3.2. Calculation of Lifetime Cancer Risk Based on Toxicological Studies of Disinfection Byproducts
Assessment of human cancer risk from animal toxicology data requires a translation of the dose-response data from laboratory animals to humans, accounting for different body sizes and species-specific differences in toxicokinetics and toxicodynamics [48
]. The risk calculation can also include factors to account for heightened susceptibility of early life stages, such as the developing fetus, infants, and children [38
]. Following the U.S. EPA methodology, we calculated cancer potency for three brominated disinfection byproducts using Benchmark Dose Modeling Software (Table 3
). For this analysis, we used animal bioassay data for dibromoacetic acid, bromochloroacetic acid, and bromodichloroacetic acid from toxicology studies conducted by the National Toxicology Program [25
]. Benchmark responses of 5% or 10% are both used for benchmark modeling in risk assessment [50
]. We calculated human-equivalent cancer slope factors based on 95% lower confidence limits of the benchmark doses, following the approach of California OEHHA and the U.S. EPA. Table 3
demonstrates that calculated cancer slope factors are very similar, whether calculated based on 5% or 10% excess risk, suggesting that either approach can be used for assessing cumulative cancer risk for disinfection byproducts. Numerically smaller slope factors indicate weaker carcinogenic potency, while larger slope factors indicate higher carcinogenic potency. According to the available data, bromochloroacetic acid has higher carcinogenic potency compared to dibromoacetic acid and bromodichloroacetic acid (Table 3
We compiled the reported human cancer slope factors and one-in-a-million cancer risk benchmarks for haloacetic acids and trihalomethanes published by California OEHHA and by the U.S. EPA (Table 4
). For calculating cancer slope factors, California OEHHA used an approach with 5% extra risk, while U.S. EPA used 10%, and the two cancer slope values, where available for the same chemical, are overall similar.
Of note, California OEHHA used age-specific water intake and age sensitivity factors for defining cancer risk-specific concentrations, while the U.S. EPA used a default adult body weight of 70 kg, and default daily water intake of 2 L, and did not apply any adjustments for children’s body weight, water intake, or susceptibility to carcinogens. Differences in approaches for calculating risk-specific contaminant concentrations explain why those values are different between California OEHHA and the U.S. EPA, even though the cancer slope factors are broadly similar. The differences between calculated slope factors are likely due to the specific animal bioassay studies chosen for benchmark modeling. Our modeled cancer slope factor for dibromoacetic acid using 5% extra risk is close to the one reported by California OEHHA: a slope factor of 0.210 (mg/kg-day)−1
) versus a slope factor of 0.250 (mg/kg-day)−1
To illustrate the data options associated with the development of cancer slope factors, Figure 3
presents our derivation of human-equivalent cancer slope factors for dibromoacetic acid, bromochloroacetic acid, and bromodichloroacetic acid calculated from tumor incidence in different animal tissues. This comparison highlights how the choice of specific animal bioassay study and modeling approaches influence the calculated cancer slope factors. In National Toxicology Program studies, various types of tumors were observed in laboratory rats and mice exposed to these chemical substances [25
]. Among those tumors and tumor sites, the liver was the most sensitive organ for exposure-related tumor development, as indicated by the highest slope factors. As noted in Table 3
, we adapted cancer slope factors for dibromoacetic acid, bromochloroacetic acid and bromodichloroacetic acid based on the liver data. For all subsequent analyses in this study, we used cancer slope factors calculated with a benchmark response of 5% extra risk as the most sensitive approach for the detection of excess cancer risk.
To assess cancer risk due to the presence of haloacetic acids in drinking water, we were faced with a data limitation in the UCMR4 dataset whereby the U.S. EPA reported only the group concentrations for HAA5, HAA6Br, and HAA9 and not the concentrations of individual haloacetic acids. We developed concentration-weighted cancer risk benchmarks for the three HAA groups following Equation (5), as previously described [16
]. Calculated group risk benchmarks are listed in Table 5
and used for analyses in Table 6
Cancer bioassay animal studies are not available for chlorodibromoacetic acid and tribromoacetic acid. However, they were both listed by the National Toxicology Program as “reasonably anticipated to be a human carcinogen” based on metabolism data and similarity to other haloacetic acids, which have been tested in animal bioassays [57
]. Following a recent study from the scientists at the National Toxicology Program [58
], we used a read-across approach for these two haloacetic acids whereby the same cancer risk benchmark is applied to chlorodibromoacetic acid as calculated for bromochloroacetic acid and to tribromoacetic acid as calculated for dibromoacetic acid (Table 5
). With the parameters listed, we calculated an additive, concentration-weighted cancer risk benchmark for each group (Table 5
). We note that monochloroacetic and monobromoacetic acids are not considered carcinogenic and do not have a cancer risk benchmark. For chloroform, California OEHHA calculated a risk-specific concentration from the geometric mean of cancer slope factors modeled from five different animal bioassays [42
]. The benchmark dose from a study published by the U.S. National Cancer Institute [56
], with the cancer slope factor closest to this geometric mean, is included in Table 5
and used in later analysis for comparison (Figure 4
Based on group cancer risk benchmarks and UCMR4 data for the average concentrations of HAA5, HAA6Br, and HAA9 in each water utility in the UCMR program, we calculated the estimated number of lifetime cancer cases attributable to these chemical groups. The upper bound estimate presented in Table 6
represents the number of attributable cases calculated using the lower bound estimate of the benchmark dose, and vice versa for the lower bound estimate of cancer cases. The previously reported value of 47.9 thousand cases for trihalomethanes calculated for the dataset spanning 2010–2017 [16
] falls in this calculated range (Table 6
For the HAA5 group, we compared the estimates of attributable lifetime cancer cases from a comprehensive national dataset and from the UCMR4 dataset. Calculations for the UCMR4 dataset results in approximately 72% of cases expected for the HAA5 group from the national dataset. This finding makes sense, as the UCMR4 program includes all the large community water systems with a combined population of approximately 183 million people. It is reasonable to hypothesize that the UCMR4 data for HAA6Br and HAA9 underestimate the total nationwide HAA6Br- and HAA9-attributable cases to a similar extent as we report for the HAA5 group. As noted earlier, cancer slope factors and cancer potency are greater for brominated haloacetic acids compared to chlorinated haloacetic acids, and this is reflected in the greater number of attributable cancer cases due to the HAA6Br group relative to the HAA5 group. As expected, the greatest number of attributable cases is calculated for the HAA9 group. Finally, we calculated cumulative cancer risk due to HAA9 and THM4 groups for the subset of systems included in the UCMR4 program. Central estimates for this cumulative lifetime risk range between 7.0 × 10−5
and 2.9 × 10−4
, depending on whether default parameters or age-sensitivity parameters are used to estimate risk (Table 6
The read-across approach for chlorodibromoacetic acid and tribromoacetic acids represents an area of uncertainty for our study. Given that trichloroacetic acid is a more potent carcinogen compared to dichloroacetic acid, we hypothesize that a similar cancer potency relationship would apply for dibromoacetic and tribromoacetic acids. Further, the population-weighted concentrations of chlorodibromoacetic acid and tribromoacetic acid, at 0.28 and 0.21 µg/L, respectively, are lower than the population-weighted concentrations of other haloacetic acids associated with cancer risk, which range from 1.1 to 7.8 µg/L. Thus, we anticipate that the uncertainty around cancer potency of chlorodibromoacetic acid and tribromoacetic acid would not significantly affect the estimated cumulative cancer risk due to the haloacetic acids presented here.
3.3. Calculation of Annual and Lifetime Cancer Risk Based on Epidemiological Studies of Disinfection Byproducts
To assess cancer risk based on human data, we relied on epidemiological studies that found an association between the presence of disinfection byproducts in drinking water (marked by the THM4 levels) and increased risk of bladder cancer. Trihalomethane concentrations correlate with HAA9 concentrations in drinking water (Figure 1
), and an assumption is made that trihalomethane concentrations may correlate with levels of other carcinogenic disinfection byproducts in tap water. For this portion of our analysis, we used the THM4 occurrence data for 2014–2017 for 48,363 community water systems in the U.S., serving an estimated 86% of the U.S. population. All estimates for attributable cases were calculated together with upper and lower boundaries based on epidemiologically-derived odds ratios for bladder cancer risk.
The largest number of estimated lifetime bladder cancer cases, approximately 400,000 (Table 7
), is calculated for large surface water systems serving communities with more than 0.1 million residents. Combined, these systems serve 40% of the total population served by community water systems in the U.S. As expected, a lower risk is associated with small groundwater systems that serve a smaller proportion of the population and generally have lower disinfection byproduct levels. Converting the estimates of disinfection byproduct-attributable lifetime bladder cancer cases into cancer risk, we estimate that lifetime cancer risk from disinfection byproducts is 3.0 × 10−3
(2.1 × 10−4
–5.7 × 10−3
) for the 279 million people served by community water systems in the U.S. These estimates are significantly greater than the de minimus risk of one-in-a-million (10−6
). We completed the same analysis just for the water systems included in the UCMR4 monitoring program and obtained very similar risk estimates of 3.2 × 10−3
(2.1 × 10−4
–6.1 × 10−3
). This similarity in estimates makes sense given that the UCMR4 program includes all large community water systems in the U.S.
We note that different risk estimates are obtained from calculations based on either annual bladder cancer incidence or lifetime risk of a bladder cancer diagnosis. Analysis based on the annual incidence of bladder cancer results in an annual risk of 2.4 × 10−5
due to disinfection byproducts. Multiplying the annual cancer risk estimate by 70, the “statistical length of life” used by U.S. EPA for human health risk assessments, results in a lifetime risk of 1.7 × 10−3
. This value is lower than an estimated lifetime attributable risk of 3.0 × 10−3
calculated from data in Table 7
because the lifetime bladder cancer probability is estimated with the National Cancer Institute DevCan software, which uses a longer lifetime of 95 years and accounts for the higher risk of developing bladder cancer with age.
3.5. Comparison of Epidemiologically-Based and Toxicologically Based Cancer Risk Estimates
To compare risk estimates using toxicological and epidemiological approaches, we focused all analyses on the 3579 systems in the UCMR4 program. For the toxicological estimates, we calculated the combined cancer risk for THM4 and HAA9 based on lower bound, central, and upper bound estimates of the benchmark dose-derived cancer slope factors. The overall toxicologically-based cancer risk was estimated as 7.0 × 10−5
(3.5 × 10−5
–1.3 × 10−4
) using default body weight and water intake factors and 2.9 × 10−4
(1.7 × 10−4
–6.2 × 10−4
) when incorporating age sensitivity factors (Figure 4
). The epidemiologically-based assessment of cancer risk in Figure 4
is calculated from lower, central, and upper estimates of bladder cancer risk reported in human studies.
In a side-by-side comparison of toxicological and epidemiological risk estimates, it is important to acknowledge tumor site non-concordance between animal and human studies. While epidemiological research finds an association of disinfection byproducts in drinking water with bladder cancer, tumors at different sites are observed in laboratory animals (see Figure 3
). Using the same exposure data, the central estimate of risk based on human data is five-fold greater than the upper bound estimate of risk calculated from animal data with included age sensitivity factors. This comparative analysis suggests that human epidemiological data must be used to the greatest extent possible to capture risks based on real-world exposures. Risk calculations based on animal bioassays are an essential part of risk assessment and mitigation. As Figure 4
demonstrates, the inclusion of age sensitivity factors brings toxicological risk estimates closer to epidemiological risk estimates, while the default weight and water intake factors produce risk estimates that are smaller than epidemiological estimates by approximately 40-fold.