Ann. occup. Hyg., Vol. 46, No. 1, pp. 5-13, 2002
© 2002 British Occupational Hygiene Society
Published by Oxford University Press
Article |
The Combination of Effects on Lung Cancer of Cigarette Smoking and Exposure in Quebec Chrysotile Miners and Millers
1Department of Epidemiology and Biostatistics, McGill University, Montreal, Canada; 2Environmental Epidemiology Unit, London School of Hygiene and Tropical Medicine, Keppel Street, London WC1E 7HT, UK
Received 14 February 2001; in final form 1 June 2001.
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONAPPENDIXREFERENCES |
|---|
Although it is well known that both cigarette smoke and microscopic airborne asbestos fibres can cause lung cancer, evidence as to how these two agents combine is nebulous. Many workers have believed in the multiplicative theory, whereby asbestos increases the risk in proportion to the risk from other causes. However, evidence against this theory is mounting: a recent review concluded that the multiplicative hypothesis was untenable, and that the relative risk of lung cancer from asbestos exposure was about twice as high in non-smokers as in smokers, a finding largely independent of type of asbestos fibre. The criteria for entry to the current study were met by 7279 men in the 1891–1920 birth cohort of Quebec chrysotile miners and millers. The data consisted of date of birth, place of employment, smoking habit, asbestos exposure accumulated to age 55 and, for those 5527 who died between 1950 and June 1992, date and cause of death; 533 of the deaths were from lung cancer. For the principal analyses, ex-smokers were excluded from the study cohort, which comprised 5888 men, of whom 473 died of lung cancer. The conventional form of analysis is simply of the double dichotomy: non-smokers of cigarettes, ‘unexposed’ and exposed; all others, ‘unexposed’ and exposed. The respective standardized lung cancer mortality ratios (SMRs) were 0.29 and 0.62; and 1.37 and 1.72. Thus, the differences in relative risk, due to exposure, were closely similar, 0.33 and 0.35. On the other hand, the effects of asbestos measured by the corresponding ratios, 2.12 and 1.25, did differ, being 1.7 times as high in non-smokers as in others. The principal analysis was much more penetrating: the method was to fit models to a ‘disaggregated’ 6 x 10 array, by smoking habit (excluding ex-smokers) and asbestos exposure, of lung cancer SMRs. Both linear and log-linear models were fitted: the former included the additive and linear-multiplicative; the latter embraced the more conventional multiplicative form. The additive model fitted much the best. The fit of each multiplicative model was improved by the introduction of an interaction term that implied a less than multiplicative relationship. Thus smoking and exposure to chrysotile appear to have acted independently in causing lung cancer, with 10 cigarettes a day having an effect roughly equivalent to exposure amounting to 700 million particles per cubic foot x years. The refutation of the multiplicative hypothesis in these data reinforces its inapplicability in general; but the additive hypothesis is not generally applicable either. Indeed, there seems to be no good reason to believe that interactions conform to any simple theory. The implications are important.
Keywords: additivity; chrysotile asbestos; cigarette smoke; combination of effects; disaggregation; independent action; lung cancer; multiplicativity; relative asbestos effect; Quebec; synergy index
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONAPPENDIXREFERENCES |
|---|
It has long been accepted that both cigarette smoke and asbestos dust are potential causes of lung cancer, but the effects of the combination of these two agents remain in doubt, if only because assessment is a matter of great difficulty, dominated by the fact that the condition is rare in non-smokers.
Doll (1971) examined the very sparse data then available, from a single quite small study, in relation to two hypotheses about the way asbestos and cigarette smoking interact: ‘In one, it is assumed that asbestos produces the same additional risk in men who smoke cigarettes as in those who do not; in the other it is assumed that asbestos produces an effect that is proportional to the effect of the other agents’. The first of these, which has been termed the additive hypothesis (formulations of hypotheses are given below), explains the data less well than the second, but the fit to this, the multi-plicative hypothesis, is not particularly convincing. Nevertheless, no other specific hypotheses have been seriously considered since. In the first review of interactions (Saracci, 1977), covering three more studies, the indications were of still greater synergism. However, formal tests of significance did not indicate the need to reject the multiplicative hypothesis. Thus, the multiplicative hypothesis, although just one of infinitely many putative forms of synergism, was generally taken as ‘accepted’.
The simplest approach to describing the joint effect of asbestos and smoking on cancer risk is to examine risks, e.g. SMRs (SMR is the standardized mortality ratio D/E, where D is the number of deaths observed and E the number expected), in the following 2 x 2 array:
Non-smokers Smokers
Unexposed to asbestos SMRUN SMRUS
Exposed to asbestos SMREN SMRES
Departure from multiplicativity can be examined by methods introduced by Berry et al. (1985). These workers define the asbestos effect (AE) in non-smokers as the ratio of SMRs in asbestos-exposed non-smokers and unexposed non-smokers: AEN = SMREN/SMRUN; analogously, for smokers, it is: AES = SMRES/SMRUS. The relative asbestos effect (RAE) is defined as the ratio of the asbestos effects, i.e. RAE = AEN/AES. Under the multiplicative hypothesis, the asbestos effect in smokers is the same as that in non-smokers, so that RAE = 1. Departure from muliplicativity is measured by RAE: values less than and greater than one indicate interactions greater than and less than multiplicative, respectively. In a recent review, Liddell (2001) found that the estimate of RAE averaged over the seven cohort studies with adequate information was 2.04, with a 95% confidence interval (CI) of 1.28–3.25, indicating that the multiplicative hypothesis is not generally applicable.
Departure from additivity can be considered in terms of an index of synergy, S, due to Rothman (1976): S is the ratio of excess risk in asbestos-exposed smokers to that obtained by adding the excesses in non-exposed smokers and exposed non-smokers. For example, working from the same 2 x 2 table of risks,
S = [SMRES – SMRUN]/[SMRUS + SMREN – 2SMRUN].
On the additive hypothesis, S takes the value one; S > 1 indicates interaction greater than additive, while S < 1 indicates an antagonistic interaction. Erren et al. (1999) calculated the ‘weighted summary value’ of S from 12 studies as 1.66 (95% CI = 1.33–2.06), and concluded that the combined effect of smoking and asbestos exposure was greater than the sum of their separate effects. Despite concerns expressed by Liddell (2001) about several aspects of this report, it provides considerable evidence that the additive hypothesis also is not generally tenable.
For the large birth cohort of miners and millers of chrysotile in Quebec, relevant evidence has been presented concerning mortality before 1976 and 1976–88; in these two periods, there had been 11 and 14 lung cancer deaths in non-smokers, respectively, whereas in the other six cohort studies combined only 16 such deaths had occurred. Results of two forms of analysis of mortality up to 1975 were given by McDonald et al. (1980), but the substantive report was of a matched case–referent analysis (Liddell et al., 1984), in which the interaction was ‘closer to multiplicative than to additive’; Berry et al. (1985) calculated the RAE as 1.8, although with a wide confidence interval, and, in a return to subjective judgement, Saracci (Saracci, 1987) classified it as ‘near additive’. Reporting on mortality for 1976–88 in the same cohort, McDonald et al. (1993) gave RAE = 1.6. A summary of this evidence is given in Table 1, which also includes the values of Rothman’s S. For 1976–88, S suggests antagonism rather than synergism, although it too has a wide confidence interval.
|
Despite the wide differences that exist in any study group in smoking habits and in degree of occupational exposure, most attempts to evaluate the synergy have concentrated on the double dichotomy: smokers/non-smokers; exposed/not exposed to asbestos. While this has little importance if the interaction is indeed multiplicative, there is otherwise at least the possibility of ‘aggregation bias’; on the other hand, it is essential to distinguish non-smokers from all others. These points have been discussed by Liddell (2001).
The current paper examines the combination of the effects on lung cancer mortality of smoking and of asbestos exposure in the Quebec cohort from its inception until 1992. Seven categories of smoking habits and ten groups of exposure accumulated to age 55 are distinguished.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONAPPENDIXREFERENCES |
|---|
This investigation makes use of the materials detailed in a comprehensive review (Liddell et al., 1997) of the development of mortality, especially of that due to cancer, over almost 90 yr, from the inception of a large birth cohort of male chrysotile workers in Quebec. Only a brief description of relevant parts is given below; more detail can be obtained, if required, from Liddell et al. (1997), or from earlier reports referenced therein.
The criteria for admission to the cohort were birth between 1891 and 1920 and employment for at least 1 month in the Quebec chrysotile-producing industry. Almost 11 000 men were registered in November 1966, and they were followed from first employment (the earliest in 1904) to, effectively, 31 May 1992; copies of death certificates were obtained for the great majority of those who had died. Reports have been published of mortality to the following end-points: 1966; 1969; 1973; 1975; 1988; and 1992. In the latest report (Liddell et al., 1997), 10 918 men were under review, but 236 men who had not been employed in the industry before age 45, and so had probably had previous experience in mining, together with five men whose work histories were inconsistent, were eliminated, so that effectively the cohort consisted of 10 677 men.
As cause-specific reference rates were available only from 1950, we had to omit those who before then had been lost to view (1012, almost all after employment of only 2 or 3 months, usually before 1939) or had died (856), leaving 8809 men. Smoking habit had been elicited by questionnaire in 1970: of those alive then, all but a tiny handful responded personally; and proxies completed the questionnaire for over 90% of those who had died between 1950 and 1970. However, 891 questionnaires, mainly among those provided by proxies, were found to be unreliable, and these subjects also had to be disregarded—leaving 7918 men. Because exposures had been calculated up to age 55 (see below), each man’s ‘study interval’ (Liddell et al., 1977) had to be from age 55, and so we had to exclude the 630 men who died before that age; also nine whose work histories (see below) were unreliable. Thus, what we term the eligible cohort, described briefly in Table 2, contained 7279 men, of whom 5527 had died, 533 from lung cancer [SMR = 1.32 (Table 4), fully compatible with that of 1.36 in table 8 of Liddell et al. (1977)]. As explained below, it was necessary for most purposes to exclude the 1391 ex-smokers, reducing the study cohort to 5888 men, among whom there were 473 lung cancer deaths in a total of 4598 (with the SMR increased to 1.52; see Table 4).
|
|
The 7279 valid smoking histories were categorized into non-smokers of cigarettes, ex-smokers, and five classes of current cigarette smokers; however, for the preliminary analysis, they were dichotomized into non-smokers of cigarettes and all others. From the work histories which had been obtained from every man, asbestos exposure accumulated to age 55 was calculated, in terms of million particles per cubic foot x years (or mpcf.y); the men were then placed in ten broadly even classes. For the preliminary analysis a dichotomy was made at 30 mpcf.y.
The great majority of the men were first registered either at the main complex in the Eastern Township of Thetford Mines (43%) or at the mine and mill in Asbestos, another of the Eastern Townships (41%). The former was considerably the dustier, and contamination by tremolite much more substantial; thus it is desirable to analyse data for these places separately. The other 16% were employed in seven much smaller operations, none large enough for separate analysis; nor could they be considered as one entity, because they were highly variable in many ways, in particular in dustiness.
Smoking habit depended markedly on when the subjects were born: nearly half of those born between 1891 and 1895 were non-smokers, and less than one-quarter smoked 16 or more cigarettes a day; these proportions changed steadily and dramatically with each quinquennium of birth, so that 10% of those born 1916–20 were non-smokers and almost half of them smoked more than 16 cigarettes a day. There were corresponding increases in the proportional mortality ratios for lung cancer, namely from 3.9% to 19.0% (see Table 2), by quinquennium of birth. It was therefore important to discover whether findings were affected by this phenomenon.
Just half the men had been exposed by age 55 to less than 60 mpcf.y, one-third to 60 but less than 400 mpcf.y and one-sixth to at least 400 mpcf.y. These proportions were fairly similar for each of the seven smoker classes, varying between 44% and 56%, between 29% and 39%, and between 15% and 20%. Smoking habits varied only slightly by place of employment.
The principal analyses were by Poisson regression (Armitage and Berry, 1994), the data for which consist of observations (Delwiche and Slaughter, 2000), one for each subdivision of the study cohort by smoking category and exposure category, and containing the numbers of lung cancer deaths (D) observed and expected (E) in the subdivision. The values of D and E were obtained for each sub-division by the subject-years method (Berry, 1983), using the PYRS program (Coleman et al., 1989) with, as reference, death rates for the male population of Quebec in 16 age groups, by quinquennia from 1950. As previously, and following usual practice, each D/E ratio is termed a standardized mortality ratio (SMR); the Appendix shows, for the eligible cohort, the numerator and denominator of the SMRs in the 10 x 7 array after categorization by both asbestos exposure and smoking habit. For any further subdivision, the correponding array had to be obtained separately for each sub-category, e.g. for each place of employment.
The models fitted are described below. The linear models have the equation:
SMR = 
+ ß · s + 
· x + 
· s*x
where s and x represent, respectively, the number of cigarettes smoked a day and mpcf.y, in each subdivision of the cohort, and s*x is their product; and where , ß, and are four fitted regression coefficients, or parameters. Reduced to three parameters by omitting the last term, the model becomes
SMR = + ß · s + · x
which is clearly additive. When = (ß)/, expression (1) reduces to
SMR = (1 + [ß/] · s) · (1 + [/] · x)
or
SMR = (1 + ß* · s) x (1 + * · x)
In this form, with the three parameters , ß* and *, the model is called linear-multiplicative.
The log-linear models have the equation:
SMR = exp(a + b · s + c · x + d · s*x)
or
SMR = exp(a) x exp(b · s) x exp(c · x) x exp(d · s*x)
Omitting d and writing k for exp(a), this becomes
SMR = k · exp(b · s) x exp(c · x)
which is the generally accepted, three-parameter, form of the multiplicative model.
The GENMOD procedure of the SAS/STAT© Software (SAS Institute Inc., 1997) is controlled by statements (Delwiche and Slaughter, 2000), each of which produces the parameters of the fitted model, together with the deviance (see below). For the full linear model (equation 1), the statement is:
model D = E V W Z/dist=poisson link=identity noint;
with new variables: V = E*s; W = E*x; and Z = E*s*x. [The need for the new variables, and the meanings of ‘link’ and ‘noint’, are explained elsewhere (Hanley and Liddell, 1985).] The analysis can proceed in stages, by including, first, E alone; then E and V; E, V and W; and lastly, E, V, W and Z. The first stage provides the overall SMR (and the deviance before any modelling); the third and fourth stages fit models (2) and (1), respectively. The corresponding SAS statement for the basic log-linear model, (4), is
model D = h s x p/dist=poisson link=log offset=q;
in which h is any numerical constant, and p and q stand for the product s*x and for ln(E). The modelling proceeds in analogous stages: the first produces the identical overall SMR and initial variance; the last two fit models (5) and (4), respectively. The linear-multiplicative model, (3), needs the constraint = (ß)/, which requires a non-linear procedure (Hanley and Liddell, 1985) not in GENMOD, so EPICURE (Preston et al., 1993) was used to fit this model, with associated deviance.
In a model that fits well, the deviance is distributed approximately as 
2 with the specified degrees of freedom (df), so that deviances much above df indicate poorly fitting models; at any stage, improvement in fit can be assessed by the reduction in deviance, treated as a 2 statistic which has 1 df for each new variable introduced (Armitage and Berry, 1994).
Model (1) may be thought of as an extension of the additive model, (2); the reduction in deviance assesses the importance of the parameter in causing departures from additivity. Correspondingly, model (1) may also be considered as an extension of the linear-multiplicative model, (3); the reduction in deviance assesses the importance of { – (ß)/} in causing departures from multiplicativeness.
As preliminaries to the principal analyses, we calculated SMRs for the complete eligible cohort, and for each of the four sections of the double dichotomy. We also examined the relationships in the study cohort between lung cancer SMRs and, separately, the amount smoked and the degree of asbestos exposure. In the basic analysis, all five models, (1) to (5), were fitted to the data for all 5888 men in the study cohort. Thereafter, we excluded also the 905 men first employed at the various smaller operations and analysed the data for all 4983 remaining; also for the same men according to place of employment and to decade of birth.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONAPPENDIXREFERENCES |
|---|
Preliminary results
Amongst all 7279 men, lung cancer was the cause of 533 deaths by 31 May 1992, while the number expected on the basis of the mortality of the male population of Quebec was 405.23; thus, the SMR was 1.32. For the double dichotomy already described, the four SMRs shown in Table 3 were obtained—without resort to modelling—and both RAE and S were calculated from them; each index lies between the values in Table 1. Although the confidence interval around RAE is narrower than for either of the RAEs in Table 1, the multiplicative hypothesis cannot be rejected on this evidence. The synergy index is, however, remarkably close to the value the additive hypothesis would produce; yet it too has a wide confidence interval.
|
For each of the seven classes of smoking habit, the average number of cigarettes smoked a day and the values of D, E and D/E, are listed in Table 4 (the values can be obtained as the column totals of the appendix table). Also given are the SMRs calculated from the fitted linear and log-linear models; the corresponding deviances, both with 4 df, were 6.7 and 22.1, respectively. Although the linear fit is not particularly good, the log-linear is very poor.
Correspondingly, Table 5 lists the observation—mean exposure, D and E—for each of the ten groups of exposure accumulated to age 55, together with the values of observed and fitted SMRs. The deviances after fitting the linear and log-linear models were, respectively, 5.2 and 5.8, each with 8 df; both fits are remarkably good, and there are no grounds for choice between them.
|
Findings
The results of the principal analysis are in Table 6. Model (1) yielded the lowest deviance, but as this was only 0.3 less than that for model (2), at the expense of an extra degree of freedom, the fit of the latter was in fact the better. In other words, far from causing departures from additivity, the parameter
damaged the additive fit. The extension of model (3) to model (1) reduces the deviance by 4.4, indicating that { – (ß)/} caused significant departures from multiplicativeness. The fact that the deviance associated with model (4) was 3.2 less than that for model (5) is suggestive of improvement due to the introduction of the product term; however, the parameter d was negative, so that the modelled interaction was less than multiplicative. None of the log-linear models fitted as well as the linear models.
|
The parameters in the additive model, (2), are presented, with standard errors, in Table 7. The analyses were repeated for the two principal complexes combined, but excluding all the smaller places of employment: as was to be expected, the results were closely similar to those in Tables 6 and 7. Further subdivision by place of employment or by decade of birth led to rather less clear-cut patterns, but there was no sign that the deviances associated with the additive model could be reduced significantly by any other model. Further, for these analyses, all the values of the parameters
and ß are compatible with those in Table 7. A test of the difference between the values of for the two places of employment yields P = 0.094; the other values of are satisfactorily homogeneous.
|
The risks of lung cancer due to the combination of smoking and exposure to chrysotile, calculated from the fitted linear additive model applied to the complete cohort, apart from ex-smokers, and expressed relative to the risk for a male smoker of 20 cigarettes a day in the Quebec population, are shown in Table 8. Although relative asbestos effect (RAE) was originally defined for evaluating the combination of effects in a double dichotomy, it can also be calculated in the current circumstances; the relevant values, given in brackets in Table 8, are seen to vary greatly with the levels of both smoking and exposure. However, no matter what levels are used to create a double dichotomy, the value of S will be unity—indicating additivity—as, of course, determined by the model.
|
In a subsidiary analysis to take account of ex-smokers, ‘dummy variables’ differentiating all seven smoking habits defined in Table 4 were created; these replaced the single variable, average number of cigarettes smoked a day, in analysis which now could be carried out for all 7279 men. The linear model fitted well (deviance = 64.1 with 62 df). The parameters for all six classes with known current consumption were in monotonic progression, roughly proportional to average cigarettes a day; that for ex-smokers fell between those for non-smokers and the lightest smokers (4.4 cigarettes a day). Thus, it appears not unreasonable to estimate ex-smokers’ risks by assuming a notional consumption of the order of 2 cigarettes a day. Estimates of risk, corresponding to those in Table 8, could then be found by using the same values of
, ß and . Such estimates would, of course, be averages without regard for the important factors of how many cigarettes an ex-smoker had smoked, for how many years, nor of when he had abstained. Nevertheless, there is strong confirmation that quitting smoking decreases greatly the risk of lung cancer, while the combination of risks conforms to the general pattern. |
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONAPPENDIXREFERENCES |
|---|
Despite the necessary exclusions, this is a large study involving over 102 500 subject-years of observation of former asbestos workers; it is unique in that for each worker we had measures both of the number of cigarettes he smoked and of his cumulative exposure. Nevertheless, the conventional approach to assessment of the combination of the effects of smoking and asbestos, through the double dichotomy—non-smokers and others; very lightly and more severely exposed—yielded equivocal results. The relative asbestos effect was similar to those observed in relation to mortality before and after 1976, differing from unity, which would have indicated multiplicativity, but not ‘significantly’; and, although the synergy index was midway between the two earlier values and therefore close to the value given by additivity, it also had a wide confidence interval.
The need for ‘disaggregation’ (see Liddell, 2001) of both factors is obvious. Increasing the numbers of levels, from two for each, to seven for smoking and 10 for exposure increased the number of degrees of freedom from one to 69; using the daily number of cigarettes as the measure of smoking forced the exclusion of ex-smokers from the main analyses, and df became 59. In other words, there was a 6 x 10 array of observations, each consisting of the observed number of lung cancer deaths (D) and the number expected (E) from provincial mortality. Even when a model with three parameters (in addition to or A) was fitted, the residual deviance had 56 df.
In the data for the 5888 men in the study cohort, the fits of the log-linear models for lung cancer risk—given history of smoking and asbestos exposure—were very poor relative to those of the linear model. The main reason for this is that the relationship between number of cigarettes smoked and lung cancer risk was linear rather than log-linear. The relationship between cumulative asbestos exposure and risk also fitted a linear model well, although the fit of the log-linear model was quite insignificantly worse. Among the three linear models, the linear multiplicative fitted worst; the fit was improved by inclusion of a fourth parameter, which changed the model away from the multiplicative towards, and very close to, the additive, and the fit of the simple (three-parameter) additive model was even better. In summary, the multiplicative hypothesis, whether expressed in log-linear or linear terms, can be discarded, and there is strong evidence that the effects of smoking and of asbestos combined additively; in other words, the effects were independent, or very nearly so.
The disaggregation permits the calculation of the equivalent of the relative asbestos effect for any specific number of daily cigarettes (s) and exposure (x), as
RAE = 1 + (ß · s)( · x)/()( + ß · s + · x)
This form brings out the dependence on the levels of both smoking and exposure that have been adopted, as is illustrated in Table 8. Any double dichotomy can be simulated, and the corresponding RAE calculated: for instance, the levels in Table 3 (for which s = 0 and 15.0, and x = 8.3 and 367) yield the ‘modelled’ RAE as 1.65; its closeness to the observed value, 1.69, provides further confirmation of the model.
The separate analyses for the two main places of employment and for the three birth decades required subdivision of the (already disaggregated) data, and this led to considerable statistical instability. In other words, the numbers in the ‘average’ observation were so small, despite the great size of the cohort, that the 2 approximation to the distribution of each deviance became unreliable (Armitage and Berry, 1994), and that results could be seriously affected by a very few misclassifications. The evidence in favour of additivity was substantiated by most of these analyses.
This evidence is in accord with the finding by Liddell (2001) that the multiplicative hypothesis for the combination of the effects on lung cancer of smoking and asbestos is untenable in general. The implications—for, at the least, research on mechanisms, protection, attributability and compensation—are beyond the scope of this paper; it should be noted that they are likely to be difficult to elucidate, especially as additivity is certainly not a universal alternative (see Erren et al., 1999).
A spin-off, however, is that a rough comparison of the two hazards can be made from the values of ß and in Table 7. Thus, the SMR for a non-smoker exposed to 100 mpcf.y was increased by 0.102, an excess achieved by an unexposed man who smoked 0.102/0.0727 = 1.4 cigarettes a day, or one pack in a fortnight. Equally, the SMR for an unexposed man who smoked 10 cigarettes a day was increased by 0.727; for a non-smoker’s SMR to be as high, his exposure would have had to be 0.727/0.00102, or >700 mpcf.y (which corresponds to a working life of 50 yr exposed to an average of 14 mpcf, or very roughly 40 fibres/ml). This ‘equivalence’ is of the same order as 10 daily cigarettes to 1100 mpcf.y, which can be inferred from figure 4 in McDonald and Liddell (1979). It may, however, be slightly inflated: the proportion of lung cancers among non-smokers of cigarettes is somewhat higher than might have been expected in the general population and it is possible that some of these men were in fact smokers, if only of pipes or cigars. Nevertheless, the proportion was no different in subjects with proxy responses than in those responding directly, and it seems likely that the effect on our primary analyses was small, for it would have increased the ‘cigarettes’ variable from zero to a very few.
Because each ‘equivalence’ is obtained as a ratio of estimated parameters, it is of low accuracy. Just as the values of the three parameters were similar in all analyses but one, the corresponding ‘equivalences’ can also be considered similar. However, at the mine and mill in Asbestos, the low value of leads to a rather higher ‘equivalence’. This is counterbalanced by a considerably lower ‘equivalence’ at the main complex in Thetford Mines, probably due to the much higher contamination by tremolite.
A final comment: the purposes of Table 8 were to show how asbestos and smoking combined, in these data, in affecting lung cancer risk, and how the relative asbestos effect depends on where the dichotomies are made; the risk estimates are not intended for use in setting control limits.
Acknowledgements—Sincere thanks are due to all members of the team led by Corbett McDonald who, over more than three decades, carried out the mortality investigation of the Quebec cohort, initiated in 1964.
|
APPENDIX |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONAPPENDIXREFERENCES |
|---|
9
|
|
FOOTNOTES |
|---|
* Author to whom correspondence should be addressed.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONAPPENDIXREFERENCES |
|---|
Armitage P, Berry G. (1994) Statistical methods in medical research, 3rd edn. Oxford: Blackwell Scientific Publications.
Berry G. (1983) The analysis of mortality by the subject-years method. Biometrics; 39: 173–84.[Web of Science][Medline]
Berry G, Newhouse ML, Antonis P. (1985) Combined effect of asbestos and smoking on mortality from lung cancer and mesothelioma in factory workers. Br J Ind Med; 42: 12–18.[Web of Science][Medline]
Coleman MP, Hermon C, Douglas L. (1989) Person-years (PYRS): a Fortran program for cohort study analysis. IARC Internal Report No. 89/006. Lyon: International Agency for Research on Cancer.
Delwiche LD, Slaughter SJ. (2000) The little SAS book: a primer, 2nd edn. Cary, NC: SAS Institute Inc.
Doll R. (1971) The age distribution of cancer: implications for models of carcinogenesis. J Roy Statist Soc A; 134: 133–66.
Erren TC, Jacobsen M, Piekarski C. (1999) Synergy between asbestos and smoking on lung cancer risks. Epidemiology; 10: 405–11.[Web of Science][Medline]
Hanley J, Liddell D. (1985) Fitting relationships betwen exposure and standardized mortality ratios. J Occup Med; 27: 555–60.[Web of Science][Medline]
Liddell FDK. (2001) The interaction of asbestos and smoking in lung cancer. Ann Occup Hyg; 45: 341–56.
Liddell FDK, McDonald JC, Thomas DC. (1977) Methods of cohort analysis: appraisal by application to asbestos mining. J Roy Statist Soc A; 140: 469–91.
Liddell FDK, Thomas DC, Gibbs GW, McDonald JC. (1984) Fibre exposure and mortality from pnemoconiosis, respir-atory and abdominal malignancies in chrysotile production in Quebec, 1926–75. Ann Acad Med Singapore; 13(2 suppl.): 340–4.
Liddell FDK, McDonald AD, McDonald JC. (1997) The 1891–1920 birth cohort of Quebec chrysotile miners and millers: development from 1904 and mortality to 1992. Ann Occup Hyg; 41: 13–36.
McDonald JC, Liddell FDK. (1979) Mortality in Canadian miners and millers exposed to chrysotile. Ann NY Acad Sci; 330: 1–10.
McDonald JC, Liddell FDK, Gibbs GW, Eyssen GE, McDonald AD. (1980) Dust exposure and mortality in chrysotile mining, 1910–75. Br J Ind Med; 37: 11–24.[Web of Science][Medline]
McDonald JC, Liddell FDK, Dufresne A, McDonald AD. (1993) The 1891–1920 birth cohort of Quebec chrysotile miners and millers: mortality 1976–88. Br J Ind Med; 50: 1073–81.[Web of Science][Medline]
Preston DL, Lubin JH, Pierce DA. (1993) EPICURE user’s guide. Seattle, WA: Hirosoft International Corp.
Rothman KJ. (1976) The estimation of synergy or antagonism. Am J Epidemiol; 103: 506–11.
Saracci R. (1977) Asbestos and lung cancer: an analysis of the epidemiological evidence on the asbestos-smoking interaction. Int J Cancer; 20: 323–31.[Web of Science][Medline]
Saracci R. (1987) The interactions of tobacco smoking and other agents in cancer etiology. Epidemiol Rev; 9: 178–93.
SAS Institute Inc. (1997) SAS/STAT® software: changes and enhancements through release 6.12. Cary, NC: SAS Institute Inc.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  D. Wraith and K. Mengersen A Bayesian approach to assess interaction between known risk factors: the risk of lung cancer from exposure to asbestos and smoking Statistical Methods in Medical Research, April 1, 2008; 17(2): 171 - 189. [Abstract] [PDF]
|
|
|
|
 |
|
 M Y N C K Cheyip, G Nelson, M H Ross, and J Murray South African platinum mine employees reduce smoking in 5 years Tob. Control, June 1, 2007; 16(3): 197 - 201. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 A Reid, N H de Klerk, G L Ambrosini, G Berry, and A W Musk The risk of lung cancer with increasing time since ceasing exposure to asbestos and quitting smoking. Occup. Environ. Med., August 1, 2006; 63(8): 509 - 512. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M. R. Cullen, M. J. Barnett, J. R. Balmes, B. Cartmel, C. A. Redlich, C. A. Brodkin, S. Barnhart, L. Rosenstock, G. E. Goodman, S. P. Hammar, et al. Predictors of Lung Cancer among Asbestos-exposed Men in the {beta}-Carotene and Retinol Efficacy Trial Am. J. Epidemiol., February 1, 2005; 161(3): 260 - 270. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
F D K Liddell and P Lee Joint action of smoking and asbestos exposure on lung cancer * Author's reply Occup. Environ. Med., July 1, 2002; 59(7): 494 - 495. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||