Introduction

Hematopoietic and lymphoid malignancies (also known as lymphohematopoietic malignancies, or LHM) arise from stem and progenitor cells derived from hematopoietic stem cells. These diseases, though, represent several heterogeneous groups of neoplasms that are biologically, etiologically or clinically distinct [1]. LHM are classified based on the progenitor cells from which they arise, the vast majority being of lymphoid (i.e., derived from the lymph and lymphatic system) or myeloid (deriving from the bone marrow) origin, although much rarer malignancies may arise from dendritic or histiocytic cells.

Lymphoid malignancies generally are associated with lymphoid progenitor cells that mature into cells of the immune system, including B lymphocytes [B-cells], T lymphocytes [T-cells], and Natural Killer [NK] cells), but are categorized by the stage of differentiation of the tumor cells rather than the cell in which the initial transforming event occurred [2]. Lymphoid malignancies include various lymphomas, as well as acute lymphoblastic leukemia (ALL) and chronic lymphocytic leukemia (CLL). Myeloid malignancies arise from myeloid progenitor cells and include all granulocytic (e.g., erythrocytes, or red blood cells) and mast cell lineages [3]. Myeloid malignancies include myelodysplastic syndrome (MDS), acute myeloid leukemia (AML, which has replaced the term acute nonlymphocytic leukemkia, ANLL), myeloproliferative neoplasms (MPN), chronic myeloid (or “myelogenous”) leukemia (CML) – and myelodysplastic/myeloproliferative neoplasms (MDS/MPN) [2]. Multiple myeloma is a malignant disorder involving plasma cells which originate from B-cells. Most of these sub-groups of LHM contain multiple entities with diverse etiologies and possible underlying risk factors.

The 2008 revision of the World Health Organization (WHO) classification of LHMs led to changes in the classification of leukemias and especially myeloid leukemias for epidemiological research based on improved understanding of the lineage of the cells, as well as the molecular genetics and pathologic characteristics of the different malignancies. The WHO classification was further updated in 2016 for lymphoid [4] and for myeloid malignancies [5].

The primary objective of this paper is to evaluate the published epidemiological evidence on the myeloid malignancies for chemical agents classified by the International Agency for Research on Cancer (IARC) as Group 1 carcinogens (that is, “carcinogenic to humans,” commonly referred to as “known human carcinogens”) and for which the epidemiological evidence of a causal association was considered sufficient. The epidemiological and toxicological evidence for associations with exposure to certain chemicals (e.g., benzene) appears to be stronger for specific myeloid malignancies – especially AML and MDS – than for leukemias as a group or lymphoid malignancies, which are generally more closely related to infections and immunological functions [6].

PART I: overview of the myeloid malignancies

Since 2001, the WHO has included genetic information relevant to the diagnosis and classification of LHMs, and the 2008 WHO classification of myeloid neoplasms built on the 2001 classification. The underlying pathology in myeloid malignancies is based on clonal proliferations arising in hematopoietic stem or progenitor cells, and specific diseases are often associated with genetic or epigenetic changes in genes involved in regulation of cell growth. The 2016 update to the 4th Edition of the WHO Classification of Tumors of the Hematopoietic and Lymphoid Tissues additionally incorporated clinical features, morphology, immuno-phenotyping, cytogenetics, and molecular genetics to classify both acute and chronic myeloid leukemias into subtypes and discrete disease entities of clinical significance [7]. A brief review of the current pathology and classification of the myeloid malignancies illustrates several ways in which specific myeloid malignancies differ and a basis for epidemiologically examining them separately (Part II).

Myelodysplastic syndromes (MDS)

MDS refers to a heterogeneous collection of clonal disorders of pluripotent hematopoietic progenitor cells (HPC) that demonstrate lower than normal blood cell counts (cytopenias), an increased percentage of blasts in bone marrow, and dysplasia in erythroid cells, granulocytes, or megakaryocytes [5]. MDS generally has an insidious onset, often diagnosed due to vague symptoms arising as a manifestation of cytopenias, and a variable prognosis, depending upon the molecular genetic profile of the subtype and individual response to therapy. Approximately 20–30% of MDS patients over the age of 65 go on to develop AML, suggesting that at least some proportion of these cases may represent the same underlying disease processes or share causal factors [8]. Some acquired mutations seen in the development of MDS include those in genes involved in RNA splicing (SRSF2), DNA methylation (DNMT3a, TET2, IDH 1/2), chromatin modification (ASXL1) or the cohesion complex (STAG2) [9].

MDS is more prevalent in older adults, with the majority of cases diagnosed in individuals over the age of 60 [10]. Rates of MDS appear to be increasing, which may be due to improvements over time in diagnostic specificity combined with clearer diagnostic criteria for MDS [11].

Acute myeloid leukemia (AML)

The classification of AML includes 20 definitive and 2 provisional subtypes [5]. AML generally has a rapid onset, often diagnosed due to the development of infections, bleeding, or fatigue that result from pancytopenia, and a variable prognosis, depending upon the molecular genetic profile of the subtype and individual response to therapy.

AML is a genetically diverse disease, with 40–55% of patients having chromosome abnormalities that can be identified using conventional analysis techniques [12,13,14]. The most common genetic change is the loss of genetic material in chromosome 5 or chromosome 7 [13]. Others include deletions in parts of chromosomes (e.g., the long arms of chromosomes 5, 7, and 9), insertion of genetic material, inversions of genetic material (e.g., involving chromosome 16), duplications, and translocations (e.g., t[8;21], t[15;17], and 11q23 translocation]) [13]. Approximately 40–50% of AML patients have a normal karyotype and harbour mutations within specific genes including IDH1, IDH2, FLT3, and NPM1.

Some AMLs develop secondary to MDS, and these occur in patients with acquired mutations in genes encoding for myeloid transcription factors (RUNX1, CEBPA) or signal transduction proteins (FLT3) [9]. However, de novo AMLs are also diagnosed in patients with mutations in RUNX1, CEBPA, FLT3 or MLL, but these patients do not have mutations in the genes associated with prior MDS (described above) [9]. Estey (2018) estimated that one-third of patients clinically diagnosed with de novo AML will exhibit genetic mutations specific for secondary AML [9].

AML is more common in the elderly, with more than 58% of cases diagnosed among those 65 years of age or older [10].

Myeloproliferative neoplasms (MPN)

MPNs (previously known as myeloproliferative disorders, or MPD) are a group of clonal hematopoeitic neoplasms, including polycythemia vera (PV), essential thrombocythemia (ET), and myelofibrosis (MF). These conditions are associated with the proliferation of one or more of the myeloid lineages (i.e., increased blood cell counts), without dysplasia. CML shares several features with these disorders, e.g., dysregulated production of a particular lineage of mature myeloid cells, a tendency to progress to acute leukemia, and abnormalities in thrombosis and hemostasis. Many diagnoses of MPNs occur in patients that have acquired mutations in the Janus kinase 2 (JAK2) gene, seen in 95% of patients diagnosed with PV and over 50% of patients diagnosed with MF and ET) [15]. Other mutations seen in patients with MPN include calreticulin (CALR), myeloproliferative leukemia virus oncogene (MPL) [16].

SEER data are limited for MPN, however, ET represented 45.5% of the cases and PV accounted for 41.5% of the cases. The incidence rate was slightly higher in males compared to females, 3.3 vs. 3.0 per 100,000, respectively. Incidence rates increased with age from 0.5 per 100,000 for under age 40 to 18.6 per 100,000 for ages 80 and over [10].

Chronic myeloid leukemia (CML)

In CML, the proliferating cells are mature cells of the myeloid lineage, which have differentiated into functional formed elements of the blood. The development of CML involves an acquired cytogenetic abnormality in the pluripotent hematopoietic stem cells (HSCs) or myeloid progenitor cells located in the bone marrow. Ninety-five percent of CML cases involve the reciprocal translocation of genetic material between chromosome 22 and chromosome 9 [t(9;22)(q34;q11)]. This translocation results in an abnormally shortened version of chromosome 22, known as the “Philadelphia (Ph) chromosome” [17, 18].

In the United States, the median age at diagnosis of CML was 65 years, while the median age at death was 77 years. The incidence rate among males, for all races and ethnicities and all age-groups, was 2.4 per 100,000 population, while among females the rate was 1.4 per 100,000. Incidence among white males was 2.5 per 100,000 while the incidence was 2.2 per 100,000 among black males. Incidence among those under 65 years of age was 1.1 per 100,000 population, but nearly seven times higher (i.e., 7.6 per 100,000) among those 65 and over. Incidence among the population aged 65 and over was highest among white males (11.1 per 100,000), followed by black males (8.2 per 100,000), white females (5.7 per 100,000) and black females (4.9 per 100,000) [10].

Myelodysplastic syndrome/Myeloproliferative Neoplams (MDS/MPN)

The 2016 Classification of LHM includes a category for MDS/MPN. These neoplasms are characterized by both dysplastic and proliferative features. Examples include chronic myelomonocytic leukemia (CMML), atypical chronic myeloid leukemia (aCML) and juvenile myelomonocytic leukemia (JMML) [7]. However, these specific myeloid neoplasms are very rare and infrequently considered in epidemiological studies; therefore, they are not discussed further.

PART II: epidemiological evaluation of four environmental agents and specific myeloid malignancies

Methods

We reviewed the list of carcinogenicity classifications by cancer site published on the IARC Monographs website [19, 20]. The IARC has identified 28 agents as having sufficient evidence of carcinogenicity in humans for neoplasms the IARC grouped as “leukemia and/or lymphoma”. We assessed the human evidence summaries in the “Evaluation” sections of the relevant IARC monographs for each agent.

We excluded from our review 10 pharmaceutical agents (azathioprine, busulfan, chlorambucil, cyclophosphamide, etoposide with cisplatin and bleomycin, melphalan, MOPP [vincristine-prednisone-nitrogen mustard-procarbazine], semustine [methyl-CCNU], thiotepa, and treosulfan) because most of these are chemotherapy agents in which exposure is voluntary and the expected benefit likely offsets the possible leukemogenic effect. We also eliminated radioactive (e.g., X- and gamma radiation, fission-products radionuclides [including strontium-90], thorium-232 and its decay products) and microbiological (Epstein Barr virus, helicobacter pylori, hepatitis C virus, Human immunodeficiency virus type 1, Human T-cell lymphotropic virus type 1, Kaposi sarcoma herpes virus) agents. We excluded two pesticides - pentachlorophenol and lindane - because IARC identified the human evidence as sufficient for causing NHL (lymphomas). We also excluded IARC’s evaluation “occupational exposures in the rubber-manufacturing industry” because workers in the industry are exposed to multiple chemicals and it cannot be determined which specific agents may be causally related to leukemia.

After these exclusions, four leukemogenic chemical agents remained: 1,3-butadiene, formaldehyde, benzene and tobacco smoking. For each of these, we conducted a focused systematic review of the literature using searches of the relevant IARC Monographs and key word searches of PubMed to identify epidemiological studies that reported results separately for specific subtypes of myeloid malignancies. Keywords included “benzene”, “1,3-butadiene,” “formaldehyde,” “cigarette,” “smoking,” “leukemia,” “myeloid,” “AML,” “CML,” “MDS,” and “MDN.” Where results of independent studies of acceptable quality were available, we conducted meta-analyses using random-effects models [21]. For each study, the following characteristics were extracted consistent with PRISMA guidelines [22]: study design, study population, geographic location, study period, exposure categories, number of deaths observed or number of cases in exposed and unexposed groups, relative risk measures (SMRs, HRs, RRs, and ORs) 95% confidence intervals (CI) and covariates adjusted for in models. Using meta-analysis, summary relative risk estimates were calculated by specific categories of myeloid malignancy including AML, CML and MDS. Cohort studies and case-control studies were analysed separately as well as overall and where possible for the highest exposure categories. When multiple results were published on the same study population, we preferentially selected for meta-analysis those based on incidence data, those representing the most complete results, or results reported for higher exposure categories. Publication bias was assessed using a visual inspection of the funnel plots as well as Egger’s test (see supplemental file). Heterogeneity was evaluated using the I2 statistic, which provides a measure for quantifying inconsistency of effects across studies. All meta-analyses were conducted using R version 3.6.1 (2019-07-05).

Results

1,3-butadiene (butadiene)

The IARC last reviewed the carcinogenicity of butadiene in 2009 [23]. The epidemiological evidence for exposure to butadiene and risk of leukemia is based primarily on studies conducted among workers in the butadiene monomer industry and workers in the styrene–butadiene rubber (SBR) manufacturing industry. However, results on specific types of leukemia are available only from studies conducted in the SBR manufacturing industry.

A study of approximately 17,000 workers from eight SBR facilities across the United States and Canada reported an increased risk of leukemia among 16,610 workers (12,412 exposed to butadiene), based on 58 leukemia deaths [24]. Because standardized mortality ratio analyses were not conducted, it is not clear whether excess mortality from leukemias occurred. A positive dose-response was reported between cumulative exposure to butadiene and risk of leukemia. Despite the individual exposure estimates and the relatively large number of leukemia deaths, results by leukemia subtype were not reported.

The mortality follow-up was extended through 1998 for 15,649 men employed since 1943, 75% of whom were exposed to butadiene [25]. A total of 71 deaths from leukemia was observed (SMR 1.16, 95% CI, 0.91–1.47). No consistent patterns were observed by categories of years since hire or by years worked. The excess leukemia mortality was concentrated among men hired in the 1950s (31 deaths; SMR, 1.50; 95% CI, 1.01–2.11). In the analysis by leukemia subtype the SMR was 1.02 (95% CI 0.56–1.71, 14 deaths) for AML and 1.67 (95% CI 0.83–2.99, 11 deaths) for CML. Mortality from AML was elevated in maintenance laborers and from CML in laboratory workers; however these were based on only five and three deaths, respectively.

Time-dependent exposure-response relationships between several butadiene exposure indices and leukemia (81 decedents) as well as all myeloid neoplasms (56 decedents from myeloid and monocytic leukemia, myelofibrosis, myelodysplasia, myeloproliferative disorders and polycythemia vera) were evaluated [26]. The butadiene exposure indices included cumulative exposure in ppm–years, total number of exposures to peaks (> 100 ppm) and average intensities of exposure in parts per million. All three exposure indices were associated positively with the risk for leukemia whereas the myeloid neoplasms were more clearly associated with peak exposures. This highlights the potential additional role choice of exposure metric may play in evaluating risk [27].

Only two studies evaluated the risk of myeloid malignancies [25, 26], based on the same study cohort. Risks of AML were not increased, and risk of CML was increased but not statistically significantly. For myeloid leukemias (including CML), a relationship was reported for peak exposure, but not cumulative exposure.

Formaldehyde

The IARC last reviewed the carcinogenicity of formaldehyde in 2009 (23). The epidemiological literature on exposure to formaldehyde and risk of leukemia published since the IARC meeting was reviewed in detail [28]. Twenty studies that reported results for leukemia overall were included, three of which also reported results for myeloid leukemia. Since then, some of the occupational epidemiological studies have been updated or re-analyzed, and new studies have been published that examine myeloid leukemias in relation to formaldehyde exposure (Table 1).

Table 1 Select characteristics of included formaldehyde studies
Full size table

Peak exposure in a cohort of workers employed in six plants producing formaldehyde in the United States was re-defined and analysed with respect to specific leukemia types. Absolute peak exposure, duration of time worked at the highest peak or time since highest peak exposure generated no clear associations with myeloid leukemia or AML. Cumulative exposure also was unrelated to risk of leukemia, myeloid leukemia, AML, or CML. The authors concluded, “Findings from this re-analysis do not support the hypothesis that formaldehyde is a cause of AML” [31]. The use of peak exposure in this and other epidemiological studies presents specific challenges that have been explored separately [27].

The other occupational cohort study of formaldehyde producers also reported no clear associations between different metrics of formaldehyde exposure and myeloid leukemia [30]. In a study of garment workers in the United States [29, 35], moderately elevated relative risks for myeloid leukemia were associated with duration of employment, a surrogate for cumulative exposure, and duration of follow-up, a surrogate for latency. A large cancer registry study in the Nordic countries “did not provide clear evidence for an association between occupational solvent exposure and AML” [34].

There were no deaths from myeloid leukemias among a cohort of laminated plastic workers from Italy [36]. A European community-based cohort study [33] found no increased risks of AML or CML among study subjects with low-level occupational exposure to formaldehyde (no study subjects were reported to have high occupational exposure to formaldehyde).

SMR results for myeloid leukemia, AML and CML, including those for the highest categories of exposure from the most recent updates of the industrial cohorts, are summarized in Table 2. Overall, the updated cohort study analyses demonstrate no clear or consistent excess risk of myeloid leukemia or AML or CML. None of the formaldehyde studies evaluated MDS or MPN. Table 3 presents meta-analysis results by myeloid malignancy, specifically ML, AML and CML. No statistically significant increased meta-relative risk estimates were seen. Based on the I2 test, heterogeneity was low, and based on Egger’s test, publication bias appears unlikely.

Table 2 Formaldehyde exposure and risk of specific types of myeloid malignancy by exposure category
Full size table
Table 3 Meta-analysis of formaldehyde exposure and risk of specific types of myeloid leukemia in the most informative cohorts
Full size table

Benzene

The IARC last reviewed the carcinogenicity of benzene in 2018 [6]. Risk estimates for one or more myeloid malignancies were reported in 31 independent studies. Characteristics of the design of these studies are summarized in Table 4, and results are summarized in Table 5.

Table 4 Select characteristics of included benzene studies
Full size table
Table 5 Results of studies on benzene exposure and specific type of myeloid malignancy*
Full size table

A cohort exposed to benzene in a variety of manufacturing and user industries, including paints and painting, printing, footwear, paints, chemicals in 12 cities in China was followed for mortality. In the most recent update of this cohort, 73 leukemia deaths were observed, including 60 among benzene-exposed workers [51]. Similar risks were reported for AML and CML, while the risk of MDS was inestimable due to zero cases of MDS among the unexposed group (Table 5). A case-cohort analysis of combined AML/MDS (44 cases) and CML (18 cases) from the 12-city China cohort examined the timing of exposure [66]. The investigators found that high cumulative exposure or high intensity exposure experienced 2 to 10 years before diagnosis increased the risk of MDS/AML among workers who were first exposed under 30 years of age, but not for workers first exposed 30 years of age or older [66].

Pooled results for AML, CML, MDS and MPN were reported using data from three separate nested case-control studies of petroleum workers from Canada, the UK, and Australia. No significantly elevated risks of AML by cumulative exposure, average exposure intensity, maximum exposure intensity, duration of employment, and peak exposure were reported [55]. Increased relative risk of MDS for cumulative exposure greater than 2.93 ppm-years and peak exposure less than 3 ppm were reported, but not for CML or MPN [58]. The most recent follow-up of incidence in the UK petroleum distribution and oil refinery workers reported deficits of MDS [67]. Similarly, the most recent mortality follow-up of the Canadian Petroleum Workers cohort reported no excess of MDS [68].

An increased mortality risk of MDS associated with 25 ppm-years or more of benzene exposure was reported, however, this finding was based on only one death [42]. A much larger registry-based study of occupational exposure to benzene and incidence of AML indicated no increased risk [34]. Age-stratified analyses indicated a possible increased risk of AML in workers under age 50 and in the highest benzene exposure group [34].

Since studies did not report results for the same subtypes of leukemia, it is problematic to combine all results using meta-analysis. We, therefore, conducted meta-analyses of results reported for myeloid leukemias combined or specifically for AML, MDS and CML (Table 6). The meta-analysis of results for AML was based on 27 estimates from 26 publications and generated a summary RR of 1.30 (95% CI 1.09–1.55; I2 = 48.91%) with similar increases, but some variation in the summary RR across exposure categories (i.e., high, low, any exposure). The results for low and any exposure to benzene were not statistically significantly elevated. Egger’s test was significant for the overall result and for cohort studies, but not for the other meta-analyses (Table 6). Visual inspection of funnel plots indicated possible publication bias favoring negative results (see supplemental file).

Table 6 Meta-analyses of studies of specific myeloid maligancy type for benzene.*
Full size table

For CML, the meta-analysis of overall results was based on 18 estimates from 17 studies resulting in a summary RR of 1.25 (95% CI 1.00–1.55; I2 = 0%) with large variation in the summary RR across exposure categories. Publication bias appears unlikely. The meta-RR for myeloid leukemias combined, based on seven studies, was 1.56 (95% CI 1.10–2.20; I2 = 45.06%) with wide variability by exposure category (high, low, any exposure). Evidence of publication bias was present for the overall and cohort meta-analyses, but small numbers of studies hindered results for the exposure categories (Table 6). Visual inspection of funnel plots indicated that publication bias favored positive results (see supplemental file).

The meta-analysis for MDS was based on nine studies and generated a summary RR of 1.87 (95% CI 1.39–2.52; I2 = 40.73%) with similar risks for the low exposure category (m-RR = 2.29, 95% CI 1.51–3.48, I2 = 0%) and the high exposure category (m-RR = 1.80, 95% CI 1.18–2.75, I2 = 51.97%). Publication bias appears unlikely. The meta-analyses for the overall category for each outcome were also calculated by study type (Table 6). The meta-analyses for CML by study type revealed a large difference between the case-control (m-RR = 1.93; 95% CI 1.05–3.56, I2 = 25.76%) and cohort studies (m-RR = 1.13; 95% CI 0.89–1.45, I2 = 0%), possibly reflecting reporting bias, as many of the case-control studies were population-based and dependent on self-reported exposure. This difference by study design was not observed for AML, MDS or the category of all myeloid leukemias combined.

The interpretation of results on risk of specific leukemia types from exposure to benzene is complicated by the heterogeneity in exposure circumstances. However, the evidence indicates a similar association between occupational exposure to benzene and specific myeloid neoplasms, but the association appears strongest for MDS, especially among more recent studies. This raises the question of whether earlier studies identifying associations between generally very high benzene exposure and AML might have reflected the occurrence of secondary AML following unrecognized (or misdiagnosed) primary cases of MDS. It is noteworthy that a very large record linkage study from the Nordic countries reported no association between benzene exposure and AML incidence [34].

Tobacco smoking

The IARC last evaluated the carcinogenicity of tobacco smoking in 2009 [69]. From the IARC review and the PubMed search, we identified 42 studies on tobacco smoking and risk of myeloid malignancies, 27 of which reported results for current or ever smokers (current and former smokers combined) as summarized in Table 7. The remaining studies reported results for different groups of smokers, defined according to dose (cigarettes per day), duration (years of smoking) or cumulative consumption (pack-years). Whenever possible, we selected results by duration of smoking, since this is the exposure metric most strongly associated with lung cancer risk (Table 8).

Table 7 Select characteristics of tobacco smoking studies
Full size table
Table 8 Results of studies on tobacco smoking and specific type of myeloid malignancies
Full size table

One of the largest studies to examine leukemia risks among smokers was a prospective cohort of 1.3 million middle-aged women recruited for breast cancer screening during 1996–2001 and followed for mortality through 2009 [84]. The investigators identified death due to myeloid neoplasm in 831 cohort members and reported a statistically significantly increased risk (RR = 1.33, 95% CI 1.24–1.42). The increase was driven by a significant RR for MPN/MDS (RR = 1.42, 95% CI 1.31–1.55), whereas the RR for AML was not elevated (RR = 1.10, 95% CI 0.96–1.26). Relative risks also increased with increased intensity of smoking for myeloproliferative/myelodysplastic disease, but not for AML [84].

A cohort study of over 330,000 Swedish construction workers with follow-up for mortality through 2004 reported a statistically significant association between “current” smoking and AML risk (RR = 1.50, 95%CI: 1.06, 2.11), but no association with CML (RR = 0.69, 95% CI 0.42–1.14). For AML, relative risks did not increase with increasing intensity of smoking [78].

Table 9 presents summary risk estimates for myeloid malignancies by various smoking exposure metrics and study type. Several meta-analyses demonstrated significant heterogeneity, as indicated by the high I2 statistic and associated low p-values. However, publication bias generally was not indicated.

Table 9 Meta-analyses of studies specific leukemia types for smoking
Full size table

The meta-analysis for smoking and AML was based on 28 studies and resulted in a summary RR of 1.43 (95% CI 1.25–1.62; I2 = 56.25%) with slightly higher meta-RR for current smokers and a lower meta-RR for ever smokers. Meta-analysis of results for CML was based on 14 studies, generating a summary RR of 0.93 (95% CI 0.74–1.16; I2 = 40.44%) with similar results for current and ever smokers. The meta-RR for myeloid leukemias combined (i.e., based on eight studies not differentiating by leukemia type) was 1.54 (95% CI 0.79–3.01; I2 = 81.80%) with similar results for current and ever smokers. The meta-analysis of overall results on MDS was based on 22 studies and resulted in a summary RR of 1.66 (95% CI 1.38–2.00; I2 = 61.89%) with similar results for current and ever smokers. Only three studies were identified for MPN resulting in a summary RR of 1.70 (95% CI 1.23–2.34; I2 = 0%) for current smokers. Publication bias appears unlikely for the meta-analyses on smoking except for the Egger’s test results for the category of current smoker and CML. Visual inspection of the funnel plot, however, did not suggest any publication bias (see supplemental file).

In contrast to benzene, the evidence on the risk of specific leukemia subtypes from tobacco smoking indicates an association with AML, but not with CML. Similar to benzene, there is evidence of an increased risk of MDS. Although only three studies were identified, risk of myeloproliferative/myelodysplastic neoplasm (MPN) appears to be increased among smokers.

Discussion

The myeloid malignancies clearly have different clinical features and characteristic genetic aberrations and therefore, they should be evaluated separately in epidemiological studies intended to identify risk factors and potential causes.

We found little consistency in the way leukemias were evaluated, and they often were analyzed in aggregate, mixing myeloid and lymphocytic leukemias. The more recent benzene cohort studies were the exception, as they specifically evaluated AML, CML and MDS separately. Some analyses evaluated myeloid malignancies separately from the lymphocytic neoplasms, but still combined AML and CML, despite evidence of different mutations in genes and other risk factors that indicate different etiologies. Despite the determination that the epidemiological evidence was sufficient for purposes of establishing causation for leukemia, our review identified only small numbers of studies that actually reported results for specific types of myeloid neoplasms. Furthermore, where specific diseases were considered, small numbers of observed events often limited the precision of risk estimates.

For example, for butadiene, only one study analyzed risks by specific leukemia type, and findings were mixed: statistically significant associations were reported for CML among laboratory workers (based on only three deaths) but not for AML [25]. That results on butadiene exposure and myeloid malignancies are based on a single study and do not allow any causal conclusions does not necessarily mean that the IARC conclusion of “sufficient” human evidence is incorrect. Rather, it indicates that the relationship, if any, with one or more specific type of leukemia cannot be discerned based on available epidemiological evidence.

The updated meta-RRs for formaldehyde showed no consistent relationship with AML, CML or myeloid leukemia overall, confirming an earlier meta-analysis [28]. A further meta-analysis of the highest exposure groups was not conducted due to a lack of a common exposure metric. A very large registry-based linkage study demonstrated no increased risk and, in fact, reported a statistically significant deficit of incident AML cases among groups potentially occupationally exposed to formaldehyde [34]. Our findings suggest that the updated human evidence for formaldehyde and leukemia (of any type) may not be sufficient as determined by IARC and should be revisited. Combined with the lack of support from animal and mechanistic studies, it is unlikely that formaldehyde causes leukemia in humans [112].

A causal relationship between benzene exposure and AML has been recognized for decades and our meta-analyses indicate a significant increased risk overall and at high levels of exposure, yet the largest study, the Nordic registry study, demonstrated no association [34]. Other recent high-quality studies also indicate no clear association between benzene and AML risk [33, 42, 48, 55], possibly due to generally low exposure concentrations that do not exceed a possible exposure threshold for risk. The most comprehensive study on benzene and incident leukemia and myeloid neoplasm risks demonstrated a stronger association between benzene and risk of MDS than for AML, especially among workers with high peak exposures [58]. However, subsequent analyses of the Canadian sub-cohort were not consistent with these findings [68]. An association was reported between benzene and CML, but this was limited to case-control studies and may reflect potential reporting bias. Nevertheless, these findings epidemiologically underscore the importance of examining and contrasting results for specific malignancies (at least initially) and that exposure metric may play an important role in identifying causal associations [27].

Findings for tobacco smoking and myeloid leukemia were consistently positive for AML and negative for CML. The meta-RR for AML demonstrated a 50% statistically significantly increased risk, whether based on seven studies reporting individual leukemia types or the six in which AML and CML were reported separately. These results, and specifically the statistically significant positive meta-RR for AML and null findings for CML, further underscore the importance of examining epidemiologically narrowly defined or disease-specific relationships.

An ancillary finding of this evaluation is the surprisingly limited body of epidemiological studies aimed at addressing and differentiating risks by specific types of myeloid malignancy. While observing small numbers of any specific leukemia will plague all but the largest studies (or studies in which a strong association is indicated), we would argue that arbitrarily combining possibly discrete disease entities to improve “statistical power” will not help elucidate their specific causes; rather, this technique likely will dilute any true malignancy-specific associations and may lead to erroneous conclusions. One exception might be the subset of MDS cases that progresses to AML: these may reflect different clinical stages in the progression of the same disease. Nevertheless, publishing results based even on small numbers will facilitate combining results across studies using meta-analysis. It is conceivable that apparently negative findings based on small numbers may not be published, leading to potential “small numbers” or “negative study” publication bias, of which we found some evidence. While concerns of uncontrolled confounding arise in many occupational epidemiological settings, it is unlikely to be problematic in this context, primarily because there are no common exposures or risk factors that are strongly associated with all (or even multiple) types of leukemia. Progress in understanding the genetic factors underlying each of the myeloid neoplasms likely will guide future epidemiological studies to improve their ability to define appropriate combinations of myeloid malignancies and to isolate environmental risk factors that may be among their causes.

Nevertheless, our detailed evaluation of the four environmental chemical agents summarized here highlights important differences in risks by myeloid malignancies and provides support for reporting disease-specific findings from studies of environmental agents and risk of specific myeloid leukemias or other LHM. They also build on clinical observations that treatments with chemotherapy drugs lead to high incidence of AML and MDS (and possibly ALL) and that the genetic changes in therapy-related myeloid neoplasm reflect some specificity for the type of chemotherapy administered, but that chemotherapy does not lead to appreciable increases in CML, MPN, or lymphoid malignancies [2]. Meanwhile, epidemiological findings based on small numbers of specific LHM should be reported but appropriately caveated and not over-interpreted, as these results statistically will be unstable with likely false-positive and perhaps more likely false-negative relative risk estimates. Similarly, findings based on analyses of multiple types of leukemias and other LHM should be examined further and if possible, groups of LHM deconstructed, to identify the specific neoplasms that may be driving an observed association or situations where true associations may be diluted or masked.