Document BXRnqq6go5mDmJR86RY1K0xL

W* .f*. * * i - ' 4km. %,, Vul. 44. Nrt. t. (V- 2000 Cmwi Copyrtft** 6 2000 puMiAnf bf EUcvkr Science U4 oo hetulf of BriiUh Ooj>ikwul Hygiene Soviet/ PU: S0003-4878(00)00045-4 *" "" re*CTV"t The Quantitative Risks of Mesothelioma and Lung Cancer in Relation to Asbestos Exposure JOHN T. HODGSON* and ANDREW DARNTON Epidemiology and Medical Statistics Unit, Heatth and Safety Executive, Magdalen House, Stanley precinct, Bootle L20 3QZ, UK ,, Mortality reports on asbestos exposed cohorts which gave information on exposure levels from which (as a minimum) a cohort average cumulative exposure could be estimated were reviewed. At exposure levels seen in occupational cohorts it Is concluded that the exposure specific risk of mesothelioma from the three principal commercial asbestos types is broadly in the ratio 1:100:500 for chrysolite, amosite and crocldolite respectively. For lung cancer the conclusions are less clear cut. Cohorts exposed only to crocldolite or amosite record similar exposure specific risk'levels (around 5% excess lung cancer per t/ml.yr); but chryso lite exposed cohorts show a less consistent picture, with a clear discrepancy between the mortality experience of a cohort of chrysotile textile workers in Carolina and the Quebec miners cohort Taking account of the excess risk recorded by cohorts with mixed fibre . exposures (generally<l%), the Carolina experience looks uptypically high. It is suggested that a best estimate lung'cancer risk for chrysotile alone would.be 0.1%, with a highest ' reasonable estimate of 0.5%. The risk differential between chrysotile and the two amphibole fibres for lung cancer is thus between 1:10 and 1:50. ' Examination of the inter-study dose response relationship for the amphibole fibres suggests a non-linear relationship for all three cancer endpoints (pleural and peritoneal mesothell- . ' omas, and lung cancer). The peritoneal mesothelioma risk is proportional to the square of cumulative exposure, lung cancer risk lies between a linear and square relationship and ' pleural mesothelioma seems to rise less than linearly with cumulative dose. Although these nonlinear relationships provide a best fit to the data, statistical and other uncertainties mean that a linear relationship remains arguable for pleural and lung tumours (but not for perito neal tumours). ' ' Based on these considerations, and a discussion of the associated uncertainties, a series of quantified risk summary statements for different levels of cumulative exposure are presented. Crown Copyright 2000 Published .by Elsevier Science Ltd on behalf of British Occupational Hygiene Society. All rights reserved -- . . ' Keywords: asbestos: amphibole hypothesis; exposure-response; lung cancer; mesothelioma: quantified risk assess ment ' INTRODUCTION There has been much debate on the relative hazard ousness of the three main asbestos types; crocldolite. amosite and chrysotile (commonly known as blue, brown and white asbestos respectively), but no sys tematic attempt to quantify the differences. Existing published quantitative risk assessments have mostly not distinguished between the fibre types, and none Received 17 September 1999; in final form 5 iune 2000. * Author to whom correspondence should be uddressed. Tel.: +44.151-9514566; fax: +44-151-95114703; e-mail: john .hodgson@hse.gsi.gov.uk has produced quantified estimates of the risk from amphiboles (a collective minerulogica! term covering crocidolite and amosite). A review commissioned by the HSE in the 1980s from Professors Richard Doll and Julian Peto (1985) gave estimates for chrysotile alone; more recently a review by the Health Effects Institute (1991) produced estimates for an unspecified mixture of fibre types. An INSERM review (1996) also ignored differences in fibre type, and drew heav ily on the HEI review. The studies included in this review were selected by reviewing the material referenced in the Doll and Peto, HEi and INSERM reports and identifying all cohort mortality reports for which quantified data on SAS SCF-FA-8701 SC-ALL-25699 P-EXHIBIT-228 566 I. T. Hodgson and A. Oarmon I exposure was available either as an average for the Excess lung cancer measure cohort as a whole, or for individual subgroups. Seven teen such cohorts were identified (Albin el at., 1990a: de Klerk er at.. 1994; Dement ft til., 1994: Enterline , er at., 1987: Fmkelstein. 1984; Hughes era/.. 1987: Liddell er at.. 1997; McDonald er al.. 1983b. 1984; Excess overall lung cancer mortality has been expressed as a percentage excess of expected lung cancer mortality per unit of cumulative exposure. /?c = m(0L-EL)KEL.X) Neubttget and Kundi. 1990; Newhouse and Sullivan, 1989; Peto er al.. 1985: Piolatto er at.. 1990; Seidman Where 0L and j. are the numbers of observed and er /., 1986; Seidman and SelikotV. 1990; Sluis- expected lung cancers, respectively and X is cohort Cremer er til., 1992; Talcott er a/.. 1989). Three of mean exposure. This estimate of the lung cancer risk . the selected cohorts have been split into sub-cohorts is described as the 'cohort average' estimate. 95% which have been separately treated in this review: the confidence limits for the cohort average estimate South African crocidolite and amosite mining cohorts have been calculated assuming a Poisson distribution have been treated separately; the New Orleans asbes for 0,.. tos cement cohort has been split into the two separate plants covered, since the mix of fibres used in the two plants was different: and the Carolina textile cohort has been split by sex, since the results for men and women were rather different. The cohorts have been referred to by their geographical location except for cohorts 3 (Enterline er al., 1987) and 17 (Newhouse Mesothelioma measure Mesothelioma mortality was expressed as a per cent of expected mortality from all causes (adjusted to an age of first exposure of 30) per unit of cumulat ive exposure. . uad Sullivatt. 1.989) which arc identified by a com Rm= lOQO^E^pO pany name, and cohort 15 (Albin er al., 1990a) where the name of the principal author on the cohort has been used. Where 0M is the number of mesothelioma deaths, AJj the total expected deaths from all causes adjusted to an age of first exposure of 30, and X the mean Information extracted Information was extracted from the identified reports on the following: The number of deaths in the cohort from all causes and from lung cancer, and the corresponding cumulative exposure. (See Appendix A for a dis cussion of this measure, and the calculation of EMl). When the expected all causes mortality was not avail able, the denominator was taken to be the total observed deaths less the total of asbestos-related deaths (mesothelioma, osbestosis and any excess lung SMRs; cancer deaths). A 95% confidence interval for RM was Dose specific lung cancer SMRs (or rates), calculated assuming a Poisson distribution for 0M. where available; ' The number of mesothelioma deaths in the cohort (for pleural and peritoneal mesothelioma Treatment of `best evidence ' cause of death data separately); - In some studies causes of death have been assigned The rates of mesothelioma by categories of time in two ways, one based purely on data given on the since first exposure; death certificates (DC), the other using other data (e.g. The process/type of work being carried out; -- autopsy reports) to establish a 'best evidence' (BE) Cohort recruitment period and duration of follow cause of death. For lung cancer this review has gener up; ally used the DC data, since this preserves compar Average age at first exposure, when available; ability with the reference rates, and with the majority The type(s) of asbestos fibre used in the process; of other studies. For mesothelioma however, the BE The average fibre levels for the entire cohort and data has been used, since reference rates are inappro the average employment duration for workers in priate, and most studies use some sort of best evi the cohort, or simply the average cumulative dence judgement to identify mesotheliomas. exposure for the entire cohort; It might be thought that where reference rates ore Information about the smoking habits of the workera in the cohort where available; derived from DC data (as in the SMR analyses in this report) the observed deaths on a DC basis should always be used. The argument is not os clear cut as The sex of the workers. it seems. The coding of death certificates is subject to a range of errors, and the net error in the count of Some general issues on the summaiy of outcome deaths coded to lung cancer on national death certifi and exposure measures are discussed below. A more cates will be determined by the balance of these errors detailed discussion on some of these points is given across the whole population. One of these errors is in Appendix A, and the extracted data is shown in the tendency of pleural mesothelioma deaths to be full in Tables 12 and 13. coded to lung cancer. In the population as a whole. 11 .... vjfrr.rs: Quantitative risks of mesothelioma and lung cancer 567 this etTor is very small, but in an asbestos exposed cohort it may have a substantial effect. Leaving the miscoded mesotheliomas in the lung cancer count will overstate the true lung cancer SMR. Excluding them will in theory understate it, but only to the small extent that this error affects the population as a whole. The best available approximation to a true estimate of the risk is therefore to exclude the miscoded'mesotheliomas, and this has been done for this review. of 10. Details of this reassessed data were to be pub lished, but these have not so far appeared in print. It is therefore difficult to know whether to make an adjustment to the published estimates, and if so by how much. Similar comments may of course apply to other cohorts and introducing a correction might then diston rather than correct the overall picture, de Klerk and colleagues, developing estimates of environmen tal risk at Wittenoom (1992) use a factor of 4 without detailed discussion. The effect of using this adjusted Derivation of cohort mean exposure estimates Mean exposure for cohorts was calculated in differ ent ways, depending on the available information. exposure level is examined as a variant of the main analyses. When data was given for separate exposure groups, the cohort mean was calculated by weighting the indi Exposure-specific risk estimates vidual group means by the expected deaths from lung It is generally assumed that the most reliable guide cancer in the group. On the assumption that excess to dose-specific risk is provided by exposure analyses risk is proportional to cumulative exposure, this using estimates of individual exposure. This is clearly weighting preserves the same proportionality when the case when these individual exposure values can the results from subgroups with different exposures be accurately determined. However this assumption are aggregated, it is therefore the optimal statistical is very much not the case in the studies in this review. measure of aggregate exposure. Not only are there the inevitable problems of extrapo Where mean exposure values for individual dose lating earlier exposures on the basis of more recent categories were not given, the midpoints were used. measurements; there are also problems of convening The top exposure category was usually given as an the most usual historic measurements (in terms of open interval (e.g. exposures>!00 f/ml.yr): in these panicle counts) to the more relevant measure of fibre cases a value was chosen based on a view of the high counts. Direct fibre counting only became generally est likely exposure and the distribution of individuals used in the 1970s. across all exposure categories. It was assumed that In these circumstances it is at least arguable that where the highest category contains a relatively small global assessments of average exposure, set against proportion of the population, the category mean will overall mortality outcomes, should be preferred. be a smaller multiple of the lower band than other Exposure-response regressions with inaccurate indi wise. ' For cohorts where results for exposure specific sub vidual exposure assignments will produce a slope estimate biased downwards. Use of an overall assess groups were not given, the cohort mean was either ment will also minimise the error introduced by con given directly (cohorts 4, 13 and 15); derived from version from panicle counts to fibres, since these information given on the distribution of individual average conversion factors will represent a more doses (cohorts I and 7), or on the'exposure of internal accurate conversion for the totality of exposure than controls (cohort 17), or by multiplying a mean exposure level by mean exposure duration (cohorts 8 and 14). Exposure estimates given in particle counts were converted to counts of `regulated fibres' (fibres with for a particular individual; However, the arguments are not all one way. Over all mortality outcomes can only be assessed against some outside reference--usually (he regional or national population--and this may not represent a an aspect ratio greater than 3:1, and length>=5 microns), using conversion factors calculated by the true baseline level for the exposed population in ques tion. Assessment of an internal exposure response gives some check on (his issue. A complete absence report authors where possible. The most commonly of exposure response must cast some doubt on any used conversion was I mppcf (million panicles per overall excess being counted as a measure of risk (the cubic foot)=3 f/ml (fibres per millilitre), and this was Albin and Connecticut cohorts are examples of this). the value adopted for the Johns Manville cohort, Cohort-level risk measures were chosen for this where a conversion was not given. For the Massachu review both because these allow a wider range of data setts cohort, where the fibre involved was crocidotile i to be assessed than if attention is restricted to internal (rather than chrysolite as in the other cohorts with exposure response analyses and since (as argued particle counts), an independent expen hygienist was above) cohort-level exposure estimates ore likely to asked for an assessment (see Appendix B). be more accurate than individual exposures. The exposure estimates for Winenoom have been questioned by Rogers (1990) who hus suggested-- having rc-exumined some of the original samples Smoking using modem light and electron microscopy--that the The evidence on the joint effect of smoking and levels may have been underestimated by up to a factor asbestos exposure on lung cancer has been reviewed / l 368 1. T. Hodgson and A. Oarmnn recently (Vainio and Bofetta, 1994) who conclude that the overall evidence indicates an interaction in the multiplicative region. This implies that the rela lation. and this was the only study where an explicit adjustment for smoking was made. Unadjusted data was used for all other studies. tive risk of lung cancer due to asbestos exposure will be the same for smokers and non-smokers alike. Thus SMRs for lung cancer based on a reference popu lation with the same smoking habits as the cohort members should only reflect the effect on mortality due to asbestos exposure. An earlier review by Berry et al. (1985) estimated that the effect of asbestos exposure was about 1.8 times greater in non-smokers than in smokers (though with confidence limits which Fibre type and industry process For the purpose of summarising the information given in the studies, each cohort was given a fibre type classification of I, 2 or 3 letters according to the type of fibre used, with the letters y, a and o rep resenting chrysolite, amosite and crocidolite exposures respectively. For example: did not exclude a simple multiplicative interaction). If this is the case the observed effect of asbestos on lung cancer rates will be greater in populations with lower smoking prevalence. However, given the rela tive lung cancer risks typical of smoking (about 15 fold) and asbestos exposure (about 2-fotd) together with the generally high prevalence of smoking in the observed populations, the scope for bias--if there is indeed a differential effect of the scale suggested-- is limited. In either case, a problem arises when the smoking habits of the cohort members differ from those of the reference population, which is the case 'yao' 'yo' 'a' means all three commercial asbestos types were used in the cohort means chrysotile and crocidolite were used means only amosite was used The order of the letters indicates the relative itnportanceof the fibres used. Very small quantities of fibre were ignored in some cohorts (Carolina, New Orleans plant I, Connecticut), the reasoning for this in each cose is set out in Appendix A (Table 14). In a similar way, for display in tabular and graphical data sum maries, industry process was coded as follows. for some of the cohorts reviewed. For this reason, any information about smoking given in the studies was summonsed. The amount of information given was very variable, and could be categorised os follows: M C T I Mines Cement Textiles Insulation Products * 1. No information given, (Ferodo, US Insulators. Pat erson, South Africa, Johns Manville, Albin). 2. The percentage of the cohort that smoked, usually F L O Friction Products Lagging and work with insulation Other based on a cross sectional survey conducted in a particular year, (Connecticut, Balangero, Quebec, Mela-analytic issues Pennsylvania. Rochdale, Wittenoom). 3. Comparison of the prevalence of smoking in the The aim of a meta-analysis is to identify where evi dence from different studies is discrepant; ideally, to cohort and the reference population. (New Orle explain the reasons for the discrepancies; and where ans, Massachusetts, Carolina). data from different studies are coherent to combine 4. Estimation of the effect of any differences in them into a common summary which will be more prevalence--for example calculation of smoker precise and soundly based than the estimate from any adjusted lung cancer SMRs, (Vocklabruck) single study. For this review the coherence of esti 5. Data on prevalence of smoking within exposure- mates of Rl and J?M from different studies has been categories--but with no external comparison assessed in a Poisson regression framework, fitting a (Ontario).. common value of the parameter of interest across a group of studies and testing the residual deviance Most studies fell within the first two of the above between the observed and predicted numbers of categories. In these cases only subjective judgements events (mesothelioma or lung cancer deaths) in the could be made by the authors about the smoking hab studies in the group. Confidence limits around the its of the cohort members. Also, cross sectional stud group estimates were calculated by profile likelihood ies were often based on a small proportion of the cohort and may not be very representative. For most studies which addressed the issue the authors con methods. Confidence limits are not shown for the means of groups which show very significant hetero geneity, since such limits have no ready interpret cluded that there was no major difference in smoking ation. Indeed, in this situation it is not clear that the prevalence or that the slight differences in prevalence mean' itself has any natural meaning. Faced with were not likely to change the expected number of clearly discrepant data, purely statistical criteria can lung cancer deaths in a substantial way. Of the studies not be used to decide on a `correct' summary or . where comparative smoking data were given, the compromise estimate. .> Vocklabruck cohort showed the largest difference in The statistical analyses in this report only take cohort smoking habits and those of the general popu account of the statistical variability of the mortality \ ' '! . mmm Quantitative risks of mesothelioma and lung cancer 569 outcomes. The statistical variability in expected mor tality levels and cohort average exposures are ignored. This means that calculated confidence inter vals will be narrower and statistical distinctions sharper than they would be if these variabilities were known and allowed for. This needs to be borne in mind in the interpretation of these analyses. RESULTS Overview Figure 1 shows a graphical comparison of the mesothelioma and lung cancer risk coefficients. In order to plot zero values (which convert to minus infinity on the log scale), convenient nominal positive values smaller than any real non-zero value in the (relevant) data have been used. These ore in the range 0.001-0.002 for and between 0.0001 and 0.0003 for Ru. The three panels of Fig. I display the same data, with each cohort represented by its cohort code, fibre type and process. Cohorts which did not show a statistically significant excess of lung cancer (RL) are shown in brackets. ' Both risk measures cover about three orders of magnitude. For the bulk of the data risk estimates for lung cancer and mesothelioma are strongly correlated with Rl, very roughly equal to 100 RM. This hetero geneity seems more readily explicable in terms of fibre type than process. For example there are mining and asbestos cement cohorts at both extremes of the risk scale, while all the amphibole cohorts are at the high risl^end of the scale. But there are not really enough examples within each category statistically to driw definitive conclusions of this type. Total mesothelioma The summarised data for total (pleural and peritoneal) mesothelioma mortality are shown in Table I and Fig. 2. The estimates of RM for crocido lite cohorts are closely grouped around an average value of 0.S1. Similarly, the two amosite cohorts show results statistically consistent with their average of 0.10. The results from mixed fibre cohorts cover a wide range from a value close to that seen for the crocidolite cohorts (/?m=0.59 for Ontario), to values nearly three orders of magnitude lower, close to those seen in the chrysotile mining cohorts. The test for het erogeneity is very clearly significant (P<0.00l). The ranking of mixed cohorts by mesothelioma risk does not appear to correspond either to process or fibre mix. If the exposure estimate for Witienoom is increased by a factor of 4, the summary value of falls to 0.15, and the consistency of the three crocidolite values is completely lost (P<0.00l). Three of the six chrysotile cohorts had no observed mesothelioma deaths. The rates in the two chrysotile mining cohorts are similar at around 0.0013, while the (a) cohort code 0.1 0.01 [17) 5a) 0.001 15 1? (13a) All 5c 2m (10) 3 zero ____ . 16 21 <b) 1 r o.i rVi E .2 0.01 (yo) |e lya) S 0.001 fibre type o**> yao .a (a) VC V (y) J yao zero (c) 1 A------- process 0.1 0.01 (F> (C| 0.001 (M) M yy m"c c (M) f0T1 CT 1 zero A____ zero 0.01 0.1 F 1 Lung cancer risk (RJ T 10 Fig. I. Comparison of exposure-specific risks of mesothelioma and lung cancer (% per f/ml.yr), with cohorts labelled by cohort code, fibre type and process. (Note: the two coincident cohorts in the top right of the chart are Ontario (4. yo, Cj and SA crocidolite mines (13o, o. M). Symbols in brackets indicate a non-significant lung cancer excessl.' Cohort number Cohort name Table I. Summary of mesothelioma mortality (lata and exposure-specific risk estimates g Process Fibre Mesothelioma deaths Total expected mortality Adjustment factor fur age first exposed Average cumulative exposure (f/inl.yr) Mesollteliuma risk expressed as /`-value for percentage total expected mortality per licicrogeneity f/tnl.yr (KM) Total number Number peritoneal Unadjusted Adjusted fur age at first exposure 95%CI J. T. Hodgson and A. Oamton 14 Massachusetts 00 1 Wittenoom Mo 13o SA crocidolite mines M 0 Total - Crocidolite cohorts 12 Paterson 1a 13a SA amoxite mines Ma Total - amosite cohorts 4 Ontario IS Albin 7 Vocklabruck C yo C yuo C yu 8 US/Canada insulators L yuo II Pennsylvania TF ira 9 Rochdale 17 Ferodo T y F yo So New Orleans (plam 2,yo) C yo Sa New Orleans (plant 1) C ya 3 Johns Manville retirees 1 yoo 2m Carolina (men) T 10 Bolongero - M 6 Quebec M 2f Carolina (women) . T Sy New Orleans (plant 2. y) C 16 Connecticut Pooled chrysolite F estimates Total - excluding Carolina men - excluding mines y y y y y y 5 72 20 97 17 4 21 17 13 5 453 14 10 13 3 1 8 1 2 33 0 0 0 37 35 70 3 8.3 0.74 10 601.8 1.08 2 223.2* 0.93 15 9 355.9 0.63 1 305.7" 0.93 10 8 62.2 0.77 0 493.3 1 1 530.2 1.0.5 282 3170.6 1.09 4 821.1 1.08 0 602-5 1 0 2646.3 1 0 217* 1.26 0 294.5 0.85 2 762.5 1 1 410.1 1.34 <1 225.4 1.2 0 5912.7 1 - 299.2 1.34 - 397.1" 1.26 - 550.7 0.93 0 0 0 120 23 16.4 65 23.6 60 13 25 500 60 74 35 93 ' 79 750 28'300 61X1 26 22 46 0.50 t 0.52 0.55 0.073 0.056 0.46 0.2 0.038 0.029 0.029 0.022 0.014 0.015 0.004 0.(X)t 0.017 0.1X13 o.noi 0 0 0 0.6X 0.4 K 0.59 0.51 0.12 0.060 0.10 0.59 0.2 0.0.16 0.026 0.027 0.022 0.014 0.012 0.1X15 0.1X11 0.021 0.013 0.1X125 (MXXI9 0 0 0 10.22.1.6) (0.38.0.60) (0.36.0.91) (0.41.0.61) 0.6 (0.068,0.19) (0.016.0.015) (0.062,0.15) 0.2 (0.34.0.9) 1(1.11.0.35) (0.012.0.084) 10.024,0.029) (0.014.0.044) (0.011.0.U4I) (0.0075.0.024) (0.0024.0.034) (0.0001.0.028) (0.(KX)5,0.0028) . PCU.OOI (0.0016.0.047) (0.0003.0.009) (0.0006,0.0013) (0,0.035) (0,0.033) (0.0.016) 0.0010 0.0010 0.0033 (0.0007,0.0014) 0.11 (0.000731.0013) 0.69 (0.0006,03)10) 0.14 'Reduced by a factor.of 0.67 to exclude expected deadut less Utan 10 yr from firvl exposure (see Appendix C). `Expected all cause mortality in plant 2 partitioned in proportion to share of expected lung cancer. Quantitative risks of mesothelioma and lung cancer J7I Fig. 2. Exposure-specific mesothelioma mortality (/tM) by cohort and fibre type groupings, showing 9J% confidence intervals. Group means labelled in capitals. Confidence intervals not shown for groups with very significant heterogeneity. two cases seen in among men in (he Carolina cohort produce an estimate, with wide confidence limits, of 0.013--^>out an order of magnitude higher than for the mines cohorts. The very wide confidence limits for the three cohorts where no cases were observed are statistically consistent with either end of this range. Indeed there is no significant heterogeneity between RM estimates in the chrysotile group, although (he total shows some tendency to heterogen eity (F=0.l1). If the mines cohorts are excluded, the central combined estimate of RM increases to 0.0033, but with wide confidence limits (0.0006-0.01) and with a similar level of heterogeneity (F=0.I4). With the Carolina men excluded, the remaining data are coherent (P for heterogeneity=0.69), and the mean estimate of ffM is 0.001 (95% Cl 0.0007 to 0.0013) No summary estimate of has been calculated for the mixed fibre cohorts, since these are so clearly stat istically heterogeneous. This heterogeneity is plausi bly explicable by variations in the mix of fibres encountered. The estimates from the pure fibre cohorts suggest a difference in potency approaching two orders of magnitude between chrysotile and amosite, and a further five-fold difference between amosile and crocidolite. If these gross differences are even approximately correct, quite small variations in the fibre mix in the cohorts exposed to several fibre types could have important effects on the mesotheli oma risk in the cohort. This would have (he conse quence (hat the generally measured fibre levels would be an unreliable estimate of the true risk status. This will be particularly true where the history of usage of different fibre types has varied'over time. Lung cancer The summary data for lung cancer is shown in Table 2 and Fig. 3. The pure fibre groupings are less coherent for RL than for although the general pic ture is similar, with higher values for the amphibole cohorts, lower values for most of the chrysotile cohorts and intermediate values for the mixed exposure groups. The Carolina cohort is the one clear exception to this pattern. The mean estimate for the three crocidolite cohorts is 4.2% per f/ml.yr (95% Cl 2.8--5.8). The two amosite cohorts give somewhat dif ferent results, and despite (he wide confidence limits on the South African data they are not statistically consistent (1=0.022). Their joint mean is 5.2% per f/ml.yr (95% Cl 4.0-65). The five amphibole cohorts taken together are also not a statistically consistent group (P=0.027), with a joint mean of 4.8% per f/ml.yr (95% Cl 3.9--5.8). The heterogeneity is mainly due to the SA amosite cohort, and if this is set aside the remaining four amphibole cohorts are just statisti cally consistent (F=0.072) with a joint mean of 5.1% per f/ml.yr (95% Cl 4.1-6.2). If the exposure estimate for Wittenoom is increased by a factor of 4, the sum mary value of Rt falls to 2 for the combined amphi bole cohorts and to l.l for (he three crocidolite Cohort number Cohort name 14 Massachusetts l3o SA crocodolite mines ' 1 Wittenoom Total - croddolite cohorts All amphiboie cohorts ex. SA amosile 12 Paterson 13a SA amosite mines Total - amosite cohorts IS Albin 4 Ontario 5o New Orleans (plant 2. yo) n Pennsylvania 8 US/Canada insulators 7 Vocklabrtick 9 Rochdale 3 Johns Manville retirees 5a New Orleans (plant 1) 17 Ferodo AU mixed Mixed cxd. Ontario, Insulators and JM 2f Carolina (women) 2m Carolina (men) sy New Orleans (plant 2, y) 16 Connecticut - 6 Quebec 10 Balengeno ' AU pure chrysotUe Pure chrysotUe excluding mines Pure chrysotUe excluding mines Table 2. Summary of lung cancer monality daia and exposure-specific risk estimates Process Fibre Observed Lung cancer dcallts Expected SMR Excess % Excess Average cumulative exposure * (f/ml.yr) Lung cancer risk (% P-value for expected lung cancer jtet heterogeneity f/ml.yr) K,. (95% Cl) 00 M '0 M0 8 19 87 Ia Ma 98 21 C yao 35 C yo 22 C yo 31 TF ya 50 L yao 934 CTl1 yo 47 yo . 56 I yut> 73 C ya F yo 21 241 0.6 10.2 48.7 13.1 7.4 1.86 8.8 1.79 38.3 1210 85.5 78.6 20.5 14.5 19.4 5.3 17.7 33.8 256.8 42.2 37.1 28.4 22.5 242.5 4.78 1.45 1.8 4.14 1.75 1.48 3.64 Ml 1.51 2.57 (1.93 0.99 77.5 6.5 15.6 16.7 13.3 16.2 677 4.8 18.9 44.6 -1.5 -1.5 378 44.8 80 314 75.1 47.9 264 11.4 51 157 -6.7 -0.6 120 16.4 23 65 23.6 13 60 93 60 5tX) 25 138 750 79 35 10 5.2 3.4 4.2 4.8 5.1 5.8 1.9 5.2 6.2 5.2 0.81 0.8 0.53 0.45 0.37 0.21 0 0 0.47 (1.32 (3.9.21) (0.71.12) (I.9.5.2) (241,5.8) (3.9441) (4-1.6J) (4.4,07.4) (-0.44.5.1) (4.0,6.5) (-0.77.21) (27.8.8) (0.21.1.6) (0.16.1.6) (0.48.0.58) (-0.72.1.9) (0.10.0.70) (0.14.0.30) (-0.53.0.54) (-0.36.0.36) (0.16,0.50) 0.090 0.027 0.072 OX)22 P<0.001 0.056 Ty 38 13.8 2.75 24.2 i75 26 Ty 74 32.2 2.3 41.8 130 28 6.7 (3.6.11) 4.6 (2.9.6.7) Cy 42 32.4 1.3 9.6 29.6 22 1.3 (-0.29.3.4) Fy 49 35.8 1.37 13.2 36.9 46 0.80 (0.029.1.8) My 587 431.6 1.36 155 36 600 0.06 (0.042.0.079) yM 19 17.3 l.l 1.7 9.8 300 0.U3 (-0.11,0.24) 0.062 /< 0.001 2_J 0.0013 0.0/d) (0.043,0.079) 0.91 I. T. Hodgson and A. Damton I Qualitative risks of mesothelioma and lung cancer 573 rig. 3. Exposure-specific excess lung cancer mortality (Au) by cohort and fibre type groupings, showing 93% confidence intervals. Croup means labelled in capitals. Confidence intervals not shown for groups with very significant heterogeneity. cohorts, but both groupings now show very signifi cant heterogeneity (^<0.001). Among themixed cohorts, two stand out with parti cularly high'Values (Ontario and Albin). Both are asbestos cement cohorts, and both also had high lev els of mesothelioma mortality. The values for for these two cohorts are both more than six times the level of the next highest observation. - The heterogeneity among the mixed fibre cohorts is driven principally by three of them: Ontario. US/Canado Insulators and the Johns Manville retirees. Other reviewers (Doll and Peto, 1983; Hughes and Weill, 1986), have remarked on the unusually high risk estimate implied by the Ontario cohort and have suggested that the exposure estimates for this group may have been underestimated. Another potential contribution to the high risk of lung cancer in this cohort is exposure to silica: 8 out of 26 workers with post mortem examinations showed signs of silicosis (Finkelstein and Vingilis, 1984). There are clearly considerable uncertainties in the estimation of average exposures for the US/Canada Insulators cohort, since this is averaged over a very ige cohort with no doubt very variable exposure .periences and over a long time period. The size of ..his group means that the value adopted for it will detetmine statistically the average risk in this group. The study of retirees from the'Johns Manville asbes tos products company is unusual in basing its esti mates exclusively on follow-up of retired individuals from the age of 63. There is no obvious theoretical reason why this should produce a seriously biased estimate of risk, though asbestos related mortality at ages below 63 will be missed. This cohort has been followed up almost to extinction, and if the impact of asbestos exposure on mortality eventually declines after the cessation of exposure, then cohorts with near complete lifetime follow up will tend to show rather lower excess mortalities than those where survivors form a substantial proportion of the cohort. In addition, the Johns Manville cohort was one where the authors had not suggested a conversion factor from particles to fibres, and this review has used the most commonly used value of 3 l/ml=l mppef. If this' conversion implies higher exposure than in fact took place (the recent review by Lash el al. (1997), used a value of 1.4 borrowed from the New Orleans cohort), then the risk coefficient implied here would be too low. If these three cohorts are excluded from the group the remaining eight are just statistically consistent (/*=0.056), and their joint mean is 0.32 (95% Cl 0.16-0.50). The six chrysolite cohorts fall into two groups: the two Carolina cohorts give values around 6% per f/ml.yr, the other four, including the two mines cohorts and dominated by the large Quebec cohort, are consistent with a joint AL estimate of 0.06% per f/ml.yr (95% Cl 0.043-0.079). The Connecticut and New Orleans (chrysotile only) cohorts give central estimates of /?L substantially above this value, (0.80 I 574 J. T. Hodgson uml A. Darnion and 1.3 respectively) but both confidence intervals are very wide. Even if the mines cohorts are excluded there is still very clear statistical inconsistency between the Carolina results and those from Con necticut and New Orleans (/,=0.00l 3). The Carolina results are also out of line with the two other (mixed fibre) textile cohorts--Rochdale and Pennsylvania-- whose 93% confidence intervals for RL have no over lap with those for Carolina. mineral oil (slight, moderate, high). The coefficients for these categories in the joint model were not reported, but were as follows, expressed as odds ratios relative to 'slight' exposure: Mineral oil exposure Odds ratio Moderate 1.12 High 1.47 (Dement, personal communication) 95% Confidence interval 0.57-2.21 0.3-2.75 RISK ASSESSMENT AT MODERATE AND HIGHER CUMULATIVE EXPOSURES Mesothelioma The quantified risk for mesothelioma at the kinds of cumulative exposure levels recorded in the reviewed cohorts--say. from 10 f/mt.yr upwards-- presents a reasonably coherent picture, with values of Rm, in round figures, of 0.S, 0.1 and 0.001 (at most 0.003) for crocidoliie, amosite and chrysotile respect ively (see Fig. 2). Lung cancer It is more difficult to come to a clear view of the quantified risks of lung cancer, because of the incon sistency of the results especially for the chrysotile cohorts (see Fig. 3). The omphibole estimates ore reasonably consistent In round figures the estimates fall in the range 2-10% per f/ml.yr. The mean for the crocidolite group is rather lower (4.2) than that for the amosite group (3.2), though their confidence lim its overlap substantially. The mean risk for all amphibole cohorts is 4.8% per f/ml.yr (95%Cl 3.9-5.8), but with some evidence of heterogeneity (P=0.027). If the SA amosite cohort data ore set aside, the remaining data are reasonably consistent (P=O.Q72), and the mean estimate becomes 3.1 (93%CI 4.1-6.2). In round figures, a value of 3% per f/ml.yr would rep resent a reasonable risk estimate for both omphibole fibre types. The pure chrysotile cohorts produce estimates of Rl spanning two orders of magnitude, from a value of 6.7 for the* Carolina women to 0.03 for Balengero mine. How should this very wide range of est>* mates be interpreted? As far as evidence, from 'pure' exposure goes there are only two strongly informative cohorts: Quebec and Carolina. The differences between these two has been studied and discussed extensively but, finally, inconclusively. The hypoth esis that mioerai oil used to suppress dust in the Caro lina plant may have contributed to the lung cancer excess has been addressed by an internal case-control analysis of this factor reported by Dement el al. (1994) and Dement (1991)). The most recent report (Dement et al., 1994), shows that the odds ratios for different cumulative asbestos exposure categories are essentially unchanged by the addition of a variable representing subjects' typical level of exposure to Although these ORs are not statistically significant (and do not form a statistically significant trend), there is some suggestion that mineral oil may have a role in enhancing the asbestos effect, particularly since all the effect of exposure duration is absorbed in the asbestos measure (workers were assigned to oil exposure categories according to the assessed oil exposure level at which they had spent the longest proportion of their employment in the plant). Early results from this cose control study showed a cross tabulation of cases and controls by asbestos exposure and mineral oil category (Dement, 1991), without for mal modelling. Crude odds ratios on (his data suggest that the asbestos response is progresdtaly steeper with increasing mineral oil category. If mineral oil does have an enhancing effect, (he anomalous increase in estimated exposure specific lung cancer risk for men in the Rochdale cohort first exposed after (930 could be explained, since dust suppression using mineral oil was introduced from that date (Peto et al.. 1983). The regression slope estimate of /?L for the men first exposed after 1950 is 1.3 (95%CI 0.37-2.6). three times the value for men first exposed between 1930 and 1950. The plausible suggestion that the longer fibre used in textile processes are responsible seems to be con tradicted by the comparative analyses of lung fibre burdens in Quebec and Carolina cohorts reported by Sebastien et al. (1989). They found that the projtortionate distribution of fibres by length' was very similar in Quebec and Carolina lungs. Nevertheless, (he notion that the longer fibres used in textile pro cesses do represent a higher risk, is consistent with experimental evidence that longer fibres are more car cinogenic (Meldrum. 1996; Stanton et al., 1981; Miller et al., 1999). Green et al. (1997) have shown that the mean length and aspect ratio of chrysotile fibres in the lungs of Carolina workers are greater than in a local population control series: and than in the lungs of workers from the Albin cohort (Albin et at., 1990o.b). Both studies on (he lung content of Carolina work ers have found omphibole (crocidolite or. amosite) fibres in an appreciable proportion of them, (hough at much lower levels than for chrysotile and its associa ted tremolite. Sebastien etal. (1989 report that amphibole fibres at concentrations >0.1 f/pg (fibres >5 microns long) were only found in the lungs of work- M i Quantitative risks of mesothelioma and lung cancer 575 en hired before 1940, which conflicts with the period of known use of crocidolite yam (in very small quan tities--see Appendix A) in the plant after 1950. This raises the possibility that some amphibole formed part of tbit exposure mix in this cohort in an early period. Green et at. (1997) show that the levels of amphibole are higher in Carolina workers than in local controls (2-fold difference in geometric mean, /*=0.031) but much less strikingly than 'for chrysotile (5-fold, f5<0.0001) or tremolite (14-fold, 0.0001). They also report that amphibole at levels > 1.0 f/pg (all fibre lengths) were found in only one of the ten lung cancer cases for whom this datum was available. This last observation limits the extent to which amphibole exposure--perhaps unrecognised--might play a role in this cohort Whatever mechanism is in play does not appear to apply--to the same extent, at least-- to the other two textile cohorts reviewed. As already pointed out, the Pennsylvania and Rochdale cohorts (with mixed fibre exposures) both give substantially lower estimates of RL. If it is accepted that some such feature of the pro cessing in the Carolina cohort has genuinely produced a much higher risk than seen in other chrysotile 'aborts the question can be asked how typical these ..eatutes are of the bulk of applications? Looked at in the wider conieti of cohorts with mixed fibre exposure, the RL value for Carolina looks untypically high. Seuing aside the possibility that amphibole presents a higher risk of lung cancer, the observations of RL from mixed fibre cohorts can be taken as informative.^ the RL level for chrysotile. This sug gests that in typical applications (including other tex tile processes) RL for chrysotile is generally lower than the value derived from the Carolina cohort The median RL for the 16 cohorts with some chrysotile exposure is 0.5, compared to 4.5 for Carolina men and 6.7 for Carolina women. All but two of the mixed fibre cohorts give an RL estimate less than I. and of the two exceptions one (Albin) has a confidence limit including zero, and the other (Ontario) shows features suggestive of significant exposure to crocidolite (see below, Fig. 4 and related text). To the extent that amphibole fibres make a dispro portionate contribution to the lung cancer risk in the mixed exposure cohorts--and the evidence presented here suggests that they do--the typical risk of lung cancer from chrysotile exposure would be even lower. In most circumstances a value of 0.5% per f/ml.yr should probably be regarded as an upper limit to the lung cancer risk from pure (commercial) chrysotile. The mean RL estimate for mixed fibre cohorts excludng the three with particular interpretationol difficult ies is 0.32% per f/ml.yr with an upper 95% confi dence limit of 0.50. It should be noted that a value of 0.5% per f/ml.yr is not as far out of line with the Carolina observations as jl might seem. The `cohort average' risk estimate from this cohort (6.7 for women, 4.7 for men) prob- Fiy. 4. Comparison of excess monalily from pleural and pcriioneat mesothelioma, showing fibre type. ably overestimates the risk, which from internal analysis is I for women and 3 for men (Dement et at.. 1994, p. 439). The exposure response regressions on this cohort give an intercept close to zero excess risk at zero dose, and there is thus no reason to sus pect serious error in the reference rates (with conse quential doubts about interpreting the slope). There is also the possibility of inaccuracies in the conversion of particle counts to fibre counts. One early report on this cohort (McDonald et ai, 1983a) suggested that the average conversion factor should be about 6 f/ml to I mppef. If this were true, the risk per f/ml.yr would be halved. A `best estimate' of the lung cancer risk would be lower than 0.5% per f/ml.yr. Noting that the mean risk of (he mixed fibre cohorts (excluding the three mentioned above) is 0.32% per /ml.yr, and chat the amphibole risk is over 10 times higher, it is possible that virtually all (he observed risk could be explained by rather less than 10% of amphibole in the mixed exposures. However there is no direct evidence on which an estimate of the risk of `pure' chrysotile could be based. Apan from the Balangero cohort, all (he chrysotile evidence considered here effectively relates to Canadian chrysotile. since this was the dominant source of fibre for the other chrysotile cohorts. The risk of 'commercial' chrysotile as esti mated from the mining cohorts is 0.06% per f/ml.yr. Given that the processing of chrysotile may produce some additional risk, the best estimate should be set higher than the mines level, say at 0.1% per f/ml.yr. The overall risk, of a mixture of 96% chrysolite with a risk of 0.1, and 4% amphibole with a risk of 5.1 would be 0.3% per. f/ml.yr. EXTRAPOLATION TO LOW EXPOSURES All these cohort observations reflect the effect of exposure to high levels of asbestos. The main interest J7A J. T. Hodgson and A. Damion in quantitative risk assessment in current conditions (and the pure fibre quantified cohorts) is shown in is to apply this evidence to (he estimation of the risks Fig. 5. associated with exposure levels 100-1000 limes There is still an apparent separation between cro lower. The standard assumption is (hat, other things cidolite and amosite cohorts, (hough the segregation being equal, the risk will be proportional to dose: but is now less clear cut (as might be expected given the this is more a cautious default assumption than any small numbers often involved). There is. of course thing more soundly based. To quote from the HEI considerable statistical uncertainty in both of these review: 'The assumption of dose-linearity for low- variables, and a simple regression (in which uncer dose assessment purposes is thus a widely accepted tainty about \c` values is ignored) would be mislead and scientifically reasonable compromise rather than ing. Table 4 summarises the results of regressions in an established scientific principle of carcinogenesis". which the fit is optimised in both variables simul However, if the true relationship between exposure taneously (fit being measured by deviance..assuming and response was not linear, the impact on low dose Poisson variation for the numbers of mesotheliomas extrapolations could be dramatic. There is some indi at each site/. cation in the present data suggesting a non-linear Fitting a single line through all the data produces exposure response, particularly for peritoneal meso a line with a slope (on the log-log scale) of 1.2, but thelioma. and the next sections examine this question. the overall fit is unsatisfactory (P<0.00l). Allowing the two fibres to have separate fits makes a very sig Relationship of pleural and peritoneal mesothelioma nificant improvement to (he fit (P<0.001), and both Figure 4 plots the percentage excess mortality from fits have steeper slopes (2.3 for crocidolite and 3.1 peritoneal mesothelioma against that from pleural for amosite -- not shown in table). These slopes ore mesothelioma. Cohorts with no mesothelioma cases not very precisely determined, and constraining them of either kind ore excluded. Cohorts with no perito to be equal does not materially degrade (he fit neal mesotheliomas ore plotted on (he peritoneal scale (P=0.7J). The central estimate for this common slope on or close to the 0.01 ordinate. The positioning of is 2.4. ;? the cohort points strongly suggests a pattern of two This model provides a very close statistical fit to alignments, one defined by the pure crocidolite all but two of the cohorts. The two exceptions are (he cohorts, the other by the two pure amosite cohorts. gas mask cohorts in Canada (McDonald and McDon Four mixed exposure cohorts lie very close to the ald. 1978) and in Leyland (Acheson et at., 1982). amosite line: the US/Canada Insulators, New Orleans which contribute 6.1 and 4.S respectively to (he total plant I, the Johns Manvitle retirees and the Albin deviance. Possible reasons for these cohorts to be cohorts. All but the lost of these clearly had amosite untypical can be identified. The Leyland cohort was as the main amphibole fibre. The point representing not ascertained from employment records, but from the Ontario cohort lies very close to the crocidolite occupational details recorded on (he wartime popu line, suggesting perhaps (hat the anomalous results lation register compiled in September 1939. If the from this cohort may be explained by underestimated numbers directly involved with gas mask assembly exposure to crocidolite. have been over estimated the percentage excess mor The position of the (male) Carolina cohort seems talities will be proportionately under estimated, (f, for somewhat anomalous. The single peritoneal meso example, only 2/3rds of the identified women were thelioma in (his group -is the only one in a cohort in fact exposed, the expected mortality denominator without material amphibole exposure, and the equal would fall to around 120, and the residual falls from ity between pleural and peritoneal numbers (one of -6.1 to 4.3--still an outlier, but materially less extreme each) is only otherwise seen in cohorts with much (P=0.038 instead of 0.014). The overall excess mor higher levels of mesothelioma (and substantial amphi tality from mesothelioma recorded in the Leyland bole exposure). The possibility of unrecognised cohort is much lower than in the Nottingham cohort amphibole exposure again suggests itself, but too engaged on (he some process: 2.7% at Leyland and much should not be read into this single peritoneal 16.3% at Nottingham, again suggesting the possibility case. It is dear (bat the three fibre types produce dif of underestimation (eg by dilution of the exposed ferent mesothelioma responses overall. The question population), perhaps substantial. of differential responses by mesothelioma site can The assessment of mesothelioma in the Canadian really only be addressed for the amphibole fibres. gas mask cohort was particularly exhaustive, involv This relationship does not depend on quantified ing review of pathological data for all cancer coses. exposure data, and if it is real it should be reproduced Three of the six peritoneal cases were only identified in other cohorts with predominant - amphibole.. after this review. If the number of peritoneal meso exposure. The most informative cohorts will be those theliomas is reduced by three, the residual for (his with crocidolite or amosite exposure, but not both. cohort falls from 4.5 (P=0.034) to 2.0 (P=O.I6). A Medline search identified eight such cohorts. The However these are post-hoc rationalisations, and it relevant data ore summarised in Table 3, and a plot is not clear whether it is better to remove these of the percent excess mortalities from these cohorts cohorts from the model or not. Despite the large VY Quantitative risks of mesothelioma and lung cancer 577 Table 3. Additional data on pleural and peritoneal mesothelioma from cohorts with predominant exposure to crocidolite or amosite (but not both), and without reported quantified cumulative exposures Cohort No. Reference Process Fibre Sex Expected all cause morality Pleural No. % Excess mortality Peritoneal No. % Excess mortality 18 Jones et al. (1996) 0r 400* S3 13 14 3.5 19 Acheson et al. (1982) Gas masks 0 f 185 3 1.6 2 1.1 (Leyland group) 20 McDonald and oy mf 41* 3 7.3 6 14.6 McDonald (1978) 21 Hill et al. (1981) 0 0m S' 1 20 1 20 . 22 Levin et al. (1998) 1 am 133.6 4 3 2 . 15 23 Parolari et al. (1987) 1 a mf 115.1 2 1.7 i 0.87 24 Finkelsiein (1989) 1 a m 1.89 2 106 25 Acheson et al. (1984) 1 *y m 298.8 4 1.3 1,. 053 'Estimated as observed deaths less asbestos related deaths. ^Estimated assuming 75% mortality from age 31 to 68. tissue. If true, it is presumably related to the dynamics controlling the distribution of asbestos fibres around the body. Note that this relationship does not depend on the cumulative exposure, and is therefore not sub ject to the uncertainties attached to exposure esti mation. Whatever its physical/biological explanation, these observations imply that at least one of these out comes has a non-linear relationship with exposure. Fig. 5. Joint distribution of excess mortality from pleural and peritoneal mesothelioma, showing fibre type. (Note: Label size (area) roughly proportion to total mesothelioma numbers in ' each cohort). residuals for these two cohorts, the overall residual deviance for the inclusive data (model 2) indicates a satisfactory fit (/M>.22). If the two outliers are removed, the separate fibre model fits the data almost exactly, and the slopes for the two fibres are very similar (model 3) and higher (around 3.2) than the y for the fit including them. In either case the single >e model is rejected in favour of separate fits to the iwo fibre types, with similar slopes. The peritoneal rate is proportional to at least the square--perhaps as much as the cube--of the pleural rate. The. form of the relationship is unusual and some what surprising, since both outcomes reflect the effect of the same carcinogenic insult to the same type of Pleural mesothelioma and cumulative exposure To examine this question more closely. Fig. 6 shows a plot of excess mortality from pleural meso thelioma against cumulative exposure with cohorts represented by their fibre type code. Figure 7 shows a similar plot for peritoneal mesothelioma. The points for the pure amphibole cohorts show a clear pattern of alignment, with the slopes for pleural mesotheli oma less than 1 and those-for peritoneal mesotheli oma greater than 1. Table 5 summarises the results of Poisson regression fits to the relationship between percentage excess mortality from pleural cancer and cumulative exposure, and the observed data points and selected regression lines are shown in Fig. 6. The relationship is modelled as linear on a log scale for each variable, and therefore has the form P^ = A^C where Pp, is the percent excess mortality from pleural cancer, X is cumulative exposure and Aft and r are regression parameters. The corresponding predicted number of pleural cancers for a given cohort is Att)CEAi/\0Q (where EAdJ is expected all cause deaths adjusted to an age at exposure of 30). The parameters were esti mated by minimising the residual deviance between the observed and predicted numbers of pleural cancer for each (pure fibre) cohort It is clear that a wide range of slopes (r) are statisti cally consistent with the data. With independent fits to each fibre type the slopes are 0.62, 1.2 and 0.72 for crocidolite, amosite and chrysolite respectively. 578 I. T. Hodgson and A. Damcon T"t>le . Join Poisson regression (slnicture model) of relationship between pleural and peritoneal mesotheliom; '\ (Sfcperitoneal=A.%pleurall`) f Model A b Residual deviance Degrees of P freedom 1. All data 2. By fibre, common slope 0 u Overall .1. Fit excluding Ley I and and Canadian gas mask data By fibre 0 a Overall 0.21 0.0089 0.26 0.00074 0.17 1.2 2.4 2.4 2.2 3.1 21.6 12.0 l.l 12.1 0.1 1.0 l.l II <0.001 6 0.06 5 0.95 10 0.22 3 0.98 4 0.91 7 0.99 Fig. 6. Excess monatiiy from pleural mesothelioma against cumulative exposure, showing fibre type. Regression tines fit ted to pure fibre cohort. Bold lines indicate fits with slope con strained to be common across fibre types, narrow lines are unconstrained fits. (The fit for amosite is of course completely determ ined since there are only two observations.) The total residual deviance is 3.93. Moving to a model in which the three slopes are constrained to be equal, the residual deviance increases marginally to 4.53, an increase of 0.6 with a corresponding increase of 2 degrees of freedom (df), clearly not a statistically sig nificant change in overall fit (F=0.74), nor for any individual fibre type. The best fitting common slope is 0.75. Using deviance differences to construct a 95% confidence limits for the common slope gives esti mated upper and lower limits of 0.27 and 1.3. Peritoneal metothelioma and cumulative exposure Figure 7 and Table 6 show similar regression analyses for peritoneal cancer. Again the crocidolite and amosite points align themselves on two parallel lines. The small numbers of observed events means that the statistical uncertainties are quite wide. There Rg. 7. Excess monality from peritoneal mesothelioma against cumulative exposure, showing fibre type. Regression lines fit ted to pure fibre cohorts. Bold lines indicate fits with slope. constrained to be common across fibre types, narrow lines are unconstrained fits (the slopes are identical for crocidolite). is very little difference between the slopes (r) for the two fibres, and the best common slope is 2.1, with a deviance based 95% confidence interval from 1.2 to 2.9, The single peritoneal mesothelioma among the Carolina men, together with zero cases in the other chrysotiie cohorts generates a negative value of r. If a common slope is imposed over all three fibres the best estimate is 1.6, but with significant heterogeneity (F=0.0025--data not shown). Only the amphibole cohorts have enough data to draw valid conclusions on peritoneal mesotheliomas. The comparison of pleural and peritoneal slops independent of exposure levels suggested a ratio of slopes between 2.4 and 3.2. If the ratio of the esti- Quantitative risks of mesothelioma and lung cancer 579 Table 5. Possion regression of pleural cancer against cumulative exposure by fibre type Fit/fibre type r 9591 Cl for r Residual " Degrees of deviance freedom P 1. Independent fits 0 1.4 a 0.02 y 0.0057 Overall 2. Best common slope 0 0.93 a 0.13 y 0.0047 Overall 3. Common slope, amphiboles only 0 0.88 a 0.120 Overall 0.62 1.2 0.72 0.75 0.77 (-0.54, 1,43) (-0.32, 3.5) (0.17. 1.79) (0.27.1.3) (-0.069. 1.62) 0.25 3.68 3.93 0.36 0.49 3.68 4.53 0.39 0.44 0.83 1 0 4 5 2 1 6 7 2 1 2 0.62 0.45 0.56 0.84 0.48 0.72 0.72. * 0.82 0.51 - 0.66- Table 6. Possion regression of peritoneal cancer against cumulative exposure by fibre type Fil/fibre type / 9591 Cl for i Residual Degrees of deviance freedom p 1. Independent fits A 0.0022 2.1 (0.93, 2.9) 0.10 1 0.75 0.00018 2.4 (0.41,6.4) 1.4 -1.7 (-22. 0.91) 2.60 4 0.63 Overall 2.70 5 0.75 2. Common slope, amphiboles only 0 0.0022 0.10 2 0.95 2.1 (1.2,2.9) 1 0.76 a 0.0006 0.09 2 0.91 Overall. 0.19 e mates of the peritoneal and pleural slopes is con strained to be 2.4, the best fit pleural and peritoneal slopes are : 0.86 (95%CI 0.51-1.15) and 2.1 (95%CI 1.2-3.6). If the ratio of slopes is constrained to be 3.2. the estimated values are r=0.67 (95%CI 0.40 0.90) and r=*2.I (95%CI 1.3-2.9). Support for a convex (r<l) increase of pleural mesothelioma risk with exposure can be found in the detailed dose-specific analyses of the Wittenoom mesotheliomas by Berry (1991). Most of these cases (62 of 72) were pleural. Figure fi plots the constant terms in the four exposure categories of Berry's analysis against their mean cumulative exposure. The slope is very close to 0.5. In addition, Goggon et at. (1995), concluded from a comparison of the ranking of occupations by mortality from pleural and perito neal cancers and from asbestosis that "a more plaus ible explanation (of the different rankings) is that the jsure response relations for mesothelioma and xxtosis ore non-linear, with the risk of pleural mesothelioma rising relatively more steeply at low exposures, but less steeply at high exposures". A non-linear relationship between exposure and the rales .of pleural and peritoneal mesothelioma means that the percent excess mortality per f/ml.yr (Ru) will tt Fig. 8. Scaling constant in the four exposure groups of Berry (1991) analysis of the Wittenoom crocidolite cohort, plotted against the mean cumulative exposure in each group. The plot ted line is proportional to the square root of cumulative exposure. not provide a consistent summary of the effect for mesothelioma at the two sites considered individu ally. Each additional unit of exposure will add--pro gressively--less risk for pleural tumours, and more for peritoneal tumours. The point at which the ubsol- 580 I. T. Hodgson and A. Damien (a) amphibole and mixed cohorts, (b) chrysotlle and mixed cohorts, (it to amphibole data tits to chrysolite >0 r y (yo| --------- -- vi (vl 1 fvol y<> yto F <v> lv> Cumulative exposure (f/ml.yrs) 10 100 1000 Cumulative exposure (f/ml.yrs) Fig. 9. Percent excess lung cancer by cumulative exposure, showing librc type, with regression lines fitted to.pure fibre cohorts (A: combined amphibole data. (I) slope free, (2) slope-fixedol: Y: chrysolite data. (I) all data, slope free (2) excl. Carolina. slope (tee, (3) excl. Carolina, slope fixed=l). uce risks for tumours at the two sites are predicted to cumulative exposure. There is no significant differ be equal is around 90f/ml.yr for crocidolite, around ence between the regressions for crocidolite and 55f/ml.yr for amosite. Below these values pleural amosite points, so these are treated together. Using all tumours are more common, and at higher levels per the data, independent fits for amphibole fibres gives itoneal tumours dominate. It happens that across the a concave relationship (n= 1.6). and for chrysotile a scale of cumulative exposure values in the reviewed negative slope (r=-0.25). These are clearly inconsist cohorts (from about 10 to nearly 1000 f/ml.yr). the ent with each other, and both depart very significantly relationship between exposure and total mesotheli from linearity (P<0.001). oma risk is not far from linear, so the summary index The negative slope for chrysotile depends endrely Rm does provide a reasonable index of the overall on the Carolina data, and if this is removed the slope mesothelioma risk over this range. is just posidve (r=0.039) with a Cl that just includes 1. Clearly the data for chrysotile-only cohorts do not provide a coherent basis for direct esdtnadon of the Lung cancer --exposure-response slope, and some appeal to the evi If pleural and peritoneal mesothelioma have a non dence provided by cohorts with' mixed exposure is linear relationship with asbestos exposure, the ques necessary (os in the discussion of Table 2 and Fig. 3). tion arises as to whether the relationship for lung can The concave slope for amphibole cohorts is largely cer is linear. Figure 9 shows a plot of percent excess dependant on the two extreme points, the Massachu lung cancer against cumulative exposure and Table setts and SA amosite cohorts. The lung cancer excess 7 summarises regression results for lung cancer by in the SA amosite cohort is quite small and stadsd- Fit/fibre type Table 7. Poisson regression of lung cancer against cumulative exposure by fibre type r 95* a for r Residual Degrees deviance of freedom P Combined amphibole 0.49 ...excluding Massachusetts and SA amosite: 1.1 Chrysotile ...excluding Carolina 195 27.5 1.6 1.4 -0.27 0.030 (1.1.2.I) (0.89,2.0) (-0.44. -0.07) (-0.26. l.l) 2.35 0.83 19.8 0.91 3 0.50 1 0.36 4 <0.001 2 0.63 Quantitative risks of mesothelioma and lung cancer SSI cally unstable, and the exposure estimate for the Mas sachusetts cohort is based on fairly slender evidence, if these two cohorts are removed the best (it slope becomes 1.4, with a confidence interval that includes I. The Massachusetts cohort with its very high levels of excess mortality, and as cohort with the highest estimated mean exposure to crocidolite, has an important--though not determining--impact on the estimates. It is unfortunate that the exposure estimates are somewhat speculative (see Appendix B). At the same time it should be noted that in relation to a prior expectation of a linear dose response the effects of (his observation on the pleural and lung cancer esti mates are opposite: (he pleural slope is flattened and the lung slope is steepened. This does not of course prove (hat the exposure estimate is correct, but if it is materially in error then either the pleural or the lung slope is even further from linear than suggested by (he present analyses. DEVELOPMENT OF NON-LINEAR RISK ESTIMATES Mesothelioma . The data in Tables 5 and 6 and Figs. 6 and 7 sug gest (he following model with separate components for pleural and peritoneal tumours: where is the percent excess mortality, r and r are the pleural and peritoneal slopes of the exposure response onra log-log scale, Ap, and Ap, are constants of proportionality for the pleura) and peritoneal elements of (he risk respectively, and X is cumulative exposure in f/ml.yr. If the information about the ratio of r and r from the non-quantifled cohorts is ignored, the best fit values using all the data are r=0.75 and r=2.1. With out the chrysolite data, the estimate of r is essentially the same (0.77). Analysis of the ratio r/r including the non-quantified cohorts (Table 4) indicates values for this ratio around 2.4 with all the data, around 3.2 excluding the two outlying cohorts. If a simultaneous fit is made to the full data with the ratio of pleural and peritoneal slopes fixed at 2.4, the resulting esti mates (using only the amphibole data) are r=0.86 and f=2.1. If the ratio of slopes is constrained to be 3.2, the estimated values are r=0.67 and r=2.1. There is little to choose between values of r from 0.67 to 0.86. We will use a slope of 0.7S as our best estimate for r. The estimates for / are less variable, ' in any case have no bearing on risk estimates at A levels. We will lake f=2.1 as the best estimate. How wide a margin of uncertainty should be allowed on these slopes? On purely statistical criteria, values of r between 0.4 and 1.2 could be chosen. However a slope as low os 0.4 seems unlikely on physical grounds. Berry's unalysis of Wittenoom data using individual doses implies a slope of about 0.3, but the uncertainties of individual dose assignment are likely to have biased this estimate downwards. The argument above suggests that the lower end of range should be set at 0.67 or lower. We will take 0.6 to represent the lower end of the plausible slope range. There are quite strong a priori reasons for using a slope of 1. It is the value that all previous risk esti mations have used, and represents a natural assump tion (effect is proportional to cause) in the absence of evidence to the contrary. A linear relationship is also (in most models) consistent with the data. We therefore take r=i as the upper end of the slope range. Different slopes imply different best fit values for A, and Apr. These estimates and their 93% confidence intervals for the three fibre types are shown in Table 8. Effects of exposure duration and age at first exposure This formulation does not take duration of exposure or age at first exposure into account. The HEI (and similar) risk models (see Appendix A) imply that for equivalent cumulative exposures, short exposure times produce larger risks than long exposure times, (in other words 10 f/ml for I yr is worse than I f/ml for 10 yr); and (hat exposure at younger ages will produce higher excess mortality rates. All the amphibole cohorts considered here had short exposures (averaging about 2 yr). The suggested risk model for amphiboles is therefore appropriate for short exposures, but will overstate the risk from extended exposure periods. The chrysotile coef ficients are effectively determined by the Quebec cohort, where the average exposure durations were quite long (averaging about 10 yr). A given cumulat ive exposure accrued over 2 yr (starting at age 30) produces about 40% more deaths as the same exposure accrued over 10 yr. For general risk assess ment purposes, where short exposures are more likely to be at issue, the chrysotile coefficient should be increased by a factor of 1.4. Reductions in the exposure accrual time below 2 yr have very little impact on the risk. The risk estimates summarised above apply to exposure stoning at age 30. Table 9 shows adjustment factors derived from (he HEI model to convert risk estimates for an age at exposure of 30 to other exposure ages. Predicted effects at very long follow up It can reasonably be questioned whether a given asbestos exposure will continue to generate a constant excess mesothelioma mortality beyond 40 or 30 yr follow up. The evidence from cohorts with long fol low up is that the incidence eventually falls. In the Paterson cohort a significant fall is seen for follow up beyond 33 yr. In the US/Canada insulators there I I 582 / T. Hodgson tind A. O&nuon Table 8. Estimated coefficients* with 95% confidence intervals for constants in the risk prediction equation for PM at three levels of the slope coefficient r ' Slope/Fibre ** 95% Cl V5% Cl Best estimate slope (reO.75, r=2.l) Crocidolite Amosite Chrysotile High slope (re I, r=2.5) Crocidolite Amosite Chrysotile Low slope (re0.6. f=l.7) Crocidolite Amosite Chrysotile 0.94" 0.13* 0.0047* 0.43 0.052 0.000970 1.5 0.24 0.012 10.71,1.2) (0.060.0.25) (0.0030.0.0069) (0J3".0.54) I0.()22*,0.099) 10.00064",0.0014) (1.1.1.9*) (0.11.0.44*) <0.0078.0.018*) 0.0022 0.0006 0.00053 0.00012 0.0083 0.003 (0.0011.0.0039) (0.00025.0.0012) (0.00029.0.00087) (0.000049.0.00024) (0.0043.0.014) (0.0013.0.0058) `Coefficients used for risk extrapolation at low doses shown in bold: " `best estimate, "lowest arguable, `highest arguable (see Table II). Numbers of peritoneal mesotheliomas at low doses are negligible. For short exposure, chrysotile coefficients should be multiplied by 1.4. Table 9. Adjustment factors to convert estimates of meso thelioma mortality due to asbestos exposure starting at age ' 30 to other exposure start ages Age Factor 20 25 35 40 2.1 1.5 0.6 0.4 Table 10. Estimated coefficients with 95% confidence intervals for constants in the risk prediction equation for PL for chosen levels of the slope coefficient r Fibre/model 95% Cl Amphibole Linear (r-l) Best (re 1.3) Steepest (rel.6) Chrysotile* Best (r=l .3) Cautious model-max of: . Linear (rel) Steepest (rel',6) 4.8 1.6 0.49 0.028 0.5 0.039 Jt (1.2. 1.9) (0.37.0.62) *A linear model is not strictly statistically consistent with the observed .data. The line with AL=4.8 is the single best fit "Non-staiistical uncertainties dominate choice of chrysotile models, 95% confidence intervals cannot be properly calcu lated. See text for discussion. - is a fall beyond 50 yr. Qualitatively it seems clear that the risk does not increase indefinitely, but there is insufficient evidence on very long follow up to fix the risk profile in this period. A rough and ready way of limiting the predicted risk at very long follow up periods is to truncate the predictions at some age. The Doll and Peto and HEI reports both truncated their predictions at age 80, and we will follow this conven tion. It is likely that this would still overstate the risk from exposure at ages below 20, and truncation of the predicted effect at 60 yr follow up might then be appropriate. Lung cancer ' The data in Table 7 and Fig. 9 suggest that the relation between lung cancer and cumulative exposure may be concuve--i.e. that the excess lung cancer risk is proportional to a powec.greater than' I of cumulative exposure. Statistically* the range of powers consistent with all the amphibole data is from 1.1 to 2.1. Without the two extreme cohorts the range becomes 0.89-2.0 with a central estimate of 1.4. No previous analysis of the epidemiological data has sug- : gested a concave relationship,' though experimental data for a wide range of carcinogens (Hoel and Portier, 1995) suggest they may be quite common. Across the range of exposures in a single study, and given the uncertainties in individual exposure estimation, a moderate degree of non-linearity will be difficult to detect. The reasonably arguable values for r fall in the interval l to 2: a degree on conservatism and some doubts about the two extreme cohorts lead us to prefer the lower end of this interval. We will take r=l (a linear relationship) and r=l.6 to represent the flattest and steepest slopes for risk assessment, and the mid point of this range (r=l.3) os our best estimate assumption. The estimates and 95% confidence limits for the constant term AL in a model for lung cancer PL = ALr with rel (linear) 1.3, and 1.6 based on amphibole data are shown in Table 10. As already discussed, the inconsistencies in the pure chrysotile data rule out a direct estimate of the exposureresponse slope based on this data. The dominant uncertainties for chrysotile are the reasons tor the observed differences in exposure-specific lung cancer risk, rather than the statistical uncertainties in estimat ing this risk level. This uncertainty is already reflected in the five-fold difference between our 'best' and: 'cautious' estimates of Rt (0.1 and 0.5 respectively). (n the absence of a better approach we will assume i ..........-- ~..M' Quantitative rirki of mesothelioma and lung cancer 583 the same range of possible slopes for the chrysolite ' lung cancer relationship as for the amphiboles, and determine the scaling constant by fixing the predicted excess mortality at the median exposure for chrysotile cohorts (70 f/ml.yr) to 0.1 % for the best estimate and 0.5% for the cautious estimate. The resulting values are shown in Table 10. The pattern of excess lung cancer--broadly con stant relative excess from 10' to 40 (perhaps more) yean from exposure (see Appendix A) implies that for exposure starts between 20 and 40 yr of age there is very little difference in the predicted risk. There may be some decline for very long follow up, but the rate of decline is unknown. As for mesothelioma we . address this possibility approximately by truncating the predicted excess at age. 80. IS THERE A THRESHOLD? Another question with important implications for risk at low levels of exposure is whether, there is a threshold for cancer initiation by asbestos. The HSE's recent Review of fibre toxicology (Meldrum, 1996), presents arguments mainly on a toxicological basis t believing that there may be a threshold for asbesios induced lung cancer. The argument is essentially based- on a view of (he carcinogenic process induced by asbestos as being an extension of the chronic inflammatory processes producing fibrosis. It is widely' agreed that heavy doses of chrysotile are required to produce lung fibrosis. And some evidence has been de^Ved from the New Orleans cohort sug gesting a threshold dose of about 30 f/ml.yr for radio logical fibrosis (Weill, 1994). Analysis of necropsy material from the Carolina cohort also shows a dis tinct step increase in fibrosis score for cumulative exposures around 20-30 f/ml.yr (Green et al., 1997). This does not apply to omphibole exposure: radiologi cal fibrosis which progressed after the cessation of exposure has been documented (Sluis-Cremer, 1991), in South African amphibole miners under medical surveillance and with cumulative doses less than 5 f/ml.yr. This suggests that if a threshold applies to the lung cancer effect of amphibole asbestos, it is very "'' low. The adoption of a slightly concave exposure response slope entails a moderately threshold-like behaviour. - Several lines of argument also suggest that any threshold for mesochelioma is at a very low level. Some cohorts (Neuberger and Kundi, 1990; New- house and Sullivan, 1989; McDonald and McDonald, '978; Thomas et al., 1982: Rossiter and Coles, 1980), re produced mesotheliomas in conditions where no . excess lung cancer was seen. Occupational PMRs for British men suggest that the range of jobs for which mesothcliomu rotes ure above background levels is very wide (Hutchings el al., 1995: Hodgson et at., 1997). Also the proportion of mesothelioma cases in population studies for whom no likely source of asbestos exposure can be identified is often quite high. All these observations suggest that relatively brief exposures may carry a low, but non-zero, risk of causing mesothelioma. . Some authors (llgren and Browne, 1991; Liddell. 1993) have argued for a mesothelioma threshold, or threshold-like behaviour of the dose-response. Such arguments are fraught with statistical and logical dif ficulties. The attempt (llgren and Browne, 1991) to deduce a 'threshold' by identifying the lowest esti mated dose received by any observed case is a logical nonsense. Furthermore, the existence of zero cases in a dose category (human or animal) should not be automatically interpeted as zero risk. Direct statistical confirmation of a threshold from human data is vir tually impossible. One would need'accurate'assess ment of very low doses across a large population with long term follow up. Case-control studies with lung content measures of exposure (McDonald et al.. 1989; ROdelsperger et al., 1999; Rogers el al., 1991) do not suggest any threshold, or downward inflexion of the dose response at the lower end of their exposure scales. Some of the animal data cited by llgren and Browne are suggestive of a threshold-- particularly that from intra-pleural and intra-perito- neal injection--but it is not clear how this would translate into a estimated human effect threshold for exposure by inhalation. Taking this evidence together we do not believe there is a good case for assuming any threshold for mesothelioma risk. QUANTIFIED RISK ASSESSMENT Under current conditions, the main interest in the heal'h risks of asbestos relates to exposure circum stances well outside the range for which we have direct observations. The statements we can make about risk therefore incoiporate two kinds of uncer tainty. First there is the usual statistical uncertainty of inferring underlying risk from observations in particular groups. This kind of uncertainty depends essentially on the number of events (in this case can cer deaths) observed. The uncertainty can therefore-- given some assumptions--be quantified: the more observed events, the less the statistical uncertainty. Statistical uncertainty is expressed as a confidence interval (a range of values with--conventionally--a 95% probability of covering the true value). The second kind of uncertainty relates to the ques tion whether the relationship between exposure and outcome seen in the observed range continues to hold outside that range. This kind of uncertainty cannot be quantified statistically. Qualitatively one can reason ably argue that the agreement will be better for exposures close to. the observed range, but with increasing distance from the observed range our con fidence that we know whut tu expect decreases. For example, previous assessments of cancelr risk from asbestos have all ussumed that the effect is linear. i'1 ** !., 584 I. T. Hodgson and A. Damion This review has presented evidence suggesting that this may not be the case. Uncertainty about the slopes of exposure-response tines has an increasing impact with increasing distance from the observed range. by 0.054/>| . For women with typical past smoking habits the figure would be 0.028AL. Table 11 makes statements about the lifetime risks of exposures accumulated over short (up to 3 yr) per Also the strength of qualitative arguments such as those advanced in the HSE review (Meldrum, 1996), in favour of a threshold for the lung cancer effect increase as exposure falls. All the above implies that simply to present a table iods from age 30. The factors given in Table 10 can be used to apply the mesothelioma estimates to other ages at exposure. The lung cancer estimates are based on 1997 male lung cancer rates. They are not sensi tive to age at exposure. of risk estimates--or even risk ranges--for different For the lung cancer risk due to chrysotile two prin cumulative exposures cannot capture the changing cipal figures are given: a best estimate and a cautious balance of the different kinds of uncertainty. Table estimate. A risk estimate derived from the Carolina 11 gives a verbal assessment of risk at a range of cohort is also given, with the qualification that this representative cumulative exposures. No estimates might be arguable in 'exceptional circumstances'. have been given for lifetime risks lower than l in These exceptional circumstances cannot be defined 100 000, and this level is referred to as 'insignificant'. with any certainly since the features of exposure at A lifetime risk of 1 in 100000 corresponds to an this plant responsible for the very high lung cancer annual risk well below 1 in a million, which HSE has risks there ore not known. Exposure to textile grade suggested (Health and Safety Executive, 1999) as a (i.e. long fibre) chrysotile is presumably necessary, "guideline for the boundary between the broadly but does not seem to be sufficient, since other textile acceptable and tolerable regions (of fatal risk to an plants have recorded much lower exposure-specific individual]." It is also well below the level at which risk (even with additional exposure to amphibole it is suggested that mesothelioma would occur in the fibre). The spraying of the taw fibre with mineral oil absence of asbestos exposure: a clear majority of the (as a dust suppression measure) has teen suggested very few mesotheliomas that would occur at this level as a possible explanation. This hypothesis seems to would not be caused by asbestos. be supported by a case-control study of lung cancers Mesothelioma risks in the observed cohorts have at Carolina (though the relevant results have not teen been expressed as a percentage (Pm) of total expected fully repotted), and by observations from another mortality in order to standardise observations from asbestos textile plant (Rochdale), where men first different follow up configurations. To make predic employed after oil spraying was introduced had three tions of risk this measure must be converted back into times the exposure-specific risk of those first absolute terms, and this is done using the average employed in earlier periods (though still lower than male life table discussed in Appendix A. For the Carolina risk). exposures starting at age 30 the excess mortality esti The main uncertainties in this picture relate to the mate Pm is applied to the total expected mortality effects of chrysotile, particularly at low doses. The from age 40 to age 79 (allowing a 10 yr minimum application of these estimates in the assessment of a latency, and truncating risk at age 80). The life table particular risk situation will depend on the purposes predicts that about 70% of survivors to age 30 will of that particular assessment, and the extent to which die between the ages of 40 and 80. Absolute risk esti a precautionary approach is appropriate. mates can therefore be derived from the PM value for a given exposure by multiplying by a factor of 0.7. _ Lung cancer risks have been expressed as a percent DISCUSSION age excess of expected lung cancer mortality. The There have been a number of papers (Cullen, 1998: major determinant of this underlying lung cancer risk Stayner r a/., 1996; Nicholson and Landrigan, 1996; is smoking--especially cigarette smoking--and the Smith and Wright, 1996), in the literature recently number of asbestos-related lung cancers will be affec which directly or indirectly consider whether there ted by the prevalence of smoking in the exposed are differences in potency between the fibre types as population. Currently (in 1997) about 9.3% of male causes of mesothelioma and lung cancer. The claim deaths between the ages of 40 and 79 are due to lung that there are important differences is often described cancer. For women the figure is 7%, reflecting differ as `the amphibole hypothesis'. In its strongest form ences in past smoking. Total survival to age 80 is this has been said to claim that pure chrysotile (i.e. lower in men than in women, and combining data for without any associated tremolite fibre) would p/esent survival and proportionate mortality from lung cancer little or no carcinogenic risk. At the other extreme, it it can be predicted that for 1000 30-yr-old men 34 has teen argued (Smith and Wright. 1996), that there will die of lung cancer between the ages of 40 and is virtually no difference between the risks presented 79. For women the number is 28. Thus for a popu by the different fibre types. Most commentators (e.g. lation with the post smoking habits of British men Doll and Pcto, 1983; Hughes and Weill, 1986; Health aged 60+ (the ages at which most lung cancers occur), Effects Institute, 1991) have considered that the the lung cancer risk from asbestos exposure is given amphibole fibre types are more dangerous, parti- I Fibre Table 11. Summary statements of the quantitative cancer risks from asbestos exposure at different levels of cumulative exposure*><4 Mesothelioma Lung cancer Risk summaries for cumulative exposures between 10 and 100 t/inLyrs Crocidolite Best estimate about 400 deaths per 100 000 exposed for each f/ml.yr of cumulative Rising from about ISO (range 100 to 250) excess lung cancer deaths per 100000 exposure. Up to 2-fold uncertainty. exposed for each f/ml yr of cumulative exposure at 10 f/ml.yrs to 350 (range 250 to 550) at 100 f/ml.yrs. Amosite Best estimate about 65 deaths per 100000 exposed for each f/ml.yr of cumulative exposure. 2-fold to 4-fold uncertainty. Chrysolite Best estimate about 2'deaths per 100 000 exposed for each f/ml.yr of cumulative Best estimate about 5 excess lung cancer deaths per 100 000 exposed for each f/ml exposure. Up to 3-fold uncertainty. yr of cumulative exposure. Cautious estimate 30. In exceptional circumstances (see note c) it is arguable that an estimate of 100 might be justified. Risk summaries for cumulative exposures of I f/inl.yrs Crocidolite Best estimate about 650 dcatlts per 100000 exposed. Highest arguable estimate 1500. lowest 250. Best estimate about 85 (range 20 to 250) excess lung cancer deaths per 100000 exposed. Amosite Best estimate about 00 deaths per 100000 exposed. Highest arguable estimate 300. lowest IS. ' Chrysolite Best estimate about 5 deatlis per 100 000 exposed. Highest arguable estimate 20, Best estimutc about 2 excess lung cancer deaths per 100000 exposed. Cautious lowest I. . estimate 30 per 100 000. In exceptional circumstances (see note c) it is arguable tluit an estimate of 100 per 100000 might be justified. The case for a threshold--ic zero, or at least very low risk--is arguable. Risk summaries for cumulative exposures of 0.1 f/mLyrs Crocidolite Best estimate about 100 deaths per 100000 exposed. Highest arguable estimate 350. lowest 25. Amosite Best estimate about 15 deaths per 100000 exposed. Highest arguable estimate 80, lowest 2. Chrysolite Risk probably insignificant, highest arguable estimate 4 deaths per 100000 exposed. Risk summaries for cumulative, exposures of 0.01 f/mLyrs ^ Best estimate about 4 (range <1 to 25) excess lung cancer deaths per 100 000 exposed. Excess lung cancer deaths probably insignificant. Cautious estimate 3 per 100000. In exceptional circumstances (see note c) it is arguable that an estimate of 10 per 100 000 might be justified. The case for a threshold--ie zero, or at least very low risk--is strongly arguable. Crocidolite Amosite Chrysolite Best estimate about 20 deaths per 100 000 exposed. Higitesl arguable estimate 100. lowest 2. Best estimate about 3 death per 100 000 exposed. Highest arguable estimate 20, lowest insignificant. Risk probably insignificant, highest arguable estimate I deatlis per 100000 exposed. Risk is probably insignificant (range <1 to 3 excess lung cancer deaths per 100 000 exposed). Mesothelioma is now the dominant risk, so precise estimation of die lung cancer risk is not critical. Risk of excess lung cancer very probably insignificant except in exceptional circumstances (see note c) when it is arguable that an estimate of. I death per 100000 might be justified. The case for a threshold--ie zero, or at least very low risk--is strongly arguable. ' (Continued an next /*>ge) oa Quantitative risks o f mesothelioma and lung cancer I .I I I i ' ' I Wr. < uonucQ -y pue uosSpoH 7l Table 11. (continued) Fibre Mesothelioma Lung cancer Risk summaries for. cumulative exposures of 0.005 f/mLyr and lower At these levels only mesothelioma need be considered. The absolute risk is low--, but quantitative uncertainties'are very considerable. Crocidoliie Best estimate about 10 deaths per 100000 exposed. Highest arguable estimate 55, Insignificant, possibly zero lowest. Best estimate fulls to insignificant level at 0-0002 f/ml.yr, and highest arguable risk becomes insignificant at 6x10'" f/uil.yr Amositc Best estimate about 2 deaths per ItJOUOO exposed liigliest arguable lifetime risk 15, falling to <1 (ie. insignificant) at 7x'IO*s t/ml.yr Chrysolite Insignificant , Insignilieanl, very possihjy zero `Exposure assumed to be accumulated over short--up to 5 yr periods storting ul age 50. Isir exposure at oilier ages adjust llto predicted mesothelioma figures using die factors in 'lalile 9. Estimates for longer periods of exposure can be approximated by making separate estimates for successive 5-year periods and adding die resulting risks (this will slightly overestimate , risk). Estimates have been rounded to nearest 5 in second significant digit (or to one significant digit wlicn less than 10).' , ' . "The lung cancer risk Is based on British male mortality in 1997 when 9.5% of male deaths at ages 40-79 were due to lung cancer. Tills represents an average for a imputation with a past pattern of smoking similar to that of older British men. lit 1996 23% of men aged 60+ had never (or only occussiotuilly) smoked, and 25% were current smokers, fur lifetime smokers the lung cancer risk will be about double the stated levels, for non-smokers about a sixth (if the interaction with asbestos is multiplicative) or about a third if the relative risk is higher than in non-smokers os suggested by Berry el at. (1985). "The lung cancer risk arguable in `exceptional circumstances' is derived from the Carolina cohort using a .value of R,. of 2.19 taken from the analysis of Stayncr ri id. It diuuld only tic considered where there is simultaneous exposure to textile grade (i.e. long fibre) chrysolite and mineral oil or some analogous co-exposure. The simple pro rata formulae proposed in this table do not take account of the impact of competing causes of mortality. The impact will he trivial so long as lire predicted asbestos related mortality is low, and limited for predicted (individual cause) mortality below about 30 pereem. Above litis level the individual asbestos related diseases (including asbestosis. which is not covered by this analysis) will reduce each other's observed impact. In this situation all that cun usefully be predicted is that total asbestos related mortality will be very- high indeed. v Quantitative risks of mesothelioma and lung cancer 587 cularly for mesothelioma, but some (Cullen. 1998: Stayner el al., 1996) have regarded the extent of these differences as unimportant, particularly since chryso lite has been overwhelmingly the most commonly used fibre. The interpretation of the whole body of evidence depends importantly on the interpretation of results from cohorts with predominantly chrysotile exposure together with a minority contribution--usually a few per cent--from araphiboles. As long as the difference in potency is not extreme these cohorts con be reason ably interpreted as indicating the risk of chrysotile exposure. But if the differences in potency are very substantia) this is no longer the case. Furthermore, in this situation an additional source of error in the estimation of exposure will be introduced, since the measured exposure (mainly of chrysotile) will often be a poor proxy for the relevant exposure. The data in this review suggest that order of magni tude differences in potency may indeed apply for mesothelioma, and probably also for lung cancer. The main reason this review differs from earlier similar reviews is in its use of the information from the amphibole mining cohorts in South Africa and Aus- alia. The publication of mortality results from the oouih African mines seems to have gone almost unnoticed. The Australian cohort has been the subject of a series of publications with varying analytical approaches and varying results. One of these analyses gave a lung cancer risk from the cohort of around I % per fibre/ml.yr, and this is the value that has been most usually|^oted, but this is probably an underesti mate due to incomplete follow up at older ages. This review is also the only one to have exploited the (admittedly uncertain) quantitative exposure infor mation in the Massachusetts cohort Implications of the non-linear exposure response for mesothelioma A non-linear relationship between the rates of pleu ral and peritoneal mesothelioma is more readily explicable if the cancer risk is proportional to some function of the concentration of fibres in the target tissue, rather than the simple number burden. If concentration rather than number burden is the relevant parameter, then the possibility of a threshold type relationship becomes much more plausible, since if the effect depends on fibres acting together, there must presumably be some point at which individual fibres ore simply too far apart to exert any joint efTed. Of course, if the mechanisms of distribution of fibres thin the lung and pleura are such that fibres tend .0 be delivered preferentially to particular areas--and there is evidence that this is the case in the pleura (Boutin et al.. 1996)--the effective threshold level may be very low. In any case such a threshold is unlikely to be a sharp cut-off. Random variations in the distribution of fibres in particular lungs, und dif ferences in individual susceptibility will mean that the exposure response curve simply starts to descend more steeply from some point on the cumulative exposure scale. Also, fibre concentration is the more plausible exposure metric for the production of fibrosis, so this interpretation is consistent with the link suggested by the HSE fibre review (and by other authors) between the two processes. It should be noted that the sugges tion is not that tumours arise directly from fibrosis, but that both are products of an underlying inflamma tory process. If fibre concentration in tissue is the k?y risk meas ure, the extreme sensitivity in animal experiments to intra-peritonea! and intra-tracheal instillation of mass ive fibre doses is also readily explicable. . Combined with the knowledge of the much greater solubility of chrysotile in the lung, this may also explain why asbestos related diseases have only been clearly seen with, heavy chrysotile exposures. If exposures are heavy and sustained a sufficient con centration of fibre in the lung may be maintained to trigger both fibrosis and malignancy. The extreme rar ity of peritoneal mesothelioma in cohorts exposed to chrysotile alone may also be explained. If the route by which asbestos reaches the peritoneum is from the pleural cavity, it may well be that chrysotile fibres do not survive long enough in body tissues to make the journey in sufficient numbers. Chrysotile and asbestos related malignancy Smith and Wright (1996), showed a ranking of 25 cohort studies by proportional mortality from pleural mesothelioma and argued that since chrysotile was the primary exposure for two of the top 10 cohorts and present as part of the mix in six of them, and that the picture for crocidolite jn terms of its presence in the mix was similar, while amosite was less evident than either of the other two fibre types, that chrysotile must therefore be similarly potent as a cause of pleu ral mesothelioma. What this argument ignores is any quantification of exposure. Without quantification it is very difficult to draw any conclusion about relative risk from a simple ranking by mesothelioma rate. In relation to (he 25 cohorts identified in this review an equally pertinent observation might be that all of them involved exposure to one or other of the amphi bole fibres. Smith and Wright also present arguments based on the relative levels of mortality from pleural mesothelioma and from excess lung cancer to suggest that there is only moderate difference between the potency of chrysotile and the amphibole fibres for causing mesothelioma--they suggest a factor of three or four. However this argument is based on the assumption that all fibre types are equally potent for lung cancer. If this review is correct in suggesting that this is not the case, these arguments are not valid. Nicholson and Lundrigan (1996). present similar 588 J. T. Hodgson and A. Oamton arguments based on the assumed equivalence of the required to explain the data. Low levels of amphibole fibre types to cause lung cancer. They also show un do have a disproportionate effect. analysis of the mesothelioma mortality of a small sub set of the US insulators study which shows that the pattern of deaths over time implies that members of this cohort were exposed to a pleural carcinogen before 1935. Since, reportedly, amosite was first used from around 1935, and prior to this date only chrysotile was used, some of these deaths must have been due to chrysolite. The authors do not mention the possible role of crocidolite, but if we accept that no amphibole fibre was used before 1935 by US insu lation workers, these observations do show that some REFERENCES Acheson. E. 0.. Gardner. M. J.. Pippard. E. C and Grime. L. P. (1982) Mortality of two groups of women who manufac tured gas masks from chrysolite and crocidolite asbestos: a 40-year follow-up. British Journal of Industrial Medicine 39. 344-348. Acheson. E. D.. Gardiner, M. I.. Winter. P. D. and Bennett. C. (1984) Cancer in a factory using amosite asbestos. Inter national Journal of Epidemiology 13. 3-10. Albin. M.. Jacobson, K.. Attawcll. R,, Johannson. L. and Wel- of the cases in this cohort must have been caused by exposure to chrysolite prior to 1935. These exposures will often have been heavy. In fact the contemporary lindcr. H. (I990u) Mortality and cancer morbidity ia cohorts of asbestos cement workers and referents. British Journal of Industrial Medicine 47. -602-610. Albin, M., Johansson. L.. Pooley, F. 0., Jakobsson. K., Allow US trade journal `Asbestos' (published monthly from ed. R. and Mitha, R. (1990b) Mineral fibres, fibrosis and July 1919) makes it clear that both amosite and cro cidolite were used in the US through the 1920s, though probably in limited quantities, since the US industry seems to have been resistant to their use on asbestos products in the lungs from deceased asbestos cement worker*. British Journal of Industrial Medicine 47. 747-774. Armstrong. B. K.. de Klerk, N. H.. Musk. A. W. and Hobbs. ' M. S. T. (1989) Cancer mortality in relation to measures of technical grounds, and chrysolite was the fibre of choice for most applications. . Slayner er al.'i review of the issues (1996), sets occupational exposure to crocidolite at Wittenoom Gorge in Western Australia. British Journal of Industrial Medicine 46. 329-536. . Berry G. (1991) Prediction of mesothelioma.rPung cancer, and out similar arguments, and also points to the evidence asbestosis in former Wittenoom asbestosworkers. British that all three commercial fibre types have produced a similar level of lung tumours in animal inhalation experiments. This is the most problematic evidence to reconcile with the human evidence that amphibole Journal of Industrial Medicine. 48. 793-802. Berry. G. (1995) Environmental mesothelioma incidence, time since exposure to asbestos- and level of exposure. Environmetrics 6, 221-228. Berry. G. (1999) Models for mesothelioma incidence following fibres are substantially more potent lung carcinogens. However, the time periods needed to induce cancer in humans (yr) and in rats (months) are very different, and it is at least plausible that all fibres are equally exposure to fibres in terms of timing and duration of exposure and the biopenistenee of the fibres. Inhalation Toxicology 11. 101--120. Betty. G. and Newhouse, M. L. (1983) Mortality of workers manufacturing friction materials using asbestos. British potent in rats because none of them ore materially . cleared from the rat lung over the months needed to. initiate a rat lung tumour. By contrast, in humans Journal of Industrial Medicine 40, 1-7. Beny. G., Newhouse. M. L. and Antonis. P. (1985) Combined effect of asbestos and smoking on mortality from lung cancer and mesothelioma in factory workers. British Journal of chrysolite (cleared in months) might have less effect Industrial Medicine 42, 12-18. than the amphibole fibres (cleared in yean). A detailed elaboration of this argument has recently been published by Beny (1999). It may also be rel Boutin. C- Dumortier, F- Rey. F, Violin. R. and DeVuyst. P. (1996) Black spots concentrate oncogenic asbestos fibres in the parietal pleura-thoroscopic and mineralogies! study. American Journal ofRespiratory and Critical Care Medicine evant that the animal experiments were made with ~ 153. 444-449. exposure concentrations massively in excess of those represented m the human data. The differences in the human data summarised in this review seem reason Coggon, D., Inskip. R, Winter. P. and Panned, B. (1995) Dif ferences in occupational mortality from pleural cancer, per itoneal -cancer and asbestosis. Occupational and Environ mental Medicine 52, 775-777. ably clear (certainly in respect of mesothelioma), and Cullen. M. R. (1998) Chrysotile asbestos: enough is enough. are based on a range of independent data. In the end, if a choice has to be made between animal and human evidence as a basis for assessing human risk, adequate human data must be given priority. Lancet 351. 1377-1378. de Klerk. N. H,, Musk. W. A.. Armstrong. B. K. and Hobbs. M. S. T. (1991) Smoking, exposure to crocidolite, and Use incidence of lung cancer and asbestosis. British Journal of Industrial Medicine 48. 4(2-417. Many of the arguments presented against the `amphibole hypothesis' in connection with mesotheli oma are variants on the basic theme that it is simply unbelievable that such a small component of the de Klerk, N. R, Musk, A. W. and Armstrong, B. K. (1992) The epidemiology of asbestos and mesothelioma. In Malignant Mesothelioma, eds D. W. Henderson, K. B, Shilkirn S. L Longlois and D. Whitaker, pp. 223-250. Hemisphere Pub - lishing Corp, New York. exposure could be responsible for the observed risk. If it is true that the mesothelioma risk is proportional . to a less than unit power of exposure, then these argu ments are correct in their basic perception that a dis proportionate effect of the amphibole component was de IGetk, N. R, Musk, A. W,, Armstrong, B. K. and Hobbs. M. S. T. (1994) Diseases in miners and millers of crocidolite from Wittenoom, Western Australia: a further follow-up to December 1986. Annals of Occupational Hygiene 38(supp I). 647-655. Dement. J. M. (1991) Carcinogenicity of chrysotile asbestos: Quantitative rirks of mesothelioma and lung cancer 589 evidence from cohon studies. Annals of Hew York Academy of Science 643. 15-23. Dement. 3. M., Brown. D. P. and Okun. A. (1994) Follow-up study of chrysotile asbestos textile workers: cohon mortality and .cue-control analyses. American Journal of Industrial Medicine 26, 431-447. Doll, R. and Peto, J. (1985). Effects on Health Exposure to Asbestos. HMSO. London. Ernestine, P. E. Harley, I. and Henderson, V. (1987) Asbestos and cancer a cohon followed up to death. British Journal pf Industrial Medicine 44, 396-401. Finkelstein, M. M. (1984) Mortality among employees of an Ontario asbestos-cement factory. American Review of Res piratory Disease J29. 750-761. Finkelstein, M. M. and Vingilis. J. I. (1984) Radiographic abnormalities among asbestos cement workers: and exposure response study. American Review of Respiratory Disease 129, 17-22. ` Finkelstein MM. (1989) Mortality among employees of an Ontario factory manufacturing insulation materials from . amosite asbestos. American Journal of Industrial Medicine. IS. 477-481. Green, F. H. Y., Harley, R., Vallyathan. V., Allhouse. R,, Fick. G, Dement, J., Milha, R. and Pooley, F. (1997) Exposure and mineralogies! correlates of pulmonary fibrosis in chryso lite asbestos workers. Occupational and Environmental Medicine 54, 549-559. Health and Safety Executive (1999). Reducing Risks. Protect ing People--Discussion Document. Health and Safety .) Executive. ealth Effects Institute (1991). Asbestos in Public and Com mercial Buildings: A'.Literature Review and Synthesis of Currym Knowledge. Health Effects Institute--Asbestos Research. Cambridge. MA. Hilt B, Rosenberg J. Langard S. (1981) Occurrence of cancer in a small cohon of asbestos-exposed workers. Scandanavian Journal of Work. Environment and Health. 7, 185-189. Hodgson, J. T., Peto, J., Jones, J. R. and Manhews, F. E (1997) Mesothcliorm mortality in Britain: patterns by birth cohon and occupation. Annals of Occupational Hvgiene 41(supp 1). 129-133. Hoel, D. J. and Portier, C J. (1995) Non-linearity of dose- response functions for carcinogenicity; Environmental Health Perspectives 102(supp I), 109-113. Hughes. J. M. and Weill, H. (1986) Asbestos exposure--quan titative assessment of risk. American Review of Respiratory Disease 133, 5-13. Hughes. J. M., Weill, H. and Hammad. Y. Y. (1987) Monality of workers employed in two asbestos cement manufacturing plants. British Journal of Industrial Medicine 44, 161-174. llgren, E B. and Browne, K. (1991) Asbestos related meso thelioma: evidence for a threshold in animals and humans. Regulatory toxicology and pharmacology 13, 116-132. INSERM (I996) Effect sur la Xante des principaux types d'ex- posilion a I'amiante. Paris. ' Jones, J. S. P., Gibbs, A. R.. MacDonald. J. C and Pooley. F. D. (1996) Mesothelioma.following exposure to crocidolite (blue asbestos). A 50 yew follow up study In Proceedings of second International congress on lung cancer, pp. 407 411. Monduzzi Editorc SpA, Bologna, Italy. Longer, A. M. and Nolan, R. P. (1989) Fibre type and burden in parenchymal (issues of workers occupationally exposed to psbesto* in the United States. In Non-occupational Exposure pa Mineral Fibres. Scientific Publication no. V0, eds J. Bignon and J. Peto, pp. 330-335. IARC, Lyon. Lash. T. L, Crouch, E ^ C. and Green, L. C (1997) A meta analysis of the relation between cumulative exposure to asbestos and relative risk of lung cancer. Occupational and Environmental Medicine 54, 254-263. Levin. J. L., McLarty. J. W,, Hurst, G. A., Smith. A. N. and Frank. A. L. (1998) Tyler asbestos workers: mortality experi ence in a cohort exposed to amosite. Occupational and Environmental Medicine JJ, 155-160. Liddell. F. D. K. (1993) Exposure-response: asbestos and mesothelioma. European respiratory review 3, 98-99. Liddell, F. D. K., McDonald, A. D. and McDonald, J. C. (1997) The 1891-1920 cohort of Quebec chrysolite miners and mil lers: development from 1904 and monality to 1992. Annals of Occupational Hygiene 41, 13-36. Liddell. F. D. K.. McDonald. A. D., McDonald, J. C. (1998) Dust exposure and lung cancer in Quebec chrysolite miners and millers. Annals of Occupational Hygiene. 42. 7-20. McDonald, A. D.. Fry, J. S., Woolley. A. J. and McDonald, J. (1983) Dust exposure and monality in an American chryso lite textile plant. British Journal of Industrial Medicine 40. 361-367. McDonald, A. D.. Fry. J. S., Woolley, A. J. and McDonald. J. C. (1983) Dust exposure and monality in an American fac tory using chrysotile. amosite, and crocidolite in mainly tex tile manufacture. British Journal of Industrial Medicine 39. 368-374. . McDonald. A. D., Fry, J. S., Woolley, A. J. and McDonald. J. C. (1984) Dust exposure and monality in an American chrysotile asbestos friction products plant. British Journal of Industrial Medicine 46. 151-157. McDonald, A. D. and McDonald, J. C. (1978) Mesothelioma after crocidolite exposure during gas mask manufacture. Environmental Research 17, 340-346. , McDonald, J. C. Armstrong. B., Case, B.. Doell, D., McCsughey, W. T. E.. McDonald, A. D. and Sebastien, P. (1989) Mcsothcliama and asbestos fibre type: evidence from lung tissue. Cancer 63, 1544-1547. Meldrum, M. (1996) Review of Fibre Toxicology. HSE Books. Sudbury. UK. Miller. B. G.. Jones. A. D.. Searl, A., Buchanan. D., Cullen, R. T,, Soutar. C. A., Davis, J. M. G. and Donaldson. K. (1999) Influence of characteristics of inhaled fibres on the development of tumours in the rat lung. Annals of Occu pational Hygiene 43, 167-179. Neuberger. M. and Kundi. M. (1990) Individual asbestos exposure: smoking and monality--a cohon study into the asbestos cement industry. British Journal of Industrial Medi cine 47. 615-620. Neuberger, M. and Kundi, M. (1991) Zum Berufskrebsrisiko durch Asbest in Osterreich. In 2. Kotloquium air der Krebsgefahr durch Asbest Bundesgesunheitsamt, Berlin. Neuberger, M. and Kundi. Mr (1993) Cancer in asbestos cement production and implication for risk management. In: Proceedings of the 24th Congress of the International Com mission on Occupational Health, 26 September-1 October, Nice, 1993. International Commission on Occupational Health and French Ministry of Labour. Employment and Professional Training. Ncwhouse. M. L. and Sullivan, K. R. (1989) A mortality study of workers manufacturing friction materials: 1941-86. Bri tish Journal of Industrial Medicine 46, 176-179. Nicholson, W. J. and Landrigan. P. J. (1996) Asbestos: a status report.' Current Issues In Public Health 2, 118-123. Hutchings, S., Jones, J. and Hodgson. J. (1995) Asbestos related disease. In Occupational Health Decennial Sup plement. Office of Population Censuses and Surveys/Health and Safety Executive {series DS no. J0l. ed. F. Drever. HMSO. London. Parolari, G., Merler. E. Bertazzi. P.A.. Zocchelli. C.. Camev- ale, F.. Berrino. F. (19*7) Gli effetti dell'esposizione ad amianto amosite tra gli ex lavoratori di una fabbrlca di coibenti: 2. Indagine epidemiologica suite cause di morte. Poges 123 134 in Parolari. G,, Gherson. G.. Cristofolini. A., Merler. E (Edx) II rischio neoploxlico da amianlo neu luoghi di lavoro e ncll'ambienie di vita. Verona. Bi & Gi Editore. Peto. J.. Doll, R.. Hermon. C., Binns, W., Clayton. R. and Goffe, T. (1985) Relationship of mortality to measures of / V. iiiw. 590 I. T. Hodgson and A. Damron environmental asbestos pollution in an asbestos textile fatlory. Annuls uf Occupational Hygiene 29. 305-335. Appendix A -| Piolatto, G.. Negri. E.. La Vecchia. C.. Pirn. E.. Occarli. A. and Peto, i. (1990) An update of cancer mortality among chrysenile asbestos miners in Balangtro. northern Italy. Bri EXTRACTED COHORT DATA tish Journal of Industrial Medicine 47. 8IO-SI4. Rddelspcrger. K.. Woitowiiz, H. -I.. Brtlckel. B.. Arhelger. R.. Pohlabein. H. and idckel. K. -H. (1999) Dose response The details of extracted data is shown in Tables 12 and 13. with explanatory notes in Tables 14 and 15. relationship between amphibole Hber lung burden and meso thelioma. Cancer Detection and Prevention 23. 183-193. Rogers. A. (1990) Cancer modality and exposure to crocidolitc. British Journal of Industrial Medicine 47. 286. Rogers. A . Leigh.).. Berry. G., Ferguson. 0. A.. Mulder. H. B. and Ackud. M. (1991) Relatioaship between lung asbestos SUMMARISING MORTALITY AND EXPOSURE MEASURES - THE CHOICE OF FOLLOW UP PERIOD IN RELATION TO EXPOSURE PERIOD fibre type and concentration and relative risk of mesotheli oma. Cancer (17, 1912-1920. LUNG CANCER Ross iter. C. E. and Coles, R. M. (1980) KM Dockyard. Devon- For most of the studies considered in this review, pod: 1947 modality study. In Biological Effects of Mineral Fibres. Scientific Publication no. 30. ed. J. C. Wager, pp. 713-721. (ARC. Lyon. Slbastien. P.. McDonald. J. C., McDonald. A. O.. Case. B. and Harley. R. (1989) Respiratory cancer In chrysolite textile and mortality was reported for the period 20 or more years from first exposure. The reason for this is that it is assumed that this represents a period in which the effect of exposure will be fully expressed. Deaths mining industries: exposure inferences from lung analysis. British Journal of Industrial Medicine 4<. 180-187. Scidman. H. and SelikofT. I. ). (1990) Decline in death rates among asbestos insulation woken 1967-1986 associated with diminuation of work exposure to asbestos. Annals of New York Academy of Science 609. 30(3-318. Seidman. H.. Selikoff, I. /. and Gelb, S. K. (1986) Mortality experience of amosite asbestos factory workers: doseresponse relationships 5 to 40 yean after onset of shod-term work exposure. American Journal ofIndustrial Medicine 10. 479-514. Sluis-Cremer. G. K..099I) Asbestos disease at low exposures Observed fn the period immediately following first exposure will be unaffected by that exposure, and the effect of exposure will become progressively more apparent as follow up time increases. Comparisons of observed and expected deaths that include periods immediately after first exposure wilkintroduce some downward bias to the assessed riskievel. Evidence on the levels of excess in different per iods after exposure suggests that between (0 or 20 and about 40 yr from exposure the lung cancer r.' \ after long residence times. Annals of New York Academy of Science 643. 182-193. Sluis-Cremer, G. K.. Liddell. F. D. K.. Logan, W. P. D. and Bczuidenhout. B. N. (1992) The modality of amphibole minen in South Africa. 1946-80. British Journal of Indus trial Medicine 49. 566-575. Smith, A. H. and Wright. C. C. (1996) Chrysotile asbestos is the main cause of pleural mesothelioma. American Journal of Industrial Medicine 30. 252-266. Stanton. M. F.. Layard, M.. Tegeris, A.. Miller, E-, May, M.. Morgan, E. and Smith, A. (1981) Relation of particle dimen sion to carcinogenicity in amphibole asbestos and other is reasonably stable. Beyond 40 or 45 yr follow . there may be some decline,in risk, but the extent to which this may have diluted recorded lung cancer risk in these cohorts seems limited (for example, there is very little difference between the US/Canada insu lators lung cancer SMR calculated over all follow up from 20 yr and one restricted to the period between 20 and 40 yr). No adjustment for differences in maxi mal follow up between cohorts has therefore been applied. fibrous materials. Journal of the National Cancer Institute The impact of follow up less than 10 yr being 67. 965-975. Stayncr, L. T- Donkovlc, D. A. and Lemen. R. A. (1996) Occu pational exposure to chrysotile asbestos and cancer risk: a review of the amphibole hypothesis. American Journal of Public Health 86. 179-186. included in the reported results, and the related prob lem of choosing on average exposure appropriate to the observed mortality needs to be considered. If . observed and expected mortality from observadons Stayner. L.. "Smith. R,, Bailer. J,, Gilbert. S.. Siccnland, K.. . less than 10 yr from first exposure are uninformative Dement. I.. Brown, D. and Lemen. R. (1997) Exposureresponse analysis .of risk of respiratory disease associated with occupational exposure to chrysotile asbestos. Occu pational and Environmental Medicine 54. 646-665. Talcott, I. A., Thurber. W. A- Kantor. A. F.. Oaetuler, E A.. of the possible effects of the exposure the inclusion of such observations in the reported results for a cohort, will dilute any actually occurring effect Pro vided there is a reasonable amount of informative fol Danahy, I. F.. Amman, K. A. and U, F. P. (1989) Asbestos-' low up on every cohort member this dilution effect associated diseases in a cohort of cigarette-filter workers. New England Journal of Medicine 321, 1220-1223. Thomas. H. F.. Benjamin. I. T.. Elwood. P. C. and Sweetium. P. M. (1982) Further follow-up study of workers from on asbestos cement factory. British Journal of Industrial Medi cine 39. 273-276. can probably be ignored, since the observed and expected deaths generated from the early (uninformative) follow up will' be outweighed for all individuals by their later observations. But if a high proportion of a cohort generates mainly uninforma Vainio, H. and Bofetta. P. (1994) Mechanisms of the combined effect of asbestos and smoking in (he aetiology of lung can cer. Scandinavian Journal of Work Environment and Health 20. 235-242. Weill, H. (1994) Biological effects: asbestos cement manufac turing. Annals of Occupational Hygiene 38. 533-338. tive follow up. reported overall results may be seri ously distorted. This con happen if recruitment to. cohort continues to the end of follow up. Subject, starting within 10 yr of the end of follow up will con tribute no informative mortality data. Quantitative risks o f mesothelioma and lung cancer Cohort identification Process Table 121 Summary of cohort (liba (part I )* Fibre Sc* Average' Typical Predicted mortality Number age at' follow for typical follow up of yean first up" from avg. age lint latency exposure ' exposed* Deaths by all causes Mesothelioma deaths Code Name lleferenee Proportion Relative to age at fust exposure . of 30 Ohs Exp SMR Total 9k Excess number mortality 1 Whtenoom de Klerk et at. (1994) 2m Carolina Dement rt at. (1994) M o m 29 35 0.24 1.08 10 719 601.8* 1.19 72* 12 T y* m 26 41 0.32 1.34 IS 607 410.1 1.48 2 0.49 (men) 2f Carolina (women) T y* r 26 41 0.32 1.34 15 362 299.2 1.21 0 0 J Johns Entcrline et at. (1987) hlanville 1 yao in 37 65 1 1 age>65 944 762J 1.24 8 1 retirees 4 Ontario Finkclstcin (I9K41 5a New Oilcans Hughes ci at. (I*>H7) C y in 27 0.17 0.77 20 108 62.2 1.73 17 27 C yo* in 32 35 0.3 0.83 20 259 294.5 0.88 1 0.34 (plant 1) 5o New Orleans c yo m 27 38 0.27 1.26 20 603 614.1 0.98 3 0.49 (plant 2) b Quebec Liddell el at. (19971 7 Vocklahruck Neuhergcr and Kundi yM in 23 72 1 1 age>55 6S87 5912.7* 1.11 33* 0.56 c yo inf 24 65 0.95 1.05 20 540 530.2* -- 5 0.94 (1990) a US/Caiiada Seklntan and SctikufT L yao* in 24 62 0.9 13)9 20 4626 3170.6 1.46 453 14 insulator* 0990) 9 Rochdale Peto n at. (1983) T yo* in 30 42 0.45 1 20 727 602.5* 1.21 10 1.7 10 Balangero Piolalto et at. (1990) yM in 27 47 0.31 1.2 20 317 225.4 1.41 2 0.89 II Pennsylvania McDonald et at. <l981a,b) TF ya* ni 29 40 0J5 13)8' 20 895 821.1 1.09 14 1.7 12 Paterson Seidman et </. (1986) 1 D in 37 40 0.6 0.63 5 593 355.9 1.67 17 4.8 IJa SA amosite Sluis-Cremer et at. (1992) M a m 31 40 0.4 0.93 0 648 456.3* 1.42 4 0.88 mines 1 Jo SA M 1o in 31 40 0.4 0.93 0 423 333.1* 1.27 20 6 crocidolite mines 14 Massachusetts Talcott ft at. (1989) IS Alhin Albin et at. (I990a.b) 16 Connecticut McDonald el at. (1984) 17 Ferodo Newhouse and Sullivan O 0 m 34 37 0.4 0.74 0 28 8.3 3.37 5 60 c yao* in 30 62 0.98 1 20 ' 592 493* 1.2 13 2.6 F y* in 31 40 0.41 0.93 20 557 . 550.7 1.01 0 0 F yo* tnf 30 45 0.56 1 10 2577 2646.3 0.97 13 0.49 (1989) `Notes un starred entries in Table 14. "Follow up duration which divides the observation licld beyond the minimum latency into two equal areas. *Sec text. I* j !i i Cohort code Lung cancer death* - Observed Diagnostic Expected basis (see . codes below*') SMR Table 13. Summary of cohort data (pari 2)* Excess %F Average Oasis for cumulative average exposure exposure (f/ml-yr) estimate (see codes . below"') Lung cancer risk expressed as % expected lung cancer per f/ml.yr Mesothelioma risk expressed as Coltort Average SMR regression slope Regression slope internal % Total expected Estimated mortality per I'/nil.yr Ui,,) A',,, in IIEI Unadjusted Adjusted for age al first exposure i. T. Hodgson and A. Damton 1 2m 2f 3 4 Sa So 6 7 8 9 10 II 12 13a l3o 14 IS 16 17 87 74 38 73 22 21 73 587* 47* 934* 56* 19 50* 98* 21* ' 19* 8 35 49 241* DC DC DC DC BE DC DC DC DC BE DC DC DC BE DC DC DC BE DC DC 48.7 32.2 13.8 28.4 5.3 22.5 50.1 431.6 42.2 2S6.8 37.1 17.3 33.8 20.5 14.5* 10.2* 0.6 19.4* 35.8 242.5 1.79 2.3 2.75 2.57 4.14 1 0.93 1.46 1.36 1.11 3.64 1.51 l.l 1.48 4.78 1.45 1.86 13.1 1.8 IJ7 0.99 38.3 41.8 24.2 44.6 16.7 -1.5 22.9 155.4 4.8 677.2 18.9 1.7 16.2 77.5 6.5 8.8 7.4 15.6 13.2 -1.5 79 130 175 157 314 -6.7 46 36 II 264 51 10 48 378 45 86 1210 80 37 -0.6 23* 28 26 750 60 79 47 600 25* 500 138* 300* 60 65 23.6* 16.4* 120* 13 46 35* ind CatAv CatAv CatAv tcxl CatAv CatAv CalAv ind L*D CatAv CutAv CatAv CutAv text text L*D text CatAv con 3.4 4.6 6.7 0.21 5.2 0 0.97 0.060* 0.45 0.53 0.37 0.03 0.80 5.8 1.9 5.2 10 6.2 0.80 0 3.1* 3* 1* 0.18 _* 0.03 0.76 0:037* - . 0.51 2.7* _* - 0.7 to 5.4* - - - - - - 0.012* - - - .0 0.058* 0.52 0.017 0 0.0014 0.53 0.0024 0.0069 0.0009 0.038 0.029 0.022 0.003 0.029 0.073 0.056 0.55 0.5 0.18 0 0.014 0.48 0.013 0 0.0014 0.68 0.0028 O.IXI53 0.0009 0.036 0.027 0.022 0.1X125 0.027 0.117 0.06 0.59 0.68 0.18 0 0.014 13.8 - - - 13 - ami - 1.7 0.9 - - 3.0 2.7 34 - `Notes on starred entries in Table IS. ^ *DC: Death certificate; BE: Best evidence: CatAv: Exposure category mid points weighted by expected mortality; ind: Mean of individual dose distribution (person weighted); L*D: Mean exposure level times mean exposure duration; text: Value given in text, assumed to be mean of individual dose distribution (person* weighted); con: Mean exposure of controls matched to lung cancer cases. 1 Quantitative risks o f mesothelioma and lung cancer Cohort identification General cohort notes Table 14. Explanatory notes on . t data Fibre type . Deaths By All Causes Mesothelioma mortality Code Name Obs fit. Exp 1 Wiltenoom 2m Carolina (men) 2f Carolina (women) Deaths to age 65 (evidence of incomplete death ascertainment a older ages) Very small amounts of crocidolite yam were used, but raw crocidolite fibre was not processed. The quantity of crocidolite used was about 0.002% of the total 3 Johns Manville retirees 4 Ontario Production and maintenance workers only 5a New Orleans (plant 1) Exd. men with <3 months Mainly chrysolite, amositc used from early 1940s; crocodolite used occasionally Two additional employment So New Orleans (plant 2) Exd. men with <6 months from 1962 (too recent to affect observed mortality) Mainly chrysolite, crocidolite used in pipe production process mesotheliomas at <20 yr latency. A further four employment occurred after end of formal follow up 6 Quebec Excluding Asbestos factory, but induding cases at ages <55 and the smaller Thetford mines Table 9 and text 7 Vocklabruck Observed less 5 mesotheliomas and 4.8 excess lung cancers 8 US/Canada insulators Nicholson and Landrigan (1996) estimate the exposure to have been 60% chrysolite and 40% amositc, based on published product compositions. Analyses of lung burden in lungs from 16 US insulation workers given by Langer and Nolan (1989) found amositc in all 16. chrysolite in half of them and crocidolite in 9 Rochdale three. - Predominantly chrysolite, Expected all cause deaths cannot be restricted to men with >1 yr employment but from mid 1930s a excess mesothelioma - and mean exposure - therefore expressed for whole consitent 5% or so was cohort crocidolite. 10 Balangero II Pennsylvania About 10% of the fibre processed was amositc. Very small amounts of crocidolite were also used, but little was handled as raw fibre (.Continued on next page) uo u J. T. Hodgson and A. Daemon Table 14. (continued) Cohort identification Code Name General cohort notes > Fibre type Deaths By All Causes Obs Exp Mcaxulniioma inonuliiy 12 Paterson 13a SA amosite mines Reported .expected number reduced to estimate lire number relating to 10+ yr from first exposure (details in Appendix C) 13c SA crocidoliie mines 14 Massachusetts 13 Albin Mainly chrysolite (>95%). Expected number eslinuitcd from observed and all cause RR with smaller amounts of crocidoliie and amosite. Recent crocidoliie use 1 cl*, never > 3-4%. Up to 18% of amosite used briefly during the 1950s. It is not clear wliich ' umphibole should be seen as the most important - - amosite because of the .. relatively high percentage ' - used during a limited -. period or crocidoliie - '- because of its regular ' 16 Connecticut historic use at low levels. Data on men exposed< 1 year excluded Chrysolite was die only lyj>e of asbestos used until 1937 wltcn some anllui|ihytlite was added lur some . product lines. About 400 lbs of crocidoliie was handled experimentally on a lew occasions in the laboratory, but only between 1964 and 1972. Given the small scale and liming of this umphibole use. the cohort has been treated ns chrysolite only 17 Ferodo Apart from two periods before 1944, when crocidoliie asbestos was used on one particular contract, only chrysotlc asbestos has been user]. i ! Colion code .I 2nt 2f 3 4 5a So 6 7 8 9 10 Tabic IS. Explanatory notes on cornet-data1 Lung cancer mortality ' Average cumulative exposure (f/mi y) From Berry (1995) Table 5 Lung cancer risk coefficient Mesothelioma risk coefficient Ku in HEI model ". \ SMR regression slope Regression Slope -- inte_r_n_al___ ___ Coefficient for exposure group Inferred from analyses in From de Klerk a al. (1992) covering cotiort mean in Berry Armstrong et at. (1989) and de (1991) Klerk tld. (1991) Given in text of main reference; Slayner el ul. (1997) give an overall estimate of 2.19 Exd. 4 mesotheliomas - Excl. an estimated 4 mesotheliomas Exluding I mesothelioma Table 2 Tables 8. 10 and 11 Reported value of 0.56, conversion factor 3 Exposure specific rates inconsitcnl, wide range of slopes can be lilted. Data for asbestos factory cannot Tabic 8 he consistently excluded, but Poisson regression (data in Table From IJddcll el <//..( 1998) 8) (Table 3) lulled valucx33/38 to adjust for factory cases impact on risk estimate is minor Smoking adjusted data for 20+ Estimated from Fig. 4 in Neuberger yr latency from Neuberger and and Kundi (1991) Kundi (1993) Excluding an estimated 42 mesotheliomas (half the. difference between DC and BE mesothelioma numbers) The range of exposure conditions likely to have been met by this group, and die long lime period over which these exposures were accrued means that any estimate of the levels of exposure will be very approximate. Other reviews have adopted notional values of average exposure level os 15-30 f/ml and average exposure duration as 25 yr, implying a mean cumulative exposure around 500 f/ml.yr, this is likely to be very uncertain estimate. Excluding workers with <1 yr Overall mean is 73 f/ml.yr (Table 16); mean exposure duration 7.9 yr employment (Table 7); thus mean level around 10 f7ml; half the man years contributed by short term men (Table 8); assuming exposures of 5 f/ml.yr for slum tenn men implies exposure of 138 T/ml.yr for men with >1 yr exposure. From Table 3: the rdsult is sensitive Reported dose specific data has not been used here, since comparison of the man-years in (he dose : to the choice of mean dose in the specific categories of Table 3 of Piolotto ct al. (1990) with the number of men in these category highest category (which covers 30% strongly suggests that all the man-years of follow-up for men in each category lias been assigned to of men). With tire top dose mean the dose category they eventually achieve, lire exposure response is thus biased downwards by an set-at 500 f/ml.yr the overall mean unknown amount. r is 250, increasing to 400 if the top dose mean is set al 1000 f/ml.yr. A . ' value of 300 has been adopted for this review. . (Continued oil next /ige) S P n<' s I o3* .J i * I I J. T. Hodgson and A. Dagtton Table 15. (continued) Cohort code Lung cancer mortality . Average cumulative exposure (f/ml y) Lung cancer risk coefficient SMR regression slope Regression Slope -- internal Mesothelioma risk coefficient Ku in HEI model 11 Excluding three (McDonald, Table 5 personal communication) The dose categories of the three mesotheliomas coded as respiratory cancer not known, so an adjusted exposure-response ' mesotheliomas. Expected cannot be fitted. . multiplied by reported by SMR for lowest exposure category (0.669) 12 Excluding an estimated four Table XVI Poisson regression (Table VI) mesotheliomas 13a Observed an expected reduced Reported means adjusted to reflect weighing of expected mortality >10 yr from first exposure isce !3o by the estimated expected Appendix C) numbets within 10 yr from first exposure (See Appendix C). Observed numbers also reduced by the estimated number of mesotheliomas coded to lung . cancer (1 in 13a. S in J3o) 14 Burden, personal communication (see Appendix D) - 15 Lung cancer taken to be respiratory cancer excluding pleural mesothiiomas, RR interpreted as SMR ' 16 Table 3 Mortality data were reported by cumulative exposure, but high mortality in short term workers makes interpretation of this data difficult. The highest respiratory cancer mortality is seen in the lowest exposure category. This remains the case if short term workers are excluded: low-dose men in all employment duration categories had high levels of respiratory cancer mortality (Table 8). The most likely explanation seems to be that exposures have been wrongly assigned at tlie individual level. The available job details generally only allowed worker histories to be described by department, rather than process. Because of this the exposurc-res|>unse data is nut used here. 17 Total observed lung cancer=lung From Derry and Newhouse (1983) From Berry and Newhouse (1983) (case control, analysis) and pleural cancels less pleural (mean exposure of lung cancer mesotheliomas controls) "Tables or figures referred to are those found in the main references listed in Table 12, unless otherwise specified. Quantitative rijkj of mesothelioma and lung cancer 597 DEFINITION OF AVERAGE EXPOSURE The inclusion of large numbers of cohort members who contribute no informative follow up may also bias the average exposure. An appropriately weighted average exposure will give zero weight to individual exposures in this group. If only a simple mean is used, and if this late entrant group is large and has-- as is likely--systematically lower exposures, the apparent cohort exposure will be too low in relation to the observed mortality, and the estimated risk per unit exposure will be exaggerated. REVIEW OF COHORTS WITH POTENTIAL EFFECT DiLUTION OR BUSSED EXPOSURE AVERAGES The potential biases discussed in the preceding paragraphs will not apply where the reported mor tality excludes observations before the tenth year of follow up (or a later year), and where the average exposure has been weighted by expected lung cancer mortality. This leaves the following cohorts as poten tially affected: Wittenoom, Ontario, Vocklabruck, US/Canada insulators, Balangero. Paterson, SA mines, Massachusetts, Albin and Ferodo. Table 16 summarises the relevant data. The possibility of dilution due to uninformative fol low-up needs to be considered for the SA mines and for the Massachusetts and Paterson cohorts. This can certainly be ignored for Massachusetts and Paterson cohorts, because of their combination of limited recruitment-period with long follow up. It cannot be dismissed for the SA mines, and an adjustment will be developed below (Appendix C). The possibility that a simple mean of individual exposures (the available figure) will be a poor proxy for the desired average weighted by expected lung cancer mortality needs to be considered for all the cohorts listed in Table 16. For all but one there are reasons (summarised individually below) for believ ing that the available figure is an acceptable proxy. For Wittenoom, the measure of excess mortality used has been truncated at subjects' 6Sth birth days. The effect of this is broadly to equalise the follow up durations (and therefore the expected mortality weights) of different first exposure groups. The recruitment period of the Ontario cohort is relatively short, and no mention is made of major variations in exposure conditions. Only }8% of the Vocklabruck cohort started their exposure after 1969: described as `the decisive year in improving the dust situation*. The basis for the 'mean' exposure in the US/Canadiao insulators cohort (drawn from pre vious reviews) is very uncertain. It is not based on averaging known or estiniated individual exposures. It is plausible that conditions may not have changed greatly over the relevant period (up to 1966). The narrow range of first exposure dates for the Massachusetts cohort implies limited scope for changes in average levels, and the long minimum follow up also means that even if there were such changes, the weighting applied to early and late entrants would be similar. Comparison of the most recent follow up report on the Balangero mine cohort with a previous report (recruitment to 1965, follow up to 1975), suggests that only a relatively small proportion of the latest cohort were first exposed after 1965; though it is not entirely clear how the two cohorts relate to each ocher, and the minimum employment quali fication time was more restrictive (I yr) for the later report than for the earlier (1 month), so the comparison is not straightforward. There was reportedly little change in exposure conditions between 1946 and 1960. Some downward bias in the derived exposure average is possible, but the extent of this is difficult to quantify. Even on an unadjusted basis, the derived risk per unit exposure for this cohort is one of the lowest seen. Table 16. Recruitment and follow up configurations for cohorts with potential effect dilution or biassed exposure averages Recruitment follow up Cohort Numbers of From To From To Maximum Follow up years follow up on latest latency (yr) entrants (yr) Wittenoom Ontario Vocklabruck US/Canada insulators Balangero Paterson SA mines Massachusetts -Albin Ferodo 10 20 20 20 20 5 0 0 20 10 1943 1948 1907 1907 1930 1941 1925 1951 1907 1920 1966 1959 1979 1966 1986 1945 1980 1953 1977 1977 1943 1948 1950 1967 1946 1941 1946 1953 1927 1942 1986 1977 1990 1986 1937 1982 1980 1988 1986 1979 44 30 84 80 58 42 56 38 80 60 20 18 11 20 1 37 0 35 9 2 f 598 /. T. Hodgson and A. Camion For the Albin cohort, major exposure changes started to apply only from the (ate 1960s. the last 10 yr of 70 yr of intake. The scope for bias is therefore limited. The average exposure used for the Ferodo cohort is that of controls matched to lung cancer coses. It is therefore--indirectly--weighted in the appro priate way. case. For those cohorts for which the mean age at first exposure is given or can be estimated, it ranges from 23 for Quebec to 37 for Paterson, with a mean across cohorts of about 30. We have therefore standardised the expected all cause mortality figure given for each cohort to an assumed mean age at first exposure of 30. The amount of adjustment applied has been calcu lated using the following formula: The one exception is the SA mines cohort. The report on this cohort shows that a large proportion of the cohort (amosite and crocidolite workers combined) were first exposed less than 10 yr from the end of follow up. Illustrative exposure data is also shown which implies that these workers were exposed to levels 4-6 times lower than those which applied before about 1950. Some adjustment to the reported individual mean exposure is therefore indi cated. This adjustment, and the related adjustment to exclude observed and expected mortality arising from uninformative follow up are described in detail in Appendix C. Briefly, we conclude that both the observed excess lung cancer and the associated cumu lative exposure should be adjusted upwards, the exposure by rather more than the mortality excess. The implied dose specific risk is reduced by about a quarter. SUMMARY MEASURES FOR MESOTHELIOMA Mesothelioma incidence rate rises very steeply with time since exposure, and this complicates the choice of a summary measure that will be properly comparable across cohorts. Comparisons between cohorts with different follow up times (or different mixes of follow up times) should be adjusted to allow for the impact of those differences on the observed mesothelioma mortality. One solution is to fit a stat istical model. The following formulation was used in the HEI report, and is fairly typical: r=*ArML(f/-10)H'-10-I>)5) where L is exposure level expressed in 0ml, D is exposure duration in yr and the contents of the curly brackets (} are set to zero if<0. However not all cohorts have the data needed to fit the HEI (or similar) models. A pragmatic way of making an equivalent adjustment is to. express observed mesothelioma numbers as a percentage of expected mortality from all causes, since this too is a measure which increases steeply with follow up time. The expected mortality from all causes has one drawback os a denominator for mesothelioma risk: it is dependent on age at first exposure. This would not be a serious problem if the mean age at Arst exposure was similar in different cohorts, but this is not the wj = AiVf,,/(K,, Where AJJ is the adjusted expected all cause mor tality to be used as denominator for the observed mesothelioma mortality; a is the mean age at first exposure for the cohort in question; A is the actual expected all cause mortality from the person years in which the mesotheliomas arose: Mm and M. are proportional expected all cause mortality estimates for.the `typical' follow up duration for the cohon (the follow up duration that divides the observation field beyond the minimum latency into two equal areas) from ages 30 and a respectively. The schedule of all cause death rates used to calculate M,0 and M,,, was rate=exp(-9.6l+.0936a)--where a is age in yr-- which provides a close fit to male al^ouse mortality in Australia. Austria, USA and Great Britain (using data taken from the mid 1970s). The fit is less good for South African and for Swedish death rates, but the adjustment depends on the ratio of expected deaths in different--and quite wide--age ranges, a measure that is not sensitive to the precise underlying life table. So for convenience the same life table approximation was used for all cohorts. For the two cohorts where mean age at first exposure was not available (Rochdale and Albin), a mean age of 30 was assumed. A similar argument to that set out above in relation to the effect of uninformative follow up on recorded lung cancer mortality in the SA mines cohorts also applies to the excess mortality from mesothelioma. All the recorded mesotheliomas in these cohorts occurred more than 10 yr from first exposure. An esti mate of the expected all cause mortality arising from follow up less than 10 yr from first exposure has been subtracted from the reported total expected all cause mortality, and this adjusted figure used as the denomi nator for excess mesothelioma mortality in these cohorts. Details of .this calculation are given in Appendix C. COMPARISON OF COHORT AVERACE RISK MEASURES WITH ALTERNATIVES BASED ON INTERNAL COMPARISONS ' For reasons explained in the main report, this review has taken cohort level measures of exposure and outcome as the basic units of observation. In the next two sections these cohort-level measures ore compared to the corresponding internal analyses for those cohorts where both are available. . Quantitative rules of mesothelioma and lung cancer 599 -- ' COMPARISON OF RISK MEASURES-- MESOTHELIOMA For cohorts where details of mesothelioma deaths and person years by time from first exposure were given, the HEI model fitted to these rates. The model was fitted using values for individual calendar yean of time from first exposure aggregated to give the reported latency categories. Best fit was assessed by maximum likelihood methods assuming a Poisson distribution, and the resulting estimates of KM are shown in Table 12. Figure 10 compares the two alternative measures of mesothelioma risk: the HEI coefficient and the percent excess mortality per f/mi.yr index RM. There is good agreement between these measures. The most discrepant point relates to the Quebec cohort (code 6), though this is on either measure clearly the lowest value. It may be relevant that the HEI parameter KH for the Quebec cohort was calculated using data based on age at death as a proxy for time since exposure, since this will have introduced additional inaccuracy. The HEI formula may be preferred for the purposes of risk projection, but the alternative measure seems provide an equally valid summary of the relative yels of mesothelioma risk in these cohorts. ' COMPARISON OF RISK MEASURES--LUNG CANCER Where exposure response regressions were reported by authors, the regression slope has been noted: this provides the 'regression slope* estimate of the lung cancer risk. For cohorts where dose-specific SMR data had been reported, but no regression analy sis was reported, a Poisson regression fit was calcu lated. . The association between the cohort average esti mate of Rl with the regression slope estimate, for studies where both measures were available is shown in Fig. II. There is a clear overall relationship, viewed across the whole risk scale. The discrepant points are those with substantial statistical uncer tainty, either because they are based on small differ ences between observed and expected cases (Sa-- New Orleans, plant 1; and 17--Ferodo) or because of uncertainties deriving from the small sire of the reference population (15--Albin). The most discrepant point relates to the Albin cohort, which was analysed as an unmatched casecontrol study in relation to a control cohort of non asbestos exposed industrial workers from the same area. The overall RR for respiratory cancer excluding mesothelioma was 1.8 (though with a wide confi dence interval: 0.9-3.7) and the mean cumulative exposure was 13 f/ml.yr, giving a cohort average esti mate of of 6.2% per f/ml.yr (with an even wider confidence interval: -0.8-21). The value of the internal regression slope in relation to exposure is not reported, though we are told that it was not statisti cally significant (F=0.5). Inspection of the RRs for the three exposure categories implies that the slope would have been about 0.05. Whether this discrep ancy reflects inaccuracy in the baseline, or in the exposure measurements (or a mix of these) is difficult to say. The high mesothelioma risk in this cohort tends to suggest that the cohort average measure is nearer the truth, but substantial uncertainty must, remain. A further discrepant point relates-to women in the Carolina cohort (2f), where the regression implies /?L=1, while the cohort average gives RL=6.7. The authors suggest that the low regression slope may reflect uncertainties in women's employment histories (which would tend to flatten the regression slope). There is some tendency for the cohort average esti mate to be larger than the .corresponding regression Fig. 10. Companion of alternative measures of mesotheliomu Fig. 11. Comparison of alternative measures of lung cancer . modality. mortality. I :ii9i<ii!lti'iur.o--`t I 600 1. T. Hodgson and A. Damion, slope for (hose cohorts with clearly positive results. This might be predicted from the flattening of regression slopes by inaccuracies in exposure esti mates. But it can also reflect inadequacies in baseline rates. For example, (he two-fold difference between the cohon average and regression slope measures for the Quebec cohort reflects the SMR of about 1.3 seen in all the low dose subgroups, and which the authors interpret as non-asbestos related. Nevertheless, the broad agreement between the two measures across studies suggests that valid conclusions can be drawn from the cohon average measure. ably present assuming some exhaust control on thr bag opening, carding and mixing areas. Unfortunately no mention of the control system is made. Also it is probably an average value that has been given for both the wet and dry methods as both were in use. It is probable given that the sampling locations are unknown that higher concentrations occurred in the dry areas: around 100 f/ml as measured by the current method. Of course this is very approximate, but 100 f/ml looks to be a good maximum exposure with TVVA of 60 f/ml. Appendix B cApfwndlx FIBRE-PARTICLEJ20NVERSION FOR CROCIDOLITE CIGARETTE FILTER COHORT NOTE BY DR G. BURDETT The measurements in 1932 which gave an average of 80 particles per ml, within the Massachusetts stan dard of 173 particles per ml, almost certainly refer to impinger measurements, which were frequently made for insurance company purposes. The normal units are millions of panicles per cubic foot (mppcf). As one cubic foot is equivalent to 28 316.8 ml the value of 80 panicles per ml is equiv alent to 2.263 mppcf and 173 panicles per ml is equi valent to 3 mppcf. Five mppcf was the threshold value in force from the 1930s to the 1960s (maybe even until 1972) when it was replaced by a membrane Alter limit of 10 f/ml, which has been falling ever since. As the units suggest, the method only counted par ticles using relatively low powered microscopy and would overlook many of the respirable fibres and is a very indirect measurement of the fibre level. It should also be remembered that impingers have poor capture efficiency below 1 pin. It is also noted that cotton and acetate fibres were mixed, carded and deposited on crepe paper under dry conditions. This would suggest that fibres made up many of the particles but I have not referred to the patent to work out quantities used to estimate the fibre percentage. My best guesstimate is that 30% of the particles were fibres but only about 10% of the fibres seen would be crocidolite (it is more dusty, but has very few >1 pm fibres compared to the other dusts). This would mean about 3% of the count was cro cidolite. fibres or about 2.3 f/ml>l pm wide. To con vert to the current index we generally find one can assume only some 4% of the >3 pm long crocidolite fibres were visible os compared with the current index. This is equivalent to a concentration of about 60 crocidolite fibres per ml using a modem version of the membrane filter method. This is several times higher than the better factories at (his time but not too far away from whai was prob- DEVELOPMENT OF ADJUSTMENTS TO THE ... SOUTH AFRICAN MINES COHORT DATA ' The starting point for the adjustment of the reported results from this cohort is the data given in Tables 1 and 2 of the published paper, which give illustrative data on exposure levels in different per iods (Table l) and a breakdown of the whole cohort by year of birth and date of first exposure (Table 2). The average age at first exposure'of the groups rep resented by the cells of Table 2 con be estimated using the mid points of the year of birth and year of first employment categories (1900 and 1933 were assumed for the earliest birth and employment categ ories respectively). Age specific all cause and lung cancer rates for white South African men in 1933. 1965 and 1975 were then used to calculate the distri bution of expected lung cancer deaths by time since first exposure in each cell. The rates for 1955 were also used for (he cells relating to first employment between 1941 and 1950, but the expected number was reduced by a factor of 0.64 to allow for the fact that cause specific follow up was only recorded from 1949. The total expected lung cancers calculated in (his way (39.3) agrees quite closely to the value reported in the paper (36.6) and the proportion of expected lung cancer deaths arising from follow up less than 10 yr from first exposure is 0.23. The reported observed and expected tung cancers in the two pure fibre subcohorts have therefore been reduced by 0.23 times the expected numbers given. The data reported in Table 1 was used to estimate approximate relative exposure levels at ten year inter vals from 1943. Taking 1943 as 1, the numbers used were 1, 0.6, 0.35, 0.23 for amosite; and 1, 0.5, 0.23, 0.15 for crocidolite. Exposures in the 1930s were assumed to be the same as in the 1940$. To derive an expected lung cancer weighting for this relative exposure pattern, the expected lung cancers in each birth-start cell from the 10th anniversary of first employment to the end of follow up in 1980 was cal v culated in a similar way to that described for the first 10 yr of follow up. The resulting distribution, of Quantitative risks of mesothelioma and lung cancer 601 Table 17. Derivation of correction factor for reported mean exposure using assumed relative exposure and weighting factors by year of first employment in asbestos mines Data item Year of lint employment Mean weighted by Before 1940 1941-50 1951-60 1961-70 1971-80 Persons Expected lung cancer Relative exposure Crocidolite Amosite Weighting factors Persons Expected lung cancers >10 yr from 1st exposure 1' 1 62 1.68 I .1 404 6.75 0.5 0.6 2355 17.59 0.25 0.35 2408 4.36 0.15 0.25 2088 0 0.35 0.44 Totals 7317 30.38 0.6 0.68 expected lung cancers in the five date of start groups is shown in Table 17. Table 17 also shows the numbers of individuals in each group, and the assumed relative levels of exposure. Mean exposures are calculated using the two alternative weightings. The expected lung cancer weighted means are larger than the corresponding person weighted means by a factor of 1.71 for crocidolite and 1.55 for amosite. The reported mean exposures have been adjusted using these factors for use in this review. Similar calculations for all cause deaths imply that the proportion of expected all cause deaths falling in the first 10 yr of follow up is 33%. The all cause mortality denominator for the observed mesotheli omas in the two subcohorts has therefore been reduced by a factor of 0.67. r I \