Tobacco Documents Online

Cancer of the Larynx: A Case-Control Study1,2.3 J. David Burch, 4 Geoffrey R. Howe, 4 Anthony B. Miller, 4.5 and Robert Semenciw 4.6 ABSTRACTwA case-control study of laryngeal cancer was conducted in southern Ontario between 1977 and 1979 with 204 ~ubjeCtS with newly diagnosed cancer and 204 controls, in- dividually matched by sex, age, and residence. Tobacco products and alcohol showed strong associations with cancer of the larynx for males, with relative risks (RR) for users of cigarettes, cigars or cigarillos, pipes, and alcohol of 6.1, 2.9, 1.6, and 5.2, respec- ~ely. The population attributable risk percent for males using tobacco products and alcohol together was estimated to be 94%. Cigarette smoking was also an important risk factor for females, although the small number of female pairs (20) precluded any meaningful detailed analysis of other possible risk factors. The RR for males for exposure to asbestos after the effects of cigarette smoking were controlled was 2.3, and the effects ~eemed restricted to cigarette smokers. The findings on asbestos were based on small numbers of cases and controls exposed and consequently were subject to large sampling errors. The estimate was consistent, however, with that from other studies and ~upported a causal role for asbestos exposure and cancer of the luynx. The RR for males for exposure to nickel was 0.9.--JNCI 1981; 67:1219-1224. Margaret Hospital in Toronto and the Hamilton Cancer Clinic, which are the only resources for radio- therapy in the study areas. The few cases not treated by radiotherapy were determined from individual ear, nose, and throat physicians. Of 258 patients ascertained, 204 (79%) were interviewed and 54 were not (41 refused, 6 died before the interview could be conducted, and 7 had physicians who withheld permission for inter- views). Control acquisition.mFor each case interviewed, an individually matched neighborhood control was se- lected. Control selection started from the house or apartment four doors to the right of the case and proceeded in a defined and systematic manner until an eligible control agreed to be interviewed. To be eligible, a control had to be of the same sex and similar age (yr of birth, +5 yr) as that of the matching case. Of a total of 315 eligible controls contacted, the first-approached controls (corresponding to 77.5% of the cases), the second-approached controls (corresponding to 11.3% of the cases), and the potential controls approached third .... or further in line (corresponding to 11.2% of the cases) Tobacco and alcohol are well recognized as impor- tant causal factors for laryngeal cancer (1, 2). The possible role of occupational factors has been less dearly defined. Several studies have implicated asbestos as a possible risk factor (3-7), but others have failed to confirm this (1, 8). Exposure to nickel has been implicated as increasing the risk of cancer of the larynx in one study only (9). This paper reports the results of a case-control study 0[ larynx cancer conducted in Ontario between 1977 and 1979. We studied the possible role of asbestos and nickel exposure in an occupational context and quan- ~fied the effects of tobacco and alcohol and their interaction. The quantitative relationships of these two [actors must be defined while asbestos and nickel are ~'aluated as risk factors to take into account the POssible confounding or modifying effects of tobacco and alcohol. MATERIALS AND METHODS .Case acquisition.--The potential case series con- s~Sted of all subjects with histologically confirmed ~qrcinoma of the larynx newly diagnosed from March s77 through July 1979 These patients were resident at ~e time of diagnosi~ in the Toronto, Hamilton, .~tadbury, and North Bay areas of Ontario. Cases from '~dbury and North Bay were included due to the ~e, sence there of large nickel mining and processing '~%stries Cases were ascertained through the Princess agreed to be interviewed. We minimized the problem of "not-at-home" controls by seeking a control at the same time of day that the corresponding case would normally be at home. Interview and questionnaire.mAll subjects were in- terviewed in their own home by a specially trained interviewer. Questions were asked on demographic data, smoking, and alcohol use. The format for these questions was detailed in (10). To obtain information on asbestos and nickel exposure, the interviewer took a detailed occupational history, including a record of substances to which a respondent was exposed during ABBREVIATIONS USED: LR = linear logistic regression; PARP = popula- tion attributable risk percent; RR=relative risk(s). ~ Received February 19, 1981; accepted July 22, 1981. ~ Supported by g-rants from the Workmen's Compensation Board of Ontario and from the Nadonal Cancer Institute of Canada. ~ Research procedures were in accord with the ethical standards of the Human Experimentation Committee, University o~ Toronto. * National Cancer Institute of Canada Epidemiology Unit, Faculty of Medicine, McMurrich Building, University of Toronto, Toronto, Canada MSS IA8. ~ Address reprint requests to Dr. Miller. ~ We gratefully acknowledge the assistance of Dr. J. Stopps, the University of Toronto. for classifying the asbestos and nickel ex- posures and of Mrs. P, Morland and Mrs. E. de'Seta for conducting the interviews. 1219 JNCI. VOL. 67, NO. 6, DECEMBER 1981

I~ IIIr .... III IIIII I1 ..................... I[lll~T .......r~l ....... I~1111 I 1220 l~urch, Howe, Miller, and Semenciw the course of any particular job. The respondent was then asked specifically about exposure to asbestos and nickel either in an occupational or nonoccupational context. Questions were also asked regarding the pos- sible exposure of the respondent's husband or wife to either asbestos or nickel. These questions were asked in conjunction with a probe list of substances containing asbestos or nickel. Those subjects who had reported some asbestos or nickel exposure were subsequently reinterviewed in an attempt to define more clearly what such exposure involved. However, not all case-control pairs in which either the case or control had reported asbestos or nickel exposure could be reinterviewed. For the pur- pos~ of the present analysis, therefore, data from the reinterviews were only used for the 40 pairs where it was available from both the case and the corresponding control. Statistical analysis.mRR estimates and tests of sig- nificance are based on the LR model as applies to individually matched case-control studies (11). With this model, use of dummy variables to represent various categories of a risk factor does not impos~ any particular form of dose-response relationship for that variable, whereas the use of continuous variables im- poses a very specific form. The use of categorized variables as grouped, continuous variables, where a particular range of the values ,of the variable is represented by a particular score, represents an inter- mediate situation in that it does impose the log-linear relationship. However, the data are smoothed by the grouping process so that the extreme values have less effect than when continuous variables were used. The categorized approach was used extensively in the presentation of results because it is a convenient single parameter representation of an effect. Grouping of continuous variables was based on approximately equal numbers in user categories. When the scores of 0, I, and 2, for example, represented categories in increasing order of exposure to a variable, the RR for those with score 2 was the square of the risk for those with score I, the RR for those with score 3 was the cube of those with score I, and so on. When only a single dichoto- mous variable wgs included in the model, the RR estimate was reduced to the ratio of discordant pairs (No. of pairs with cases positive for exposure and controls negative/No, of pairs with cases negative for exposure and controls positive) and the corresponding overt ~io-nlflc:~nco to~t v.,~ h,~ed cm tho hinc~rninl distribution. All P-values quoted were one sided, inas- much as the observed effects were, in general, in the direction expected on the basis of a prior hypothesis. RESULTS Case-Control Comparability The mean ages at interviews of 184 mate cases and 184 male controls were 62.4 and 63.1 years, respectively. The corresponding values for the 20 female cases and controls were 56.8 and 57.6 years, respectively. The few female case-control pairs precluded detailed analysis; therefore, the results presented were mostly confined to males. Cases and controls were very similar with respect to education, income, and marital status. On the average, cases took longer to complete the interview than did the controls, the difference being accounted for by longer smoking and drinking histories of the cases and by communication difficulties of those with laryn- gedomies. More than 95% of all interviews was assessed satisfactory by the interviewer, and the exclusion of interviews with cases or controls considered not satis- factory from the analysis produced-virtually identical results with those interviews including all subjects. Smoking Table 1 shows the distribution of male case-control pairs by the average number of cigarettes smoked per day. The RR estimates presented were those obtained by inclusion of a term for lifetime consumption of alcohol in the LR model. Risks for all categories of smokers were significantly elevated, and there was a highly significant dose-response relationship (P = 0.001). An alternative measure of cigarette smoking was lifetime consumption, defined as the product of average frequency and years of use. This measure gave a dose- response relationship of 1.0 for nonsmokers, of 2.0 (1.0, 4.1) for those with lifetime consumption of less than 150,000 cigarettes, of 4.5 (2.1, 9.5) for those who smoked 150,000-299,000 cigarettes, and of 5.4 (2.6, 11.3) for those who used 300,000 cigarettes or more; lifetime consumption was used subsequently as the measure of cigarette smoking because it included both a duration and frequency component. Study subjects were questioned separately regarding the use of nonfilter and filter cigarettes. The RR for the use of filter cigarettes were very similar to those for use of nonfilter cigarettes. Respondents were also asked their extent of inhaling. Table 2 shows RR estimates for lifetime consumption of nonfilter and filter ciga- TABLE 1.--Case--control pairs distribution and RR for cigarette use~.• Average frequency per day for males Cigarettes per day Cigarettes per day for cases ~ for controls 0 <15 15-24 >_25 0 8 i0 18 16 <15 3 6 11 17 15-24 3 11 23 21 >_25 3 3 17 19 RRb 1.0 3.0 3.4 4.5 90% confidence 1.4, 6.3 1.7, 6.8 2.2, 9 2 interval b a Filter and nonfilter combined. ~ Estimates are based on LR model: cigarettes per day (~ dummy variables)+alcohol lifetime consumption (1 categorizeu variable). : jNci. VOL. 67. NO 6. DECEMBER 1981

TABLE 2.-RR° for light and heavy inhalers of nonfilter and filter cigarettes for males RR Cigarette lifetime Nonfilter Filter consumption, in thousands Light L Heavy Light L Heavy inhaler° inhaler inhaler° inhaler 0 1.0 1.0 1.0 1.0 <150 1.1 1.8 2.2 1.7 150-299 1.1 3.4 4.7 2.8 >300 1.2 6.2 10.0 4.8 Coefficient" 0.05 0.61 0.77 0.52 P-value~ 0.432 <0.001 0.002 <0.001 " Estimates based on LR model: nonfilter lifetime consumption, light inhalers only (1 categorized variable)+nonfilter lifetime t0nsumption, heavy inhalers only (1 categorized variable)+filter lifetime consumption, light inhalers only (1 categorized vari- ~le}+filter lifetime consumption, and heavy inhalers only (1 categorized variable)+alcohol lifetime consumption (1 categorized variable). b Includes noninhalers. rettes classified by the usual extent of inhaling. The reduction in risk observed for light inhalers as com- pared to heavy inhalers was restricted to users of nonfilter cigarettes. When ex-cigarette smokers were compared to current smokers, a decrease in risk was seen only for those who had stopped smoking for at least 30 years (e.g., RR=0.59 for those who had smoked between 150,000 and 299,000 cigarettes). However, only 2 cases apd 9 controls had quit 30 years or more ago, and these small numbers made the decrease in risk nonsignificant IP=0.377). RR estimates for cigarette smokers classified by the number of years since they first started smoking are shown in table 3. Only 1 case and 6 controls reported TSBL~ 8.--RR" for cigaretteb smoking by years since smoking started for males" Cigaretr~ lifetime consumption, in thousands RR by yr since smoking startedd 15-34 35-49 _>50 0 1.0 1.0 1.0 <150 2.3 2.2 1.5 150-299 5.2 4.9 2.1 .~300 11.7 10.8 3.0 Coefficient a 0.82 0.79 0.37 ~.___P-valuea <0.001 <0.001 0.006 a Estimates are based on LR model: cigarette lifetime con- ~UraPtion <15 yr (1 categorized variable) l-cigarette lifetime con- suraption, 15-34 yr (1 categorized variable)+cigarette lifetime c~nsUraption, 35-49 yr (1 categorized variable)+cigarette lifetime nsuraption >50 yr (1 categorized variable)+alcohol lifetime C°nsumption'b _. ,:{~- catec, orized~, variable ). , ~"dter and nonfilter combined. ,~ ,Y.ear of diagnosis (interview) minus year first started °,~°~ng_'" To~ few subjects" reported smoking <15 yr to be included in this table. Etiology of Larynx Cancer 1221 to have started smoking less than 15 years before patients were diagnosed for cancer and controls were interviewed. All others who started smoking 15 years or more before showed significantly elevated risks relative to risks of lifetime nonsmokers, though a fall off in risk was seen for those who started smoking 50 years or more before. The difference in risk between those who started smoking 50 years or more before and those who started 15-50 years before was statistically significant (P= 0.007). The few female pairs did not permit the detailed analysis presented for male cigarette smokers. When the 20 female pairs were dichotomized into ever- smoked cigarettes and never-smoked cigarettes, the discordant pair ratio was 12:0 (P<0.001). The number of males who reported the use of either cigars (and/or cigarillos) or pipes was much smaller than the numbers who reported cigarette smoking. Therefore, the analysis was restricted to the dichotomy "ever used" versus "never used." The discordant pair ratio for subjects who used cigars and/or cigarillos was 16:11 (RR= 1.5). However, when this term was included in an LR model with cigarette lifetime consumption (1 categorized variable) and alcohol lifetime consumption (1 categorized variable), the RR estimate became ].5 (P=0.006). This indicated that there was negative confounding between cigarette smoking and cigar and/or cigarillo smoking; i.e., the .two practices were nega- tively correlated. Similarly, for pipes the discordant pair ratio was 15:18 (RR=0.8). However, controlling for cigarette smoking and alcohol resulted in an RR estimate of 2.0 (P=0.072). A similar effect with respect to pipe smoking was noted in a case-control study of bladder cancer conducted previously in Canada (10). Alcohol Table 4 shows RR estimates for males for various categories of average frequency per day for the con- sumption of beer, spirits, and wine. Since variables TABLE 4.~RR~ for use of alcoholic beverages: Averdge frequency per day for males~ RR Drinks Wine, ever per day Bee~ Spirits used:never used <2 3.6 1.7 (1.8, 7.0) (0.9, 3.0) 2 or 3 2.7 3.4 0.5 (1.4, 5.3) (1.5, 7,8) (0.2, 0.9) >_4 4.8 1.3 (2.4, 9.8) (0.5, 3.4) Ounces RR for per day ethanol <1.04 4.4 (2.2, 8.5) 1.04-2.5 3.9 (2.1, 7.3) >2.6 4.8 (2.3, 9.9) ~ Estimates for beer, spirits, and wine are based on LR model: beer per day (3 dummy variables}+spirits per day (3 dummy variables)+wine user (1 dummy variable) and cigarette Iifetime consumption (1 categorized variable). Estimates for ethanol are based on LR model: ounces ethanol per day (3 dummy vari- ables}+cigarette lifetime consumption (1 categorized variable). ~ Numbers in parentheses are 90% confidence intervals. JyCI. VOL. 67, NO. 6, DECEMBER 1981

1222 Burch, Howe, Miller, and Semenciw representing all three types of alcoholic beverages were included in the same LR model, the estimates for each ..beverage were not confounded by the other beverages. Both beer and spirits showed evidence of increased RR to lifetime nondrinkers, although no evidence was shown that wine use resulted in increased risk. A variable was also created representing the estimated consumption of ethanol from.all sources, and RR for this variable were also presented in table 4. Although there is no suggestion of a dose-response relationship, the RR appear to be more stable than the estimates for use of the individual alcoholic beverages. When alcohol consumption was represented as estimated lifetime consumption, the RR showed some indication of a dose-response relationship, the estimates having been 3.5, 4.7, and 4.8, corresponding to a lifetime consump- tion of less than I0,000 ounces, 10,000-26,000 ounces, and 26,000 ounces or more. Interaction of Cigarettes and Alcohol Consumption The possibility of an interaction between cigarette smoking and alcohol consumption is complex. When the data were analyzed with the use of an LR model and an interaction term was included, the observed effect was manifested in a negative coefficient for the interaction term, which implied departure from simple muhiplicativity, the departure increasing with increas- ing dose for both cigarette smoking and alcohol. The estimates of risk presented in table 5 were based on the model including the interaction term. However, the LR coefficient for the interaction term (-0.10) was not statistically significant (SE = 0.11; P = 0.177), and there- fore one cannot reject the hypothesis of simple multi- plicativity. Asbestos Exposure Individuals were categorized by their exposure to asbestos in two ways: first, by whether they "reported" exposure in the interview ("reported" exposure), and second, by whether an occupational epidemiologist "classified" them as exposed or nonexposed ("classified" exposure or nonexposure). This classified assessment was done in ignorance of the subjects' case or control status. Among male subjects, a total of 36 cases and 27 TABLE 5.--RRa for combination of cigarette lifetime consu.mption and alcohol lifetime consumption Ounces of ethanol, in thousands RR for the following No. of cigarettes, in thousands 0 <150 150-299 >_300 0 1.0 2.0 3.9 7.6 <10 2.0 3.5 6.3 11.1 10-25 3.9 6.3 10.1 16.3 _>26 7.7 11.2 16.3 23.7 " Estimates are based on LR model: cigarette lifetime con- sumption (1 categorized variabie)+alcohol lifetime consumption (1 categorized variable)+interaction term. JNCI. VOL. 67. NO. 6, DECEMBER 1981 TABLE 6.--RR estimates for asbestos and nickel exposure Exposure and parameter Reported Classified exposurea exposurea (P-value) (P-value) Asbestos Discordant pair ratio 31:22 RR uncontrolled 1.4 (0.109) RR controlled for smokingb 1.6 (0.069) Nickel Discordant pair ratio 19:14 RR uncontrolled 1.4 (0.193) RR controlled for smokingb 1.0 (0.491) 13:8 1.6 (0.140) 2.3 (0.052) 13:11 1.2 (0.342) 0.9 (0.452) ° For definitions, see "Results.;' b Lifetime cigarette consumption (1 categorized variable). controls reported exposure to asbestos, and these num- bers were reduced to 14 cases and 9 controls when classified as exposed. Of these 23 subjects, all cases and 7 controls had occupational exposure. One female case and 1 female control reported exposure to asbestos, but both were classified as nonexposed by the occupational epidemiologist. Table 6 shows the discordant pair ratios and RR estimates for both exposure variables for males. Ciga- rette smoking was the only variable that showed any confounding effect for asbestos exposure. Smoking was negatively confounded with asbestos exposure in the present sample, and therefore the controlled RR esti- mates were higher than the uncontrolled RR. The higher estimates seen for the classified as compared to the reported exposure suggested that subjects tended to report some exposures as asbestos that were not and that this occurred equally for cases and controls. This possibility was supported by inspection of the actual reported exposures, because in several instances the substance involved, for example, "rock wool" for home insulation, would unlikely be asbestos. Further results were therefore restricted to the classified exposure variable. The occupational epidemiologist subdivided those with asbestos exposure into "possibly exposed" and "definitely exposed." Ten cases and 8 controls were classified in this way as possibly exposed, and 4 cases and 1 control as definitely exposed. RR estimated from an LR model including cigarette lifetime con~ sumpuon (1 continuous variable) tot posslt)le a,u definite exposures, respectively, were 1.5 (P=0.225) and 5.1 (P--0.08B). The few subjects classified as exposed to asbestos did not permit a definite analysis in terms_of duration of exposure or period from hrst exposure. With respect to duration o{ asbestos exposure, 1 case reported a period of 1 week (although intensive), and 1 control reported a period of 2 days. All other subjects reported at least 1 year of asbestos exposure. Only 1 case and I control reported a period from first exposure of less than 10 years.

Etiology of Larynx Cancer 1223 The possibility of interaction between asbestos expo- sure and cigarette smoking was difficult to examine in vie~' of the small numbers. Only 3 controls and no cases were lifetime nonsmokers. The RR estimates for asbestos exposure restricted to cigarette smokers (con- trolled by cigarette lifetime consumption) was 2.5 (p=0.051). Although this was suggestive of the risk being restricted to cigarette smokers, the interaction term was not statistically significant. Exposure to Nickel Exposure to nickel was classified in the same way as exposure to asbestos. Of the 23 cases and 18 controls who reported exposure to nickel, 13 cases and 11 controls were classified as exposed by the occupational epidemiologist. All of these had occupational exposure. Discordant pair ratios and RR estimates are shown in table 6. There was no suggestion in these data of any increased risk of larynx cancer associated with nickel exposure. Consideration of those classified as possibly exposed and those definitely exposed gave RR esti- mates of 2.5 (P=0.131) and 0.4 (P=0.093) (controlled for cigarette lifetime consumption). Other Occupational Exposures A general search was made of the occupational histories and occupational exposures to fumes reported by the study subjects. Elevated risks were observed for individuals who had worked for some time as moulders or coremakers (discordant pair ratio, 6:0), pipefitters or plumbers (8:0), and metal processors (21:4). Increased risk was also associated with occupational exposure to textile dusts (5:0), vehicle fumes (16:4), and foundry fumes and metal dusts (25:5). Considerable overlap was [0und between the metal processing occupations and exposure to foundry fumes. Elimination of those individuals who reported expo- sure to both occupation and substance left 8 cases and 1 control in the category of metal processing occupa- tions involving exposure to foundry fumes. Population Attril~utable Risk Percent Estimates of PARP (12) for males from the present study are: cigarette smoking (79.3), alcohol (70.1), Cigars and/or cigarillos (I1.2), pipes (6.3), and as- Used:never used or exposed. The importance of cigarette Smoking and alcohol was derived from the large RR associated with each and the high proportion of the controls exposed. PARP estimates were not additive in a. simple algebraic fashion. However, by forming a S~mple 2×2 table corresponding to cases and controls C!assified by exposure to any combination Of the five ~sk factors, we were able to estimate the cumulative PARP When alcohol and all three types of smoking Were included in the risk factor positive category, the cumulative PARP was 94%, and this estimate was not changed by the inclusion of asbestos exposure. DISCUSSION This study confirmed the major role played by tobacco products in the etiology of laryngeal cancer, cigarettes alone being responsible for almost 80% of all cases occurring in the population (assuming causality of the association)• Rothman et al. (2) reported that data_from other studies show a good linear relationship between RR and the number of cigarettes smoked per day. The data from the present study were fitted to a simple linear regression model, constrained to pass through the point RR= 1.0, dose-0 cigarettes per day, the other points being weighted inversely by their variance. The slope of the line is 0.127 (increase in RR per cigarette per day). This compares with a value of 0.647 reported by Rothman et al. (2) from the data from the case-control study of Wynder et al. (1). Thus, although the dose-response relationship in the present data is highly statistically significant, the effect is weaker than that in the study of Wynder et al. (1). .A relevant comparison between studies can be made in terms of the PARP. For example, from the data provided by Wynder et al. (1) the PARP can be estimated to be 87% for all types of tobacco products, which is very similar to the value reported here. This is a reflection of the fact that the maximum possible value of the PARP is 100%. For factors with large RR and a high proportion of the population exposed, values of the PARP will tend to converge to 100% even though the estimates of risk and proportion exposed vary from study to study. In contrast to the results reported by Wynder and Stellman (13), in the present study no important difference in risk was found from filter as opposed to nonfiher cigarettes. This thus raises the question of the effectiveness of the use of filters as a protective measure for larynx cancer. A puzzling observation was the decrease in risks seen for noninhalers of nonfiher cigarettes but a lack of differential for smokers of filter cigarettes. This anomaly could possibly be due to chance. RR for larynx cancer of the order of 2.5-4.0 have been observed in several cohort studies of alcoholics and heavy drinkers (2). AlcohoI was also implicated as a risk factor in the case-control studies reported by Wynder et al. (1, 14). The present data suggest that indicated by previous studies. In particular, evidence exists that even among nonsmokers, alcohol use in- creases risk (table 5). Providing the estimates for alcohol as a risk factor obtained in the present study reflect causality, the importance of alcohol in terms of its attributable risk is almost as great as that for tobacco products, ~ whereas elimination of both alcohol and smoking hhbits in the general population could prevent more than 90% of the observed cases of laryn- geal cancer. jNcr VOL 67. NO. 6. DECEMBER 1981

122~4 Burch, Howe, Miller, and Semenciw There are a number of methodologic problems involved in the attempt to assess any relationship between asbestos exposure and larynx cancer. Three cohort studies (4, 6, 7) of larynx cancer occurring in workers exposed to asbestos have been reported. The standard mortality ratios observed in the studies ranged from 1.8, to 5.4, the comparisons being based on expected rates from the general population. However, these studies were unable to control for the possible confounding effects of cigarette smoking, and the relatively low rate of larynx cancer means that the observed number of cases is generally subject to large sampling errors. The alternative approach, namely that of the case-control study, is subject to three meth- odologic problems: the relative infrequency of larynx cancer and exposure to asbestos in the population, the difficulty in the assessment of whether or not an individual really has been exposed to asbestos when one relies on interview data, and the need to ade- quately control for the possible confounding elects of tobacco and alcohol.. In view of these problems it is hardly ~urprising that RR estimates ranging from 1.4 to 13 have been obtained from case-control studies (3, 5, 8, 15). However, the estimate obtained from the present study (2.3), although not quite statistically significant at the 0.05 level with the use of a one-sided test, is consistent with the observations of other case- control studies. Further, the increased risk seen for plumbers and pipefitters is also consistent with an association between asbestos and laryngeal cancer. The present study, therefore, supports the hypothesis that asbestos may be a cause of larynx cancer. In contrast, the RR observed in the present study for nickel exposure is not elevated. A study (9) that suggested an increased cancer .risk for nickel workers did not indicate that a specific group of workers was affected. Nevertheless, any risk could possibly be re- stricted to a limited number of specific occupations in the nickel processing industry. Under these circum- stances cohort studies on the basis of accurate occupa- tional exposure records may be of more value than the case-control approach. REFERENCES (I) WYNDER EL, COV~Y LS, MABUCH1 K, MUSHtNS~] M. Environ. mental factors in cancer of the larynx. A second look. Cancer 1976; 38:1591-1601. (I) ROTHMAN KJ, C~N~ CI, F~,NDERS D, FRtED MP. Epidemiolo~. of laryngeal cancer. Epidemiol Rev 1980; 2:195-209. (3) STELL PM, McGtLL T. Asbestos and laryngeal carcinoma. Lan- cet 1973; 2:416-417. (4) NEWHOUSE ML, BERRY G. Asbestos and laryngeal carcinoma. Lancet 1973; 2:615. (5) MORGAN RW, SHET~GAgA PT. Occupational asbestos exposure. smoking and laryngeal carcinoma. Ann NY Acad Sci 1976: 271:308-310. (6) SELmOW IJ, HAMMONO EC, SEmMAN H. Latency of asbestos disease among insulation workers in the United States and Canada. Cancer 1980; 46:2736-2740. (7) E~LUNO A, EN6HOLM G, OSTLU~n E. Asbestos cancer in the construction industry. In: Advances in medical oncolo~'. research and education. Vol III. Oxford: Pergamon Press, 1979:89-95. (8) HINDS MW, THOMAS DB, O'REILLY HP. Asbestos, dental X-ra.vs, tobacco and alcohol in the epidemiology of laryngeal cancer. Cancer 1979; 44:1114-1120. (9) PEDERSEN E, HOGETVEIT AC, ANDERSEN A. Cancer of respiratory organs among workers at a nickel refinery in Norway. Int J Cancer 197~; 12:32-41. (10) Howe GR. BURCH JD, MILLER AB, et al. Tobacco use, occu- pation, coffee, various nutrients, and bladder cancer. JNCI 1980; 64:701-71~. (11) BRESLOW NE, DAY ME, HALVORSEN KT, PRENTICF~ RL, SASAI C. Estimation of multiple relative risk functions in matched case-control studies. Am J Epidemiol 1978; 108:299-~07. (12) COLE P, ~MACMAHON B. Attributable risk percent in case-control studies. Br J Prey Soc Med 1971; 25:2't2-244. (1.3) WYNDER EL, STELLMAN SD. Impact of long-term filter cigarette usage on lung and larynx cancer risk: A case-control study. JNCI 1979; 62:471-477. (14) WYN~ZR EL, BI~OSS IJ, DAY E. A study of environmental factors in cancer of the larynx. Cancer 1956; 9:86-110. (15) BIANCHI C, DIBONITO L, CASTELLI M, BROLLO A. Exposition a l'amiante darts le cancer du larynx. Pathologica 1978; 70: 40~-408. j.,,,;cl. VOL. 67. NO. 6. DECEMBER 1981