© 2001 World Health Organization (WHO). Water Quality: Guidelines, Standards and Health. Edited byLorna Fewtrell and Jamie Bartram. Published by IWA Publishing, London, UK. ISBN: 1 900222 28 08Risk assessmentChuck Haas and Joseph N.S. EisenbergThis chapter introduces the technique of microbial risk assessment and outlinesits development from a simple approach based upon a chemical risk model to anepidemiologically-based model that accounts for, among other things,secondary transmission and protective immunity. Two case studies arepresented to highlight the different approaches.8.1 BACKGROUNDQuantifiable risk assessment was initially developed, largely, to assess humanhealth risks associated with exposure to chemicals (NAS 1983) and, in itssimplest form, consists of four steps, namely:x hazard assessmentx exposure assessmentx dose–response analysisx risk characterisation.Risk Assessment, Haas C and J. Eisenberg (2001) In: Water quality - Guidelines, standards and health: Assessment of risk and risk management for water-related infectious disease, L. Fewtrell and J. Bartram, World Health Organization-Geneva and Intl. Water Assoc.-London162 Water Quality: Guidelines, Standards and HealthThe output from these steps feeds into a risk management process. As will beseen in later sections this basic model (often referred to as the chemical riskparadigm) has been extended to account for the dynamic and epidemiologiccharacteristics of infectious disease processes. The following sub-sectionselaborate on the chemical risk paradigm as outlined above.8.2 CHEMICAL RISK PARADIGM8.2.1 Hazard assessmentFor micro-organisms, hazard assessment (i.e. the identification of a pathogen asan agent of potential significance) is generally a straightforward task. The majortasks of Quantitative Microbiological Risk Assessment (QMRA) are, therefore,focused on exposure assessment, dose–response analysis and riskcharacterisation. The task of risk management is one of deciding the necessityof any action based upon the risk characterisation outputs, and incorporatessignificant policy and trans-scientific concerns.One outcome of the hazard analysis is a decision as to the principalconsequence(s) to be quantified in the formal risk assessment. With micro-organisms, consequences may include infection (without apparent illness),morbidity or mortality; furthermore, these events may occur in the generalpopulation, or at higher frequency in susceptible sub-populations. Althoughmortality from infectious agents, even in the general population, cannot beregarded as negligible (Haas et al. 1993), the general tendency (in watermicrobiology) has been to regard infection in the general population as theparticular hazard for which protection is required. This has been justified basedon a balance between the degree of conservatism inherent in using infection asan endpoint and the (current) inability to quantify the risks to more susceptiblesub-populations (Macler and Regli 1993).8.2.2 Exposure assessmentThe purpose of an exposure assessment is to determine the microbial dosestypically consumed by the direct user of a water (or food). In the case of watermicrobiology, this may necessitate the estimation of raw water micro-organismlevels followed by estimation of the likely changes in microbial concentrationswith treatment, storage and distribution to the end-user (Regli et al. 1991; Roseet al. 1991). A second issue arising in exposure assessment is the amount ofingested material per ‘exposure’. As a default number, two litres/person-day isused to estimate drinking water exposure (Macler and Regli 1993), although thismay be conservative (Roseberry and Burmaster 1992). For contact recreationalRisk assessment 163exposure, 100 ml/day has often been assumed as an exposure measure (Haas1983a), but actual data to validate this number are lacking.8.2.3 Dose–response analysisIt is generally necessary to fit a parametric dose–response relationship toexperimental data since the desired risk (and dose) which will serve to protectpublic health is often far lower than can be directly measured in experimentalsubjects (at practical numbers of subjects). Hence it is necessary to extrapolate afitted dose–response curve into the low-dose region.In QMRA, for many micro-organisms, human dose–response studies areavailable which can be used to estimate the effects of low level exposure tomicro-organisms. In prior work, it has been found that these studies may beadequately described by one of two semi-mechanistic models of the infectionprocess. In the exponential model, which may be derived from the assumptionof random occurrence of micro-organisms along with a constant probability ofinitiation of infection by a single organism (r), the probability of infection (PI) isgiven as a function of the ingested dose (d) by:(8.1)For many micro-organisms, the dose–response relationship is shallower thanreflected by Equation 8.1, suggesting some degree of heterogeneity in themicro-organism-host interaction. This can be successfully described by the beta-Poisson model, which can be developed from Equation 8.1 if the infectionprobability is itself distributed according to a beta distribution (Furumoto andMickey 1967a,b; Haas 1983b). This model is described by two parameters, amedian infectious dose (N50) and a slope parameter (D):(8.2)Figure 8.1 depicts the effect of the slope parameter on the dose–responserelationship; in the limit of Į ĺ , Equation 8.2 approaches Equation 8.1.)exp(1 rdPI164 Water Quality: Guidelines, Standards and HealthFigure 8.1. Comparison of exponential and beta-poisson dose–response functions.The exponential and beta-Poisson models are two dose–responserelationships that can be developed from biologically plausible assumptionsabout the infection process (Table 8.1 outlines the best-fit dose–responseparameters for these models for a number of human pathogens). A generalframework for plausible models can also be derived.In addition to such quasi-mechanistic models, a variety of empirical modelsare possible, three models which have been used (primarily in chemical riskassessment), are the log-logistic, the Weibull, and the log-probit.Generally, several models may fit available data in a statistically acceptablesense, and yet provide very different estimates for the risk at an extrapolatedlow dose. This situation is one that has frequently been encountered in chemicalrisk assessment (Brown and Koziol 1983).
View Full Document
Unlocking...