Formaldehyde (CH₂O) is a simple one-carbon molecule, found in most human and other living cells as a normal product of the metabolism of serine, glycine, methionine, and choline, and is generated in the demethylation of N-, O-, and S-methyl compounds. It is also an essential intermediate in the biosynthesis of purines, thymidine, and various amino acids [¹]. Consequently, formaldehyde is present in virtually all cells in the body at varying concentrations.

Formaldehyde is also produced commercially and is valuable as a biocide, preservative, and basic chemical in the manufacture of common materials such as plastics, building materials, glues and fabrics, and many household and consumer products, including medicines, health, and beauty aids. Formaldehyde is also a product of organic matter combustion.

Common exposure sources include some laboratories, indoor air (e.g., carpets), vehicle emissions, cigarette smoke, and workplaces manufacturing or using resins, various wood products (e.g., particle board), adhesives, textiles, and numerous other consumer products [²]. High concentrations of formaldehyde were found inside some of the temporary housing units built for victims of hurricane Katrina in the US in 2008, which raised the public awareness of the chemical and its potential acute health effects [³, ⁴].

Inhalation is the predominant route of exposure to exogenous formaldehyde. Following inhalation, formaldehyde rapidly reaches cells in the upper respiratory tract and reacts virtually instantaneously with primary and secondary amines, thiols, hydroxyls, and amides [⁵]. Formaldehyde is swiftly metabolized by erythrocytes [⁶–⁹]. Formaldehyde forms adducts with DNA and proteins and also produces DNA cross-links [¹⁰].

The most common acute health effects of exposure to formaldehyde include eye and upper respiratory tract irritation. Reversible declines in lung function have also been observed, although the evidence that it causes asthma and other chronic respiratory diseases is inconsistent [¹¹]. There is inadequate evidence to assess other potential adverse effects of formaldehyde in humans, such as immunotoxicity, neurotoxicity, and reproductive and developmental toxicity [¹², ¹³].

Our methods were consistent with those used by IARC [²⁸] and others [²⁹–³¹]. Briefly, we identified published, peer-reviewed epidemiologic studies specifically addressing formaldehyde exposure and risk of the LHM. Searches were conducted in PubMed, the US National Library of Medicine’s primary research tool that indexes most of the world’s health and medical peer-reviewed journals since at least 1966. All years indexed were searched to identify these studies using the following key words in various combinations: cancer, leukemia, non-Hodgkin’s lymphoma, lymphoma, lymphocytic, Hodgkin’s lymphoma, hematopoietic, multiple myeloma, hematological neoplasm, formaldehyde, embalmer, garment, laboratory workers, epidemiology, case–control, cohort, case-referent, occupational, chemical, exposure, risk, review, meta-analysis, and commentary. We identified a total of 1,441 potentially relevant articles from the literature searches. Of these articles, 126 were retained as relevant to formaldehyde exposures and LHM. Articles were excluded if they (1) were not epidemiological studies, (2) did not focus on formaldehyde, (3) focused on outcomes other than cancer, or (4) did not present results for specifically for LHM. Additionally, references cited in other publications, including reviews, were checked to ensure the thoroughness of the literature review. We did not attempt to identify unpublished reports. The final review included a total of 37 articles—22 cohort studies and 17 case–control studies.

We comprehensively reviewed the identified literature, including studies of occupational groups and population-based case–control studies of specific LHM that presented results for formaldehyde-related exposures. Most emphasis was placed on findings from occupational cohort studies, which, because of the greater potential for exposure to substantial concentrations of formaldehyde, provide the best evidence for possible associations. We limited the review to the most recent updates of occupational studies, although we include findings from earlier reports where results have changed materially with successive updates.

Defining the outcome of interest is an important aspect of the design of epidemiologic studies, and the LHM are particularly challenging in this regard. Much of the information about LHM and formaldehyde exposure derives from mortality data in occupational cohort studies that spanned several LHM classification schemes. The principles of the nosological classification of this group of neoplasms have changed during the past 40 years, following the increasing understanding of the pathological and clinical characteristics of the different diseases. The most substantial changes in the International Classification of Diseases (ICD) have occurred for the non-Hodgkin lymphomas (NHL). Until the 9th Revision of the International Classification of Diseases (ICD), NHL was classified under two rubrics: “lymphosarcoma and reticulosarcoma” and “other neoplasms of the lymphoid tissue” (Hodgkin lymphoma had a separate code) [³²]. In ICD-10, which follows a new WHO classification, chronic lymphocytic leukemia (CLL), the most common type of leukemia among the elderly, is classified as a form of NHL, and other changes were made to the classification of NHL. The InterLymph Consortium of lymphoma epidemiology has made an effort to adapt the last two versions of the WHO classification to epidemiologic studies, following a hierarchical approach [³³, ³⁴]. Unfortunately, the majority of epidemiologic studies, in particular occupational cohort studies, which based outcomes on death certificates, do not follow the WHO classifications (or its InterLymph adaptation).

We present and discuss findings for specific LHM to the extent allowed by published data. We do not discuss results for all LHM combined because diseases in this group are clinically and pathologically heterogeneous, and thus probably etiologically distinctive.

We did not perform meta-analyses because our evaluation of the individual studies determined that the literature is too heterogeneous, that is, inconsistent, with respect to disease classification and exposure assessment, and therefore, quantitative risks are not appropriately combined. Moreover, the number of independent studies with comparable exposure circumstances (i.e., the same industry or occupation) and similar exposure assessments was too small to justify meta-analyses of these subsets of results. We were especially concerned about combining studies of different groups of workers with poorly characterized circumstances of exposure to formaldehyde. Several previous meta-analyses [³⁵–³⁸] have been performed, yielding variable conclusions, which may result from different methods and the underlying heterogeneity of exposure and health outcome data specificity and validity among published studies. In our opinion, the apparent gain in precision from a meta-analysis would be offset by problems in the interpretation of the summary results. We do, however, provide Forest plots of overall study findings as Figs. 1, 2, 3, 4, and 5.

The main considerations pertinent to assessing epidemiological evidence for a causal relation between formaldehyde exposure and the leukemias or other specific LHM are consistency of findings across studies, evidence for exposure–response associations, accuracy of exposure and health outcome assessment, and minimal confounding and bias. The extent to which exposure assessment in a given study is valid, accurate, and, ideally, permits quantitative dose–response estimation is a critical aspect of research quality. Secondarily, epidemiologic findings suggestive of an association should be interpreted in relation to available evidence of mechanisms of pathogenesis.

The epidemiologic literature provides little or no evidence indicating excess risks overall or exposure–response associations between formaldehyde and any of the LHM, including leukemias, myeloid leukemias, and acute myeloid leukemias. In the majority of occupational cohort studies, which we regard as most informative, specific LHM risk estimates were consistent with the null value, with few exceptions, where the excesses were generally small (i.e., RR < 1.5) and statistically imprecise.

The NCI producers cohort [³⁹] and the nested case–control analysis of the embalmers and funeral directors group [⁵⁴] found elevated risk estimates based on some exposure metrics compared with an internal reference group. However, the increased relative risk for myeloid leukemia noted in an earlier follow-up of the NCI producers cohort [⁴⁷] had diminished in the most recent update [³⁹].

The strongest associations for myeloid leukemia observed in this cohort were with peak exposures; whereas cumulative exposure and average exposure intensity were unrelated to risk. As described in the original publication on the exposure assessment of the NCI producers study [⁷³], there was no uniform definition of peak exposure. Instead, peak was defined on a job-specific basis as an excursion (usually of short duration, e.g., <15 min) relative to the estimated average exposure for the job. Moreover, epidemiologic associations of a specific disease with peak exposure can be difficult to interpret in the absence of prior mechanistic support, such as the requirement for acute above-threshold exposures. In general, established human carcinogens show strong and consistent associations between unbiased measures of cumulative exposure and cancer risk, and cumulative exposure is the default dose metric that is mostly used to assess cancer risk for etiologic exposures. A re-analysis of the data from the previous follow-up [⁴⁷] corroborated the absence of associations with cumulative exposure but also indicated no consistent associations between myeloid leukemia and either duration of time worked at the highest peak or time since highest peak exposure [⁴⁵]. Findings from similar re-analyses have not been reported for the most recent follow-up. In the other relatively strong occupational cohort study [⁴⁸], there was no association between formaldehyde exposure and leukemia.

Among the other occupational studies, the US embalmers study generated elevated odds ratios for some formaldehyde exposure metrics [⁵⁴]. However, as noted by Cole et al. [⁶⁸], this study has notable limitations—including a lack of overall excess leukemia risks (based on PMR analysis), exposure assessment uncertainties, and a poorly defined study base originating from a convenience sample of death certificates obtained from previous proportionate mortality studies. In the study of US garment workers [⁵³], the only support for an association with formaldehyde was the observation of moderately elevated relative risks for myeloid leukemia associated with long-term exposures and longest follow-up that are very crude exposure metrics correlated with older age. The results of the remaining lower-quality studies are not supportive of an association between formaldehyde exposure and leukemia risk. Another recent review of the literature reached similar conclusions for associations with the leukemias [⁷⁴].

The pattern of epidemiological results for the lymphomas is inconsistent. In the NCI producers cohort, there were some notably elevated relative risks (in the range of 2.5–4.0) observed for exposure categories of highest peak for HL and MM [³⁹, ⁴⁷], yet null or at most very small excesses for these diseases were reported in the other studies of occupational formaldehyde exposure.

Existing epidemiologic evidence does not provide convincing support that formaldehyde causes any of the LHMs, including myeloid leukemia. Findings among the highest quality occupational cohort studies are largely null, the positive findings are inconsistent in terms of strength and specificity of association, and there are only isolated instances of exposure–response relations. Epidemiologic evidence from other formaldehyde-exposed occupational cohorts is similarly inconsistent, is often based on small numbers of events, and suffers from a greater likelihood of exposure misclassification and other potential limitations than the two large industrial cohort studies that we regard as highest quality. Available community-based studies, which generally have superior diagnostic classification but poorer quality exposure assessment than in the occupational cohort mortality studies, provide no support for etiologic associations of formaldehyde with any of the LHM.

Although we conclude that a causal connection between formaldehyde exposure and LHM is not supported by existing epidemiologic findings and that the evidence is further weakened by the absence of established carcinogenic mechanisms for the LHM, we nevertheless encourage further epidemiologic research on this topic. We make this recommendation with the caveat that, in order to be informative, further research should offer substantive improvements over the existing body of studies, especially in terms of application of modern diagnostic criteria for specific LHM and individual level quantitative exposure assessment. Well-defined occupational cohort studies should offer the best opportunities to evaluate associations between formaldehyde exposure and LHM risks. Because formaldehyde exposure is ubiquitous, accurately characterizing exposures from the many possible sources, including combustion, household furnishings, automobiles, and consumer products, is essentially impossible. Workplace exposures, on the other hand, are typically substantially higher than exposures from other environmental sources. Continued follow-up of the established high-quality occupational cohorts would be worthwhile, although the scientific yield may be limited because exposure and health outcome misclassification limitations can probably not be remedied. Re-analyses, including sensitivity analysis, of existing datasets may add insight into reported findings, as evidenced by previous re-analyses of the NCI producers cohort data [⁴⁵]. Specifically, additional statistical analyses of risks of specific LHM in relation to the various exposure metrics in the original NCI producers study [⁷³] are warranted.

A more attractive—but also more complicated and expensive—option would be to enumerate and follow new occupational cohorts exposed to formaldehyde. Professional groups, such as anatomists, pathologists, funeral directors, and embalmers, may be the most appropriate study populations because their exposures are frequent, generally remain at relatively high intensity, and may not be confounded by other potential exposures to leukemogens, such as benzene. Another advantage to studying such professions is that they are comprised of persons with comparable socioeconomic status, a characteristic often associated with baseline rates of LHM in the population.

In contrast, new cohort studies of industrial workers would likely encounter problems related to vastly reduced exposures in large workplaces during the past several decades in many high-income countries, and the resulting reduced capacity to test exposure-related associations rigorously. New occupational cohort studies in developing economies may offer opportunities for further research. Any new occupationally based studies should strive to obtain incidence data with modern LHM classification, and to incorporate valid, thorough exposure assessments for formaldehyde and potential confounders. Cross-sectional and, preferably, prospective investigations of biomarkers of bone marrow toxicity relevant to carcinogenesis that have adequate statistical power would also be worthwhile and might be incorporated into cohort studies where feasible (e.g., on subsets of workers).

In summary, we find insufficient epidemiologic evidence to support a causal relation between formaldehyde exposure and leukemia, including myeloid leukemia. We find no clear evidence of an excess risk of leukemia or myeloid leukemia in any large, well-conducted study. Furthermore, we find the occasional positive associations between various exposure metrics and leukemia or myeloid leukemia risk to be inconsistent, and in some instances, contradictory to results based on more conventional exposure characterization approaches. We also find no epidemiologic basis on which to conclude that formaldehyde causes any of the lymphomas. Further weakening arguments for causal associations is the absence of well-defined plausible models of pathogenesis. Nevertheless, in view of the ubiquitous presence of formaldehyde in the population and experimental evidence indicating high-dose carcinogenic potential, at least for portal-of-entry sites, we recommend improved epidemiologic research on potential risks for the LHM.

1Previous publications of the NCI producers study include [⁴⁰–⁴⁷].

2Previous publications of the UK producers study include [⁴⁹, ⁵⁰].

3Previous publications of the US garment workers include [⁵¹, ⁵²].

495 % CI calculated from data presented in manuscript based on method described by Rothman and Boice [⁶⁹].

595 % CI calculated from data presented in manuscript based on method described by Rothman and Boice [⁶⁹].

ENVIRON acknowledges receipt of a grant from the Research Foundation for Health and Environmental Effects® (RFHEE), a 501(c)(3) tax-exempt organization established by the American Chemistry Council (ACC) to conduct this review. RFHEE had no role in reviewing or interpreting the epidemiologic literature or in preparing or revising the manuscript. Drs. Checkoway and Boffetta received financial support from ENVIRON for their time in reviewing the literature and drafting and revising the manuscript. The authors are grateful for the contributions of the following colleagues who provided helpful comments on the draft manuscript: Dr. Philip Cole, University of Alabama, Dr. Jack Mandel, Exponent, Dr. Gary Marsh, University of Pittsburgh, Dr. Morel Symons, du Pont, Dr. Patricia Buffler, U California, Berkeley, Dr. David Coggon, University of Southampton, UK, Dr. David Savitz, Brown University. We especially appreciate the contributions of Dr. Robinan Gentry and Dr. James Swenberg to the section summarizing the mechanistic evidence and of Ms. Farah Chowdhury for creating summary tables.