Infectious organisms resistant to disinfectants, such as protozoa and viruses have caused several outbreaks around the world, including in developed countries. The most noticeable pathogen is Cryptosporidium which caused the Milwaukee outbreak [1, 2], and several episodes in the US, the UK and Canada [3–6]; but Giardia lamblia, the most common intestinal protozoan in the US, is also frequently reported in association with waterborne diseases [7–9]. Cysts may be found in water as a result of fecal contamination from both man and animal. Thus, protozoa pose major challenges to design and maintenance of safe water supplies. Microbial risk assessment is an important tool to manage these risks because it allows water quality standards and other management decisions to be based on quantitative estimates [6, 10, 11]. The US-EPA, for instance, has recommended that a treatment be provided to ensure that populations are not subject to a yearly risk of infection greater than 10^-4 [7, 12, 13].

The quantitative microbiological risk assessment approach is now well established [9, 14–16]. Dose-response functions have been proposed and reviewed for waterborne pathogenic micro-organisms and for microbial food safety [17–23]. The dose-response functions that are currently used in risk assessment studies for protozoa stem from clinical trials where adult healthy volunteers have been exposed to known amounts of well characterized micro-organisms and followed to ascertain the development of infection (faecal excretion, seroconversion). Rendtorff (1954) [24] and Dupont et al (1995) [25] performed controlled human feeding trials respectively for Giardia lamblia and Cryptosporidium parvum (oo)cysts. The exponential and the beta-Poisson model are commonly used to describe these relationships [9, 15, 17, 22]. They are based on the assumptions of independent, single hit action, and random distribution of the micro-organisms in the inoculum. But it must be kept in mind that infection was measured by cyst excretion, not by symptoms. This experimental approach is faced with severe ethical restrictions that limit trials to healthy adult volunteers. Also, issues regarding strain selection make generalization of such data questionable [9]. Hence, variability related to infectivity of different strains and to the immune response of hosts could not be accommodated. These limitations must be taken into account in risk assessment studies [7].

Epidemiological data offer the advantage of being based on natural infection events, to bear on a variety of host populations and on microorganisms that are found in real life situations [9]. Epidemiological studies have already been used to assess dose-response functions in the context of outbreaks, but serious difficulties are encountered for retrospective exposure assessment. For instance, a risk assessment was conducted on the basis of disease risk parameter estimated after the dose-response model developed by Dupont on healthy human volunteers [25], and compared to the observed attack rates during the Milwaukee Cryptosporidium outbreak [1] to derive concentration estimates of the pathogen in drinking water; the results proved consistent with the levels that were found in ice samples (1.4 oocysts/L versus 0.79 oocysts/L) [18, 26]. Similar conclusions were drawn in another outbreak in Bradford (UK) [15].

The present work uses epidemiological data collected in a non epidemic setting and compares the risk estimates derived from the published dose-response parameters for Giardia to the observed incidence data in a general population exposed to environmental strains of micro-organisms.

The E.MI.R.A (Epidemiology and MIcrobial Risk Assessment) study was carried out between October 1998 and June 1999, in south-east France. It combined a daily epidemiological follow-up of digestive morbidity among a panel of volunteers, and a microbiological surveillance of drinking water. The E.MI.R.A study design has been described in detail elsewhere [27–29].

Total tap water intake follows a normal distribution and varies according to season (average in winter = 1.6 L/d, SD = 0.7 L/d; average in spring = 1.8 L/d, SD = 0.8 L/d). For Monte Carlo simulation, we used our data set on un-boiled tap water intake (0.8 L/d, 0.3 L/d). Detailed description of daily drinking water consumption of E.MI.R.A volunteers has been published [32].

When comparing the risk estimates derived from the Rendtorff model and from our observations, the best correspondence is found with a 1/5 abatement factor : for hypothetical Giardia cysts counts ranging between 1 to 200/L (i.e. in the range of measured concentrations), modeled risk values exceed the estimated incidence rate by a factor spanning from 6.5 to 3.0 for a 1/1 abatement factor, and from 3.3 to 1.5 for a 1/2 abatement factor; this ratio did not exceed 1.6 for a 1/5 abatement factor, with values close to 1 for low Giardia concentrations. Assuming a daily water intake of 2 L and a Giardia concentration of 10 cysts/100 L, the estimated relative excess risk would be 12 % according to the Rendtorff model (with the 1/5 abatement factor), against 11 % applying the corresponding OR to the baseline rate of ADC in absence of Giardia cysts. Figure 1 exhibits risk estimates (expressed in percent increase above the baseline incidence rate, with the 95 % CI limits) from our study results and from the Rendtdorff model values after we applied a 1/5 abatement factor to the cysts counts; given this correction, it shows a good agreement between the two slopes for contamination levels in the domain of observed data (up to 110/L). The underlying assumption in figure 1 is that 20 percent of the detected cysts are capable to induce disease.

The hazard that Giardia and Cryptosporidium pose to communities supplied by surface water has been early recognized by the US legislation, which requires a minimum of three log removal/inactivation of Giardia and two log removal of Cryptosporidium [36, 37]. These objectives were derived in order to ensure that consumers would not be exposed to a risk greater than one per 10 000 individuals per year [11, 19]. In the UK, the "Drinking Water Inspectorate" also takes into account Cryptosporidium risks, requiring a maximum level of 10 oocysts/100 L. UK regulations also require that water providers perform a risk assessment on their water treatment facilities and implement continuous monitoring for Cryptosporidium [38]. In continental Europe, no rules exist to date for Giardia or Cryptosporidium in drinking water [39].

While risk assessment is an important tool for management of waterborne infectious risks, it is affected by several uncertainties : there are knowledge gaps about pathogens concentrations, and weaknesses in specification of dose-response functions, amongst other factors [8, 9, 40]. Dose-response data for infection with Giardia lamblia were derived from an experiment on healthy adult [24]. Now, healthy adults are not the most important group from a public health perspective [8, 41]. The E.MI.R.A study included some sensitive subpopulations (i.e., children, elderly subjects, pregnant women, but not subjects with chronic diseases), which are excluded from experiments for evident ethical reasons. Gerba et al assessed that this subgroups represent about 20 percent of the US population [42].

Our survey provided directly drinking water data; so we did not need to formulate additional hypotheses on pathogens survival at early steps of the water supply chain, a critical uncertainty factor in the waterborne microbial risk assessment procedure [8, 40]. The detection method we used, however, does not differentiate between live, virulent and non infectious micro-organisms; nor is it strictly specific of human strains of parasites. On the other side, the imperfect recovery of cysts from water samples may allow for some cysts counts underestimation. The recovery rate of 50 % of the technique we used had not been assessed in a different framework than that of the present study [31], but it stands in the range of values that were published using the same U.S. Environmental Protection Agency method 1623, in the low side [43]. Given these different factors, only a fraction of the detected parasites are viable, infectious and virulent to humans. Several measures of viability have been proposed in the literature, such as germ morphology, in vitro excystation assays, infectivity tests in animal or cellular models [8, 44, 45]. No assay on viability/infectivity was conducted within the E.MI.R.A framework. Teunis et al reported that Giardia cysts shedding might occur at very low infectious doses (3–10 cysts) [8]. In our study, Giardia cysts were found at non negligible levels in tap water (10, 20, 30 and 110 cysts per 100 L).

Other critical issues in exposure assessment should also be discussed. We hypothesised, in this analysis, that one water sample represented the average parasitical quality of water during 7 days (day of sampling and 3 days about). Backing this assumption, a Canadian study showed small differences in Giardia cysts counts along consecutive days [46]. It was conducted in highly polluted surface water, however, and whether this observation applies in the low contamination level of the present study is unclear. While we are not aware of published papers describing short term variability of (oo)cysts concentrations in ground or surface water used for drinking water, seasonality has been assessed in several studies, with uneven findings according to watershed vulnerability and to the time pattern of contaminants discharge into surface water [47–51]. Hence, the time representativity of our water samples is not fully characterised despite our repetitive sampling. The latency for Giardia infection and disease is commonly reported to be comprised between one and 14 days. Thus, the seven days morbidity period we chose might be incomplete; it showed the best statistical associations, however, compared to 2 weeks or one month sampling frames. Exposure to waterborne pathogens depends upon infectious micro-organisms concentration, but also on the amount of water that is ingested. Most risk assessment studies use 2 L as the default value for daily drinking water intake. Here, we applied the consumption values that were declared by the study volunteers for un-boiled tap water. Finally, one cannot rule out some degree of confounding of the effect attributed to Giardia due to concomitant presence of other pathogens, a situation that occurred during the true outbreak that was observed during the study. It should be noted that our risk estimates for Giardia are adjusted for other measured concomitant pathogens. In a recent report of the large Bergen waterborne giardiasis outbreak, Norway, the authors underline how difficult it may be to assign cases to a given pathogen, even using sophisticated Giardia and Cryptosporidium genotyping, both of cases stools and water samples [52, 53]. The two protozoans were not found concomitantly in our study, however, but this might be partly explained by imperfect recoveries.

It is reasonable to think that the association between digestive morbidity and Giardia exposure differs according to age. Unfortunately, we could not assess this heterogeneity, because of lack of statistical power, the reason why we split our data in only three age categories. The literature does not provide such information either and this calls for further research. Our risk estimates for Giardia are consistent with those derived after the Rendtorff model provided that we down-weight the measured germ counts by a 1/5 factor. This parameter encompasses several dimensions: (i) the recovery rate of the analytical techniques is not 100 percent (with the Gelman^® protocol used here this rate averages 50 percent); (ii) some detected cysts might be non viable and non infectious to humans; (iii) some infected subjects might be asymptomatic. It is difficult, and out of the scope of this study, to devise what role these different factors play, in the net estimated 20 percent weigh. For instance, one could hypothesise that the incidence rate could have been greater in this general population, because the viability and infectivity characteristics of the pathogens in the natural environment might be larger than those assayed in the Rentdorff experience. In their 1991 work, Le Chevallier et al had found that 13.3 percent of the detected cysts had a viable morphology [44]. A probability of disease, when infected, of about 70 percent (95 % CI : 35–100) has been assessed for Cryptosporidium [54] and 37 % of oocysts found in six US watersheds showed viable and infectious [36]. For Giardia, symptomatic infection proportions have been reported ranging between 50 % and 67 %, and as high as 91 % [7]. On the other hand, there might be some degree of population immunity, associated with the prevalence of cysts in drinking water [55, 56]

While substantial efforts are ongoing to improve risk assessment of Cryptosporidium, because of its well established virulence for immunocompromised subjects, fewer risk assessment data concern Giardia. The E.MI.R.A study had been designed to deal with the 2 protozoa, but presence of Cryptosporidium in tap water of our French communities was too rare an event for analysis. Evidence of consistency between estimated and observed germ counts after the Milwaukee Cryptosporidium outbreak also gives some support to the validity of risk parameters that were computed through clinical infection assays; we suggest, however, that studies along the same rationale as the E.MI.R.A. study could be designed to assess dose-response functions for this pathogen in a non-epidemic context, because waterborne exposure to protozoa may result in sporadic cases and not necessarily in outbreak situations, as it was seen in the present study. In an Australian study in Sydney, raw and finished water were contaminated by Cryptosporidium and Giardia without discernable increases in illness in the general population [57]. It is well known, however, that only a very small fraction of individuals with acute digestive conditions ask for medical attention and are adequately diagnosed and registered [5].

Quantitative microbial risk assessment is a useful tool for the management of waterborne infectious risks. This study, conducted in a general population in a natural setting, suggests that the exponential model for waterborne Giardia risk derived from clinical experiments is consistent with epidemiological data. However, much remains to be learned about key characteristics that may heavily influence quantitative risk assessment results, such as germ viability, infectivity and virulence, or analytical recovery performances in natural environments.