases per year, DALYs pppy and total DALYs per year (according to published burden estimate for gastroenteritis and fatality rates and adapted to the average life expectancy of 72 years in Vietnam ( Salomon et al., 2012;Gibney et al., 2014 ).

Dose-response models
The same dose-response models for the various pathogens are used as described elsewhere ( Fuhrimann et al., 2016d ) to determine the relationship between quantity of exposure (i.e. number of organism ingested) and the effective health outcome (i.e. infection and illness) ( Haas et al., 2014 ). For the QMRA, Beta-Poisson doseresponse models for rotavirus, Campylobacter spp., pathogenic E. coli , pathogenic Salmonella spp. and A. lumbricoides were employed ( Haas et al., 1999( Haas et al., , 2014Teunis and Havelaar, 20 With a median infectious dose defined as For norovirus, a hypergeometric function was used ( Teunis et al., 2008 b). We employed an approximation for the mean probability of infection ( Haas, 2002 ) of the Beta-Poisson dose-response model: where Г(.) represents Eulers gamma function ( McBride et al., 2013 ). For Cryptosporidium spp., an exponential model was used ( Westrell et al., 2004;de Man et al., 2013;McBride et al., 2013 ).
In brief, P I ( d ) represents the probability of infection, d is a single dose of the pathogen, whereas the pathogen infectivity constants α, β and r characterise the dose-response relationship.
To account for the proportion of infections that turn into symptomatic cases ( P ill ( d )), we used for each pathogen a constant value ( λ) (i.e. illness to infection ratio):  log 10 (CFU/100 mL) Normal (3.0;0.2) * Ascaris spp. log 10 eggs/1 L Uniform (0;1.4) * * , prevalence = 0.08 ( C water ) Wastewater-fed agricultural fields in Bang B and Tam Hieap Escherichia coli log 10 (CFU/100 mL) Normal (6.0;0.7) * Salmonella spp. log 10 (CFU/100 mL) Normal (2.0;0.6) * Ascaris spp. log 10 eggs/1 L Uniform (0;1.4) * * ; prevalence = 0.08 ( C water ) Drainage system Duyen HA Escherichia coli log 10 (CFU/100 mL) Normal (5.4;0.5) * Salmonella spp. log 10 (CFU/100 mL) Normal (2.6;0.6) * Ascaris spp. log 10 eggs/1 L Uniform ( ( Fuhrimann et al., 2016d ). When assuming that each exposure event i is independent and that there is no acquired immunity after an infection, with P ill/ h ( d i ) representing the probability of illness, which is a function of the ingested dose d i at exposure event i for each of the seven pathogens (or hazards) h ( Eq. 6 ), then for an average of n h exposures to the hazard per person per year, with population size Pop E , the expected number of cases is

Estimation of disease burden
The disease burden due to human exposure is expressed in DALYs. This metric combines morbidity (years lived with disability) and premature death (years of life lost) ( Murray et al., 2012 ). For each pathogen h , DALYs per case of gastrointestinal illness ( DALY h ) are calculated as the sum of the product of the probability of developing disease symptom j (i.e. j = mild, moderate and severe diarrhoea or death) given the illness occurs, relative frequency of the symptom ( f j ), duration of the developed symptom in years ( D j ) and the respective severity factor ( S j ) ( Salomon et al., 2012 ): Mortality is calculated according to the average life expectancy at birth in Vietnam of 72 years ( Katukiza et al., 2013 ). Note that sequelae such as HUS, Guillain-Barré syndrome, reactive arthritis or irritable bowel syndrome are not considered in the model. The total disease burden ( Total DALYs,h ) per hazard is the product of cases ( Cases h ) and DALYs per pathogen: The total disease burden for all hazards together is

Incidence of gastroenteritis
The combined estimated incidence ( Inc comb ) for gastroenteritis episodes per year due to the exposure to contaminated water was highest for urban farmers ( S farming U ), and sanitation workers ( S working ) who suffer, on average, 2.0 and 1.0 gastroenteritis episodes per person per year ( Fig. 3 , Table 3 ). Incidence estimates were considerably lower for people exposed to flooding events ( S flooding ) and peri-urban farmers ( S farming PU ) ( > 0.1 episodes per person per year for both groups) ( Table 3 )

Number of gastroenteritis cases per year
Among the 7,125 exposed people in the districts of Hoang Mai and Thanh Tri, a total of 11,073 cases of gastroenteritis caused by any of the seven pathogens were estimated due to exposure to wastewater for the four scenarios and the duration of one year ( Fig. 3 , Table 2 ). Gastrointestinal infection due to rotavirus, E. coli and Campylobacter spp. contributed most to the total cases (59%, 18% and 13%, respectively). Taken together, 95.0% of all cases were concentrated in the 5,300 urban farmers ( S farming U : 10,511 cases).

Total disease burden per year
Across all 7,125 exposed people and four scenarios, our model estimated a burden of 62.6 DALYs per year due to exposure to wastewater in Hoang Mai and Thanh Tri district. The main responsible pathogens were E. coli , rotavirus and Campylobacter spp., accounting for 42%, 39% and 14% of the burden, respectively. Urban farmers ( S farming U ) were most vulnerable with a burden of 59.5 DALYs.

Disease burden per person per year
Combined DALYs pppy for all scenarios (summed-up for the seven pathogen and all exposed individuals) were far above the revised WHO reference level of 0.0 0 01 DALYs pppy ( Table 1 ). The highest disease burden was estimated at 0.011 DALYs pppy in urban farmers ( S farming U ), followed by sanitation workers ( S working = 0.0057 DALYs pppy), urban communities at risk of flooding events ( S flooding = 0.0 0 05 DALYs pppy) and peri-urban farmers ( S farming PU = 0.0 0 04 DALYs pppy). In terms of different pathogens, E. coli, rotavirus and Campylobacter spp. had the largest share with 0.0037, 0.0035 and 0.0013 DALYs pppy, respectively.

Disease burden estimates and comparison with standards
Our estimated gastrointestinal disease burden among urban farmers and sanitation workers who are exposed to wastewater are 5.6-and 2.8-fold higher than the diarrhoea burden estimates for the average Vietnam citizen (0.002 DALYs pppy) ( Institute for Health Metrics and Evaluation, 2015 ). The high number of cases and the resulting burden due to E. coli , rotavirus and Campylobacter spp. together cause most of the disease burden (88%), which is in line with prior QMRA investigations in Asia and Africa ( Pham-Duc, 2012; Katukiza et al., 2013;Machdar et al., 2013;Fuhrimann et al., 2016d ). To our knowledge, we present the first specific QMRA to estimate disease burden due to exposure to wastewater for highrisk groups in Southeast Asia. Compared to a prior study conducted in Kampala, Uganda, using the same methodology as presented here, the current estimates are considerably lower (e.g. the estimated burden for urban farmers in Hanoi was 0.011 DALYs pppy compared to 0.073 DALYs pppy in Kampala) ( Fuhrimann et al., 2016d ). Very high estimates were obtained by QMRA models with regard to people exposed to wastewater in a typical slum area in Kampala (15,015 people; burden 10,172 DALYs) ( Katukiza et al., 2013 ) and an urban wastewater systems in Accra (286,833 people; burden 31,979 DALYs) ( Labite et al., 2010 ). These differences might, at least partially, be explained by specific social-ecological contexts, pathogen spectrum, DALY estimates per pathogens (e.g. we excluded sequelae) and methodological differences (e.g. calculation of risk of illness).

Comparison of QMRA estimates with epidemiological survey data
Only few studies have compared QMRA estimates with diarrhoea episodes assessed in epidemiological surveys ( Bouwknegt et al., 2014;Haas et al., 2014 ). We aimed to fill this gap, and hence, conducted a cross-sectional survey, assessed water quality and obtained data on self-reported diarrhoeal episodes with a recall period of 2 weeks ( Fuhrimann et al., 2016b ). Extrapolating the self-reported 2-week diarrhoeal prevalence rate to 1 year (52 weeks, considering no seasonality), the annual incidence would be slightly higher compared to the combined incidence rates of all seven pathogens estimated in the QMRA (i.e. 2.6 versus 2.0 in urban farmers; 2.3 versus 0.09 in urban community members at risk of flooding events; 2.1 versus 1.0 in sanitation workers; and 1.0 versus 0.07 in peri-urban farmers). These obtained differences for urban farmers, communities and sanitation workers might be explained by additional risk factors for diarrhoeal disease, which drive the higher estimate (e.g. contaminated food crops, drinking water, human-to-human transmission, heavy metals or pesticides). Hence, future QMRA or chemical risk assessment should account for such factors, along with the potential effect of vulnerability and potentially acquired immunity to infectious agents within the exposed population groups in LMICs ( Minh et al., 2004;Holm et al., 2010;.
Further, when focusing on helminthiases, the QMRA estimated that 74 people would be infected with A. lumbricoides over the course of a year, with 71 cases among urban farmers. These low incidence rates for urban farmers (1.4% of total cases) are in line with our cross-sectional survey, which did not detect a single A. lumbricoides infection in famers using wastewater. As deworming is done on a regular basis and a significant protective effect was found in our cross-sectional survey, this low incidence might indeed remain undetected in epidemiological surveys ( Fuhrimann et al., 2016b ). In the light of our findings, for further cross-comparison between epidemiological surveys and QMRA, a standardised way to assess incidence and burden of diarrhoeal episodes and intestinal parasitic infections should be proposed and validated to make use of both tools. This statement can also be underlined by a similar comparison of QMRA estimates with findings from an epidemiological survey in Kampala ( Fuhrimann et al., 2016b( Fuhrimann et al., , 2016c.

Sensitivity analysis
A sensitivity analysis of the QMRA model employed here has been presented by Fuhrimann et al. (2016d) . There is a considerable effect of different volumes of water accidentally ingested. Such accidental ingestion may not be very accurate, as it is highly dependent on individual behaviours, which are influenced by age, sex, educational attainment and socioeconomic status ( WHO, 2006;Haas et al., 2014 ). Moreover, pathogen ratio for rotavirus, E. coli and Campylobacter have shown a considerable effect on the total number of gastroenteritis cases, as there is considerable difference in microbial contamination between seasons in urban wastewater systems this might have considerable effect on the number of cases ( Ensink, 2006;Katukiza et al., 2013;Fuhrimann et al., 2015Fuhrimann et al., , 2016a. It is important to note that E. coli is secreted by humans and animals continuously, whereas pathogens Table 2 Disease burden due to gastroenteritis expressed in disability-adjusted life years (DALYs) calculated by means of severity (mild, moderate, severe and fatal), probability and duration of the respective severity grade, stratified by pathogen.

Gastroenteritis
Severity weights Reference are secreted only by a proportion of infected people over a short period of a few days. This might affect the E. coli to pathogen ratio change over time ( Mara, 2004;Haas et al., 2014 ).

QMRA limitations
Our model framework has several limitations, which are offered for consideration. First, the estimates of the ingested volumes of water are based on literature values for certain human behaviours ( Labite et al., 2010;Schets et al., 2011;Katukiza et al., 2013 ). Second, the dose-response models applied in our model are based on feeding studies (e.g. norovirus) or rely on epidemiological evidence (e.g. Ascaris spp.) collected in high-income countries ( de Man et al., 2013 ;Barker et al., 2014 ). Third, in scenario 1 we do not differentiate between adults and children. Especially for rotavirus, burden is known to be substantially different between age groups . Fourth, the model excluded other exposure pathways such as dermal contact (e.g. hookworm and skin disease) and inhalation. It is reported that other water-borne pathogens such as hepatitis, cholera and adenoviruses occur in high prevalence and cause severe health implications ( Karagiannis-Voules et al., 2015 ;Fuhrimann et al., 2016b ). Fifth, partially acquired immunity due to exposure history or vaccination may result in lower estimates ( Sunger and Haas, 2015 ). Sixth, the duration of a fatal case was assumed to be equal to the life expectancy at birth, which is likely to result in an overestimation of the burden estimates ( Katukiza et al., 2013;Fuhrimann et al., 2016d ). Finally, long-term sequelae is often contributing to a significant proportion of DALYs when assessing burden to intestinal pathogens in high-income countries, hence, for next QMRA in LMICs sequelae should be included ( Havelaar et al., 20 0 0 ).

Conclusions
Employing a QMRA approach, we found high incidence of water-borne pathogens among urban farmers in Hanoi who use wastewater in agriculture and aquaculture. The validation of the QMRA estimates with findings from epidemiological surveys showed that QMRA estimated incidence of gastroenteritis was considerably lower, which can be explained by additional transmission routes (e.g. food crops) which can result in gastroenteritis cases. The high incidence of gastrointestinal infections results in considerable disease burden, which is 5.6-fold higher in urban farmers compared to an average Vietnamese and more than hundred times above the revised WHO tolerable level of 0.0 0 01 DALYs pppy. In turn, exposure to wastewater around Hanoi has considerable public health implications for these three population groups and call for actions, especially to bring down the burden due to E. coli , rotavirus and Campylobacter spp. infections.
Against this background, health-based targets should be set according to local idiosyncrasies after validation with epidemiological findings. For example, as a first step towards setting local targets could be the consideration of estimates provided by the Global Burden of Disease study 2010. These findings are especially interesting in the frame the Sustainable Development Goals (SDGs), as they address sustainable and safe wastewater reuse and recovery systems, from the point of generation to the point of disposal and (re)use for minimising adverse health impacts associated with water-borne disease, while maximising gains from safe wastewater use in agriculture and aquaculture in LMICs. Finally, the study provides a first example on how to link quantification of health risk using a QMRA to an SSP approach in an Asian context. Table 3 QMRA estimates for annual probability of illness, number of cases per year, total DALYs per year, DALYs per person per year across all four exposure scenarios ( S flooding , S working , S farming U and S farming PU ) and seven pathogens along the major wastewater system in Hanoi, Vietnam.