Document V3L7neKMrQJnpYrVqNL5GD16j

ii >4) 44:- Uiilvi l.iloycnli.in. Ini nv i> Icnc in >ps. En- I. Lebrel ".ure by .eup En- iman exIcalth 42 schelein rc io n9 (1982) >oper S, al expo unds in I breath. weig M, exposure remenls exhaled alaiion. siology. iecione: Emiron soriable il of incirom 8 i breath :i--22. Physiol . ORIGINAL ARTICLES Scum! J Hiirk Environ Health 12 (1986) 193--202 'Q - (A) - (aJ CftkJ . m1Sm A case-referent study of lung cancer, occupational exposures and smoking I Comparison of title-based and exposure-based occupational information by Helge Kjuus, MD,' Rolv Skjaerven, MSc,s Sverre Lang&rd, MD, MSc,1 Jan T Lien, MD,3 Terje Aamodt, MD4 KJUUS H. SKJjCRVEN R. LANGARD S. LIEN JT. AAMODT T. A case-referent study of lung cancer, occupational exposures and smoking: I Comparison of title-based and exposure-based occupational information. Scand] Work Environ Health 12 (1986) 193--202. The role of occupational exposures and smoking in the development of lung cancer has been studied among 176 male incident lung cancer cases and 176 referents admitted to two county hospitals in southeast Norway during 1979--1983. After the allocation of all occupational titles in the Nordic Classification of Occupations into three exposure groups according to potential exposure to respiratory carcinogens and other contaminants, each subject was clas sified according to exposure status of main occupation and number of years in each exposure category. An excess risk of lung cancer was observed both among those in possibly exposed occupations and among those definitely exposed. A more than threefold excess risk was observed among subjects with more than 30 years in exposed occupations. Exposure to 22 agents/processes was further assessed by a separate ques tionnaire and estimated simultaneously in a logistic regression model. Elevated risks were associated with exposure to asbestos and several other agents/processes, which largely correlated to each other. Smoking was strongly associated with all histological subtypes of lung cancer, while for occupational exposures the risk ratio was highest for small cell carcinoma and lowest for adenocarcinoma. Very high risk ratios for lung cancer were observed among heavy smokers in exposed occupations. Key terms: chemicals, logistic regression, occupational titles, respiratory carcinogens, tobacco life time dose. A major challenge in studies of the role of occupational exposures in relation to lung cancer is to form an ap propriate operational scale for the determinant in ques tion. This challenge refers in particular to case-referent studies of limited size, in which the pooling of dif ferent occupational exposures into broad groups of the determinant is often a prerequisite for the analysis. Traditionally some kind of occupational title-based, information has been applied to such studies, although an exposure-based approach would conceptually be more relevant (II, 13, 30). By combining these ap proaches, the present study aimed to examine the role of occupational exposures in the development of lung cancer among men in two counties in southeast Norway. In a previous study, a crude dichotomous deter minant scale related to length of stay in a polluted workplace was applied to 103 cases of cancer and 103 1 Department of Occupational Medicine, Telemark County Hospital, Porsgrunn, Norway. Section for Medical Informatics and Statistics, University of Bergen, Bergen, Norway. 1 Lung Division, Telemark County Hospital, Skien, Norway. ` Lung Division, Vestfold County Hospital, Tensberg, Norway. Reprint requests to: Dr H Kjuus, Department of Occupational Medicine, Telemark County Hospital, N-3900 Porsgrunn, Norway. referents (18). Among 32 lung cancer cases [Interna tional Classification of Diseases (1CD) 162, 163], a threefold increase in relative risk was observed for ex posed subjects (2 10 years in a polluted workplace) in comparison to unexposed subjects (< 10 years in a polluted workplace). The lung cancer part of that study has been continued, and in the present report we give the results from 176 male lung cancer cases and 176 referents, selected during a five-year period from the Telemark and Vestfold County Hospitals. The study has been divided into three parts. This paper describes the design and the main results of the study, while the role of asbestos exposure and the quantitative importance of all occupational exposures in the development of lung cancer are subjects of separate presentations (17, 19). Subjects and methods The neighboring counties Telemark and Vestfold are located in the southern part of Norway and have 162 000 and 190 000 inhabitants, respectively. The in dustrial areas are mainly located near the towns along the coast, with a predominance of chemical industry in Telemark and shipbuilding and, previously, whaling in Vestfold. The remaining parts of the counties are dominated by farming and forestry. During the period 3 193 1000329? 1979--1983, 136 male incident lung cancer cases and 136 referents were identified in the medical ward of Telemark County Hospital. In the last two years of the period, all 40 incident lung cancer cases admitted to the medical ward of Vestfold County Hospital in Tensberg were also included in the study. Besides the Vestfold County Hospital in Tensberg, which covers both county and local hospital functions, there are three smaller hospitals in this county. The proportion of urban and rural parts of the referral area for each of the four hospitals is fairly similar. Subject identification Male patients who were under 80 years of age and admitted to the medical ward with a recent diagnosis of lung cancer (1CD 162--163) were selected as cases. For each case, one age-matched reference patient ( 5 years) was randomly selected from the admissions lis: or from the inpatient list of the same department. Patients with conditions implying physical or mental handicaps which would have precluded employment in heavy industry were not accepted, neither were patients with poor general health or those who were obviously mentally diminished. Patients admitted be cause of chronic obstructive lung disease were excluded from the reference series. The diagnoses among the referents accepted for the study were ischemic heart disease (N = 64), other cardiovascular disease (N = 37), other malignant neoplasms (N = 6), other lung disease (N = 12), gastrointestinal disease (N = 9), hepatobiliar disease (N = 6), urological disease (N = 3), musculoskeletal disease (N = 5), diabetes mellitus (N = 6), and miscellaneous (N = 28). The distribution of the histological subtypes of lung can cer (WHO, 1980) is shown in table 1. Table 1. Histological type ol respiratory cancer among the cases. Histological type Squamous cell carcinoma Small cell carcinoma Adenocarcinoma Large cell carcinoma Alveolar cell carcinoma Mesothelioma Uncertain histology Not histologically verified Number 94 38 21 4 1 fi 6 8 Percentage 53.4 21.5 11.9 2.3 0.6 2.3 3.4 4.5 Subject interviews All the subjects were interviewed at the bedside by one of us (HK). They were not informed of the purpose of the study, which was presented as a ``survey of oc cupational exposures among patients in a medical ward." The subjects were instructed not to give any information about their disease to the interviewer. In most cases, the interview took place either before the diagnosis had been confirmed or before the patient had been informed of his diagnosis. Thus 14 subjects in terviewed as potential cases were found not to have Table 2. Exposure classification of 99 occupational groups. (The No-cic Classification of Occupations, 1965. two-digit code precedes the occupational group.) Oefinite exposure (OT +) Possible exposure (OT?) No exposure (OT-) 50 Mining and quarrying 52 Mineral treating work 59 Other mining and quarrying work 73 Smelting, metallurgical and foundry work 75 Iron and metalware 78 Painting and paperhanging work 83 Chemical and related process work 01 Chemical and physical work 51 Well drilling and related work 60 Ship officers and pilots 61 Deck and engine-room crew work 63 Railroad locomotive engineers and firemen 70 Textile work 72 Shoe and leather wort 74 Precision mechanical work 76 Electrical work 77 Wood work 79 Construction work 80 Graphic work 81 Glass, ceramic and ciay work 84 Tobacco work 85 Other production process work 86 Packing and wrapping 87 Stationary engine anc motor- power work 88 Longshoremen 89 Miscellaneous laborir; work 93 Building caretaking ard cleaning work 94 Hygienic and beauty t-eatment work 95 Laundering, dry-clear -g and pressing work 00 02 03 04 05 06-29 30-33 40--44 62 64 65 66 67 68 69 82 90 91 92 97 98 99 XI Technical work Biological work Medical work Nursing care Other professional health and medical work Administrative and clerical work Sales work Agriculture, forestry, fishing Air transport work Road transport work Conductors, dispatchers and freight assistant work Traffic supervising Postal and telecommunicalion work Postal and other messenger work Other transport and communication work Food and beverage work Public safety and' protection work Hotel, restaurant and domestic work Table waiting Photographic work Funeral service Other service work Military work 194 10003240 lung cancer and were initially excluded. Four of these were accepted as referents, however. The interview "was, as far as possible, performed "blind" as to the subject's status as case or referent. Information on smoking habits was collected at the end of the inter view, and referents with ischemic heart disease were not interviewed as long as they were in the coronary care unit. For approximately 10 % of the participants the case-referent status was unintentionally revealed, and for another 10 % their status was suspected. If potential study subjects were discharged from the hos pital before the interview had taken place, they were not recalled. Some cases were interviewed however dur ing a subsequent hospital stay. The average length of an interview was 30 min. One potential case and two potential referents refused to participate in the study. Assessment of exposure For each subject the interviewer obtained a complete work history of all jobs held since the age of 14 years. For each job, the occupational title (OT), the duties usually performed, and detailed information on rele vant exposure factors were ascertained. Characteriza tion of exposure has partly been based upon an ap proach based on occupational titles and partly on an approach based on exposure. Exposure-based approach From I January 1982 on, a standardized questionnaire concerning past exposure to 17 types of chemical agents and five specific work processes was used. For the 68 cases and 68 referents ascertained before this date, the questionnaire was filled out retrospectively on the basis of the exposures and occupational tasks stated in the detailed occupational history. A dichotomous scale of exposure was used in which regular exposure for more than three years was classified as "exposed." Smoking habits were classified according to status (never a smoker, present smoker and exsmoker (> 1 year since cessation of smoking)], type (cigarettes, pipe and cigar), and amount, measured as lifetime con sumption of tobacco (in grams). The categories for this lifetime dose were chosen so that they corresponded to traditional categories for the average number of cigarettes smoked daily, assuming a 43-year smoking period, which was the average smoking period among the present and former smokers in the study. One cigarette was estimated to be equal to 1 g of tobacco. As a crude indicator of possible domestic radon ex posure, the subjects were also asked if they had lived in a dwelling of wood or brick/stone for most of their lives. The distribution of cases and referents according to age, hospital, urban/rural place of residence, and smoking habits is shown in table 3. Occupational title-based approach Initially in this study, the 99 occupational groups (twodigit level) and all of the over 4 000 occupational titles (five-digit level) in the Nordic Classification of Occu pations (1963) were allocated into the following three exposure groups according to possible exposure to respiratory carcinogens or other contaminants experi enced in a typical job in that particular occupation: definite exposure (OT +), possible exposure (OT?), and no exposure (OT -). To cover the possibility that pollutants not known as specific carcinogens might act as cocarcinogens, eg, in combination with smoking, we classified occupations with definite exposure to most kinds of dust, gases, and fumes as exposed oc cupations. Obviously, several jobs were classified into different exposure groups at the two levels, as the most comprehensive classification corresponded more close ly to the actual job performed. For example, furni ture makers, who were allocated to the OT? category at the two-digit level (wood workers according to code 77), were classified in the OT + category at the fivedigit level, due to the possible carcinogenic effect of hardwood exposure. The classification of the 99 occu pational groups (two-digits) into the three exposure groups is shown in table 2. Exposure status of the occupational title of the longest job held (two-digit level), together with the number of years in exposed occupations (five-digit level) up to 1970, was then recorded, omitting the last 9 to 13 years of possible exposure. Statistical methods Multivariate analyses were performed with logistic regression models with the rate of cases within the sample of cases and referents as the dependent vari able (4, 6). The program for unconditional logistic regression as in the BMDP statistical program pack age was used (8). Following the proposals of Holford et al (13), conditional analyses of the matched pairs were also performed. The results were, however, sim ilar to those obtained with the unconditional analy ses, and, as the number of concordant pairs reduced Tibia X Number of cases and referents according to age, hospital, urban/rural place of residence, and smoking habits. Indicator Cases N% Referents N% Age (years) 40-59 60-69 70-79 Hospital Telemark County Hospital Vestfold County Hospital Urban/rural status Urban Rural Smoking habit status Present smoker Former smoker Never a smoker Type of tobacco used Cigarettes Pipe, other 32 87 57 136 40 125 51 135 39 2 151 23 18.2 49.4 32.4 77.3 22.7 71.0 29.0 76.7 22.2 1.1 86.8 13.2 36 20.5 86 48.9 54 30.7 136 77.3 40 22.7 101 57.4 75 42.6 77 43.8 75 42.6 24 13.6 127 83.6 25 16.4 195 10003261 the material and therefore gave wider confidence in tervals, we chose to report the results of the uncon ditional estimation procedure, which is known to give conservatively biased estimates (4). Since age was the most important matching criterion in the material, selected age-specific results have also been presented. Odds ratios were used as estimators of the risk ratio (RR) (14). The significance of the models was tested for goodness of fit, using likelihood ratio tests (8, 29), Table 4. Risk ratio (RR) for lung cancer according to exposure status (OT - = no exposure, OT? = possible exposure. OT + = definite exposure) based upon title tor mam lifetime occupation (two digits), adjusted for smoking and urbanfrural status. OT- OT? OT + Total Cases (N) Referents (N) 44 63 69 176 79 52 45 176 Risk ratio* Unadjusted Adjusted 1 Adjusted 2 Adjusted 3 Adjusted 4 Adjusted 5 10 1.0 1.0 1.0 1.0 1.0 95 % confidence interval for RR, adjusted 2 22 2.8 1.8 2.5 1.7 2.3 1 8 24 1.7 2.2 1.9 2.6 0.9--3.0 1.3-4.2 J Adjusted 1 Adjusted 2 Adjusted 3 Adjusted 4 Adjusted 5 smoking, three levels (0--9,10--19, s 20 ciga rettes/d). smoking, three levels (0--9.10--19. 2 20cigarettes/d) and urban/rural status. smoking, five levels (0--4, 5--9. 10--19. 20--29, 2 30 cigarettes/d). smoking, five levels (0--4, 5--9. 10--19. 20 --29. 2 30 cigaretles/d) and urban/rural status smoking, linear adjustment and the models were compared from the differences of the likelihood ratios (G), which were assumed to be chi-square distributed. Factors were introduced as additive terms within the logistic scale. Departure from additivity was evaluated through first-order interac tion effects in a backward stepping procedure with a removal level set at 15 ^. Ninety-five percent con fidence intervals (95 % Cl) were calculated with the use of the estimated coefficients and standard errors from the regression models. For a few analyses (table 7 in the Results section, exposure group OT -), a testbased method was used (25). Results The risk ratio for lung cancer according to exposure status (OT -, OT?, OT + ) of main lifetime occupa tion is shown in table 4. After adjustment for smok ing (three levels) and urban/rural status, there was a risk ratio for lung cancer of 2.3 among persons in definitely exposed occupations (OT +) when they were compared to those in unexposed occupations (OT -). After the same type of adjustment, the main age group (60--69 years) gave the estimates 2.0 and 3.5 for the OT? and OT + category, respectively. There were only small changes in the risk ratio with further adjustment for smoking. Although it was not possible to assess the significance between the different models in terms of the likelihood ratio tests, it seemed that the risk ratio estimates increased somewhat when the control for smoking was more elaborate (10 levels or linear ad justment) in that it gave more weight to the tails in the smoking distribution. Table 5 shows the risk ratio for Table 5. Risk ratio for lung cancer, according to number of years in definitely exposed occupations (OT * j for the total material and the main age group (60--69 years). exposed Number of subjects Referents Unadjusted isk ratio Adjusted* Cat6 Lmc Age group 60--69 years -- 1-- 9 10-19 20-29 230 22 13 20 21 11 Total material 1- 9 10-19 20-29 2 30 47 29 38 35 27 43 18 13 7 5 83 33 25 24 11 i.O 1.0 1.0 1 4 1.2 1.6 3.0 2.7 26 5.9 4.5 41 4 3 5.5 66 1.0 1.0 10 1 6 13 1 4 2 7 2.2 19 2.6 2.2 2.5 4 3 3.8 34 * Smoking, three levels (0--9. 10--19. 2 20 cigarettes/dl and urbanfrural status. 0 Cat = effect of OT estimated in five separate categories c Lin = effect of OT + estimated assuming a linear relation 0 Confidence interval based on categorical estimates Test lor trend: 60--69 years: G = 13.5, 4 df, p = 0.009 Total: G = 12.8, 4 df. p = 0.012 Test for departure from linearity: 60-- 69 years: PG = 0 63, 3 df. p = 0.89 Total: pG = 0.66. 3 dt, p = 0.88 95 % confidence interval0 0 4- 3.2 10-71 1.5-13.6 1.5-19 9 0 7- 2.7 1 1-- 4.3 1 1-- 4.4 16- 8.9 1% 10003262 lung cancer according to the number of years in ex posed occupations. There was a statistically significant effect with length of exposure [G = 12.8,4 degrees of freedom (df), p = 0.012], with a risk ratio of 3.8 for those with more than 30 years in OT + occupations. The effect was nearly linear in the log scale (AG = 0.66. 3 df, p = 0.88). The association between occupational exposure and lung cancer was the most pronounced in the 60- to 69-year age group, followed by the age group 70--79 years. When half the years in possibly exposed occupations ('/'OT?) was added to the number of years in OT + occupations, the risk ratios became even higher, with a more than fivefold risk of lung cancer among those with more than 30 years of exposure (table 6). The test for trend was significant (G = 16.9, 4 df, p = 0.002), and, again, the age group 60--69 years showed the strongest effects. The trend in this age group, how ever, seemed to depart from exponentiality, although not significantly (AG = 5.73, 3 df, p = 0.13). The results for some selected occupational groups are shown in table 7. The unexposed group (OT -) was used as the reference category for exposed occupations (OT + and OT?), while for the unexposed occupations (OT -) the remaining occupations in the same group were used as reference. Elevated risk ratios were ob served for several occupational groups, being highest among miners and quarriers (RR = 10.2). Many of these results were, however, based upon small num bers and were consequently subject to large sampling errors. The risk ratios for lung cancer according to smok ing habits were estimated for lifetime consumption of tobacco, iranstormcJ into the equivalent number ol cigarettes smoked pe' dav and assuming a 45-year smoking period. Onh :wo olTlic lung cancer cases had been "nonsmokcry," and even one of these was a sporadic smoker ior^-lcw. years in his yQpih The as- sociations observed between smoking and lung can cer were strong, with risk ratios of 3, 6, 16, 40, and 100 for the smoking levels I--4, 5--9, 10--19. 20--29, and 30 cigarettes or more per day, respectively. The data fit closely a model of exponential dose-response relationship (AG = 2.48, 4 df, p = 0.65). There were no obvious differences between the histological subtypes of lung cancer, with risk ratios for adenocar cinomas at the same level as for the other subtypes. Among subjects in definitely exposed occupations (OT +), however, the highest risk ratio was observed for small cell carcinoma (RR = 4.2), followed by squa mous cell carcinoma (RR = 2.3), and adenocarcinoma (RR = 1.5). These relationships were also reflected when the lung cancers were divided into type-1 tumors (squamous cell carcinoma and small cell carcinoma) and type-II tumors (here: adenocarcinoma), according to Kreyberg (20). The group l:group II ratio was highest among those definitely exposed, being 7.4, 6.9, and 4.6 for exposure status OT +, OT?, and OT -. respectively. No difference in the group I: group II ratio was found between the highest and the lowest smoking level, being 6.8 for both groups. When exposure-based information from the ques tionnaire was considered, statistically significant in creases in the risk ratios were found for persons exposed to asbestos, rock drilling, quartz/stone dust, cement dust, `'other gases," and for fertilizer produc- Table 6. Risk ratio (or lung cancer, according to number of years with combined exposure ^number ol years m definitely exposed (OT +1 occupations plus half the years in possibly exposed occupations (Vi OT?)| for the total material and the main age group (60--69 years). Years exposed (OT + and Vi OT?) Number of subiects Cases Referents adiusted Risk ratio Ac;us:ed* Cat" Lm = 95 /. confidence interval0 Age group 60--69 years _ 12 1- 9 15 10--19 24 20-29 21 2 30 15 Total material _ 1-- 9 10-19 20-29 >30 25 30 46 41 34 30 16 23 11 6 57 33 36 37 13 1.0 1.0 2.3 29 2.6 2.6 4.8 3 4 6.3 7.8 1.0 1.0 2.1 2.5 2.9 2.8 2.5 2 2 6.0 5.6 Smoking, three levels (0--9, 10--19. a 20 cigarettes/d) and urban/rural status Cat = estimated in five separate categories. c Lin = estimated assuming a linear relation. 0 Confidence interval based on categorical estimates. Test for trend: 60--69 years: G = 11.5, 4 df, p = 0.022 Total: G = 16.9, 4 df, p = 0.002 Test for departure from linearity: 60--69 years: dG = 1.91, 3 df, p = 0.59 Total: aG = 5.73, 3 df, p = 0.13 1.0 1.5 23 3.6 5.4 1.0 14 1.9 2.6 35 1.0- 8.9 0.9- 7.1 1.1 -- 108 2.1-28.5 12- 55 1.3- 5.7 11--46 2.3-136 197 10003263 Table 7. Risk ratios for lung cancer m selected occupational groups accz'dmg to exposure status based upon title tor mam lifetime occupation (two-digit code) (Reference category no exposure --cup) Occupational group (two-digit code) Number ol subjects Cases Referents Risk ratio Unad justed Ad justed* 95 % confidence interval No exposure (OT-)" Academic, artistic, administrative and clerical work (00. 02--29) Salesmen, commercial travelers (31--33) Farming, torestry (40--42, 44) Road transport work (64) Food, hotel, public salety, other specified work (82, 90-92, 96-99) Possible exposure (OT?) Ship offices, deck and engineroom crew work (60--61) Electrical work (76) Wood work (77) Construction work (79) Longshoremen, freight handlers (88) Packing and wrapping, stationary engine and motor power work, various production process work, building caretaking (81, 84--87, 93- 95) Definite exposure (OT*-) Mining and quarrying work (50-52. 59) Smelting, metallurgical, foundry work (73) Iron and metalware work (75) Painting and paperhanging work (78) Chemical and related process work (83) 6 7 12 8 8 18 6 7 9 11 11 14 4 22 5 24 20 7 31 9 6 9 1 16 6 4 10 2 -w 20 5 18 0.5 0.5 19 1.7 06 0.7 1.7 1.4 2.7 2.0 3.6 1.5 10.9 9.8 0.8 0.7 2.7 3.0 50 5.7' 2.0 1.7 12 7 10.2- 2.0 17 1 8 1.7 2.4 2.6' 0 <o 1 u nCO 1 6CO 0.2--1.2 0.7-4 1 0.4--1.2 0.7--2.6 0.6--4.2 1 0-100 9 0.2--2.3 0.9-102 1.5-21.2 0.6--4.8 2.0-52.1 04-73 1.2--5.7 * Adjusted for smoking, three levels (0--9. 10--19. 2 20 cigareties d) 3 Reference category: remaining occupational groups in the no exposure category p<0.05 lion and the welding of stainless steel (table 8). The risk ratio for wood dust exposure was significantly reduced (RR = 0.4). Several of the exposures were highly correlated, however. For instance, half of the cases exposed to either stainless steel welding fumes or fertilizers also reported moderate to heavy asbestos exposure. A stepwise logistic regression procedure was therefore applied to the material: it included the dif ferent variables into the model according to signifi cance level. Risk ratios for asbestos, wood dust, rock drilling, and fertilizer production remained significant after this procedure (table 9). With regard to the combined effect of smoking and occupational exposures, the logistic regression model used in this study yielded an independent, additive estimation of In RR, which implies multiplicative effects at the risk ratio level. Table 10 shows that the risk of lung cancer was considerably increased among heavy smokers who had also been occupationally ex posed. Or.iy three cases and three referents had lived the longes: part of their life in a dwelling mainly made out of br:ck or stone. Discussion When either title-based or exposure-based occupational information was used, this study showed a statistical ly significant increased risk of lung cancer among persons judged to have experienced exposure to re spirators carcinogens and other industrial pollutants at their workplace. This finding is in accordance with the results of a nationwide cohort study of 12 000 Norwegian- which will soon be published by K'vale and co-workers iG Kvile. personal communication). They used the same exposure categorization for main occu pation as ;r the present study and found a risk ratio of 1.6 among subjects in the OT? category and one of 2.6 among those in the OT + category w hen adjust- 198 10003264 Table 8. Risk ralios for lung cancer according to occupational exposures, based upon questionnaire information (Exposed >3 years of relevant exposure) Exposure factor Number of exposed subiects Cases Referents Risk ratio Unadiusted Adiusled* 95 % confidence interval Asbestos Rock drilling Tunnelfmmmg Quartz/stone dust Cement dust Rockwool/glassfiber Wood dust Hardwood dust Cellulose/paper Fertilizer production Plastic/rubber dust Metal dust Tar/aspnaft Coke/coal Fuel oilsfgasoline Paints, glues, lacquer Solvents or degreasing liquids Welding, all types Welding, stainless, acid proof Nickel/cbromium (excluding welding) Other gases (ammonia, nitrogen oxides, chlorine and sulfur dioxide) 55 33 19 59 29 13 20 -- 13 16 7 56 11 23 42 17 47 28 16 7 50 24 17 10 35 13 11 39 9 10 5 7 43 8 12 35 17 35 19 6 6 32 29 2 7` 2.2 2 02.0 1.9 2.0 1 9# 2.5 26' 1.2 1 0 0.5 0.4 * 1.3 1.0 3 4 4 2* 1.0 0.9 1.4 1.6 1.4 1.1 2.1 1 5 1 3 10 1.0 1.2 1.5 1.5 1.6 1.9 2.8 3.3* 1.2 1.4 1.8 1.9* 1 5-4.8 1 0-3 9 0.8--4 6 1.1-3.2 1 2-5 7 0 4-2 5 0.2--0.8 0 4-2 6 1 4 --12.7 0.3--2 8 1.0--2.7 0.4--3.2 0 7--3.2 0 6-1.8 0.6--2.6 09--26 0.9--3 7 1 2--9.3 0 4--4 4 1.1 --3.3 * Smoking, three levels (0--9, 10--19, e20 cigarettes/d). ' p<0.05. Table 9. Simultaneous estimation of occupational exposures, logistic regression. Expcsure factor Smoking Urban/rural status Asbestos Rock drilling Quartz Cement Wood dust Hardwood dust Fertilizer production Coke/coal Welding/stainless Other gases Step number 0 1 2 3 45 -- X X X XX *1 c * X X X X XX zz * zc * * X XX z o0 * X c 1 z zc " z 00 z c0 Risk ratio. last step 4.4 (10--19 cigarettes/d) 15 4 ( a 20 cigarettes/d) 27 2.4 0.5 3.7 Other factors: not significant at step zero (p >0.05). :e p< 0.0001 p<0.001 ' p<0.01 p < 0.05 x Factor included in the model. : Factor lost significance (p>0.05). mem was made for age, region, smoking, and urban/ Kreyberg (20). It is, however, in contrast to the findings rural place of residence. Time-related information of Vincent et al (34) and Kvile (G Kv&le, personal com based upon five-digit job titles, however, as used in munication), who report the strongest association the present study, would be a more sensitive method between adenocarcinoma and occupational exposures. :h for exposure characterization. With this approach, the Different criteria for the classification of histological O risk ratios among those in the highest exposure cate subtypes of lung cancer among pathologists, and in gory increased, particularly when the possible expo particular of adenocarcinoma, may have contributed sure category (OT?) was included (tables 5 and 6). to some of this apparent discrepancy (31). Occupational exposure was more strongly associated The observed risk ratios for several of the separate .0 with small cell carcinoma and squamous cell carci occupational groups are in accordance with those of noma than with adenocarcinoma, a finding which cor other studies (3, 27, 37). Although these data were responds to the results of Stayner & Wegman (32) and adjusted for smoking, we are cautious in the inter- 199 10003269 Table 10. Ris ratio lor comDined expos-'e 10 smoking ith-ee leveisi anc c ^panonai exposure based upon exposure staius ol the mam occupational title and num;e- o' >ears in definitely exposec occupations Number of subjects Cases Referents Risk ratio Unadjusted Adjusted* 95 % confidence interval Main occupation No exposure (OT -) 0-- 9 cigarettes/d 10--19 cigarettes/d a20 cigarettes/d 12 52 21 23 11 4 Possible exposure (OT?) 0-- 9 cigarettes/d 10--19 cigarettesld a20 cigarettes/d Definite exposure (OT +) 0-- 9 cigareties/d 10--19 cigarettes/d >20 cigarettes/d 11 31 30 17 22 4 u 24 44 20 11 1 Number of years in definitely exposed occupations (OT - ) < 10 years 0-- 9 cigarettes/d 10--19 cigarettes/d a20 egareties/d 19 74 36 34 21 8 10 years 0-- 9 cigarettes/d 10--19 cigarettes/d 20 cigarettes/d 18 33 59 26 23 1 10 40 11.9 1.5 77 23.8 25 79 47 7 1.0 4.1 10.2 2.1 8.8 89 6 10 41 13 2 17 68 21 9 23 96 306 10 41 14 4 24 97 340 2.5-68 5.8-30.2 09-3.0 2.3-20.3 5 3-89.5 13-4 2 3.2-28.5 7.5-125.6 2.5-68 6.3-32.6 15-38 3.7-26.0 9.3-124.3 * Urban/rural status pretation given ihem because of the sparse number of subjects in each group. This is also the case when the separate occupational exposures are studied (table 8). The interpretation of these data is further hampered by the high interde pendence between several of the exposure factors. Ex posure to stainless steel welding fumes, together with several other variables initially of statistical signifi cance, lost their significance when smoking and as bestos first entered the model. It is of interest that sub jects who reported exposure to rockwool in the present study.-did-aoi carry. anv_excess risk of lung cancer (RR = 1.0). We have no definite explanation for the "protective" effect of wood exposure in this study (RR = 0.4). The result demonstrates the necessity for cautious interpretation of these data, and the possi bility of statistically significant results occurring bychance is also close at hand in this study of multiple exposures (2, 7). All subjects not exposed to the particular factor under study were used for the risk ratio estimates in table 8, while a "cleaner" reference category consisting of subjects not exposed to any of the factors would be another choice (26, 35). With this latter approach, all the rate ratios increased 50 to 100 o, which gave several additional statistically significant results. Be cause of the aforementioned limitations in the inter pretation of the data, wc chose to present the most conservative figures. For a given level of daily cigarette consumption, the relative risk of lung cancer has been show n to increase with age (12). The constant relative risk observed across the age strata in (he present study indicates that this age effect is merely an expression of a cumulative con sumption of tobacco (33). The present smoking data suggest, furthermore, that the relative risk is an ex ponential function of the lifetime tobacco dose; this finding is in accordance w ith those of previous studies (9, 33). All histological subtypes of lung cancer showed a dose-response relationship to smoking. This relation ship is in accordance with the findings of several other studies (32. 34, 36), but it is in contrast with the results of Kreyberg's study (20), in which adenocarcinoma was shown not to be associated with smoking. When the rate ratio estimates for smoking and oc cupational exposures were combined, occupationally exgosedjheav v_smokers carried a more than 30-fpld excess risk for lung cancer in comparison to unexposed light smotmfP--9rigareffe'sAty The'mafn factor in these dramatic figuresThowever, was that of smoking. As regards the selection of cases, histological veri fication of the diagnosis was not confirmed for eight of the persons. The results for the separate histologi cal subtypes of cancer do not support the possibility of a "diluted" case group in the present study how-ever. Furthermore, we have no information which might support the possibility of an exposure-related referral of lung cancer cases to the study hospitals. In Vestfold, however, 73 o of the referents were recruited from the town of Tonsberg and neighboring com munities, while only 55 % of the cases came from this region. Thix occurrence indicates a somewhat larger referral area for the cases than for the referents at this hospital. When urban/rural status was included into the model, the risk ratio estimates among the occu pationally exposed was slightly reduced, but the con- 200 tributinn ol tliix viiriahIc w;is muimicuIIv nonsiL'iuli- cant lor all the associations studied. Patients admitted to ilie hospitals because ol chronic obstructive lung disease were excluded from the ref erence series because of the possible association between this condition and occupational exposures (16). Patients admitted to the hospital because of an other condition, but who also happened to have chronic obstructive lung disease as a concomitant diag nosis, were not excluded, however, as recommended by Lubin & Hartge (23). Furthermore, as all potential exposures after 1970 were omitted, it seems unlikely that the reference diagnosis in 1979--1983 could have influenced participation in industrial activities in the 1960s. However, as smoking is associated with cardiovas cular disease and a wide variety of other diseases, the effect of smoking on lung cancer might have been underestimated in the present study. The exclusion of 80 subjects with potentially smoking-related diagnoses in the reference group (ICD 410--414 ischemic heart disease, ICD 531--535 stomach disease, and ICD 485--519 other respiratory disorders) did not change the observed risk ratios for smoking. Neither was there any change in the risk estimates for occupational ex posures when the smoking-related diagnoses were ex cluded from the analyses. With regard to possible bias in the estimation of ex posure, occupational title-based information is as sumed to be less subject to recall bias than exposurebased information (1, 28). As both job-title informa tion and exposure-based information were available for all the subjects in this study, a retrospective method for evaluating possible recall bias has been used (I). A further study of asbestos exposure among metal workers (code 75), of solvent exposure among paint ers (code 78), and of fertilizer exposure among chemi cal process workers (code 83) did not reveal any "protective effect" among the unexposed subjects within the given job categories which could indicate an exaggeration of these exposures among the cases in these particular groups. The allocation of occupations into different expo sure categories will always involve some degree of arbitrarily. For example, if woodwork, which was found to be "protective" in the present study, had been allocated to the OT + instead of the OT? category, the risk ratios would have been 2.0 for both groups when adjusted for smoking (three levels) and urban/rural status. On the basis of an a posteriori judgement, however, woodwork might sooner be placed in the unexposed category, which in turn gives the estimates 2.2 and 2.5 for the OT? and OT + categories, respec tively. Random and independent errors in the estimation of the exposure would furthermore tend to diminish the apparent degree of association between exposure and effect (5). This tendency is even greater when the measure used for confounder control, such as for minking in the present >iudy. is more reliable than (lie surrogate measures used to characterise the exposure (21). As multiple categories ol exposure were used for both variables based on job titles in the present study, the impact of this error is assumed to be considerably reduced, however, in contrast to studies in which ihe exposure variable is dichotomized (24). Although the observed crude risk ratios were somewhai reduced when controlled for smoking at various levels, the results do not suggest smoking as a very strong confounder in the present study. Furthermore, other uncontrolled unknown confounding would need to be strongly associated both with the disease and the exposure variables in order to explain the observed as sociations (38). Domestic radon exposure has been suggested as a potential confounder in studies of the association between lung cancer and occupational exposures (10). Although information on building material in main dwellings is a very crude indicator of such exposure, this factor can hardly have confounded the results of the present study, as the vast majority of both the cases and referents had mainly lived in wooden houses. Several of the occupation and exposure factors allocated to the exposed categories in the present study may not represent a cancer hazard in themselves, eg, exposure to several types of dust and gases. However, such exposure factors may promote the effect of es tablished carcinogens, such as those found in tobacco smoke. If that was the case, one would expect a risk ratio of unity for the occupationally exposed nonsmokers and an elevated risk ratio for the occupa tionally exposed smokers. The smoking distribution among the cases prevents any meaningful study of this relationship however. Although interview-based information on occupa tional exposures is probably more likely to be biased by disease state than information on occupation, the former is obviously to be preferred when the aim is to identify etiologic agents in the workplace. However, such information should be supplemented in future studies by exposure data of a quantitative nature, which would increase the validity of the exposure es timation (22). As to the use of occupational titles in occupational epidemiology, the present study suggests that such information might not only be utilized for hypothesis-generating purposes, but also for quan titative assessments, through the evaluation of the role of a combination of all occupational exposures in the development of a disease. However, a prerequisite for such an evaluation would be detailed information on all occupations held in a lifetime, together with a relevant categorization of the exposure variable. Acknowledgments We are indebted to Mr T Torgrimsen, an industrial hygienist, for contributing to the exposure classifica tion of occupational titles and to Dr PT Lyngdal for 201 10003247 selecting subjects for the study. We also thank Ms P Flor for her linguistic assistance and Ms U Danielsen for typing the manuscript. Thanks are also due to the Norwegian Cancer So ciety and Norwegian Society for Fighting Cancer for financial support. References 1. Axelson O. A note on observational bias in case-referent studies in occupational health epidemiology. Scand J Work Environ Health 6 (1980) 80--82. 2. Axelson O, Flodin U, Hardell L. A comment on the reference series with regard to multiple exposure eval uations in a case-referent study. Scand J Work Environ Health 8 (1982): suppl I. 15--19. 3. Bonassi S, Ceppi M, Puntoni R, Valerio F. Vercelli M, Belli S. Biocca M. Comba P. Ticchiarelli L, Mariotto F. Taddeo D, Zuccherelli D. Casalini A, Perini C, Nelli L. Sansoni C, Pupp N. Farina M. Luciani A. Mortality studies of dockyard workers (longshoremen) in Italy. Am J Ind Med 7 (1985) 219--227. 4. Breslow L, Day NE. Statistical methods in cancer re search. Volume I (The analysis of case-control studies). International Agency for Research on Cancer, Lyon 1980. (1ARC scientific publication no 32). 5. Bross 1. Misclassification in 2 x 2 tables. Biometrics 10 (1954) 478--486. 6. Cos DR. Analysis of binarv data. Methuen, London 1970. 7. Dinman BD, Sussman NB. Uncertainty, risk, and the role of epidemiologv in public poliev development. J Occup Med 25 (1983) 511--516. 8. Dixon WJ, ed. BMDP statistical software. University of California Press. Berkeley, CA 1983. 9. Doll R. Pelo R. Cigarette smoking and bronchial car cinoma: Dose and time relationships among regular and lifelong non-smokers. J Epidemiol Community Health 32 (1978) 303--313. 10. Edling C, Kling H. Axelson O. Radon in homes -- A possible cause of cancer. Scand J Work Environ Health 10 (1984) 25--34. 11. Gamble J. Spirtas R. Job classification and utilization of complete work histories in occupational epidemiol ogy. J Occup Med 18 (1976) 399--404 12. Hammond EC. Smoking in relation to the death rates of one million men and women. Natl Cancer Inst Monogr 19(1966) 127--204. 13. Hoar SK. Morrison AS, Cole P. Silverman DT. An oc cupation and exposure linkage system for the study of occupational carcinogenesis. J Occup Med 22 (1980) 722-726. 14. Hogue CJR, Gaylor DW, Schulz KF. Estimators of relative risk for case-control studies. Am J Epidemiol 118 (1983 ) 396--407. 15. Holford TR, White C. Kelsey JL. Multivariate analysis for matched case-control studies. Am J Epidemiol 107 (1978) 245-256. 16. Kjuus H. Istad H. Langdrd S. Emphysema and occu pational exposure to industrial pollutants. Scand J Work Environ Health 7 (1981) 290-297. 17. Kjuus H, Langird S, Skjaerven R. A case-referent study of lung cancer, occupational exposures and smoking: III Enologic fraction of occupational exposures. Scand J Work Environ Health 12 (1986) 210-- 215. 18. Kjuus H, Lislerud A. Lyngdal PT. Omland H, Stave O, Langird S. Cancer and polluted work places: A casecontrol study. Int Arch Occup Environ Health 49 (1982) 281-292. 19. Kjuus H, Skjxrven R, Langlrd S, Lien JT, Aamodl T. A case-referent study of lung cancer, occupational ex posures and smoking: II The role of asbestos exposure. Scand J Work Environ Health 12 (1986) 203--209. 20. Kreyberg L. Aetiology of lung cancer: A morphologi cal. epidemiological and experimental analysis. Universitetsforlaget, Oslo 1969. 21. Kupper LL. Effects of the use of unreliable surrogate variables on the validity of epidemiologic research studies. Am J Epidemiol 120 (1984) 643--648. 22. Lemasters GK, Selevan SG. Use of exposure data in occupational reproductive studies. Scand J Work En viron Health 10(1984) 1--6. 23. Lubin JH. Hartge P. Excluding controls: Misapplica tions in case-control studies. Am J Epidemiol 120 (1984) 791--793. 24. Marshall JR, Priore R, Graham S, Brasure J. On the distortion of risk estimates in multiple exposure level case-control studies. Am J Epidemiol 113 (1981) 464--473. 25. Miettinen OS. Estimability and estimation in casereferent studies. Am J Epidemiol 103 (1976) 226--233. 26. Pastorino U, Berrino R, Gervasio A, Pesenti V, Riboli E, Crosignani P. Proportion of lung cancers due to oc cupational exposure. Int J Cancer 33 (1984) 231--237. 27. Pukkala E, Teppo L, Hakulinen T, Rimpela M. Occu pation and smoking as risk determinants of lung can cer. Int J Epidemiol 12 (1983) 290--296. 28. Sackett DL. Bias in analytic research. J Chronic Dis 32 (1979) 51-63. 29. Schlesselman JJ. Case-control studies: Design, conduct, analysis. Oxford University Press, New York, NY 1982. 30. Siemiatycki J, Day NE. Fabry J, Cooper JA. Discov ering carcinogens in the environment: A novel epidemi ologic approach. J Natl Cancer Inst 66 (1981) 217--225. 31. Spencer H. Pathology of the lung. Volume 2. Fourth edition. Pergamon Press, Oxford 1985. 32. Stayner LT, Wegman DH. Smoking, occupation, and histopathologv of lung cancer: A case-control study with the use of the Third National Cancer Survey. J Natl Cancer Inst 70 (1983) 421--426. 33. Stevens RG. Moolgavkar SH. A cohort analysis of lung cancer and smoking in British males. Am J Epidemiol 119 (1984) 624--639. 34. Vincent RG. Pickren JW, Lane WW, Bross I, Takita H, Houten L. Gutierrez AC, Rzepka T. The changing histopathologv of lung cancer: A review of 1682 cases. Cancer 39 (1977) 1647--1655. 35. Wang J-D, Miettinen OS. Occupational mortality studies: Principles of validitv. Scand J Work Environ Health 8 (1982) 153--158. 36. Weiss W, Boucot KR. Seidman H, Carnahan WJ. Risk of lung cancer according to histologic type and cigarette dosage. J Am Med Avsoc 222 (1972) 799--801. 37. Wiklund K. Swedish agricultural workers: A group with a decreased risk of cancer. Cancer 51 (1983) 566--568. 38. Winkelstein W, Shillitoe EJ. Brand R, Johnson KK. Further comments on cancer of the uterine cervix, smok ing. and herpesvirus infection. Am J Epidemiol 119 (1984) 1--22. Received lor publication: 13 September 1985 202 10003268