Use of microbial risk assessment to inform the national estimate of acute gastrointestinal illness attributable to microbes in drinking water

Jeffrey A. Soller Soller Environmental 3022 King St, Berkeley CA 94703, USA E-mail: jsoller@sollerenvironmental.com Microbial risk assessment (MRA) evaluates the likelihood of adverse human health effects that occur following exposure to pathogenic microorganisms. This paper focuses on the potential use of MRA to provide insight to the national estimate of acute gastrointestinal illness (AGI) in the United States among persons served by public water systems. This article defines MRA, describes how MRA is implemented, provides an overview of the field of MRA and discusses how MRA may be useful for characterizing the national estimate. Communities served by drinking water systems with relatively contaminated source waters, sub-standard treatment facilities, and/or contamination problems in their distribution systems are subject to higher risks than communities where such issues are less of a concern. Further, the risk of illness attributable to pathogens in drinking water in each community can be thought of as the sum of the risk from the treated drinking water and the risk from the distribution system. Pathogen-specific MRAs could be developed to characterize the risk associated with each of these components; however, these assessments are likely to under-estimate the total risk from all pathogens attributable to drinking water. Potential methods for developing such MRAs are discussed along with their associated limitations.

quantitative characterization and estimation of potential adverse health effects associated with exposure of individuals or populations to hazardous materials and situations (NRC 1983;Hoppin 1993).The risk management component weighs policy alternatives in light of the results of risk assessment and, if required, selects and implements appropriate control options, including regulatory measures.The risk communication component is the interactive exchange of information and opinions concerning risk and risk management among risk assessors, risk managers, consumers and other interested parties (WHO 1999).This paper focuses on the risk assessment component of the risk analysis process.
MRA (also known as pathogen risk assessment) is a process that evaluates the likelihood of adverse human health effects that can occur following exposure to pathogenic microorganisms or to a medium in which pathogens occur (ILSI 1996).To the extent possible, the MRA process includes evaluation and consideration of quantitative information, however, qualitative information is also employed as appropriate (WHO 1999).
Quantitative risk assessment has been used since the 1970s to assess human health effects associated with exposure to chemicals (Hammond & Coppick 1990).
The principles, processes and methods for carrying out risk assessments for chemical agents were formalized in 1983 by the National Research Council (NRC) resulting in a four step process or framework (NRC 1983).The steps outlined by the NRC include hazard identification, doseresponse assessment, exposure assessment, and risk characterization.Many of the early MRAs employed the NRC conceptual framework to provide a structure from which the assessments could be conducted (Haas 1983a;Regli et al. 1991;Rose et al. 1991;ILSI 1996).
As the field of microbial risk assessment developed, it became clear that there were some complexities associated with modeling the infectious diseases that are unique to pathogens, such as person-to-person transmission of infection and immunity.Therefore, the conceptual framework for chemicals may not always be appropriate for the assessment of risk of human infection following exposure to pathogens (ILSI 1996).To address this concern, the EPA Office of Water sponsored a series of workshops to develop a conceptual framework to assess the risks of human infection associated with pathogenic microorganisms.
Those workshops resulted in a published framework (ILSI 1996) that was then tested through the conduct of two case studies (Soller et al. 1999;Teunis & Havelaar 1999) and subsequently revised (ILSI 2000).The EPA/ILSI framework for assessing the risk of human infection following exposure to water-and food-borne pathogens is comprised of three principal components: problem formulation, analysis, and risk characterization.At this time, both the NRC and EPA/ILSI frameworks are currently employed for the conduct of MRAs.Following is a brief summary of those frameworks.

NRC risk paradigm for microbial risk assessment
As applied to microbial risk assessment, the four steps comprising the NRC risk paradigm are summarized below:

Hazard identification
For microbial agents, the purpose of hazard identification is to identify the microorganisms or the microbial toxins of concern.Hazards can be identified from relevant data sources such as scientific literature, databases, and solicitation of expert opinion.Relevant information for the hazard identification often includes review of clinical studies, epidemiological studies and surveillance, laboratory animal studies, investigations of the characteristics of microorganisms, interaction between microorganisms and their environment, and studies of analogous microorganisms and situations (WHO 1999).

Dose -response assessment
The dose-response assessment provides a quantitative or qualitative description of the likelihood, severity and/or duration of adverse effects that may result from exposure to a microorganism or its toxin.Dose -response relationships can be developed for different end points, such as infection or illness, depending on the microorganism of interest.In the absence of appropriate dose-response data, risk assessment tools such as expert elicitations could be used to consider factors such as infectivity that may be necessary to characterize the host's response to a dose of pathogens (WHO 1999).
There are several important factors related to both the microorganism and the human host in the dose-response assessment.Relative to the microorganism the following may be important: the virulence and infectivity of microorganisms can change depending on their interaction with the host and the environment; genetic material can be transferred between microorganisms, leading to the transfer of characteristics such as antibiotic resistance and virulence factors; and/or low doses of some microorganisms can in some cases cause a severe effect (WHO 1999).
Relative to the human host the following may be important: genetic factors; increased susceptibility due to breakdowns of physiological barriers; individual host susceptibility characteristics such as age, pregnancy, nutrition, overall health, medication status, concurrent infections, immune status and previous exposure history; and population characteristics such as population immunity, and access to and use of medical care (WHO 1999).

Exposure assessment
An exposure assessment describes the magnitude and/or probability of actual or anticipated human exposure to pathogenic microorganisms or microbiological toxins.For microbiological agents, exposure assessments might be based on the potential contamination in water by a particular agent or its toxins, and on other exposure pattern information (for example, the frequency and/or duration of exposure).
Factors that must be considered for exposure assessment include the frequency of human exposure to the pathogenic agents and the associated concentrations of those pathogens over time.Other factors that could be considered in the assessment could include the potential impact of environmental conditions and/or water treatment reliability (WHO 1999) as well as factors influencing the patterns of exposure (such as socio-economic status, ethnicity, seasonality, population demographics, regional differences, and/or consumer preferences and behavior).

Risk characterization
Risk characterization represents the integration of the hazard identification, dose-response assessment, and exposure assessment components to obtain a risk estimate.The risk characterization process results in a qualitative or quantitative estimate of the likelihood and severity of the adverse effects which could occur in a given population, including a description of the uncertainties associated with these estimates.
Risk characterization depends on available data and expert interpretation of those data.The weight of evidence integrating quantitative and qualitative data may permit only a qualitative estimate of risk.The degree of confidence in the final estimation of risk will depend on the variability, uncertainty, and assumptions identified in all previous steps (WHO 1999).Differentiation of uncertainty and variability may be important for subsequent risk management considerations.However, experience indicates that, for MRAs, it is possible that variability and uncertainty will be confounded to such an extent that it is difficult or impossible to consider them independently.

EPA/ILSI paradigm for microbial risk assessment
The EPA/ILSI MRA (ILSI 2000) is conceptually similar to the NRC paradigm for human health risk assessments (NRC 1983) and the ecological risk assessment framework (US EPA 1992).The framework emphasizes the iterative nature of the risk assessment process (Figure 1), and allows wide latitude for planning and conducting risk assessments in diverse situations (Soller et al. 1999).This framework consists of three principal components: problem formulation, analysis, and risk characterization.The analysis phase is further subdivided into the characterization of exposure and human health effects.
The problem formulation stage involves all stakeholders and is used to identify: (1) the purpose of the risk assessment, (2) the critical issues to be addressed, and (3) how the results might be used to protect public health.Once identified, initial descriptions of the exposure and potential health effects are described and then a conceptual model is developed.This conceptual model is used as a starting point for the analysis phase of the risk assessment and later as an interactive tool along with components developed in the analysis phase to initiate the risk characterization.
In the analysis stage information about both the exposure and the health effects is compiled and summarized.This compilation of quantitative and qualitative data, expert opinion, and other information results in exposure and host/pathogen profiles that explicitly identify the data to be integrated into the risk characterization and the associated assumptions and uncertainties.These two elements, while separate, must also be interactive to ensure that the results are compatible.
The final stage, risk characterization, results in a statement of the likelihood, types, and/or magnitude of effects likely to be observed in the exposed population under the expected exposure scenario, including all of the inherent assumptions and uncertainties.Often, the risk characterization phase includes data integration through parameterization of a mathematical model, numerical simulation and interpretation.

MRA METHODS
Quantitative methods to characterize human health risks associated with exposure to pathogenic microorganisms were first published in the 1980 s (Haas 1983a, b;Cooper et al. 1986).
Over the last 20 years, the field of microbial risk assessment has been developing and maturing.Since that time, doseresponse relations have been developed for various pathogenic microorganisms (Haas et al. 1999;McBride et al. 2002), and microbial risk assessment investigations have been carried out for a number of those pathogens in water, food and other

Exposure profile
Hostpathogen profile

Analysis
Iterative refinement

Risk characterization
Host & pathogen characterization

Conceptual model
Figure 1 | EPA/ILSI generalized framework for assessing the risks of human infection following exposure to pathogens.
media.This section reviews and summarizes the doseresponse and risk characterization methods that have been most commonly employed in MRAs.

Evaluation of dose -response relations
The dose-response assessment provides a qualitative or quantitative description of the likelihood of adverse effects that may result from exposure to a microorganism or its toxin.
For MRA, the adverse health effects most commonly of interest are infection and illness.Infection may be defined as the invasion, colonization and multiplication of a pathogenic microorganism (Teunis et al. 1996).Since humans may come into contact with a wide range of pathogen concentrations, it is therefore useful to develop a functional relationship between a quantitative measure of microorganism ingestion and the risk of infection.A dose is defined as a quantitative measure for the intensity of exposure of the host to the pathogen of interest.The units of dose are usually given as the number of organisms ingested.By studying the effects of various doses, it is possible to determine a dose-response relationship between the dose and the frequency of infection within the exposed population of hosts.Researchers have studied quantitative descriptions of the dose-response relationships of organisms to provide insight into the risk of becoming infected after the ingestion of a certain dose of organisms (Teunis et al. 1996).(Dupont et al. 1995;Chappell et al. 1996;Moss et al. 1998;Okhuysen et al. 1998Okhuysen et al. , 1999;;Messner et al. 2001), Giardia lamblia (Rendtorff 1954a, b;Rendtorff & Holt 1954a, b), Vibro cholera (Hornick et al. 1971), rotavirus (Ward et al. 1986;Regli et al. 1991), poliovirus (Lepow et al. 1962;Katz & Plotkin 1967;Minor et al. 1981), Escherichia coli O157:H7 (Powell et al. 2000), and coxsackieviruses B4 and A21 (Suptel 1963;Couch et al. 1965).Generally, these data were obtained from feeding studies of healthy adults.Thus, certain portions of the population including children, the elderly, and individuals with compromised immune systems are not well represented by these data.

Dose
Typically the reported dose-response data have been fit to models that relate the probability of infection to the mean dose ingested.In some cases, illness as an end-point was also investigated; however, the conditional modeling of illness given infection has proven to be difficult (Teunis et al. 1996).The most common models, although not the only models used to relate an ingested dose to infection, are the exponential and beta-Poisson models (Haas et al. 1999).
Those models are summarized below.
The exponential model is based on the following assumptions (Haas et al. 1999): microorganisms are distributed in water randomly and thus follow the Poisson distribution for infection to occur, at least one pathogen must survive within the host the probability of infection per ingested or inhaled organism is constant.
Mathematically, the probability of infection (P inf ) as expressed by the exponential model is as follows: In the exponential model each microorganism has the same fixed probability (r) of surviving and reaching a host Unfortunately, in this approximation to the beta-  (2002).
New methods for dose-response assessment relying on Bayesian approaches have begun to appear in the literature over the last several years (Messner et al. 2001;Englehardt 2004;Englehardt & Swartout 2004).A detailed description of these methods is beyond the scope of this paper.
However, it is noteworthy that these methods attempt to address a major limitation of likelihood-based approaches (that data validation is not possible for the doses of pathogens that are consistent with public health goals).

Risk characterization methods
A literature review was recently conducted to document the status, advantages, and limitations of different types of microbial risk assessment risk characterization techniques (Soller et al. 2004).The literature review of approximately 1100 citations indicated that at the broadest level there was a distinction between direct estimates of risk or illness using epidemiologic data and indirect estimates using models.
Direct estimates entail collecting infection or disease outcome data, for example, prospective studies or outbreak investigations.Indirect estimates employ exposure data as input to numerical models to compute estimates of illnesses.
Based on the available literature, it appears that direct methods are most commonly used to assess the public health impact associated with a specific and known (or identifiable) exposure pathway.Those methods may not, however, provide the regulatory and management information for making decisions regarding changes in environmental conditions.For this purpose indirect methods can play a useful role.The literature review indicated that MRA methodologies vary primarily in the manner in which they address the unique properties of an infectious disease transmission system.The fundamental difference between these risk assessment techniques is that NRC paradigm models (static models) do not account for the properties that are unique to a dynamic infectious disease process (Table 1).In static models, the number of individuals that are assumed to be susceptible to infection is not timevarying, whereas in dynamic models that number is timevarying.

Static microbial risk assessment models
Assessments using a static model for evaluating microbial risk typically focus on estimating the probability of infection or disease to an individual as a result of a single exposure event.These assessments generally assume that multiple or recurring exposures constitute independent events with identical distributions of contamination (Regli et al. 1991).
Secondary transmission and immunity are typically not considered, assumed to be negligible, or that they effectively cancel each other out.In this context, secondary transmission would increase the level of infection/disease in a community relative to a specific exposure to pathogens, and immunity would decrease the level of infection/disease in a community relative to a specific exposure to pathogens.
In static MRA models, it is assumed that the population may be categorized into two epidemiological states: a susceptible state and an infected or diseased state.Susceptible individuals are exposed to the pathogen of interest and move into the infected/diseased state with a probability that is governed by the dose of the pathogen to which they are exposed and the infectivity of the pathogen (Figure 2).In variation in pathogen virulence and host susceptibility (Eisenberg et al. 1996(Eisenberg et al. , 2004)).The probability of infection is often multiplied by the number of exposed individuals to estimate the expected number of infected individuals for the exposure scenario under consideration.

Dynamic microbial risk assessment models
In a dynamic risk assessment model, the population is assumed to be divided into a group of epidemiological states.Individuals move from state to state based on Potential for secondary or person-to-person transmission of infection or disease exists.
Immunity to infection from microbial agents is typically not considered.
Exposed individuals may not be susceptible to infection or disease because they may already be infected or may be immune from infection due to prior exposure.
Dose -response function is the critical health component.
The dose-response function is important; however, factors specific to the transmission of infectious diseases may also be important.strategy involving a hierarchy of models from simple to increasingly complex models which could be traversed to make microbial risk assessment analyses more realistic while remaining mathematically tractable.It seems quite reasonable to anticipate that the issue of model complexity for microbial risk assessments will be an area of future research that will receive substantial attention.

REPRESENTATIVE MRAS FOR WATERBORNE PATHOGENS
A number of microbial risk assessments have been carried and self-limiting.The MRA method employed was similar to that described above for rotavirus (Gerba et al. 1996).Point estimate values were employed for the concentration of adenovirus in drinking water (0.01/l and 0.001/l), the volume of water ingested (2 l/day and 4 l/day), and the exponential dose-response parameter (r ¼ 0.4172).
The risk of illness was determined by multiplying the probability of infection by 0.5.The probability of mortality was calculated by multiplying the probability of illness by 0.01%.Annual risks were computed using a similar methodology as described above (Gerba et al. 1996).Mena et al. (2003) employed static MRA methods to evaluate the public health risk associated with drinking waters contaminated with coxsackieviruses.Coxsackieviruses are the most common non-polio enteroviruses found in domestic wastewater and in contaminated surface water, groundwater and drinking water (Mena et al. 2003).
Most coxsackievirus infections result in mild febrile illness, although coxsackieviruses are also capable of causing a wide range of more serious illnesses.The methods employed were similar to those described above for rotavirus and adenovirus with point values used to estimate exposure and the exponential dose -response relation.
In addition to the point estimate virus specific assessments described above, static MRA methods have also been When accounting for the different susceptibilities in the population, the model did over-predict disease incidence; however, the findings assume that the surveillance data used for comparison reflect actual waterborne cryptosporidiosis incidence.

Risk assessments employing dynamic models
Dynamic microbial risk assessment methods have been used to characterize the potential public health effects associated with rotavirus in drinking water (Soller et al. 1999), obtain insight into the epidemic process related to drinking water treatment failures (Eisenberg et al. 1998), characterize risks from microbiological contaminants associated with recreational activities (EOA 1995a, b;Eisenberg et al. 1996;Soller et al. 2003Soller et al. , 2006)), and estimate the bias associated with modeling the infectious disease process using a static model (Eisenberg et al. 2003;Soller et al. 2004).In all of these investigations probabilistic simulations were employed to account for variability and uncertainty in model parameters.
The fundamental difference between the investigations cited above and those described in the previous section is Mathematically, the epidemiological status of the population was modeled as a series of ordinary differential equations.Using this approach and assuming that the primary (drinking water) and secondary (person-to-person) transmission processes are independent, the change in the fraction of the population in any state from one time period to the next was computed.For example, the relative change in state S from one time period to the next due to primary infection was where: b SC1 is the rate at which the population moves from This investigation demonstrated that the risk of viral gastroenteritis attributable to the wastewater treatment facility from recreation was related not only to treatment efficacy of the wastewater treatment facility (Figure 8), but also viral loading from people recreating in the river (Figure 9).The primary advantage of employing a simulation based approach in this type of investigation is the ability to evaluate the potential benefits of proposed management options (Soller et al. 2006).Although MRA methods inherently do not characterize the cumulative risk associated with all pathogens potentially present in an environment, this study illustrated that it is possible to synthesize a model organism that captures the salient features of a class of pathogens of interest, and thus frame an investigation in a manner such that practical risk management decisions can be made.and a sanitary survey for all systems using surface water.

MICROBIAL RISK ASSESSMENTS CONDUCTED BY EPA OFFICE OF WATER
The economic benefits of the IESWTR are assumed to result entirely from the decreased probability of incidence of cryptosporidiosis as determined through MRA, and the avoidance of resulting health costs.Exposure to pathogenic protozoa such as Giardia or other waterborne bacterial or viral pathogens is almost certainly reduced by this rule, but was not quantified through MRA (Regli et al. 1999).
In the IESWTR analysis, a static risk assessment methodology was employed to quantify the number of infections, illnesses, and deaths of Cryptosporidium in drinking water.Data from human ingestion trials were used to derive the best fit value (k ¼ 239) and 95% confidence interval (132 -465) for the exponential doseresponse function.The exponential dose -response parameter was then modeled as a lognormally distributed variable.The rate of daily water ingestion was assumed to be lognormally distributed with mean 1.95 l and a standard deviation of 0.8 l.Assumptions were made about the performance of existing drinking water treatment processes in removing oocysts to estimate Cryptosporidium concentrations in finished water.It was estimated that the existing standard treatment resulted in a normal distribution of reduction with mean of 2.5 logs and a standard deviation of 0.63 logs.An alternative removal distribution was also evaluated (mean 3.0 logs with standard deviation of 0.63 logs).The viability -infectivity of Cryptosporidium in finished water was modeled as a uniform distribution with a minimum value of 5% and a maximum of 15%.The percent of infections that result in symptomatic illness was modeled as a triangular distribution with mean 39%, low of 19% and high of 0.62.
The authors used the above data to calculate the expected number of annual infections.Monte Carlo simulations were carried out to estimate the probability distribution of the estimated number of infections and illnesses.This approach is fundamentally consistent with the approach developed earlier for Giardia and viruses in drinking water (Regli et al. 1991).Discussions on the magnitude of waterborne disease associated with drinking water supplies continued throughout the 1990 s. Two epidemiology studies suggested that drinking water consumption might be responsible for approximately one third of all gastrointestinal illness (Payment et al. 1991(Payment et al. , 1997)).It is also noteworthy that a large waterborne disease outbreak in the city of Milwaukee, WI occurred under conditions that did not violate the drinking water regulations in effect at the time.That outbreak resulted in an estimated 403 000 cases of illness.(Imhoff et al. 2004).This estimate is based on data collected during 1998-1999 and translates to a rate of 0.72 illnesses per person-year (Imhoff et al. 2004).

The proposed Long
This CDC estimate is based on an analysis of random-digit dialing (RDD) telephone survey results from a sample population of 29 million persons (, 11% of the total US population) in 8 states during a 12-month period (Imhoff et al. 2004).This estimate is similar to an earlier CDC estimate of AGI in the US (0.75 illnesses per person-year from all sources) based on 1996-1997 data using the same RDD methodology, similar questions, and covering most of the same sample population.

Microbial risk assessment methods for the national estimate
In considering the national estimate of AGI specifically attributable to microbes in drinking water, it is reasonable to assume that communities served by systems with relatively contaminated source waters, sub-standard water treatment facilities, and/or contamination problems in their distribution systems are subject to a higher risk than communities where such issues are less of a concern.
Further, the risk of illness attributable to pathogens in drinking water in each community can be thought of as the sum of the risk from pathogens present in the water as it leaves the drinking water treatment facility (treated water risk) and the risk from pathogens in the distribution system (distribution system risk).
Based on the discussions presented earlier in this paper, it seems reasonable that MRA methods could be used to provide some insight into the national incidence of AGI attributable to drinking water provided that appropriate data are available as input to MRA models.If it is assumed that the risk attributable to drinking water in each community is the sum of the treated water risk and the distribution system risk, it seems feasible that MRAs could be developed to characterize the risk associated with each of these components.It should, however, be recognized that MRA-based infection or illness estimates derived from pathogen-specific data will inherently under-estimate the total risk attributable to drinking water because the total risk will be a function of all pathogens present in drinking water whereas the MRA estimates will likely be based on data for specific pathogens.Thus, MRA methods may be most useful for providing a reasonable lower bound characterization of the national estimate.

Risks from pathogens in treated drinking water
To identify the types of MRA models that may be of most use for providing insight into the national estimate, it is necessary to consider treated water risk and distribution system risk separately.Characterizing the risk associated with pathogens in the treated water (at the point that the water enters the distribution system) could be accomplished either with a static model or a dynamic model as described previously in this paper.For example, if a static model were used, an assessment similar to those conducted by EPA for previous regulations (IESWTR or LT2ESWTR) may be appropriate.On the other hand, if a dynamic model were used, in addition to considering source water quality and drinking water efficacy, the relative importance of personto-person transmission of disease and/or immunity to the pathogenic agent of particular concern could be investigated and characterized (Eisenberg et al. 2003;Soller et al. 2003).
At the present time, the potential bias associated with modeling treated drinking water risk as a static process compared to a dynamic process is unknown and has not been investigated.Previous work for exposure to pathogens from reclaimed water exposures indicated that there is a substantial potential for person-to-person transmission and immunity to impact the results of an assessment in a meaningful way relative to the results obtained using similar assumptions and a static model (Soller et al. 2004).
However, since exposures to pathogens from a drinking water route of exposure may occur as frequently as daily for a large portion of the population, the results from the reclaimed water investigation (which investigated less frequent exposures and a smaller proportion of the population exposed) may not be applicable.Thus, in selecting a model for an assessment of the risk associated with pathogens in treated drinking water, it appears that some consideration of and/or justification for not including person-to-person transmission and immunity would be appropriate.

Risks from pathogens in the distribution system
Risks associated with exposure to pathogens from contamination in the distribution system of a public water supply site at which infection may result.Under this model the dose required to cause infection in half the exposed population is N 50 ¼ 2ln(0.5)/r.The dose-response relation for many protozoans and viruses tend to follow this model.The biological implication of this model is that differential susceptibility in the challenged population tends not to be strong (McBride et al. 2002).The beta-Poisson model is based on similar assumptions to the exponential model except that the third assumption (that the probability of infection per ingested organism is constant) is relaxed.This model allows the probability of infection per ingested or inhaled organism to vary with the population.In this model the probability of surviving and reaching a host site ("r" in the exponential model) is beta distributed, and thus the model contains the two parameters (a and b) of the beta distribution.The most commonly used approximation to the beta-Poisson model is as follows:

Figure 2 ,
Figure 2, the solid lines represent the movement of individuals from one epidemiological state to another, and the dotted lines represent the movement of pathogens.Although humans may be exposed to pathogens from a number of potential environmental sources, static models typically employ the assumption that susceptible individuals are exposed to pathogens from the specific pathway under

Figure 2 |Figure 3 |
Figure2| Conceptual model for a static risk assessment.
out for waterborne pathogens including assessments employing both static and dynamic models.This section summarizes static and dynamic MRAs representative of those found in the literature with an emphasis on MRAs carried out for drinking water exposures.Risk assessments employing static modelsStatic microbial risk assessment methods have been used to evaluate the potential public health effects associated with drinking water contaminated with a range of waterborne pathogens including both viruses and parasites.The methods employed in those assessments have varied from relatively straightforward assessments using point estimate values for model parameters to more complex assessments relying on stochastic (probabilistic) models.For example, in evaluating the public health impact from exposure to human rotavirus in drinking water, Gerba et al. (1996) used point estimate values for the concentration of rotavirus in drinking water (0.004/l and 100/l) based on surface water concentrations from previously published studies and an assumed 99.99% reduction of rotavirus through drinking water treatment.The volume of water ingested (2 l/day and 4 l/day) and beta-Poisson dose-response parameters (a ¼ 0.26, b ¼ 0.42) were also based on point estimate values.The probability of clinical illness was determined by multiplying the resulting probabilities of infection by 0.5.The probability of mortality was determined by multiplying the probability of illness by 0.01% for the general population and 1% for the elderly.Yearly risks were calculated as a function of daily risks: P ¼ 1 2 (1 2 P daily ) 365 .Crabtree et al. (1997) employed a static risk assessment model to evaluate the potential heath effects associated with adenovirus from drinking water exposures.There are 47 types of adenoviruses with infections resulting in conjunctivitis, pharyngitis, pneumonia, appendicitis, bronchiolitis, and gastroenteritis.Adenovirus infections are usually acute used (1) in assessments for viruses in which combinations of characteristics of different viruses have been assumed to determine appropriate level of drinking water treatment for viruses as a class of contaminants, and (2) in conjunction with Monte Carlo simulation techniques to account for variability and uncertainty in model parameters.For example,Regli et al. (1991) suggest that the enteroviruses (a subgroup of enteric viruses) for which a standard analytical method has been available for some time, could serve as an indicator of worst-case potential occurrence for any specific virus.Similarly, the dose-response relation for rotavirus has been used to derive upper-limit risk estimates for viruses in water as rotavirus is the most infectious waterborne virus for which dose-response information is currently available(Haas et al. 1993).Haas et al. (1993) accounted for uncertainties in exposure assessments (lognormal distribution for volume ingested) and the doseresponse relationship (95% confidence intervals about the maximum likelihood estimate for a and b) for viruses in drinking water by applying Monte Carlo simulation techniques.Static microbial risk assessment methods have also been used to evaluate the potential public health effects associated with drinking water contaminated with G. lamblia(Rose et al. 1991;Teunis et al. 1997) and Cryptosporidium(Perz et al. 1998;Teunis & Havelaar 1999;Makri et al. 2004).Rose et al. (1991) conducted a static risk assessment investigating the potential health risks associated with G. lamblia in drinking waters.The methodology employed was similar to that described above for rotavirus and adenovirus.Point estimates were used to characterize the volume of water consumed daily (2 L), average levels of cysts in surface waters (0.22 -104/ 100 l), reduction of cycts due to drinking water treatment (99.9%), and the dose-response relation.Annual risks were computed as described above and source water concentrations corresponding to annual risks of 1/10 000 were derived.Teunis et al. (1997) conducted an assessment of the risk of infection by Cryptosporidium and G. lamblia in drinking water from a surface water supply in which the major contributing factors to risk were each treated as stochastic variables.The stochastic variables investigated included the concentration of cysts (G.lamblia) and oocysts (Cryptosporidium) in raw water, the recovery of the detection method, the viability of recovered cysts or oocysts, the removal of organisms in the treatment process, and the daily consumption of unboiled tap water.A frequency distribution for the probability of infection was developed based on the results of the probabilistic simulations.The advantage of this stochastic approach was clearly evident in the results that indicated that the uncertainty in the estimated removal efficiency of the treatment process dominated the uncertainties in all other contributing factors.In a similar investigation to that described above, Teunis & Havelaar (1999) conducted a case study in which the risk of human infection from Cryptosporidium parvum in drinking water was characterized.Exposure was assessed by considering the different stages from river water to consumed tap water.Oocyst counts in the river water were corrected for the performance of the detection method via a probabilistic process.Before treatment, the water was assumed to be stored in storage basins for several months.The removal and inactivation of oocysts during this process was modeled as a stochastic process.Assessment of the performance of a drinking water treatment process was modeled using spores from sulfite reducing clostridia as the surrogate organism.Inactivation by disinfection was estimated by using a process model from the literature.Consumption of unboiled tap water was modeled using the lognormal distribution based on a Dutch survey.The daily ingested dose was then calculated by means of Monte Carlo simulation.For dose-response assessment, the beta-Poisson model was employed.The dose-response relations for infection and illness were used to generate, via Monte Carlo methods, distributions for the risk of daily, annual, and lifetime infection and illness.A static MRA model was employed by Perz et al. (1998) to examine the potential role of tap water in the transmission of endemic Cryptosporidium parvum infection.Their model had two components: an exposureinfection component relate low-dose exposure to infection; and an infection -outcome component to include the probabilities of clinical outcomes leading to case detection and reporting.A concentration of 1 oocyst/1000 L was assumed for the treated and distributed drinking water.The population was divided into four subgroups which included adults and children both with and without AIDS.The computed risks were used in conjunction with the 1995 New York City population to estimate the number of cases due to ingestion of tap water.Makri et al. (2004) carried out a case study static MRA investigation to develop and test a predictive model for waterborne cryptosporidiosis risk on a regional scale, accounting for persons living with AIDS, and to determine if the model predicted regional patterns of cryptosporidiosis incidence.Similar to the investigations carried out by Teunis et al. (1997) and Teunis & Havelaar (1999), probabilistic simulations were employed to account for variability and uncertainty in model parameters.Analytic recovery, source water concentration, viability, consumption, dose-response relation, and the probability of illness given infection were all modeled as probabilistic distributions for the exposure component.The Monte Carlo routine resulted in daily and annual probability estimates of Cryptosporidium infection, illness, prolonged illness, and case detection.This study differs from prior work as it compares predictions based on water quality data with endemic cryptosporidiosis surveillance and accounts for differential susceptibility.
Figure 4.The population was divided into four states as follows:(1) individuals susceptible to infection (S); (2) individuals

Figure 6 |Figure 7 |
Figure 6 | Representative output from a dynamic MRA model.
Term 2 Enhanced Surface Water Treatment Rule (LT2) was published in the Federal Register on August 11 2003 and the final rule was published on January 5, 2006.This rule was intended to reduce the occurrence of viable waterborne pathogens, particularly Cryptosporidium, in public drinking water delivered by surface water systems (US EPA 2003).The quantified health benefits estimated for this rule result from reducing the incidence of adverse health effects caused by drinking water containing Cryptosporidium.Similar to the IESWTR, microbial risk assessment was used to characterize the expected incidence of adverse health effects associated with exposure to Cryptosporidium and to estimate the benefits of actions taken to reduce the exposure.The static risk assessment methodology, comprised of a two-dimensional Monte Carlo simulation model, was employed to quantify the benefit estimates of the LT2 by calculating the difference between illness and death estimated for the baseline condition (pre-LT2) and after implementation of the LT2.Benefit estimates are the predicted number of illnesses and deaths avoided because of the implementation of the regulatory requirement.The risk assessment model is conceptually similar to that implemented for the IESWTR; however, the model is more sophisticated in its treatment of variability and uncertainty and more detailed in its exposure assessment.The LT2 risk model integrates dose-response and exposure assessment components into a Monte Carlo simulation model which is implemented in two steps.The first step of the risk assessment model was structured as a two-dimensional Monte Carlo simulation.A two-dimensional simulation is used when the model includes both uncertainty and variability components in the inputs, and where it is necessary to clearly distinguish the influence of these elements in the model output.Uncertainty components included the dataset representing source water concentrations, the true distribution of source water oocyst concentration, the true distribution of pre-LT2 oocysts removal, the morbidity factor, the fraction of oocysts that are infectious, and the true mean of the exponential dose-response infectivity parameter.The form of the dose -response model employed was one that predicted the probability of acquiring one or more infections over the time period studied.Variability components included in the model were the source water concentration, pre-LT2 Cryptosporidium removal, log reduction achieved due to treatment, and the volume of water ingested daily.Of note is the fact that the volume of water ingested was assumed to have mean of 1.07 l per day (1.23 liters/day minus 0.16 liters/day, the mean consumption by those who reported bottled water as their main source of drinking water).Variability in individual consumption was reflected by calculating the individual risk reduction estimates; however, that variability was not reflected in the overall estimates of variability in illness.By structuring the first step in the modeling process in this way, it was possible to characterize both the distribution of the individual annual risk of illness in the affected population and the overall population average annual risk of illness.In the second step of the risk assessment model, the number of cases of illness and mortality and the confidence bounds on those estimates were computed for the various pre-LT2 and post-LT2 assumptions regarding Cryptosporidium in source water.During this second step, the number of illness cases was adjusted to account for secondary transmission.To account for uncertainty in the secondary spread factor, a triangular distribution was used with a low of 10%, a high of 40% and a most likely value of 25%.These estimates were derived based on a summary of available outbreak data.The risk characterization process was implemented to describe the reduction in general population risk for unfiltered systems and filtered systems.Morbidity risk in the analysis was based on studies of infectivity and morbidity on healthy volunteers.No data were available to characterize differential infectivity or morbidity for the immunocompromised or other sensitive subpopulations.The mortality risk was based on data from the 1993 Milwaukee outbreak which indicated that all fatalities were in sensitive subpopulations.Thus, all of the quantified deaths avoided due to LT2 were presumed to be lives saved in sensitive subpopulations.USE OF MICRIBAL RISK ASSESSMENT TO INFORM THE NATIONAL ESTIMATE Previous estimates of waterborne disease in the US Questions about the magnitude of waterborne disease and the importance of different risk factors associated with waterborne disease occurrence in the US have been a subject of interest for many years.A comprehensive review of this topic was completed and can be found in this issue (Roy et al. 2006).The following description is a very brief summary of previous estimates of waterborne disease in the US.A number of different estimates of the national occurrence of waterborne disease were published in the 1970 s and 1980 s.These estimates are highly uncertain as they were not based on studies specifically designed to obtain such an estimate nor are they based on MRA methods.For example, Hauschild & Bryan (1980) estimated the annual incidence of both food-and waterborne disease in the US for 1974 and 1975 to be 1400 000 to 3400 000 cases, Morris & Levin (1995) estimated that 1.8 million cases of waterborne disease and 1800 deaths occur annually, and Bennett et al. (1987) estimated the annual incidence of waterborne disease to be more than 900 000 cases and almost 900 deaths.

Mead
et al. (1999) estimated the annual total number of illnesses in the US caused by known pathogens by adjusting for under-and non-reporting of illnesses.Based on their methodology, an estimated 38.6 million cases of GI illness occur annually due to known pathogens in the US from all exposures including food, water and other routes of exposure.Of those 38.6 million cases, it was estimated that 5.2 million cases are from bacterial pathogens, 2.5 million are from parasitic pathogens, and 30.9 million are from viral pathogens.The most recent CDC estimate of AGI from all types of exposures (e.g.food, person-to-person transmission, water, air, etc.) in the US is 195 million episodes per year Haas et al. (1999)or McBride et al.

Table 1 |
Comparison of static and dynamic risk assessment models

Table 2 |
Epidemiological states for a representative dynamic model

Table 3 |
Rate parameters for a representative dynamic model Rate of movement from a diseased state (infectious and symptomatic) to an asymptomatic (carrier) state.1/dcorresponds to the duration of symptoms during infection.gRate of movement from a post-infection state (not infectious, asymptomatic, and not susceptible to infection) to a susceptible state.1/gcorresponds to the duration of immunity or protection from infection.b1Rate of movement from a susceptible state to an exposed state due to exposure to pathogens from an environmental source (i.e.not person-to person transmission).Function of the number of pathogens to which an individual is exposed and the infectivity of the pathogen of interest.The infectivity is described quantitatively through a dose -response function which is comprised of one or two dose-response parameters.
a Rate of movement from an exposed state to a carrier (infectious and asymptomatic) state or a diseased state (infectious and symptomatic).1/a corresponds to the latency period prior to infection for the pathogen of interest.s Rate of movement from a carrier state to a post infection state.1/s corresponds to the duration of infectiousness, or equivalently, the duration of asymptomatic shedding of pathogen in feces.d b2 Rate of movement from a susceptible state to an exposed state due to exposure to pathogens from secondary (person-to-person or person-to-environment-person) transmission.P sym Probability of a symptomatic response.Clinical data describing the proportion of infected individuals that develop symptoms.Occam's Razor 1 is a useful starting point in considering model complexity; however, the selection of a model type does involve tradeoffs.Biological or demographic "realism" can come at the cost of analytical complexity that distances the model from available data.Further, each model form involves certain types of assumptions that may or may not be realistic or appropriate for a particular situation.With the perspective that different model forms and accompanying analytical approaches may be necessary for different applications, Koopman et al. (2001) suggest an analysis