Exposure to formaldehyde and cancer mortality in a cohort of workers producing resins.

to and cancer mortality in a cohort of workers Scand The mortality of I 332 male workers employed at least 30 days in 1959-1980 in a resins-manufacturing plant was examined . Ambient measurements taken in the plant between 1974 and 1979 documented a potential for exposure to levels of formaldehyde as high as ~ 3.0 mg/rn-. Vital status was ascertained for 98.6 % of the cohort members, and their mortality was compared with expected deaths drawn from the national and local population rates. A statistically significant increase in lung cancer was observed, based on 18deaths, which was not fully accounted for by possible confounding factors linked to personal habits or sociocultural characteristics. This elevated risk, however, could not be attributed specifically to exposure to formaldehyde. Mortality from digestive cancer (14 deaths observed) and hematologic neoplasms (5 deaths observed) was not substantially higher than expected . The study was limited by in formati on bias due to incomplete work histories, by the small number of deaths from some relevant causes, and by the possibly insufficient length of the observation period . Therefore these results do not provide sufficient grounds either to link formaldehyde exposure in the plant to the increased cancer risk noted or to exclude that formaldehyde might pose such a risk to humans under certain exposure circumstances.

Formaldehyde (HCHO) is an almo st ubiquitous compound in the natural and man-made environment, and a normal metabolic product in most life form s. Its chemical identification and preparation dates back to the middle of the past century. The estimated current worldwide volume of formaldehyde as a commodity chemical is over 4.5 million tons per year. Exposure to formaldehyde has been documented in a large variety of industrial settings, ranging from medical and laboratory environments to agriculture and wood, plywood and particle board, textile, and paper industries. In the chemical industry, exposure to formaldehyde occurs in rubber and explosives manufacturing, tanning , dying, etc; the major (and increasing) use is in plastics production (eg, urea-, phenol-, and melamine-formaldehyde resins) (9).
Adverse health effects on the human respiratory tract and the skin associated with exposure to formaldeh yde have been reported for a rather long time (5,15). Recently, results of experimental inhalation studies performed in different laboratories showed that formaldehyde was carcinogenic. Squamous cell carcinomas of the nasal cavity were induced in mice and rats with a sharp dose-response relationship (3,12). Such reports prompted new investigat ions on the long-te rm effects possibly associated with hum an exposure to formaldehyde. Epidemiologic studie s were carried out in 1 Institute of Occupational Health "Clinica del Lavoro L Devoto, " University of Milan , Milan, Italy.
Reprint requests to : Dr PA Bertazzi, Istituto di Medicina del Lavoro , Un iversita ' degli Stud i, Via San Barnaba 8, 1-20122 Milano , Ital y. various occupationally exposed popu lations . Different designs were used , ie, proportional mortality ra tio (PMR) studies (14,16,20,22), standardized mortality ratio (SMR) studies (I, 7,8,13,23), and case-referent studie s (4, II, 17,18). Their results have been inconclusive so far. Case reports have also been published which yielded contrasting results (6,10). Given such background, further epidemiologic researc h seemed full y ju stified . This study was initiated upon a joint request of the management and unions of a plant manufa cturing resins, where a reportedly high number of cancer deaths had occurred in recent years. The stud y was designed as a cohort mortality investigation covering the period from the beginning of production up to the end of 1980. Although the period of time elapsed from the beginning of production of formaldehyde-based resins was perhaps not suffic iently long (22 years) and the exposed population not very large , still the study seemed worth conducting as, accord ing to historical environmental data, the potential for human exposure to formaldehyde in similar occupational settings was among the highest (9).

Plant description and exposure
The plant had been manufacturing different types of resins for varying periods of time, starting in 1959. The sole production constantly carried on, and the largest , had been that of urea -and melam ine-formaldehyde resins. According to the workers and plant technicians, there was, at least in past years, a cont inuous pungent odor of formaldehyde in the ambient air of the area where formaldehyde-based resins were produced during the entire workshift. The olfactory threshold for formaldehyde is 0.05 ppm, and mild irritating symptoms have been described in association with a wide range of airborne concentrations (from 0.1 to several parts per million) depending on the duration of exposure (5). The second largest production was styrenebased resins (polystyrene and copolymers). A third, and much more limited, production involving a potential for exposure to a suspected carcinogen (epichlorohydrin) was epoxy resins. Other resins manufactured in the plant were polyesters and colophony resins. Each type of production occurred in a separate building, with the exception of epoxy and colophony resins. In several production areas there was a potential for exposure to various solvents, eg, xylene, toluene, methyl isobutyl ketone, etc.
Few environmental measurements of airborne contaminants had been taken between 1974 and 1979. Table I presents the formaldehyde results, which were judged by plant technicians as "typical" ambient air concentrations in those years. Technological changes and improvements of the production processes were undertaken only after 1975. It is rea sonable, then, to assume that, in preceding years, expo sure to the mentioned compounds had been even higher.

Study population and follow-up
All male workers ever employed in the plant at any time and in any department between 1959 and 1980 for a period of at least 30 days were considered eligible for the study. Meeting these criteria were I 332 For each worker the plant Personnel Office was able to suppl y (besides personal data such as name, residence, date of birth , etc) only the date of hire and date of resignation. Specific jobs held in the plant were comprehensi vely recorded only for current employees. Reconstruction of work history for past employees was thu s attempted in interviews of retired workers still living in the area, current workers employed for a long time, and foremen. A work history was considered acceptable only after the data were double-checked with the Personnel Office, which provided information according to its best knowledge and the existing, although incomplete, records. In this way, work histories were collected for all but 18 070 of the workers in the study.
The fixed area measurements available could not be used for estimating the job-specific exposure of individual workers, either because of their very nature (area samples) or because of their limited number.
Notwithstanding the limited information, workers were subdi vided into the following three exposure categories: " exposed to formaldeh yde," "exposed to other compounds," " exposure unknown. " Apart from the group with unknown exposure, the classification of workers according to their having or having not been operating in jobs with exposure to formaldeh yde was considered acceptable for the study purposes, thanks to the fact that the job mobility in the plant turned out to have been rather low. As a matter of fact , 79 % of the workers in the study had been employed in a single job during their career in the plant, and almost all workers engaged in the formaldehyde-based resins production had not spent relevant periods of time in other departments.

Analysis
The information collected was analyzed with a computer program specifically developed for cohort studies (24). Each subje ct contributed to the person-year calculation from the day he reached the minimum defined exposure period (30 da ys) until the end of the stu dy (31 December 1980) or , if deceased, until the date of death . Expected deaths were calculated from two sets of mortality rates (national and local) specific by cause, sex, age (five years), and calendar time (five years) . Local rate s were used in an attempt to control at least for the major , possibly confounding, social , cultural, and geographic factors. The rat ios o f ob served to expected deaths x 100(SMR) were then calculated . Fisher exact confidence limits were computed on the assumption of a Poisson distribution for the observed deaths, used in calculating the expected number of deaths. The overall mort ality did not depart significantly from ex-

Status Number Percentage
Adm itted 1 332 100.0 Th e follow-up was successful for 98.6 0J0 of the stu dy popul ation, and 112 workers had died durin g the study period (table 2). A total of 20 336 person-years of observation were accumul ated. The comparison of workers' mortality from selected causes with the expected figures drawn from the national and local rates yielded the results shown in table 3. A changing pattern of the observed: expected ratios is appa rent , depending on which set of rates was with publi shed desk calculator programs (19). SMR values were also employed to compare mortality across the thr ee subcohorts with different exposure, although the person-years distribu tions were somehow different. From figure 1, one can note that work ers exposed to form aldeh yde were, on the average, a little older but, in ou r judgement , not enough so as to invalidate the analysis. Th e compariso n carrie d out after the direct standardization of the rat es supported this view. pectations. Deaths due to cancer (all sites) were significantly higher than expected onl y in comparison with the national rates. Among cancers of specific sites, the 18 lung cancer deaths ob served made up a significa nt excess above expectations from either reference population. Hematologic neoplasms were from 1.5 to 2.0 time s more numerous than expected , but the increase was not sta tistica lly significant. The suggestion of an   increased mortality from cancer of the alimentary tract result ing from the comparison with the national rates disappeared when the ob served deaths were compared with deaths expected from the local population rates. Deaths due to nonmalignant diseases of the cardiovascular and alimentary systems were fewer than expected. Mortality from resp iratory diseases was similar to both the nat ional and local expected figure s. Mortality from external causes (accidents, violence, etc) was slightly increased above expectations. Table 4 presents the result s of the anal ysis by year since first expo sure for selected cancer cau ses. No definite trends were visible according to the increasing number of years since th e beginning of expo sure (latency); in particular, the increased lung cancer mortalit y seemed to be evenly distributed across the categories. Analysis by length of employment in the plant (table 5) showed a higher than expected mortality from all cancers and can cer of the lung in the catego ries of shortest duration . A weak sugge stion of a risk increasing with duration of employment was obtained for cancers of the alimentary tract.
The cohort was then split into expo sure categories. Th e results of the analysis regarding these categories and selected can cer sites are reported in table 6. The SMR values of the workers known to ha ve been exposed to formaldeh yde were consistently elevated , but none of them significantly. Workers exposed to chemicals other than formaldehyde in the plant experienced a cancer mortality lower than expected , with the sole exception of lung ca ncer. Workers who se expo sure could not be ascertained exh ibited th e mo st elevated risk of d ying from cancer , and the increased SMR for lung cancer turned out to be statistically significant. As already mentioned, th e person-years distribution in the groups seemed to allow the comparison of their SMR values. If we ignore, for the moment, workers with unknown expo sure, it is apparent that form aldehyde wo rkers experienced an all cancer mortality more th an twice as high as that of the workers with oth er types of exposure. Th is increase was mainly due to digestive cancer and hematologic neoplasms. There was no differen ce, instead , between the two groups with regard to lung cancer mortality. The ratio of the "directly" standardized (standard = total cohort) mortalit y rates (SRR) was also calculated in Table 9. Mortality from selected cancer causes in the subcohorts composed unde r the alternative assumption that all workers with unknown exposure had actually been exposed to formaldehyde (formaldehyde plus unknown) or that they all had had other exposure (o t her plus unknown). Further analysis was carried out on the subcohort of formaldehyde-exposed workers. Results by years since first expo sure are shown in table 7. Admittedly, with the low number of observed deaths, if a pattern is discernible , it is a decreasing one with increasing time since the beginning of exposure. Even analysis by duration of expo sure (table 8) failed to reveal an y definite pattern or trend; again, the numbers were small.
In the analysis stage, an attempt was made to overcome the bias due to the lack of information on the expo sure history of a number of workers by making "extreme" assumptions, ie, that all workers with unknown expo sure had in fact or , alternatively, had not been exposed to formaldehyde . The results of this anal ysis are shown in table 9. As expected, attributing all workers with unknown exposure to the subcohort of workers exposed to formaldehyde produced a further increase in the SMR values of this group . (For comparison see table 6.) And the excess risk from lung cancer reached statistical significance. The addition of these workers to the subcohort with "other" exposure produced an analogous increase of the relevant risk estimates with no change, however, in the direction (above or below 100) of the SMR values. (See table  6.) Even in this exposure group, under this assumption, the increased lung cancer risk became statistically significant.

Discussion
During the study period the examined cohort of resins production workers experienced a lung cancer mortality significantly higher than expected, in comparison either with the national population of the same sex and birth cohort or with the local one. The comparison with the local rates makes it unlikely that personal habits or environmental or sociocultural factors could explain the increased risk noted. On the other hand, no definite association of the increased lung cancer risk with either duration of employment or years since first employment in the factory was found .
An excessive number of deaths from cancer of the alimentary tract and from hematologic neoplasms was also noted in the entire cohort. Howe ver, the SMR values were not statistically significant, and the excesses became very modest or disappeared when the expected deaths were calculated from local rates. The analysis by latenc y and duration of emplo yment failed to suggest an association between employment in the plant and an increased risk from such types of cancer.
No cases of nasal cancer were recorded on the death certificates; the expected figure was extremely low however (0.0327 expected).
Expo sure in the plant was, as mentioned, complex, and the available exposure histories were not complete or exhaustive. It was, nonetheless , possible to identify workers who had held only jobs with exposure to formaldehyde. The comparison of the mortality of this subcohort with the local rates suggested an increased risk from cancer (all sites), cancer of the alimentary tract, lung cancer, and hematologic neoplasms. The increased SMR values were not however statistically significant, nor did the analysis by duration of exposure to formaldehyde and latency reveal clear-cut trends. Scanty significant results were obtained, but they could well be explained by chance alone, given the high number of comparisons performed.
The internal comparison of the formaldehydeexposed workers against the workers in the same plant with known exposure of other types showed no difference in lung cancer mortality, while mortality from cancer (all sites) and from digestive cancer was significantly higher among the formaldehyde-exposed workers. An elevated risk ratio was also noted for hematologic neoplasms. It should be mentioned that the very low number of observed deaths makes the estimates of the "directly" standardized mortality rates 466 unstable. In addition the comparison of workers exposed to formaldehyde with worker s with other exposure was certainly affected by the information bias due to the lack of detailed work histor ies for a number of former employees.
The assumption that workers with unknown exposure had actually been exposed to formaldehyde or, alternatively, to other compounds did not provide any definite clue to the interpretation of the results. In fact, a significantly elevated risk from lung cancer was obtained in both exposure groups after the workers with unknown exposure were added. Instead, under the conservative assumption that they all had not been exposed to formaldeh yde, one could derive the suggestion of an association of digestive and hematologic malignancies with formaldehyde exposure. (Compare the SMR values in the subcohort of formaldehyde-exposed employees in table 6 with those of the subcohort composed of workers with other plus unknown exposure in table 9.) However such a suggestion is not supported by any other result of our study.
Among the epidemiologic studies so far published, only one, carried out on British chemical workers, provided a very weak suggestion of a possibly increased lung cancer risk (I), but its relationship with formaldehyde exposure turned out to be highly improbable after thorough anal ysis (2). Other studies suggested an elevation of the cancer risk of the upper respiratory tract in several occupational cohorts with a potential for exposure to formaldehyde (14,17,20), but none could draw definite conclusions about the specific ity of the association. As regards cancer of other sites, an increased risk of cancer of the aliment ary tra ct was suggested by epidemiologic studies of embalmers (22), chemical workers (14, l6) , and perm anent-press garment workers (20); and the risk of hematologic neoplasms was elevated in two investigations (8,20). All these studies suffered from limitations such as confounding by other exposure, small sample size, etc. The interpretation of these findings is further hindered by the lack of a biologically plausible explanation of the effect of formaldehyde on these body sites, while a sound hypothesis exists for the possible carcinogenic effect of formaldehyde on the upper respiratory tract (21). Other studies failed to reveal any increased risk of cancer (4, I I, 13,18). Despite their different design, all the studies were affected by limitations such as mixed expo sure, short latency, low statistical power, short duration of exposure or observation period, and so on . Thu s, despite their being "negative," they were not able to demonstrate conclusively that an increased cancer risk from formaldeh yde exposure did not actually exist.
Our study evidenced a significantly increased risk of lung cancer in a plant with multiple exposure s, where con centrations of formaldeh yde abo ve 3.0 mg/rn! were not infrequent , particularly in the past. However, no sufficient evidence emerged, either from the external or from the internal comparison per-formed , capable of specifically associating the excessive lung ca nce r risk observed with exposure to formaldehyde. The lung cancer excess observed in the entire cohort stem med mainly from the mortality experience of workers wi t h unknown exposure, who, as a group, exhi bited a rather elevated risk from an y of the rele vant causes o f death. This phenomenon could be a reflectio n of the selective feature of their being sho rt-ter m employees (90 070 of the person-yea rs accumulated by this su bc oh o rt was in the employm ent duration category of < 5 yea rs) rather than a cons eq ue nce of the ir speci fic, alt hough not ascertained, exposure. The interp ret ation of the rather modestl y increased risk of di gestive cancer and hematologic neoplasms remained in co ncl usive , although suggestions emerged of its po ssible a ssociation with formaldehyde exposure. Thus , even in our study, the sm a ll number of deaths from so m e of the relevant causes , the approximate quantification and loose categorization of the individual exposure histories, along with a possibly still limited length of the follow-up period, pr evented any firm conclusi on from being drawn about the hypothesis test ed .
Finally, a few com m en ts regarding other possibly re levant ex p os ures in the plant. -Long-term health effec ts possibly linked to sty re ne and epichlorohydrin were not in vestigated fo r seve ra l reasons , the major ones being the specific hypothesis of th is study and the limited number of workers ex p os ed to these compounds . The very low job mobility of the formaldehyde workers renders it improbable th at these co mpounds caused an appreciable confounding effect in the results presented and di scu ssed. It sho u ld be m entioned that th e results for worker s with exposure other than formaldehyde did ac t ua lly sugg es t a moderately increased risk from lung ca ncer (ta ble 6), but too many d ifferent jobs, a nd hence d ifferent exposure s, were included in th is category to allow a n y h ypothesis to be drawn from the findings re garding sty rene or epich lo ro hyd rin effects.
The stu d y is now continuing by means of an extension of the follow-up period and a sea rch for more detailed and exhaustive information on ind ividu a l exposures . Since several independent st u di es are still under wa y in different countries, we trust that in a few yea rs their results will co ncur in clarifying th e possible carci nogenic risk, if an y, posed by fo rm aldehyd e to man under certain exposu re co n d itio ns .