Document omR6NDKdDv626yG6v4dz3NoZr

grcWhe- n/'T C' llr^irS Some Effects of Cigarette Smoking, Arsenic, and S02 on Mortality Among US Copper Smelter Workers Philip E. Enterline, PhD; Gary M. Marsh, PhD; Nurtan A. Esmen, PhD; Vivian L. Henderson, MS; Catherine M. Callahan, MS; and Myunghee Paik, MS Previous studies of the relationship between arsenic levels and respiratory cancer among copper smelter workers have not directly accounted for possible effects ofSO* exposure and cigarette smoking. This is a report on the 1949-1980 mortality experience of 6,078 white male workers who worked at least 3years between 1 January 1946and 31 December 1976 at one or more of eight US copper smelters. The completeness of the cohort was verified statistically, and worker exposures to arsenic, SOa. dust, nickel, cadmium, and lead were estimated from retrospective industrial hygiene surveys reported else where. By using internal controls, a dose-response relationship for lung cancer was observed with exposure to arsenic and SOn. When cigarette smoking data were included with arsenic and SOb exposure data in a nested case-control analysis, only smoking and arsenic were statistically significant factors. The arsenic-lung cancer relationship was confined to a single smelter associated with high content feed. In the remaining smelters mortality for all causes of death and for all cancer was not high based on comparisons with national, state, and local rates. There were in recent years 16 copper smelters oper ating in the United States. Of these, two with high airborne arsenic levels--one in Tacoma, WA and one in Anaconda, MT--have been the subject of epidemiologic investigations in which levels of air arsenic were related to deaths from respiratory cancer.1"3 A third smelter in Salt Lake City, UT has been studied, but levels of air From the Departments of Biostatistics (Dr Enterllne. Dr Marsh. Ms Henderson, Ms Callahan, and Ms Paik) and Industrial and Envi ronmental Health Sciences (Dr. Esmen). Address correspondence to Department of Biostatistics. Graduate School of Public Health, University of Pittsburgh, Pittsburgh, PA 16861 (Dr Enterline). 0096-1736/87/29UMJ831S08.00/0 Copyright by American Occupational Medical Association arsenic were not reported.4 Studies of all three smelters revealed an excess in respiratory cancer. For the two in which levels of air arsenic were estimated this excess was directly related to arsenic exposure. This is a report of an historical prospective mortality study and a case-cohort study of workers from eight copper smelters in the United States including the smelter studied earlier for which levels of air arsenic were not estimated. Males who worked a cumulative 3 years or more in copper smelting during the years 1946 1976 were observed for deaths through 1980. Smelters were selected for study because they had been in oper ation for a long time and had adequate employee records on which to develop a cohort. Previous studies of copper smelters that show a re lationship between arsenic levels and respiratory cancer did not directly account for possible effects of SO2 ex posure and cigarette smoking on this relationship. SO has been reported to be an animal carcinogen, and its irritant properties have been considered as a possible mechanism for a cocarcinogenic or cancer promoting effect.3,8 Since levels of S02 and arsenic often vary together in the environment of smelters, a role for SG2 in the production of lung cancer among arsenic-exposed workers has not been eliminated. Cigarette smoking is a potential confounder in any study of respiratory can cer. High exposures to S02 are common to nearly all smelting operations since most ores smelted are sulfide ores and the smelting process involves the conversion of these sulfides to oxides. However, arsenic exposure varies depending mainly on the arsenic content of the feed. It was hoped that by studying several smelters with varying levels of arsenic in the feed, contributions made by arsenic to respiratory cancer could be sepa rated from any contributions made by S02. Also, by *t adding a nested case-cohort study, any confounding effects of cigarette smoking could also be accounted for. Methods The primary data collection phase of this study was conducted between August 1981 and October 1982. Dur ing, this time a University of Pittsburgh data collection team traveled tb each plant and screened' personnel, payroll, and medical records for all male production employees and ex-employees who met the study eligibil ity criteria. All pertinent demographic and work history information was then microfilmed so that time consum ing data abstraction activities could be conducted off site. The completeness of the study population as assem bled from plant records was verified using the method developed by Marsh and Enterline.7 The verification scheme involved cross-checking a self-weighted random sample of names that appeared on two or more triannually successive first quarter returns filed by plant management with the Social Security Administration against personnel records abstracted from plant files. Any names appearing on the social security forms (form 941) but not appearing in the cohort werei checked against plant records to determine the reason for exclu sion. For the most part names on form 941 not in the cohort were ineligible for the study (eg, female, worked <8 years, or no potential for plant exposure such as cafeteria worker, secretary, etc). However, for some, records were in the plant files but missed during the initial survey, and for some, the employee record could not be found. The number of sample records that were in the plant files but had been missed divided by the total sample size adjusted for those missed for accepta ble reasons forms the basis for a "missed rate." All plants had "missed rates" below 5%, and this was considered acceptable. However, some names on the social security forms could not be located in plant files. The number of such names divided by the number in the sample is called the "undetermined rate." This was well below 5% for ail plants except one (Plant 4). For this plant the rate was 24.6%. It was subsequently found that records from this plant had been transferred to a different location for the purpose of an epidemiologic study. AH these records were recovered and added to the study cohort bringing the "undetermined rate" for that plant well below 5%. One special feature of this study is that for one or more of the smelters it was possible for one of us to estimate exposures to arsenic, SOa, dust, nickel, cad mium, and lead by job and by year, thus enabling workers to be assigned an exposure level to these con taminants. By utilizing the strategy developed by Marsh,6 the diverse occupational histories and multi agent exposure data from the various smelters were merged into a uniform format suitable for statistical anniygig Several exposure-disease relationships could be examined as the result of this format and are reported elsewhere.9 This report summarizes the results of the study only as they relate to exposures to arsenic and pf SO and to mortality of white males'1 for all' causes, all cancer, and lung cancer. There were no females identi fied as eligible for study and only a small number of blacks, too few to make separate analysis worthwhile. The total and cause-specific mortality experiences of the white male copper smelter workers were examined during 1 January 1949 to 31 December 1980. All pri mary analyses were conducted utilizing the Occupa tional Cohort Morality Analysis Program developed by Marsh and Preminger10 and later extended by Caplan et al" and Marsh et ai.IE All cause mortality reflects the general levels of health of copper smelter workers, whereas cancer may be more directly related to contaminants in the working environment of copper smelters. Six of the eight copper smelters in this study had a mine and mill operating in conjunction with the smelter, and one of these also had a refinery. Only the mortality experience of workers whose only work experience was at the copper smelter is reported here. Results Table 1 shows the copper smelters studied along with some information about each smelter. The oldest of these had a start-up date of 1901, and the newest had a start up date of 1950. Six of the eight smelters had start-up operations in 1940 or earlier. Although the age of the smelters complicates attempts to calculate exposures it ensures that exposures occurred in the sufficiently dis tant past and for sufficiently long periods to permit meaningful observations with regard to cancer. Esti mated arsenic concentrations of the feed for seven of the eight smelters are also shown by time. Feed includes the processing of spiess as well as copper ore. Plant 3 is the smelter previously studied by Rencher et al4 and before 1959 operated as a custom smelter reprocessing some materials high in arsenic content from other smelt ers. Clearly the arsenic content of the feed for Plant 3 is different from the other smelters in this study. Table 1 also shows the number of workers studied, person years studied, mean years worked, the mean level of cumulative and average arsenic exposure, and mean years worked in jobs where SOz exposure peaks exceeded 6 and 3 ppm. A total of 6,078 white male workers were identified as having worked a cumulative 3 years or more between January 1946 and December 1976--ranging from 189 for Plant 5 to 2,288 for Plant 3. As suggested by the arsenic content of the feed, Plant 3 had a considerably higher mean worker arsenic exposure than the other copper smelters studied. For Plant 2 levels of arsenic were too low to warrant estimates of individual worker exposure. For Plant 5 no data were available to make individual exposure estimates for arsenic or any other substance. Plants also differed somewhat in terms of exposure to ;f S02. It was felt that SOB exposures cannot be meaning fully expressed as averages, as can arsenic exposure, so || that for this analysis worker exposure to SOg has been Want NO. State 1 A2 2 TN 3 UT 4 NV 5 A2 6 AZ 7 AZ 6 AZ Start-up Date 1912 1901 1906 1910 1924 1950 1904 1940 % As in Feed (1950) .02 <.01 .63 .05 .08 .035 .03 TABLE 1 Selected Data for Eight US Copper Smelters No. of Workers Studied Person Yr Studied Mean Yr Employed Mean Level of Arsenic (pg/ma) 563 497 2.288 602 189 599 965 375 9.016 10.119 46.347 11,474 3,492 11.747 19,891 6.518 16.7 21.6 21.1 18.1 15.4 16.3 .21.5 15.7 12.9 68.6 12.0 8.1 7.0 7.5 Mean TimeWeighted Arsenic (pg/m'-yr) 177.9 959.7 176.2 124.5 114.5 133.4 Mean Years Exposed to SO, in Jobs with Peak SO, >6 ppm >3 ppm 6.1 10.4 7.0 11.2 8.0 10.2 0.0 4.4 3.7 6.0 4.4 7.9 4.2 4.2 expressed as years in jobs where peak exposures ex ceeded various levels. For Table 1, 6 and 3 ppm were chosen somewhat arbitrarily to illustrate differences among smelters. The mortality experience of workers meeting the study criteria was determined for the period 1949 through 1980 by first examining plant records to estab lish whether these workers were still employed or death was known to the company. The remaining worker records were submitted to the Social Security Adminis tration (SSA), and it was determined whether individ uals were alive as of 31 December 1980 as evidenced by the fact that they were still contributing to or were receiving benefits from the SSA. Deaths known to the SSA were also identified. For the remaining workers who were neither known to be dead nor contributing to or receiving benefits from the SSA, clearance was made with state driver's license bureaus, the Veterans Ad ministration, the US Postal Service, and ultimately di rect telephone contacts. Of the total 6,078 white male workers with experience only in copper smelting 73.9% were alive at the end of the study, 24.5% dead, and the status of 1.6% unknown. For those known to be dead, death certificates were located for 94.1%. All deaths were coded by a nosologist to the revision of the International Classification of Diseases in effect at the time of death to facilitate comparisons with official vital statistics tabulations. For this report, cause of death groupings are as defined in the eighth revision of the international lists of causes of death. Table 2 shows standardized mortality ratios (SMR) for all causes of death, all cancers, and lung cancer where the expected numbers of deaths are based on the national mortality experience for white males, the state mortality experience for white males, anS--the mortality experience of counties in the area in which the smelter was located, also based on white males. SMR is the ratio of observed to expected number of deaths (XlOO) for persons of the same age, sex, and race and living during the same period, where the expected is based on some reference population. In this case, the reference popu lation is the total national population, the population of the state in which the smelter is located, and the popu lation of the county or group of counties-in the area in which the smelter is located. Whereas for Table 2 it was possible to examine mortality for cancer for the years TABLE 2 Observed Deafhs'ancJ SMRs for Selected Causes of Death Plant No. 0brrlNOof Deaths SMRs 8ased on Expected National State Local Area All causes (1960-1980) 1 93 2 92 3 491 4 118 5 52 6 144 7 - 204 6 41 All cancer (1949-1980) 1 22 2 20 3 128 4 21 5 10 6 32 7 41 83 Lung cancer (1949-1980) 18 27 3 48 48 54 6 'a 7 10 80 97.7 96.5 88.8t 78.Of 128.9 94.2 74.5f 87.3 101.1 88.8 101.6 74.8f 134.2" 98.2 78.2t 87.2 101.2 89.7 101.6 70.8f 124.3 98.5 78.4t 93.7 98.2 84.9 98.2 56.6t 94.6 87.6 53.4-f 27.5- 109.9 89.5 I37.6t 58.8' 106.4 98.3 71.030.4- 107-9 95.0 132.1t 68.5 108.7 96.2 74.9 35.0 118.1 90.3 119.7 73.9 141.1 70.3 51.4- 0.0' 127.5 86.3 226.91 71.4 149.9 76.0 55.3 0.0 125.0 89.4 210.9| 97.6 160.8 74.1 60.5 0.0 *P< .05. tP<.01. 1949-1980, mortality for all causes could be examined only for the years 1960-1980 since detailed data by county are available only for those periods. This type of analysis is made possible by the mortality and population data system maintained by the Department of Biostatis tics, Graduate School of Public Health, University of Pittsburgh. This system was developed primarily from the National Center for Health Statistics data tapes furnished to the Department of Biostatistics that contain detailed demographic and cause of death information for cancer deaths starting in 1949'and for all deaths starting in 1960. The MPDS is periodically updated through a cooperative arrangement with the Environ mental Protection Agency Office of Biometry in Re search Triangle Park, NC. That office has also provided Journal of Occupational Medicine/Volume 29 No. 10/October 1987 833 population data needed for calculation of the rates used for calculating expected numbers of deaths. For most plants the all causes SMBs based on national data approximated the SMBs based on state or county mortality experience. For Plant 3, which is located near Salt Lake City, the use of US mortality data produced an all cause SMB considerably lower than mortality data for Utah and the area around Salt Lake City. For Plant 5 the all cause SMBs were high, and the SMB was statistically significant when a comparison was made with death rates for the state where the smelter is located. This is a plant that closed in 1972 and is the plant for which no exposure information was available. For all cancer and for lung cancer the SMBs based on national mortality were fairly close to SMBs based on state or local data, except for Plant 3. For Plant 8 the cancer mortality experience is remarkable. At this time we have no explanation for the extremely low observed cancer mortality for this plant aside from the possibility that this could be a chance occurrence. There was nothing learned in the verification process that suggests a reason for the low cancer mortality for workers from this plant. A total of 196 names were sampled from forms 941, as described above. Of these 139 were not in the cohort. However, for 135 of these the reason was acceptable (i.e., females, worked <3 years, etc). For the four with unacceptable reason three were found in the plant files but had been missed at the time of the initial microfilming. The missed rate for this plant was 4.9% whereas the undetermined rate was .5%, using definitions given above. Note, however, that only a sample of names appearing on forms 941 were actually used in the verification process, and it would be easy to miss a few workers who died of cancer since they would constitute only a smali part of the population. Thus, the verification procedure is useful primarily for the identification of the entire sets of records that are missing or were missed during microfilming, as was the case in Plant 4 as described above. Only a check of all names appearing on forms 941 with a complete follow up for missing names would be likely to establish the reason for the low cancer rate in Plant 8 if, in fact, this was other than a chance occurrence. Because of the variation in the background mortality rates, particularly those for cancer, in the areas where the copper smelters were located and because copper smelter workers may not have a mortality experience like others in the areas in which they live, it is unclear what external reference population should be used. As an alternative to making further external comparisons we have chosen a method for analysis of exposures in relation to mortality using the study population as an internal control. The procedures used are those origi nally described by Mantel and Haenszel13 and Mantel14 and later adapted for use in cohort studies by Hakulinen.13 The computing method, which has been incorpo rated into the Occupational Cohort Mortality Analysis Program system,18 was modeled after those developed by Gilbert and Buchanan.16 These computations utilized person years by age, calendar time, time since first exposure (latency), and exposure level. Expected num- OOA ber of deaths in each exposure level were calculated by multiplying the number of person years in that exposure level and stratum (age, calendar time, latency, etc) by the death rate for the combined exposure categories in that stratum. For example, if there are a thousand person years and ten deaths from a cause of interest in a particular stratum and if there are 400, 300, 200, and 100 person years, respectively, in each of four exposure categories within that stratum the number of expected deaths in these categories would be 4, 3, 2, and 1, respectively. The total number of deaths expected in a given exposure category is obtained by summing over strata. For arsenic exposure, person years at risk rela tive to exposure categories were lagged for 5 years since it was not believed that arsenic would produce any cancers in < 5 years. However, for SOa we are uncertain what (if anything) is produced, and no lag was intro duced. For each exposure category a summary measure was calculated as a Mantel-Haenszel relative risk. This can be thought of as risk in a particular exposure category relative to the risk for pooled exposure categories summed across strata. A test of whether the number of deaths observed in a particular exposure category dif fers significantly from that expected is the MantelHaenszel test statistic as adapted by Hakulinen,13 and this was used for assessing overall differences in the relative risk for the various exposure categories. In addition. Mantel's test for trend14 was used to assess whether the relative risks increase significantly with increasing exposure. For this test an exposure value or score was assigned to each exposure category, and observed and expected means of these scores (obtained from weighting the observed and expected deaths, re spectively) were calculated and compared. The variance of the difference in the observed and expected scores was then calculated to obtain an approximately nor mally distributed test statistic. Table 3 shows the results of an analysis of deaths from lung cancer in relation to individual arsenic ex posure expressed as a time-weighted exposure. That is, air arsenic levels in each job and period (ug/m3) times the amount of time workers spent in those jobs (years), summed across all jobs and time intervals. Only Plants 1, 3, 4, 6, 7, and 8 are used since no exposure data were available for Plant 5, and for Plant 2 the arsenic content i of the ore was so iow that estimates of individual expo- f sure levels were not attempted. Also, only those men j whose work histories indicated some exposure to arsenic were included in this analysis. Table 3 shows that the hypothesis of no difference 1 among exposure categories could not be rejected with a two-sided probability of .28. However, the exposure! score for the trend test had a P value of .059, which | approaches statistical significance. A trend in relative f risks is very clear, rising from .58 to 1.60, and when | plotted on an arithmetic scale is concave downward. Exposures to arsenic reported in the six copper smelt ers represented by Table 3 were low relative to expo-J sures at two copper smelters studied earlier and not| included here.8'17 For example, the smelter with the! Effects of Tobacco, Arsenic, and S02/Enteriine et all TABLE 3 . Observed and Expected' Deaths from Lung Cancer and Relative Risks by Cumulative Arsenic Exposure for Copper Smelter Workers Exposed to Arsenic (Lagged 5 yr) Arsenic Time-Weighted Exposure (g/m3 yr) <100 100-249 250-999 1,000+ Test of Heterogeneity)- Exposure Score for Trend Test Observed Expected Trend Test P Value$ Observed 12 18 22 18 Expected 17.99 20.19 19.30 12.51 .28 892.86 682.52 .059 Relative Risk 0.58 0.85 1.21 1.60 ' Expected deaths based on experience of ail white copper smelter (only) workers adjusting for age. calendar year, and latency period, t Two-sided probability based on three degrees of freedom x2 test. One-sided probability of trend arising due to chance. TABLE 4 Observed and Expected- Deaths from Lung Cancer and Relative Risks by Duration of Exposure to SO* Greater than or Equal to Selected Peak Values for Copper Smelter Workers Exposed to SO* ' Peak SO* (ppm) <5 Duration (yr) Test of Exposure Score for Trend Test Trend Test Heterogeneityt P Values 5-9 10-14 15+ Observed Expected 0.75 Observed Expected Relative risk 3.00 Observed Expected Relative risk 6.00 Observed Expected Relative risk 12.00 Observed Expected Relative risk 24.00 Observed Expected Relative risk 14 15.5 0.88 21 21.4 0.97 18 20.9 0.76 10 13.1 0.55 2 4.4 0.23 10 11.0 0.89 9 11.5 0.75 3 6.1 0.44 2 2.6 0.73 2 1.1 2.27 11 9.2 1.24 9 7.5 1.24 7 5.6 1.31 6 1.6 5.21 3 0.8 5.59 43 42.3 1.04 25 23.6 1.10 17 12.4 1.62 3 3.7 0.76 0 0.8 0.00 .95 .95 .58 .97 .98 14.20 11.45 11.00 8.33 8.21 13.87 11.03 9.24 6.96 6.41 .46 .40 .13 .03 .01 * Expected deaths based on experience of all white copper smelter only workers adjusting for age. calendar year, and latency period, t Two-sided probability based on three degrees of freedom *1 test. ' One-sided probability of trend arisirradue to chance. . highest exposure (Plant 3) had a mean worker exposure of 68.6 fig/m3 and a mean time-weighted level of 957.7 lig/m3-yr. In a recent reanalysis of data from the copper smelter at Tacoma, WA the mean worker exposure was 580 f*g/m3, and the mean time-weighted exposure was 6,530 ng/m3-yr.17 The highest exposure category shown in Table 3 roughly corresponds to the two lowest expo sure categories in the recent Tacoma report. A recent report on the Anaconda smelter yields exposure levels similar to those for Tacoma. The mean worker exposure was 685 fig/ma, and the mean time-weighted exposure was 10,400 jtg/m8-yr.s Table 4 is similar to Table 3 but deals with respiratory system cancer in relation to SO* Only those men whose work histories indicated some exposure to S02 are in cluded. As in Table 8 none of the tests for heterogeneity are statistically significant; however, the trend test is statistically significant (P = .03) at S08 peak exposure levels exceeding 12 ppm. To examine the separate effects of arsenic and SO* giving consideration to cigarette smoking, a case-cohort study was nested in the study already described. All copper smelter workers who died of lung cancer between 1 January 1949 and 31 December 1980 were identified as cases. Controls were a random sample of all workers with stratification by plant and year of birth. In select ing controls certain members of the original cohort were considered ineligible. Excluded were members of the cohort who died of lung cancer, those dying of any cause before 1 January 1949, those with unknown date of death, those dying before .age 45, those <45 years old as of the termination.of the study (31 December 1980), those who terminated employment before 1 January 1949, those whose year oftermination of follow-up minus year of birth was <45, those with unknown date of termination, and those with unknown date of birth. Controls represented a 5% stratified random sample given restrictions outlined above. A case-cohort study design has been described by Prentice et al18 and was chosen partly because it allowed interviewing for smok ing histories to start before lung cancer cases were1 known and because it yields seif-weighted estimates of smoking for workers from each smelter studied. Smoking histories were obtained through telephone .Journal of Occupational Medicine/Volume 29 No. IQ/October 1987 835 interviews with the worker (if still living) or a knowl edgeable informant, ideally a member of the worker's immediate family. Smoking data were obtained for 76% of the lung cancer cases and for 85% of the control group. The difference was probably because most con trols were alive as of the date the study was conducted, whereas ail cases were deceased and it was more difficult to obtain information on dead individuals than living individuals. When cases were compared with controls who had died before the end of the study, completion rates were quite similar. Cases and controls for whom smoking data were ob tained were compared with those from whom smoking data was not obtained on a large number of variables related to both lung cancer and probable occupational exposure levels, and it was concluded that the discrep ancy in noncompletion rates between cases and controls has not introduced any significant bias. Data were ana lyzed by logistic regression with backward stepwise elimination using the statistical package BMDP.19 There were two basic areas of interest--matching or confound ing variables, including cigarette smoking and demo graphic variables, and occupational exposures. Cases and controls were initially analyzed to identify those smoking and demographic variables in combinations that best explained lung cancer mortality. The smoking variable chosen was a combination of duration of smok ing and time since smoking began, whereas the only important demographic variable was age at death or age on 31 December 1980. There were 183 copper smelter workers for whom data regarding smoking, arsenic exposure, and SOa exposure were available. These workers were from Plants 1. 3, 4. 6, 7. and 8, represented in Tables 3 and 4. As in Table 3, arsenic was expressed as a timeweighted measure. S02 exposure was somewhat arbi trarily expressed as duration of exposure to SOa in jobs where peaks exceeded 6^t>pm. Other exposure levels were examined, and results were generally supportive of the results at 6 ppm. An examination of the data set revealed that for 181 of 183 copper smelter workers, arsenic exposure levels ranged from zero to 4.196 ng/ ms-yr, whereas two workers had exposures of 9.982 and 10,503, respectively, one a case and one a control. It TABLE 5 Statistically significant log odds ratios for lung cancer with backward elimination after entry of years smoked, years since started smoking, timeweighted exposure to arsenic, years in job with SO* exposure greater than 6 ppm, all four-way. three-way, and two-way interactions, and age Variables Original Units (x) AH workers (181) Plant 3 woikers (86) Transformed Units ('i) All Plant 3 workers workers (181) (88) Years smoked (xi) Years since started (x2) Interaction (xi. x2) Arsenic (x3) SO* (x4) Interaction (x3, x4) P (goodness of fit) 0.0799 0.0197 -0.00132 0.00104 NSNS .642 0.114 0.0203 -0.00221 0.000816 NS NS .666 0.199 NS NS 0.0439 NS NS 343 1.27 -00294 -0.15 0.0093 -0.697 0.0192 .474 * NS not significant at 5% level. 836 was decided that these two outliers could distort the analysis, and they were dropped. The results ofa logistic regression analysis performed on copper smelter workers are shown in Table 5. All major variables shown in Table 5 plus age were entered into the regression analysis, along with all four-way, three-way and two-way Interactions among exposure and smoking variables (16 terms). Table 5 shows terms that were statistically significant (P <.05) for smelter workers in the case-cohort study and smelter workers from Plant 3. Data in original units and square roots of the original units were used. The fit of the model is shown by the P values at the bottom of each column. For data in original units the logistic coefficients for smoking and arsenic exposure were statistically signif icant and not much different for workers from all plants and workers from Plant 3. When only workers from plants other than Plant 3 were used the only statistically significant variables were those related to smoking (not shown). Thus the relationship between arsenic and lung cancer is apparently driven by Plant 3. Since cases and controls were unmatched, some additional matching variables (original units) were added to those in Table 5. Only Plant 3 was used here since all information about arsenic and lung cancer appears to be contained in this data set. Matching variables entered together were year of birth and year of hire. These had iittie effect on the coefficients shown in Table 5. The coeffi cient for arsenic, for example, was changed from .000816 to .000921. Also, removal of smoking variables had little effect on the coefficient for arsenic. When original units were used, the coefficient for arsenic exposure was surprisingly large, predicting un realistically high relative risks when arsenic exposures like those for the smelter at Tacoma. WA were entered. Moreover, as noted for Table 3, when relative risks were plotted against dose the dose-response curve was con cave downward, tike that reported for the Tacoma, WA and Anaconda. MT smelters, whereas the predicted dose-response curve from a logistic regression for ex posures in the range of interest is concave upward.17 To alter the shape of the dose-response curve to give more realistic projections at higher ieveis of arsenic exposure, data were transformed by taking square roots. A shown in Table 5, when this was done only years smoked and arsenic exposure were statistically significant for the full data set. When analysis was : confined to workers from Plant 3, some two-way inter action terms were statistically significant. This trans form gives a- dose-response curve for arsenic that is J slightly concave downward and gives fairly realistic | response estimates at extrapolated higher arsenic ex posure levels. < Because the controls were actually a random samp!e| of the eligible population, a profile of smoking habits for| copper smelter workers may be obtained. There were? 126 controls for whom smoking histories were obtained? Of these, 70.6% gave a history of cigarette smoking? For Plant 3 this percentage was 64.7, while for the remaining plants taken together this percentage war 74.7 The difference in smoking habits between Plant $ Effects of Tobacco, Arsenic, and S02/Enterline ef i * 0 > located in Utah, and all other plants was not as large as might have been expected in view of the fact that the lung cancer death rate for Utah is only about half the rate for the rest of the United States, and the difference is probably due to cigarette smoking. This suggests that smelter workers at Utah were not typical of the Utah population with regard to smoking behavior and that SMRs for lung cancer using Utah rates to calculate expected deaths are probably overstated. 1 Discussion This report examines a part of a larger data set generated as a result of examining the mortality expe rience of copper smelter workers in the United States. It shows that all cause mortality rates in copper smelter workers are not unusually high and for lung cancer the one smelter that appeared to have a problem was the one with an appreciable exposure to arsenic. Arsenic clearly emerged as a primary cause of lung cancer in smelter workers after adjusting for cigarette smoking and S08 exposure. An important question is whether there is an inter action among smoking, arsenic, and SOe. The logistic model used for the case-cohort study is a multiplicative model so that lack of interaction in a logistic model specifies lack of interaction other than multiplicative interaction. Since no significant interactions could be detected when data were analyzed in their original form it can be inferred that interactions examined did not deviate from simple multiplicative interactions. What we can say, therefore, from the case-cohort analysis is that we are unable to reject the notion that arsenic. SOa, and smoking have multiplicative effects. However, we did not test for multiplicative effects so that some other form of analysis is needed to provide more direct information on this subject. This study illustrates the difficulty in finding a suit able population for calculating expected numbers of deaths in an occupational cohort. For the copper smelter located in Salt Lake City, UT where there are large numbers of Mormons who abstain from cigarette smok ing, calculations of standardized mortality ratios for lung cancer using the US mortality experience as the standard seemed inappropriate. Moreover, workers at the Salt Lake City plant did not appear to be represent ative of the population of the area in which the smelter is located. Thus, the true SMR for lung cancer for the Utah smelter workers apparently lies between 119.7 (US expected) and 226.9 (Utah expected). A rough estimate ofthis can be obtained by calculating the excess in respiratory cancer predicted at the Utah smelter given data on arsenic levels at that smelter and knowl edge ofthe relationship between arsenic and lung cancer from studies of other smelters. A recent reanalysis of data from the copper smelter located in Tacoma, WA shows that a power function provided a good fit of the dose-response data available for that smelter.17 This was confirmed when a similar data set was reanalyzed from a smelter at Anaconda, MT. The average time- weighted exposure to arsenic for workers at the Sait Lake City smelter was 959.7 fjg/m*-yr. By using the power function derived from the Tacoma data, this predicts a standardized mortality ratio 154.0, an SMR that falls between the SMR off 119.7 calculated for the Utah smelter using the US white male population to calculate expected numbers of deaths and the SMR of 226.9 using the Utah population to calculate expected deaths. This study also suggests that if there are a few missing records in an occupational cohort study their absence is extremely difficult to detect using methods proposed to date.7 For one smelter in this study the death rate for cancer was very low, and this could have been due to missing records for workers who died of that disease. Although this smelter does not greatly influence overall results, the fact that such situations might exist is disturbing and detracts from the validity of occupational epidemiology. Finally, this study represents an experience in apply ing logistic regression analysis to occupational data and demonstrates the need to actually test the resulting model not only by goodness-of-fit tests but also by seeing if it has believable predicting ability. Since the logistic model is multiplicative and the dose-response curve concave upward, extrapolation to arsenic levels much above those from which the model coefficients were derived resulted in extraordinarily high predicted risks for lung cancer. An attempt was made to correct for this by a transformation of variables. However, what transformations to make and what variables to trans form is a difficult decision. Clearly there is no reason to believe that all independent variables have the same relationship to the dependent variable nor with each other. Care must be exercised therefore in extrapolating coefficients developed from logistic or any other form of regression analysis. This is a study of the mortality experience of workers from eight US copper smelters. For each copper smelter an attempt was made to estimate worker exposure for various contaminants. Presented here is the mortality experience for all causes of death, all cancer, and lung cancer. For lung cancer there was a dose-response relationship with exposure to arsenic and with exposure to SOa using an internally generated control group. When data on cigarette smoking were added to exposure data for arsenic and SOs in a nested case-cohort analysis only smoking and arsenic were statistically significant factors. The relationship between arsenic and lung can cer was apparently confined to a single smelter where the feed had been relatively high in arsenic content. Arsenic exposure did not appear to be related to lung cancer in the other smelters, and mortality for all causes of death and for all cancer was not high. Acknowledgments This research was supported under a contract between the Smelter Environmental Research Association and the Department of Biostatis tics, University of Pittsburgh. i ^ ' The authors greatly acknowledge the assistance and cooperation of Donald Robbins and Gerald Schurtz as well as others in the Smelter Environmental Research Association member companies who partici pated in this research effort. In addition we acknowledge with great appreciation the staff at the University of Pittsburgh: Michele LaValley, Andrea March, and Donald Fisherowaki who collected and coded demographic and work history data and conducted follow-up and verification, computer programmers Mary Preininger, Marcia Goetsch, and Jay Graham, Cathy Champagne who collected the data on tobacco smoking, and student assistants Ella Lyons who partici pated in the statistical analysis, Donald Tuchman who assisted in collecting and evaluating industrial hygiene data, and Gloria Claus and Linda Kobus who patiently typed and corrected this manuscript. n9i@r0iCss 1. Lee AM, Fraumeni JF: Arsenic sod respiratory cancer in man: An occupational study. J Natl Cancer Inst 1969:48:1045-1052. 2. Higgins ITT, Welch KB, Ob MS, et al: Arsenic exposure and respiratory cancer in a cohort of 6,044 Anaconda smelter workers. Report to the Chemical Manufacturers Association and the Smelter Environmental Research Association, Apr 1986. 3. Enterline PE, Marsh GM: Cancer among workers exposed to arsenic and other substances in a copper smelter. Am J Epidemiol 1982;116:895-911. 4. Rencher AC, Carter MW, McKee DW: A retrospective epide miological study of mortality at a large copper smelter. J Occup Med 1977;19:754-758. 5. Peacock PR. Spence JB: Incidence of lung tumors in LX mice exposed to (1) free radicals; (2) SO. Br J Cancer 1967:21:606-618. 6. Amdur MO: Effects of sulfur oxides on animals, in Nriagu NO (ed): Suitor la the Environment, Part II: Ecological Impacts. New York, John Wiley and Sons, 1978, pp 68-74. 7. Marsh GM. Enterline PE: A method for verifying the complete ness of cohorts used in occupational mortality studies. J Occup Med 1 1979;21:655-670. | 8. Marsh G: A strategy for merging and analyzing work history 1 data in industry-wide occupational epidemiology studies. Am Ind Hyg 1 Assoc J 1986:48:414^-419. I 9. Enterline PE, Marsh GM, Esmen NA. et al: Mortality among 3 copper and zinc smelter workers in the United States. Final Report to 1 the Smelter Environmental Research Association. University of Pitts- % burgh, Pittsburgh, PA, Jan 1986. 1 10. Marsh GM, Preininger ME: OCMAP: A user-oriented occupa- j tional cohort mortality analysis program. Am Stat1980:34:245. f 11. Caplan RJ, Marsh GM. Enterline PE: A generalized effective 1 exposure modelling program for assessing dose-response in epidetoi- | oiogical investigations. Computers Biomed Res 1983:16:567-596. I 12. Marsh GM, Ehland J, Paik M, et al: OCMAP/PC: A user- 1 oriented occupational cohort mortality analysis program for the IBM/ 1 PC. AM Stat 1866:40:303. s 13. Mantel N, Haenszei W: Statistical aspects of the analysis cf I data from retrospective studies of disease. JNOI 22(1958)719-748. g 14. Mantel N: Chi-square tests with one degree of freedom: Exten- g; sions of the Mantel Haenszei procedure. J Am Stat Assoc 1969:56:690-- p 700. 15. Hakulinen T. Mantei-Haenszei statistic Cor testing the associ ation between a poiychotomous exposure and a a rare outcome. Am J Epidemiol 1981:113:198-196. 16. Gilbert ES. Buchanan JA: An alternative approach to analyzing occupational mortality data. J Occup Med 1984:26:688-688. 17. Enterline PE. Henderson VL, Marsh GM: Exposure to arsenic and respiratory cancer. A reanalysis. Am J Epidemiol 1987;125:929- 938. 18. Prentice RL. Self SG. Mason MW: Design options for sampling within a cohort in Moolgavkar SH. Prentice RL (eds): Modern Statis tical Methods in Chronic Disease Epidemiology. New York. John Wiley and Sons, 1986. 19. Dixon WJ: BMDP Statistical Software. Berkeley, OA. Univer sity of California Press, 1981.