Document 71BKQ3geKpvZ4waaJEx9EadR8

A M E R I C A N JOURNAL OF INDUSTRIAL MEDICINE 36:75-82 (1 999) Smoking as a Confounder in Case-Control Studies of Occupational Bladder Cancer in Women Andrea 't Mannetje, MSC,' Manolis Kogevinas, BAD,'*JennyChang-Claude, M D , ~ ' Sylvaine Cordier, P L D , ~Carlos-Albert0 Gonzalez, M D , M~ artine Hours, M D , ~ Karl-HeinzJockel, M D , ~Ulrich Bolm-Audorff, M D , ~Elsebeth Lynge, PhD,8 Stefan0 Porru, M D , ~ Francesco Donato, MD,'O Ulrich Ranft, PhD,' Consol Serra, M D , ' ~Anastasia Tzonou, PhD,13 PaoloVineis, MD,I4 JurgenWahrendorf, mD,* and Paolo Boffetta, M D I ~ Background: In studies in men, risk estimates on occupation and bladder cancer are distorted by about 10% when not adjusting for smoking. We examined the degree to which occupational risk estimatesfor bladder cancer in women are confounded by smoking, and the degree of residual confounding by inadequate control of this effect. Methods: Primary data of I1 case-control studies on occupation and bladder cancerfrom Denmark, France, Germany, Greece, Italy, and Spain were pooled. Informationfor smoking and lifetime occupational history for 700 female cases and 2,425 female controls ages 30-79 was abstracted and recoded. Logistic regression was used to calculate odds ratios (OR) by occupation, applying five models which differed in their degree of adjustment for smoking. Results: In major occupational groups, risk estimates were distorted by less than 10%when not adjustingfor smoking. A statistically significant excess riskfor bladder cancer wasfound in I3 specific occupations and industries.In most occupations, adjustmentfor smoking led the ORs towards the null value, but all statistically signiJicantassociations were maintained after adjustment. In three occupations (lathe operators,field crop workers, and wood manufacturers), a statistically signijcant excess risk was masked when not adjustingfor smoking. In six occupations, estimates were distorted by more than 10% (-22% up to +40%). In occupations where smoking acted as a positive confoundel; the proportion of confounding removed using a dichotomous smoking variable (everlnever) was around 60%. In one occupation (buyers), controllingfor smoking status (eve6 never) led to over-adjustment, because the percentage of smokers was high but the quantity smoked was low. 'Respiratory and Environmental Health ResearchUnit, IMIM, Barcelona, Spain 2Deutsches KrebsforschungszentNm, Abteilung Epiderniologie, Heidelberg, Germany 31nstiM National de la Sante et de la Recherche Medicale-INSERM, Villejuif, France 41nstitutde,recerca epidemiologica clinica, Matarb, Spain 51nstitutd'Epidemiologie, UniversiteClaude Bernard, Lyon, France 61nstitut fur Medizinische Informatik, Biometrie und Epidemiologie, Essen, Germany 7HessischesMinisteriurn fur Frauen, Arbeit und Sozialordnung, Wiesbaden, Germany 81nstiite of Public Health, Universityof Copenhagen, Kobenhavn, Denmark glnstituteof OccupationalHealth, University of Brescia, Italy 'OUniversity of Brescia, Italy 'lMedical Instituteof EnvironmentalHygiene, Heinrich-HeineUniversity,Dusseldorf, Germany 12Centred'Estudis, Prograrnesi Serveis Sanitaris, Sabadell, Spain 13Departmentof Hygiene and Epidemiology,MedicalSchoolof Athens, Athens, Greece 14Unitof Cancer Epidemiology,Turin, Italy 15Unitof Environmental Cancer Epidemiology, International Agency for Researchon Cancer, Lyon, France Contract grant sponsor: the European Commission, Europe against Cancer Programme; Contract grant number: SOC 962007420sfO2; Contract grant sponsor: FIS, Spain; Contract grant number: 9711105E. *Correspondence to: Dr. Manolis Kogevinas, Respiratoly and Environmental Heaith Research Unit, lnstitut Municipal d'lnvestigacio Medica (IMIM), 80 Doctor Aiguader Rd., Barcelona08003, Spain. E-mail: kogevinas@imim.es Accepted: 7 March 1999 o 1999Wiley-Lis, Inc. 76 Mannetje et al. Conclusions:Tobacco smoking was not found to be a major confounderfor the association between occupation and bladder cancer in women. Most of this confounding effect could be removed by adjustment by smoking status (everhever),without consideration of amount or duration of smoking. Am. J. Ind. Med. 36:75-82, 1999. o 1999 Wiley-Liss,Inc. KEY WORDS: confoundingfactors; smoking; occupational exposure; bladder neoplasms; women INTRODUCTION Information on smoking is usually available in casecontrol studies examining occupational exposures, and adjustment for the potential confounding effect of smoking is, therefore, easily done. The magnitude of confounding by smoking has, however, rarely been quantified. This makes it difficult to evaluate to what extent risk estimates are confounded by smoking in occupational studies where smoking informationis not available,such as industry-based studies or studies based on cancer incidence or mortality statistics. In the absence of smoking data, methods for sensitivity analyses and external confounding adjustment have been described, enabling a rough estimate of the potential confounding by smoking [Axelson and Steenland, 1988; Gail et al., 1988; Greenland, 19961. Although such evaluations on a theoretical basis are useful, they have also been qualified as "educated guessing" [Greenland, 19961. The confounding effect of smoking in women has not been studied, although it could be different from that in men because of differences in smoking habits and employment characteristicsbetween genders. Confounding by smolung in studies examining occupational exposures can either be negative (smoking rate in occupation is relatively low, resulting in an underestimation of the RR) or positive (smoking rate is relatively high, resulting in an overestimationof the RR). In many countries, smoking behavior varies considerably by occupation [Stellman et al., 1988; Brackbill et al., 19881. In the US, the percentage of women having ever smoked regularly ranged from 59.4% in waitresses to 21.3% in farmers. Smoking rates also appear to be significantly higher in groups exposed to occupational hazards compared to groups less exposed [Stellmanet al., 19881. The degree of potential confounding depends on the difference in smoking habits between risk and reference group and on the strength of the association of the disease in question with smoking. Population surveys indicate that there is a wide range in occupation-specific smokingrates in both men and women of industrialized countries, suggesting that confounding by smoking might be considerable in specific diseases. The confounding effect of smoking on the RR for lung cancer has typically been found to be more important than that for bladder [Axelson and Steenland, 1988; Siemiatycki et al., 1988; Blair et al., 1988; Scherg, 1995; Chiazze et al., 1995; Greenland, 19961. An empirical investigation in a case-control study in Montreal [Siemiatycki et al., 19881looked at the degree of confounding bias related to smoking for bladder and lung cancer. For lung cancer, considerable distortions of unadjusted odds ratios (ORs) were found. For bladder cancer, the differences between adjusted and unadjusted ORs were not larger than 10% in 23 of 25 occupations. A higher than 20% distortion was seen in only one occupation (teachers). The difference between bladder and lung cancer can be attributed to the weaker association with smoking for bladder compared to lung cancer. Even for bladder cancer, however, the risks for smoking are higher than those typically found for occupation. The residual confounding by smoking that can be caused by using a restrictively categorized variable for smoking (Le., `without taking into account quantity and duration of smoking) has been found to be limited [Cordier et al., 1993; Siemiatycki et al., 19881. In this study, we quantified the magnitude to which the occupational risk estimates for bladder cancer are confounded by smoking in women using a large pooled dataset of 11 case-control studies with detailed occupational and smoking information. We also estimated the degree of residual confounding when limited rather than detailed information on smoking is available. MATERIAL AND METHODS We performed a pooled analysis of case-control studies on bladder cancer that were conducted in European countries between 1976 and 1996 (Mannetje et al., in press). A total of 11 studies were identified using as criteria the availability of detailed occupational and smoking information and the accessibility of the primary data. Three additional European studies with detailed information on occupation were identified, one each from the UK [Cartwright, 19821, Sweden [Steineck et al., 19901, and Finland [Tola et al., 19801 but for various reasons it was not possible to incorporate them in the pooled analysis. The pooled file comprises data from three studies from Germany [Claude et al., 1988; Bolm-Audorff et al., 1993; Greiser and Molzahn, 19971, two each from France [Cordier et al., 1993; Hours et al., 19941, Italy [Vineis and Magnani, 1985; Porm et al., 19961, and Spain [Gonzilez et al., 19891 (Serra, unpub- Occupational Bladder Cancer in Women 77 TABLE 1. Case-Control Studies of Bladder Cancer Cases Controls ~n(hosPim)~ Mean year -n -% -n -% n-h)) All Country ofenrollment 700 2,425 (781/1,664) Jensen et al. Denmark 1980 96 13.7 193 8.0 (01193) Claude et al. Germany 1980 95 13.6 120 4.9 (120/0) g, Rebelakos et al. Greece 1982 47 6.7 44 1.8 (4410) a1 Vineis et at. Italy 1982 39 5.6 183 7.5 (183/0) li- Cordier et al. France 1985 83 11.9 101 4.2 (101/0) as Gonzalezet al. Spain 1986 49 7.0 120 4.9 (57163) 'g Hours et al. France 1986 19 2.7 38 1.6 (3810) >S Bolm-Audorffet al. Germany 1991 51 7.3 55 2.3 (5510) ZS Porn et al. blY 1992 30 4.3 163 6.7 (16310) in Greiser et al. Germany 1993 176 25.1 1390 57.3 (011390) )n Sem (unp.) SDain 1995 15 2.1 18 0.7 (0/18) :e le Age 30-39 years 12 1.7 109 4.5 17u37) to 40-49 years 26 3.7 240 9.9 (105/135) 33: 50-59 years a- 60-69 years 119 17.0 253 36.1 573 23.6 (152/421) 760 31.3 (221/539) le 70-79 years 290 41.4 743 30.6 (2111532) 3r Smoking id Nonsmokers 383 55.1 1649 68.2 (60811041) 3"I7 Ex-smokers 111 16.0 322 13.3 (49/273) Current smokers le l-10,000packs 201 28.9 111 16.0 446 18.5 (103/343) 316 13.1 (86/230) 1- 10,00~15,00p0acks 46 6.6 81 3.4 (1lffO) :t >15,000 packs id 44 6.3 49 2.0 (6/43) If Paldiclpatingstudieswith number offemale cases and controls, age distribution,and smokinghablts. .d lished) and one each from Greece [Rebelakos et al., 19851 more than 2 years time between diagnosis and interview. and Denmark [Jensen et al., 19871. This final pooled dataset comprised 700 cases and 2,425 Primary data from these 11 studies were combined hospital or population controls. Controls were individually :S using common coding and classification schemes for all or frequency matched to cases on age (within 5 years) and 1- variables. Originally used occupational codes were recoded geographic area (Table I). The case-control ratio differed per A to 5-digit ISCO (International Standard Classification of study, ranging from 1:l to 1 9 , with Germany forming the .e Occupations) 1968 codes [ILO, 19691and industry codes to largest national component for cases as well as controls 1- ISIC (International Standard Industrial Classification of All (Table I). 1- Economic Activities) revision-2 codes [Statistical Office of Unconditional logistic regression was applied, since 1- the United Nations, 19711.All studies recorded the occupa- most studies used frequency matched controls. ORs were t, tional and industry code of every job done more than 6 calculated for major occupations and industries (one-digit :t months and year of starting and end of each job, except for ISCO and ISIC codes), and for those specific occupations 0 one study [Rebelakos et al., 19851, for which only informa- and industries showing a statistically significant excess risk e tion for the longest held job was available. All studies for women (Mannetje et al., in press). Five alternative :t included detailed smoking information, including duration models were used, differing in their degree of adjustment for 1, and quantity smoked. We limited the analysis to cases and smoking: :t controls within the age range 30 up to 79 years and excluded subjects outside this age range (ncase=s 161, ~ , , m l s = 357) Model 1. (Crude OR), no adjustment for smoking. I- and also prevalent cases (ncasc=s 92), defined as those with Model 2. Smoking adjusted as a binary variable (neverlever). 78 Mannetje et al. Model 3. Smoking adjusted as a 3-categoryvariable (never/ edcurrent). Model 4. Smoking adjusted as a 5-categoryvariable (never/ edcurrent stratified in three groups by cumulative tobacco consumption). Model 5. Smoking adjusted as three variables, of which two binary (smokerhonsmoker; currenthot current), and one continuous variable (ln(1ifetimeconsump- tion in packs+1)). In addition, all ORs were adjusted for age (5-year age groups) and study. The large multicenter study in Germany [Greiser and Molzahn, 19971, was considered as three separate study areas (former East and West Germany and Berlin). The interaction term between center and age (continuous) was found to be statistically significant and therefore was also included in all models. Confoundingrisk ratios [Axelsonand Steenland, 19881, were calculated dividing the crude OR (for bladder cancer due to occupation) without adjustment for smoking (Model l), by the "true" OR fully adjusted for smoking (estimated by Model 5). Residual confounding was quantified by calculating the percent adjustment obtained in each model [Savitz and Bar&, 19891: OK(,) = crude OR for exposure and disease (c = crude) OR,(e) = OR adjusted with the misclassified confounder (pa = partially adjusted) O&,(e> = OR adjusted with the correctly classified confounder (estimatedby model 5 ) The percent adjustment reflects the percentage of confounding removed by using a smoking variable in the model, which partially adjusts for smoking (i.e., does not sufficiently take the variation in tobacco smoking into account), compared to full adjustment. In this case, we calculated the percent adjustment achieved by using Models 2, 3, and 4, assuming that the use of Model 5 yielded to complete confounding adjustment. The percent adjustment was not calculated when the confounding by smoking was smaller then 10%and when partial and complete adjustment showed different confounding directions. RESULTS Smoking Compared to never smokers, current smokers had more than a 3-fold risk for bladder cancer (OR = 3.61, 95% CI 2.80-4.65) and ex-smokers had an OR of 2.52 (95% CI 1.90-3.36). The ORs for "ever" smoking ranged between studies from 1.42 to 8.07, but the interaction term between smoking and study was not statistically significant. Major Occupational and Industry Groups (One-Digit ISCO and lSlC Codes) The prevalence of women having ever smoked for 1 year or more was 45% among cases and 32% among controls. Prevalence of smoking in the main occupational groups ranged from 22-59% in cases and between 1 8 4 8 % in controls (Table 11). Smoking prevalence was lowest in agriculturalworkers (ISCO 6) and highest in administrative/ managerial workers and clerical workers (ISCO 2 and 3). A similar pattern was observed for the main industry groups, although extreme smoking prevalences were seen for two industry groups (miners, electricity workers) including only a few cases. These high prevalences were not found in the larger control groups, indicating they were due to sampling variation because of small numbers rather than due to job characteristics. Crude ORs (Model l), smoking-adjusted ORs (Models 2-5), and the confounding RR (Model 1 divided by Model 5) for the major occupationaland industry groups are shown in Table II. Smoking appears to be a positive confounder in most major occupation groups, although in nearly all groups the confounding effect was not more than 10%. Higher confounder RRs were seen for major industry groups, particularly mining (confounder RR = 1.24), transport (confounder RR = 1.10) and finance (confounder RR = 1.17). Among these major occupationshndustries, agricultural workers and those in community, social, and personal services had a statistically significant lower risk for cancer of the bladder. Only the association of agricultural workers (ISCO 6, ISIC 1) and bladder cancer appeared to be negatively confounded by smoking (confounder RR around 0.90). The percentage of positive confounding removed by using basic smoking information was high for those occupations and industries most confounded by smoking. For these major occupation and industry groups, at least 78% of confounding was removed by using Model 2 (i:e., smoking entered in the model as a dummy variable, nevedever). High-Risk Occupations and Industries In the specific occupations and industries showing statistically significant excess risk ( P < 0.05),the confounding effect of smoking was more substantial (Table 111).The deviations in crude ORs (Model 1) compared to adjusted ORs (Model 5) ranged between -22% and +40% (confounder RR 0.78-1.40). In approximately half of these occupations confounding was 10% or less. In 10 of 13 high-risk occupationshndustries,the confounding effect was positive. In the three occupations negatively confounded by Occupational Bladder Cancer in Women 79 TABLE II. Risk for Bladder Cancer in Major Occupations and Industries. Adjustment for Smoking in Four Different Cases N(ever.m)/ n1-r mhsd) (% ever smoked) Controls N ( s v s r ~ ) / Model 1 n(lMwr Crude OR (% ever smoked) (95%CI) Model 2 Model 3 Model 4 Model 5 Confounding Adjusted OR Adjusted OR Adjusted OR Adjusted OR RR (95%CI) (95%CI) (95%Cl) (95%CI) OR(,)/ORn Occupations ISCOl Professional,technical, and related ISCO2 Administrativeand managerial workers ISCO3 Clerical and related ISCO4 Sales workers ISCO5 Serviceworkers ISCO6 Agricultural,animal husbandry, forestry workers, fishermen, and hunters ISCO9 Production and related workers Industries ISlCl Agriculture, hunting, forestry, and fishing ISIC2 Mining and quarrying ISC13 Manufacturing ISIC4 Electricity, gas, and water ISIC5 Construction ISIC6 Wholesale and retail trade, restaurants, hotels ISIC7 Transport, storage, communications SIC8 Finance, insurance, real estate, business services ISIC9 Community, social, and personal services 35135 (!%yo) 716 (54%) 106/75 (59%) 78/81 (49%) 1231109 (53%) 2013 (22%) 1231132 (48%) 14135 (29%) 2/0 (1OOYO) 1461137 (52%) 411 (80%) 816 (57%) 93/78 (54%) 18110 (64%) 25111 (69%) 1401117 (54%) 1771283 (38%) 23125 (48%) 3301497 (40%) 2231343 (39%) 3031566 (35%) 571262 (18%) 0.81 (0.60-1.08) 1.07 (0.54-2.11) 1.15 (0.92-1.44) 1.17 (0.93-1.46) 0.97 (0.79-1.19) 0.86 (0.65-1.14) 0.80 (0.59-1.08) 0.98 (0.49-1.98) 1.05 (0.84-1.33) 1.14 (0.91-1.44) 0.92 (0.75-1.13) 0.96 (0.72-1.28) 0.81 (0.60-1.09) 1.01 (0.50-2.03) 1.06 (0.84-1.33) 1.16 (0.92-1.46) 0.92 (0.74-1.1 3) 0.96 (0.72-1.28) 0.80 (0.59-1.08) 1.oo (0.49-2.01) 1.06 (0.84-1.33) 1.16 (0.92-1.46) 0.92 (0.74-1.13) 0.95 (0.72-1 27) 0.81 (0.60-1.10) 1.04 (0.52-2.09) 1.06 (0.84-1.33) 1.16 (0.92-1.46) 0.91 (0.74-1.13) 0.96 (0.72-1.27) 1.oo 1.03 1.08 1.01 1.07 0.90 3071587 (34%) 1.14 1.10 1.09 1.10 1.08 (0.94-1.39) (0.90-1.34) (0.89-1.34) (0.90-1.34) (0.88-1.33) 1.06 , 571242 (19%) 6114 (Soyo) 384/712 (35%) 8115 (35%) 22/39 (36%) 2841447 (39%) 69186 (45%) 84191 (48%) 441/779 (36%) 0.59 (0.42-0.85) 0.92 (0.18-4.60) 1.25 (1.02-1 53) 1.46 (0.50-4.22) 1.43 (0.76-2.71) 1.oo (0.80-1.25) 1.01 (0.64-1.58) 0.83 (0.56-1.25) 0.78 (0.63-0.96) 0.64 (0.45-0.92) 0.74 (0.14-3.99) 1.20 (0.97-1.48) 1.37 (0.47-4.00) 1.34 (0.69-2.59) 0.97 (0.77-1.22) 0.91 (0.57-1.44) 0.73 (0.48-1.10) 0.74 (0.59-0.91) 0.64 (0.45-0.91) 0.73 (0.13-3.99) 1.20 (0.97-1.48) 1.39 (0.48-4.00) 1.37 (0.71 -2.66) 0.98 (0.78-1.23) 0.90 (0.57-1.43) 0.71 (0.47-1.08) 0.74 (0.60-0.91) 0.64 (0.44-0.91) 0.77 (0.14-4.16) 1.21 (0.98-1.49) 1.41 (0.49-4.07) 1.38 (0.71-2.68) 0.97 (0.77-1.22) 0.90 (0.57-1.44) 0.71 (0.47-1.09) 0.73 (0.59-0.91) 0.64 (0.45-0.92) 0.74 (0.13-4.13) 1.20 (0.97-1.48) 1.44 (0.50-4.19) 1.35 (0.69-2.62) 0.96 (0.76-1.21) 0.92 (0.58-1.47) 0.71 (0.47-1.08) 0.75 (0.61-0.93) 0.92 1.24 1.24 1.04 1.01 1.06 1.04 1.10 1.17 1.04 'Subjects having worked for morethan 6 months in a job. Subjects may contributeto morethan one occupation or industry. *Model 1:Crude OR, No adjustmentfor smoking; Model 2: Smoking adjusted as a binary variable (nevedever); Model 3: Smokingadjusted as a 3-categoryvariable (never/ex/current); Model 4: Smoking adjusted as a 5-category variable (never/ex/currentstratifiedin three groups by cumulativetobacco consumption); Model 5:Smoking adjusted as three variablesof which two binary +(srnokednon-smoker;currenthot current), and one continuousvariable (In (lifetimeconsumption in packs 1)). smoking (lathe operators, field crop workers, and wood manufacturers), a statistically significant excess risk was masked when not adjusting for smoking. The widest differences between crude and adjusted ORs were mostly seen for occupations and industries with a small number of cases. For the-high risk occupations with 20 cases or more, the deviation in OR was not found to be more than lo%, with the exception of the relatively large group of field crop workers (qcasCs=) 33), which showed a negative confound- ing of - 17%. 80 Mannetje et al. TABLE 111. Occupations and industries With a Statistically Significant High Risk for Bladder Cancer1,*;Adjustment for Smoking in Four Different Ways3 Cases Controls N(mm N(-, -/ n(- w) n(mr -1 (% ever smoked) (% ever smoked) Model 1 Crude OR (95%CI) Model 2 Model 3 Model 4 Model 5 Confounding Adjusted OR Adjusted OR Adjusted OR Adjusted OR RR (95%CI) (95%Cl) (95%CI) (95%CI) ORfl)/ORO Occupations ISCO83 Blacksmiths, tool- makers, and machine- tool operators ISCO83420 Lathe operator ISCO622 Fieldcrop and vegetable farm workers ISCO78 Tobacco preparers and tobacco product makers ISCO791 Tailors and dressmakers ISCO45190 Other salesmen, shop assistants, and demonstrators ISCO37020 Mail sorting clerk ISCO422 Buyers Industries ISIC314 Tobacco manufacture ISIC369 Manufacture of other non-metallic mineral products, except products of petroleum and coal ISIC331 Manufacture of wood and wood products, except furniture ISIC3 Manufacturing ISIC371 Iron and steel basic industries 9111 45% 013 0% 1I32 3% 513 63% 30/30 50% 314 43% 410 100% 311 75% 515 50% 513 63Ya 2J4 33% 1461137 52% 412 67% 17/34 33% 2J6 25% 3/38 7% 315 38% 57/90 39% 10111 48% 1I 5 17% 513 63% 2.05 1.96 1.91 1.94 1.87 (1.13-3.70) (1.07-3.59) (1.04-3.51) (1.06-3.57) (1.02-3.45) 3.42 (0.83-14.0) 1.48 (0.87-2.53) 3.81 (1.25-1 1.7) 4.55 (1.08-19.2) 1.a2 (1.05-3.14) 3.35 (1.1 1-10.1) 4.57 (1.09-19.1) 1.a0 (1.04-3.12) 3.26 (1.09-9.78) 4.61 (1.11-19.2) 1.78 (1.03-3.08) 3.1 2 (1.05-9.28) 4.36 (1.02-18.6) 1.78 (1.03-3.08) 3.08 (1.03-9.22) 1.47 (1.04-2.08) 2.74 (1.06-7.06) 1.42 (1.OO-2.03) 2.58 (0.99-6.74) 1.42 (0.99-2.02) 2.65 (1.02-6.89) 1.44 (1.01-2.06) 2.63 (1.01-6.85) 1.43 (1.00-2.04) 2.51 (0.95-6.61) 5.56 (1.39-22.2) 5.54 (1.31-23.4) 4.75 (1.10-20.5) 4.08 (0.89-18.6) 4.72 (1.09-20.4) 4.49 (0.96-20.9) 4.43 (1.01-19.5) 4.56 (1.OO-20.9) 4.35 (0.99-19.1) 5.03 (1.12-22.5) 1.10 0.78 0.83 1.24 1.03 1.09 1.28 1.10 5110 2.96 3.12 3.09 2.98 2.92 1.01 33% (1.20-7.29) (1.25-7.75) (1.24-7.67) (1.20-7.36) (1.18-7.25) 616 4.78 3.47 3.48 3.55 3.41 1.40 50% (163-14.0) (1.15-10.5) (1.16-10.5) (1.18-10.7) (1.12-10.3) 1l a 11% 3841712 35% 513 63% 3.28 3.54 3.38 3.52 3.53 (0.97-1 1.1) (1.02-12.3) (0.98-11.7) (1.02-12.1) (1.02-12.2) 0.93 1.25 (1.02-1.53) 3.93 (1.11-13.9) 1.20 (0.97-1.48) 2.78 (0.80-9.69) 1.20 (0.97-1.48) 2.90 (0.83-10.2) 1.21 (0.98-1.49) 3.00 (0.85-1 0.5) 1.20 (0.97-1.48) 3.01 (0.83-11.O) I 1.04 1.31 'All OCCUpatiOnS and industries (ISCO 1968and ISIC-rev2codes) with statistically significant high risk (P-value < 0.05). %ubjects having worked for more than 6 monthsin a job. Subjects may contribute to more than one occupation or industry. 3Model1: Crude OR,No adjustmentfor smoking; Model 2: Smoking adjusted as a binary variable (nevedever); Model3: Smoking adjustedas a 3-categoryvariable (never/ex/cunent); Model 4: Smoking adjusted as a 5-category variable (never/ex/current stratified in three groups by cumulativetobacco consumption); Model 5:Smoking adjusted as three variables of which two binary (smokednon-smoker;currenthot current),and one continuous variable (In (lifetimeconsumption in packs + 1)). In most high-risk occupations/industries,adjustment for smoking using detailed information did not lead to important changes in the crude ORs (Table 111).In some occupations, however, residual smoking confounding was observed when using a dichotomous (nevedever) smoking variable, particularly for tobacco workers (ISCO 78) and mail sorting clerks (ISCO 37020). In these two occupations, the proportion of confounding removed using a dichotomous smoking vari- Occupational Bladder Cancer in Women 81 able (nevedever) was around 60% (63% and 67%, respec- Deviations of more than 10% were, however, mainly found tively). When ex- and current smokers were separately for occupations and industries with less than 20 cases, classified (Model 3), the proportion of confoundingremoved suggesting an effect of sampling variation. was 76% and 69%, respectively, while using a 5-category Residual confounding can arise in various ways: be- smoking variable (Model 4) it was 95% and 93%. Using a cause of error in the measurement of the confounder, or due dichotomous smokmg variable (nevedever) caused an over- to less than optimal stratification or modeling of the adjustment of the risk in several occupations/industries.This bias was more pronounced in buyers (ISCO 422). In this occupation, adjustment for smoking as a dichotomous variable led to a lower OR, though full adjustment led again to a higher one. This occurred because the percentage of "ever smokers" among buyers was high, while the amount of cigarettes they smoked was low. confounder [Siemiatyckiet al., 19881.In this study, we were only able to look at the latter using different degrees of smoking information. In most occupationshndustries, residual confounding was limited when adjusting for smoking using a dichotomous variable. It would, however, be recommendable to incorporate additional information including the quantity and duration of smoking, since under- as well as over-adjustment can occur when this information is not DISCUSSION taken into account. For two high-risk occupations, ORs were not sufficiently adjusted for smoking when using only Various criteriahave been proposed regarding confound- smoking status in the model (residual confounding of 40%) ing selection policies, particularly in situations where it is and for one occupation (buyers) the OR was clearly over- not known a priori whether an exposure may confound adjusted when the quantity smoked was not taken into another exposure-effect estimate [Maldonado and Green- account. Using pack-years as a continuous variable together land, 1993; Breslow and Day, 1980; Rothman and Green- with smoking status variables did not appear to give better land, 19981. Often, a variable is retained in a model only if adjustment results than using pack-years in strata (in this v" its inclusion seems to make an "important difference" in the case, three strata of cumulative exposure). exposure4fect estimate, "important" having sometimes In most cases, the confounding effect appeared to be been defined as a bigger than 10%change in the rate ratio of positive (10 out of 13 high-risk occupations and industries). interest [Maldonado and Greenland, 19931. The results This supports the notion that in most high-risk occupations presented here show that the crude ORs deviate from the the smokingprevalence is also higher [Stellman et al., 19881 smoking adjusted ORs by more than 10% in approximately half the occupationshndustries statistically significantly associated with bladder cancer. Although the results of our analyses refer to the specific population examined in the case-control studies, they seem to have a wide applicability since this population included women living in several European countries and having been exposed during long time periods. Two mechanisms underlie the variation in crude and adjusted ORs [Siemiatycki et al., 19881: 1) the variation in smoking rates between the occupation under study and the reference population, due to job-specific differences in smoking habits, in which case smoking is associated with disease and occupation; 2) the variation in smoking rates between the occupation under study and the reference and that in these cases positive confounding can be expected. In none of the occupationshndustries showing a statistically significant excess risk for bladder cancer could this be due to confounding by smoking. Even if less frequent, negative confounding by smoking appears to be equally important as positive confounding, since it can mask the excess risk of specific occupations, as can be illustrated by the group of "field crop workers." This occupation would not have been selected as being a statistically significant high-risk occupation if adjustmentfor smoking had not been made. In conclusion, tobacco smoking was not found to be a major confounder for the association between occupation and bladder cancer in women, although in a few occupations population, due to sampling variations, in which case risk could be under- as well as overestimated by more than smoking is not necessarily associated with occupation. 20%. Although adjustment for smoking taking into account When studying large occupational groups for their bladder both smoking status and quantity and duration of smoking is cancer risk, the effect of the second mechanism is negligible preferable, it should be noted that most of the confounding and this study showed that the true confounding effect is effect could be removed by adjustment by smoking status mostly limited to a deviation in crude ORs of less than 10%. (everhever), without considerationof amount or duration of This is in concordance with findings from other studies on smoking. confounding by smoking on occupation-bladder cancer associations. When studying small occupational groups REFERENCES (which is mostly the case of high-risk occupations, particu- larly in women, where the number of subjects exposed is Axelson 0,Steenland K. 1988. Indirect methods of assessing the effects of often small), it is not possible to separate the two effects. tobacco use in occupational studies.Am J Ind Med 13:105-118. i 82 Mannetje et al. Blair A, Steenland K, Shy C, O'Berg M, Halperin W, Thomas T. 1988. Control of smoking in occupational epidemiologic studies: methods and needs.Am J Ind Med 13:34. Bolm-Audorff u, Jockel m, GlguSS B. Pohlabeln H, Siepelkothen T. 1993.Schriftenreiheder Bundesanstalt fiir Arbeitsschutz(Occupationalrisk factors for cancer of the lower urinary tract). Bremerhaven: Wirtschaflsverlag NW. p 1-179. Brackbill R, Frazier T, Shilling S. 1988. Smoking characteristics of US workers, 1978-1980. Am J Ind Med 13:5-41. Breslow NE,Day NE.1980. Statisticalmethods in cancer research. Vol. I. The design and analysisof case-controlstudies.Lyon:IARC Sci Pub No 32. InternationalAgency for Research on Cancer. Cartwright R. 1982. Occupational bladder cancer and cigarette smoking in West Yorkshire. Scand J Work Environ Health 8(Suppl 1):79-82. Chiazze L Jr, Watkins DK, Fryar C. 1995. Adjustment for the confounding effect of cigarette smoking in an historical cohort mortality study of workers in a fiberglass manufacturing facility. J Occup Environ Med 37:744-748. Jc,Frentzel-Beyme RR9 Kmze E' 1988' Occupation and risk Of cancer of the lower urinary tract among men. A case-control study. Int J Cancer 41:371-379. Cordia S, Clavel J, Limasset JC, Boccon-Gibod L, Moua m E , Mandereau L, Hemon D. 1993. Occupational risks of bladder cancer in France: a multicentriccase-controlstudy. Int J Epidemiol22:403411. Gail MH, Wacholder s,Lubin JH. 1988. Indirect COlECtiOnS for confound- ing under multiplicative and additive risk J Ind Med 13:119-130. Gonzitlez CA, L6pez-Abente G, Errezola M, Escolar A, Riboli E, I=zugaza I, Nebot M. 1989. Occupation and bladder cancer in Spain: a multi-centre case-control study. Int J Epidemiol 18:569-577 Greenland s. 1996. Basic methods for sensitivity analysis of biases. Int J Epidemiol25:1107-1116. Greiser E, Molzahn M, editors. 1997.MultizentrischeNieren- und UrothelCarcinom-Studie (Multicenter urothelial and renal cancer study). Bremerhaven: Wirtschaftsverlag NW, Verlag fir Neue Wiss. (Schriftenreihe der Bundesanstaltfiir Arbeitsschutzund Arbeitsrnedizin: Forschung; Fb 780.) Hours M, Dananche B, Fevotte J, Bergeret A, Ayzac L, Cardis E, Etard JF, Pallen C, Roy P, Fabri J. 1994. Bladder cancer and occupational exposures. Scand J Work Environ Health 20:322-330. ILO (International Labour Office). 1969. International standard classification of occupations.Revised edition 1968, Geneva. Jensen OM, Wahrendorf J, Knudsen JB, Sorensen BL. 1987. The Copenhagen case-referent study on bladder cancer. Risks among drivers, painters and certain other occupations.Scand J Work Environ Health 13:129-134. Maldonado G, Greenland S. 1993. Simulation study of confounderselection strategies.Am J Epidemiol 138:923-936. Mannetje A, Kogevinas M, Chang-Claude J, Cordier S, Gonz&lezCA, Hours M, Jockel KH, Bolm-Audoff U, Lynge E, porn S, Donato F, Ranft u, Sema c,Tzonou A, Vineis p, W&endod J, Boffem p. Occupation and bladder cancerin E~~~~ cancecrausecosntrol (in press. Porn S, Aulenti V, Donato F, Bofetta P, Fazioli R, Cosciani Cunico S , Alessio L. 1996. Bladder cancer and occupation: a case-control study in northern Italy. Occup Environ Med 5 3 6 1 0 . Rebelakos A, Trichopoulos D, Tzonou A, Zavitsanos X, Velonakis E, TrichopoulouA. 1985.Tobacco smoking, coffeedrinking and occupation as a risk factors for bladder cancer in Greece. JNCI75:455461. Rothman KJ, Greenland S. 1998. Modem epidemiology,Znd ed. Philadelphia: Lippincot- Raven Publishers. Savitz DA, Bar6n AE. 1989. Estimating and correcting for confounder misclassification.Am J Epidemiol 129:1062-1071. Scherg H,1995. Die fehleinschatzungbemflicher K&,sris&en in epidemi- ologischen studien durch unkenntnis des raucherstatus am beispiel von hamblasen und lungenkrebs der baumaler (Confounding of occupational cancer risk in epidemiological SNdieS due to ignorance Of smoking data as exemplified by bladder and lung cancer in painters). Soz Raventivmed 40:302-308. SiemiatyckiJ, Wacholder s,D~~~ R, Cardis E, Greenwood C, Richardson L, 1988. Degree of confoundingbias related to smoking, ethnic goup, and socioeconomic Status in estimates of the associations between occupation and cancer. OccUpMed 30:617425. Statistical Office of the United Nations. 1971. International standard industrial classification of all economic activities. Statistical papers series M, number 4, revision 2. New yo&: United Nations, Steineck G, Plat0 N, Gerhardsson M, Norell SE, Hogstedt C. 1990. Increased risk of urothelial cancer in Stockholm during 1985-87 after exposure to benzene and exhausts. Int J Cancer 45: 1012-1017. Stellman SD, Boffetta P, Garfinkel L. 1988. Smoking habits of 800.000 American men and women in relation to their occupations. Am J Ind Med 13:43-58. Tola S, Tenho M, Korkala, ML, Jminen E. 1980. Cancer of the urinary bladder in Finland. Association with occupation. Int Arch Occup Environ Health 46:43-5 I . Vineis P,Magnani C. 1985. Occupation and bladder cancer in males: a case-control study. Int J Cancer 35:599-606.