INTRODUCTION
Escherichia coli O157 is a gastrointestinal zoonotic pathogen of public health importance in a number of countries including North America and the UK [Reference Chase-Topping1]. The disease is relatively rare with the highest annual incidence worldwide being reported in Scotland (e.g. 4·7 cases/100 000 in 2008 [Reference Locking2]). The symptoms of the disease include bloody diarrhoea in ~90% of cases, of which 10–15% progress to haemolytic uraemic syndrome (HUS) [Reference Tarr, Gordon and Chandler3], long-term risk of hypertension [Reference Clark4, Reference Moist5] and occasionally death. The disease is most common in young children aged <5 years and incidence is highest in rural rather than urban areas [Reference Strachan6].
The main reservoir for this pathogen is cattle and sheep [Reference Hancock7, Reference Solecki8] but it can also be found in a number of other animal species (e.g. deer). Humans can become infected by three primary and one secondary transmission pathway: foodborne infection via contaminated food including beefburgers, dairy products and produce (e.g. vegetables) [Reference Rangel9]; waterborne primarily through untreated, contaminated water supplies [Reference Licence10–Reference Schets12]; ingestion of faecal material by direct contact with animals, their faeces and contaminated soil [Reference Locking13] and the secondary transmission pathway is person-to-person contact which is particularly prevalent in nursery and care-home settings [Reference Reiss14].
A number of major outbreaks have occurred with this pathogen, e.g. the Central Scotland outbreak in 1996 where 512 people were infected and 17 died which involved contaminated meat products [Reference Cowden15]; the Walkerton outbreak in Canada in 2000 where 2300 people were infected from wells contaminated with ruminant faeces [Reference Hrudey11] and in the USA in 2006 where there was an outbreak associated with spinach consumption that involved 205 cases [Reference Grant16]. The trend in Scotland is increasingly towards outbreaks associated with sources other than foods such as contact with animals and drinking contaminated water [Reference Strachan6]. However, the majority of human E. coli O157 cases are sporadic [Reference Snedeker17] and case-control studies in Scotland [Reference Locking13], USA [Reference Kassenborg18, Reference Slutsker19] and Wales [Reference Parry20] confirm the importance of the infection pathways outlined above.
The spatial epidemiology of this pathogen has been investigated in Scotland [Reference Strachan6, Reference Innocent21], Sweden [Reference Kistemann22], Canada [Reference Michel23–Reference Pearl25] and The Netherlands [Reference Friesema26]. These studies demonstrate the higher incidence in rural areas compared to urban populations as well as the importance of risk factors such as contact with livestock and local socioeconomic status. One approach to determining the relative importance of each infection pathway would involve development of a regression model that combines the risk factors for each pathway and links this to the reported number of human cases. To the best of our knowledge this has not been previously attempted.
Stochastic quantitative microbiological risk assessments (QMRAs) have been developed to estimate the risk of illness associated with consuming beefburgers [Reference Cassin27–Reference Signorini and Tarabla30] and from visiting or camping on pasture that has recently been grazed by cattle or sheep [Reference Strachan, Dunn and Ogden31]. A deterministic model has also been developed which estimates the number of cases associated with drinking from private water supplies (PWS) and visiting bathing waters [Reference Vinten32]. One of the major advantages of using QMRAs is that they can evaluate the relative efficacy of mitigation strategies in silico prior to application in the real world. Combining epidemiological data with QMRAs (e.g. to develop dose–response relationships from outbreak data) has been reported previously [Reference Miliotis33]. However, combining risk assessments for each transmission pathway to predict the total number of human cases and the subsequent comparison with the number of reported cases in order to obtain an estimate of the importance of each infection pathway has not been attempted. This is because a human exposure survey of the risk factors (e.g. number of glasses of water drunk per day by the population on a PWS) is not available. Further, foodborne, waterborne and environmental QMRAs have not been simultaneously parameterized with data from a particular geographical area.
The aims of this paper are to assess the importance of the different transmission pathways of human E. coli O157 infection in the high-incidence Grampian region of Scotland by (i) developing a simple regression model parameterized with spatial data and (ii) developing and combining QMRAs to enable comparison with reported human case data. We then compare the results of these modelling techniques to obtain a greater understanding of E. coli O157 infection in Grampian. Finally, we evaluate a number of mitigation strategies for each transmission pathway to determine the efficacy of reducing E. coli O157 infections.
MATERIALS AND METHODS
Data
Anonymized human case data of E. coli O157 infection (n=770) were collected from National Health Service (NHS) Grampian for the years 1997–2008 inclusive. Of these, 694 had information containing the postcode district of residence, the date of reporting, and the age and gender of the case. The 27 cases identified from one primary school outbreak, is an outlier and these data were removed from the analysis, leaving 667.
Data on the human population at postcode district level were collected from the 2001 Scottish census [from SCROL (http://www.scrol.gov.uk/)]. Data on cattle and sheep numbers at the resolution of 2 km×2 km tetrads were obtained from the 2000 Scottish agricultural census (from EDiNA® (http://edina.ac.uk/agcensus) revised for 2004). Data on properties on PWS (n=13578 in 2006) were obtained from Aberdeenshire and Moray local authorities.
Empirical epidemiology
The incidence of E. coli O157 was calculated, as was the incidence ratio of rural to urban cases. Urban cases were defined as postal districts with a population density >200 persons/km2 [Reference Strachan34]. Tests of statistical significance and 95% confidence intervals were determined using a bootstrap method that involved resampling the original data with replacements using statistical software Poptools (www.cse.csiro.au/poptools/). The bootstrap sample size was 10 000. This was also used to determine whether rural/urban ratios were statistically significantly different from 1. The percentage of cases that were general outbreaks was determined by consultation of reports from Health Protection Scotland (www.hps.scot.nhs.uk/ewr/index.aspx).
Data for risk assessment
A telephone-based exposure assessment questionnaire of a representative (by age, population density, socioeconomic status) subsample of the Grampian population (1068 questionnaires from a population 519 979) was conducted by Aberdeen University and from this the following information was obtained: the number of ‘well done’ and ‘rare’ burgers consumed per person per year; the number of visits to camp-sites on pasture grazed by cattle and sheep per person per year; the number of glasses of water drunk per person per year and the proportion of treated PWS.
Spatial risk-factor regression model
A multivariate linear regression model was developed to link together proxy risk factors for each of the three primary transmission pathways [foodborne, environmental (e.g. direct contact with ruminants and their faeces) and waterborne (i.e. PWS)]. Person-to-person transmission was not included as it was assumed in the first instance to be similar for each pathway.
The proxy risk factors (defined below) were calculated for 2 km×2 km tetrads. The proxy risk factor for food (F) is given by:
where P is the number of people in the human population under consideration. This assumes that the same food is consumed and the same variation in hygiene/cooking processes are performed across Grampian. Hence the risk of contracting E. coli O157 infection from food is the same for each person in the population.
The environmental (E) proxy risk factor has the following form:
where N E.coliO157 is the total number of E. coli O157 excreted by cattle and sheep (calculated by estimating the average number of E. coli O157 excreted per gram of faeces multiplied by the average weight of faeces excreted per animal per day, multiplied by the number of animals in the postal sector – this was done separately for cattle and sheep and the results were summed). Hence, the greater the number of people and the greater the number of E. coli O157 excreted, the greater is the risk of human infection.
The waterborne (W) proxy risk is given by:
where N PWS is the number of properties on a PWS.
The regression model correlates the proxy risk factors with the actual number of reported cases (N cases) and has the following form:
where a, b, and c are the regression coefficients.
The linear regression was performed in SPSS 17 (www.spss.com) at the postal district level (n=38) and to achieve this the 2 km×2 km proxy risk-factor tetrads were integrated up to this spatial resolution using ArcView 3.3 (ESRI, USA). Normality of data was tested using XLSTAT (www.xlstat.com).
The number of predicted cases attributed to each transmission pathway for each postal district was determined by calculating the appropriate regression coefficient multiplied by the risk factor proxy (e.g. aF for food). This was then summed for all the postal districts in Grampian. The percentage attribution for each transmission pathway was then determined.
QMRAs
The QMRAs were written using @Risk (version 4.0, Palisade Corporation, USA). A manual giving full details of each of the models is available upon request to the corresponding author.
Food
The Cassin model for beefburgers was used [Reference Cassin27]. Briefly, this process-based model has initial inputs which include the prevalence and concentration of E. coli O157 excreted by cattle as they enter the abattoir (see Table 1). It determines the level of contamination on the meat and how this is affected by the grinding process. It simulates the survival of the pathogen in the burger, and models the die-off of the organisms during the cooking process. It then uses a beta-binomial dose–response model to determine the probability of illness. This QMRA was developed for the situation in Canada and here it is re-parameterized using data from Scotland (Table 1).
Table 1. Parameterisation of quantitative microbiological risk assessments (QMRAs) using Scottish datasets
CI, Confidence interval; PWS, private water supply.
Environment
This process-based QMRA model was used to estimate the environmental risk of infection for day visitors and campers using pasture grazed by animals. Full details of the QMRA have already been reported [Reference Strachan, Dunn and Ogden31] and here the data used in its parameterization were updated (Table 1). Briefly, this model uses data for human ingestion of soil that is contaminated with E. coli O157 shed by cattle or sheep. The model is run for the scenarios of either a day visit or an overnight camp to pasture. The same dose–response model used by Cassin et al. [Reference Cassin27] was employed to calculate the resulting probability of infection. The model was run four times (overnight camping and day visits on fields each recently grazed by cattle or sheep) to cover each human exposure scenario.
Water
A process-based QMRA was developed to determine the risk of infection from drinking a glass of water from a PWS. The flow diagram in Figure 1 indicates the main steps in the process to determine the probability of illness. Briefly, a property on a PWS is randomly selected. The supply is identified as being contaminated or not using a beta distribution based on a prevalence survey of E. coli O157 in PWS in the Grampian region (Table 1). However, this does not give the concentration of E. coli O157 present and the following process is used for estimation purposes. If the PWS is negative the probability of illness from this source is zero. If positive, the number of cattle/sheep in the proximity (2 km×2 km square) is determined from the agricultural census and from this the ratio of the average concentrations of E. coli O157 to E. coli excreted (Table 1) into the environment is calculated (see Supplementary material, available online). This ratio (R) is then multiplied by the concentration of E. coli (randomly sampled from the survey of PWS which are positive for E. coli) to determine the concentration of E. coli O157 in the PWS (see Supplementary material). The dose ingested from consuming a glass of water is calculated. The dose response is then applied to determine the probability of illness.
Fig. 1. Flow chart showing the private water supply (PWS) quantitative microbiological risk assessment (QMRA) which determines the probability of illness (P ill) from drinking a single glass of water.
Incorporating exposure data into the QMRA models
The annual number of exposures, estimated from the survey, for the whole of the Grampian population (n=519 979) for the different pathways were: 3·45 million beefburgers consumed, 2·48 million day visits [this excludes individuals who have contact daily (e.g. farmers who were considered to be immune to E. coli O157 in this study)], 0·1 million camps on pasture and 34·7 million glasses of water drunk from a PWS. Computational time and computer memory limitations meant that only a subsample of these exposures could be simulated (~2 million). Resulting values were then used to determine the number of illnesses in Grampian with associated confidence intervals. These data were then compared with the average number of cases reported in Grampian per year and also with the attribution of cases to one of the four pathways determined by the spatial risk-factor regression model.
Application of mitigation strategies in QMRAs
Risk mitigation strategies that were identified in the literature and readily operational in the QMRAs were simulated to determine their efficacy in reducing the number of cases of E. coli O157 infection.
The following mitigation strategies were evaluated (Table 2):
Table 2. Efficacy of risk mitigation strategies
PWS, Private water supply; QMRA, quantitative microbiological risk assessment.
* The beefburger QMRA assumes a uniform distribution of between 0 and 2 log reduction using existing processes within the abattoir for the base scenario. This was re-run using 0 log as a baseline.
Food
(1) Removing high shedding animals prior to slaughter, (2) reducing concentration (numbers of E. coli O157) shed, (3) a consumer education policy that ensures that burgers are cooked properly, (4) decontamination of carcasses in the abattoir and (5) reducing the individual animal prevalence in cattle.
Environment
(1) Washing hands after contact with animals and/or their faeces, (2) banning camping in the countryside in fields recently grazed by farm animals, (3) reducing (a) individual and (b) group prevalence in cattle and sheep, (4) reducing concentration shed, (5) keeping farm animals off pasture for a fixed period directly prior to day visit or camping visit and (6) physical removal of faeces from pasture prior to visit.
Water
(1) Increasing the proportion of PWS treated, (2) treatment of only those PWS which have a high load of commensal E. coli, (3) banning the use of PWS in areas of high livestock density and (4) reducing the individual animal prevalence in cattle.
RESULTS
Empirical epidemiology
The average incidence during this time period was 11·1 cases/100 000 and was highest in the youngest age groups (Fig. 2). The ratio of rural to urban incidence was 1·96 for the population as a whole and was largest for the age groups between 1 and 4 years.
Fig. 2. (a) The average incidence of E. coli O157 infections in Grampian stratified by age for the 12-year period 1997–2008 and (b) the rural/urban incidence ratio with 95% confidence intervals [asterisks indicate statistically significant (P<0·05) ratios >1·0 determined by randomization test].
Spatial risk factor regression model
Figure 3 a presents a map of the number of reported E. coli O157 cases in each postcode district. Figure 3(c–e) show the proxy risks for food, environment and water. All three factors were included in the multivariate regression because they were all found to have correlation coefficients >0·4 (P<0·05) compared with the number of reported cases in univariate analysis. The multivariate regression model gave a significant fit (R 2=0·87, P<0·0001). The predicted number of cases by the regression model allocated to each postal district is illustrated in Figure 3 b and the regression coefficients as well as the percentage of cases attributed to each pathway are given in Table 3. Environment was found to be the most important source while the model reported that PWS contributed the fewest cases. Cross-correlation was found between two combinations of the proxy risk variables used in the regression: food-environment (R 2=0·56), food-PWS (R 2=0·04), environment-PWS (R 2=0·67).
Fig. 3. Choropleth maps of the postal districts in Grampian showing (a) reported cases, (b) predicted number of cases using the regression model and proxy risks for (c) food, (d) environment and (e) water. The proxy risks are normalized to values between 0 and 1.
Table 3. The mean number of cases attributed annually to each transmission pathway by the spatial risk-factor regression model and quantitative microbiological risk assessment (QMRA)Footnote *
CI, Confidence interval; P ill, probability of illness.
* The average of the annually reported number of cases is 55·6.
QMRAs
The probability of illness (P ill) for each iteration of the three QMRAs was determined and is displayed in Figure 4. The highest average probability of illness was from food (P ill=2·65×10−4, i.e. about 1 in 4000), which was greater than from environment (P ill=2·15×10−4, i.e. 1 in 5000) and PWS (P ill=1·31×10−5, i.e. 1 in 100 000). When this was combined with exposure information the number of cases could be determined (Fig. 4). Incorporating the exposure information permitted the number of cases associated with each transmission pathway to be determined. The food model predicted most cases (915) compared to 554 from the environment and 162 from PWS. The total number of predicted cases (1631) from the three QMRAs was ~30-fold higher than the average number of cases reported in Grampian per year. Both QMRA and spatial risk-factor regression approaches indicate that PWS are the least important pathway in terms of human cases but they order differently the importance of food and the environment (with overlapping confidence intervals).
Fig. 4. The distribution of the probabilities of illness for (a) food (consuming a burger), (c) environment (four scenarios: camping/visiting pasture with sheep/cattle) and (e) drinking from a private water supply (PWS) in a single day (note: 99·95% of individuals have zero probability of infection – not illustrated). Inclusion of exposure information enables the number of cases to be determined: (b) food, (d) environment and (f) water.
Risk mitigation
Table 2 presents the efficacy of the risk mitigation strategies for the three QMRAs. With regard to food, to achieve a reduction of ~50% in cases for that transmission pathway would require the removal of either those animals shedding high numbers of target bacteria (between 3 and 4 log10, which is equivalent to 1·35% of cattle) prior to slaughter, or reducing the concentration shed by 1 log, or reducing carcass contamination by 1 log10, or achieving a 75% reduction in individual animal prevalence. Ensuring all burgers are cooked properly would result in a 29% reduction in cases.
With regard to the environment, to reduce cases by 50% would require: reducing pathogen numbers ingested by 0·5 log10 through improved hand washing, or achieving a 50% reduction in individual animal prevalence, or reducing the group prevalence by 50%, or reducing concentration shed by 0·5 log10, or keeping farm animals off pasture prior to human visit for at least 7 days, or physical removal of faeces from pasture. Banning camping on pasture would only achieve a reduction of 23%.
Mitigations for water would achieve ~50% reduction in cases by increasing the proportion of PWS treated from 48% (current level) to ~80%, or treating all highly contaminated PWS (>100 c.f.u. commensal E. coli/100 ml). Banning the use of PWS in areas of high density of animal numbers (>250 animals/km2) would only reduce the number of cases by ~22%.
DISCUSSION
Within Grampian, E. coli O157 infection is dominated by cases from rural areas, particularly in children. This high incidence is probably due to two main factors: the immature immune system of the child and/or the child's increased exposure from one or more of the infection pathways. It is interesting to note that children aged <1 year old, although having a relatively high incidence do not over-present with the disease in rural areas compared with their urban counterparts. This could be because they are less likely to come in to contact with the pathogen through direct contact with farm animals. However, it would be expected that some of these children would be drinking from untreated PWS. Foodborne transmission is possible in this group (post-weaning) but this is most likely to be from meat products or contaminated foods other than milk which cannot be sold unpasteurized in Scotland. Children aged >1 year present on average much more frequently in rural areas. There is evidence from outbreaks (e.g. camping in Scotland [Reference Howie35]) which demonstrates that direct contact with animals and their faeces are significant risk factors in this age group. This is further exacerbated with person-to-person transmission being particularly important in young children [Reference Parry and Salmon36] (with poor hygiene awareness) within the household or to the wider community (e.g. nurseries [Reference Snedeker17]).
The QMRA approach estimates an individual's probability of risk associated with particular exposures while the spatial risk factor regression method operates at a group level (e.g. the postal district). The QMRA is based on probability distributions whereas the regression model uses risk factors and is essentially deterministic. Further, the regression model requires the number of cases to be known to generate a model fit while the QMRA actually predicts this. Hence, the two approaches are quite different in trying to answer the same question of determining the relative importance of each of the primary infection pathways.
There is good agreement with the spatial regression and QMRA approaches in attributing the fewest cases to waterborne infection; however, it is unclear whether food or environment is most important. The risk associated with eating a burger is similar to visiting a field recently grazed by animals. This is at odds with our previous study [Reference Strachan6] which found the pasture visit was a much greater risk (100-fold). However, here the burger model predicts that 29% of cases are associated with undercooking (eating rare) whereas previously we had assumed that they were well cooked. Empirical analysis of the epidemiological data (Fig. 2) shows that the incidence is approximately twice as high in rural areas compared to urban ones, indicating that the environment plays an important role. There is a degree of cross-correlation between the risk proxies and this could be due to the geography of Grampian (e.g. PWS tend to be in rural farming areas where cattle and sheep are reared). The regression model could be applied at higher spatial resolution and seasonality could also be included. However, breaking the dataset down into smaller groups leads to deviation from the normal distribution. This would result in nonlinear regression techniques being required from which the number of cases associated with each pathway cannot be readily estimated. The regression model generates a good correlation between the predicted and actual number of cases but a disadvantage is that because of its simplicity it is not readily operational to evaluate risk mitigation strategies. This can of course be achieved by QMRA [Reference Cassin27]. An error would occur in cases where the illness is contracted outside their home postal district [e.g. the person may have visited the countryside, acquired the disease from coming into contact with animals there, then returned to an urban home and subsequently fallen ill (there is a period of 3–4 days before symptom onset [Reference Karch, Tarr and Bielaszewska37]). This would result in an underestimate of the importance of the environment while food would be overestimated. It is not possible to resolve this false attribution without further epidemiological data.
All of the QMRAs over-predicted the number of human cases. This suggests one or more common factors associated with the models may be error prone, e.g. the assumption that all E. coli O157 are equally infective (recently demonstrated not to be the case [Reference Laing38]) and that all organisms are released effectively from faecal material (this is reported not to be the case [Reference Hodgson39]) [Reference Snedeker17]. In addition, a proportion of the human population is likely to be immune [Reference Belongia40] (e.g. from repeated exposure to a contaminated PWS or working on a farm that has cattle or sheep shedding this pathogen) and asymptomatic cases may be common (e.g. up to 40% of secondary [Reference Snedeker17] and between 4–15% of sporadic [Reference Parry and Salmon36] cases). Further, the dose–response model chosen has an influence; the Cassin dose–response model was used here and another dose–response model (determined from E. coli O157 outbreaks) was also implemented [Reference Teunis, Ogden and Strachan41]. This latter model predicted even more cases (60-fold higher than reported; data not presented). The Cassin model is a surrogate model parameterized using data from Shigella feeding studies. The Teunis model is based on E. coli O157 outbreak data which are relatively few and could involve strains that have higher than average infectivity and by definition expose a susceptible population.
The QMRAs described do not cover all the potential pathways of infection in humans. For example, foods other than burgers were not considered in this study, but are known to be associated with E. coli O157 infection in the Grampian region of Scotland (e.g. lettuce (www.eurosurveillance.org/ViewArticle.aspx?ArticleId=3273) and cheese [42]). Further, application of manure to pasture would pose a risk to visitors. However, if the manure had been sufficiently stored/treated [Reference Jones43] prior to application then it is likely that this risk would be much smaller than that from freshly voided faeces. Hence it is likely that in this light the QMRAs overestimate the number of cases. For the waterborne pathway, risks associated with swimming and water sports are not included. The average annual exposure to this pathway is more than one order of magnitude lower than from drinking water from a PWS [Reference MacRitchie, Hunter and Strachan44]. In summary, combining the three QMRAs in this paper is a simplification of the real world. To obtain a more realistic representation, QMRAs should be developed for other vehicles and included in the analysis. This should be a focus of further research. Thus, the approach taken here will actually underestimate the number of human cases as all possible vehicles are not included. Regarding person-to-person transmission it is assumed in this paper to be a constant factor by pathway and age. It is known that secondary transmission is more important in children [Reference Snedeker17, Reference Parry and Salmon36] and that this is also likely to be the case in the elderly. It would be useful to incorporate data into the risk models on not just age but living conditions (e.g. attending a nursery or living in a care home) and also behaviour (hand washing, etc.). This would be possible to implement in a QMRA but would need a dynamic component [Reference Soller and Eisenberg45]. Consumption of burgers has been found not to vary between urban and rural areas [Reference MacRitchie, Hunter and Strachan44] and hence this does not explain the excess number of cases in rural areas (Fig. 2). Further, burger consumption is greatest in children aged >5 years old and so cannot explain the high incidence in children aged <5 years. Finally the ‘riskier’ practice of eating rare burgers is more associated with the adult section of the population.
The QMRAs facilitated the evaluation of a range of putative mitigation strategies in silico. However, these strategies need to be considered not only in terms of their efficacy to reduce E. coli O157 but also in terms of their practical implementation, acceptability and cost.
In terms of foodborne infection by burgers, a mitigation step is the removal of high-shedding animals prior to slaughter (identified using a cowside test) or alternately reducing shedding levels in ruminants. A cowside test, based on immunoassays, currently has a sensitivity limit of about 104 c.f.u./ml [Reference Naylor46] which is insufficient to identify animals to achieve the potential 50% reduction in cases. A 1 log10 increase in sensitivity is preferable. Therapeutic agents [Reference Naylor46] can reduce shedding by 2 log10 units when administered orally and by 4 log10 units with rectal application. However, timing and practicality of application would have to be considered. A review of the effects of diet and dietary supplements has been carried out [Reference Callaway47] investigating the roles of fasting, feed additives (probiotics and antimicrobials) as well as dietary changes but concluded that much of the research was unclear with conflicting results. However, abruptly changing diet from a high proportion of grain to a high proportion of hay did reduce E. coli O157 shedding, but the practicality of carrying this out prior to slaughter might be difficult. Further, the dynamics of shedding E. coli O157 is unclear; animals tested (at the farm or lairage) as being above a safe threshold might be below that value at slaughter, and vice versa.
In terms of treating the carcass at the abattoir, a number of treatments have been evaluated, e.g. hand trimming followed by spray washing achieved a reduction of between 1·4–2·5 log10 of E. coli counts [Reference Gorman48]. In small slaughter plants [Reference Algino, Ingham and Zhu49] a range of treatments were investigated [e.g. dry ageing, sprays of hot water (low and high pressure) or acetic acid] and found to have similar effects resulting in the reduction of Enterobacteriaceae by 0·6–2·0 log10. Steam treatment [Reference Trivedi, Reynolds and Chen50] applied to beef carcasses achieved 0·8 log10 reduction. Implementation in the abattoir would inevitably be costly in terms of equipment, labour and possible reduced processing speed. Finally, an educational campaign explaining the importance of correct cooking procedure for minced beef might achieve a reduction in cases and would also mitigate the risk from other gastrointestinal pathogens (e.g. Salmonella) that can be found in these products.
The environmental risk mitigations demonstrate that washing hands can potentially lead to a significant reduction in infection rates. Indeed, if there is a >3 log10 reduction in E. coli concentration by hand washing using antimicrobial soap [Reference Fischler51] it is predicted that this could lead to a reduction of 99·9% of environmental infections. However, it is unknown how many existing cases had previously washed their hands before falling ill. It has been reported [Reference Howie35] that washing hands prior to eating food in a camp-site contaminated with E. coli O157 by sheep faeces reduced the infection rate significantly compared with those who had not. Implementing hand washing in rural settings with limited clean water may be problematical. Alternatives, e.g. ethanol-based antibacterial gels may be more appropriate and have been demonstrated to be equally effective as soap in an agricultural setting where >1 log10 reduction was achieved in more than 50% of those participating [Reference Davis52]. Further, it is worth noting that those aged <5 years are at greatest risk and stopping hand–oral contact would be very difficult to achieve in this age group.
Reducing E. coli O157 concentration shed by animals in the environment is a greater challenge than prior to slaughter as a long-term solution is preferable as opposed to a short-term measure, e.g. abrupt feed change. Banning of camping in areas recently grazed by farm animals is a potential strategy but is difficult to enforce and compulsory measures might be considered counter-productive. However, advising people that camping in this situation is a health risk does offer the opportunity of allowing individuals to make an informed decision.
Reducing infection rates by 50% from the PWS pathway would involve increasing the percentage of PWS treated from 48% to ~80%. Grants for water treatment are currently available but many owners/users do not perceive PWS as a significant health concern [53]. Specifically, treating those PWS with high contamination levels could also potentially lead to a 50% reduction of cases but requires testing prior to targeting installations; further, one-off tests are inadequate to quantify microbiological quality. In the QMRA a 2 log10 reduction is simulated for treated PWS; however, some treatment methods are more effective than others. For example, UV treatment can reduce E. coli or E. coli O157 by 1·5–6 log10 [Reference Zimmer and Slawson54, Reference Sommer55], chlorination by 3–5 log10 [Reference Rice, Clark and Johnson56] and sand filtration by 0·3–4 log10. Banning the use of PWS in areas with high animal densities appears to have a relatively small effect on the number of cases and might be totally impractical.
A reduction in prevalence of between 50% and 75% in the ruminant population would potentially reduce infection by ~50% in all of the QMRAs studied. There are a number of potential strategies to consider. A two-dose vaccine regimen reduced positive faecal samples from feedlot cattle and hides by ~50% [Reference Smith57]. Farming practices also have the potential to reduce prevalence, e.g. not mixing young and old stock and not mixing cattle and sheep both achieved a 50% reduction in prevalence [Reference Ellis-Iversen, Smith and Pritchard58] while reducing cattle stocking density by 50% resulted in a 25% reduction in prevalence [Reference Ellis-Iversen59].
CONCLUSIONS
The results presented here demonstrate that the sources of E. coli O157 infection in humans are multifactorial but that environmental and foodborne routes are the most important. It should be borne in mind that the models do not account for all the possible transmission pathways and further research is required to fill this gap. The epidemiology reinforces the point that young children are at greatest risk. To reduce infection rates it is important that people wash their hands properly after contact with farm animals or their faeces and that educational campaigns highlight the importance of correct cooking practices and the risks associated with visiting the countryside (e.g. contact with farm animals). Although waterborne infection appears to be the least important pathway, encouragement to improve take-up of grants to treat PWS (particularly those with high indicator counts) will still have an effect on reducing human cases of disease. Despite their limitations, the combination of QMRAs, spatial regression models and epidemiology all contribute to an improved understanding of how humans become infected by this pathogen.
NOTE
Supplementary material accompanies this paper on the Journal's website (http://journals.cambridge.org/hyg).
ACKNOWLEDGEMENTS
The authors thank Dr Tom Reid (Foresterhill Hospital, Aberdeen) for supplying the clinical E. coli O157 case data, and Anne Thomson (University of Aberdeen) for carrying out E. coli microbiological counts from cattle and sheep faeces. The work was funded through the Economic and Social Science Research Council programme on Rural Economy and Land Use (RELU) and comments on the manuscript from the E. coli O157 RELU team are acknowledged.
DECLARATION OF INTEREST
None.
