2
Hazard Identification

This chapter discusses considerations to inform cancer hazard identification conclusions. It addresses the first charge to the committee, specifically to consider:

the overall weight of the evidence for a causal relationship between ethylene oxide and breast cancer risk in humans at occupational concentrations and at environmentally relevant concentrations of ethylene oxide in ambient air.

To address this first charge, the chapter discusses the following topics: prior cancer assessments of ethylene oxide, the systematic review process, epidemiological evidence for cancer hazard identification (including attention to the Healthy Worker Effect [HWE] and characterization of statistical significance), incorporation of experimental animal evidence, and mode of action (MOA) considerations. Findings and recommendations concerning identified deficiencies are then provided.

In its evaluation of the Texas Commission on Environmental Quality’s (TCEQ’s) 2020 Ethylene Oxide Carcinogenic Dose-Response Assessment: Development Support Document (hereafter referred to as the TCEQ DSD), the committee also considered the broad guidance for the development of toxicity factors provided in TCEQ Guidelines to Develop Toxicity Factors (TCEQ 2015; hereafter referred to as TCEQ Guidelines).

PREVIOUS CANCER ASSESSMENTS

The International Agency for Research on Cancer (IARC) classified ethylene oxide in Group 1 (carcinogenic to humans) in 1994 (see IARC 1994). The Group 1 classification was confirmed in subsequent IARC evaluations of ethylene oxide, most recently in 2012 (IARC 2012). The IARC evaluation found limited evidence in humans for a causal association of ethylene oxide with lymphatic and hematopoietic cancers and breast cancer. Even though the human evidence was not conclusive, ethylene oxide was classified as carcinogenic to humans (Group 1) based on sufficient evidence that ethylene oxide causes cancer in experimental

animals and strong evidence from exposed workers and in experimental systems that ethylene oxide, a direct-acting alkylating agent, operates by a genotoxic mechanism.

The U.S. Environmental Protection Agency (EPA) classified ethylene oxide as carcinogenic to humans (U.S. EPA 2016a). This classification was based on human occupational studies showing elevated incidence of lymphoid cancer and breast cancer in female workers; tumors in experimental animals; and that the weight of evidence (WOE) supported a mutagenic MOA (U.S. EPA 2016a).

TCEQ relied on both the IARC (2012) and U.S. EPA (2016a,b) assessments as its starting point for the carcinogenic WOE for ethylene oxide (TCEQ 2020, p. 12), consistent with TCEQ guidance (TCEQ 2015, p. 154). TCEQ gathered relevant evidence from these existing evaluations performed by IARC (2012) and U.S. EPA (2016a,b) and supplemented it with peer-reviewed studies published after 2016, the date of the most recent U.S. EPA evaluation.

SYSTEMATIC REVIEW PROCESS AND CONSIDERATIONS

For more than a decade, systematic approaches to the evaluation of scientific evidence have been considered the state of practice for conducting risk assessments of chemicals. Systematic review is defined as “a scientific investigation that focuses on a specific question and uses explicit, prespecified scientific methods to identify, select, assess, and summarize the findings of similar but separate studies” (IOM 2011, p. 1). Systematic review methods are designed to promote transparency and reproducibility in the identification, selection, and assessment of studies, and to include approaches for weighing evidence within and across evidence streams in support of hazard identification and dose-response assessment. National and international working groups have developed guidance and tools for systematic reviews relating to environmental health (Descatha et al. 2018; Morgan et al. 2018; NTP 2019; Schaefer and Myers 2017; Woodruff and Sutton 2014) as well as uniform approaches for evidence integration to support cancer hazard identification (Samet et al. 2020). Numerous consensus reports of the National Academies have outlined, endorsed, or applied the systematic review process to environmental risk assessments, as outlined in Box 2-1.

The committee applauds TCEQ’s decision to align the TCEQ DSD’s approach with the standard of practice for toxicological assessments by applying systematic review methods. The following sections detail where best practices were followed and where opportunities for improvement were identified.

Problem Formulation and Protocol

Human health risk assessment includes an initial scoping and problem formulation step to inform the development of the research questions and the best evidence-based approach for answering them (NRC 2009). A protocol is established before the assessment begins, and it plays a critical role in ensuring transparency,

BOX 2-1
National Academies’ Consensus Advice on Systematic Review in Risk Assessment

Authoritative guidance across several consensus reports of the National Academies has outlined a rigorous systematic review process for identifying and characterizing the human health hazards of environmental exposures. This guidance extends from the 2011 review of the assessment of formaldehyde by U.S. EPA’s Integrated Risk Information System (IRIS), and includes subsequent recommendations for U.S. EPA’s IRIS program and Toxic Substances Control Act evaluations and to the U.S. Department of Defense (NASEM 2018, 2019, 2021, 2022a,b, 2023a,b; NRC 2011, 2014). In addition, numerous National Academies’ reports have applied systematic review methods in evaluations of environmental agents (e.g., Application of Systematic Review Methods in an Overall Strategy for Evaluating Low-Dose Toxicity from Endocrine Active Chemicals [NASEM 2017]).

The steps of a well-conducted systematic review encompass methods to identify, select, assess, and synthesize the relevant body of evidence. As shown in Figure 2-1-1, Review of EPA’s Integrated Risk Information System (IRIS) Process (NRC 2014) provides a framework for the parallel, but separate, reviews for the three relevant evidence streams: human, animal, and mechanistic. The framework describes the flow of evidence utilization through the steps of evaluation of studies, evidence integration for hazard identification, and derivation of toxicity values.

consistency, and objectivity in the review process. The protocol provides comprehensive documentation of inclusion and exclusion criteria for compiling evidence, as well as the methods for the subsequent steps of the systematic review. Implementation of the protocol helps ensure reproducibility and prevent bias.

TCEQ’s approach to identification and evaluation of the ethylene oxide literature was not consistent with the state of practice for problem formulation and

protocol development. A short problem formulation section is included in the main text of the TCEQ DSD that provides procedural details of TCEQ’s decision-making concerning whether to accept U.S. EPA’s 2016 toxicity values for ethylene oxide or develop its own. While a purpose for the assessment is stated, the problem formulation statement lacks several key elements needed to scope the assessment properly (NASEM 2017, 2019). Specifically, the document does little to define the scope of the problem (beyond noting that ethylene oxide is emitted in Texas), presents no specific questions to guide analyses, and does not establish (or mention) specific assessment methodologies it plans to use to address the questions. A somewhat more detailed problem formulation is provided in Appendix 1 but not called out in the main document. The fuller problem formulation in Appendix 1 does provide a broad set of questions (e.g., “What are the physical and chemical properties of ethylene oxide?” “Is ethylene oxide carcinogenic, and if so, is it carcinogenic by a specific route of exposure?” “What is the critical effect following exposure to ethylene oxide?”) and mentions using systematic methods, but both the main text and Appendix 1 are overly general and limited in specificity.

Appendix 1 (TCEQ 2020, A1.1) of the TCEQ DSD appears to conflate problem formulation with an abbreviated systematic review protocol that is significantly limited in detail and focuses primarily on dose-response assessment and derivation of toxicity factors. The protocol provides no specifics regarding the types of cancers being considered, the methods for assessment of bias or external validity, or plans to employ frameworks for evidence synthesis and integration. The lack of a detailed protocol in the TCEQ DSD is inconsistent with prior guidance from the National Academies, which suggests development of an a priori, prepublished protocol for assessments (NASEM 2023b).

Overall, a major flaw in TCEQ’s approach is that systematic processes were not used to guide the identification and evaluation of evidence in support of the hazard determination for cancer but were instead only employed for the dose-response assessment. This choice resulted in inadequately supported hazard identification, which may have implications for downstream decision-making and confidence in conclusions built upon these bases.

Guidance for the Hazard Assessment

As systematic review approaches were initiated only for the TCEQ DSD’s dose-response assessment, no formal approach or methodology is presented for hazard assessment. TCEQ Guidelines is cited at the start of Chapter 3 of the TCEQ DSD to support the statement that TCEQ only performs carcinogenic dose-response assessments for chemicals it considers to be “Carcinogenic to Humans” or “Likely to Be Carcinogenic to Humans”; however, no guidelines are provided in this guidance document for hazard assessment of carcinogens (TCEQ 2015). Another TCEQ guidance document on systematic review and evidence integration (TCEQ 2017) is cited later in the TCEQ DSD, but, upon inspection,

the committee found that this document only provides guidelines for systematic review and evidence integration when conducting dose-response assessment. The committee recommends that formal guidance for the use of systematic review in hazard assessment should be followed (e.g., see Box 2-1); see Recommendation 2-1.

Best Practices for Systematic Review for Hazard Identification

TCEQ mentions prior hazard evaluations from IARC (2012) and U.S. EPA (2016a) as “starting points for the carcinogenic weight of evidence hazard assessment” and notes that they added relevant studies published after 2016 (TCEQ 2020). No methods for identifying relevant studies published after 2016 are presented, and only three additional publications are discussed regarding human evidence. Two of these were narrative reviews (Marsh et al. 2019; Vincent et al. 2019) aimed at reevaluating the prior epidemiologic literature. Throughout the hazard assessment, the reviews by both Vincent and colleagues and Marsh and colleagues were clearly an influential basis for TCEQ’s decision-making, particularly in the case of hazard determination for the breast cancer endpoint, leading TCEQ to conclude that there was insufficient evidence for the determination of a breast cancer hazard (TCEQ 2020). However, TCEQ’s approach does not reflect best practices for evaluation of a prior review for quality and rigor using established tools, such as ROBIS¹ or AMSTAR 2 (see Shea et al. 2017), as have been implemented in National Academies’ reports (e.g., NASEM 2021, 2022b, 2023a). Based on this, the committee found that the process for evaluation of prior evidence in TCEQ’s hazard assessment fails to provide a credible basis for its hazard conclusions.

The committee notes that human epidemiologic evidence was excluded from TCEQ’s systematic review for dose-response if the study did not include known exposure concentrations or if exposure concentrations were not able to be estimated. While quantified exposure is a necessity for dose-response assessment, the lack of such information does not preclude a study from informing hazard identification. Evidence with semi-quantitative or grouped exposure levels can inform hazard identification, and available studies include those evaluating cancer endpoints in cohorts environmentally exposed to ethylene oxide in which exposures were categorized according to emission concentrations near study participant residences (Hart et al. 2018; Jones et al. 2023). Such studies are particularly relevant given the committee’s statement of task to consider “the overall weight of evidence for a causal relationship between ethylene oxide and breast cancer risk in humans at occupational concentrations and at environmentally relevant concentrations of ethylene oxide in ambient air.”

A more suitable approach for the hazard assessment is to perform a systematic review for both lymphoid and breast cancers, including an updated systematic

___________________

¹ The website for ROBIS is https://www.bristol.ac.uk/population-health-sciences/projects/robis/robis-tool/, accessed August 22, 2024.

review of the literature in a manner consistent with the state of practice (i.e., appropriate scoping and problem formulation and a prepublished systematic review protocol) (see Recommendation 2.1). Short of that, established tools can be applied to evaluate critically the quality of existing reviews upon which conclusions are drawn in evidence synthesis and integration to inform hazard determinations (NASEM 2017).

EPIDEMIOLOGICAL EVIDENCE FOR BREAST CANCER

TCEQ concluded that there was epidemiological evidence, albeit inconsistent, for associations between ethylene oxide exposure and lymphoid cancer in exposed workers, and those data were used to derive a unit risk factor.² This decision was informed by TCEQ’s review of two epidemiological studies performed by the National Institute for Occupational Safety and Health (NIOSH) and Union Carbide Corporation. The findings of these studies are described in more detail in Chapter 3.

The following section discusses these studies in the context of TCEQ’s hazard assessment for breast cancer. TCEQ concluded there was insufficient epidemiologic evidence for a relationship between ethylene oxide exposure and breast cancer.³ However, several studies provide some evidence for a causal relationship between ethylene oxide and breast cancer risk in humans at occupational concentrations that may prompt TCEQ to reconsider its hazard determination for breast cancer.

The NIOSH studies of sterilizer workers (Steenland et al. 2003, 2004) demonstrated excess breast cancer risk in association with ethylene oxide exposure. Analyses of breast cancer incidence showed significant exposure-response effects with occupational exposure using internal analyses in both categorical and continuous cumulative exposure models. These findings were further supported by breast cancer mortality analyses.

The breast cancer incidence study by Steenland and colleagues (2003) examined data for 7,576 women workers in sterilization facilities. Breast cancer incident cases (n = 319) were ascertained via interview, death certificates, cancer registries, and medical records. Interviews were only available for 68% (n = 5,139) of the cohort. The standardized incidence ratio (SIR) for the whole cohort relative to external referent rates (Surveillance, Epidemiology, and End Results Program) was 0.89 (95% confidence interval [CI] 0.78–1.01) for a 15-year lag (Steenland et al. 2003, Table 3).⁴ In categorical exposure analyses, the breast cancer SIR in the highest exposure quintile was elevated, albeit not significantly, when compared to the external referent group (SIR 1.27, 95% CI 0.94–1.69) but showed a statistical-

___________________

² This sentence was changed after release of the report to clarify TCEQ’s conclusion on the relationship between ethylene oxide and lymphoid cancer.

³ This sentence was changed after release of the report to clarify TCEQ’s conclusion on the relationship between ethylene oxide exposure and breast cancer.

⁴ This sentence was changed after release of the report to correct the terminology.

ly significant trend across exposure quintiles (SIRs of 0.77, 0.77, 0.94, 0.83, and 1.27) on a linear (p = 0.002) or log scale (p = 0.05; Steenland et al. 2003, Table 3). Trend parameters were not reported in the study; only the p-values were reported. It is also worth noting that for both breast cancer mortality (Steenland et al. 2004) and incidence (Steenland et al. 2003), rates among the workers were below national values for women without attributable exposure, while rates for the high-exposure group were above the national average. Furthermore, the exposure-response trends provide support for an exposure effect.

In addition to analyses using an external comparison, internal exposure-response analyses using a nested case–control design were conducted for the entire cohort (n = 7,576) and for the subcohort with interviews (n = 5,139). Odds ratios (ORs) for the entire cohort using Cox regression showed a statistically elevated breast cancer incidence rate for the highest exposure quintile versus unexposed, given a 15-year lag (OR 1.74, 95% CI 1.16–2.65; Steenland et al. 2003, Table 4). There was also a positive effect in continuous exposure models of cumulative exposure (coefficient [standard deviation]: 0.0000054 [0.0000035]; p = 0.12) and log cumulative exposure (coefficient [standard deviation]: 0.037 [0.019]; p = 0.05), with a 15-year lag in Cox regression models for the entire cohort (Steenland et al. 2003, Table 4). Among women with interview data, associations were even stronger. The advantage of using the subcohort with interview data is that analyses were controlled for individual-level risk factors including parity and family history of breast cancer. Cox regression analyses showed that risks were elevated in the highest cumulative exposure categories with ORs (95% CI) over increasing exposure categories of 1.00 (lagged out), 1.06 (0.66–1.71), 0.99 (0.61–1.60), 1.24 (0.76–2.00), 1.42 (0.88–2.29), and 1.87 (1.12–3.10), with a 15-year lag (Steenland et al. 2003, Table 5). Significant associations were also observed in the subcohort between breast cancer incidence and continuous cumulative exposure (p = 0.02) and log cumulative exposure (p = 0.03), with a 15-year lag (Steenland et al. 2003, Table 5).

Another consideration for the subcohort analyses is that several cancer cases were identified only by interview. Two hundred thirty-three people with a history of breast cancer or their next of kin had interviews. Medical records or cancer registry confirmation of breast cancer were found for 189 of these, and 44 cases were identified by interview alone. It is likely that if interviews were available for the complete cohort, then more cases would have been identified. The external comparison analyses are also likely biased by further under-ascertainment of breast cancer cases, as incidence data were not linked for all states. TCEQ did not rely on results from the subcohort with interview data; rather, it relied solely on external comparison analyses.

In the updated NIOSH mortality study, Steenland and colleagues (2004) examined cancer deaths in a cohort of 18,235 ethylene oxide sterilization facility workers. There were 103 breast cancer deaths, with no excess in breast cancer mortality among all workers compared to the general population (standardized mortality ratio [SMR] 0.99, 95% CI 0.84–1.17; Steenland et al. 2004, Table 1). However, when evaluating cumulative exposure, both external and internal

comparisons found a statistically significant association between the highest exposure level and breast cancer mortality. In the highest quartile of exposure, when compared to the external referent group, there is an indication of excess risk for breast cancer among females using a 20-year lag (SMR 2.07, 95% CI 1.10–3.54; Steenland et al. 2004, Table 5). Results for internal Cox regression analyses for breast cancer and categorical lagged data (20-year lag) showed an increased rate in the highest quartile (OR 3.13, 95% CI 1.42–6.92; Steenland et al. 2004, Table 8). ORs were elevated in all exposure categories (1.00, 1.76 [0.91–3.43], 1.77 [0.88–3.56], 1.97 [0.94–4.06], and 3.13 [1.42–6.92]). The model likelihood was 8.69, and the p-value for the exposure-response trend across categories was p = 0.07. When modeled continuously, the log of cumulative exposure with a 20-year lag was significantly associated with breast cancer mortality (coefficient [standard deviation] = 0.084 [0.035], model likelihood = 5.69, p = 0.01). These results support the findings of the incidence study (see Steenland et al. 2003).

Healthy Worker Effect

There are two distinct aspects of HWE—Healthy Hire Effect and the Healthy Worker Survivor Effect (HWSE; Eisen et al. 2006). Both aspects lead to underestimation of the health effects caused by long-term occupational exposures. It is critical to distinguish between the two aspects because the methods needed to avoid bias due to each aspect differ:

• The Healthy Hire Effect occurs because, on average, healthier members of the general population are hired into the workforce. This is a problem for studies that compare workers to the general population (e.g., for SMRs and SIRs). The most straightforward way to avoid this bias is to compare high- to low-exposed workers in an internal analysis.
• HWSE occurs because healthier workers tend to stay at work longer, resulting in an active worker cohort of survivors. HWSE is a problem for studies comparing high-exposed workers to low-exposed workers. In contrast with the Healthy Hire Effect, there is an extensive literature on how to avoid bias due to HWSE. The state-of-the-art approach acknowledges the possibility that any internal analysis of a cohort study of a long-term exposure and chronic health outcome is likely to be impacted by HWSE. Estimates of association may be biased toward the null due to HWSE because healthier workers who stay at work longer often have greater cumulative exposure compared to those who leave the workplace earlier. Over the past 20 years, appropriate methods needed to address HWSE have become more developed (Buckley et al. 2015). Because TCEQ does not have access to the original data, these methods cannot be applied in its assessment.

In an occupational study, an elevated SMR indicates that exposures at work are associated with an increased risk of death; however, since workers are

healthier at hire than the general population of the same age, SMRs often indicate that workers are less likely to die than the general population, particularly from chronic heart and lung disease. An SMR less than 1.0 for all causes of death combined (not for specific causes of death) is often cited as evidence that there is a Healthy Hire Effect operating in a particular cohort study. The SMR does not capture how long individuals remain at work once hired, and so it is not impacted by HWSE.

TCEQ’s Misconceptions About HWE

TCEQ relied only on external (as opposed to internal) analyses for breast cancer because it concluded that HWE was not relevant for this outcome. This decision was based on two misconceptions about HWE, which the committee clarifies here: (1) The lack of a deficit in an SIR for female breast cancer in a general Norwegian worker study (Kirkeleit et al. 2013) is not evidence that HWE would not impact that specific outcome in a study of a particular hazard, and (2) an internal reference group comparison of workers is always preferred over an external reference group to reduce confounding due to the Healthy Hire Effect.

TCEQ relied on a paper by Kirkeleit and colleagues (2013) that purported to evaluate HWE for individual endpoints based on SMRs and SIRs by comparing rates between the general working population and the general population of working-aged adults in Norway. However, that paper has two major flaws. First, combining workers across different work environments obscures any hazard-specific risks and makes it difficult to interpret specific SIRs. Second, it conflates HWE with the Healthy Hire Effect and ignores the potential contribution of HWSE. The Healthy Hire Effect often causes SMRs and SIRs comparing workers to the general population to be underestimates of worker risk for all causes combined or all cancers combined. Sometimes the underestimate will appear as an actual deficit in the observed mortality rate relative to that which is expected—an SMR or SIR less than 1.0. However, the converse is not true; in the presence of a true hazard, an SMR greater than 1.0 may still be an underestimate of true risk due to the Healthy Hire Effect. By conflating the Healthy Hire Effect with HWE, ignoring HWSE, and over-interpreting specific SIRs, reliance on the paper by Kirkeleit and colleagues (2013) leads TCEQ to falsely conclude that HWE does not exist for female breast cancer.

Occupational epidemiologists assume that the Healthy Hire Effect exists in any worker study and therefore that the best way to avoid that potential bias is to conduct internal analyses whenever possible. Comparing risk between higher-exposed and lower-exposed workers is the optimal design for an etiologic study. In this way, both groups are equally likely to be healthy when they are hired, and there may be less variability in socioeconomic status between internal comparison groups versus the general population, so any risk differences observed between groups are likely due to differences in their specific work exposures. Moreover, the exposure-response information provided by internal analyses is critical to the development of recommended exposure limits. Finally, cohort studies with time-

varying measures of exposure offer the only viable way to address the second component of HWE, namely bias due to HWSE. Therefore, internal analyses are always preferred over external analyses to avoid bias due to the Healthy Hire Effect and to allow HWSE to be assessed if given access to original data, although the smaller size of the unexposed reference group will reduce power.

A reliance on SMRs and SIRs to examine breast cancer risk related to ethylene oxide inappropriately led TCEQ to ignore existing exposure-response evidence. TCEQ preferred external analyses in this assessment because the reference group is larger than that in internal analyses that rely on unexposed (or less-exposed) workers. Studies relying on a general population reference group have more statistical power but are biased by confounding and selection. In the face of a bias–variance tradeoff, bias is never preferred.

Lack of Attention to HWSE in the Study by Steenland and Colleagues

HWSE was not accounted for in the cohort study of female breast cancer and ethylene oxide by Steenland and colleagues (2003). Some results from Cox models suggest an elevated risk of breast cancer, OR = 1.74 (95% CI 1.16–2.65), in the highest quintile of cumulative ethylene oxide exposure, lagged 15 years (Steenland et al. 2003, Table 4). When restricted to the subcohort with interview data, parity was added to the model and the OR increased slightly to 1.87 (95% CI 1.12–3.10; Steenland et al. 2003, Table 5). It stands to reason that more cumulative exposure to a true hazard should predict higher risk. However, because of HWSE, this intuitively compelling difference in risk between longer- and shorter-term workers is not often observed in occupational studies. The OR in the highest category of cumulative exposure may underestimate risk because workers who “survive” longest at work are generally a healthier subset of those hired. At the time the study by Steenland and colleagues was published in 2003, new methods to evaluate and address HWSE had been published but were nascent.⁵

Characterization of Statistical Significance

The TCEQ DSD does not follow current best practice when evaluating and interpreting effect estimates and dose-/exposure-response data from the literature. TCEQ relied on statistical significance testing when interpreting effect estimates and dismissed clear evidence of exposure-response in categorical analyses, including as occurring with appropriate internal reference groups. Given the extensive deficiencies in statistical significance testing, current best practice is to consider the magnitude, direction, precision, and consistency of effect estimates, especially those derived from studies with relatively small sample or case sizes (Greenland et al. 2016; Lash et al. 2021; Savitz et al. 2024). These considerations are also standard practice in systematic review. Examples of TCEQ’s reliance on

___________________

⁵ This paragraph was changed after release of the report to more accurately reflect Steenland et al. (2003).

statistical significance testing include the following from the TCEQ DSD: (1) “consistent with the lack of a statistical difference as in Mikoczy et al. (i.e., SIR of 1.35 (0.54, 2.78)” (TCEQ 2020, p. 14) and (2) “Steenland et al. (2003) found no excess of breast cancer incidence among the cohort as a whole compared to the US population; only finding an increase in the highest exposure quintile in certain internal analyses; that is, categorical with exposure lagged 15 years for cumulative exposure and duration of exposure (see Tables 4 and 5 of Steenland et al. 2003) the RR [relative risk] for even the highest exposed group (>14,620 ppm-days) was not statistically increased (i.e., 1.27 (0.94, 1.69)” (TCEQ 2020, p. 25). As such, some interpretations of available epidemiologic data are inappropriate/incomplete in the hazard assessment section.

EXPERIMENTAL ANIMAL DATA TO INFORM HAZARD ASSESSMENT

TCEQ Guidelines states, “When relevant and sufficient human studies are not available, laboratory animal data are used to develop toxicity factors. Several factors are considered when selecting key animal studies. For example, the adverse effect observed in laboratory animals must be relevant in humans as discussed as part of an International Programme on Chemical Safety framework illustrated in Figure 3-1 (Boobis et al. 2006)” (TCEQ 2015, p. 46). The committee supports following this TCEQ guidance.

Per this guidance, TCEQ may need to reconsider animal data in the context of its reevaluation of breast cancer data for hazard assessment. The TCEQ DSD based its cancer assessment on human evidence; thus, a cursory evaluation of the relevant animal bioassay data was provided (see Section 3.1.2 “Summary of Animal Studies” in TCEQ [2020]).

MOA CONSIDERATIONS

Animal data also can be used to support a WOE evaluation of the MOA of a carcinogen (TCEQ 2015). TCEQ concluded that ethylene oxide most likely caused tumors via a mutagenic MOA, which can be applied to all tumor types (TCEQ 2020). TCEQ’s use of animal data to support this conclusion is provided in Section 3.2.2 “WOE for a Mutagenic MOA” in the TCEQ DSD (TCEQ 2020). Section 3.2.2 does not provide any details concerning how the scientific literature was searched and evaluated for study quality. TCEQ Guidelines discusses the hierarchy of evidence used by U.S. EPA for determining a mutagenic MOA (TCEQ 2015). For example, the highest level of evidence is provided from the detection of cancer-relevant oncogene/tumor suppressor gene mutations in the target tissue following chemical exposure. The TCEQ DSD would be strengthened if a similar approach was used to organize data tables summarizing the available evidence in support of a mutagenic MOA. In addition, in vitro evidence to support a mutagenic MOA has been largely ignored by TCEQ and should also be considered.

Although mutagenicity is widely accepted as one MOA, TCEQ did not consider alternatives (e.g., endocrine-mediated effects, cytotoxicity with regenerative cell proliferation, immune suppression, or epigenetic mechanisms) in its evaluation of the MOA. It is unclear to the committee how TCEQ drew the conclusion that other MOAs were critically considered and evaluated.⁶ TCEQ should include evaluated information from more recent authoritative ethylene oxide risk assessments, including Toxicological Profile for Ethylene Oxide (ATSDR 2022). Evidence and possible mechanisms for alternative MOAs should be included along with any search terms used. By conducting a systematic review of the evidence, these alternatives could be identified and evaluated.

FINDINGS AND RECOMMENDATIONS

Finding 2.1: TCEQ did not perform a systematic review of the evidence in support of the hazard assessment.

Recommendation 2.1 (Tier 1): The Texas Commission on Environmental Quality (TCEQ) should conduct a systematic review of all relevant endpoints for the hazard assessment. Multiple National Academies of Sciences, Engineering, and Medicine reports have provided best practices for the conduct of a systematic review that TCEQ could follow, such as Review of U.S. EPA’s ORD Staff Handbook for Developing IRIS Assessments: 2020 Version (2022) and Review of EPA’s 2022 Draft Formaldehyde Assessment (2023).

Finding 2.2: TCEQ inappropriately excluded human evidence lacking dose-response data that could have contributed to the hazard assessment for breast cancer. Exclusion of this evidence reduces confidence in its hazard assessment of breast cancer.

Recommendation 2.2 (Tier 1): The Texas Commission on Environmental Quality should evaluate all relevant human evidence for the hazard assessment of breast cancer, without exclusion of studies that did not present quantitative data that would be adequate for dose-response assessment.

Finding 2.3: Consideration of HWE was flawed, leading to the reliance on an inappropriate external reference group, which contributed to the exclusion of breast cancer as an endpoint.

Recommendation 2.3 (Tier 1): The Texas Commission on Environmental Quality should rely on the internal analyses from the breast cancer incidence study conducted by Steenland and colleagues and published in

___________________

⁶ This sentence was changed after release of the report to more accurately reflect the committee’s understanding of TCEQ’s approach to other MOAs.

Cancer Causes & Control in 2003 and interpret those results in light of the potential underestimation of risk due to the Healthy Worker Survivor Effect.

Finding 2.4: TCEQ relied exclusively on statistical significance testing and p-values when interpreting epidemiologic effect estimates, which contributed to misrepresentation of some of the evidence.

Recommendation 2.4 (Tier 1): The Texas Commission on Environmental Quality should follow current best statistical and epidemiological practices by considering the magnitude, direction, consistency, and precision of effect estimates, as well as considering evidence of dose-/exposure-response relationships in categorical analyses.

Finding 2.5: TCEQ predominantly relied on human epidemiology studies to evaluate breast cancer.

Recommendation 2.5 (Tier 2): The Texas Commission on Environmental Quality should reevaluate all relevant animal and mechanistic evidence for breast cancer in reaching conclusions about the evidence for this cancer endpoint.

Finding 2.6: TCEQ only considered mutagenicity as an MOA.

Recommendation 2.6 (Tier 2): The Texas Commission on Environmental Quality should consider alternative modes of action when completing its assessment.

These recommendations build off each other and should be synthesized during reevaluation of the hazards of ethylene oxide. For example, TCEQ should incorporate these recommendations along with other relevant recommendations, including but not limited to (1) evaluating evidence that uses an internal reference group to account for bias due to the Healthy Hire Effect; and (2) considering Bradford Hill and other criteria to evaluate the WOE, including evaluating evidence of dose-/exposure-response relationships, among others.