Benzene Epidemiology and Leukemia and Dose Response

Chemical-Specific Epidemiology Page — Benzene

Benzene Epidemiology

RoseyFields evaluates the human epidemiological evidence for Benzene as follows. This page summarizes the study populations and data sources underlying the human evidence base, organizes the full range of investigated health outcomes by strength of evidence, discusses study quality considerations relevant to weighing that evidence, and explains how RoseyFields integrates this evidence with animal toxicology to support causation opinions in litigation.

2. Overview of Study Populations and Data Sources

The human evidence base for benzene draws on several structurally distinct types of study populations, each contributing a different strength to the overall causal picture.

Occupational cohorts

The strongest and most extensively studied populations are occupational cohorts with sustained, relatively high, and often individually reconstructed benzene exposure. These include the Pliofilm (rubber hydrochloride) manufacturing cohort in Ohio, followed for mortality since 1940 and repeatedly reassessed as exposure reconstruction methods have improved (Rinsky et al., 1987, 2002; Rhomberg et al., 2016); the joint National Cancer Institute/Chinese Academy of Preventive Medicine (NCI/CAPM) cohort of over 74,000 benzene-exposed and 35,000 unexposed workers across 12 Chinese cities (Hayes et al., 1997; Linet et al., 2015); the Dow Chemical cohort in Michigan (Bloemen et al., 2004; Collins et al., 2015); pooled petroleum distribution worker cohorts from Australia, Canada, and the United Kingdom (Schnatter et al., 2012); and Norwegian offshore oil industry workers (Kirkeleit et al., 2008; Stenehjem et al., 2015). These cohorts generally offer individual or job-title-level quantitative exposure estimates (in ppm or ppm-years), long follow up, and, in the more recent analyses, careful diagnostic subclassification of hematological malignancies by trained hematopathologists.

Community and contamination-site cohorts

A second category involves populations exposed via contaminated environmental media rather than direct occupational handling of benzene, such as residents near hazardous waste sites, populations served by contaminated drinking water supplies (e.g., the Camp Lejeune Marine Corps Base cohort; Ruckart et al., 2014), and children living near roadways, petrol stations, or industrial emitters whose ambient benzene exposure has been estimated using dispersion modeling (Raaschou-Nielsen et al., 2001). These studies typically involve lower, more chronic, and less certain exposure levels than occupational cohorts but are directly relevant to the general population and to residential exposure litigation.

General-population biomonitoring and cohort studies

A third category uses general-population samples with biomarkers of benzene exposure — most commonly urinary trans,trans-muconic acid (t,t-MA) or S-phenylmercapturic acid (SPMA) — rather than direct workplace measurements. Examples include the Shanghai Women’s Health Study, a large population-based cohort with a validated job-exposure matrix (Bassig et al., 2015), and cross-sectional analyses of U.S. National Health and Nutrition Examination Survey (NHANES) participants examining associations between urinary benzene metabolites and outcomes such as hearing loss (Benedict et al., 2024) or insulin resistance (Choi et al., 2014). These studies extend the exposure range studied to levels much closer to typical ambient and non occupational exposure but face greater uncertainty because the biomarkers used are not always specific to benzene.

Using multiple, structurally different populations strengthens causal inference in several ways. Occupational cohorts provide the statistical power and exposure range needed to detect and quantify dose-response relationships; community and contamination-site cohorts test whether effects observed at high occupational exposures extend to more chronic, lower-level, and environmentally relevant exposures; and general population biomonitoring studies extend the observed exposure-response curve toward the low end relevant to ambient exposure while capturing effect modification by sex, age, and co-morbidities not well represented in largely male industrial cohorts. Because each study design has different, largely non-overlapping sources of potential bias (healthy-worker survivorship in occupational cohorts; exposure misclassification in ambient/biomarker studies; recall bias in case-control designs), consistent findings across these different designs and populations substantially strengthen the case for a causal, rather than confounded or artifactual, association.

3. Health Outcomes by Body System and Strength of Evidence

The table below summarizes the full range of health outcomes investigated in the human benzene literature, organized by RoseyFields’ assessment of evidentiary strength. Sections 4 discusses the strong-evidence outcomes in greater depth.

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

Acute myeloid leukemia / acute nonlymphocytic leukemia

AML (historically reported as ANLL) is the health outcome with the strongest, most consistent, and most thoroughly quantitatively modeled association with benzene exposure in humans, and is the basis for IARC’s classification of benzene as carcinogenic to humans (Group 1) and for its cancer classification under EPA’s IRIS program. The evidence derives from at least five largely independent cohorts spanning different countries, industries, and time periods, summarized below.

Across these cohorts, the direction of association is consistently positive, and several show a clear monotonic exposure-response gradient (most notably the reassessed Pliofilm cohort and the pooled petroleum distribution-worker MDS analysis). Effect sizes vary considerably in magnitude — from relative risks in the range of 1.1–2.2 in the larger, lower-exposure cohorts to standardized mortality ratios exceeding 10 in the highest-exposure Pliofilm subgroup — a pattern consistent with a genuine dose-response relationship rather than chance findings, since the studies with the highest reconstructed cumulative exposures also show the largest effect estimates.

Hematotoxicity (cytopenia)

Reduction in circulating blood cell counts — white blood cells, lymphocytes, platelets, and, at higher exposures, red blood cells — is the most extensively documented non-malignant human health effect of benzene and is the basis for EPA’s noncancer toxicity values. The single most influential study is Lan et al. (2004a), a cross-sectional study of 250 benzene-exposed shoe manufacturing workers and 140 matched controls in China, which found dose-related decreases in white blood cells, granulocytes, lymphocytes, B-cells, CD4+ T-cells, and platelets beginning at a mean exposure of only 0.57 ppm, with decreases of 7–15% at that exposure level and progression to a 36% decrease in B-cell counts (representing clinical B-cell deficiency) at 28.7 ppm. Effects were also observed in a subgroup with negligible co-exposure to other solvents, strengthening the causal attribution to benzene specifically. This finding of hematological effects at sub-1- ppm exposure has been corroborated, at somewhat higher exposure levels, by numerous other occupational studies, including cross-sectional analyses of Chinese workers (Qu et al., 2002; Rothman et al., 1996a; Schnatter et al., 2010; Wang et al., 2021a) and studies using urinary metabolite concentrations (SPMA) rather than air monitoring as the exposure metric (Zhang et al., 2020; Wang et al., 2021b). By contrast, several large, well-conducted studies at mean exposures below roughly 0.6 ppm — including a cohort of 8,532 workers (Swaen et al., 2010) and a longitudinal study of 1,200 workers (Tsai et al., 2004) — found no association with blood cell counts, suggesting a lower boundary near 0.5–0.6 ppm below which hematological effects become difficult to detect epidemiologically, consistent with the LOAEL identified in Lan et al. (2004a).

Myelodysplastic syndrome

MDS, a bone marrow failure disorder that can progress to AML, shows the clearest monotonic dose-response relationship of any benzene-associated hematological outcome in the pooled petroleum-distribution-worker analysis, with an odds ratio of 4.33 (95% CI 1.31–14.3; P for trend 0.01) comparing the highest to lowest cumulative exposure tertile (Schnatter et al., 2012). This finding is considered particularly informative because MDS ascertainment in this study benefited from independent review and reclassification of diagnoses by trained hematopathologists, addressing a historical limitation in which MDS cases were often misclassified as AML or leukemia more generally in older cohort analyses relying on death-certificate coding.

5. Study Quality Considerations

RoseyFields weighs the epidemiological evidence for benzene in light of several recurring methodological limitations that vary in severity across the outcome-specific literature.

Confounding

Co-exposure to other solvents and industrial chemicals is the most pervasive source of potential confounding, since benzene is rarely used in isolation in occupational settings and is a component of complex mixtures such as gasoline and cigarette smoke. Several early lymphoma and leukemia studies were excluded from IARC’s most recent review specifically because benzene exposure could not be adequately distinguished from co exposure to other solvents (e.g., studies reviewed but excluded for inadequate exposure characterization). Smoking is a related, frequently under-controlled confounder; some studies (e.g., Linet et al., 2015; Bassig et al., 2015) explicitly adjust for smoking status, while others (e.g., Wong, 1995, in the Pliofilm cohort) lack complete smoking data, a limitation noted explicitly by IARC’s Working Group.

Healthy-worker effect and reverse causation

Occupational cohort mortality studies are also subject to the healthy-worker effect, whereby employed populations tend to show lower overall mortality than the general population because severely ill individuals are less likely to be employed, which can bias effect estimates toward the null for outcomes with long latency, such as lung cancer (noted explicitly as a limitation in the Korean petrochemical cohorts of Koh et al., 2011, 2014). For hematological endpoints assessed cross-sectionally, reverse causation is a more limited concern because there is no plausible pathway by which incipient cytopenia would cause increased benzene exposure, but selection effects (e.g., workers who develop early hematological symptoms transferring out of high exposure jobs) could attenuate observed associations in surviving-worker cross-sectional samples.

Exposure misclassification

Exposure misclassification is a graded concern across study types. Occupational cohorts with individual air monitoring and detailed work-history reconstruction (e.g., the reassessed Pliofilm cohort; Rhomberg et al., 2016) offer the most reliable exposure estimates, though even these involve substantial retrospective uncertainty for exposures occurring decades before measurement technology existed. Studies relying on job title alone without quantitative measurement (e.g., Koh et al., 2011, 2014; Linet et al., 1996) are more prone to non-differential misclassification, which generally biases associations toward the null. General-population biomarker studies face a distinct issue: urinary t,t-MA is not specific to benzene and is also a metabolic product of dietary sorbic acid/sorbate food preservatives, meaning associations between urinary t,t-MA and outcomes such as insulin resistance (Choi et al., 2014; Park et al., 2022) or hearing loss (Benedict et al., 2024) cannot be confidently attributed to benzene exposure alone.

Statistical power and diagnostic classification

Many individual cohort studies of rarer outcomes (NHL, MM, CML, childhood leukemia) are underpowered, with fewer than 20 exposed cases in a substantial fraction of the studies reviewed by IARC, producing wide confidence intervals that are compatible with both a null and a moderately elevated true effect. Diagnostic classification has also improved materially over time; because leukemia was historically treated as a single diagnostic category and because MDS was often not separately identified, older studies likely misclassified some MDS cases as AML or vice versa, complicating comparison of risk estimates across different eras of the same cohort (as explicitly discussed for the Pliofilm and pooled petroleum-distribution cohorts by Schnatter et al., 2012, and Rhomberg et al., 2016).

6. Independent Expert Panel and Agency Weight-of-Evidence Conclusions

The International Agency for Research on Cancer’s most recent (2018) Working Group review concluded that there is sufficient evidence in humans that benzene causes acute myeloid leukemia in adults, and that positive associations have been observed for non-Hodgkin lymphoma, chronic lymphocytic leukemia, multiple myeloma, chronic myeloid leukemia, acute myeloid leukemia in children, and cancer of the lung, while noting these associations do not yet rise to the level of sufficient evidence. On this basis, IARC classifies benzene as carcinogenic to humans (Group 1), a classification first assigned (with sufficient evidence) in 1982 and reaffirmed through successive reviews. Notably, IARC’s Working Group specifically declined to conclude that positive associations for lung cancer were fully established, with a small minority of the Working Group considering that the lung cancer association was not observed at all — illustrating that even within the sufficient-evidence framework, the strength of evidence is explicitly graded outcome by outcome rather than treated as uniform across all benzene-associated cancers.

EPA’s Integrated Risk Information System has independently classified benzene as a known human carcinogen for all routes of exposure (Group A under EPA’s earlier carcinogen classification framework), based on convincing human evidence of leukemia risk together with supporting animal evidence, and has derived quantitative unit risk estimates for inhalation and an oral cancer slope factor extrapolated from the inhalation dose-response data. EPA’s noncancer toxicity values (RfC of 0.03 mg/m3; RfD of 0.004 mg/kg-day) are likewise based directly on human hematological effects data, reflecting the same body of occupational hematotoxicity evidence discussed in Section 4, rather than on animal data — a notable departure from the more common practice of basing noncancer toxicity values on animal bioassays, and one that reflects the depth and quality of the human hematological database for this chemical specifically.

7. Use of Epidemiological Evidence Alongside Toxicology to Support Causation Opinions

RoseyFields integrates the epidemiological evidence summarized above with the animal toxicological and mechanistic evidence (addressed in RoseyFields’ companion benzene toxicology page) to support both general and specific causation analyses.

For general causation — whether benzene exposure is capable of causing a given disease in humans at all — RoseyFields relies primarily on the convergence of the strong-evidence occupational cohort findings (Section 4) with the mechanistic and animal bioassay evidence demonstrating a biologically plausible pathway (CYP2E1-mediated bioactivation to genotoxic quinone metabolites acting on hematopoietic stem cells) and reproducible tumor induction in multiple rodent bioassays. This convergence of independent human epidemiological cohorts, mechanistic data, and animal carcinogenicity findings — the same lines of evidence relied upon by IARC’s Working Group in reaching its Group 1 classification — is the strongest form of evidence available to support general causation opinions for AML and is used by RoseyFields as the anchor for that determination.

For specific causation — whether benzene exposure caused a particular individual’s disease — RoseyFields uses the quantitative exposure-response relationships from the strong-evidence cohorts (Section 4) to evaluate whether a claimant’s reconstructed cumulative or peak exposure falls within, below, or substantially above the exposure ranges associated with elevated risk in the epidemiological literature, informed by dose reconstruction methods described in RoseyFields’ toxicology page. Because the epidemiological cohorts differ in exposure assessment method, latency window, and diagnostic classification, RoseyFields documents which specific cohort(s) are most comparable to a claimant’s exposure circumstances (route, intensity, duration, co exposures, and diagnosed subtype) rather than applying pooled or generic risk estimates indiscriminately. For outcomes in the suggestive-but-inconsistent tier (Section 3), RoseyFields’ causation opinions are correspondingly more qualified, and RoseyFields explicitly identifies the specific sources of inconsistency (confounding, exposure misclassification, diagnostic heterogeneity) that limit confidence in any given finding, consistent with the weight-of-evidence approach used by IARC and EPA themselves.

References