5.1.1. PRIOR UNCERTAINTY IN TOXICITY

In order to apply the VOI framework discussed in Section 2, it is necessary to establish a prior distribution of uncertainty in chemical toxicity in the absence of any specific knowledge about the toxicity of the chemical to be tested. The original VOI framework used data previously considered by Krewski et al. (1993) on variation on the potency of the chemical carcinogens as well as data used by Chiu et al. (2018) on variation in toxicity reflected in distribution of PODs for non-cancer, critical effects. To characterize the prior distribution of chemical toxicity in this case study, we utilize data on 608 chemicals considered previously by Chiu et al. (2018). These chemicals include substances evaluated under the EPA Integrated Risk Information System (IRIS), the Office of Pesticide Programs (OPP), Superfund Regional Screening Levels (RSLs) programs, as well as substances evaluated by California EPA Office of Environmental Health Hazard Assessment. With more than one endpoint evaluated for many of these chemicals, there are a total of 1,522 chemical-endpoint combinations in this database. As indicated in Figure 5-1, the 1,522 chemicals and endpoints considered by Chiu et al. (2018) reflect a variety of subchronic and chronic toxicity (non-cancer) endpoints.⁷

Figure 5-1

Overview of toxicological effect types represented by 1,522 chemicals and endpoints. [Reproduced from Chiu et al. (2018), Figure 5A, originally published in Environmental Health Perspectives, with permission from the authors.].

The median human dose (HD_M^I) associated with an effect of magnitude (M) and population incidence (I) across the 1,522 chemicals and endpoints considered by Chiu et al. (2018) is ${\mathrm{H}\mathrm{D}}_{\mathrm{M}}^{50}=$ using the average human body weight of 80kg (EPA 2011c), corresponding to a value of μ_tox=log₁₀(HD_M⁵⁰)=0.51. Excluding chemicals with extremely high potencies, such as 2,3,7,8-tetrachlorodibenzo-p-dioxin and those chemicals tested above the limit dose of 1,000 mg/kg-day, the distribution of toxic potency spans approximately 6 orders of magnitude (OM). Assuming this represents approximately 99.9% of the variation in the PODs for non-cancer endpoints expressed by these chemicals, the log₁₀ prior uncertainty standard deviation in chemical toxicity of an untested chemical u⁰(μ_tox) is given by 6/(2z_0.9995)=6/6.58=0.912.

Information on σ_tox, which is the logarithm of the geometric standard deviation [log₁₀(GSD)] of human susceptibility, is provided by the International Programme on Chemical Safety (IPCS) [WHO (2017), Table 4.4]⁸. The value of σ_tox is calculated to be 0.424 using the midpoint of 5^th and 95^th percentiles (P₀₅=0.151 and P₉₅=0.697), with uncertainty about er_tox being u(σ_tox)=0.166. Since σ_tox cannot be negative, the VOI framework currently integrates uncertainty distribution between ±6u(·) about the mean, u(σ_tox) as σ_tox/6=0.0706.

5.1.2. CONDITIONAL POSTERIOR UNCERTAINTY ABOUT μ_tox FOR THHA

In order to extrapolate the results of two-year rodent bioassays to humans, it is necessary to translate the value of μ_tox to an HED. This is done using the HD_M^I, as outlined by Chiu et al. (2018). In addition to providing a best estimate of the HED, the uncertainties associated with the various steps in this extrapolation need to be considered. As indicated in Table 5-1, key sources of uncertainty include variation among animals within a given bioassay, uncertainties in allometric scaling in extrapolating from animals to humans, and differences in toxicokinetics (TK) and toxicodynamics (TD) between animals and humans, after adjusting for differences in body weight.

Table 5-1

Sources of uncertainty for two-year rodent bioassays in THHA.

The sources of uncertainty summarized in Table 5-1 are characterized by the ratio, P₉₅/P₅₀, of the 95^th to 50^th percentiles of the respective uncertainty distributions. Intra-study variation is reflected in the BMD/BMDL ratio. The BMDL is the lower confidence limit of the estimated BMD. As a result, toxicity tests with less uncertainty will result in the BMDL being closer to the BMD, and correspondingly smaller BMD/BMDL ratios. To estimate the uncertainty associated with the BMD values derived from animal bioassay, data from 584 two-year rodent bioassays are used to determine the distribution of BMD₁₀/BMDL₁₀ ratios (Sand et al. 2011), where BMD(L)₁₀ corresponds to BMD(L) values associated with a benchmark response (BMR) of 10% extra risk. The mean BMD₁₀/BMDL₁₀ ratio across 584 datasets was 1.803 and is used as the estimated intra-study variability shown in Table 5-1. The uncertainty due to allometric scaling (1.235) is based on results reported by Chiu et al. (2018), and the uncertainty in differences in TK/TD between animals to humans (3.000), is taken from published work by the IPCS [WHO (2017), Table 4.3].

Let HD denote the HD_M⁵⁰. Following Chiu and Slob (2015), the uncertainty standard deviation for HD using the three sources of uncertainty about HD in Table 5-2 can be obtained by first calculating the ratio P₉₅/P₅₀ for the HD as

Table 5-2

Sources of uncertainty for five-day transcriptomic studies in ETAP.

The sample standard deviation σ_B for the bioassay is obtained by taking the logarithm of the GSD, with

Under the assumption of lognormality in both the prior uncertainty in μ_tox and variability in the sample information, application of Bayesian updating leads to the conditional posterior uncertainty standard deviation of

5.1.3. CONDITIONAL POSTERIOR UNCERTAINTY ABOUT μ_tox FOR ETAP

The information required to derive the posterior conditional uncertainty about μ_tox following a five-day transcriptomic study is summarized in Table 5-2. This source of uncertainty is gauged in terms of intra-study variability, estimated by the average of the transcriptomic BMD₁₀/BMDL₁₀ ratios for 14 chemicals for which both five-day in vivo transcriptomic and chronic rodent bioassay dose-response data are available and analyzed as described in the ETAP scientific support document (EPA 2024c; Gwinn et al. 2020)⁹. Since both the two-year bioassay and five-day transcriptomic studies are in vivo toxicity testing strategies using rodents, the uncertainty due to allometric scaling and animal-human TK/TD remains the same for both methods.

Using the VOI parameters in Table 5-2 and performing the same calculations presented in Section 5.1.2 for the bioassay, the sample standard deviation of the five-day transcriptomic studies in ETAP is obtained as

It should be noted that, for both methods, toxicity testing reduces the uncertainty in μ_tox, but not σ_tox, as the latter parameter reflects inter-individual variation in susceptibility in the target population.

5.3.1. ADVERSE HEALTH OUTCOME VALUATION

The economic value associated with adverse health outcomes varies widely with the severity of the outcome, including whether the outcome is acute or chronic, non-life-threatening or fatal, and irreversible or reversible. Health economists have estimated and applied a value of a statistical life (VSL) of $8.8M¹² (EPA 2022b). Most valuations of non-fatal adverse health effects focus on direct medical costs, lost productivity, and direct non-medical costs such as education or transportation.¹³ In an international review of the economic costs associated with childhood disabilities, Shahat and Greco reported a range of annualized values from $450 to $69,500, corresponding to $36,000 to $5.6M over the course of an 80-year lifetime (Shahat and Greco 2021). While the range of valuations is informative, the lower values in this range may not be directly relevant in the U.S. context, because of the higher cost of the U.S. health system compared to other developed countries. Economic values have also been estimated for specific diseases in the U.S. including autism at $69,530 annually (Ganz 2007), asthma at $36,500 (Belova et al. 2020). Down syndrome at $15,311 (Peng et al. 2009), and pervasive developmental disorders at $10,538 (Peng et al. 2009), Buescher et al. (2014) estimated the annual costs (in 2011 dollars) for children with autism spectrum disorders (ASD) and intellectual disability at $86,000 - $107,000 annually (depending on age) and for children with ASD and no intellectual disability at $52,000 - $63,000. Although these examples do not constitute a systematic review of the literature on health effects valuations, they provide useful benchmarks for determining the range of economic values for adverse health outcomes to be used in the present VOI analysis.

In the previously published VOI analyses, Hagiwara et al. (2022) considered annualized valuations of $110,000 for a fatal outcome.¹⁴ Acute adverse health effects, such as a restricted airway event, were valued at $70 per occurrence¹⁵ (EPA 2013, 2021). Assuming an expected rate of occurrence of one event per week, this corresponds to an annualized cost of $2,600. Considering the range of valuations discussed above, values of $110,000, $10,000, and $1,000 are used in the present VOI analysis to capture the potentially broad range of adverse outcomes following exposure to an untested chemical. As the great majority (>98%) of the adverse health effects caused by the 1,522 chemicals and endpoints considered by Chiu et al. (2018) are not fatal, a value of $10,000 is used in the baseline scenarios in the analysis, with the other two values presented as sensitivity analyses (Section 6.2).

5.3.2. CONTROL COSTS

Control costs were evaluated that would capture both the high and low ends in the range of costs for a chemical that may be evaluated by ETAP. At the high end, the maximum annualized control cost, denoted by ACC_max, for reducing emissions of key air pollutants in the United States was considered. EPA estimated an average annual control cost of $2.0B for individual air pollutants such as acid gas, mercury (Hg), particulate matter less than 2.5 microns in aerodynamic diameter (PM_2.5), and sulfur dioxide (SO₂) (EPA 2011b). Trends in emission rates demonstrated a reduction of 25% between 1990 and 2021 across the following seven key air pollutants: carbon monoxide (CO), ammonia (NH₃), nitrogen oxides (NO_x), PM_2.5, PM₁₀, SO₂, and volatile organic compounds (VOCs) (EPA 2022). As the cost of exposure reduction is expected to increase as exposure is reduced to lower and lower levels, the annualized control is modelled as

At the lower end of the range of control costs, a recent evaluation of the costs of chemical restriction proposals between January 2010 to May 2020 under the European Union's Registration, Evaluation, Authorisation and Restriction of Chemicals (REACH) indicated an annualized total expenditure of €1.7B across all the proposals (ECHA 2021). The annualized control cost associated with each of the 33 risk management programs considered in this evaluation are shown in Figure 5-3. Annualized control cost ranges from a low of €12K associated with substitution of trifluoroacetate salts in spray products to a high of €955M for reformulation and compliance cost for intentionally added microplastics.¹⁶ The mean and median control cost across all chemical control programs included in this program were €53.3M and €6M, respectively, corresponding to $50.6M and $5.7M, based on average 2022 exchange rates. Excluding those programs with zero cost, the mean and median values increased to €60.7M ($57.6M) and €17.6M ($16.7M), respectively. Based on the diverse set of chemical control programs, a control cost using ACC_max of $578M (using Eq. (8) with a $50M ACC at 25% reduction and η=1) is included for the sensitivity analysis in addition to the $23.1B used for the baseline scenarios discussed above.

Figure 5-3

Boxplot of annualized control cost associated with 33 risk management programs under the REACH registration (in €M). The overall median and mean control costs are €6M and €53.3M, respectively.

5.3.3. AFFECTED POPULATION SIZE

For the baseline scenarios, the size of the affected population is set to be N=330M people, representing essentially the entire U.S. population. To investigate the impacts associated with only a subset of the population being exposed, two sensitivity analysis scenarios were evaluated for population sizes in which 165M (50% of the U.S. population), and 33M (10%) people are exposed.

5.3.4. DISCOUNT RATE

To account for the fact that costs and benefits realized in the future are of less economic value than those realized immediately, a discount rate of r=5% is applied in the calculation of the TSC to standardize all costs and benefits to their net present values at the beginning of the time horizon. The present case study employed a discount rate of 5%, consistent with the recommendation of the EPA Science Advisory Board (SAB)(EPA 2004) and current economic conditions. Other discount rates could also be considered, such as the value of 3% subsequently recommended by the 2010 EPA economic analysis guidelines (EPA 2010). Sensitivity analyses previously conducted by Hagiwara et al. (2022) using discount rates of 3, 5, and 7% showed that choice of discount rate did not dramatically alter VOI.

5.3.5. TOXICITY TESTING AND ASSESSMENT DURATION

The baseline analysis relies on estimates that the THHA requires 8 years from the start of toxicity testing to reaching a regulatory decision. This timeline was based on observations that the traditional two-year rodent bioassay takes an average of 4 years to complete (Faustman and Omenn 2015; NTP 1996; Pastoor and Stevens 2005), and the typical human health assessment process is estimated to take an additional 4 years (Krewski et al. 2020). To investigate the effect of shortening the testing and assessment time of THHA, a sensitivity analysis scenario was considered that reduces the human health assessment process to 2 years, thereby encompassing a 6-year timeframe for testing and decision making. The 2-year timeframe is consistent with the targets for implementation of the streamlined IRIS assessment process for less challenging assessments¹⁷. The effects of a longer time frame for THHA were also considered. The NTP routinely establishes and conducts high quality toxicity testing studies on chemicals selected for analysis via a nomination process. For the most recent chemicals tested that evaluated effects of chronic exposure to adult animals as the primary focus, the completion time from the start of the 2-year bioassay and issuance of the report was on average 8 years, with a standard deviation of approximately 1 year, and with a maximum duration of 9.5 years¹⁸. Based on this information, sensitivity analysis was performed on the approximate maximum of 10 years for toxicity testing and the 4 years required to complete the assessment with the total time resulting in a delay in decision making as long as 14 years with THHA.

For the ETAP, the baseline analysis assumes that the studies and human health assessment can be completed in 6 months¹⁹. It is possible that delays may be encountered when performing an ETAP due to issues such as chemical purity, stability, or selecting the appropriate dose range. To investigate the effects of lengthening the time required for testing and developing the human health assessment for ETAP, additional sensitivity analyses were performed assuming 1-year and 2-year timeframes.

5.3.6. TIME HORIZON

The time horizon, denoted by TH, over which the TSC is calculated, is set to 20 years in the baseline scenarios since it is a time frame commonly used in health economics. In sensitivity analysis scenarios, additional time horizons of 40 years and 75 years were considered for evaluating impacts of health effects that may be encountered by the exposed population over a generation or over a lifetime.

5.3.7. STUDY AND ASSESSMENT COSTS

In the baseline scenarios, the costs of performing a chronic rodent bioassay was estimated to be $4M and derived from multiple sources (Faustman and Omenn 2015; NTP 1996; Pastoor and Stevens 2005). To evaluate the impact of alternative assumptions about the cost of toxicity testing, an additional sensitivity analysis was conducted using different study costs. Bottini and Hartung estimated the cost of a traditional chronic rodent bioassay to be €780K, or $845K (Bottini and Hartung 2009). Typically, a traditional chronic bioassay is conducted following a subchronic or other short-term study to set the appropriate dose range. When combined with a subchronic toxicity study (€116K, or $126K) or a short-term study (€49K, or $53K), the total cost of testing for THHA may be between $898K and $971K. For the sensitivity analysis, the cost of THHA was reduced to $1M.

For the ETAP, the five-day in vivo transcriptomic study and ETAP process currently costs approximately $200,000 for the baseline scenario²⁰. The most variable cost component of an ETAP assessment is chemical procurement, which can range from less than $5K to over $50K. In the sensitivity analysis, the effect of changing the study cost was investigated by assuming that the cost associated with ETAP is increased to $250K.

Although it is recognized that there are labor costs associated with the review of available data and with the composition and issuance of the human health assessment, in the absence of authoritative quantitative information on the relevant average labor costs, the cost associated with development of the traditional human health assessment is presumed to be $0 for the purposes of the case study. In the case of ETAP, the assessment labor costs are expected to be significantly less than with THHA due to the ETAP standardized reporting format and significantly shorter time required for data review and quality assurance steps, which would impact the VOI to favor ETAP.

5.3.8. ADDITIONAL CONSIDERATIONS FOR TARGET-RISK DECISION-MAKING

The decision rule for the TRDM requires specification of the q_L and q_U to determine whether the chemical of interest warrants exposure mitigation action. In the present analysis, the q_L and q_U are set to 5% and 95%, respectively. Unlike the BRDM, who considers the cost of exposure reduction, leading to an optimal reduction in exposure, decisions taken by the TRDM are driven by achieving the TRL regardless of exposure reduction cost. Acknowledging that it is not always possible to eliminate exposure, it is assumed that the TRDM reduces the GM of the population exposure by 90% when exposure mitigation is deemed necessary (Hagiwara et al. 2022). While EPA does not have a recommended TRL for non-cancer effects, a value of 10⁻⁶ is often used for cancer-related effects (EPA 2009). The baseline scenarios were conducted using a TRL=10⁻⁶. For the sensitivity analysis, a TRL=10⁻⁴ was also considered. Chiu et al. (2018) noted that the median residual risk at the traditional reference dose (RfD) for most of the 1,522 chemicals and endpoints included in their analysis is less than 0.01%.

A summary of the parameter values governing prior information on toxicity and exposure, toxicity testing and assessment, economic valuation of health impact, and decision making contexts used in the baseline scenarios of ETAP versus THHA presented in Section 6.1 is given in Table 5-4. Results from the variations in these parameter values were used in the sensitivity analyses presented in Section 6.2.

Table 5-4

Summary of parameter values used in the baseline VOI analysis of ETAP versus THHA.