For the three studies combined, the observed number of leukemia deaths increased from 42 in the early reports to 120 in the current updates. The expected number increased from about 47 to 132. For all three studies combined, for all forms of leukemia among formaldehyde-exposed subjects, the standardized mortality ratio (SMR) has been consistent at about 0.90. Data for ML are provided in none of the early reports and in only two of the current updates. For these two studies combined, the SMR for ML is 0.86 (our estimate). The positive study from the National Cancer Institute (Hauptmann et al., 2003) is presented first and the negative study from the Medical Research Council of the United Kingdom ( Coggon et al., 2003) is last.

3. The NCI study

The 2003 report by Hauptmann et al. of the National Cancer Institute (hereafter, the "NCI" study), describes the mortality experience of workers employed at 10 industrial plants in the USA. Exposure to formaldehyde was estimated from work histories collected through 1980 on the basis of job titles, tasks, plant visits by industrial hygienists, discussions with workers and plant managers, and monitoring data (Blair et al., 1986; Stewart et al., 1986 and Stewart et al., 1987). On the basis of this information, four exposure metrics were estimated as time-dependent variables: cumulative exposure to formaldehyde (in ppm-years), average formaldehyde exposure intensity (in ppm), duration of formaldehyde exposure (in years), and highest peak exposures to formaldehyde (in ppm). Peak exposures were not measured, but defined as short-term exposures (generally <15 min) that exceeded the 8-h time weighted average formaldehyde exposure intensity.

This study, both in its original (Blair et al., 1986) and now in its extended follow-up, is negative for lung cancer (the respective SMRs are 1.10 and 0.97) and for CNS cancer (SMRs are 0.80 and 0.92). Both the original and the current report are negative for cancer of the nose and positive for cancer of the nasopharynx (SMRs=5.00 and 2.10, respectively). However, in neither paper was the positive finding for nasopharynx cancer determined to be causally related to formaldehyde exposure.

The original report included 20 leukemia deaths observed and 27 expected among exposed subjects, giving an SMR of 0.74 with a 95% confidence interval (CI) of 0.45–1.14. There was no dose–response relationship and subjects with the highest cumulative exposure to formaldehyde had a deficit of leukemia deaths (SMR=0.88). The original paper (Blair et al., 1986) used a five-category scale to designate the "intensity" of exposure to formaldehyde. The categories (in ppm) were: 0, >0–<0.1, 0.1–0.4, 0.5–1.9, and >2.0. The respective SMRs for leukemia were: 0.00, 0.95, 0.44, 1.04, and 0.78 (none of which was statistically significant). The "cumulative" exposure of subjects was designated on a four-category scale (in ppm-yrs): 0, 0.1–0.5, 0.51–5.5, and >5.5. The respective SMRs for leukemia were: 0.00, 0.84, 0.79, and 0.83 (These analyses were limited to white men, 78% of the subjects).

The current report (Hauptmann et al., 2003) describes deficits of leukemia deaths both among unexposed (SMR=0.38) and among formaldehyde-exposed subjects (SMR=0.85). In this regard it is similar to the early report. However, the further analyses of leukemia findings, which address dose–response relationships are not based on SMRs and the attendant comparison with general population rates. Rather these analyses consist of internal comparisons and are expressed as "relative risks" (RRs).

The paper provides findings for four measures of formaldehyde exposure: "peak" (ppm), average "intensity" (ppm), "cumulative" (ppm-yrs), and "duration" (yrs). A further difference between the two papers is that the current paper uses a four, not five-category scale for peak exposure.

The internal analyses show no relationship of the risk of leukemia or of ML with cumulative exposure or with duration of exposure to formaldehyde. There is a weak relationship of ML with average intensity of exposure. The RR is 2.49 (CI=1.03–6.03) in the highest exposure category and, while this value is statistically significant, the attendant trend is not (p=0.09). The only clearly positive finding is for ML and with respect to "peak" exposure; RRs are 0.67 (unexposed subjects), 1.00 (low-exposure and referent group), 2.43 and 3.46. The RR of 3.46 (CI=1.27–9.43) and the trend (p<0.01) are statistically significant. SMRs are not provided for the dose–response analyses, nor are RRs or SMRs available using the five exposure categories employed in the original study. Findings for lymphatic leukemia and for "other/unspecified" leukemia are unremarkable.

The NCI study has considerable strengths and its several published reports comprise a major source of knowledge regarding the health effects of exposure to formaldehyde. Principal strengths are the large number of subjects included and an average duration of follow-up of 35 years. This duration is long enough to encompass typical cancer induction periods for more than 68% of the subjects. The total number of deaths observed and expected is considerable. These include sixty-nine leukemia deaths, about 60% of the total of leukemia deaths in the three studies reviewed here. The analysis is thorough, with internal (RR) measures of association provided for specific forms of leukemia at each of the four levels of exposure to formaldehyde for each of four measures of exposure. Unfortunately, SMRs are not provided for the analysis of any of the four measures of exposure.

Support for a causal relationship between formaldehyde and ML, in this paper, rests upon two things—a perception of an overall excess of ML and the acceptance as meaningful of the dose–response relationship even though it is seen for only one (peak exposure) of four metrics evaluated. In fact, it is unlikely that there is any excess of ML among exposed subjects. The all leukemia SMR for these persons is 0.85 (0.67–1.09), based on 65 deaths. Of these deaths 28 (43%) are ML. In the United States, among white men aged 20 and over, the comparable percentage, based on deaths in 1979–1981, is 46%. Thus, there is if anything, a 15–20% deficit of ML among formaldehyde-exposed workers in the NCI study.

With respect to the one dose–response relationship:

(1) Peak exposures were only inferred. To quote the paper itself, "No actual measurements of peak exposure were available in this study." Moreover, all person-years at risk of death were assigned to the highest peak category which they followed. By assigning person-years in this way, an instantaneous peak estimate experienced few times—possibly only once—is treated as if it were descriptive of some, possibly considerable, at-risk period. In the current NCI study, 45% of the person-years of peak exposure are 2.0 ppm (their Table 3). However, only 4% of the original cohort had average exposures 2.0 ppm (Blair et al., 1986). Neither paper indicates how many subjects had peak exposure over 2 ppm. It appears that a very small number of people were exposed above 2 ppm, but the method of classification of peak exposures resulted in assignment of a large number of person-years to the high peak exposure categories.

(2) What actual interpretation can be placed on a "trend" that is based solely on internal, or relative, analyses—for a group of subjects whose absolute risk of death from leukemia is below the norm? The trend with peak exposure in the NCI paper indicates only that the "high peak" individuals have a higher risk of ML than do the "low peak." But it may well be that the "low peak" subjects have a greatly reduced risk of ML—for whatever reason, including chance—as compared with the general population that provides the reference value for the SMR analyses.

Findings with respect to two other measures of formaldehyde exposure in this report, which generally are more reliable indicators of risk—at least in occupational epidemiology, cumulative exposure (in ppm-years) and duration of exposure (in years) are negative. And, as we shall see in the findings of the NIOSH study below, duration of exposure was most descriptive of risk.

A second fundamental question to address before making a causal interpretation of the trend in the NCI study is this: why was the present report prepared? There has been no finding or expression of particular interest in lymphatohematopoietic cancers in earlier reports from the NCI study (Blair et al., 1986, 1990). Further, the present report was followed almost immediately by an additional report ( Hauptmann et al., 2004 Draft Manuscript) based on the same investigation as the report reviewed here. This additional report provides data on the full spectrum of malignancies usually evaluated in a large-scale retrospective follow-up study. Thus, it appears that the findings on leukemia and ML now reported by the NCI comprise an entirely a posteriori result selected for early publication from among a very large number of findings. Thus, the present findings on ML may be the result of the making of multiple comparisons.

4. The NIOSH study

The 2004 report by Pinkerton et al. of the National Institute of Occupational Safety and Health (hereafter the "NIOSH" study) followed up persons employed in 1955 or later at any one of three plants, where garments were produced from fabrics treated with formaldehyde resins. Exposure estimates were derived from industrial hygiene measurements accumulated during the period 1981–1986 (Elliott et al., 1987; Stayner et al., 1988). The average level of exposure to formaldehyde was relatively low, about 0.15 ppm (geometric mean) in the early 1980‘s. The present authors repeat the suggestion in an earlier report (Stayner et al., 1988) that exposure levels "may have been substantially higher" prior to the 1980‘s.

The early report (Stayner et al., 1988) from this study was largely negative for lung cancer and for CNS cancer. There was no overall positive finding for leukemia (and no data for ML). Yet, among workers with the longest duration of employment (20+ years) the SMR for leukemia was statistically significant at 3.10. However, the SMR was even higher, and also statistically significant, at 9.26 among workers with less than four years of exposure.

The current report (Pinkerton et al., 2004) updates the number of deaths (all cause) from 635 in the initial report to 2206. The major findings include a deficit of all cause mortality (SMR=0.92 (0.88–0.96)) and of all cancer mortality (SMR=0.89 (0.82–0.97)). SMRs for cancer of the lung, CNS, buccal cavity and pharynx were unremarkable. No form of cancer was statistically significantly in excess.

For all leukemia, the SMR was 1.09 (0.70–1.62) based on 24 deaths observed and 22 expected. There were 15 deaths from ML, giving an SMR of 1.44 (0.80–2.37). Of these 15 deaths, 9 were due to acute myelogenous leukemia (SMR=1.34 (0.61–2.54)). We can estimate then, that the findings for chronic myeloid leukemia would have been 6 deaths observed and 3.7 expected giving a similar SMR of 1.63 (0.60–3.58).

In this study, there were three categories of "dose" for the metric "duration of exposure." These categories and the related SMRs for ML were <3 years (SMR=0.83), 3–9 years (SMR=1.26) and 10+ years (SMR=2.19). None of these SMRs, or the trend, was statistically significant. The variable "time since first exposure" also was evaluated. For <10 years the SMR was 0.90; for 10–19 years it was 0.40; and, for 20+ years it was 1.91. Only the last figure was statistically significant; the trend was not. The variable "year of first exposure" was assessed. The SMR for ML for "prior to 1963" was 1.61 (not significant) and was 1.15 and 1.02 for the other categories. There were moderate to slight deficits of deaths from lymphocytic and from "other and unspecified" forms of leukemia.

This study appears well-done, but has several inherent limitations: although the study‘s extended follow-up has made it precise with respect to "all cause" and "all cancer" categories of death, it remains very imprecise with respect to individual disease categories including ML, and even all leukemia. The actual exposures to formaldehyde are unknown. They were low after 1980 and probably higher—possibly "substantially" higher—in early years. The exposure levels were probably quite uniform at all three plants and over time, other than for the likely long-term decline. There were probably no peak exposures.

5. The MRC study

The 2003 report by Coggon et al. of the Medical Research Council‘s Environmental Epidemiology Unit at the University of Southampton, UK (hereafter the "MRC" study) (Coggon et al., 2003) describes the follow-up of 14,014 men employed since 1957 at any one of six chemical plants in the UK. As compared to the original report ( Acheson et al., 1984), this report extends the follow-up period by 19 years, through December 31, 2000. Follow-up was 99% complete and the number of deaths observed was increased from 1619 to 5185.

The exposure assessment conducted for this study was similar to that performed for the NCI study. Subject identification data and occupational histories were abstracted from employment records, and each job was classified as belonging to one of five exposure categories of exposure to formaldehyde (background, low, moderate, high, or unknown). No measurements of formaldehyde were available before 1970. However, from later measurements and from workers‘ recall of irritant symptoms, it was estimated that background exposure corresponded to a time-weighted average concentration of less than 0.1 ppm; low exposure to 0.1–0.5 ppm; moderate exposure to 0.6–2.0 ppm; and high exposure to greater than 2.0 ppm. A separate job-exposure matrix was constructed for each factory, and within a factory each job was assigned the same exposure category for all time periods.

Both among the total group of subjects and among men in the "high exposure" category there were small excesses for all cause mortality (SMRs=1.04 and 1.15) and for all cancer mortality (SMRs=1.10 and 1.31). In both groups of subjects the cancer excess is due almost exclusively to excesses of stomach and lung cancer. A slight deficit was found of deaths from sino-nasal and from nasopharyngeal cancer. There was a slight deficit of deaths from CNS cancer (SMR=0.85 (0.57–1.21)). There was no excess of deaths due to leukemia (SMR=0.91) including a small deficit (SMR=0.71 (0.31–1.39)) among the high exposure group. No data are provided for ML. However, we have estimated the maximum likely SMR for ML among the high exposure group to be 0.89. (We took the observed number to be 43% (from the NCI study) of the 8 "all leukemia" deaths actually observed and the expected number to be 40% (national statistics) of the 11.3 deaths for "all leukemia" actually expected. The result was SMR=3.4/4.5=4/4.5=0.89.)

This study is well-done and has a number of strengths. These relate primarily to the probability that exposure to formaldehyde was very high and to the exceptionally long duration of follow-up (up to 60 years for some subjects). Further, the analyses are broad-based and thorough, the absence of information on ML notwithstanding. Limitations of the study include the lack of evaluation of dose–response relationships, though there are data for all subjects and for those with high exposure.

6. Commentary

The criteria, or guidelines considered most relevant to evaluating a hypothetical causal relationship relating to cancer were advanced in 1965 by Hill (Hill, 1965). They were modified and updated by Cole in 1997 ( Cole, 1997). This evaluation of the formaldehyde-ML hypothesis is based on Cole‘s guidelines for causation in the General Case with emphasis on the three papers reviewed.

A usual premise for considering general causation is that there are one or more studies that provide reasonably firm support for causation. That is not so in the present instance as neither the NCI nor the NIOSH study appear, individually—or combined—to sustain a causal argument. However, it is useful to consider the General Case nonetheless. The first guideline is "replicability"—the consistent finding of similar, positive results in all the available epidemiologic studies. This guideline is not met in that the MRC study, which probably involved the highest formaldehyde exposures and, probably the highest peak exposures as well, is negative. The NIOSH study is less positive than is the NCI study. Peak exposures probably did not occur in the garment factories studied by NIOSH. Nonetheless, the authors imply, probably correctly, that formaldehyde exposure was high for some period prior to 1980. It already has been pointed out that the NCI study is not highly consistent even within itself in that only one of four measures of exposure—and that the least reliable—was significantly associated with ML, the internal and external comparisons produce disparate results, and the number of exposure categories for peak exposure has been revised from the original study.

The second guideline is "strength" of association—the extent to which the collective body of data yields an SMR greater than 1.00. For leukemia as a whole the SMR is actually <1.00. Even if the MRC study is ignored, the NCI and the NIOSH studies combined yield, for ML, an SMR of 0.86—also less than 1.00 when based on external analyses (our estimate).

The third guideline is "coherence." the extent to which one or more (assumedly positive) epidemiologic studies are consistent with positive findings for leukemia from other types of investigations, especially animal, and other laboratory investigations. In fact, the generally negative epidemiologic findings are highly coherent with substantial experimental data showing that, under controlled exposures, there is a lack of increase in the concentration of formaldehyde in the blood of exposed humans (2 ppm), monkeys (6 ppm), or rats (15 ppm) (Casanova et al., 1988; Heck et al., 1985), and the inability of inhaled formaldehyde to induce cancer at sites other than the respiratory tract ( IARC, 1995). These results imply that inhaled formaldehyde is rapidly metabolized in the respiratory tract, does not reach the bone marrow and is, therefore, unlikely to induce distant-site toxicity including leukemia.

The fourth guideline is "response to manipulation" and implies that, if a causal relationship were to exist, the removal (or reduction) of formaldehyde from an environment—in this case a work environment—would be followed eventually by a reduction in a previously observed excess risk of leukemia or of ML. In fact, the NCI findings actually support the opposite: the long-term trend in formaldehyde exposure in the plants studied has been down—yet it is only the recent report which—unlike the earlier report—suggests a ML excess. It is possible that the NIOSH findings meet this final criterion in that a leukemia excess, and an ML excess, was restricted to persons exposed prior to 1963. However, that excess was not large or significant.

In sum, then, the formaldehyde–leukemia hypothesis fails each of the four guidelines of general causation. This is hardly surprising in view of the weak and inconsistent findings in the most recent epidemiologic research and the consistent negative findings in animal studies.

The current report of the NCI study has limitations which several different investigators are now exploring (Collins and Lineker, 2004; Heck and Casanova, in press; Marsh et al., 2002). Further, the current NCI report had a termination date of December 31, 1994. It is reasonable to consider that the follow-up, even now, could be extended through 2001 and possibly through 2002. This extension would likely increase the number of leukemia deaths by about 40% from 64 to about 90. This may allow resolution of the discrepancies now evident in the study.

References

Acheson, E.D., Barnes, H.R., Osmund, C., Pannet, B. and Taylor, C.P, 1984. Formaldehyde in the British chemical industry: an occupational cohort study. Lancet 1, pp. 611–616.

Blair, A., Stewart, P.A., O‘Berg, M., Gaffey, W., Walrath, J., Ward, J. et al., 1986. Mortality among industrial workers exposed to formaldehyde. J. Natl. Cancer Inst. 76, pp. 1071–1084.

Blair, A., Stewart, P.A. and Hoover, R.N., 1990. Mortality from lung cancer among workers employed in the formaldehyde industries. Am. J. Ind. Med. 17, pp. 683–699.

Blair, A., Saracci, R., Stewart, P.A., Hayes, R.B. and Shy, C., 1990. Epidemiologic evidence of the relationship between formaldehyde exposure and cancer. Scand. J. Work Environ. Health 16, pp. 381–393.

Collins, J.J. and Lineker, G.A., 2004. A review and meta-analysis of formaldehyde exposure and leukemia risk. Regul. Toxicol. Pharmacol. 40, pp. 81–91.

Casanova, M., Heck, H.d‘.A, Everitt, J.I., Harrington Jr., W.E. and Popp, J.A., 1988. Formaldehyde concentrations in the blood of Rhesus monkeys after inhalation exposure. Food Chem. Toxicol. 26, pp. 715–716.

Clary, J.J., Gibson, J.E. and Waritz, R.S., Editors, 1983. Formaldehyde: Toxicology, Epidemiology, Mechanisms, Marcel Decker, New York.

Coggon, D., Harris, E.C., Poole, J. and Palmer, K.T., 2003. Extended follow-up of a cohort of British chemical workers exposed to formaldehyde. J. Natl. Cancer Inst. 95, pp. 1608–1615.

Cole, P., 1997. Causality in epidemiology, health policy and law. Environ. Law Rep. 27, pp. 10279–10285.

Department of Labor, Occupational Safety and Health Administration (OSHA), 1987. Occupational Exposure to Formaldehyde, Final Rule, Federal Register 52 (233), 46168–46132

Elliott, L., Stayner, L.T., Blade, L., Halperin, W. and Keenlyside, 1987. Formaldehyde exposure characterization in garment manufacturing plants: a composite summary of three in-depth industrial hygiene surveys. , U.S. Department of Health and Human Services, Public Health Service, National Institute for Occupational Safety and Health, Division of Surveillance, Hazard Evaluation, and Field Studies, Cincinnati, OH.

Harrington, J.M. and Oaken, D., 1984. Mortality study of British pathologists 1974–80. Br. J. Ind. Med. 41 2, pp. 188–191.

Hauptmann, M., Lubin, J.H., Stewart, P.A., Hayes, R.B. and Blair, A., 2003. Mortality from lymphatohematopoietic malignancies among workers employed in formaldehyde industries. J. Natl. Cancer Inst. 95, pp. 1615–1623.

Hauptmann, M., Lubin, J.H., Stewart, P.A., Hayes, R.B. and Blair, A., 2004. Mortality from solid cancers among workers in formaldehyde industries. Am. J. Epidemiol. 159 12, pp. 1117–1130.

Heck, H.D‘A., Casanova, M., in press. The implausibility of leukemia induction by formaldehyde: a critical review of the biologic evidence on distant-site toxicity. Regul. Toxicol. Pharmacol.

Heck, H.D‘A, Casanova-Schmitz, M., Dodd, P.B., Schachter, E.N., Witek, T.J. and Tosun, T., 1985. Formaldehyde (CH₂O) concentrations in the blood of humans and fischer-344 rats exposed to CH₂O under controlled conditions. Am. Ind. Hyg. Assoc. J. 46, pp. 1–3.

Hill, B.A., 1965. The environment and disease: association or causation?. Proc. R Soc. Med. 58, pp. 295–300.

IARC Monograph on the Evaluation of Carcinogenic Risks to Humans, 1995. Wood Dust and Formaldehyde. International Agency for Research on Cancer, vol. 62. pp. 282–94

Marsh, G.M., Youk, A.O., Buchanich, J.M., Cassidy, L.D., Lucas, L.J., Esmen, N.A. and Gathuru, I.M., 2002. Pharyngeal cancer mortality among chemical plant workers exposed to formaldehyde. Toxicol. Ind. Health 18, pp. 257–268.

National Toxicology Program (NTP), 2002. Report on Carcinogens, Tenth Edition U.S. Department of Health and Human Services, Public Health Service

Pinkerton, L.E., Hein, M.J. and Stayner, L.T., 2004. Mortality among a cohort of garment workers exposed to formaldehyde: an update. Occup. Environ. Med. 61, pp. 193–200.

Stayner, L.T., Elliot, L., Blade, L., Keenlyside and Halperin, W., 1988. A retrospective cohort mortality study of workers exposed to formaldehyde in the garment industry. Am. J. Ind. Med. 13, pp. 667–681.

Stewart, P.A., Blair, A., Cubit, D.A., Blase, R.E., Kaplan, S.A., Ward, J. et al., 1986. Estimating historical exposures to formaldehyde in a retrospective mortality study. Appl. Ind. Hyg. 1, pp. 34–41.

Stewart, P.A., Cubit, D. and Blair, A., 1987. Formaldehyde levels in seven industries. Appl. Ind. Hyg. 2, pp. 231–236.