Tobacco Documents Online

The Asbestos example Prof. J. Corbett McDonald

Linear extrapolation for risk estimation at low level exposure: the asbestos example J Corbett McDonald MD FRCP, Professor Emeritus National Heart & Lung Institute London University and School of Occupational Health McGill University, Montreal 1. Introduction Asbestos exposure can cause two kinds of malignant disease - cancer of the bronchus (lung cancer) and primary tumours of the pleura or peritoneum (mesothelioma). It remains uncertain whether asbestos can cause lung cancer in the absence of tobacco smoke or other co-carcinogen and also whether pure chrysotile or only certain amphibole fibres - crocidolite, amosite and tremolite, in particular - can cause mesothelioma. These doubts lie at the heart of all etiological questions on asbestos and carcinogenecity, and cannot be avoided when considering the validity of extrapolation. Equally unavoidable is the concept of a threshold or so called "safe" level of exposure. Conceptually and biologically both linear and non-linear relationships - sigmoid for example - may or may not have a threshold. Although the four possibilities are probably indistinguishable, threshold and non-threshold models provide quite different estimates of risk at very low exposure levels. 1

2 The measurement of response in terms of morbidity or mortality in response to toxic or carcinogenic agents seldom presents a serious problem but concepts of dose and exposure are f ar more difficult to deal with. In human populations dose is very seldom known or usable in epidemiological studies. Exposure as a surrogate for dose is inevitably more complex and requires more precise definition, qualitatively and quantitatively, than is ever possible. With asbestos fibres, for example, there are reasons to believe that their biological activity will depend on mineralogical type, fibre dimensions, airborne concentration and pattern of exposure in and over time. Of the dozen or so cohort studies of asbestos workers in which an attempt has been made to assess exposure for each cohort member all were based on dust counts, not fibre concentrations; fibre size distributions were not known and reliable information on fibre type was generally lacking. In every study exposure has been expressed as the product of duration and intensity (cumulative exposure), implying the biologically unlikely assumption that these two variables, one fairly precise and easily measured and the other vague and approximate, are interchangeable. It should thus be evident that the carcinogenic risk associated with the inhalation of asbestos is an extremely complex multifactorial problem requiring the appropriate statistical analysis of individual exposure in terms of fibre type, size, concentration, duration and timing together with comparable information on smoking habit. Small wonder that present views on the subject are essentially speculative; however, although health policies

3 and decisions cannot wait for certainty, note should be taken of the nature, size and implications of possible error. 2. Exoosure restionse 2.1 The epidemiological data Present opinion is based largely on findings from a relatively small number of cohort studies of lung cancer mortality in workers exposed occupationally. The main studies are summarized in Table 1(1); as clearly stated, exposure was expressed in them all as the product of duration and intensity, the latter based on dust particle counts. The slopes shown in the last column are for lines fitted to relative risks derived from observed SMRs on the assumption of a linear non-threshold model. In fact, straight lines do not necessarily provide the best fit and, as shown in Figure 1, none of the lines describing the SMRs themselves actually pass through the origin (2). Nor do the slopes listed in Table 1 or shown in Figure 1 take any account of smoking; indeed, only in one was smoking habit known. The simple view is often taken that as the interaction between asbestos and smoking can be assumed to be multiplicative, linearity will not be affected by the contribution of either factor. However, in six studies where it has been possible to study the interaction (3), only one showed the relationship to be multiplicative and that in the cohort of American insulation workers where asbestos exposure was not quantified (see Table 2). If, as seems somewhat more probable, the interaction is generally more than additive but less than ~ multiplicative, the linear hypothesis for asbestos per se is t.n ~ ~ ..~ further undermined. -~1 ~ N -~ W

4 Just as the epidemiology of lung cancer is dominated by smoking, that of mesothelioma is strongly affected by fibre type; as a result the quantitative information available on exposure- response for the latter is scanty. Several studies (4) suggest that the risk is related to duration of exposure and that few if any cases occur within less than several months of first exposure for the amphiboles or several years for commercial chrysotile*. Indirect evidence of a systematic relationship amphiboles is afforded by analyses of lung tissue at autopsy f rom mesothelioma cases and controls in Canada (5) and for Australia (6). The nature of this evidence, the interpretation of which entails several assumptions, is illustrated in Figures 2 and 3. 2.2 Statistical extraoolations Until fairly recently occupational epidemiology was primarily concerned with identifying health hazards at work and with providing a rationai basis for setting a 'threshold limit value' or other hygienic control measure sufficient to reduce risks to an acceptable level. More recently, concern has grown about the possibility of risks to the general public at exposure intensities well below those found in the workplace. Underlying this concern is the fear, probably related to experience with ionizing radiation, that there might be no safe threshold for carcinogens. As occupational epidemiology is not able to N Ln v ~ .~ -.~ * the term 'commercial' is used to indicate that the chrysotile ~ N -.~ is f requently contaminated with fibrous tremolite. ~

5 provide a direct answer to this type of question, several valiant attempts have been made to obtain some measure of possible risk by linear extrapolation. The assumptions underlying these efforts are very large and subject to many uncertainties which can be considered under three headings:- a) ,Exoosure-resoonse The nine cohort studies in eight industrial groups summarized in Table i show that the difference in gradient between those for textile and friction product workers was about 50 fold. The experience of American insulation workers and of men engaged in the manufacture of amosite insulation products, are not shown in the Table because exposure was not assessed individually. However, with certain assumptions, especially as to linearity, it seems likely that the gradients for these two groups lay somewhere between those for cement workers and textile workers. There are at least two possible explanations for the variation. First, some of the exposure estimates may have been seriously incorrect; if so, the error was systematic or the response relationships would have been lost. Second, neither the original dust particle measurements nor the usual conversion to fibres, countable with the optical microscope, may adequately reflect the biological hazard; experimental work on fibre size and the dynamics of penetration and retention all suggest that this could be an important part of the explanation, perhaps all of it. 0 b) Fibre tyQe Differences between the various types of ~ -~.1 ~ asbestos fibre can probably be ignored in predicting risks of ~ Crt

6 lung cancer and asbestosis, but mesothelioma is another matter. The evidence that virtually all peritoneal and most pleural cases are attributable to amphibole exposure, rather than to chrysotile, is strong though not conclusive. The uncertainty is compounded by the lack of adequate exposure-response information for mesothelioma. In none of the nine cohorts shown in Table i was the relationship of inesothelioma to exposure examined. Despite this, some reports have suggested that an indication of risk can be obtained from a small number of other cohort studies, in which only average group exposure had been roughly estimated. All the cohorts used for these estimates entailed exposure to pure amphiboles or to amphibole-chrysotile mixtures; generally excluded f rom consideration were those in which the mesothelioma risk was low. c) Conversion All the available exposure-response data f rom occupational cohorts are based on total respirable dust measurements. Determination of the equivalence of these measurements in terms of fibres (>5 µm long) per millilitre (f/mL) is a difficult and dubious operation. Even in chrysotile mining and milling, the range of conversion ratios is at least 40-fold. A problem of similar magnitude concerns the equivalence in fibre terms of measurements made in the general environment, nearly all of which are gravimetric and usually expressed in nanograms per cubic metre (ng/m3). The conversion factor relating mass to optical fibre concentration had a range of 5 to 150 and probably varied with fibre type (1).

7 On taking these three types of uncertainties into account, the possible error in any estimate made by extrapolation could range over five orders of magnitude. Even this would not take account of such questions as sampling error in environmental measurement, fibre type, or fibre size distributions. In a paper by Enterline in 1981 (7), estimates of lung cancer deaths, based on extrapolation from linear and curvilinear exposure-response models, were made. Using conversion factors of 3 for f/mL per mpcf and 40 x 103 for f/mL per ng/m3, and linear extrapolation from his own exposure-response data (SMR = 100 + 0.658 mpcf-yr), he estimated that continuous lifetime exposure at 5 ng/m3 (approximately the average outdoor level in urban areas of the US) would result in 4.6 lung cancer deaths . per million population. On the other hand, a curvilinear model, for which there was experimental but not epidemiological support, would result essentially in zero deaths. Several other estimates of current and lifetime risk of lung cancer and mesothelioma for the US population have been made purely by extrapolation. A simplified comparison of these estimates is set out in Table 3. To achieve a measure of comparability, some liberties were taken with the published data, and the figures shown are therefore approximate. The differences between the lung cancer estimates are mainly due to the idiosyncratic selection of exposure-response data from industrial cohorts. The NRC committee used three of the nine cohorts included in Table 1 and added six others, in all of

which only group estimates of exposure had been made. Schneiderman used only two of the nine and included three of the six added by the NRC committee. Nicholson used four of the nine cohorts and not the other five. Other estimates of lifetime risk associated with non- occupational chrysotile exposure were made by Doll & Peto (8) who calculated that exposure for 40 hours a week for 20 years would result in 10 excess deaths per million population. Hughes & Weill (9) published similar estimates for school children and asbestos cement production workers (see Table 4). 2.3 Validity of risk estimates Any hope of being able to validate estimates of mortality risk of the magnitude shown in Table 3 and Table 4 by any form of planned epidemiological survey would appear quite impossible. There are so many other known and unknown confouhding factors which affect human health to a greater degree than very low level asbestos exposure, the allowance could not be made for these even in the largest conceivable prospective or retrospective study. Nevertheless there are some data which throw light on the problem. On the basic question of the linear-exposure-response model, exploratory analyses of several sets of cohort data (10,11) have now been made by multivariate relative risk methods. These have shown that the pattern of risk in relation to exposure intensity differs substantially from that with cumulative exposure and 8

9 might well be sigmoid rather than linear; further work on these lines is in progress. More direct evidence is afforded by observed trends in mesothelioma mortality in Canada, the United States, Australia, Great Britain and Scandinavia. In all these countries, with a combined population approaching 400 million, the incidence in males and females began to separate in about 1950. Since then there has been a steady upward trend of about 10% per annum in men but little evidence of an increase in women. These national trends in mesothelioma mortality were reviewed in detail in the recent comprehensive report by the Health Effects Institute - Asbestos Research (10); their main conclusion, summarized below, expresses the implications of these observations very well:- "The risk assessment model ...... predicts that the number of background (that is, not asbestos-related) mesothe7iomas in the United States might be increased by 10, from approximate7y 400 per year (200 each in men and women) to about 410 per year, if the whole population were exposed for 20 years in buildings to the a verage level of 0.0002 f/mL....., or for 13 years to the average leve l of 0.0005 found for schoo Is. Th is sma ll increase would not be - detectable by an analysis of national age-specific trends in mesothelioma incidence nor could such an analysis confirm the risk assessment mode7........ . The data do, howe ver, i nd i ca te tha t the overa l l r i sk to the popu l a t i on for mesothelioma in buildings is 7ikely to be smal7er in comparison