Combined of silica dust exposure and tobacco smoking on the prevalence of respiratory impairments among gold miners.

OK. Combined effect of silica dust exposure and tobacco smokingon the prevalenceof respiratory impairmentsamong goldminers. Scand J Work Environ Health 1990;16:411-22. The combined effect of underground gold mining dust with a high content of free silicaand tobacco smokingon the prevalenceof respiratoryimpairment wasexaminedamong 2209 South African gold minersand 483 nonminers. The subjectsweregrouped as having normal function; minimal, moderate or marked obstruction; marked obstruction with restriction; or pure restriction on the basis of their lung function profiles. Each profile group was compared with the normal group for exposure prevalences,and additive and multiplicativerelative risk models wereapplied to test for departure from the additivity of individual effects. Departure from additivity was found to increase progressively with the severityof obstructive impairment. The results indicated that approximately 94 010 of the caseswith the most severerespiratory impairment (N = 191) could have been prevented through the elimination of tobacco smoking.In conclusion, tobacco smokingwasfound to potentiate the effect of dust on respirato ry impairments.

virgin rock is silica, and approximately 30 070 of the respirable dust is free silica (I). The average current levels of respirable silica representative of the gold mining industry ranges from 0.05 to 0.84 mg/rn! for underground air (I). In addition, tobacco smoking is highly prevalent among the white miners. About 70 % were found to be current smokers, and only 12 % were nonsmokers (2,3).
Occupational exposure to underground dust in South African gold mines has been found to be associated with an increased prevalence of respiratory symptoms (2,3), chronic obstructive lung disease (3), and emphysema identified at autopsy (4, unpublished paper "Emphysema Type in Relation to Silica Dust Exposure, in South African Gold Miners" by Hnizdo et al). Experimental evidence from rats also indicates that exposure to crystalline silica causes emphysema and abnormally thick walls in the small airways with associated functional changes of airflow obstruction in rats (5,6).
There is, however, little evidence in the literature to indicate whether tobacco smoking interacts with silica dust to increase the amount of respiratory dysfunction (7). The objective of this study was to examine the effect of combined exposure to silica dust and tobacco smoking on the prevalence of respiratory dysfunction. It has been pointed out that, in the assess- ment of the combined effect of tobacco smoking and occupational exposure, the pattern of lung injury in pulmonary function, rather than a single lung function measure, should be examined (8,9). Thus, in this study, both, the individual lung function measures [forced expiratory volume in 1 s (FEVl.o), vital capacity (VC), and forced midexpiratory flow (FEF 25 _ 75 ) ] and the spirogram pattern were analyzed. We reanalyzed the data from a study (3) of 2209 white South African gold miners and 483 nonminers to (i) examine the individual lung function measurements FEVl.O' VC, and FEF 25 _ 75 for any dose-response trend and for the presence of interaction between dust and smoking; (ii) examine the patterns and degree of ventilatory impairment as reflected in the spirogram, their relation to silica dust and tobacco smoking, and the presence of interaction; (iii) estimate the attributable risk for tobacco smoking and underground gold mining dust exposure; (iv) examine the relevance of the pattern of ventilatory impairment to mortality from respiratory and cardiovascular diseases.

Study subjects
The 2209 study subjects consisted of all white gold miners who fulfilled the following criteria: (i) a number from the Medical Bureau for Occupational Diseases within the range of B9000 to C8999; (ii) age 45-54 years; (iii) minimal underground service of 10 years; (iv) residence in South Africa for at least 20 years; (v) service in mines other than gold less than two years; and (vi) attendance for a medical examination at the Medical Bureau for Occupational Diseases during the four-year period between 1968 and 1971. An annual medical examination is compulsory for all miners working in dusty occupations. Miners who were retired but attended the medical examination because they were seeking compensation or a medical examination were also included in the selection procedure. Only one of the medical examinations which the miners attended during the study period was used for the purposes of the study. A group of 483 male nonminers of the same age group, drawn at random from various municipal and government departments through the inclusion of all the workers from a particular office, was included in the original study of Wiles & Faure as a reference group but was not reported on in their paper (3). They were included in this reanalysis of the data.

Mining exposure
The mining service was estimated from records of the Chamber of Mines. These records show the exact number of shifts worked in each occupation. The occupations were classified on the basis of dust counts done by Beadle (10) into 11 groups for which the average surface area of respirable dust particles ( < 5 urn), after heat and acid treatment that leaves mainly silica particles, was calculated. For each miner the number of shifts worked in each occupation was weighted by the mean respirable dust count for that occupational group. The dust exposure was analyzed in terms of particle-years cumulated to the date of the medical examination (GOLDPY), and categorized particleyears/IO 000 (0, 1-1.5, 1.6-2.5,2.6-3.5,3.6-4.5, 4.6) (GOLDPYC), for which the means of the ranges were used to estimate the dose-response relationship. In a preliminary analysis particle-years were more strongly associated with the spirogram measurements than years of exposure.

Smoking habits
Full details of the smoking habits were obtained by means of a smoking questionnaire, and the reliability of the answers was checked against smoking histories recorded in the medical files of the Medical Bureau for Occupational Diseases. The information on smoking habits was utilized in the analysis in three different ways. First, the subjects were grouped into nonsmokers, ex-smokers, and current smokers, and indicator (zero, one) variables SMKI (ex-smoker) and SMK2 (current smoker) were created to represent the grouping. Second, the tobacco consumption and smoking status were combined, and the subjects were divided into the following five categories: 0 =nonsmoker, 1 = ex-smoker, 3 = current smoker with < 20 pack-years, 4 = current smoker with 20-39 packyears, and 5 = current smoker with~40 pack-years (SMOKEC). Third, lifetime tobacco consumption was analyzed as cigarette equivalent pack-years (SMOKEPY).

Lung function tests
The spirograms were measured on a Godard watersealed spirometer with the subject seated and wearing a nose clip. Spirograms judged to be of poor technical quality were excluded by a pulmonary physiologist. The lung function parameters measured were VC, FEVl.O' and FEF 25 _ 75 • For the analysis of the pattern of ventilatory impairments the miners were grouped according to the presence of an obstructive or restrictive lung function profile, and those with the obstructive profile were further subdivided according to the degree of obstruction.
A method of categorization used for asbestos-exposed workers (9) was adapted through the incorporation of features that are used by the Pulmonary Unit of Medical Bureau for Occupational Diseases when the types of lung dysfunction are categorized. The observed lung function tests were considered low if they fell below a 95 or 99 070 confidence limit. The confidence limits were calculated as (predicted value -SEE· z), where the predicted value was calculated from the equations given later in the report, SEE was the standard error of the estimate, and z = 1.645 or 2.326 for the 95 and 99 % confidence limit, respectively. The lung function tests were considered normal if they were above the lower 95 % confidence limit.
The prediction equations were derived by the Lung Function Unit of the Medical Bureau for Occupational Diseases and are routinely used for the assessment of lung function impairment in white miners who come to the Bureau for their medical examination. The equations were derived from 731 healthy miners consisting of 34 % smokers, 41 % ex-smokers, and 25 !t/ a nonsmokers who came to the Bureau during 1984. They averaged 51 years in age, 174.5 ern in height, and 81 kg in weight. The 95 % confidence limits estimated from Crapo's prediction equations [derived from a nonsmoking, white male population (11)] and from equations of the Medical Bureau for Occupational Diseases are very comparable for the ages considered in this study (45-54 years). The lower 95 !t/ a condifence limits derived from Crapo's equations and those of the Medical Bureau for Occupational Diseases for a man 51 years of age, 174 cm tall, and weighing 81 kg are 3.64 and 3.76 for VC, 2.97 and 2.97 for FEV 10' 2.08 and 2.32 for FEF 2s _ 7s , and 71.85 and 70.51 for FEV10/VC, respectively. The main difference in the two sets of predictive curves was the weight factor included in the equations of the Medical Bureau for Occupational Diseases. Weight was found to be a significant negative predictor for lung function among white South African gold miners, who are often overweight. The midexpiratory flow time (MEFT) was calculated as VC/(2 . FEF 2S _ 7S ) ' The prediction equations and standard errors of the estimate (SEE) were as follows: Th e subjects were grouped into six cat ego ries according to the lung funct ion profile as follow s: (i) normal (reference category) [those with normal lun g fun ction tests (ie, normal YC, FEY l.o' FEY l.oIYC , and MEFT Is)]; (ii) minimal obstruction (those who had a normal YC and one or two features of obstruction, but not all three, present, the three features being low marked pure obstruction (those with a normal VC, an FEV l.o below the 99 % con fidence limit, a low FE Yl.o/YC, and an MEFT of > I s); (v) marked obstruction with restr iction (those with marked pure obstruction except th at the YC was below the 99 % confidence limit) (all the subjects with an obstructive and a restri ctive element were found to ha ve an FEV l.o value that was below the 99 % confidence limit); and (vi) pure restrictio n (those with a low VC, a norm al FEV l.o, a normal FEY l.o/YC ratio, and an MEFT of :s; I s). Four men were found to have restr ictive impairment with a minimal obstructive element, and they were combined with the pure restrictive group.
There were 202 men whose lung function tests were excluded from the or iginal study by a pulmonary physiologist, as they were con sidered to be submaximal efforts, and the lung functions were not recorded in the computer file. In addition, the lung funct ion tests of 21 men were rejected from the analysis on the ad vice of one of the authors (EB), as they were found to have very unusua l parameters and did not belon g to any of the aforementioned categories.

Other medical and compensation data
Th e medi cal examination included the assessment of (i) simple chronic bronchitis by a questionna ire, (ii) rhonchi, (iii) sputu m prod uction, and (iv) radiological silicosis (3).
The medical files of the Medical Bureau for Occupational Diseases record dat es and types of compensation awarded to miners. Compensation for pneumoconio sis or chr on ic ob structive lung disease can be awarded if rad iological silicosis or specified lung dysfunction is found and the person ha s worked at least 10 years in highly dusty occupations or 20 years in less du sty occup ations. 3

Mortality data
The follow -up period for the collection of mortality data was from the date of the examination to the end of 1987, by which time 859 of the 2209 miners whose lung fun ction s were mea sured had died. The deaths were identified through a miner s' pension fund and the Department of Interior. The causes of death were coded according to the ninth revision of the International Cla ssification of Diseases (ICD) of the World Health Organizat ion by an independent physician , using the best available information deri ved from the death certificate , autopsy report , a nd med ical file of the Medical Bureau for Occupational Diseases. The cau ses of death of interest to this analysis are chronic ob structive lung disease (ICD 416, 491, 492 and 496) and ischem ic heart d isease (lCD 410-414).

Statistical methods
Analysis of lung function measurements. First, the relationship between each lung function mea surement (FEVl.o, VC and FEF 25 _ 75 ) , and du st and smo king was evalu ated by means of a linear regres sion analysis in which the effect of age, body weight, and height were adjusted for. Second, the individual measurement s were dichotomized into normal and a bnorma l if they fell above or belo w the 95 (or 99) % confidence limit of the predicted value, and the general relati ve risk model (12) was applied to assess the presence of interaction.
A nalysis of the lung function profiles. For the analysis of the lung fun ction profiles, the cases in each of the five impa ired profiles were compared with the refer ence category for differences in expo sure pre valences by means of the expo sure odds ratio. The following tests were applied: (i) a test for interaction and (ii) the estimation of att ributable risk. In the test for interaction the null hypothesis of the ab sence of an interact ion effect was postul ated in terms of the additivity of relative risks (13)(14)(15). In the test for the null hypothesis several types of exposure variables were employed. The dose-response trends in the odds ratios for each impairment and the individual du st levels were compared bet ween the non smokers, ex-smokers, and current smo kers. The indicator smo king variables, SMKI and SMK2, were employed to test whether the ex-smoker s and current smokers differed from the nonsmo kers in their do se-response trend for th e categorized dust variable GOLDPYC. The fully saturated additi ve relati ve risk model was of the form : where P was the probability of being a case, Q was a nu isanc e parameter; ( . ] estimated the odds ratio for where x, represented the dust variable (GOLDPYC or GOLDPY) and X z the smoking variable (SMOKEC or the exposed subjects relative to the subjects whose exposure was zero; the Pcoefficients estimated the excess relative risk due to each effect, coefficients pz and P 3 measuring the change in the intercept value caused by the ex-smoker and current smoker effect and P 4 and P s measuring the rate of change in the odds ratios with increasing dust level for ex-smokers and current smokers, respectively. The coefficients P4 and P s indicated whether tobacco smoking modified the effect of mining dust exposure. The likelihood ratio test for testing the null hypothesis of no interaction (ie, H o : P4 = P s =0 against H a : P4 and/or P s 7" 0) was given by X LR Z = -210g(LlL z ), where 10gL l and 10gL z were the maximized log likelihoods estimated from the model without interaction and with interaction, respectively. The multiplicative model (ie, the logistic model) with the same variables was also fitted to the data.
A test of interaction using either both categorized or both continuous exposure variables was also performed. The additive relative risk function was applied as SMOKEPY). The likelihood ratio test for testing the null hypothesis of no interaction (ie, H O:P3 = 0 against H a : Pd 0) was given by X LR z= -210g (LlL z ) as in the preceding test.
The computer program RISK (16) was used to fit the relative risk (equation 2), and the computer program MINUIT (17) was used to fit equation 1. The program MINUIT is a general minimization program and was used to fit the maximum likelihood function for grouped data, and, where zero values occurred, a value of 0.5 was added to each cell. For comparison SAS PROC CATMOD (18) was also applied to fit the additive and multiplicative models with grouped likelihood function to the data in table 4 in the Results section; the option ADDCELL =0.5 was used to replace zero cell values. The results obtained were comparable with those from MINUIT. The results obtained from the programs RISK and MINUIT were almost identical for the continuous and categorized variables.
In the estimation of attributable risk the method of calculation, as proposed by Miettinen (19) and Bruzzi et al (20), and applied by Kjuus et al (21) to the calculation of attributable risk for smoking and occupational exposure for lung cancer, was applied. The etiologic fraction for dust (EF d) was estimated as:

Results
Lung junction tests Figure I shows the mean values for the FEVl,o' VC, FEF zs _ 7S ' and FEVl,o/VC measurements plotted according to categorized dust particle-years for the nonsmokers, ex-smokers, current smokers of -s30 packyears, and current smokers of > 30 pack-years. The graphs display an interesting pattern. For FEF zs _ 75 the values for the nonminers (DUST = 0) are lower than those for the miners in the lowest exposure category (DUST = I). For all four variables there is a declining slope for the miners up to the dust level of 40 000 particle-years. If one considers the data up to that point, it can be observed that the decrease that occurs as the dust exposure increases is approximately parallel for all the smoking categories and that there is no indication that dust and smoking interact to increase the impairment. The consistent increase in the lung function measurements in the highest dust exposure category is presumably due to some systematic effect, possibly a healthy worker effect. The rate of decrease associated with increasing dust level in the region up to 40 000 particle-years is highest for FEF 2S _ 7S and lowest for ve. but the changes in FEVl.o and ve are almost parallel.
The results from the linear regression analysis done for the lung function tests confirmed the aforementioned observat ion, namely. that the interaction effect between dust and smoking is not statistically significant. The regression coeffi cients for dust particleyears/IOOO adjusted for the effect of age, height, weight, and smoking (pack-years) were -0.0067 (SE 0.0013) , -0.0052 (SE 0.0012) and -0.0118 (SE 0.0022) for FEVl.o, ve, and FEF 2S _ 7S (P<O.OOOI), respectively. For the dichotomized FEVl.o and ve variables (normal versus lower than the 95 or 99 0/0 confidence limit) the interaction effect was also not significant when tested by the additive model, even when the subjects with both measurements, FEVl.oand ve, lower than the 99 % confidence limit were compared with the rest of the subjects. Tables 1-3 show the distribution of the miners into the six lung function profile group s and their characteristics. The groups did not differ in age, but the two groups with restriction were slightly taller and those with the restr ictive profile were considerably heavier in compar ison with th e other group s. Tobacco consumption displayed an increasing trend with obstruction. The lung function values decreased, and the proportions of those having bronchitis and rhonchi increased as the level of obstruction increased. The dust particle-years increased up to moderate obstru ction only and then declined slightly. This pattern is reflected by the proportion of men with silicosis. The proportions of those who came for a benefit examination (because they were seeking compensation) and of tho se already certified also increased with the level of obstruction. The mortality from respiratory diseases showed a strong association with increasing obstruction. The restrictive profile group had the highest percentage of death from ischemic heart disease.

Lung f unction profiles
Test for interaction. Table 4 shows the distribution of the subjects in the six lung function profile groups according to smoking habits and dust exposure. In a comparison with the reference group, the observed odds ratios indicated an increase in the prevalence of each impairment as the dust and smoking exposure increased. There was some difficulty in selectingthe baseline groups for the calculation of the odds ratios. As the nonm iners (DUST = 0) had higher proportions of obstru ctive impair ments in all th ree smoking categories than the miners with the lowest dust exposure (DUST = I), they were excluded from further analyses, and the values of the miners with the lowest dust exposure (DUST = I) were used as the base line. The value of 0.5 was added to the cell frequencies for computational reasons; this procedure had the effect of increasing the prevalences among the nonsmokers, but it did not increase the dose-response trend among the smo kers . It ca n be seen that , for mo st of the profile gro ups, th e nonsmokers did not d isplay an increasing trend wit h an increase in du st exp osur e, while the exsmokers and cur rent smokers did , and the tre nd increased with the degree of obst ructive impairment. Figure 2 shows the observed odd s ratio s and the odd s ratios predicted by the additive and multiplicative mo dels for eac h ob structive imp ai rment. Th e odds ratios were plotted according to the smo king statu s and levels of du st pa rticle-years/lO 000. The predicted odds ratios for th e two mark ed ob stru ctive pro files were estima ted from th e ad dit ive mod el whic h inclu ded the interaction term . The multiplicative mod el did not include the interaction term as it was not fou nd to be significa nt. Table 5  of th e null hyp othesis of no interaction for the add itive model. Th e good ness -o f-fit test ( -2log L m ) is present ed fo r the multiplicat ive model fitt ed without an inte ra ctio n ter m. Becau se o f th e narro w ran ge, age was no t a sig nific an t pred ictor for a ny impa irme nt category . The regression coefficients estimated excess relative risk attributable to each exposure th at was abo ve the bas e-line risk, which was set at un ity. Thus, for example , the estimated odds ratio for the mar ked obstructio n wit h restrict ion impa irment for the cu rr ent smo ker with a value of 5 for      table 5 for the coefficients.) The interaction effect was found to be significant at the probability level of 0.05 for the combined marked-obstruction groups. As the interaction terms for the individual marked-obstruction groups were found to be significant for the other variable types, it is likely that adding the 0.5 value to the cell frequencies and combining the current smokers into one category decreased the interaction effect, which was significant at the probability level of 0.10.
As the values of the miners with the lowest exposure were used as the base line, the value of one was subtracted from the dust exposure to obtain the value of 1.0 for the odds ratio in the lowest exposure category; this procedure should not have influenced the strength of the association.  Table 6 summarizes the results of the likelihood ratio tests of the absence of interaction effect for the obstructive and restrictive impairments for the categorized and continuous exposure variables as included in equations 1 and 2. The interaction effect was not significant at the probability level of 0.05 for the minimal-and moderate-obstruction groups for any type of exposure variable. For the marked-obstruction groups (with and without restriction), the interaction effect was significant for the continuous and categorized variables. For the restrictive profile group the interaction effect between dust and smoking was not found to be significant.
Estimation of attributable risk. Table 7 shows the twoway.frequency distribution of all the marked-obstruction cases and of the reference subjects. The marginal odds ratios adjusted for each "other exposure" are shown. The odds ratios for the two-way cells show that the two highest smoking categories displayed the strongest dose-response trend with dust. The odds ratios in the two highest dust and smoking categories were approximately equal to the products of the marginal odds ratios and, therefore, indicated that the combined effect in these cells was almost multiplicative.
The etiologic fractions, estimated with equation 3, were 48 0J0 [95 0J0 confidence interval (95 010 CI) 40.2-55.2] for dust, 82 0J0 (95 0J0 Cl 76.2-87.0) for smoking, and 90 0J0 (95 0J0 CI 86.6-92.6) for the combined effect of dust and smoking. The preventable fraction indicated that approximately 42 0J0 of the cases could have been prevented by the elimination of tobacco smoking, 8 010 by the elimination of dust exposure, and 40 0J0 by the elimination of either dust or smoking. Table 8 summarizes the etiologic and preventable fractions. The value of the etiologic fraction is a reflection of the size of the marginal odds ratios for dust. It can be seen that the etiologic fractions for dust increase consistently as the degree of obstruction in-Tabl e 7. Etiologic fract ions f or the marked obstru ct ive prof iles com bined (marked obs t ruc tion + marked obstruction with rest ric tion). (C =cases, R =referents, OR =odds rati o, 95 % CI =95 % co nfidence in terval)   creases. For the two mark ed-ob struction gro ups the values were more than twice as high as that for the gro up with mod erate obst ruct ion, which had th e highest du st exposure and highest propo rtion of men with silicosis. Th us the results ind icate that heavy tobacco smoking potentiates the effect of du st on lung function impairment. The preventable fraction s showed tha t, as the severity of the obstructive impairment increased, smoking on its own an d the combined ef fects of smok ing and du st becam e mor e impo rtant. The etiologic fractions for the marked-obstruction profiles from which tho se persons that came for a benefit examination were excluded were found to be only slightl y lower than those obtained fro m all th e data (table 8).

Discussion
Th e results of th e present study indicate tha t for th e lung function measurements (FEV1.0' ve, and FEF 25 _ 75 )  However, in our study, when the miners were grouped according to their profile of lung function impairment, a synergistic effect between dust and smoking became apparent. The miners with the markedobstruction profiles were found to have the highest tobacco consumption and the strongest dose-response trend with dust; yet they did not have the highest dust exposure or the highest proportion of men with silicosis (table 3). The dose-response trend with dust was found to be stronger for the marginal odds ratios adjusted for smoking (see the etiologic fractions in table 8), and for the current smokers (table 4 and figure  2), the latter of which was found to be mainly due to those whose tobacco consumption was over 20 packyears (table 7). Table 7 shows the marginal odds ratios for dust for the two marked-obstruction groups combined. For the moderate-obstruction group the corresponding odds ratios for dust were 0.9 (95070 CI 0.4-1.9), 1.2 (95 % CI 0.6-2.5), 1.5 (95 % CI 0.7-3.3), and 2.0 (95 % CI 0.9-4.5). Table 7 also shows that the odds ratios in the four highest exposure categories were approximately equal to the product of the marginal odds ratios, a finding indicating a multiplicative effect. Figure 2 shows that, with increasing obstruction, the number of impaired lung functions among the nonsmokers decreased and the doseresponse trend with dust became progressively steeper for the current smokers.
The log likelihood ratio tests obtained from fitting the additive relative risk model to different types of exposure variables confirmed that the joint effect of dust and tobacco smoking on the prevalence of marked obstruction was higher than would be expected from the additivity of individual effects (tables 5 and 6). No interaction term was significant for the multiplicative model.
The etiologic fractions, estimated for each impairment profile in comparison with that of the reference group, also displayed an increasing trend with the degree of obstruction (table 8). Reducing the respirable dust to which miners are exposed in South African gold mines to a nonharmful level would theoretically have prevented the occurrence of 48 % of the impairments in the marked-obstruction profile groups (N = 402). Eliminating tobacco smoking would prevent 82 % of the impairments in these same groups and 94 % (95 % CI 87-100) of the impairment in the group with marked-obstruction plus restriction (N =191), which had the highest mortality from chronic obstructive lung disease (table 3).
The results from the present study are consistent with those obtained from the mortality study of these miners. In the mortality study, the effect of dust and smoking on mortality from chronic obstructive lung disease was assessed (24). The results showed a similar pattern, namely, that the death rates from chronic obstructive lung disease occurred among tobacco 420 smokers only and the dose-response trend with dust increased as the smoking level increased. The combined effect of the two exposures was close to being multiplicative. As the miners who died from chronic obstructive lung disease were mostly those who suffered from marked obstruction, the exposure patterns were similar.
The association between the restrictive profile and mortality from ischemic heart disease has been observed in several other studies (25), and in this study it may have been associated with obesity.
Possible biases in this study may be related to the selection of the nonmining referents , who were heavier smo kers [the average equivalent pack-years being 27.1 (SE 1.0) and 24.3 (SE 0.4) (P < 0.01) for the nonminers and miners, respectively] and had a higher prevalence of obstructive impairments. However, the nonsmoking nonminers also had a higher prevalence of obstructive impairments than the nonsmoking miners with the lowest dust level, and this circumstance could indicate either that the miners were fitter when selected originally or that a healthy worker effect was present. The use of the miners in the lowest dust category as the base-line group could have led to an underestimation of the effect of dust, but less so than if the nonminers had been used for this purpose.
A selection bias, whereby the fitter workers accumulate higher dust exposures, was probably involved also in this study and may have caused an underestimation of the effect of dust and of the synergistic effect. When we examined the cumulative particle-years by the decades (ie, the 1940s, 1950s, 1960s, and 1970s), the normal and minimal-obstruction profile groups consistently had the lowest exposure, whereas the groups with marked obstruction (with and without restriction) or restriction decreased their exposure mostly during the 1960s in comparison with the exposure of the moderate-obstruction group. Furthermore, a negative association between tobacco consumption and dust exposure was observed among the miners. The mean pack-years, according to the categories of dust particleyears, were 27.1, 25.2, 25.2, 23.2, and 23.6. This finding indicates that the miners who smoked less were fitter to continue to work in dusty occupations. Zwi (26) has also observed that gold miners doing the heaviest work with the greatest dust exposure had the best lung function, and more recently this observation was confirmed for black gold miners (27). This effect may have been responsible for the change in the doseresponse trend in the highest dust-exposure category, as shown in figure I.
The miners whose original spirograms were rejected because of inconsistencies, and consequently their lung function data were not recorded in the computer file, were compared with the miners who were included in the study. They were found to have a higher exposure to dust (in particle-years) up to 1970 (P=O.ool), but the difference in cigarette equivalent pack-years was not statistically significant. They had a higher mortality from all cau ses, 55 versus 46 0;0 (P < 0.0001), and from chro nic obstru ctive lung disease, 8 versus 3 % (P < 0.0001). Thu s the exclusion of these men from the origin al study and from thi s analysi s has very likely resulted in an und erestimation of the effect of dust.
The miners selected for thi s study could have come to the Medical Bureau for Occupational Diseases for a regular annual (so-called periodic) examin ation or a specialized (so-called benefi t) examination. A miner may attend a benefit examination one year and a periodical examination the next. By the age of 45 years many miners have already accumulated the service required for compensation and thus requ est the benefit examination to establish whether they quali fy for compensat ion. Table 9 shows the miners categorized according to the reason for attending the medical examination . The mean dust exposures, accumulated according to the individual decades, were higher and the lung fun ction measurements lower for th e benefit examinations than for the periodic examinations. No important dif ference s in age, bod y weight or height , bod y mass index, or tobacco consumption were observed between the fou r groups. Th e possibilit y that the men who came fo r their benefit examination could have biased the results of this study was considered . The anal ysis presented in table s 5 and 6 were repeated, and those who had attended the benefit examinations were excluded. The pat tern of the dose-respon se trend s and the regression results were very similar to tho se repor ted . Table 8 shows that the etiologic frac tions calculated for all mar ked obstruction changed only slight-Iy. Generally, the overall pa ttern was the same as that obt ained from all the da ta .
It can bee seen that the disability certifi cation (table 3), which leads to finan cial compensation, was relat ed to the amount of tob acco consumption (table  I). The moder ate-obstruction group , which had higher dust exposur e, a higher rate of silicosis, and lower tobacco consumption in comparison with the two mar ked-obstru ction groups had a lower rate of certifi-cation for chronic obstruc tive lung disease or pneumoconiosis than the latter two groups. Thus it appears that the heavily smoking miners had an ad vant age for compensation over the more heavily dust-e xposed miner s who smoked less.
The re are several po ssible reasons for the differ ent results obtained from the analysis of the individual tests and from the analysis of the profile groups with respect to the interaction effe ct. First , categorizing the men on the basis of their spirogram performance may have helped to eliminate misclassification errors for the disability. Second, ob structive impairment is a fun ction of FEV l.o and VC (28), and , if expo sure to dust causes parallel reduct ion in the FEVl.o and VC (29), then the FEV l.o/VC ratio, often used as a measure of obstru ction, may not change with increasing dust level. In ou r study the parallel reduct ion in airflow and lung volume was considered. Th e observed midexpiratory flow time , used in the grouping, was independent of the predicted values. Third, the impaired miners were compared with the miners who had " normal " lung function tests , and in relation to these miners the synergistic effect of dust and smoking became apparent. Finally, gro uping the men according to their lung function profiles produced more homogeneous groups with respect to exposure pattern .
It has been shown that the logistic model, often used for the analysis of dichot omous outcome variables, assumes multipl icat ivity of relat ive risks and thu s may not detect synergistic effects that are important from the point of view of public health (14). It has been proposed that, when the multiplicative model fits the data substa ntially better th an th e additi ve mod el without an interaction term, then a synergistic effect of public health importance can be assumed (15). The results from this study support the proposition. Figure 2 and the log likelihoods in tabl e 5 show that the additive model with interaction and the multiplicati ve model without interaction fit the data equally well. It has been shown that the parameter error estimates obtained from the additive model are unreliable for inference purposes and that the log likelihood ratio statistic for the absence of effect can be used to test the hypotheses of additivity or multiplicativity of relative risks (30).
In conclusion, it can be said that this study supports the hypothesis that tobacco smoking potentiates the effect of dust exposure in underground gold mining and increases the chances of developing a severe obstructive impairment in lung function that leads to increased mortality from chronic obstructive lung disease.