Ethylene Oxide (EtO) Epidemiology and Cancer Risk

RoseyFields evaluates the human epidemiological evidence for ethylene oxide as follows, drawing on the peer reviewed occupational cohort, case-control, and biomonitoring literature.

A note on sourcing: This summary cites the underlying peer-reviewed epidemiological and biomonitoring studies rather than the regulatory compilation (EPA IRIS carcinogenicity assessment) in which many of these studies were originally identified and critically reviewed. Study details (cohort composition, exposure assessment, effect estimates) are drawn from that publicly available compilation and have not been independently re-verified by RoseyFields against the original journal articles.

2. Overview of Study Populations and Data Sources

The human evidence for ethylene oxide (EtO) comes almost entirely from occupational cohort and case-control studies, supplemented by a body of biomonitoring (cytogenetic) research conducted on subsets of exposed workers. No general-population biomonitoring program comparable to NHANES has characterized background EtO body burden in the way that exists for many metals or persistent organics, and no birth-cohort or community contamination site studies of EtO were identified in the reviewed literature; this reflects EtO’s occupational exposure profile (industrial gas used in chemical manufacturing and medical/spice sterilization) rather than a gap in study design choices.

Two structurally distinct categories of occupational cohort populate this literature, and the distinction matters for interpreting the evidence: (1) chemical-manufacturing cohorts, employed at facilities that produce EtO or use it as a feedstock (e.g., the Union Carbide Corporation [UCC] cohort in West Virginia; the West German multi-plant cohort of Kiesselbach et al., 1990; the British cohort followed by Gardner et al., 1989 and Coggon et al., 2004), and (2) sterilization-industry cohorts, employed at facilities using EtO to sterilize medical equipment, hospital supplies, or spices (e.g., the NIOSH industry-wide cohort of Steenland et al., 1991, 2003, 2004; the Swedish sterilant-worker cohorts of Hagmar et al., 1991, 1995 and Mikoczy et al., 2011; the U.S. sterilization-plant cohort of Norman et al., 1995). This distinction is important because chemical-manufacturing settings, particularly those using the older chlorohydrin production process, involve concurrent exposure to other chemicals (notably ethylene dichloride, a chlorohydrin-process byproduct), whereas sterilization work involves comparatively pure EtO exposure with few other occupational carcinogens present. Findings from sterilization cohorts therefore carry less confounding uncertainty for isolating an EtO-specific effect, even though they are not immune to other limitations (small size, short follow-up, crude exposure assessment) that vary considerably from study to study.

Cytogenetic biomonitoring studies form a second, mechanistically oriented data source: rather than tracking cancer incidence or mortality, these studies measure chromosome damage (sister chromatid exchanges [SCEs], micronuclei, chromosomal aberrations) or gene mutation frequencies (e.g., HPRT locus mutations) directly in the peripheral blood lymphocytes of EtO-exposed workers, generally as a function of estimated cumulative or recent exposure. This body of evidence does not by itself establish a cancer risk but provides a mechanistic bridge between EtO’s well-established genotoxicity in experimental systems and the cancer findings observed in the cohort studies (see Sections 4 and 6).

Using multiple, structurally diverse populations — chemical-manufacturing versus sterilization cohorts, mortality versus incidence study designs, U.S., British, Swedish, German, Italian, and Hungarian populations, and cancer endpoint studies alongside independent cytogenetic biomonitoring studies — strengthens causal inference in several specific ways relevant to the EtO database. First, replication of a positive association for lymphohematopoietic cancer across cohorts with different co-exposure profiles (comparatively pure EtO exposure in sterilization workers versus mixed chemical exposure in manufacturing workers) makes it less plausible that a shared confounder specific to one

RoseyFields — Ethylene Oxide Epidemiology Summary

industry explains the pattern. Second, replication across countries and healthcare/regulatory systems reduces the likelihood that a single national cancer-registry artifact or surveillance bias is responsible. Third, convergence between the cancer-endpoint cohort studies and the mechanistically independent cytogenetic biomonitoring literature — which measures a different kind of outcome (chromosome damage rather than diagnosed cancer) using different populations and different investigators — provides a form of triangulation that is considerably more persuasive than either evidence stream alone.

3. Overview of Health Outcomes Investigated in Humans, by Body System

The following organizes the human evidence by body system and evidence tier. As detailed further in Sections 4–6, the strongest and most consistently modeled evidence concerns lymphohematopoietic cancers and female breast cancer; several other cancer sites have been investigated with inconsistent or inadequate evidence; and genotoxic/cytogenetic endpoints, while not diseases in themselves, provide strong and consistent supporting evidence of a biologically plausible pathway to the cancer outcomes.

3.1 Hematologic and Lymphatic System (Strong, Consistently Modeled Evidence)

Cancers of the lymphohematopoietic system — principally leukemia (including lymphocytic leukemia), non-Hodgkin lymphoma (NHL), Hodgkin lymphoma, and multiple myeloma — constitute the largest and most consistently positive body of cancer evidence for EtO. Of 11 independent cohorts with adequate information to calculate relative risk estimates, approximately 9 reported an increased risk of lymphohematopoietic cancer or a subcategory thereof, and the single largest and highest-quality cohort (the NIOSH industry-wide sterilization-worker study) demonstrated a statistically significant, quantitatively modeled exposure-response trend. This body of evidence is discussed in detail in Section 4.1.

3.2 Breast (Strong, Consistently Modeled Evidence in Females)

Female breast cancer is the second major outcome with strong, quantitatively modeled evidence, again anchored by the NIOSH cohort, which is unusual among occupational cancer cohorts in including a sufficiently large number of female workers (7,576) to support a dedicated incidence study with internal exposure-response modeling. Supporting evidence, though based on much smaller case counts, comes from an independent Swedish sterilant-worker cohort and from Hungarian hospital-worker studies. This evidence is discussed in detail in Section 4.2.

3.3 Genotoxic and Cytogenetic Endpoints (Strong, Consistent Mechanistic Evidence)

A substantial and consistent body of biomonitoring literature demonstrates that occupationally EtO-exposed workers show increased frequencies of SCEs, micronuclei, and chromosomal aberrations in peripheral blood lymphocytes, generally in proportion to exposure intensity, with some studies also demonstrating increased micronuclei in bone marrow and nasal mucosal cells and increased HPRT-locus gene mutation frequencies. This evidence, while not a disease outcome per se, is treated as a strong line of supporting evidence for carcinogenic hazard because chromosomal aberrations and micronuclei (though not SCEs specifically) have themselves been linked to increased cancer risk in independent large prospective general-population studies. This evidence is discussed in detail in Section 4.3.

3.4 Gastrointestinal System — Stomach and Pancreatic Cancer (Suggestive but Inconsistent Evidence)

Elevated stomach cancer mortality was reported in the pooled Swedish cohorts of Hogstedt et al. (1986) (SMR = 5.5) and elevated (nonsignificant) stomach, esophageal, and lung cancer SMRs were reported by Kiesselbach et al. (1990), but these findings were not replicated with statistical significance in other cohorts, and the studies reporting them had

RoseyFields — Ethylene Oxide Epidemiology Summary

substantial limitations (small numbers, external comparisons, multiple co-exposures, no latency analysis). Pancreatic cancer showed a statistically significant excess in one small, heavily confounded chlorohydrin-worker subcohort (Benson and Teta, 1993) and a nonsignificant excess in the parent UCC cohort (Greenberg et al., 1990) and in a chlorohydrin-specific Dow Chemical cohort (Olsen et al., 1997), but was not independently corroborated in the higher quality NIOSH sterilization-worker cohort. Overall, the EPA’s own weight-of-evidence conclusion characterizes the epidemiological evidence for these other cancer types, including stomach and pancreatic cancer, as inadequate.

3.5 Reproductive and Other Systems (Null or Inadequate Evidence)

No dedicated, adequately powered epidemiological studies of EtO and reproductive, developmental, neurological, cardiovascular, hepatic, or renal outcomes in occupationally exposed human populations were identified in the reviewed literature; the human epidemiological database for EtO is essentially confined to cancer mortality/incidence and the genotoxic biomarker literature described above. This is a data gap rather than a body of null evidence for these other systems (see Section 8).

4. Key Studies, Exposure Metrics, Effect Sizes, and Consistency for Strong-Evidence Outcomes

4.1 Lymphohematopoietic Cancer

The pivotal study is the NIOSH industry-wide cohort of 18,254 workers (45% male, 55% female) at 14 U.S. plants where EtO was used, primarily for sterilization of medical supplies and spices (Steenland et al., 1991; Stayner et al., 1993; Steenland et al., 2004). Individual cumulative exposure estimates (in ppm-days) were derived for workers at 13 of the 14 plants using a validated regression model built from 2,700 personal-breathing-zone industrial hygiene measurements, which explained 85% of the variance in an independent test dataset (Greife et al., 1988; Hornung et al., 1994). In the extended follow-up through 1998 (mean follow-up 26.8 years), the overall external SMR for lymphohematopoietic cancer was unremarkable (SMR = 1.00, 79 cases), but internal exposure-response analyses — considered the more informative comparison because they control for the healthy-worker effect — showed a statistically significant positive trend with increasing log cumulative exposure in males for both all hematopoietic cancer (p = 0.02) and the lymphoid subcategory (NHL, lymphocytic leukemia, and myeloma; p = 0.02), with a 15- year exposure lag. In categorical analysis, workers in the highest cumulative-exposure quartile showed an odds ratio of 3.42 (95% CI: 1.09–10.73) for all hematopoietic cancer and 3.76 (95% CI: 1.03–13.64) for lymphoid cancer, relative to the lowest quartile (Steenland et al., 2004).

Several other cohorts provide supportive, though individually weaker, evidence. The pooled Swedish cohort of sterilizer and production workers (Hogstedt et al., 1986; Hogstedt, 1988) reported a statistically significant SMR of 9.2 (95% CI: 3.7–19) for leukemia based on 7 deaths. A statistically significant excess of lymphohematopoietic cancer (SMR = 2.9, 95% CI: 1.3–5.8) was found in the small chlorohydrin-exposed UCC subcohort (Benson and Teta, 1993), and a statistically significant excess of hematopoietic cancer (SMR = 7.1) was reported in an Italian chemical-worker cohort (Bisanti et al., 1993), although both of these studies had substantial exposure-characterization and confounding limitations. A nested case-control study in a Belgian chemical plant found a significant association between a cluster of Hodgkin lymphoma cases and both benzene and EtO exposure (Swaen et al., 1996). By contrast, updated follow up of the UCC cohort through 2003 (Swaen et al., 2009) and reanalysis by Valdez-Flores et al. (2010) found no statistically significant exposure-response trends, and the British cohort update by Coggon et al. (2004) found only small, nonsignificant increases; both of these negative or null studies, however, relied on comparatively small case counts and, in the UCC cohort’s case, a considerably cruder exposure assessment than the NIOSH regression-model based approach.

4.2 Female Breast Cancer

RoseyFields — Ethylene Oxide Epidemiology Summary

The dedicated NIOSH breast cancer incidence study identified 7,576 women from the parent cohort with at least 1 year of sterilization-facility employment; incidence was ascertained through interviews, death certificates, and cancer registries, with 319 incident cases identified in the full cohort (Steenland et al., 2003). A statistically significant exposure-response trend was observed for standardized incidence ratio across cumulative exposure quintiles using a 15-year lag (p = 0.002), and in Cox regression analysis with categorical exposure, the top cumulative-exposure quintile showed a statistically significant odds ratio of 1.74 (95% CI: 1.16–2.65). In the subcohort with detailed interview data (233 incident cases), which allowed adjustment for established breast cancer risk factors (parity and first-degree relative breast cancer history), the top exposure quintile odds ratio was 1.87 (95% CI: 1.12–3.10), indicating the association was not substantially attenuated by control for these established risk factors. The companion mortality analysis of the same cohort (Steenland et al., 2004) found a significant excess in the highest cumulative-exposure quartile using a 20-year lag (SMR = 2.07, 95% CI: 1.10–3.54, 13 deaths) and a significant positive internal trend (p = 0.01).

Independent replication comes from a 2011 extended follow-up (an additional 16 years) of a Swedish sterilant-worker cohort originally studied by Hagmar et al. (1991, 1995): although overall standardized incidence ratios were not significantly elevated, internal analysis found statistically significant increases in incidence rate ratios in the two highest cumulative-exposure quartiles relative to lower-exposed workers, despite the comparator group itself having some EtO exposure (Mikoczy et al., 2011). Two much smaller studies are also directionally supportive: a U.S. sterilization-plant cohort found a significant excess incidence of breast cancer (SIR = 2.6, 12 cases) among female workers (Norman et al., 1995), and a small Hungarian hospital-worker cohort found 3 observed breast cancer deaths against fewer than 4.4 expected total cancer deaths of any type, a pattern the source review characterizes as indicative of an increased breast cancer risk given Hungarian national background rates (Kardos et al., 2003); this finding is further contextualized by a separately reported cluster of breast cancer cases among nurses at the same hospital (Major et al., 2001). Only one identified study — the British cohort update by Coggon et al. (2004), which had only 11 breast cancer deaths — failed to find any elevated risk.

4.3 Genotoxic and Cytogenetic Biomarkers

Human genotoxicity monitoring generally uses cumulative or intensity-based EtO exposure estimates (personal air monitoring, job-exposure matrices, or hemoglobin adduct measurements) as the exposure metric and cytogenetic endpoints in peripheral blood lymphocytes as the outcome. Multiple independent studies report increased SCE frequency in occupationally exposed workers, with a positive relationship to exposure intensity most evident in the more highly exposed groups (Sarto et al., 1987; Sarto et al., 1991; Tates et al., 1991). Increased micronucleus frequency has been reported in lymphocytes (Ribeiro et al., 1994; Tates et al., 1991), in exfoliated nasal mucosal cells (Sarto et al., 1990), and in bone marrow cells (Högstedt et al., 1983), although this endpoint appears somewhat less sensitive than SCE frequency and data from the most highly exposed workers are limited. Chromosomal aberrations — mechanistically more directly linked to mutagenesis than SCEs — have also been reported in multiple worker studies, again most consistently in the highest-exposure groups (Ribeiro et al., 1994; Tates et al., 1991; Sarto et al., 1987). The genotoxic and mutagenic properties of EtO in humans have been documented since 1979 (Hogstedt et al., 1979), and dose-dependent DNA strand-break induction (via the comet assay) has been demonstrated directly in human breast epithelial cells, lymphoblasts, and lymphocytes at concentrations as low as 20 µM, with breast epithelial cells showing particular sensitivity relative to keratinocytes or cervical epithelial cells (Adám et al., 2005) — a finding of direct mechanistic relevance to the breast cancer epidemiology described above.

The biological significance of these cytogenetic endpoints for cancer risk is independently supported by large prospective general-population studies unrelated to EtO: increased micronucleus frequency has been associated with increased overall cancer risk (Bonassi et al., 2007), and increased chromosomal aberration frequency has been associated with increased cancer risk in multiple independent cohorts (Boffetta et al., 2007; Rossner et al., 2005). This external validation is part of why the EtO cytogenetic literature is treated as meaningful supporting evidence for the cancer hazard characterization rather than as an isolated biomarker finding of uncertain relevance.

RoseyFields — Ethylene Oxide Epidemiology Summary

5. Study Quality Considerations

5.1 Exposure Misclassification

Exposure assessment quality varies enormously across the EtO cohort literature and is the single largest driver of differential confidence between studies. At one end, the NIOSH cohort exposure assessment used a validated regression model incorporating plant- and job-specific variables, built and tested on separate industrial-hygiene datasets, and explaining 85% of exposure variance in the independent test set (Greife et al., 1988; Hornung et al., 1994). At the other end, several chemical-manufacturing cohorts relied on qualitative, department-level exposure categories (e.g., “low,” “intermediate,” “high” potential exposure) without individual worker-level measurements and, in some cases, without any description of how the categories were derived (Greenberg et al., 1990; Kiesselbach et al., 1990). Non-differential exposure misclassification of this kind generally biases effect estimates toward the null, meaning that studies with cruder exposure assessment (e.g., the UCC cohort updates) are less able to detect a true exposure-response relationship even if one exists, and their null findings should be weighted accordingly, as the source review does explicitly (contrasting the crude UCC exposure assessment with the NIOSH regression-model-based approach when explaining the discordant Valdez-Flores et al. (2010) and Steenland et al. (2004) findings).

5.2 Confounding by Co-Exposure

Confounding by co-exposure to other chemicals is a first-order concern in the chemical-manufacturing cohorts, particularly those using the older chlorohydrin EtO production process, which generates ethylene dichloride and other byproducts (IARC, 1994b); the Benson and Teta (1993) chlorohydrin-worker findings were explicitly interpreted by the study authors as possibly attributable to ethylene dichloride rather than EtO itself. By contrast, the sterilization industry cohorts, and the NIOSH cohort specifically, involve comparatively pure EtO exposure with little evidence of confounding occupational carcinogen co-exposure (Steenland et al., 1991), which is a principal reason the source review treats the NIOSH findings as the most interpretable evidence of an EtO-specific effect.

5.3 Statistical Power and Small Numbers

Most cohorts other than NIOSH are small, with correspondingly small numbers of the relatively rare cancer outcomes under study; several studies with entirely null or non-significant findings (e.g., Coggon et al., 2004, with only 11 breast cancer deaths and 17 lymphohematopoietic cancer deaths; Norman et al., 1995, with a single lymphohematopoietic case) simply lack the statistical power to detect a moderate relative risk even if a true effect of the magnitude observed in the NIOSH cohort is present. This is a recurring theme in the source review’s assessment: studies that fail to demonstrate an association are generally characterized as lacking the study-design strengths (size, exposure quality, follow-up) needed to give confidence in a true null finding, rather than as strong evidence against an association.

5.4 Reverse Causation and Healthy Worker Effect

The healthy-worker effect — the tendency for employed cohorts to show lower overall mortality than the general population because sicker individuals are less likely to be employed — is a structural feature of essentially all the EtO occupational cohorts and is the primary reason internal comparisons (workers at different exposure levels within the same cohort) are considered more informative than external comparisons (workers versus the general population); several of the null or borderline overall SMRs in the EtO literature (e.g., the NIOSH cohort’s overall cancer SMR of 0.98) likely reflect this effect diluting a true, exposure-related increase that is only visible in internal exposure response analysis. Reverse causation is a less prominent concern for the cancer mortality/incidence outcomes given the long induction periods used (15–20-year lags) and the clear temporal sequence of occupational exposure preceding disease diagnosis in all reviewed studies.

RoseyFields — Ethylene Oxide Epidemiology Summary

5.5 Study-Specific Limitations Summary

Beyond the general categories above, several study-specific issues recur: lack of latency or induction-period analysis in the earliest studies (e.g., Hogstedt et al., 1986); reliance on mortality rather than incidence data, which undercounts non-fatal cancers, particularly breast cancer, which has a comparatively favorable prognosis (addressed directly by the dedicated NIOSH incidence studies); short follow-up in several sterilization-worker cohorts, limiting the ability to observe cancers with long latency; and, in the Wong and Trent (1993) reanalysis of the NIOSH cohort, an unexplained addition of subjects and expected deaths that the source review flags as making that particular reanalysis of questionable reliability.

6. Weight-of-Evidence Conclusions from Agency and Expert Review

EPA’s National Center for Environmental Assessment, in its 2016 IRIS carcinogenicity assessment, characterized EtO as “carcinogenic to humans” by the inhalation route under the Agency’s 2005 Guidelines for Carcinogen Risk Assessment, applying the Bradford Hill (1965) considerations for causal inference explicitly to the epidemiological database. The assessment concluded that of the seven relevant Hill considerations, temporality (satisfied because exposure preceded disease in every study design used), coherence, biological plausibility, and analogy (satisfied by EtO’s well-established mutagenicity, its epoxide chemistry, and its direct carcinogenicity in laboratory rodents) were “clearly satisfied.” Consistency and biological gradient (dose-response) were judged satisfied “to varying degrees” — approximately 9 of 11 informative cohorts showed a positive association for lymphohematopoietic cancer, and the two largest, highest-quality cohorts (NIOSH for both cancer types; the Swedish sterilant cohort for breast cancer) showed statistically significant internal exposure-response trends — while strength of association was characterized as the most limited criterion, given that most relative risk estimates, while statistically significant in the best studies, were of only modest magnitude (on the order of 3.0 for lymphoid cancer mortality and 1.9 for breast cancer incidence in the highest exposure groups).

The overall epidemiological weight-of-evidence conclusion reached was that the evidence for a causal association between EtO exposure and both lymphohematopoietic cancer and female breast cancer is “strong but less than conclusive” on its own. This human evidence was combined with independent lines of evidence — extensive rodent carcinogenicity (lymphohematopoietic tumors in both rats and mice, mammary carcinoma in female mice), unequivocal genotoxicity and mutagenicity across experimental systems, and a mutagenic mode of action supported by demonstrated chromosome damage in exposed human workers — to reach an overall hazard characterization of “carcinogenic to humans,” a category the Agency’s guidelines reserve for situations where epidemiological evidence, even if short of independently conclusive, is substantially strengthened by convergent mechanistic and animal evidence. The assessment characterizes its overall confidence in this hazard characterization as high, while separately noting that evidence for other cancer types investigated (e.g., stomach and pancreatic cancer) remains inadequate to support a similar conclusion.

This EtO determination sits within a broader international consensus: EtO’s classification as a human carcinogen by expert bodies applying similar weight-of-evidence frameworks (e.g., IARC) reflects substantially the same underlying epidemiological and mechanistic database summarized here, particularly the NIOSH cohort data and the extensive genotoxicity literature.

7. Use of Epidemiological Evidence, Alongside Toxicology, in Litigation

In toxic tort and product liability litigation, epidemiological evidence of the kind summarized above is generally used to support two analytically distinct causation inquiries, and the EtO database illustrates the considerations relevant to each.

RoseyFields — Ethylene Oxide Epidemiology Summary

7.1 General Causation

General causation asks whether the substance is capable of causing the type of injury at issue in the population at large, at exposure levels of the kind experienced by the plaintiff class. The EtO epidemiological literature summarized in Sections 3–6 — particularly the NIOSH cohort’s statistically significant internal exposure-response findings for both lymphohematopoietic cancer and breast cancer, corroborated by independent cohorts, by the extensive rodent carcinogenicity data, and by the mutagenic mode-of-action evidence — is the type of evidence typically offered to establish general causation for these two cancer categories specifically. Agency weight-of-evidence determinations such as EPA’s “carcinogenic to humans” characterization (Section 6) are frequently cited in this context, though courts vary in how much independent weight they give such determinations relative to requiring an independent Bradford Hill or comparable analysis from a testifying expert; the underlying primary studies, exposure-response models, and effect-size data (rather than the agency conclusion alone) are generally what withstand admissibility scrutiny under evidentiary standards such as Daubert.

For the cancer types with inadequate or inconsistent human evidence (Section 3.4–3.5 — e.g., stomach or pancreatic cancer, or any endpoint outside the cancer/genotoxicity literature entirely), the current epidemiological database does not, on its own, support a general causation opinion, regardless of what toxicological or mechanistic plausibility might otherwise be argued; the absence of adequate human evidence for these outcomes is a materially different evidentiary posture than the strong-but-not-conclusive evidence available for lymphohematopoietic and breast cancer.

7.2 Specific Causation

Specific causation asks whether the substance caused a particular plaintiff’s injury, and generally requires integrating the general-causation epidemiological framework with case-specific facts: the individual’s estimated cumulative or intensity-weighted EtO exposure (ideally reconstructed using an exposure-assessment approach comparable in rigor to the NIOSH job-exposure-matrix methodology described in Section 4.1, rather than qualitative or self-reported exposure characterization), the latency between first exposure and diagnosis relative to the induction periods (15–20- year lags) used in the key studies, the specific cancer subtype diagnosed (given that the strongest evidence is for lymphoid rather than myeloid leukemia subtypes, and for invasive rather than in situ breast cancer), the presence or absence of competing/alternative causes and established non-occupational risk factors (particularly for breast cancer, where parity and family history were shown to be independently predictive in the NIOSH subcohort analysis; see Section 4.2), and differential diagnosis of other established causes for the individual’s cancer. Because most of the epidemiological relative risk estimates for EtO are of modest magnitude (typically in the range of 1.5–3.5 in the highest-exposure groups, see Section 6), specific causation opinions in individual cases often require careful attention to where a given plaintiff’s estimated exposure falls within the study’s exposure-response curve, since relative risk near or below 2.0 raises well-recognized “more likely than not” attribution questions distinct from the general causation determination itself.

7.3 Integration with Toxicological Evidence

The epidemiological evidence summarized here is ordinarily paired with toxicological and mechanistic evidence — EtO’s demonstrated genotoxicity and mutagenicity across experimental systems, its direct DNA-alkylating chemistry, its rodent carcinogenicity at the same tumor sites showing site-concordance with the human lymphohematopoietic and breast findings, and the human cytogenetic biomarker data described in Section 4.3 — to address biological plausibility and coherence, two of the Bradford Hill considerations that epidemiological data alone cannot fully establish. This integration is precisely the analytical structure used in the EPA’s own weight-of-evidence characterization (Section 6) and is generally the same structure that a testifying expert would use to move from an agency hazard characterization to an individualized causation opinion, with the important caveat that an agency’s population-level, upper-bound regulatory risk estimate (designed to be protective under uncertainty) is not the same

RoseyFields — Ethylene Oxide Epidemiology Summary statistical quantity as a case-specific probability-of-causation estimate, and the two should not be conflated without an explicit, transparent analytical bridge between them.

8. Data Gaps

• No epidemiological studies of EtO and reproductive, developmental, neurological, cardiovascular, hepatic, or renal outcomes in humans were identified; the human evidentiary base is confined to cancer and genotoxicity endpoints.
• No general-population biomonitoring data (e.g., an NHANES-type survey) characterize background EtO exposure or its association with health outcomes outside occupational settings, limiting the ability to extrapolate the occupational cohort findings to community or environmental exposure scenarios without model-based extrapolation of the kind performed in the EPA’s unit risk derivation.
• Point-mutation studies in EtO-exposed humans remain few and comparatively insensitive; the evidence for point mutations, as opposed to chromosomal-level damage, in exposed workers is correspondingly more limited than the SCE, micronucleus, and chromosomal aberration literature.
• There are no chemical-specific data on relative susceptibility of children or the elderly to EtO carcinogenicity; the EPA’s assumption of increased early-life susceptibility is a default, mode-of-action based policy assumption rather than one derived from EtO-specific human or animal susceptibility data.
• Exposure-response modeling for lymphohematopoietic cancer relies on a comparatively small number of cases (53 lymphoid cancer deaths in the NIOSH study), and the source review itself notes the resulting risk estimates are more sensitive to modeling choices (lag time, knot placement) than the breast cancer estimates, which are based on a larger case count.

References