Epidemiologic evidence on the relationship between formaldehyde exposure and cancer.

Epidemiologic evidence on the relation ship between formaldehyde exposure and cancer. Scand J Work Environ Health 1990;16:381- 93. Over 30 epidemiologic studies have evaluated cancer risks associated with formaldehyde exposure. Excesses werereported for several sites, leukemia and cancers of the nasal cavities, nasopharynx, lung, and brain generating thegreatest interest. The excesses of leukemia and brain and colon cancer found among profes sionals may not be related to formaldehyde exposure, since similar excesses were not observed among industrial workers. Inconsistencies among and within studies impede assigning formaldehyde a con vincing causal role for the excesses of lung cancer found among industrial workers. A causal role for formaldehyde is the most probable for cancers of the nasopharynx and, to a less extent, the nasal cavities. Evidence of exposure-response relationships, the fact that direct contact with formaldehyde may occur at these upper respiratory sites, and the consistency of these findings with experimental studies make this assumption highly probable.

An imal bio assays (I , 2) indicate that formaldeh yde ca uses nasal tumors in rats. The wide spread use of for maldehyde in ind ustry a nd in commercial products ha s heightened public health co ncern s regarding this chemical. In the United States, the National Oc cupational Ha zards Survey (3) of the National Institute for Occupational Safet y and Health indicates that 1.6 million wor kers may be expo sed to formaldehyde during the manu facture o f formaldeh yde resins, textiles, ga rments, particle bo ard and plywood, insulating materials, dinnerware, and paper a nd in health-related pro fession s. In addition , off-ga ssing of formaldehyde fro m certain co nsumer products may lead to expo sure of large numbers of th e ge neral public.
The results of animal expe riments led to ep idem iologic studies on a variety of formaldehyde-exposed populations. Previous reviews of these studies have pr imarily focu sed on asses sin g th e weight of evid ence co ncern ing fo rma ldehyde as a human carcinogen witho ut a det ailed presentation of the stu dy result s. In one such review the International Agenc y for Research on Ca ncer (l ARC) concluded that th ere is sufficien t eviden ce fo r th e carcinoge nic ity of formald eh yde in 1 National Cancer Institute, Rockville, Maryland, United States. 2  animals, but limited evidence for humans (4). Another group co ncluded that th e evidence fo r the carcinogenicity of formaldeh yde in humans was less per suasive th an did IARC (5).
In our review of th e epidemiologic literature we hav e summarized results on a site-by-site basis and have focu sed on potential expo sures to form aldehyde and other expo sures that the study populations ma y ha ve experienced to eva luate the strength o f th e evidence regarding fo rmalde hyde and human ca ncer.

Study designs
Over 30 reports fro m epidemiologic studies on formald ehyde have been published (table I) . These reports hav e focused on professional groups such as funeral directors and embalmers, anatomists, pathologists, and workers in formaldeh yde facilities producing formaldeh yde , resins, plastic molding, decorative laminates , plywood , particle bo ard , and apparel. These studies, as well as relevant broad surveys of cancer risk by occupation in the United States, Canada, a nd Swed en, a re included in thi s review. The studies ha ve been conducted in eight countries (United States, Canada, Great Britain , Denmark, The Netherlands, Finland, Sweden , and Italy) and ha ve emplo yed standardized mortal ity ratio (SMR), proportionate mortality ratio (PMR), standardized incidence ratio (SIR), and case-referent des ign s. The total number of cancer cases with potential exp osur e to formaldeh yde in thes e studies is now large enough to provide considerable statistical po wer for evalu ating the ca rci nogenicity o f formaldeh yde in humans for all but rare sites. Table 1. Des c ri p t ion o f th e epidem io log ic s tu dies on formald eh yde. (PMR =pro portionate mort a li t y ra t io, SIR =standardized i n ci d e n c e ra ti o, SMR =standardized mortali ty rat io)

382
In epidemiologic studies of forma ldehyde exposure and cancer risk variou s ap proaches to exposure assessment have been taken, largely dict ated by th e cons traints of the study design. Mortality surveys, such as of profe ssional groups, have relied upo n quali tat ive assessme nt of exposure based upon job title or similar designations. Mortality studies of individual industrial facilities have generally provided more-detailed exposure estimates based on industrial exposure measurements and ancillary information. Case-referent studies composed of persons from a variety of occupations have generally relied on indirect exposure assessment based on knowledge of work practices and exposures in formaldehyde-related jobs .

Professionals
Embalmers may come in contact with a variety of chemicals, including formaldehyde, pheno l, dyes and stains, glutaraldehyde, glycol, glycerol, xylene, alcohols, and particulates, and also with biologic material and gamma radiation (14). Levels of <2 ppm for solvents (ie, isopropyl alcohol, 1,1,I-trichlo roethane, isopentane, and trichloroethylene) and of 1.8 mg/rrr' for total dust , and 1.3 rng/nr ' for respirable dust measured from hardening compo und s have also been reported (15). Published reports on exposure in this profession primarily discuss exposure to formaldehyde. Mean time-weighted average (TWA) values ranging from 0.30 to 1.30 ppm have been reported for formaldehyde during the emba lming procedure (15-1 7), but a recent evaluation by three of the authors (PAS, RBH, AB) found somewhat higher TWA values (0.8 to 2.9 ppm), with instantaneous peak s of formaldehyde or other aldehydes of up to 20 ppm . On the assumpt ion that an embalming takes about 3 h and that only one embalming occurs per day , the 8-h TWA values in all of these cases would be about 60 0/0 lower. Thus TWA values during embalming operations appear to be similar to those found in industry, but such high peak concentrations have not been reporte d in industrial facilities. Chemicals used in anatomy laboratories may be similar to those used in funeral homes, but anatomists appear to have a greater range of exposure to formaldehyde, the TWA values ranging from 0.02 to 5.87 (18)(19)(20)(21). Although peak levelsin anatomy laboratories have not been reported, they may be similar to those generated during embalming. In studies of cancer risk among profe ssional workers (ie, anatomists, pathologists, and funeral professionals), assessment of exposure to formaldehyde was largely qualitative, being inferred from job titles identified from state licensing records and from membership lists of professional societies with no specific estimates of exposure for study subjects.

Industrial workers
In contra st to the studies of cancer risk among professional groups, epidemiologic studies of industrial workers (table 2) have generally included quantitative estimates of the level of exposure to formaldehyde. Estimated levels in these investigations were similar and resembled measure ments reported in other settings (22). The six manufacturing facilities in the study by Acheson et al (23) produced formaldeh yde, formaldehyde resins, formaldehyde adhesives, paraform, and alcoforms . Exposure to formaldehyde was estimated for each job and assigned to exposure categories of <0.1,0.1-5 ,0.6-2.0, and >2.0 ppm. Monitoring Table 2. Exposure levels and potential chem ica l confounders in stud ies of indust rial workers. Twenty-five percent of the subjects were unexposed. Among those holding jobs where exposure occurred, 35 070 had their highest exposed job in the low-exposure category, 14 0J0 had their highest exposed job in the moderate category, and 51 0J0 had their highest exposed job in the high category (> 2.0 ppm). The authors specifically mentioned the presence of only one other substance, asbestos, but we assume that many of the chemicals identified in the formaldehyde and resin plants studied by Blair et al (24)(25)(26)(27), such as wood dust, phenol, melamine, and urea, may have also been present in the formaldehyde and resin plants in this study.
Blair et al (24)(25)(26)(27)(28)(29) studied IO plants producing a variety of products, including formaldehyde, formaldehyde resins and molding compounds, decorative laminates, photographic film, and plywood. The estimates of formaldehyde levels for each unique combination of job, department, plant, and calendar-year combination were based on the results of monitoring conducted by the investigators (about 2000 full-shift personal samples) (27), company monitoring results (over 4500 results dating back to the 1960s for some plants), plant visits, job descriptions, and interviews with management and long-term employees regarding changes in the process, controls, work practices, and sensory perceptions (26). With the use of these estimates jobs were assigned to categories of trace,~0.1, 0.1-<0.5,0.5-1.9, and~2.0 ppm. In addition, for each job, estimates were made for the level and frequency of peak exposures, the potential for exposure to formaldehyde in liquid or particulate form, the possibility of other chemical exposures, and the industrial hygienists' confidence in the estimates. Eight percent of the cohort never held exposed jobs. Among the exposed workers, 44 070 of the cohort had cumulative exposures which were between 0.01 and 0.5 ppm-years, 34 0J0 had exposures of 0.51-5.5 ppm-years, and 14°/0 had cumulative exposures which exceeded 5.5 ppmyears. Alcohols, melamine, urea, antioxidants, ammonia, plasticizers, dyes and pigments, asbestos, phenol, carbon black, wood dust, stearates, amides, and acetone were other chemicals to which large numbers of workers were exposed.
Bertazzi et al (30) studied a urea and melamine resin plant. The jobs identified were classified as exposed to formaldehyde, exposed to other chemicals, or not exposed. Between 1974 and 1979, the mean levels of formaldehyde (based on 187 measurements) were between 0.2 and 3.8 ppm, with individual results as high as 9.8 ppm. Other operations in the plant included a styrene resin and an epoxy resin (with possible epichlorohydrin exposure) operation in which exposures may have occurred to xylene, toluene, and methyl isobutyl ketone. No specific identification of other chemicals in the resin operation was made, but we as-384 sume them to be similar to those in the resin plants studied in the United States (25) and Great Britain (23).
Partanen et al (31) identified 19 sawmills and manufacturers of plywood and formaldehyde glue for a nested case-referent study of respiratory cancer. Exposure assessments by job were based on interviews of cases and referents (or their next-of-kin), available monitoring data, plant visits, and interviews with supervisors and long-term workers to obtain data on ventilation controls, work practices, and other relevant factors. Monitoring was conducted in four companies with few air sampling measurements. Jobs were assigned to one of four exposure categories « 0.1, 0.1-< 1.0, 1.0-<2.0, and~2.0 ppm). The investigations identified 74 0J0 of the cases and referents as exposed to formaldehyde of < 0.1 ppm, 17 0J0 as exposed to 0.3 ppm, 7 0J0 as exposed to 1.5 ppm, and 2 0J0 as exposed to 3.0 ppm. Peaks of > 2 ppm were also noted. Other exposures noted included wood dust, chlorophenols, pesticides, terpenes, phenol, caseinalbumin glues, melamine glues, solvents, exhaust gases, molds, bacteria, and possibly bis(chloromethyl)ether (32).
Stayner et al identified study subjects from the company insurance fund records of three garment manufacturing firms for a PMR study (33) and from company records for an SMR study (34) (two of these plants were also in the PMR study). In the PMR study, recent measurements in two of the plants indicated that air levels ranged from 0.1 to 1.0 ppm. Levels were presumably higher in the past. No other chemical exposures were believed to be present. In the SMR study (34) the current geometric mean level of formaldehyde was low, ranging from 0.14 to 0.17 ppm (in 557 samples). Although historical exposure levels to formaldehyde were not available, they were assumed to be substantially higher. The authors stated it was unlikely that substantial peaks or intermittent exposures occurred or that there were other potential chemical confounders.
In other industry-based studies, exposure evaluations have been less detailed. Approaches have included analyses by length of employment (35,36), ever or never exposed (35,37), and physical state of the formaldehyde (liquid or powder) (35).

Case-referent studies
In the case-referent study by Gerin et al (38,39) information obtained in an interview of the subjects consisted of a detailed occupational history including raw materials and final products, type of workroom, activities of other nearby workers, and presence of gaseous fumes or dusts. A team of hygienists and chemists translated this information into potential exposures (38). For formaldehyde, the authors indicated that the categories of low, medium, and high corresponded roughly to <0.1,0.1-1.0, and> 1.0 ppm. Among the 4249 subjects interviewed, 760 (18 0J0) were judged to have low exposures , 203 moderate exposures (5 070) , and 8 (0.2 % ) high exposures .
Fayerweather et al (40) supplemented the work history information from the personnel records of seven DuPont plant s with information in the medical records and from interviews with co-workers. Available monitoring results were used in the estimation of levels for "continuous" exposure. Wh en monitoring data were nonexistent, int erviews of long-term employees were made with regard to sensory perceptions and proce ss descriptions and control cha nges. The followin g three levels of "co ntinuous" expo sure were identified: < 0.1 , 0.1-2.0 , and 2: 2.0 ppm. Intermittently exposed jobs were classified as low (peaks < 2 ppm) or high (> 2 ppm) exposure . Other chemicals which were used included asbestos, benzene , ca rbon tetrachloride, zinc chromate, and acrylonitrile.
In most of the other case-referent studies industrial hygienists classified the jobs held by subjects into semiquantitative cat egories (41,42) or nominal categories (43)(44)(45). Estimates of possible level of expo sure were not made . In man y cases the subjects were exposed to other chemicals, including known or potential carcinogens (41)(42)(43)(44)(45). One study had two industrial hygienists independently evaluate expo sure, and risks of nasal cancer were calculated on th e basis of each assessment (42) .
In summ a ry, several facets regarding the expo sure assessments used in these stud ies are worth noting . The studies of professionals and the case-referent investigations suffer from typical limitat ions in that exposure is based entirel y on job title without detailed information on the facility. Absence of in formation on levels of exposure in these investigations prevent ed evaluation of do se-response relat ion ships. Moreo ver, in most of the studies of profession s, other expo sures were ignored. Among case-referent studies, those by Gerin et al (39) and Fayerweather et al (40) are stronger because the y assembl ed more information regarding exposures and developed estimates of expo sure levels. The expo sur e assessments in the cohort studie s of industrial worke rs are generally better because actual site visits to the plants under stud y were made by industrial hygienists. In addition monit oring data were often available. With these data, quant itative estimates were developed and used in exposure-response analyses. Only two of the industrial studies included information on exposure oth er than fo rmaldehyde (24,31) .

Summary of epidemiologic results
Tables 3 and 4 display result s from the epid emiologic studies. To provide a comprehensive review, cancers that were excessive in two or more studies or showed some expo sure-response pattern are included in the tabul ation s (II sites). The ob served and expected numbers present ed are generally for expo sed persons only. For some studi es, the expected numbers were not pro vided but could be calculated from the data in the pape r. The expected numbers from case-referent studies were obtained by dividing the number of exposed cases by the odds ratio.
The observed and expected numb ers were summed for studies of professional and industrial groups separately to create combined relati ve risk (CRR) estima tes. This summation approach weights the risk estimat es by study size. We also present the proportion of the studies hav ing relati ve risks greater than 1.0, an approach which weights the contribution from each stu dy equall y. Th e sta tistical significance of the CRR estimates was assessed acco rding to Bailar & Ederer (46).
The studies of indu str ial workers from the United Stat es were not ent irely independent, but total overlapping of the study populations did not occur because criteria for entry into the cohorts varied from study to study . In addition, the detail to which exposure to formaldehyde was estimated also varied among the overlapping studies. Stud ies by Marsh (35) and Liebling et al (37) were on a plant that was also included in the study by Blair et al (25). The plants studied by Won g (36) and Fayerweather et al (40) also included some workers that were studied by Blair et al (25). Two plants were included in both studies by Sta yner (33,34). In the summation of the observed and expected numbers, howe ver, results from the earlier study by Sta yner (33) and the studies by Marsh (35), Fayerweath er et al (40), Wong (36), and Liebling et al (37) were not included because they were generally covered in the lat er study by Stayner et al (34) or in the study by Blair et al (25) . Thu s there is no double counting of cancer deaths in the summary sta tistics pro vided in this report.
Amon g profes sionals (tables 3 and 4), significant excesses occurred for leukemia (CRR 1.6) (II of the 13 investigations showing excesses ran ging from 1.1 to 3.1) , brain cancer (CRR 1.5) (six of nine studies showing excesses rang ing from 1.2 to 3.3), and colon ca ncer (CRR 1.3) (excesses in seven of nine studies rangin g from 1.1 to 2.3). Fewer deaths from lung cancer occurred among the professionals than expected (CRR 0.9), nine of 15 studies showing a deficit. Overall, mortalit y from lung cancer among embalmers and funeral directors was about as expected (CRR 1.0), in contrast to the striking deficit among anatomists and pathologists (CRR 0.3). Onl y one death from nasal cancer occurr ed among the professionals, less than one death being expected. Observed and expected numbers were presented for nasopharyngeal cancer in onl y one report , and excesses were observed among black and white men (CRR 2.2). There was an overall deficit for Hodgkin's disease (CRR 0.5) (based on six deaths). Mortality from pro state cancer and for cancer of th e buccal cavity and pharynx was about as expected (CRR 1.0).
In contrast to professionals, indu str ial workers did not show elevated mortality from leukemi a (CRR 1.1) Table3. Mortality from canc ers o f the buccal cavity and pharyn x , na sopharynx , lung, nose, prostate, and bladder among per-   Table 4. Mortality from cancers of the brain, colon and skin , leukemia, and Hodgkin's disease among persons exposed to tate (CRR 1.0). Overall, fewer deaths occurred than expected for Hodgkin's disease (CRR 0.8) and for cancers of the colon (CRR 0.9) and skin (CRR 0.9). The risk of nasal cancer was evaluated by exposure level or duration of expo sure to formaldehyde (table 5). When the observed and expected numbers from the various studies were combined, no exposure-response gradient was evident from the CRR values. Of the four studies with sufficient information for an exposure-response evaluation, only one had a positive trend (42). On the other hand, for nasopharyngeal cancer , the CRR values rose to 2.1 in the high-exposure category (table 6), a trend which was statistically significant (X =2.02). All three of the studies with exposure-response data showed a rising risk of nasopharyngeal cancer with increasing exposure.
The relative risks for lung cancer by time since first exposure (latency) is shown in table 7. Only three studi es of professionals had information on latency.
The CRR values for lung cancer in the shorter and longer latency categories for these studies combined were 0.6 and 0.9, respectively. Risk did not vary for the embalmers by length of time since first exposure (47,48), but it did for the anatomists (49). Among the industrial workers, the CRR was 1.1 for lung cancer for the shorter latency group and 1.2 for the longer latenc y group . Three studies showed higher lung cancer risks with longer latency (23,25,31), and three b Lower level = all exp osed , high er leve l = p robab ly exposed t o high leve ls for at least one year. C On a scal e of 1-9, lower level = 1-2 and higher level = 2:3 (all among th ose exposed to lower levels of wood dusts) .
d Exposu re estimat es fr om industrial hyg ienist A. e Ex pos ure es ti mates from industrt at hygienist B. f Lower level ver sus me diu m to high level s. 9 Usin g data f rom in d ustr ial hyg ienist B for H ayes et al (42). , P ,;0.05. a Unexposed = <0.5 ppm-ye ars, lower level = 0.05-< 5.5 ppm-years, higher level = 2:5.5 ppm-ye ars (all amon g th ose also exp os ed t o part ic ulate s).
b Lower level = probab ly exposed, hi g her levels = probab ly exp osed to some hig her levels 2:20 year s befo re dea th.
C Rel ati ve exp os ure sca le, low = lo w , high = medium and hi gh.

388
showed higher risks with shorter latency (34,43,50). Risk of lung cancer by level or duration of expo sure to formaldehyde is shown in table 8. Am on g the professionals, fewer death s from lung cancer occu rred tha n expec ted in both categories, and th e relative risk was lower in the higher exposu re category (CRR 0.7) tha n in the lower expos ure category (CRR 1.0). In the studies of ind ustri al workers there was a sma ll, but statistically significant, excess of lung cancer amo ng worke rs with sho rter duration or lower levels of exposure (CRR 1.2), but not amo ng the longer or mor e heavily exposed (CRR 1.0) workers . Usin g duration as the only measure of expo sur e resulted in a C RR of 1.3 for the shor ter duration category and 0.8 fo r the lon ger one. Th e C RR values for the low and high categor ies based on level of exposure were identical (1. 1 an d l.l , respec tively), however. Neither th e ind ividual studies nor the summary statistics showed significant tr end s between level o f formaldehyde expo sure and risk of lun g cancer. f Lower = all exposed; higher = high exposure. 9 Lower = >0-<0.5 ppm-years; higher= ,,5.5 ppm-years. h Lower = >3 months-<3 years in duration; higher= ,,10 years in duration. i Lower= <10 years of exposure; higher= ,,10 years of exposure. j Lower=0.1-1 ppm; higher= ,,1 ppm. k Lower level = ,,10 years of exposure at low levels; higher level = ,,10 years of exposure at high levels .

Discussion
As noted in earlier reviews, professional and industrial workers exposed to formaldehyde have exhibited different cancer patterns (4,5,22,51). The excesses for leukemia and cancers of the colon and brain observed among professionals were not seen among industrial workers. The excesses were rather consistent for leukemia (I I of 13 studies), brain cancer (six of nine studies), and colon cancer (seven of nine studies) among professionals. Studies lacking such excesses had few reported deaths (generally less than three). Diagnostic bias (52) may be a concern for these cancers, since professionals may receive better medical care than the general population which served as the reference in the investigations. Among anatomists (49), however, the relative risks for leukemia and brain cancer remained elevated when mortality rates from psychiatrists were used to generate the expected numbers, a finding suggesting that the excess risks for anatomists cannot be entirely explained by diagnostic bias or general socioeconomic bias. In the study of anatomists (49) the SMR decreased but remained elevated when expected deaths from benign tumors of the nervous system (conditions likely to be misdiagnosed) were included in the estimate of expected brain cancers, again a finding suggesting that the excess is unlikely to be entirely due to diagnostic bias. Industrial workers appear to have TWA values of formaldehyde similar to those of professional workers. The lack of excesses of leukemia or cancer of the brain or colon among industrial workers would seem to indicate that formaldehyde is not contributing to the excesses of these tumors. Experimental investigations also suggest that formaldehyde is unlikely to induce cancer in tissues other than those in which direct contact can occur. The formaldehyde concentration was not increased in the blood of rats exposed to 14.4 ppm (53), and e 4 C)-and eH)-formaldehyde was not found in the bone marrow in labeling experiments (54). Professionals may, however, experience higher peak exposures on a more frequent basis than industrial workers. This possibility may be important if, during high peak exposures, body defenses are overwhelmed (55). In balance, these data suggest that the excesses for leukemia and brain cancer among professionals cannot be entirely explained by socioeconomic confounding or diagnostic bias, but the factors in the work environment responsible for these excesses have not been identified. Pathologists, anatomists, and embalmers differ from industrial workers in that they may be exposed to other substances used in the preserva-tion of tissues, including dyes and stains, glycol, and xylene, as well as biologic materials (14)(15)(16)(17)(18)(19)(20)(21). Additional investigations are needed to determine if these exposures alone or in combination with formaldehyde contribute to the cancer excesses noted among professionals.
Among industrial workers, cancers of the respiratory system have received the most attention. A small, but statistically significant, excess occurred for cancer of the lung (CRR 1.1), and an overall exposureresponse gradient occurred for cancer of the nasopharynx. There was a small overall nonsignificant excess for nasal cancer (CRR 1.1), but the risk of this cancer did not increase with increasing duration or level of exposure. Several lines of evidence suggest, however, that formaldehyde may be involved in the development of this cancer in humans. First, two casereferent studies noted excess risks among persons exposed to formaldehyde (42,44). The occurrence of an exposure-response gradient from the two independent assessments of exposure in the study from The Netherlands (42) is of particular interest. Exposure assessment is especially difficult in case-referent studies . Random exposure misclassification is to be expected, but it would tend to dilute exposure-response patterns. Similar estimates of risk from independent assessments of exposure increases one's confidence in the findings . Second, in the investigation from The Netherlands (42), squamous cell cancer was the histological type of nasal cancer most strongly associated with formaldehyde, and this was the type of cancer seen among rodents exposed to formaldehyde (I, 2). The Danish case-referent study, however, did not find that exposure to formaldehyde was particularly associated with squamous cell cancer of the nose (56). Third, hyperplasia and squamous cell metaplasia of the nasal mucosa, reported among woodworkers (57) and chemical workers exposed to formaldehyde (58), correspond to lesions observed among rodents in bioassays (I , 2) and lend biologic credibility to the hypothesis that formaldehyde is a risk factor for nasal cancer in humans. Potential confounding by other occupational factors remains a problem, however.
In the case-referent studies for nasal cancer (42, 44, 59) many workers exposed to formaldehyde may have also had exposure to wood dust , a recognized nasal carcinogen, and it is unclear whether its potential effects were entirely removed. In studies in Denmark (44) and The Netherlands (42), however, elevated risks occurred among persons who held jobs for which exposure to wood dusts was less likely. Other casereferent studies in the United States (60) and Nordic countries (61) found no association between formaldehyde exposure and nasal cancer, but they were not specifically designed to assess risks from this chemical. Several cohort studies of industrial workers or professional workers have also failed to detect excesses of nasal cancer (23,25,34) although the power of the studies was small.

390
The excesses for cancer of the nasopharynx occurred in a cohort study (relative risk 3.0) and in case-referent studies (CRR 1.2, based on two studies). In the cohort study (28) relative risks for nasopharyngeal cancer increased steadily to 7.S-fold among workers who had cumulative exposure to formaldehyde of 2: 5.5 ppm-years and exposure to formaldehyde-containing particulates. No exposure-response relationship was found, however, among persons who were exposed to formaldehyde but not exposed to particulates. In a case-referent study (45,62) in which exposure to formaldehyde was assessed from workplace and residential sources, the relative risks for nasopharyngeal cancer rose from 1.7 for persons with occupational exposure alone, to 2.8 for persons with residential exposures alone, to 6.7 for persons with both occupational and residential exposures (45,62). These relative risks were adjusted for ethnic origin and cigarette smoking. In the other case-referent study (59) the relative risk of nasopharyngeal cancer rose with probable level of exposure to 2.3 for those expo sed at higher levels 20 or more years before death. Contradictory results regarding nasopharyngeal cancer were obtained in other studies. No excesses occurred in most studies of professionals and no deaths from cancer of the nasopharynx were reported in the study of chemical workers in Great Britain (23) or in the investigation of garment workers in the United States (34). In these investigations, however, the expected numbers were small. The excess of nasopharyngeal cancer among black and white male embalmers in a recent study from the United States is the first report among professionals and further points to a role for formaldehyde in the origin of this tumor (63). In Denmark, a case-referent study noted an association between formaldehyde exposure and. nasopharyngeal cancer among women, but not among men (44). Despite these inconsistencies, biologic plausibility (excesses for a site where direct contact may occur) and the occurrence of exposure-response patterns in studies of different designs (cohort mortality and case-referent) suggest that a causal association between formaldehyde exposure and cancer of the nasopharynx seems credible and likely. As with nasal cancer, small numbers, inconsistency among studies, and a po ssible independent role for particulates preclude definiti vely labeling formaldehyde as a nasopharyngeal carcinogen .
The findings regarding lung cancer are difficult to interpret. The significant deficits of lung cancer that occurred among anatomists and pathologists were very likely due to a lower prevalence of smoking among these health professionals (49,64,65) than among the general population (66). Smoking habits among embalmers and funeral directors appear, however , to resemble that of the general population. In a study of United States veterans (67) the proportion of embalmers who ever smoked cigarettes (57 !tlo) did not differ from that of the entire study population (58 070) .
Smoking, therefore , does not appear to confound com-parisons between embalmers and the general population. Embalmers, however, do not appear to experience an excessof lung cancer (CRR 1.0). In addition, among funeral directors and embalmers, the risk of lung cancer did not increase with time since first exposure.
Industrial workers had a small excess of lung cancer (CRR 1.1), but the overall relative risks increased little with latency. Neither did the risk of lung cancer increase with duration or level of exposure to formaldehyde among industrial workers , findings resembling those of studies of embalmers and funeral directors.
The CRR values decreased with increasing duration of exposure and were similar in the low-and highexposure categories.
In the studies of multiple plants in the United States (24,25,28,29,33,34) and Great Britain (23, 68) excess mortality from lung cancer was not observed in all the plants. Among garment workers (34) in the United States, deficits were observed at two plants, whereas one had an excess. In the study of workers at 10 plants producing and using formaldehyde (24,25) an excess of lung cancer occurred among workers with 2: 20 years of latency for six plants, and deficits occurred for four plants. The plants with the excesses were not necessarily the plants with the highest levels of exposure. In Great Britain (23) three plants had deficits, and two had excesses. One of the plants with an excess had the highest exposure levels in the study and showed an exposure-response relationship (the SMR rising to 118 for the most heavily exposed). The other plant with an overall excess showed no such exposureresponse relationship and had levels similar to those found in the other plants in the study .
In both the American and British studies of chemical workers (23)(24)(25) the lung cancer excess was largely confined to workers involved in the production of resins and molding compounds. In the American study, lung cancer was not elevated among worker s exposed to formaldehyde unless they also came into contact with sub stances associated with the production of resins and molding compounds (24). The study of workers in an Italian resin plant also noted an excess of lung cancer, but it did not appear to be directly related to formaldehyde exposure (50). An update (69) of the Finnish cohort (31) noted an excess of cancer of the upper respiratory tract but not an excess of lung cancer.
Some investigators have speculated that the healthy worker effect may contribute to the lack of an exposure-response gradient for lung cancer (70,71). Robin s et al (70), using an analysis which controlled for this problem , concluded that the healthy worker effect was not an explanation for the failure to observe an exposure-response pattern in the Blair et al study (25). Sterling & Weinkam (71), in another reanaly sis of these data, confirmed the excess of lung cancer reported by the original authors (25) but concluded that it was caused by formaldehyde. An error in the counting of deaths resulted in an inclusion of approximate-Iy three times as many deaths in the analyses as existed in the cohort (29). In a letter addressing this problem (72) Sterling & Weinkam reported an excessof lung cancer that rose with level of exposure. In these analyses, however, more deaths from lung cancer occurred than were noted in the original analysis by Blair et al (25). These new analyses (72) may have included contributing causes of death from lung cancer, but contributing causes do not appear to be included in the analyses of all cancers combined.
Based on findings from all the reviewed studies, a causal association between exposure to formaldehyde and lung cancer cannot be entirely discounted, but many of the traditional criteria for causality in epidemiologic investigations are missing. The excess risk was small among industrial workers (CRR 1.1) and was not seen consistently in all the plants studied. In some plants there were deficits . No such excess for lung cancer occurred among the embalmers. Overall, there was a slight increase in the risk of lung cancer with time since first exposure among the indu strial workers, but thi s pattern was also inconsistent by study. Risks increased with latency in two of the three studies of professionals which included information on latency effects, but in only two of the six studies of industrial workers. A causal association was not suggested by the mortality pattern either by level or by duration of exposure. In three studies among professionals, risks were lower for the workers exposed to formaldehyde for longer durations than for those with shorter durations, and in the industry studies the risks were lower among the heavier or longer exposed workers in four of six studies. Furthermore, elevated mortality from lung cancer among workers in certain operations involving resins and molding compounds suggests that further evaluation of exposures in this process in the American, British, and Italian cohorts may help clarify the role of formaldehyde and other substances in the excess of lung cancer seen in some of these studies .
In summary, we conclude that it is likely that the excesses of nasopharyngeal cancer observed were caused by exposure to formaldehyde . The association with nasal cancer is plausible, but somewhat less persuasive t-han that for nasopharyngeal cancer. The absence of excesses for leukemia and cancers of the colon and brain among indu strial workers suggests that the associations seen among professional workers may not be due to formaldehyde . Although a role for formaldehyde in the excess of lung cancer cannot be dismissed, inconsistencies among and within studies of industrial workers suggest that this association is not causal; however , other interpretations are possible. Formaldehyde may be an effective carcinogen only in the presence of other exposures, which were not consistent from study to study, or it may be a weak carcinogen whose effect was easily masked by the presence of lung carcinogens that varied from study to study.