Risk management of Giardia exposure in South American communities

Hazard Identification

The protozoa Giardia is one of the various pathogenic microorganisms that cause acute diarrhoeal disease (ADD) and account for the largest percent of total global outbreak relate to water contamination problems (WHO^[1]) due to: a) Their persistence in water reservoirs for moderate to long periods (weeks to months) b) High resistance to disinfection procedures commonly used in water treatment stations c) Low infectious doses (a single pathogen can cause infection). d) High concentrations excreted [up to 105 oocysts per gram of feces]. e) Animal sources are important for contamination of water supplies. Consequently, the drinking water guidelines of World Health Organization (WHO) have suggested the use protozoa as pathogens to monitor drinking water quality (WHO^[1]).

Exposure Assessment

Ingestion of drinking water is the primary route of exposure in this risk assessment. The exposure pathway begins at the tap where treated water may contain Giardia cysts as a result of potentially insufficient treatment, as observed in Minas Gerais, Brazil and Caracas,Vz (Leal^[7], Razzolini^[8], Betancourt and Mena^[4]). Residents of these locations may attempt to protect themselves from Giardia by use of Point of Use (POU) water treatment devices in the home. Thus, dose to a member of these communities is the weighted by the POU-adherent population and is determined as followed: dose=cv[〖w10〗^(-r)+(1-w)]

where c = concentration (count/L), v = vol. water ingested per day (L/day), r = log removal value of POU device, and w = compliance* rate (%).

The concentrations of Giardia in treated drinking water were obtained from a study of the Minas Gerais treatment plants (Leal^[7]) and from Caracas water treatment plants II and III (Betancourt and Mena^[4]). While one water treatment plant, subsistema Itapeceria, consistently had concentrations of Giardia well above the limit of detection (LOD), the others often reported Giardia concentrations below the LOD. For subsistema (ss) Itapeceria, Giardia concentrations were bootstrapped and the median values fitted to obtain a lognormal distribution of counts per liter. Concentration data for the remaining two water treatment plants, named ss Vz (in Caracas) and ss Para (in Minas Gerais) were fit to a Poisson distribution using a Maximum Likelihood Estimation (MLE). To account for the different susceptibility experienced by children and adults, we used two point estimates for volume of drinking water ingested and for the dose response curve (detailed the dose-response tab). After Kahn and Stralka^[9], adults were assumed to ingest 1.5 L/day while children were assumed to have ingested 0.44 L/day (2009). Though log removal rates (LRV) from household water treatment was sparse, a few estimates were obtained for Ceramic filters (Lantagne^[10]; Brown and Sobsey^[11]; Michen et al.^[12]) and biosand filters (Sangya-Sangnam 2009, Kubare^[13]); the LRV was assumed to be approximately normal and the standard deviation was assumed to be approximately 10% of the mean LRV. Model input parameters were collected from the literature and are summarized in Table 1.

Table 1. Parameters for dose calculations

• Note that here, compliance means both having a POU household water treatment device and perfectly adhering to the use of the device for all ingested water.

Dose Response

Exponential, k is 0.020

 

Risk Characterization

Dynamic Transmission Model

Static risk characterization models are useful for calculating the probability of infection based on a variety of exposures and dose response curves. With statistical distributions and Monte Carlo simulations, these models can also account for uncertainty or variability in the values selected for the model parameters. This is an effective strategy for quantifying reductions in drinking water risk when using point of use devices in private and public settings. However, one of the primary limitations of static risk characterization models is that they assume a constant probability of infection over space and time. In contrast, a person’s probability of infection can be highly variable depending on a range of factors, including recent infections and secondary routes of exposure. Dynamic models are capable of accounting for baseline probabilities, but they also account for the greater complexity of real world conditions by incorporating spatial and temporal variability.

Dynamic models describe the ‘life cycle’ of disease using the following stages: susceptible (S), exposed or latent (E), infectious (I), and recovered (R). Dynamic models vary in complexity depending on how the various stages are defined. For example, models may include distinct populations of susceptibles with varying probabilities of infection. This is important because people who have previously been infected by a particular pathogen may achieve partial immunity or reduced probability of infection during future exposures. The model shown below (Figure 1) includes distinct populations of susceptibles (S1 and S2), but the same probability of infection is used for both groups to simplify the model. Each of the stages is also subdivided into compliant (e.g., SC) and non-compliant (e.g., SNC) groups with corresponding rates of exposure (βC and βNC, respectively) to account for point of use devices in the community.

After exposure, people transition to the first of three ‘exposed’ or latent stages at a rate of π/3. The sub-stages act as an artificial delay to more accurately represent real-world conditions (Eisenberg et al.^[14]). In this model, the overall latent phase (E1+E2+E3) is characterized by a 14-day duration and includes 8% of the population at time zero (Mohammed Mahdy et al.^[15]). Individuals then transition to either an asymptomatic (IA) or symptomatic (IS) infectious stage at which time they start shedding pathogens. The asymptomatic infection ratio (ρ) was estimated at 65% (USEPA^[16]). Infectious individuals are also capable of spreading disease via secondary exposure (βS). This is the only situation in which the population is shared between the compliant and non-compliant models. In other words, a non-compliant, infectious individual is capable of infecting either a compliant or non-compliant susceptible. After an estimated 14 days of infection (1/δ), people transition to a ‘recovered’ stage during which they have partial immunity. The duration of partial immunity is estimated at 90 days (1/σ) before the individuals return to being susceptible (S2). The rate parameters shown in Figure 1 are described in Table 1.

Figure 1. Dynamic transmission model for giardiasis. In the model, the non-compliant and compliant populations are independent except for the exposure to infectious individuals (INC,A+INC,S+IC,A+IC,S).

Table 1. Transmission model parameters. Some parameters are listed as the inverse of those shown in Figure 1 because they represent time periods with specific significance.

A series of ordinary differential equations (shown below) were developed and solved using Matlab to characterize the dynamic nature of giardiasis in three South American communities. The model simulated the spread of disease over 365 days.

dSNC,1/dt = -beta_NC*S_NC_1 - beta_s*S_NC_1*(I_NC_A + I_NC_S + I_C_A + I_C_S) dSC,1/dt = -beta_C*S_C_1 - beta_s*S_C_1*(I_NC_A + I_NC_S + I_C_A + I_C_S) dENC,1/dt = beta_NC*S_NC_1 + beta_s*S_NC_1*(I_NC_A + I_NC_S + I_C_A + I_C_S) + beta_NC*S_NC_2 + beta_s*S_NC_2*(I_NC_A + I_NC_S + I_C_A + I_C_S) - (pi_1/3)*E_NC_1 dENC,2/dt = (pi_1/3)*E_NC_1 - (pi_1/3)*E_NC_2 dENC,3/dt = (pi_1/3)*E_NC_2 - (pi_1/3)*E_NC_3 dEC,1/dt = beta_C*S_C_1 + beta_s*S_C_1*(I_NC_A + I_NC_S + I_C_A + I_C_S) + beta_C*S_C_2 + beta_s*S_C_2*(I_NC_A + I_NC_S + I_C_A + I_C_S) - (pi_1/3)*E_C_1 dEC,2/dt = (pi_1/3)*E_C_1 - (pi_1/3)*E_C_2 dEC,3/dt = (pi_1/3)*E_C_2 - (pi_1/3)*E_C_3 dINC,A/dt = ro*(pi_1/3)*E_NC_3 - delta*I_NC_A dINC,S/dt = (1-ro)*(pi_1/3)*E_NC_3 - delta*I_NC_S dIC,A/dt = ro*(pi_1/3)*E_C_3 - delta*I_C_A dIC,S/dt = (1-ro)*(pi_1/3)*E_C_3 - delta*I_C_S dRNC/dt = delta*(I_NC_A+I_NC_S) - sigma*R_NC; dRC/dt = delta*(I_C_A+I_C_S) - sigma*R_C dSNC,2 = sigma*R_NC - beta_NC*S_NC_2 - beta_s*S_NC_2*(I_NC_A + I_NC_S + I_C_A + I_C_S) dSC,2/dt = sigma*R_C - beta_C*S_C_2 - beta_s*S_C_2*(I_NC_A + I_NC_S + I_C_A + I_C_S)

The spread of giardiasis was simulated with three different compliance ratios (0%, 80%, and 100%) for both ceramic and biosand filtration. The model was run with primary exposure only and a combination of primary and secondary exposure for the three South American communities. Figure 2 (ceramic filter) and Figure 3 (biosand filter) summarize the results from these analyses and also compare the results of the dynamic versus static models.

The static model results are relatively straightforward and illustrate a monotonic decrease in risk as the compliance rates in those communities increase. The variability between the communities can be attributed to both varying population and also varying risk of exposure due to differing Giardia cyst concentrations in the tap water. On the other hand, the dynamic model results, which account for both primary and secondary exposure, are somewhat counterintuitive because the risks actually decrease for 0% and 80% compliance. This can be explained by the fact that the risk of exposure actually decreases at certain times because individuals have transitioned to a different stage of disease and are no longer susceptible. In basic terms, there is no risk of exposure for individuals who are already exposed, infected, or recovered. This actually attenuates the annual risk. As the compliance rate increases, the static model indicates a substantial decrease in risk via primary exposure, while the secondary exposure route in the dynamic model remains dominant. The significance of secondary exposure is also clearly illustrated in Figure 4, which compares the results of the static model, the dynamic model with no secondary exposure, and the dynamic model with secondary exposure. Again, the dynamic model attenuates the risk in some low compliance scenarios, particularly when isolating primary exposure, but there is a clear increase in risk when comparing the primary and primary+secondary exposures using the dynamic model. Furthermore, the significance of the initial latent population at time zero is evident considering that the 100% compliance scenario with no secondary exposure still predicts a relatively high number of cases.

One of the other significant benefits of dynamic models is the ability to describe the temporal variability of the spread of infection through a community. Figure 5 illustrates the number of infections at any given time in Para, Brazil when assuming no compliance and a combination of primary and secondary exposure. This figure also illustrates the peak number of infections and the time required to reach the peak of the outbreak or the ultimate steady state condition. In potable reuse applications (Figure 6) or other scenarios in which people may be exposed to treated wastewater effluent—either for body contact or consumption—this information can be used to predict the loading on treatment infrastructure. Specifically, an outbreak condition will likely cause Giardia cyst concentrations (Figure 7) to exceed the levels typically observed in the raw sewage arriving at a wastewater treatment plant. Unless the wastewater treatment train, natural attenuation in the environment, and drinking water treatment train are sufficiently robust to handle this increased load, downstream communities may be at increased risk of illness and disease.

Figure 2. Risk of giardiasis with ceramic filter intervention (with secondary exposure for dynamic data)

Figure 3. Risk of giardiasis with biosand filter intervention (with secondary exposure for dynamic data)

Figure 4. Comparison of the significance of primary and secondary exposure in the static and dynamic models for Caracas, Venezuela

Figure 5. Temporal variability of infections in Para, Brazil (0% compliance and primary+secondary exposure)

Figure 6. de facto potable reuse

Figure 7. Temporal variability in raw wastewater Giardia concentration in Para, Brazil (0% compliance and primary+secondary exposure)

Therefore, the output from dynamic transmission models can be used to identify a more appropriate estimate of risk that accounts for spatial and temporal variability, secondary routes of exposure, and endemic disease in the community. Dynamic transmission models can also be used to characterize the release of pathogens over time, which, in the case of de facto reuse, can ultimately feed into subsequent risk assessments for downstream communities.

Risk Management

Comparison of disease burden by location for each scenario Higher levels of compliance generally produce lower levels Secondary transmission caused the disease burden to increase by a factor of 5 - 10

Exposure doses corresponding to the modeled risk calculations were determined Comparison of these values were made with water concentrations producing an acceptable risk The difference of these values can be used to suggest water treatment improvement

Water treatment cost

Using costs associated with water treatment improvement, the cost-effectiveness of each option can be calculated and compared.

Point of Use Devices are insufficient to meet acceptable risk when secondary transmission is included Options to meet an acceptable level of risk can be identified and compared In addition to water treatment, secondary transmission must be controlled to decrease its significant contribution to the risk A number of stakeholders were identified for targeted risk commination

Risk communication (Stakeholders)