Lung cancer, smoking, and employment in foundries

Health 1989;15:38-42. A case-referent study on lung cancer was conducted 10 Cracow, Men dying of lung cancer within a 6-year period (1980-1985) fonn~d the case group. The reference series was selected from death registers and was fre quency-~atched with the ca~es by sexand age. Deaths due to other respiratory diseases were excluded. Information o~ the oCCup~tlon, smokl~g habits, and residency of 901 cases and 875 referents was col lected fro~ their next.-of-km. Th~ combIn~d eff~ct of smoking and industrial exposure, in particular em ployment.In. ste~1 or Iron foundnes, was investigated by multivariate analyses and was very well fitted by a ~ultlphcatlve model. Foundry.employment, in particular in the youngerage « 70 years) group, oc cupational ex~osure to known carcinogens Inother industries for more than 20 years and smoking were found to be fisk factors.

Whil e the use of tobacco products ha s been identified as a major causa tive factor for lung cancer (I), industrial and other en vironmental exposures are suspected to be contributors to the rising mortality from this disease . Chemical compounds which enter the ambient air from industrial processes ma y enhance the carcinogenicity of tobacco smoke in the lung. In recent years, it has become apparent that an excess in lung cancer risk might be associated with found ry exposures (2)(3)(4), although a relationship between exposure to specific pollutants in foundry air and an increased lun g cancer risk ha s not yet been firmly established.
A pilot study carried out earl ier (5) showed that the Cracow population, in comparison with all of Poland, had an excess of lung cancer deaths among men. Possible explanations include better reporting of deaths due to lung cancer in Cracow, a higher prevalence of smoking, or a greater degree of exposure to occupational hazards. The large case-referent study presented in this report aims at evaluating risk factors of lung cancer in a population with a high degree of exposure from occupations in the iron and steel industry while conI rolling for smoking and other factors. In addition, an att empt has been made to investi gate interactive effect s between foundry empl oyment and cigarette smo ke on lung cancer risk through the use of a thorough multivariate analysis.

Subjects and methods
All men whose death was attributed to primary lung ca ncer among male Cracow residents during 1980-1985 were identified from the death certifica tes. The men formed the case group. The referents were selected fro m th e C racow death registe r as th e first per son of the sa me sex and age ( ± 5 years) entered with a cau se of death other than respiratory cancer or chronic resp iratory disease after each case. The de ath certificat es obta ined for the cases and referents were used to locate the next-of-kin for interview and hospital confirmation of the primary diagnosis.
A stru ctured postal questionnaire was sent to I 273 next-of-kin of the cases and 1 188 next-of-kin of the referent s. If the questionnaire was not returned after one month, a reminder was sent, followed by a second one, if an answer was not rece ived . The subjects from whom the questionnaires were not returned after three contacts were treated as nonrespondents. The response rate was 70.7 070 for the cases and 73.5 070 for the referen ts, ie, responses for 901 cases and 875 referents were received . The response rates were roughly simil ar across the age groups for both the case s and the referents. The data on the type of respondents sho wed th at most of them were surv iving spouses (60 % for cases and 56 % for referents) or o ffspring (31 and 30 % , respectively) . Responses were missing for some of the questions. Although the procedure of selecting the referents wa s linked to the cases on an indi vidual ba sis for organizational purposes, thi s design was used onl y to provide sufficient frequency matching in respect to age and sex.
Th e questionnaire provided data on the deced ent's residential , occupational , and smoking history, as well as demographic and other variables o f interest. The occup ational history covered the following item s: bra nch of industry (eg, iron and steel industry, chemical industry) of longest held job, dur ation of work, job category (manual versus nonmanual), suspected exposure in job and its duration to coal, cement, asbestos du sts, metal dusts and fumes, or ionizing radiation. From these items a categorization into different occupational exposure gro ups was performed as follows: Workers from steel or iro n foun dries were divided into three exposure levels (exposure to one of the aforementioned substances for less than 20 years or an unknown duration , for 20 to 29 years, and for 30 years or more) . Workers outside the steel indu stry who were exposed to the aforementioned substance s for more than 20 years were treated as an "other exposure" category. The remaining group served as the nonexposed base-line category. This last category also included some individuals with missing information. The categorization into the exposure groups was based on information on the duration of exposure rather than on time at work in a specific ind ustry because separate periods of work in non hazardous places within the same factory cou ld then be taken into consideration .
The data concerning smoking habi ts included the year the subject started smokin g, the average number of cigarette s smoked daily, and the duration of smoking. From the smoking history data , the lifetime consumption of tobacco was estimated and expressed in pack-years . (One pack-year is equiva lent to smoking 20 cigarettes daily for one year.) Accordingly, the following th ree gro ups were for med : 1-20 pack-years, 20-40 pack-years, and more than 40 pack -years.
Smokers with unknown lifetime con sumption formed a separate group ("pack-years missing") .
For the statistical analysis, logistic regression models (6) were fitted to the data with the SAS (statistical analysis system) software package for unconditional maximum likelihood estimation of the regression parameters. This analysis is appropriate for a frequency-matched design. All the variables but age were entered as categorical varia bles into the model. The factor age was taken into consideration in some models because the selection proc edu re for the referents, and a slightly higher response ra te in the reference series in the older age groups, resulted in a slightly higher mean age for the reference group . The confidence intervals (Cl) are given on the 95 0J0 level and arc based on the normal approximation of the estimated regression coefficients. The attributable risks were computed according to a met hod suggested by Bruzzi et at (7).

Results
The age distribution of the cases and referents is shown for the entire gro ups and the foundry workers in the groups in table I . The mean age of the cases at death was 63.3 years, and that of the referents was 66.8 years. The next-of-kin of 106 cases (12 0J0) and 72 referents (8 010) reported the steel and iron industry as the usual employer of the deceased persons. The mean age of the persons having worked in this ind ustry was 59.4 years for the cases an d 64.0 years for the referents. Table 2 gives the number of cases and referents in the exposure categories of the foundry workers and the corresponding numbers of those exposed in other industries or transport (truck drivers) for at least 20 years, and of th ose unexposed. Individuals below 70 years of age at death were con sidered separately and therefore their exposure experience is also given.
In table 3 the frequency distribution of the cases and referents is presented for the different occupational categories . Foun dry workers formed the biggest sub-   group, but other frequentl y reported indu strie s were machine con struction, the chemical industry, and the building tr ade . Table 4 shows the relative risk (RR) estimates derived from the final logistic model as the antilog of the regression parameters. The relative risk estimates were excessive for the foundry workers in all the exposure categories , and they showed an increasing trend with duration of expo sure . The highest relative risk (RR 2.66) was determined for the highest exposure category (95 % CI 1.31-5.42). The test of linear trend appeared to be highly significant (P = 0.002). The category "exposure in other industries" also showed a significantly increased relative risk (RR 1.76, 95 % CI 1.37-2.25) . Among the foundry workers three cases and two referents had missing data on duration of exposure. If considered separately, the estimated relative risk of this group was 2.2 (95 % CI 0.25-19.4). In the final model these individuals ha ve been added to the lowest exposure group.
As expected, tobacco smoking was found to be the strongest risk factor. The estimated relati ve risks given in table 4 were found to increa se with the amount of cigarette con sumption. In the final logistic regression models the factor "age at start of smoking < 18 years" was also introduced, as it showed an additional significant effect (RR 1.25, 95 % CI 1.00-2.66).
In order to assess whether the relative risk patterns were more pronounced in the younger age group, we performed an additional analysis fo r the subgroup of individuals under 70 years of age. The ranges of the relative risk estimates for smoking were similar to those of the total group, but the age at the start of smo king had no effect in this subgroup. The risks for the foundry workers were considerably higher in the younger age group ( With the purpose of examining a possible bias arising from the different sources of death certificate diagnosis (hospital versus doctors outside hospitals) and from the various types of informants, we used a variety of additional logistic models. Two different mod els were constructed -one for the hospital-deceased cases and another for the hospital cases that underwent postmortem medical examination. Despite a much smaller number of cases in these two additional models, the general risk pattern for lung cancer remained stable. In all three model s the odds ratio estimates for the effects of smoking and occupational exposure appeared to be similar and significant. There were only small differences in the absolute values of the estimated relative risks .
In an attempt to elucidate a possible role of information bias, another model was constituted in which the source of "respondent" was controlled for (spouses versus others). No effect of type of respondent on the relative risk pattern was found. In order to look for other possible factors that may confound the results, we included two other variables in the model [edu cationallevel and place of birth (rural versus urban)] but neither improved the fit.
We also evaluat ed the possible interactive effects between smoking and occupational exposure among the foundry workers by adding the interaction terms into the final logistic model. The estimates were close to zero and hence indicated that the two factors act multiplicatively. The probability of foundry workers in the highest exposure category also being heavy smokers was estimated to be 32-fold (8.00 x 3.99) that of the unexposed nonsmokers.
On the assumption of a joint distribution of smoking and long-term employment in the steel indust ry the population attributable risk estimates for Cracow were as follows : 72.3 0J0 for smoking, 2.5 070 for foundry employment, and 75.2 070 for all factors, including other occupational exposures for the total male population. For the male population under 70 years of age the corresponding numbers were 75.1,4.8, and 78.0 0J0 respectively. Thus a higher proportion of lung cancer cases in the younger age group may be attributed to either of the risk factors.

Discussion
The main purpose of our study was the assessment of the effect of the occupational hazards due to employment in steel and iron foundrie s. It is necessary to mention that, in Cracow, a metallurgical complex was set up in 1955. In this complex the occupational hazards have not changed substantially over time. In our data we found that, after controlling for age and smoking habits (pack-years and age at the start of smoking), the suspected occupational exposure in the metallurgical industry affected the risk of lung cancer among men significantly. Its effects became more evident after long-term employment, ie, 30 years or more, and were more pronounced in the age group under 70 years. These finding s may be explained by a reduction in risk after the cessation of employment or by an increasing misclassification of exposure information among older respondents . However, in this study, occupational history was not obtained in detail, and thus the time since last exposure could not be taken into consideration. Therefore, a younger aged subgroup was formed, as it may be assumed that the time since last occupational exposure was relatively short in this group and a reduction in lung cancer risk may not yet have taken place. The results were in agreement with the observations of Egan-Baurn et al (8) who found an odd s ratio of 2.36 for iron foundry workers who died before the age of 65 years and an odds ratio of only 1.19 for those who died after this age.
In this study no detailed occupational histories could be obtained, and no attempt was even made to identify specific carcinogenic agents. Instead we proposed to assess the magnitude of risks attributable to occupation in the iron and steel industry in general. Socioeconomic status has been known to be a major determinant oflung cancer, but, in this study, it had no independent effect after the exposure variable was included in the model. Although there are other industries which may involve exposures to the same range of putative carcinogens, the foundry workers formed the largest group, and a detailed analysis of the other groups did not seem feasible. Of some concern was the group of subjects who were reported to have occupational exposure but whose information on the branch of industry was missing. When the relative risk of this group was estimated separately, a value of 2.53, consistent with that of the remaining group, was found.
This case-referent study in Cracow confirmed an increased risk of lung cancer previously reported among steel workers. Blot et al (9) found a significantly increased risk associated with employment in the steel industry. Adjusted for cigarette smoking, the odds ratio associated with career employment was 1.8 (95 0J0 CI 1. 2-2.8). For a subgroup of foundry workers the estimated relative risk was higher, but it was based on very small numbers.
Work in iron and steel foundries has been also linked to lung cancer in studies from several countries (2,4,8,(10)(11)(12). In these studies, the increased risk tended also to be the most pronounced among workers with the longest duration of employment. However, the responsible agent has yet to be identified. Many foundr y processes give rise to various fumes or dusts. These emissions tend to be a very complex mixture of particles, chemicals, and gases, and their chemical and physical characteristics vary with the process from which they arise. Pyrolysis products include polycyclic aromatic hydrocarbons, some of which are known as carcinogens, and airborne particulates may be composed of metals or metal oxides, silica, carbon, or organic matter. Polynuclear hydrocarbons, which occur not only in foundries and near coke ovens, but also in industrial operations involving the conversion of iron to steel, are particularly suspect.
Of some concern in this type of study is the nextof-kin's ability to provide accurate information about the study subjects (13)(14)(15)(16)(17)(18)(19). Lerchen & Samet (20) assessed the validity of surviving wives as a source of information on occupational exposure of their husbands by comparing the histories reported by the wives with those provided earlier by their husbands . Agreement for the usual job, ie, the job held the longest time, was 84 0J0 for the industries and 78 0J0 for the occupations. There was perfect agreement for cigarette smoking status (ever, never); however, wives tended to report 20 cigarettes smoked per day even when their husbands smoked substantially more or less. Pickle et al (21) evaluated the ability of several types of surrogate respondents to provide information on smoking, occupational, medical, and demographic characteristics. Spouses and offspring provided the most accurate information for events that occurred during adult life . Rogot & Reid (22) compared self-reported occupations by British and Norwegian migrants to the United States with responses provided after their deaths by nextof-kin. In a classification scheme that included five broad occupational groups , agreement was 77 0J0 .
In this study the rate of blank responses for various questions ranged from almost 0 0J0 up to 34 0J0. The response rates generally decreased with the amount of details requested and tended to be higher for those subjects over 70 years of age, particularly for questions on educational level and employment. Additional control for type of respondent (spouse versus others) did not result in substantial changes in the regression parameters. We think, however, that missing responses for th ese questions a re more likely to signify " no t applicable " than "don't know" and that an a ssignment into the "unexposed" categ o ry is th erefore appropriate.
In sum m a ry, we conclude that long-term employment in th e stee l or iro n fo u ndry industry subs tan tia lly in creases the risk o f lung ca nce r. T obacco sm o king, however, remains th e dominant ca use due to bo th higher relative risk s and higher prevalence.