Balancing the Risks and Benefits of Drinking Water Disinfection: Disability Adjusted Life-Years on the Scale.

To evaluate the applicability of disability adjusted life-years (DALYs) as a measure to compare positive and negative health effects of drinking water disinfection, we conducted a case study involving a hypothetical drinking water supply from surface water. This drinking water supply is typical in The Netherlands. We compared the reduction of the risk of infection with Cryptosporidium parvum by ozonation of water to the concomitant increase in risk of renal cell cancer arising from the production of bromate. We applied clinical, epidemiologic, and toxicologic data on morbidity and mortality to calculate the net health benefit in DALYs. We estimated the median risk of infection with C. parvum as [10.sup.-3]/person-year. Ozonation reduces the median risk in the baseline approximately 7-fold, but bromate is produced in a concentration above current guideline levels. However, the health benefits of preventing gastroenteritis in the general population and premature death in patients with acquired immunodeficiency syndrome outweigh health losses by premature death from renal cell cancer by a factor of [is greater than] 10. The net benefit is approximately 1 DALY/million person-years. The application of DALYs in principle allows us to more explicitly compare the public health risks and benefits of different management options. In practice, the application of DALYs may be hampered by the substantial degree of uncertainty, as is typical for risk assessment. Key words:, bromate, Cryptosporidium parvum, disinfection, drinking water, ozone, quality of life, risk assessment. Environ Health Perspect 108:315-321 (2000). [Online 21 February. 2000]

http://ehpnet1.niehs.nih.gov/docs/2000/108p315-321havelaar/ abstract.html

The microbiologic safety of drinking water is of paramount importance to public health (1). Source protection is generally accepted as the primary strategy to obtain microbiologically safe drinking water. However, many sources (surface waters in particular) are highly polluted and need extensive treatment before distribution to the consumer. Chemical disinfection is prominent in these treatment schemes. Highly oxidizing chemicals such as chlorine and ozone kill a variety of pathogenic microorganisms during treatment, and chlorine is applied in many countries as an additional safeguard in the distribution system. An important drawback of the use of these chemicals is the generation of disinfection by-products, which have suspected adverse effects on human health. This situation calls for a formal quantitative framework to compare the positive and negative health effects of drinking water disinfection and to assist in the design and operation of treatment plants (2).

Microbiologic risks are expressed as the annual individual probability of infection for a given consumption of drinking water. Chemical risks related to genotoxic carcinogens are usually associated with an increase in cancer incidence attributable to a lifetime exposure. The public health impact of these disease end points is very different and cannot be compared directly. Hence, decision making is difficult. We evaluated the applicability of a single measure of attributable disease burden to compare and balance health risks and benefits on a public health basis. The disability adjusted life-year (DALY) consists of the loss of healthy life years due to either premature mortality or morbidity. The DALY integrates several dimensions of the public health impact, such as the number of affected persons and the severity and duration of adverse health effects, and uses time as a unit of measurement. We illustrate our approach by a case study of ozonation of a hypothetical drinking water supply using surface water. We used Monte Carlo simulation to assess the uncertainty in our risk estimates, to account for biologic and seasonal variation in exposure and effects, and to accommodate uncertainty in parameter values. In accordance with the framework proposed by the U.S. National Research Council (3), we discuss the four consecutive elements of the risk assessment process.

Hazard Identification

We considered a hypothetical drinking water supply that is typical for drinking water in The Netherlands. In this supply, raw water is abstracted from an international river that has been polluted by domestic and industrial sources and by agricultural run-off. The water is treated by storage in three consecutive reservoirs, coagulation, sedimentation, rapid sand filtration, ozonation (only when disinfected), granular activated carbon filtration, and ultraviolet (UV) postdisinfection. The water is then distributed to the consumer without any residual disinfectant in the distribution system. Drinking water may harbor many pathogenic microorganisms. In this paper, we considered only one pathogen and one of the many potential disinfection by-products that are associated with ozonation. We illustrated the methodology but did not make a comprehensive assessment of a particular situation. We selected a microbial and chemical agent of major public health significance in surface water supplies. Other agents occur in lower concentrations and/or induce smaller health effects. Hence, this restriction does not largely affect the validity of the final conclusions.

Cryptosporidium parvum is a protozoan parasite that produces environmentally resistant and highly infectious oocysts. Other pathogenic microorganisms, such as viruses and Campylobacter jejuni, may also be present in similar concentrations in water from storage reservoirs. However, these organisms are inactivated by postdisinfection processes such as UV irradiation in our scenario and hence cause smaller public health problems. Adequate control of C. parvum is therefore of critical importance in most surface-water supplies. Infection with C. parvum may result in self-limiting gastroenteritis in immunocompetent persons. In those who are immunocompromised, the infection is not easily cleared and usually results in severe life-threatening gastroenteritis. In the immunocompromised group, we only considered acquired immunodeficiency syndrome (AIDS) patients because a severe clinical course of cryptosporidiosis is associated with CD4 counts [is less than or equal to] 53 cells/mL (4), which is typical only in the advanced stages of AIDS.

C. parvum has caused major outbreaks of waterborne disease in Europe and in North America (5,6). C. parvum oocysts are resistant to chlorination but may be inactivated by ozonation, which is increasingly used as an alternative disinfectant. However, ozone reacts with bromide ions to produce bromate (7). Bromate induces tumors in the rat kidney, thyroid, and mesothelium (8,9) and is a renal carcinogen in the mouse (9). In accordance with the World Health Organization (WHO) (1), we concentrated on renal cell cancer as an outcome of chronic exposure to bromate. Bromate is considered the most important by-product of ozonation (10). Other by-products from ozonation may include aldehydes, bromoform, and brominated acetic acids, none of which are classified as genotoxic carcinogens. Pilot plant experiments (11) have demonstrated that at usual ozone doses only formaldehyde is produced in measurable quantities, although at levels far below the WHO guideline of 900 [micro]g/L (1).

Exposure Assessment

The model used for assessment of exposure to C. parvum and bromate is described in Table 1. Table 1 also shows Equations 1-9. Not all parameter values are available for one single drinking water production plant; therefore, we used data from several plants to represent a hypothetical but typical situation in The Netherlands. When necessary, we used additional data from the international literature. Table 2 presents a point estimate for each parameter as well as the probability distributions used in the uncertainty analysis.

Table 1. Model equations for assessment of the incidence of infection by C. parvum and renal cell cancer by bromate.

Description (unit)            Equation

Conc of C. parvum oocysts     [MATHEMATICAL EXPRESSION
in drinking water (per L)     NOT REPRODUCIBLE IN ASCII]

Ozone Conc after first        [MATHEMATICAL EXPRESSION
contact chamber (mg/L)        NOT REPRODUCIBLE IN ASCII]

Ozone residual (mg/L)         [MATHEMATICAL EXPRESSION
                              NOT REPRODUCIBLE IN ASCII]

Effective ozone Conc (mg/L)   [MATHEMATICAL EXPRESSION
                              NOT REPRODUCIBLE IN ASCII]

Disinfection of C. parvum
oocysts

Without ozonation             [MATHEMATICAL EXPRESSION
                              NOT REPRODUCIBLE IN ASCII]

With ozonation                [MATHEMATICAL EXPRESSION
                              NOT REPRODUCIBLE IN ASCII]

Daily risk of infection       [MATHEMATICAL EXPRESSION
with C. parvum                NOT REPRODUCIBLE IN ASCII]

Number of persons with one    [N.sub.inf,Cp] = N x
or more infections per year   [1 - [(1 - [P.sub.inf,Cp]).sup.365]]

Conc of bromate in drinking   [MATHEMATICAL EXPRESSION
water ([micro]g/L)            NOT REPRODUCIBLE IN ASCII]

Incidence of renal cell       [MATHEMATICAL EXPRESSION
cancer per year               NOT REPRODUCIBLE IN ASCII]

                              Equation
Description (unit)              no.

Conc of C. parvum oocysts        1
in drinking water (per L)

Ozone Conc after first           2
contact chamber (mg/L)

Ozone residual (mg/L)            3

Effective ozone Conc (mg/L)      4

Disinfection of C. parvum
oocysts

Without ozonation                5A

With ozonation                   5B

Daily risk of infection          6
with C. parvum

Number of persons with one       7
or more infections per year

Conc of bromate in drinking      8
water ([micro]g/L)

Incidence of renal cell          9
cancer per year

Conc, concentration.

Table 2. Point estimates and probability distributions of model parameters for assessment of the incidence of infection by C. parvum and renal cell cancer by bromate.

Description (unit)                  Symbol

Conc of C. parvum in raw            [C.sub.Cp,RAW]
water (per L)

Decimal reduction of C. parvum by   [DR.sub.Cp,CSF]
coagulation-sedimentation-rapid
sand filtration

Ozone dose (mg/L)                   [MATHEMATICAL EXPRESSION NOT
                                    REPRODUCIBLE IN ASCII]

Ozone demand parameter              [[Beta].sub.0]
(intercept)(mg/L)

Ozone demand parameter (slope)      [[Beta].sub.1]

Contact time (min)                  T

Ozone decay parameter (per min)     d

Hom intercept parameter             k
[[(L/mg).sup.m]/[min.sup-n]]

Hom parameter for concentration     m

Hom parameter for contact time      n

Daily consumption of uncooked       [V.sub.u]
drinking water (L/person/day)

Dose-response parameter for         r
C. parvum

Population served                   N

Dissolved organic carbon (mg/L)     DOC

pH value                            pH

Conc of bromide in raw water        [MATHEMATICAL EXPRESSION NOT
([micro]g/L)                        REPRODUCIBLE IN ASCII]

Daily consumption of total          [V.sub.t]
drinking water (L/person/day)

Renal cell cancer risk parameter    q
([kg.sup.2/3]/[micro]g)

Body weight (kg)                    BW

                                        Median
Description (unit)                      value

Conc of C. parvum in raw            0.044
water (per L)

Decimal reduction of C. parvum by   1.6
coagulation-sedimentation-rapid
sand filtration

Ozone dose (mg/L)                   3.1

Ozone demand parameter              -0.194
(intercept)(mg/L)

Ozone demand parameter (slope)      0.443

Contact time (min)                  5

Ozone decay parameter (per min)     0.257

Hom intercept parameter             0.219}
[[(L/mg).sup.m]/[min.sup-n]]

Hom parameter for concentration     0.899}

Hom parameter for contact time      1.059}

Daily consumption of uncooked       0.16
drinking water (L/person/day)

Dose-response parameter for         0.0037
C. parvum

Population served                   [10.sup.6]

Dissolved organic carbon (mg/L)     3.5

pH value                            7.5

Conc of bromide in raw water        86
([micro]g/L)

Daily consumption of total          0.81
drinking water (L/person/day)

Renal cell cancer risk parameter    3.0 x [10.sup.-6]
([kg.sup.2/3]/[micro]g)

Body weight (kg)                    74.5

Description (unit)                  Probability distribution(a)

Conc of C. parvum in raw            Lognormal (-3.12, 0.97)
water (per L)

Decimal reduction of C. parvum by   Normal (1.6, 0.4)
coagulation-sedimentation-rapid
sand filtration

Ozone dose (mg/L)                   Fixed

Ozone demand parameter              Normal (pred, 0.091)
(intercept)(mg/L)

Ozone demand parameter (slope)      Normal (pred, 0.091)

Contact time (min)                  Inverse Gaussian (5, 75)

Ozone decay parameter (per min)     Normal (pred, 0.239)

Hom intercept parameter             log (DR) = normal
[[(L/mg).sup.m/[min.sup-n]]         (pred, 0.134)

Hom parameter for concentration

Hom parameter for contact time

Daily consumption of uncooked       Lognormal (-1.80, 0.99)
drinking water (L/person/day)

Dose-response parameter for         Lognormal (-5.61, 0.43)
C. parvum

Population served                   Fixed

Dissolved organic carbon (mg/L)     Lognormal (1.25, 0.03)

pH value                            Fixed

Conc of bromide in raw water        Lognormal (4.46, 0.20)
([micro]g/L)

Daily consumption of total          Lognormal (-0.21, 0.82)
drinking water (L/person/day)

Renal cell cancer risk parameter    Custom (bootstrap)
([kg.sup.2/3]/[micro]g)

Body weight (kg)                    Men: lognormal (4.39, 0.15)
                                    Women: lognormal (4.21, 0.16)

Abbreviations: Conc, concentration; Pred, predicted value.

(a) Normal (mean, SD); lognormal (mean, SD on [log.sub.e] scale); inverse Gaussian (mean, shape).

C. parvum. We estimated the concentration of C. parvum oocysts in drinking water from their concentration in raw water and their reduction by treatment. The concentration of C. parvum oocysts, measured at the outlet of a series of storage reservoirs with a mean retention time of 5 months and corrected for recovery of the detection method and for viability (12) can be described by a lognormal distribution. The removal of C. parvum oocysts by coagulation-sedimentation-rapid sand filtration is estimated by the removal of spores of sulfite-reducing clostridia under full scale conditions (13). We used Equation 1 to estimate the median concentration of C. parvum oocysts in drinking water without ozonation as 1.1 x [10.sup.-3]/L.

We used a two-step approach to estimate the effective ozone concentration in water (14). Ozone was added at a constant dose of 3.1 mg/L to the first of a series of four contact chambers, then partially consumed by initial rapid oxidation reactions (ozone demand). Parameter values in the ozone demand model and the subsequent exponential decay model were based on linear regression analysis of pilot-plant data (14), assuming a pH of 7.5 and a temperature of 10-12 [degrees] C. The SD of the residuals in the regression equation quantified the uncertainty in the predicted values (15). We modeled the variability in the residence time distribution in the ozone contact chambers by the inverse Gaussian distribution (16). The Hom formula (Equation 5B) modeled inactivation of C. parvum oocysts by ozone, using the geometric mean as an estimate of the effective ozone concentration (17). The parameters were based on linear regression analysis using the log-transformed Hom model, and were applied to data measured at approximately 10 [degrees] C (17), again taking the SD of the residuals to quantify uncertainty. Ozonation reduced the median concentration of viable oocysts in the baseline approximately 7-fold to a median concentration of 1.6 x [10.sup.-4]/L.

Bromate. Bromate is only occasionally present in raw water in The Netherlands (18); therefore, all bromate in drinking water is presumably formed during the ozonation process. Dissolved organic carbon and bromide concentrations in raw water are based on measurements at an actual plant in The Netherlands, taking annual variation into account as an uncertainty factor. We used the Song et al. (19) model to estimate the formation of bromate as a function of the ozone dose and several water quality parameters. We calculated the median bromate concentration in our scenario as 3.8 [micro]g/L (Equation 8 in Table 1).

Finally, we calculated the daily ingested dose of C. parvum oocysts and bromate ions by multiplication of their concentration in drinking water with the daily consumption [V.sub.u] of uncooked drinking water (12) or [V.sub.t] of total drinking water (20), respectively. The [V.sub.t] commonly used default value of 2 L/person/day is close to the 95th percentile of the distribution in Table 2.

Hazard Characterization

C. parvum. We used the exponential dose-response model (21) to estimate the risk of infection related to the daily ingested dose of C. parvum and to estimate the dose-response parameter, r, (22) from volunteer challenge data (23). A sample of 1,000 simulations of r, obtained by bootstrapping, is described by a lognormal distribution. We assumed a population of 1 million inhabitants in the distribution area who randomly consume drinking water. The daily risk associated with exposure to C. parvum is accumulated to estimate the annual risk of infection (Equation 7).

To assess the uncertainty in the annual risk of infection, we assumed that all model parameters were independent. Five thousand replicates of the concentration of C. parvum oocysts in drinking water without ozonation ([C.sub.Cp,DW]) were drawn from the distributions in EqUations 1 and 5A. Then we calculated corresponding estimates of the concentration of C. parvum in drinking water with ozonation ([MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]) by multiplying each value of [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] with a random value for the reduction factor by ozonation (Equation 5B). Samples from the distributions of [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] and [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] were then combined with randomly drawn samples of r and [V.sub.u], using identical indices for both series (with and without ozone). Equation 6 allows us to compare their values and assess the influence of ozonation on the risk of infection of any one individual on any 1 day. We estimated the annual individual risk of infection by taking a sample of 365 values of the daily risk using Equation 7. We repeated this 5,000 times to stimulate the distribution of the annual risk of infection. We then computed the annual incidence of infection as the annual risk multiplied by the size of the population at risk.

The annual incidence of infection without ozonation appears to be approximately lognormally distributed with a median of 1,000 cases/million person-years (mpy) and spans a relatively small range. The 95% confidence interval (CI) for annual incidence of infection is 760-1,500 cases/mpy (Figure 1A). The introduction of ozonation leads to a reduced incidence (median 180 cases/mpy; CI, 120-260 cases/mpy).

[Figure 1 ILLUSTRATION OMITTED]

The median values of the incidence of infection cannot be exactly reproduced by the simple substitution of median parameter values in the equations in Table 1. The additive nature of Equation 7 implies that the annual risk is mainly determined by the few occasions on which a high dose is encountered.

Bromate. Exposure to bromate in drinking water is mainly by ingestion; other routes (inhalation and dermal exposure) are: not described in the literature. To estimate the excess lifetime risk of renal cell cancer from exposure to bromate, the WHO (1) used data by Kurokawa et al. (24) and fitted the linearized multistage model (LMS) to extrapolate from high to low doses (25). Applying this procedure while dividing the lifetime risk by an average life expectancy of 80 years results in an incidence of 0.16 cases of renal cell cancer/mpy at the calculated exposure level. For our comparison of infection and cancer risks associated with the ozonation of drinking water, the estimated uncertainty in that risk is needed. Therefore, we reanalyzed the original data to obtain the maximum likelihood estimate of excess risk (26). We added other data by Kurokawa et al. (27) and recently published data (9) (Figure 2). The three data sets were fitted by the LMS model, in which one of the parameters, b, which denoted a dose factor, was allowed to vary between the three studies. We found that the LMS model did not give a significantly better fit than the two-stage model:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

where [F.sub.i] = probability of renal cell cancer in study i, dose = dose [mg KBr[O.sub.3]/[(kg body weight).sup.2/3]], a (background response) and c (shape) = common parameters for all datasets (for this data set a = 0.019 and c = 0.13), and [b.sub.i] = dose-response parameter fitted to dataset i ([b.sub.1] = 0.0013, [b.sub.2] = 0.0028, and [b.sub.3] = 0.0091 [kg.sup.2/3]/mg). For each study we calculated the corresponding dose associated with a 10% response ([ED.sub.10]). The slopes of the three straight lines from the origin to the [ED.sub.10] are point estimates for the three conservative dose-response parameters of the individual studies ([q.sub.1] = 0.0086, [q.sub.2] = 0.0026, and [q.sub.3] = 0.0012 [kg.sup.2/3]/mg). The geometric mean of these three values is the point estimate for the pooled dose-response parameter ([q.sup.3] = 0.0030 [kg.sup.2/3]/mg). We estimated the uncertainty in q by bootstrapping (28), which resulted in a trimodal uncertainty distribution.

[Figure 2 ILLUSTRATION OMITTED]

We obtained a Monte Carlo sample for the distribution of renal cell cancer incidence by substituting 3,000 replicates from all of the underlying distributions in Equation 9. Using this model, we estimated that the median incidence of renal cell cancer was 0.009 cases/mpy, 18 times lower than the WHO estimate, as based on the LMS model. This difference is largely (a factor of 10) because we used the maximum likelihood estimate instead of the 95th percentile of the dose-response parameter and also because we used a probability distribution for the consumption of drinking water instead of a conservative point estimate (accounting for a factor of 1.8). Using three datasets instead of one has little effect on the median risk. The renal cell cancer incidence CI is relatively wide (0.001-0.075) (Figure 1B), mainly because the three experimental dose-response data sets show large differences.

Risk Characterization

General. Disease may result in the reduction of survival time (quantity of life), the reduction of the quality of life, or both. We combined these outcomes in a single measure, the DALY, as originally proposed by Murray in the global burden of disease (GBD) study (29). The DALY concept is designed as a generic instrument that produces results on an interval scale so that meaningful comparisons between widely different disease end points can be made. The loss of healthy life-years in a population, measured in DALYs, is calculated as:

DALY= LYL + YLD,

where LYL is the number of life-years lost due to mortality and YLD is the number of years lived with a disability, weighed with a factor between 0 and 1 for the severity of the disability. LYL is calculated by accumulation over all relevant diseases (i) of the product of d (the number of deaths due to a particular disease) and e* (the standard life expectancy at the age of death due to that disease). YLD is calculated as the accumulated product of N (the number of persons affected by a nonlethal disease), L (the duration of this disease), and W (a measure for its severity). If necessary, disease processes are subdivided into several stages with different duration and severity. Murray (29) and Van der Maas and Kramers (30) provide further details. Thus, for a particular agent, the population health burden is calculated as:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII].

C. parvum. Table 3 presents the distributions used for estimation of the health burden of infection with C. parvum. Infection with C. parvum leads to gastroenteritis in 71% of immunocompetent persons (31). The duration of cryptosporidiosis is usually reported as 1-2 weeks, but these estimates are based on cases detected in laboratory surveillance and may be biased toward longer duration. In population-based outbreak studies (6,32) and in volunteer experiments (23,31), the mean duration of gastroenteritis is reported to be only 3-6 days. Both types of study (31,33) report relapses of diarrhea in 40-70% of patients. To quantify the duration of cryptosporidiosis we used a lognormal distribution with a median duration of 6 days and a range between 2 and 30 days.

Table 3. Point estimates and probability distributions of model parameters for assessment of the health burden of gastroenteritis by infection with C. parvum.

                          Immunocompetent population

                       Median            Probability
Parameter              value(b)          distribution(c)

P | infection          0.71              Normal (0.71, 0.08)
Severity(W)            0.054             Lognormal (-2.92, 0.63)
Duration (L)           0.016             Lognormal (-4.14, 0.67)
Case fatality ratio    0.06 per 10,000   Lognormal (-2.80, 1.00)
Life expectancy (e*)   8.2               Custom(d) (13.2, 15.0)

                         Population of AIDS patients(a)

                       Median     Probability
Parameter              value(b)   distribution(c)

P | infection            1.00     Fixed
Severity(W)              0.16     Lognormal (-1.81, 0.63)
Duration (L)             0.13     Lognorma] (-2.03, 0.92)
Case fatality ratio      0.70     Normal (0.70, 0.08)
Life expectancy (e*)     1.0      Lognormal (0.00, 0.83)

(a) Prevalence 4:10,000.

(b) Duration and life expectancy have unit year; all other parameters are dimensionless.

(c) Normal (mean, SD); lognormal (mean, SD on [log.sub.e] scale).

(d) See Figure 3A.

The disability weight for cryptosporidiosis is taken from the GBD study (29). In this study, a severity weight for watery diarrhea is derived using the person trade-off protocol. Panel members trade off life-years of healthy persons for healthy years gained by people in a certain state of reduced health until a point of indifference is reached. The person trade-off protocol estimates social values of different health states, and thus is considered appropriate for the evaluation of health intervention programs (34).

The mean mortality risk for cryptosporidiosis in the immunocompetent population is estimated as 1/100,000, based on experience from the 1993 outbreak in Milwaukee, Wisconsin (35). Four deaths in the nonimmunocompromised population were attributed to the outbreak, which involved approximately 400,000 persons (35). The large uncertainty in this estimate is accounted for by choosing a lognormal distribution with a high but arbitrary dispersion factor (log-SD = 1). This distribution has a median of 0.6, a mean of 1, and 95% of the distribution ranges between 0.1 and 4 per 100,000. To estimate the number of life-years lost by a fatal case of gastroenteritis, we used the age distribution for all gastroenteritis deaths in The Netherlands in 1993-1995 and computed the standard life expectancy in 5-year intervals (29). Figure 3A shows that most deaths associated with gastroenteritis occur in people older than 75 years of age.

[Figure 3 ILLUSTRATION OMITTED]

In immunocompromised persons, particularly in AIDS patients, infection with C parvum leads to gastroenteritis in virtually all cases (38). McGowan et al. (39) reported that only 30% of AIDS patients have a remission; the others suffer from cryptosporidiosis until death. We fit a lognormal model to the reported duration of disease in all patients. There is no formally derived severity weight for cryptosporidiosis in AIDS patients. Because enteritis is more severe in AIDS patients than in immunocompetent persons, leading to significant weight loss and abdominal pain, a larger weight is justified. The median survival times for patients with and without cryptosporidiosis are 11.5 and 66 weeks, respectively. Hence, active cryptosporidiosis reduces the life expectancy of 70% of AIDS patients by approximately 1 year (median value). The uncertainty in this value depends mainly on the survival of the remission group, which is poorly known because most patients were still alive when the study ended. With this restriction, we fit a lognormal distribution to the reported survival times. The average prevalence of AIDS is 0.4/1,000 as for the city of Amsterdam (40).

Drawing 3,000 replications using the probability density functions in Table 3 simulates the probability distribution of the health burden of gastroenteritis resulting from infection by C. parvum in drinking water (Figure 4A). The distribution is highly skewed to the right, with a median value of approximately 1.1 healthy life-years/1,000 cases of infection and a small but important probability of a loss of up to five healthy life-years/1,000 cases of infection.

[Figure 4 ILLUSTRATION OMITTED]

Bromate. The survival of patients with renal cell cancer depends primarily on age, fitness for resection, postoperative survival, and the presence of metastases (Table 4) (41,42). Patients with one or more unfavorable prognostic (actors have a short survival. Those patients who survive for 5 years after diagnosis have a normal life expectancy. To estimate the number of life-years lost from a case of renal cell cancer, we combined these prognostic factors with the age distribution at diagnosis (Figure 3B) and calculated standard life expectancies. The morbidity burden due to renal cell cancer is small as compared to the mortality burden because death usually occurs within a few months after diagnosis of a fatal case, and because the quality of life after a successful operation is not negatively influenced. Disability weights for the clinical phase of unrelated cancers in the Dutch Public Health Status and Forecast Report (32) are used to approximate the severity of renal cell cancer during the first year after diagnosis to account for the effects of surgery, therapy, and emotional stress on the quality of life of survivors. The health burden/case of renal cell cancer is simulated by 3,000 samples from the probability distributions in Table 4. The median health burden/case of renal cell cancer is 10 DALYs, with a relatively symmetrical distribution (Figure 4B).

Table 4. Point estimates and probability distributions of model parameters for assessment of the health burden of renal cell cancer.

                                Age class

Parameter              < 70 years      [is greater than or
                                         equal to] 70 years
Percent in class       62 (50-65)(a)   38 (35-50)(a)
No resection           18 (10-20)      37 (25-40)
Postoperative death     4 (3-5)         4 (3-5)
Metastases             19 (15-30)      14 (10-20)
Death within 5 years   25 (20-30)      48 (40-60)
  (no metastases)
Life expectancy        24.7 (7.4)(b)   10.5 (2.9)(b)

Values in percent except where noted.

(a) All triangular distributions: most likely value (minimum, maximum).

(b) Discrete distribution; mean (SD), in years. derived from Figure 3B.

Integration. Table 5 presents point estimates of the health impacts of supplying drinking water with or without ozonation. The median incidence of infection with C parvum of 1,000 cases/mpy leads to 710 cases of gastroenteritis in the immunocompetent population and an additional 0.40 cases of cryptosporidiosis in AIDS patients. The mortality in the immunocompetent population is low; the AIDS population had 0.29 premature deaths/mpy. Multiplication of these incidence estimates with disability weights and the duration of the health states results in an estimate of the health loss in the scenario without ozonation. Disability by acute gastroenteritis is the most important source of health loss for this pathogen (0.61 DALYs), whereas AIDS patients add 0.29 DALYs, mostly due to premature death. By summation of all effects, we estimated a health loss by infection of 0.93 DALYs/mpy in the scenario without ozonation. The introduction of an ozonation step in the treatment chain does not completely eliminate health risks, although it reduces the median health loss by gastroenteritis to 0.17 DALYs/mpy. Hence, the median health benefit of ozonation is 0.76 DALYs/mpy. The production of bromate leads to 0.01 additional cases of renal cell cancer/year. The health loss of morbidity from renal cell cancer is negligible; premature death adds 10 life-years lost/fatal case or 0.06 DALYs/mpy, which is [is greater than] 10-fold lower than the health benefits by reducing the risk of infection. The median net health benefit in this scenario is estimated as 0.70 DALYs/mpy.

Table 5. Balancing the risks of drinking water disinfection (point estimates based on median values).

                                            Morbidity

Hazards                         N       L          W        YLD

No ozone
  Immunocompetent persons      710    0.016      0.054      0.61
  AIDS patients                0.40   0.16       0.16       0.01
With ozone
  Immunocompetent persons      130    0.016      0.054      0.11
  AIDS patients                0.07   0.13       0.16       0.00
  Bromate, renal cell cancer   0.01   See text   See text   0.00
Health benefit(a)
  Immunocompetent persons       -        -          -       0.50
  AIDS patients                 -        -          -       0.01
  Bromate, renal cell cancer    -        -          -       0.00
Total                           -        -          -       0.51

                                       Mortality

Hazards                        d       e*    LYL     DALY

No ozone
  Immunocompetent persons      0.004   8.2    0.03    0.64
  AIDS patients                0.28    1.0    0.28    0.29
With ozone
  Immunocompetent persons      0.001   8.2    0.01    0.12
  AIDS patients                0.05    1.0    0.05    0.05
  Bromate, renal cell cancer   0.006   10     0.06    0.06
Health benefit(a)
  Immunocompetent persons        -      -     0.02    0.52
  AIDS patients                  -      -     0.23    0.24
  Bromate, renal cell cancer     -      -    -0.06   -0.06
Total                            -      -     0.19    0.70

Abbreviations: d, number of deaths (per mpy); e*, standard life expectancy at the age of death; L, duration (years); LYL, life years lost; N, number affected (per 1 mpy); W, severity weight; YLD, years lived with disability.

(a) No ozone minus with ozone.

Uncertainty and sensitivity analysis. Figure 5 shows the uncertainty in the estimated health effects. The population health loss by exposure to C. parvum is simulated by multiplying the matched pairs of incidence of infection (with and without ozonation) with samples from the health burden/case. The median without ozonation is 1.2 DALYs/mpy (CI, 0.36-4.6) and is reduced a factor of approximately 6 (equal to the reduction of oocyst concentration) to a median of 0.20 DALYs/mpy (CI, 0.06-1.1) by ozonation. These values are again different from the point estimates by simple substitution of median values as given in Table 5.

[Figure 5 ILLUSTRATION OMITTED]

Multiplication of the incidence distribution with the health burden distribution of renal cell cancer results in the estimated health loss in the population by exposure to bromate in the case of ozonation. The median health loss has a value of 0.10 DALYs/mpy (CI, 0.01-1.1).

Parallel to these simulations, we calculated the net health benefit of ozonation as the health loss by gastroenteritis (in the case without ozonation) minus the sum of the residual health loss by gastroenteritis and the health loss of renal cell cancer (in the case with ozonation). We estimated the net median health benefit by the probabilistic model as 0.82 DALYs/mpy; the benefit is mainly determined by a reduction of the risk of infectious disease. Figure 6 shows the uncertainty distribution of the net health benefit, which is skewed to the right; the CI ranges between -0.22 and 3.7. There is a 5% chance that the net effect is negative and a 50% chance that it is [is greater than] 0.82 DALYs/mpy.

[Figure 6 ILLUSTRATION OMITTED]

Univariate sensitivity analysis (data not shown) demonstrates that the net health benefit is most sensitive to (in order of importance) decimal reduction of C. parvum by coagulation/sedimentation/filtration (CSF); dose-response parameter of C. parvum; consumption of unboiled drinking water; concentration of C. parvum in raw water ([C.sub.Cp,RAW]); pH; and probability, duration, and severity of enteritis in immunocompetent individuals. We varied these parameters by wide but plausible ranges to evaluate the robustness of the overall conclusions and the magnitude of the net health benefit. In comparison to the baseline scenario, only some extreme parameter values resulted in the conclusion that ozonation leads to a net health loss in the population: high efficiency of CSF (decimal reduction of C. parvum [is greater than or equal to] 2.4) and low severity weight for gastroenteritis in immunocompetent individuals ([is less than or equal to] 0.005) in combination with a low prevalence of AIDS ([is less than or equal to] 0.5/10,000). Other variations in parameter values result in increased estimates of health gain. Increasing [C.sub.Cp,RAW] or reducing the efficiency of CSF result in health benefits up to 8-12 DALYs/mpy (Figure 7 A, B). The dose-response parameter r for C. parvum is based on experimental data with healthy adult volunteers. Other subgroups of the population (infants and the elderly) might be more susceptible to infection. A 10-fold increase in r results in an approximately 10-fold increase in the net health benefit to approximately 8 DALYs/mpy. Similarly, increasing the unboiled water consumption to 1 L/day increases the net health benefit to 3 DALYs/1 mpy. The results are less sensitive to the total water consumption; to reduce the net health gain to 0 would require per capita consumption of 5 L/day. The prevalence of AIDS was also of interest in our study. In our scenario, the health loss in AIDS patients was smaller than in the immunocompetent population. If the AIDS prevalence is higher, e.g., 4/1,000 [which is typical of larger U.S. metropolitan areas (38)], the health loss in AIDS patients is much larger, and results in a larger net benefit of 3 DALYs/mpy.

[Figure 7 ILLUSTRATION OMITTED]

Discussion

This study evaluates the feasibility of integrated measures of public health, such as the DALY, as a means to assist in the complex decision-making problem of risk balancing in drinking water disinfection. A major advantage of the proposed method is that it leads to a logical, transparent, and comprehensive evaluation of health gains and losses in terms of established public health concepts (quality and quantity of life and social magnitude) using time as a unit of measurement. The DALY measure is superior to comparisons based on annual mortality rates because it also includes nonlethal end points and explicitly addresses life and health expectancy. Information from microbiology, toxicology, epidemiology, environmental, and the clinical sciences is integrated in one single estimate that accounts for the probabilities of different end points occurring. A similar measure (the quality adjusted life-year) is extensively used in economic assessment of medical technologies and in clinical decision making. The explicit introduction of values and preferences when attributing weights to different diseases adds a new and possibly controversial dimension. Because highly prevalent health responses of limited severity were the most important variables for the final result, this aspect deserves detailed attention and communication with both decision makers and the public. Sensitivity analysis demonstrates that variation of the severity weight of gastroenteritis in the immunocompetent population does not affect the conclusion that the health benefits of ozonation outweigh the health losses, but the magnitude of the net effect varies considerably.

DALYs specifically address the health dimension of a risk management problem. They do not capture social and cultural values that are important determinants of risk perception in the public (e.g., equity, voluntariness, and dread) or economic and technical factors (e.g., costs, benefits, and feasibility). Such factors will also be taken into account when taking risk decisions.

Many potential flaws of the DALY approach are not unique for this method, but bear upon health impact assessments in general, given the present lack of knowledge and availability of data. Important but almost inevitable shortcomings are the imprecision of population exposure assessments, the unknown and possibly unknowable shape of the dose-response curves at low environmental levels of exposure, and the translation of this information from rodents to humans as well as within the human population. Another important issue is the internal and external validity of epidemiologic results.

The limited availability of data leads to extensive use of simple models to arrive at final estimates. Many assumptions are made both qualitatively (which models to use) and quantitatively (the choice of distributions and parameter values). The uncertainty in the microbiologic health risk assessment relates predominantly to the precision of model parameters and to a lesser extent to the models themselves. Because relevant health risks occur at concentrations of microorganisms that are far below the limit of detection of current analytical methods, validation of model predictions by comparison with measured data is not possible. On the other hand, the uncertainties of the chemical risks pertain to the level of construct validity. Is carcinogenesis in laboratory animals exposed to high doses of bromate really predictive of effects in humans who are exposed to concentrations many orders of magnitude lower and, if so, what is the actual risk? The difference in this study between the conventional conservative approach designed to derive safe maximum allowable concentrations, and the probabilistic approach designed to estimate the probability distribution of the response parameter for renal cell cancer, illustrates one of the consequences of this uncertainty. Not only does the risk of renal cell cancer appear to be reduced by a factor of 18, but the final conclusion is also strongly affected by the choice of the extrapolation method. None of the models in this study can be rigorously validated, but we believe that they reflect the best available knowledge. The DALY methodology adds a public health dimension to the process of risk assessment that may make the outcomes more useful for risk managers. Also, our method does not impose any constraints on allowing the incorporation of future information.

In our standard scenario, the estimated health benefits by pathogen reduction appear to outweigh the health losses by bromate formation. The absolute level of the net health benefit is approximately 1 DALY/1 ropy, which is a relatively small fraction of the total healthy life expectancy. For comparison, we refer to the commonly accepted [10.sup.-6] lifetime cancer risk by exposure to genotoxic carcinogens. Using renal cell cancer data as an example, this risk equals a health loss of [10.sup.-6] x 10 DALYs/80 life-years = one-eighth DALY/mpy. The health benefit in our case study is larger than this value; therefore, ozonation is indicated. The cost-effectiveness of achieving this benefit has not yet been evaluated.

Several factors may influence the size of the final health benefit. The microbiologic risks are most sensitive to the concentration of oocysts of C. parvum in the raw water. Many drinking water supplies in the world are challenged with considerably higher levels than our hypothetical supply, and in these cases ozonation would yield a considerable public health benefit. However, there are also many drinking water supplies in which the concentration of bromate would be higher than our estimate, leading to higher health losses in achieving the desired disinfection goal. Bromate levels can be partially controlled by the adjustment of pH and optimization of the dose/contact time combination. If a high level of inactivation of C. parvum oocysts must be achieved, it will certainly lead to increased toxicologic risks. Our results highlight the need for raw water quality protection and a multiple barrier approach in treatment to reduce the need for disinfection as much as possible.

REFERENCES AND NOTES

(1.) WHO. Guidelines for Drinking-Water Quality. 2nd ed. Vol. 2: Health Criteria and Other Supporting Information. Geneva:World Health Organization, 1996.

(2.) Craun GF, Bull RJ, Clark RM, Doull J, Grabow W, Marsh GM, Okun DA, Regli, S, Sobsey MD, Symons JM. Balancing chemical end microbiological risks of drinking water disinfection. Part I: benefits and potential risks. Aqua (Oxf) 43:192-199 (1994).

(3.) National Academy of Sciences, National Research Council. Risk Assessment in the Federal Government: Managing the Process. Washington, DC:National Academy Press, 1983.

(4.) Colford JM, Tager IB, Hirozawa AM, Lemp, GF, Aragon T, Petersen C. Cryptosporidiosis among patients infected with human immunodeficiency virus. Am J Epidemiol 144:807-816 (1996).

(5.) Richardson AJ, Frankenberg RA, Buck AC, Selkon JB, Colbourne JS, Parsons JW, Mayon-White RT. An outbreak of waterborne cryptosporidiosis in Swindon and Oxfordshire. Epidemiol Infect 107:485-495 (1991).

(6.) Mac Kenzie WR, Hoxie NJ, Proctor ME, Gradus MS, Blair KA, Peterson DE, Kezmierczak JJ, Addiss DG, Fox KR, Rose JB, et al. A massive outbreak in Milwaukee of Cryptosporidium infection transmitted through the public water supply. N Engl J Med 331:161-167 (1994).

(7.) Ozekin K, Westerhoff P. Bromate formation under Cryptosporidium inactivation conditions. Water Qual Int May/June:16-17 (1998).

(8.) Kurokawa Y, Maekawa A, Takahashi M, Hayashi Y. Toxicity and carcinogenicity of potassium bromate--a new renal carcinogen. Environ Health Perspect 87:309-335 (1990).

(9.) DeAngelo AB, George MH, Kilburn SR, Moore TM, Wolf DC. Carcinogenicity of potassium bromate administered in the drinking water to male B6C3[F.sub.1] mice and F344/N rats. Toxicol Pathol 26:587-594 (1998).

(10.) Weinberg HS, Glaze WH. An overview of ozonation disinfection by-products. In: Disinfection By-Products in Water Treatment: The Chemistry of Their Formation and Control (Minear RA, Amy GL, eds). Boca Raton, FL:Lewis Publishers, 1996;165-186.

(11.) Marinas BJ, Rennecker JL, Teefy S, Rice EW. Assessing ozone disinfection with non biological surrogates. J Am Water Works Assoc 91:79-89 (1999).

(12.) Teunis PFM, Medema GJ, Kruidenier L, Havelaar AH. Assessment of the risk of infection by Cryptosporidium or Giardia in drinking water from a surface water source. Water Res 31:1333-1346 (1997).

(13.) Havelaar AH, Van Olphen M, Schijven JF. Removal and inactivation of viruses by drinking water treatment processes under full scale conditions. Water Sci Technol 31:55-62 (1995).

(14.) Van der Veer B, Schers G-J, Wilkes DR. Optimization of an ozonation system design to achieve disinfection and oxidation. In: Proceedings of the AWWA Annual Conference, Engineering and Operations, 19-23 June 1994, New York. Denver, CO:American Water Works Association, 1994;799-821.

(15.) Vose D. Quantitative risk analysis. A guide to Monte Carlo simulation modelling. Chichester:John Wiley & Sons Ltd, 1996.

(16.) Haas CN, Joffe J, Heath MS, Jacangelo J. Continuous flow residence time distribution function characterization. J Environ Eng 123:107-114 (1997).

(17.) Finch GR, Black EK, Gyurek L, Belosevic M. Ozone inactivation of Cryptosporidium parvum in demand-free phosphate buffer determined by in vitro excystation and animal infectivity. Appl Environ Microbiol 59:4203-4210 (1993).

(18.) Versteegh JFM, Neele J, Cleven RFMJ, Van Gaalen FW. Bromaat Tijdens de Drinkwaterzuivering en in Drinkwater [in Dutch]. Rapport no. 734301001. Bilthoven:Rijksinstituut voor Volksgezondheid en Milieu, 1995.

(19.) Song R, Donohoe C, Minear R, Westerhoff P, Ozekin K, Amy G. Empirical modeling of bromate formation during ozonation of bromide-containing waters. Water Res 30:1161-1168 (1996).

(20.) Haring BJA, Karres JJC, Van der Poel P, Zoeteman BCJ. Onderzoek Near de Gebruiksgewoonten bij Drinkwater-consumptie in Nederland [in Dutch]. H 2 0 12:212-216 (1979).

(21.) Haas CN. Estimation of risk due to low doses of microorganisms: a comparison of alternative methodologies. Am J Epidemiol 118:573-582 (1983).

(22.) Teunis PFM, Van der Heijden OG, Van der Giessen JWB, Havelaar AH. The Dose-Response Relation in Human Volunteers for Gastro-Intestinal Pathogens. Report no. 284550002. Bilthoven:National Institute of Public Health and the Environment, 1996.

(23.) DuPont HL, Chappel CL, Sterling CR, Okhuysen PC, Rose JB, Jakubowski W. The infectivity of Cryptosporidium in healthy volunteers. N Engl J Med 332:855-859 (1995).

(24.) Kurokawa Y, Takayama S, Konishi Y, Hiasa Y, Asahina S, Takahashi M, Maekawe A, Hayashi Y. Long-term in vivo carcinogenicity tests of potassium bromate, sodium hypochlorite, and sodium chlorite conducted in Japan. Environ Health Perspect 69:221-235 (1986).

(25.) U.S. Environmental Protection Agency. Guidelines for cancer risk assessment. Fed Reg 51:33992-34003 (1986).

(26.) U.S. Environmental Protection Agency. Proposed guidelines for cancer risk assessment. Fed Reg 61:17959-18011 (1996).

(27.) Kurokawa Y, Heyashi Y, Maekawa A, Takahashi M, Kokubo T, Odashima S. Carcinogenicity of potassium bromate administered orally to F344 rats. J Natl Cancer Inst 71:965-972 (1963).

(28.) Slob W, Pieters MN. A probabilistic approach for deriving acceptable human intake limits and human health risks from toxicological studies: general framework. Risk Anal 18:787-798 (1998).

(29.) Murray CJL. Rethinking DALYs. In: The Global Burden of Disease: A Comprehensive Assessment of Mortality and Disability from Diseases, Injuries, and Risk Factors in 1990 and Projected to 2020 (Murray CJL, Lopez AD, eds). Global Burden of Disease and Injury Series, Vol. 1. Boston, MA:Harvard School of Public Health on behalf of the World Health Organization and the World Bank, 1996;1-98.

(30.) Van der Maas PJ, Kramers PGN, eds. Gezondheid en Levensverwachting Gewogen. Volksgezondheid Toekomst Verkenning 1997, deel III [in Dutch]. Bilthoven/Maarssen:Rijksinstituut voor Volksgezondheid en Milieu/Elsevier/De Tijdstroom, 1997.

(31.) Ockhysen PC, Chappel CL, Sterling CR, Jakubowski W, DuPont H. Susceptibility and serologic response of healthy adults to reinfection with Cryptosporidium parvum. Inf Immun 66:441-443 (1998).

(32.) Mac Kenzie WR, Kazmierczak JJ, Davis JP. An outbreak of cryptosporidiosis associated with a resort swimming pool. Epidemiol Infect 115:545-553 (1995).

(33.) Mac Kenzie WR, Schell WL, Blair KA, Addiss DG, Peterson DE, Hoxie NJ, Kazmierczak JJ, Davis JP. Massive outbreak of waterborne Cryptosporidium infection in Milwaukee, Wisconsin: recurrence of illness and risk of secondary transmission. Clin Infect Dis 21:57-62 (1995).

(34.) Nord E. The person-trade-off approach to valuing health care programmes. Med Decis Making 15:201-208 (1995).

(35.) Hoxie NJ, Davis JP, Vergeront JM. Cryptosporidiosis-associated mortality following a massive waterborne outbreak in Milwaukee, Wisconsin. Am J Publ Health 87:2032-2035 (1997).

(36.) CBS. Overledenen near Doodsoorzaak (Serie A1), 1993-1995 [in Dutch]. Voorburg:Centraal Bureau voor de Statistiek, 1996.

(37.) Schouten LJ, Van den Brandt PA, Jager JJ, Smeets LBJM. Netherlands, Maastricht. IARC Sci Publ 120:670-672 (1992).

(38.) Perz JF, Ennever FK, Le Blanc SM. Cryptosporidium in tapwater. Comparison of predicted risks with observed levels of disease. Am J Epidemiol 147:289-301 (1998).

(39.) McGowan I, Hawkins AS, Weller IV. The natural history of cryptosporidial diarrhoea in HIV-infected patients. AIDS 7:349-354 (1993).

(40.) Houweling H. Public Health Surveillance of AIDS and HIV Infection in the Netherlands and Europe [PhD Thesis]. Amsterdam:University of Amsterdam, 1997.

(41.) Coebergh JWW, Damhuis RAM. Cancer of kidney and ureter. In: Cancer Incidence and Survival in the Southeast of the Netherlands, 1955-1994 (Coebergh JWW, Van der Heijden LH, Janssen-Heijnen MLG, eds). Eindhoven, The Netherlands:Comprehensive Cancer Centre South, 1995;74-75.

(42.) Damhuis RA, Blom JH. The influence of age on treatment choice and survival in 735 patients with renal carcinoma. Br J Urol 75:143-147 (1995).

Address correspondence to A.H. Havelaar, Microbiological Laboratory for Health Protection, National Institute of Public Health and the Environment (RIVM), PO Box 1, 3720 BA Bilthoven, The Netherlands. Telephone: [phone]. Fax: [phone]. E-mail [e-mail]

We thank D. Habbema and M-L. Essink Bot. We also thank P. Kramers and J. Melse for discussions on methodology and presentation of quality-of-life-related measures in public health decision making; K. Blair for the data on the mortality risk associated with cryptosporidiosis; and C. Haas, D. Wolf, and the reviewers for helpful comments on draft manuscripts.

Received 29 July 1999; accepted 26 October 1999.

Arie H. Havelaar, Augustinus E.M. De Hollander, Peter F.M. Teunis, Eric G. Evers, Henk J. Van Kranen, Johanna F.M. Versteegh, Joke E.M. Van Koten, and Wout Slob National Institute of Public Health and the Environment (RIVM), Bilthoven, The Netherlands