Tobacco Documents Online

" Publication of the International Union Against Cancer © 1989 Alan R. Lies, Inc, 'Publication de I'Union~lnternationale. Contre le Cancer RISK FACTORS FOR ORAL CANCER IN BRAZIL: A CASE-CONTROL STUDY Eduardo L. Feo, r~co1,s, Luiz P. Kow.~r~sK~z, Benedito V. Or.Iv~Im~~, M. Paula Cueo, oo'~, Raimunda N. Pl~,~m~1, M. Estela SmVAI, Antonio S. F~,v.~~ and Humberto Toea.ON~~ ~Ludwig Institute for Cancer Research, R. Prof. Antonio Prudente, 109, 01509 S~o Paulo; ~Servigo de Cirurgia de Cabega e Pesco¢o, Hospital Heli6polis INAMPS, Ro COnego Xavier, 276, 04231 Sgto Paulo; 3Hospital Erasto Gaertner, Liga Paranaense de Combate ao Cdncer, R. Ovande do Amaral, s/n, 81500 Curitiba; and 4Hospital Ara(tjo Jorge, Associa~do de Combate ao • Cancer de Goids, R. 239, 181, 74000 Goidnia, Brazil. ! Int. J. Cancer: 43, 992-1000 (1989) A case-control study of risk factors for carcinomas of the tongue, gum, floor, and other specified parts of the mouth was conducted in 3 metropolitan areas in Brazil: S~o Paulo (South-east), Curitiba (South), and Goi~nia (Central-West). We analyzed information on demographics, occupational his- tory, environmental exposures, tobacco smoking and alcohol drinking habits, as well as diet, oral and other health charac- teristics obtained from interviews with 232 cases and 464 hos- pital non-cancer controls matched for 5-year age-group, sex, hospital catchment area and trimester of admission. Tobacco and alcohol consumption were the strongest risk factors irre- spective of the anatomical site. The adjusted relative risks RR) for ever vs. never smokers were: 6.3, 13.9, and 7.0, for ndustrial-brand cigarettes, pipe, and hand-rolled cigarettes, respectively. A strong correlation was seen between number of pack-years and risk. The RR for the heaviest vs. the lowest consumption categories (> 100 vs. < I pack-years) was 14.8. Risk levels generally decreased to those of never smokers after 10 years had elapsed since stopping smoking. The risk associated with alcohol was mostly evident for wi_ne (cancer of the tongue) and "cacha~.a" (all sites), a hard liquor distilled from sugar cane. Other important risk factors were drinking "chimarr~o" (a type of mat~), use of a wood stove for cook- ing, and frequent consumption of charcoal-grilled meat and manioc. Oral hygiene characteristics represented correlates of disease risk. A significant protective effect was observed for consumption of carotene-rich vegetables and citric fruits, but not for green vegetables in general. Incidence rates for cancer of the mouth in some metropolitan areas in Brazil are among the highest in the world. The rates for cancer of the tongue [WHO, International Classification of Diseases, 9th Revision (ICD-9) rubric 141] and other parts of the mouth, excluding lip (ICD-9 143-145) among male inhab- itants of S~o Paulo are 7.4 and 8.0 per 100,000, ~espectively; age-standardized to the world population of 1960 (Mirra and Franco, 1985). The risk of buccal cancer expressed by these rates is second only to those observed in selected areas of India and France (Muir et al., 1987). Higher risk areas in the country exhibit rates that are 2 to 3 times higher than those seen in lower-risk areas. Several epidemiologic investigations have shown that to- bacco and alcohol consumption are the 2 most important risk factors for oral cancer (Blot et al., 1988). Some dietary factors and oral health characteristics represent additional determi- nants of risk that have been identified in more recent studies (Winn et al., 1984; McLaughlin et al., 1988; Graham et al., 1977; Young et al., 1986). In contrast to findings with respi- ratory system cancers, however, the role of occupational ex- posures has not been unequivocally demonstrated for cancer of the oral cavity (Moss and Lee, 1974; Elwood et al., 1984; Young et al., 1986). In order to quantify the importance of risk factors for cancer of the oral cavity in Brazil, we conducted an epidemiologic study in 3 metropolitan areas: S~o Paulo (South-east), Curitiba (South), aM Goi~nia (Central-Western Brazil). A hospital- based, case-control study design was used to infer exposure- disease relationships for variables such as: smoking and drink- ing history, dietary factors, employment history, and gene.ral and oral health characteristics. MATERIAL AND METHODS • ,- Study subjects All patients with newly diagnosed carcinomas of the tongue, gum, floor of the mouth, and other parts of the oral cavity (ICD-9 141 and 143-145) referred to 3 head-aM-neck surgery services in S~o Paulo (Heli6polis Hospital), Curitiba (Erast0 Gaertner Hospital), and Goihnia (Arafijo Jorge Hospital) dur- ing the period February 1, 1986 to June 30, 1988 were con- sidered eligible for the study. Patients with malignant neo- plasms of the lip or of the salivary glands (ICD-9 140 and 142) were not included in the investigation. All diagnoses were confirmed histopathologically, and the anatomic site was as- certained post-surgically. With the exception of the head- and-neck surgery service in S~o Paulo, which is responsible for approximately 20% of all incident cases of the city, the patient accrual in the other 2 centers approached 100% of all incident cases in their respective areas during the period of the study. Control subjects were selected among patients of the same hospital to which cases had been admitted or from neighboring general hospitals. Two control patients were matched to each case on the basis of sex, 5-year age-group, and trimester of hospital admission. A diagnosis of neoplastic disease (ICD-9 140-239) or of mental disorder (ICD-9 290-319) was consid- ered as grounds for ineligibility among candidate control pa- tients. Interviews - All subjects were submitted to a 40- to 60-rain structured, questionnaire-based, standardized interview. Interviewers, specially trained for the study, were totally blind to all etio- logical hypotheses being tested. Interviews elicited detailed information on socio-economic and demographic variables, general health characteristics, history of environmental and occupational exposures, tobacco smoking and alcoholic bev- erage consumption, dietary habits based on past consumption frequency for selected items, and oral hygiene habits. Given the sensitive nature of some items, interviews were conducted in total privacy with complete assurance for the patient of the confidentiality of the information. Likewise, all interviews of case subjects were carried out prior to any major medical pro- cedure that could potentially affect the patient's ability to com" munieate or to recall information. Interviews were immediate~ interrupted if patients complained of (or if the interviewer sus- pected) difficulty in communication due to pain or speech problems. Such cases were eliminated from the study. No proxy respondent interviews were used in these instances. 5To whom reprint requests should be addressed. Received: November 24, 1988. o 0~

Sttdstical analysis ~e odds ratio was the measure of association used to esti- ~te the relative risk (RR) of disease due to each study factor• Point and interval estimates for the RR were obtained by mul- tiple logistic regression using conditional maximum likelihood ~stimation based on the matching factors (age, sex, study site, and admission period) (Breslow and Day, 1980; Campos-Filho and Franco, 1989)• Simultaneous adjustment by tobacco and alcohol consumption in the analysis was based on the lifetime cumulative exposure using pack-year equivalents of cigarette smoking and the sum over all beverage types of kilograms of ethanol consumption• A pack-year was defined as the cumu- lative exposure equivalent to smoking one pack of cigarettes daily during one year. Doses were calculated as follows: 20 manufactured cigarettes = 4 hand-rolled, black tobacco cigarettes = 4 cigars = 5 pipefuls with regular pipe tobacco = 1 pack; ethanol concentration in beer = 5%, wine = 10%, hard liquor and "cacha~a" = 50%. Adjust- ment for these synthetic indices was always based on the es- timation of 4 separate parameters representing the effects of 3 leveis of consumption (k-1 dummy variables per index). The 2 cut-off values for categorization of tobacco and alcohol con- sumption were selected so as to obtain the highest possible likelihood ratio statistics in individual 2-term models with each index. The statistical assessment of interaction (effect modification) was based on a multiplicative scale by fitting models contain- ing main effect variables and their cross-product terms. Infer- ence was based on the partial likelihood ratio (deviance) sta- tistic between nested models. Statistical trend in the dose-risk relationship for a given variable was assessed in models con- taining the variable treated as continuous instead of a set of indicator variables. Significance was based on the statistic rep- resenting the ratio of the estimated coefficient to its standard error assuming a standard normal distribution. All statistical inference was based on 2-sided alternative hypotheses• RESULTS A total of 236 oral cancer patients were considered eligible for the study. Four patients were considered incapacitated or t~ ill. Successfully completed interviews were obtained with ~he remaining 232 cases (98.3%). Interviews were also com- pleted with the corresponding 464 matched control patients. There were no refusals to participate among cases and controls. Distribution according to main characteristics p Case accrual by study site was as follows: 88 (37•9%) in Silo aulo, 86 (37 1%) in Curitiba, and 58 (25.0%) in Goi~mia, There were 10~ 4~.7%) cases of cancer of the tongue (ICD-9 141), 23 (9.9%) of cancer of the gum (ICD-9 143), 42 (18• 1%) ~g_eancer of the floor of the mouth (ICD-9 144), and ~ t~.3%) cases of carcinomas of other specified parts of the oral ~ity (ICD-9 145). There were 201 (86.6%) male and 31 ~.4%) female oatients. The a~ze distribution was as follows: ~40 e - 8" " y ars, 10 (4.3%); 40-49 years, 50 (21.6%); 50--59 years, " ) ~36 6%), 60-69 years, 56 (24 1%); and >-70 ears, 31 l " - " - Y (3•4%) patients• • The underlying causes of hospitalization among control pa- ,ti_ents could be grouped into 13 diagnostic categories of the ~13-9. Digestive system diseases (ICD-9 520-579) repre- • nted the most common cause (128 controls, 27•6%) followed .0y cardiovascular diseases (ICD-9 390-459, 109 controls, 23.5%) and trauma and poisoning (ICD-9 800-999, 37 con- ~ols, 8 0%) Representation among controls for the other spec- ked diagnostic categories ranged from a minimum of 2 (0.4%) ~r pregnancy-associated diseases (ICD-9 630-676) and con- ~.~i~ro~ faonr°remsa~aSto(IryCsDy~em74d0~is7ea5s9e)s }~CaD_~x~3_t~9)°.fT3hle RISK FACTORS FOR bRAL CANCER 993 cause of hospitalization was assigned as ill-defined (ICD-9 780-799) for 64 (13.8%) control patientlY Table I shows the distribution of cases and controls accord- ing to selected socio-economic and demographic variables. Monthly income categories were based on quintiles of the dis- tribution among controls with known answers for this variable. Definition of geographical region was based on standard cen- sus criteria for all Brazilian states. The proportion of white patients among cases (83%) was slightly higher than among controls (78%). Likewise, cases were slightly more likely to report lower family income values than controls, although the proportions of cases and controls with incomes in the 2 upper quintiles were similar. Although differences in proportions of rural residents were unremarkable, there was a clear distinction when stratifying for sex. Strictly urban residences were re- ported by 54 out of 201 male cases (26.9%) and 82/402 male controls (20.4%), whereas these proportions among female subjects were 3.2% (1/31) for cases and 22.6% (14/62) for controls. Occupational exposures Oral cancer RR estimates for the most often reported occu- pational categories are presented in Table II with frequencies of cases and controls by exposure type. Except for a moderate risk reduction associated with metal processing occupations [RR = 0.6, 95% confidence interval (95%CI): 0.3-1.0] cases and controls presented comparable distributions• RR estimates were unaffected when controlling for smoking and alcohol con- sumption. Table II also shows RRs according to sporadic oc- cupational or household exposures to possible sources of car- cinogens. Use of a wood stove for cooking and heating was more frequently reported by cases than by controls. The excess risk was not due to confounding by smoking or alcohol con- sumption. Stratification by anatomic site (tongue vs. other parts of mouth) did not materially change-the RR estimates for these v~ariables apart from the expected loss of precision• The only TABLE i ~ DISTRIBUTION OF CASES AND CONTROLS ACCORDING ~ SELECTED SOCIO-ECONOMIC AND DEMOGRAPHIC CHARACTERISTICS Variable Categories Cases (%) Controls (%) Ethnic background White 193 (83.2) 364 (78.4)~ Mulatto 26 (11.2) 79 (17.0) Black 10 (4.3) 18 (3•9) Other 3 (1.3) 3 (0•6) Marital status Never married 19 (8.2) 43 (9•3) Currently 170 (73•3) 351 (75.6) marred Formerly 43 (18•5) 7,0 (15.1) married Schooling level Illiterate 66 (28.4) 112 (24.1_) Grade school 141 (60•8) 303 (65.3) High school 21 (9.1) 36 (7•8) College 4 (1.7) 12 (2.6) Household income 0-37 61 (26•3) 95 (20.5)~ (in US$/month) 38-72 46 (19.8) 88 (19.0) 73-128 32 (13•8) 92 (19.8) 129-233 54 (23.3) 94 (20.3) 234+ 34 (14•7) 88 (19.0) Unknown 5 (2.2) 7 (1•5) Geographical North/North-east 48 (20.7) 102 (22.0) region South-east 81 (34.9) 182 (39•2) of birth South 71 (30.6) 117 (25•2) Central-West 27 (11.6) 50 (10.8) Other country 5 (2•2) 13 (2.8) Residence in rural No 55 (23•7) 96 (20•7) area >~5 years Yes 177 (76•3) 368 (79•3) ~Percent values may not add up to 100 because of round-off error or exclusion of subjects with missing information.

• 994 ~" FRANCO ET ALe TABLE II - ORAL CANCER CRUDE AND ADJUSTED (TOBACCO AND ALCOHOL CONSUMPTION) RR ESTIMATES ACCORDING TO EMPLOYMENT IN SELECTED OCCUPATIONAL SETTINOS AND TO SOME ENVIRONMENTAL EXPOSURES Exposure Cases/control Crude analysis Adjusted analysis Never Ever RR 95% CI RR 95,% Ci (A) Employment in specific occupational settings: Textile 220/423 12/40 0.6 0.3-1.1 0.5 0.3-I. 1 Wood 205/423 27/39 1.5 0.9-2.5 1.2 0.7-2.2 Paper 225/458 • 6/5 2.4 0.7-7.9 2.1 0.6-7.3 Mining 222/436 10/27 0.7 0.3-1.5 0.8 0.3-I .8 Leather 225/450 7/13 1.1 0.4-2.7 1.3 0.4-3.7 Metal 209/393 23/70 0.6 0.3-1.0 0.6 0.3-1.0 Sugar/alcohol 225/452 7/11 1.3 0.5-3.3 0.9 0.3-2.5 Rubber 225/452 7/11 1.3 0.5-3.6 1.5 0.5-4.8 (B) Environmental or household exposures: Pesticides 134/267 98/197 1.0 0.7-1.4 1.2 0.8-1.7 Wood stove 134/343 98/121 2.4 1.6-3.5 2.5 1.6-3.9 Asbestos 222/437 " 10/27 0.7 0.3-1.5 0.5 0.2-1.2 ~By conditional logistic regression (malching variables: age, sex, study site, and admission period). difference was in regard to use of wood stove. This exposure was considerably more often associated with cancer of the tongue (RR = 6.5, 95%CI: 2.8-15.0) than with other cancers of the oral cavity (RR = 1.4, 95%CI: 0.8-2.4) (adjusted for smoking and alcohol). Dietary variables The interview provided consumption frequency information with respect to 20 dietary items. Some items could be grouped into specific categories. Table III shows the association be- tween risk of cancer of the oral cavity and selected dietary items. Three levels of average past consumption are presented for regular-sized servings in each category. The only signifi- cant reductions in risk were associated with more frequent consumption of carotene-rich foods (carrots, pumpkins, and papaya) and citric fruits. Highly significant trends in dose-risk relationships were observed in the univariate analyses for these 2 variables. However, adjustment for smoking and alcohol reduced the strength of the associations (p = 0.0639 for car- otene, p = 0.0303 for citric fruits). Consumption of smoked meat was not associated with risk of oral cancer. On the oth6r hand, a substantial increase in risk was observed in categories of frequent consumption of char- coal-grilled meat (more than 4 times/week). This association persisted after adjustment for smoking and alcohol consump- tion. Similar, but less pronounced, higher risk levels were associated with increased consumption of manioc, a common staple food item in many rural areas in Brazil. Significant trends in dose-risk relationship were seen with consumption of pickled vegetables and pepper in univariate analyses. Further adjustment for tobacco and alcohol consump- tion reduced both the strength of these associations and the magnitude of the level-specific risks. RR estimates for all dietary items did not differ according to anatomic site. TABLE III - ORAL CANCER CRUDE AND ADJUSTED (TOBACCO AND ALCOHOL CONSUMPTION) RR ESTIMATES1 ACCORDING TO THE FREQUENCY OF CONSUMPT/ON OF SELECTED FOOD GROUPS - Cases/ Crude analysis Adjusted analysis Food group Frequency controls RR 95,% CI RR 95% CI Carotene-rich <l/me 62/101 1.0 (ref)2 1.0 (ref) 1/mo-3/wk 154/310 0.7 0.5-1.1 0.8 0.5-1.4 > =4/wk 16/53 0.4 0.2-0.8 0.4 0.2-1.0 Citric fruits <l/me 77/92 1.0 (ret) 1.0 (ref) 1/mo-3/wk 981233 0.4 0.3-0.7 0.5 0.3-0.8 > =4/wk 57/137 0.4 0.2-0.6 0.5 0.3-0.9 Green vegetables <l/me 41/58 1.0 (ref) 1.0 (ret) 1/mo-3/wk 144/299 0.7 0.4---1.0 0.8 0.5-I .4 > = 4/wk 47/104 0.6 0.4-1.1 0.7 0.4-1.4 Manioc (cassava) <I/me 41/106 1.0 (ref) 1.0 (ref) 1/mo-3/wk 118/245 1.3 0.8-2.0 1.3 0.8-2.2 >=4/wk 72/112 1.7 1.0-2.7 1.8 1.1-3.1 Smoked meat <l/me 151/327 1.0 (ref) 1.0 (ref) l/mo-3/wk 59/105 1.3 0.8-1.9 1.0 0.6--1.6 > =4/wk 13/19 1.6 0.7-3.3 1.1 0.5-2.6 Charcoal-grilled <l/me 148/312 1.0 (ref) 1.0 (ref) meat 1/mo-3/wk 63/131 1.2 0.8-1.7 1.2 0.7-1.8 >=4/wk 18/17 3.6 1.4-9.0 5.3 1.9-15.0 Pickled vegetables <l/me 160/351 1.0 (ref) 1.0 (ret) 1/mo-3/wk 31/62 1.2 0.7-1.9 1.0 0.6-1.7 > =4/wk 35/40 2.1 1.2-3.5 1.7 0.9-3.1 Pepper <l/me 83/211 1.0 (ref) 1.0 (ref) 1/mo-3/wk 52/84 1.7 1.1-2.6 1.5 0.9-2.5 > =4/wk 92/158 1.6 1.1-2.3 1.3 0.9-2.0 ~By conditional logistic regression (matching variables: age, sex, study site, and admission pedod).-2Reference category.

• . r..~, ~ RISK FACTORS FOR bRAL CANCER ' . ...... TABLE IV - O'~kL CANCER CRUDE AND ADJUSTED (TOBACCO AND ALCOHOL CONSUMPTION) RR ESTIMATES ACCORDING TO THE FREQUENCY OF CONSUMPTION OF NON-ALCOHOLIC BEVERAGES Ca~s/ - Crude analysis Adjusted analysis Beverage Frequency controls RR 95% CI RR 95% CI Coffee (cups/day) Tea (cups/month) "Chimarr~o" (matr) (cups/month) < = 1 43/113 1.0 (ref)2 1.0 (re0 2-5 113/239 1.3 0.8-1.9 1.1 0.7-1.8 > = 6 76/l 12 1.9 1.2-3.2 1.5 0.9-2.6 Trend test (p-valug): 0.0103 0.1385 < 1 114/221 1.0 (ref) 1.0 (ref) 1-30 88/187 0.9 _..0.6-1.3 0.9 0.6-1.3 >30 30/5-6 1.0 0.6-1.7 1.3 0.7-2.3 Trend test (p-value): 0.6368 0.4439 <I 180/387 1.0 (ref) 1.0 (re0 1-30 24/39 1.6 0.9-2.9 1.6 ' 0.8-3.3 >30 28/38 2.0 1.1-3.6 1.6 0.8-3.3 - - Trend test (p-value): 0.0720 0.2001 tBy conditional logistic regression (matching Variables: age, sex, study site, and admission pefiod).-~Reference ~ategory. Consumption of non-alcoholic beverages Positive associations with risk of cancer of the oral cavity were seen with consumption of coffee and "chimarr~o"--a type of mat6 infusion heavily consumed in the Southern states 0f Brazil. However, these associations were partly due to con- founding by smoking and alcohol, as further adjustment by these variables reduced the magnitude of the level-specific RR estimates (Table IV). A possible effect of beverage tempera- ture on the risk of oral cancer was investigated separately among drinkers of coffee and "chimarr~,o". There was no indication of such an effect as judged by RR estimates c~m- paring drinkersof '.'buming hot" beverages vs. lower temper- atures. Analysis of site-specific RRs for n0n-alcoholic beverages indicated that the positive associations observed for coffee and "chimarr~o" were mostly in regard to cancer of the tongue (ICD-9 141). Smoking and alcohol-adjusted RR estimates for heavy vs. low consumption of coffee were 2.0 (95%CI:I 0.8- 4.7) and 1.2 (95%CI: 0.6-2.6), for cancer of the tongue arid other parts, respectively. Likewise, adjusted tongue cancer RRs for medium vs. low and for high vs. low consumptions of "chimarr~o" were 2.5 (95%CI: 0.8-7.5) and 3.7 (95%CI: 1:3-10.7), respectively. Equivalent level-specific RRs for car- clnornas of other oral sites in regard to "chimarr~o" were close t~ unity~l.3 for medium vs. low and 0.7 for high vs. low Consumptions. Oral hygiene and other health conditions Poor dentition and infrequent tooth brushing were more Commonly reported by cases than by controls (Table V). Con- trois also reported more frequent visits to the dentist than cases, but to a less significant degree. No significant association was seen between disease status and use of dentures. However, the alcohol and smoking-adjusted RR estimates for this variable were 1.0 for cancer of the tongue and 0.7 for cancers of other parts of the mouth. The latter estimate departed from the null value in the directionof a negative association, Further mutual adjustment using these variables in addition to smoking and alcohol substantially changed the risk estimates for all oral hygiene factors towards the null value. The sole exception was tooth-brushing frequency. The RR associated with infrequent brushing was more than twice that of subjects reporting daily brushing. A number of questions sought information on personal and family history of infectious and chronic diseases. No evidence of association was seen between disease risk and these vari- ables. The proportions of persons reporting tonsillectomy were bomparable between cases and controls. Tobacco smoking Table VI shows the alcohol-adjusted RR estimates by ana- tomical site for various categories of smoking behavior. To- bacco consumption was, by any measure, the most important factor for prediction of the risk of cancer of the oral cavity. RR estimates were more pronounced for cancer of the tongue than for cancer of other oral sites. Confidence intervals were wide, reflecting the low number of cases who never smoked any type of tobacco (10 subjects, 4.3%). Likewise, there were 99 con- trois who never smoked (21.3%). The reference category for calculation of RRs by levels of cumulative exposure also in- cluded 1 case and 9 controls whosmoked less than '1 pack-year TABLE V - ORAL CANCER CRUDE AND ADJUSTED (TOBACCO AND ALCOHOL CONSUMPTION) RR ESTIMATES~ FOR CANCER OF THE MOUTH ACCORDING I'O SELECTED ORAL HYGIENE CHARACTERISTICS Cases/ Cmde analysis Adjuste~d analysis Variable Category controls RR 95% CI RR 95% CI Use of dentures No 112/194 1.0 (ref)2 1.0 (ref) Yes 120/270 0.8 0.6-1.1 0.9 0.6-1.2 Broken teeth No 113/267 1.0 (ref) 1.0 (ret) Yes 1131195 1.4 1.0-I .9 1.3 0.9~1.8 Oral health care Never 20/20 1.0 (ref) 1.0 (ret) <l/year 206/430 0.5 0.2-0.9 0.6 0.3-1.3 > = 1/year 5/12 0.4 0.1-1.4 0.6 0. I-2.3 Tooth-brushing Daily 163/388 1.0 (ref) 1.0 (ref) . Infrequent 69/73 2.6 1.7-4.0 2.3 1.4-3.7 tBy conditional logistic regression (matching variables: age, sex, ~study site, and admission period).-~Reference category. L r,0 o

996 '-~ FRANCO ET AL. _ ~f'TABLE .VI- ORAL CANCER RR ESTIMATESi ACCORDING TO TOBACCO CONSUMPTION VARIABLES. RESULTS- : ADJUSTED BY ALCOHOL CONSUMPTION AND STRATIFIED BY SITE . Tobacco Tongue (141) ,Other (143;5) Mouth (14i + 143-5) consumption ' _ ~ _,~ RR 95% CI RR 95% CI RR 95% CI Never smoker 1",0 (ref)= _ 1.0 (rd) 1.0 (ref) Other than cigarettes 18.4 2.2-151 6.8 1.9-23.7 9.1 3.3-25.6 Mixed types 15.4 1.9-122 4.0 1.2-13.1 6.3 2.4-17.1 Cigarettes only 16.4 2.1-126 3.8 1.2-12.0 6.3 2.4-16.7 Filter 16.0 2.1-122 3.8 i.2-11.9 6.3 2.4-16.3 Non-filter 16.3 2.0-134 4.5 1.2-17.3 6.9 2.4-20.0 Cigars Never 1614 2,2-124 4.4 1.4-13,4 6.8 2.6-17,7 Ever 19.4 1.5-249 2.1 0.3-17.0 5.5 1.2-24.8 Pipe Never 16.1 2.1-122 3.9 1.3-12.0 6.4 2.5-16.6 Ever 27.5 3.0-256 11.2 2.6-48.3 13.9 4.4-44.2 Hand-rolled Never 16.3 2.1-125 4.2 1.3-13.1 6.6 2.5-17.4 Ever 16.9 2.2-131 4.5 1.4-14.3 7.0 2.7-18.7 Lifetime consumption (pack-years): <1 1.0 (ref) 1.0 (ref) 1.0 (ref) 1-25 15.2 1.9-120 4.4 1.2-16.4 7.1 2.4-21.0 26-50 _ 24.7 3.1-196 4.7 1.2-18.1 9.5 3.1-28.6 51-100 27.4 3.3-227 7.3 1.9-27.8 12.5 4.1-38.3 >100 28.0 3.3-239 9.4 2.2-39.5 14.8 4.7-47.3 tBy conditional logistic regression (matching variables: age, sex, study site, and admission period) using separate model statemems.-2Reference category for all models. in a lifetime. Strong and highly significant dose-response-like relationships were found for this combined index of cumulative exposure for both anatomical sites. Differences in magnitude of risk due to the various types of tobacco were generally not substantial, regardless of the anatomical site. There was, how- ever, a higher risk associated with pipe-smoking when this habit was contrasted with the other smoking.behaviors. Such a contrast was particularly noticeable for cancer of other parts of the mouth (ICD-9 143-5). There were sufficient smokers of industrially produced and hand-rolled cigarettes in the study to allow evaluation of the effects of stopping smoking on the risk of oral cancer. Table VIi~ shows the risk estimates for current smokers and for 2 categories of former smokers contrasted with those of never- smokers. Results are also adjusted for other smoking behav- iors. The magnitude of risk for persons who smoked industri- ally produced cigarettes decreased to the same level as that of never smokers 10 years after cessation. Patterns of risk reduc- tion after cessation (industrially produced cigarettes) for the 2 anatomical sites were similar. On the other hand, although the reduction in risk subsequent to stopping smoking for smokers of hand-rolled cigarettes was also considerable, risk levels reached after 10 years were still higher than those of never smokers (Table VII). Use of smokeless tobacco, either as snuff dipping or tobacco chewing, was not associated with risk of oral cancer. Nine cases and 13 controls reported using tobacco in this form. RR estimates by matched analyses for this variable were indepen- dent of smoking or drinking status, sex, or anatomical site. Alcohol drinking Table VIII shows the smoking-adjusted RR estimates ac- cording to the cumulative consumption of each of the alcoholic beverage types probed in our study. Consumption categories were chosen to enhance comparability among beverage types. Excess risks were observed with increased consumption of wine and "cacha~a", a distilled sugar-cane .spirit. Trends in the association with the latter beverage were highly significant for both anatomical sites. However, the risk of oral cancer imputed as consequent to wine intake was practically restricted to cancer of the tongue. In addition, level-equivalent RRs were appreciably higher for wine than for "cacha~a" when consid- ering tongue lesions. Notwithstanding the effect of wine con- sumption on the risk of cancer of the tongue, RR estimates were generally higher for cancer of other parts of the mouth (ICD-9 143-145). The excess risk due to alcohol seemed to reach a plateau at the cumulative level of 1,000 kg (Table VIII). Joint effects of tobacco and alcohol Analysis of the combined effects of smoking and alcohol on the risk of oral cancer using 5 exposure cat6gories for each TABLE VII - EFFECT OF SMOKING cESSATION ON THE RISK OF ORAL CANCER STRATIFIED BY SITE. ALCOHOL-ADJUSTED RR ESTIMATES~ ACCORDING TO TYPE OF CIGARE'I'TES Other (143-5) . Mo_uth (!41_÷ 143-5) • ~ g._ .. RR : 95% CI RR 95% CI .. ~ Tobacco Tongu~ (.141). :- consumption .... _ :_~ ~ ~ ~ 95%CI Never smoker 1.0 (ref)~ 1.0 Industrial brand cigarettes: Current smokers 23.3 2.9-186 5.5 Ex-smokers, 1-10 yrs 6.3 0.7-58.7 1.8 Ex-smokers, > 10 yrs 1.2 0.1-2 !. 1 0.4 Other than cigarettes 16.5 1.9-141 6.2 Hand-rolled cigarettes: Current smokers 32.2 4.0-262 10.0 Ex-smokers, 1-10 yrs 17.5 1.9-158 2.2 Ex-smokers, >10 yrs 4.1 0.4-38.6 1.8 Other than hand-rolled 14.6 1.9-112 3.8 (reO 1.0 (ref) 1,7518.6 9.3 3.4-25.9 0.4-7.7 2.9 0.9-9.2 0.1-2.9 0.6 0.1-2.8 1.7-22.8 8.5 2.9-24.9 2.9-34.6. 14.4 5.2-40.0 0.6-8.8 4.9 1.6-14.8 0.5-7.1 2.3 0.8--7.1 1.2-11.9 6.0 2.3-15.9 ~By conditional logistic regression (matching variables: age, sex, study site, and admission period).-~Reference category.

L RR epen - ire. s ac- holic 5s in leant racer icted ~vere ~sid- con - lates outh dto able 1 on ~ach .! .) ! 1 ruSK F%CTOR~ ~)R ORAL CANCER TABLE ~ - ORaL CANCER RR ESTIMATF~_ ' BY CATEGORIES OF ALCOHOLIC BEVERAGE CONSUMPTION. " TOBACCO CONSUMPTION-ADJUSTED RESULTS STRATIFIED BY SITE Lifetime alcohol consumptj.on Tongue (141) Other (143-5) Mouth (141 + 143-5) (kg) RR .. 95% CI RR 95% CI RR 95% CI Beer <1 1.0 1-103 0.9 >100 1.3 Wine < 1 1.0 1-100 2.5 >100 7.9 Hard liquor <1 1.0 1-103 0.8 >103 1~2 "Cacha~a" <1 1.0 1-103 1.1 ,. 101-400 1.4 401-1,000 2.7 1,001-2,000 2.8 >2,000 17.5 All types -<1 1.0 1-1130 2.4 101-400 3.3 401-1,000 5.6 1,001-2,0.00 5.9 >2,000 9.7 (ref)~ 0.5-1.7 0.4-4.8 (reO 1.2-5.2 1.0-62.1 (re0 0.4-1.8 0.4-3.1 (re0 0.4-3.4 0.4-4.6 1.0-7.2 1,0-7.3 1.7-180 (re0 0.6-9.3 0.7-14.9 1.4-21.8 1.5-23.7 1.5--61.1 1.0 (reO I.O (reO 1.0 0.5-1.8 0.9 0.6-1.4 1.1 0.4-3.2 1,2 0.5-2.6 1.0 (reD 1,0 (ret) 1.0 0.6-1.8 1.4 0.9-2.2 0.4 0.1-2.3 1.6 0.6-4.2 1.0 (reO 1,0 (ref) 1.2 0.6-2.4 1.0 0.6-1.7 2.2 0.9-5.4 1.7 0.9-3.2 1.o (red 1.0 (ref) 1.4 0.5-3.9 1.2 0.6-2.7 2.3 0.8--6.1 1.9 0.9-4.0 3.3 1.2-9.5 3.1 1.5-6.3 5.9 2.2-15.8 4.2 2.1-8.3 5.1 1.2-21.0 6.7 2.2-20.3 1.0 (reO 1 .o (reO 3.1 o.7-13.2 2.7 1.0-7.2 4.0 1.0-16.5 3.5 1.3-9.8 9.5 2.2-41.1 7.1 2.6-19.5 13.7 3.2-58.1 9.2 3.3-25.1 9.0 1.6-49.2 8.5 2.5-29.4 conditional logistic regression (matching variables: age, sex, study site, and admission period).-ZReference category. 997 variable is presented in Table IX. The reference category (RR = 1) for the computation of RRs contained 5 cases and 54 controls. There was no statistical evidence of effect modifica- tion as judged by the contribution to goodness of fit of 16 cross-product terms in addition to a baseline model containing 8 main effect terms for smoking and alcohol (p = 0.5971). However, when numbers of pack-years and kilograms of alco- hol were fitted as single ordinal variables (5 levels), the con- tfibution of a cross-product term on the 2-term baseline model this model as compared to the baseline model assuming inde- pendence of effects are represented in Figure 1. It is possible to observe that the effects assuming interaction (Fig. Ib) are higher than those fitted without interaction assumed (Fig. la). Multivariate analysis A number of multifactorial models containing variables as- sociated with oral cancer risk were fitted by conditional logistic regression. These models always contained ethnic origin, rural was significant (p = 0.0018). Because of the small numbers of cases in the lowes~t con- Sumption categories, effect modification was also analyzed af- ter reclassification of pack-years and alcohol into 3-level vari- ,ab)es (Fig. 1). The categories chosen were: 1, < =25; 2, 26- i u0; and 3, >100 pack-years for smoking; and 1, <= 100; 2, !01-I,000; and 3, >1,000 kg for alcohol. Interaction as Judged by the fitting of 4 cross-product terms in addition to the b.aseline model containing 4 main effect terms was marginally slgnificant at the 10% level (p = 0.1172). Risk magnitudes for -- . =residence, a cross-product term for rural residence and sex (a matching variable), income, schooling level, smoking, alco- hol, and interaction terms for the last 2 variables. Addition to this baseline combination of control variables of all other fac- tors showing univariate associations with oral cancer risk did not provide materially different parameter estimates for these factors as compared to the tobacco- and alcohol-adjusted re- sults. In addition, interaction between study factors and tobacco and alcohol consumption was investigated in models contain- TABLE IX - JOINT EFFECTS OF TOBACCO AND ALCOHOL CONSUMPTION ON THE RISK OF ORAL cANCER. RR ESTIMATES1 ACCORDING TO CATEGORIES OF COMBINED CUMULATIVE EXPOSURE Alcohol consumption RR by tobacco consumption (Pack-years) RR for alcohol (kg) <1. ~ 1-25 26-50 51-100 > 100 Crude Adjusted 10.7 73.8 15.2 1.0 1.0 (3/14) (2/2) : (I/5) (t5/93) 25.5 45.3 220.0 3~9 2.7 (6/20) (4/11) (6/4) (2_7/94) 33.0 65.1 93.5 6.1 3.5 (6/20) (8/17) (6/9) (28/81) 85.3 111.8-67.2 13.0 -7.1 (19/24) (15/17) (5/11) (61/86) 116.2 135.2 141.6 18.4 8.8 (28/27) (32/29) (30/28) (100/106) 9.7 15.8 22.9 28.2 (49/114) (62/105) (61/76) (48/57) 7.1 9.2 12.0 13.9 <1 1.0z 2.4 (5/54)3 (4/18) 1-100 2.4 14.2 (4/26) (7/33) 101-400 0.0 31.3 (0/I0) (8/25) 401-1,030 ...... 6.3 86.9 (1/9) (21/25) >I,000 23.1 103.4 (1/9) (9/13) RR for tobacco consumption: Crude 1.0 (11/108) Adjusted 1.0 IBy conditional logistic regression (matching variables: age, sex, study site, and admission 2 period).- Reference category. - Number of cases/number of controls in stratum~

998 .. • FRANCO ET AL. A :B RR RR 3. 2 ALCOHOL FIaUR~ 1 - RR estimates for oral cancer according to categories of joint tobacco and alcohol exposure. (a) Assuming independence of effects; (b) assuming effect modification. Levels of cumulative tobacco exposure: 1, < = 25; 2, 26-100; 3, > 10t3 pack-years. Levels of lifetime alcohol consumption in kg: 1: <= 100, 2: 101-I,000, 3: >1,000. - ing each factor, terms for tobacco and alcohol, and cross- product terms between the factor considered and the latter vari- ables. No evidence of effect modification was obtained by using this approach. DISCUSSION The distributions of risk factors for the development of oral cancer-are considerably different across populations. Habits such as tobacco smoking and alcohol drinking, the prime de- terminants for this group of diseases, vary markedly_as a func- tion of economic, cultural, demographic and geographical characteristics. Furthermore, the role of dietary practices and possible occupational exposures is no less dependent on cul- tural and economic population attributes than smoking and drinking. Our present knowledge about the relative importance of these risk factors has been derived mostly from investiga- tions in Northern hemisphere populations in North America, Europe, and Asia. Apart from an earlier study on a Hispanic population in Puerto Rico (Martinez, 1969) there have not been any epidemiological studies of oral cancer in Latin-American countries. Brazil is the largest country in South America, with a population of highly diversified ethnic, economic and cul- tural characteristics. The high rates of oral cancer in some regions of the country prompted us to conduct this investiga- tion. The main determinants of oral cancer risk in Brazil, as shown by the present study, are tobacco smoking and alcohol drinking. The oral cancer risk due to smoking seems to be abnormally higher in Brazil, as judged by the magnitude of the RRs due to different categories of tobacco consumption in this study compared to others (Wynder and Stellman, 1977; El- wood et al., 1984; Blot et al., 1988). In fact, the population- based magnitude of risk due to tobacco smoking may be even higher in the populations studied since the present estimates were based on a standardized comparison with hospital con- trois. We made no attempt to exclude tobacco-related diseases from the roster of causes of hospitalization among controls. This decision was taken on the basis of the difficulty of eval- uating etiology as being due to smoking individually and be- cause of the impracticability of standardizing inclusion or ex- clusion criteria in a multi-center study of this nature~ Differences in risk as a function of type of tobacco were most notable with respect to pipe smoking. Consumption of black tobacco in the form of hand-rolled cigarettes did not seem to contribute a markedly higher risk of oral cancer than in industrialized cigarettes. Although RR estimates for hand- rolled cigarettes were generally higher than regular filter or non-filter cigarettes, the differences observed were not of the same magnitude as those seen in other studies of upper respi- ratory and digestive system cancers (De Stefani et al., 1987; Tuyns et al., 1988). A striking reduction in risk was observed among ever- smokers quitting the habit. Such a reduction brought risk level~ to those of never smokers after 10 years for smokers of indus- trial brand cigarettes. A comparable effect has been recorded in other studies (Wynder and Stellman, 1977; Blot et al., 1988). The association with alcohol also reflected differences in beverage type since level-specific categories standardized by amount of cumulative exposure to ethanol differed in risk mag- nitude and site-specificity. Combined alcohol RR estimates were generally higher for cancers of the gum, floor and other parts of the mouth than for cancer of the tongue. Similar results were obtained by Blot et al. (1988). There were sufficient numbers of light smokers in the study to permit evaluation of risk due to alcohol under little influence of smoking. Alcohol drinking was an explanatory variable for the risk of oral cancer even among individuals who were ex- posed to less than 1 pack-year of tobacco smoking. It is note- worthy that the evaluation of combined effects of tobacco and alcohol only proved informative after using matched analysis by conditional logistic regression. When the matching factors used in the study design (age, sex, study site and admission period) were deliberately neglected during the analysis by un- conditional maximum likelihood estimation, very little resolu- tion was attained in assessing effect modification (data not shown). For instance, the reported risk for the heaviest drink- ing category (> 1,000 kg) was estimated as being 23 times that of hght (< 1 kg) dnnkers among light smokers (< 1 pack-ye ). The confidence interval for this estimate excluded the null value. By unmatched analysis, however, the same comparison yielded an RR of 1.2, a value which was not significantly different from unity. Although individual indices of food nutrients could not be calculated in our study, it was possible to measure an important protective effect of carotene-containing vegetables and citric fruits from the diet. It is likely that the reduction in risk con- ferred by these food items is, in fact, a reflection of their contents of vitamins A and C. The measurable risk reduction

ects; ~ ohol l in ad- ~ or the ~ pi- g7; in by tg- .'es let Its ~4y ce "or e- )n ot at ). Y :~" ~" RIS~ FACTORS FO~'ORAL CANCER 999 ~e to these. ,,dietary items-~,~S ,n, ot d~ue to confounding" by- a hypothetical oral cancer-prone patient profile. Multivariate ~justment by ethnic, economic, demographic and cultural in- dicat0rs, in addition to the essential confounders--smoking ~xt drinking--did not change risk estimates for these food items.On the other hand, consumption of green vegetables, although protective upon crude analysis, failed to exhibit a risk-reduction effect after controlling for sm6king and alcohol. The above findings are in agreement with a recent study on diet and oral cancer risk (McLaughlin et al., 1988). : Positive associations with risk were detected for consump- tion of charcoal-grilled, but not smoked meat, and manioc (cassava). Charcoal-grilling and smoke-curing represent cook- ing methods potentially affecting the content of polycyclic ar- omatic hydrocarbons in food (Committee on Diet, Nutrition, ~_d Cancer, 1982). However, these procedures have not been found to affect oral and pharyngeal cancer risk (McLaughlin et d., 1988). Consumption of pickled vegetables and spicy foods was not associated with oral cancer risk since controlling for tobacco and alcohol showed that the RR estimates for these variables were mostly due to confounding. Other studies of oral and pharyngeal cancer have also failed to find an associ- ~on with consumption of spicy or pickled foods (Winn et al., 1984; McLaughlin et al., 1988). Mat6 ("chimarrfio") consumption was associated with risk of oral cancer, particularly of cancer of the tongue. Previous studies imputing cancer risk to mat6 consumption have dealt exclusively with esophageal and laryngeal cancers (De Stefani aal., 1987; Vassallo et al., 1985; Victora et al., 1987). Our results point to an effect other than temperature for the in- creased cancer risk observed for "chimarrfio" consumption. A recent study has also failed to detect an increase in risk asso- ciated with temperature during consumption of non-alcoholic beverages (Winn et al., 1984). It is possible, however, that our evaluation of the role of this variable may be biased because of severe misclassification. Further studies are needed to eluci- date the association with mat6 consumption, perhaps by in- eluding validation of beverage temperature reported in the in- tel'views. Oral hygiene characteristics were strongly interrelated among our study subjects. The multivariate assessment of the role of these variables indicated that only tooth-brushing fre- quency behaved as a genuine marker of oral cancer risk. An earlier study had found an important association between oral cancer risk and poor dentition, as judged by a standardized oral examination (Graham et al., 1977). However, these findings ~ not directly comnarable to ours because of fundamental differences in "effect~estimation between studies (crude, un- matched vs. fitted, multivariate odds ratios). ~e present study design was not adequate to investigate in e.tail possible occupational risk factors in regard to oral cancer sk. It was possible to learn, however, that, if any epidemio- logical importance is to be attribui~d tO these fa6tors in Brazil, it is unlikely that it will be a major one in population-based terms. Nevertheless, a striking finding from the questioning of .sporadic environmental exposures in our study was the strong association with use of a wood stove for cooking and heating. The association was most evident with cancer of the tongue and totally independent of several putative confounding factors such as smoking, drinking, residence, and diet. This finding has potential implications in terms of prevention because of the wide distribution of this characteristic in rural areas in Brazil. Approximately one third of the controls in our study reported currently living in a house with a wood stove. It is plausible to assume that indoor burning of coal or wood and the possibly inadequate ventilation in these houses could be responsible for severe domestic air pollution with subsequent increased cancer risk. A similar association has been incriminated in regard to lung cancer risk among women in China (Mumford et al., 1987). Unfortunately, the sample size of our study did not permit proper investigation of the significance of a sex differ- ential effect in regard to use of wood stove as associated with oral cancer risk. In conclusion, our study has shown that the relative impor- tance of determinants of oral cancer in high-risk South- American populations may be different from that seen in coun- tries of the Northern hemisphere. Successful prevention pro- grams in Brazil should give even greater emphasis to tobacco smoking and alcoholic beverage consumption. Regionally spe- cific lifestyle and behavioral characteristics such as "chimar- rfio" (mat6) consumption and indoor use of wood stoves for cooking and heating may be responsible for a substantial pro- portion of oral cancer incident cases. Further studies should be aimed at the elucidation of the public health importance of these factors in Latin American populations. ACKNOWLEDGEMENTS The authors are indebted to all the other participants in the • Ludwig Institute for Cancer Research's Upper Respiratory and Digestive System Cancer (LICR URDS) Study Group. The following clinical investigators participated in the Group's clinical committee: Drs. M.B. Carvalho, A. Rapoport, J. An- drade-Sobrinho, G. Ramos, J.L. Kanda, J.F. Gois, J.S. Cha- gas, and G.A. Teixeira. The Group Pathology Committee was composed of: Drs. H. Torloni (co-ordinator), W.T. Vieira, L.A. Sampaio and V.M. Cardoso. The following persons were responsible for data acquisition and management: Ms. M.E. Silva (field supervisor), Ms. R.N. Pereira and Mr. N. Campos- Filho (computer data processing), Ms. L. Fanes, Mr. V.N. Souza and Ms. M.S. Morals (interviewers). The LICR URDS Cancer Study Group's Planning Committee was as follows: Dr. E.L. Franco (study chairman), Dr. L.P. Kowalski, Dr. B.V. Oliveira and Dr. M.P. Curado (principal investigators), and Dr. A.S. Fava (project manager). REFERENCES i.~.I-oL W.J., McLAUGHLIN, J.K., WINN, D.M., AUSTIN, D.F., GREEN- 13, R.S., PRESTON-MARTIN, S., BERNSTEIN, L., SCHOENBERG, J.B., MHAGEN, A. and FRAUME~I, JR., J.F., Smoking and drinking in re- latona to oral and pharyngeal cancer. CancerRes., 48, 3282-3287 (1988). !9oI~'SLoW,N E and DAY, N E Statistical methods in cancer research, 1. I, The ~lysis of case~o~rol studies, IARC Scientific Publication ~' PP. 248--279, IARC, Lyon (1980). ~Uel A~Os'FILnO, N and FRANCO, E L., A microcomputer program for ~ple logistic regression by unconditional and conditional maximum _ aood methods. Amer. J. Epidemiol., 129 (in press) (1989). .UOMMITTEE ON DIET NUTRITION, AND CANCER. Diet, nutrition, and can- ~r, p. 329, Nationai Academy Press, Washington (1982). ~E 8TEFANI, E, CORREA, P, OREGGIA, F., LEIVA, J., RIVERO, S., ~.Imh.UbEZ, G i DENEO-PELL~GR1N1, H, ZAVALA, D. and FONTHAM, E., Risk factors fo~ laryngeal cancer Canc'~r 60, 3087-3091 (1987). I~I-WOoD, J.M., PEARSON, J.C.G., SKIPPEN, D.H. and JACKSON, S.M., Alcohol, smoking, social and occupational factors in the aetiology of can- cer of the oral cavity, pharynx and larynx. Int. J. Cancer, 34, 603-612 (1984). GRAHAM, S., DAYAL, H., ROHRER, T., SWANSON, M., SULTZ, H., SHEDD, D., and FISCHMAN, S., Dentition, diet, tobacco, and alcohol in the _ epidemiology of oral cancer. J. nat. Cancerlnst., ~9, 1611-1618 (1977). MARTINEZ, I., .Factors associated with cancer of the esophagus, mouth, and pharynx in Puerto Rico. J. nat. Cancerlnst., 42, I069-I094 (1969). MCLAUGHLIN, J.K., GRIDLEY, G., BLOCK, G., WINN, D.M., PRESTON- MARTIn, S., SCHOENB~RG, J.B., GREENBERG, R.S., STEMHAGEN, A., AUSTIN, D.F., ERSHOW, A.G., BLOT, W.J. and FRAUMENI, JR., J.F., Dietary factors in oral and pharyngeal cancer. J. nat. Cancer Inst., 80, 1237-1243 (1988). MIRRA, A.P. and FRANCO, E.L. (eds.), Cancer Incidence in Silo Paulo County, Brazil. LICR Cancer Epidemiology Monograph Series, Vol. 1, Ludwig Institute for Cancer Research, Silo Paulo (1985).

1000 ~' FRANCO ET AL. Moss, E. and LEE, W.R., Occurrence of.oral and pharyngeal cancers in textile workers. Brit. J. Indust. Med., 31, 224-232 (1974). Mum, C., WATEmaOUSE, J., MACK, T., POWELL, J. and WHELAN, S., Cancer incidence in five continents, Vol. V, IARC Scientific Publications 88, IARC, Lyon (1987). MUMFORD, .I'.L., HE, X.Z., CHAPMAN) R.S., CAO, S.R., HARMS, D.B., LI, X.M., XIAN, Y.L., JIANG, W.Z., Xu, C.W., CHUANG, J.C., WIL- SON, W.E. and COOKE, M., Lung cancer and indoor air pollution in Xuan Wei, China. Science, 235, 217-220 (1987). TUYNS, A.J., EST~VE, J., RAYMOND, L., BERMNO, F., BENHAMOU, E., BLANCHET, F., BOFFETTA, P., CROSIGNANI, P., DEL MORAL, A., LEH- MANN, W., MERLETTI, F., PI~QUIGNOT, G., RIBOLI, E., SANCHO- GARNIER, H., TEP.~.AClNt, B., Zu)~IPa, A. and ZumFa, L., Cancer of the larynx/hypopharynx, tobacco and alcohol: IARC international case-control study in Turin and Varese (Italy), Zaragoza and Navarra (Spain), Geneva (Switzerland) and C~:Ivados (France). Int. d. Cancer, 41, 483-491 (I988). VA$SALLO) A., CORREA, P., DE STEFANI, E., CENDAN, M.,'ZAVALA, D., CHEN, V., CARZOGLIO) J. and DENEO-PELLEGRINI, H., Esophageal can- cer in Uruguay: a case-control study. J. nat. Cancer Inst., 75, 1005-1009 (1985). VICTORA, C.G., MtJNoz, N., DAY, N.E., BARCELOS, L.B., PECCIN, D.A. and B~GA, N.M., Hot beverages and oesophageal cancer in Southern Brazil: a case-control study. Int. J. Cancer, 39, 710-716 (1987). WINN, D.M., ZIEGLER, R.G., P~CKLE, L.W., GPdDLEY, G., BLOT, W.J. an~t HOOVER, R.N., Diet in the etiology of oral and pharyngeal cancer among women from the Southern United States. Cancer Res., 44, 1216- 1222 (1984). WYNDER, E.L. and S'rELLMAN, S.D., Comparative epiderniology of to- bacco-related cancers. Cancer Res., 37, 4608-4622 (1977). YOUNG, T.B., FORD, C.N. and BRANDENnURG, J.H., An epidemiologic study of oral cancer in a statewide aetwork. Amer. J. Otolaryng., 7, 200-208 (1986). t r t I ( i i 1 a i