Document Em6jqvxo9RQBdpkrLY9YGVnXR

Pergamon PI!: 80003-4878(97)000537 Am. oc<vpL H\J.> Vol 42, No I. PT ?*50. !99$ C 1991 Bniiih OEcupaiionai H>picnc Society published by FJscvier &acfter Ltd Ail fhrtm rtwvcd Primed n ^ Bnum Mt\m. hk si* no - n on Dust Exposure and Lung Cancer in Quebec Chrysolite Miners and Millers F, D. K. LIDDELL*|, A, D. McDONALDt and J. C. McDONALDt * Department of Epidemiology and Biostatistics, McGill University, Montreal, Canada tDepartment of Occupational and Environmental Medicine, National Heart and Lung Institute, Imperial College of Science, Technology and Medicine, London, U.K. A large cohort of men born between 1891 and 1920 and employed for at least a month in the chrysotile producing industry of Quebec has been under study since 1966. These men were followed from first employment (the earliest in 1904) to 1992, by which time over 8000 had died, 657 from lung cancer. The current study is of 488 cases of lung cancer formerly employed at three places, viz, a major complex, here called Company 3, in the region of Thetford Mines (243 cases), the mine and mill in the town of Asbestos (206) and a small asbestos-products factory in the same town (39). For each case, four referents were sought by random selection from among survivors to a greater age, after matching on place of employment, age of starting w ork, smoking habit and date of birth. This process was highly successful, although six cases had less than four referents. For each man (the 488 cases with 1941 referents) and for each calendar year of employment, we obtained the fraction of the year worked at various levels of intensity, assessed in 13 `dust categories' of mpef (million particles per cubic foot). We then calculated how many years each man spent at these various levels; these years, adjusted for the length of the working week (66 h until 1937; 48h 1938-1949; and 40h 1950-1985), were accumulated up to ten years before the death of the case. The men were classified according as they were non- or ex-smokers, or smokers, of cigarettes. For each man at Company 3 and one referent for each, his years of work in a central area of five mines and in a peripheral area of ten mines were differentiated; contamination of the chrysotile by fibrous tremollte was known to be much greater in the central than in the peripheral area. Case-referent comparisons, within place of employment, were made by conditional logistic regression. As anticipated from earlier subject-years analyses, lung cancer risks were found to be negligible for years worked in dust categories 1 and 2 (averaging 0.5 and Impel), regardless of place; as the upper limit of category 1 is considerably higher than permitted nowadays, the lung cancer risk from exposure to chrysotile at permitted levels can be taken as extremely small. Patterns of exposure-response for higher categories were irregular. At Company 3, some risks appeared elevated for years spent in the higher dust categories: 3-4, 5-7, 8-10 and 11-13, with averages around 9,20, 36 and 92 mpef, respectively. For categories 3-4 and 8-10, the odds ratios were high for some or all work in the central area, but minimal for years spent in the peripheral area only. Odds ratios were fairly low for cigarette smokers who worked in categories 5-7 and also for years spent in the highest categories (11-13). At the mine and mill in Asbestos, all risks were low except for years worked by non- and ex-smokers in categories 7-13 (ca. 40 mpef). There were no increased risks at the factory. It was known from the subject-years analyses that most of the excess had occurred at Company 3, but it is now clear that for all practical purposes It was confined to the central area there, probably due largely to fibrous tremolite and in dust conditions of at least dust category 3. The average of this category was 7 mpef or very roughly 24 fibres/ml, about two orders of magnitude higher than today's hygiene standards. 1998 British Occu pational Hygiene Society. Published by Elsevier Science Ltd. Received I May 1997; in final form 19 August 1997, ^Author to whom correspondence should be addressed. Prof. F. D. K. Liddell. 35D Arterbcrry Road. Wimbledon. London SW20 SAG. U.K. 7 DEFENDANT'S EXHIBIT HWBUI0007916 8 F. D. K. Liddell et at. 4 4 introdlction * between 1950-54 and 1990-92. from 24.7 to 464.7. Knowledge of the mortality experienced by chrysolite workers is largely derived from observations on one cohort, namely that which comprised all men born between 1891 and 1920 and employed for at least a month in the Quebec chrysotile producing industry. These 10,918 men were followed from first employ ment (the earliest in 1904) to 1992 and no further tracing is planned. Reports of mortality to several points in time, viz, 1966, 1969, 1973, 1975 and 1988, were published between 1971 and 1993, followed by a comprehensive review of the development of the mortality investigation from its inception (Liddell et a!.. 1997). This presaged exhaustive reports on meso thelioma and lung cancer: the former is now in press Clearly, this could not be ignored. Another factor to be considered was contamination by fibrous tremolite which preliminary investigation (McDonald and McDonald, 1995) had suggested as important in the aetiology of mesothelioma. In the area of Thetford Mines, there were 15 geographically dispersed mines and mills falling into two clearly definable groups: five in a circumscribed central area and ten located in a peripheral area. Lung burden analysis (Sebastien et al, 1989) of 58 members of the cohort in the central area and 25 in the peripheral area had shown that the geometric mean concentration of tremolite fibres 5 pm or more in length was 3.8 times higher in the central area than in the peripheral area. t n (McDonald et al, 1997); this is the latter. Between 1 January 1950 and 31 May 1992. a period MATERIALS AND METHODS for which the expected numbers of deaths could be calculated on the basis of Quebec male mortality, 646 deaths due to lung cancer were recorded in the cohort and the standardized mortality ratio (SMR) was 1,37 (Liddell et al, 1997 Table 6). The same table gave the numbers of deaths (with SMRs in brackets) by place of employment* as follows; `Company 3' in Thetford Mines 280 (1.451: the mine and mill in Asbestos 253 (1.29); the small asbestos products factory also in Asbestos 49 (1.34); and the other six companies near Thetford Mines 64 (1.45). Well over half of the total excess of lung cancer deaths appeared to be due to smoking and the extent of lung cancer mortality attributable to asbestos exposure was estimated as 6065 deaths. Taking account also of dust exposure, the attributable excesses by place of employment can be approximated, from the material in Tables 7 and 8 of Liddell et ai. (1997). as close to 40 at Company 3, as near 18 at the mine and mill in Asbestos and as one at the factory there, with very few elsewhere, i.e. at the six small companies 4-9. It was thus necessary to consider each place of employment separately; and a further reason for separate consideration was that the patterns of dust exposure were quite different (see Appendix). In view of the importance of smoking in the aeti ology of lung cancer, this factor had to be taken into A birth cohort of all Quebec chrysotile miners and millers born 1891-1920 who had worked in the indus try for at least one month has been followed from 1904, when the first man started work, through 1992 (Liddell et al, 1997). Vital status was determined for 9780 men, of whom 8009 had died; for 657 deaths, the cause had been coded for official statistical purposes as malignant neoplasm of trachea, bronchus or lung (three-digit categories 162-164 in the Seventh Revision and 162 in the Eighth and Ninth Revisions of the International Classification of Diseases; see World Health Organisation, 1977), which is termed lung can cer in this report. It was decided to eliminate from the study the 66 cases who had been employed by one or other of six small companies (4-9 in Liddell et al., 1997) because of the very few cases at each. Also excluded were the 52 cases who were recorded as hav ing commenced work in the industry after age 35. These comparatively late starters might well have had substantial earlier exposure to carcinogens and, in view of the long latency averaging over 45 years, were unlikely to have developed asbestos-related malig nancy. For each of the remaining 539 cases, we sought four referents with the following criteria. Relative to the case, the referent must have: l account. Use was made of the information on smoking 1. worked at the same place of employment * that is habit that had been obtained in 1970 for almost every either at Company 3 in Thetford Mines or at the one in the cohort alive at that time. Lung cancer mine and mill in Asbestos or in the factory there; mortality rates for Quebec males. 1950-1992, which 2. survived to a greater age; had been utilised in the subject-years analysis of the 3. had a similar cigarette smoking habit; cohort (Liddell et al., 1997). exhibited marked age- 4. started work at a similar age; and period-cohort effects (Alderson, 1976); one illus 5. had a similar date of birth. tration is that the equivalent average death rate per 100,000 population aged 50-84 increased 18-fold For nearly 85% of the cases, it was possible to find a pool of potential referents and a sample of four was selected by a strictly random process, in what is termed the first tranche ofselection. In the second tranche, to The term `place of employment- is preferred in this report to the strict!) correct, namely the `company from the records of which the man was registered in the cohort" (see Liddell et al.. 1997). find referents for the remaining cases, criteria (3), (4) and (5) were relaxed; however, two cases had to be omitted because of a computing error. There were 49 cases whose smoking histories could not be reliably 'rm HWBUI0007917 -t.7 -.g 464,7. ioma. In the ographically two clearly central area -ung burden mbers of the ripheral area :entration of . as 3.8 times :pheral area. : miners and in the indusllowed from hrough 1992 termined for 7 deaths, the cal purposes .'bus or lung he Seventh Revisions of ;s: see World sed lung can.atef'" - the ed iddeb t\t a!., . each Also rded as havkfter age 35. e!i have had tens and. m v years, were slated malig- sought Four lative to the tent* that ts j ics or at the clory there, bie to find a of four was nat is termed sl tranche, to iteria (3). (4) ,,:s had to be leu- 49 sit 1 -!\ Dust exposure and lung cancer in Quebec chrysolite miner* and miller* Table 1. Place of employment, smoking Habit, age started work, and date of birth, with era of death of ease 488 Cases 1941 Referents PLACE OF EMPLOYMENT Thctford Mines: Company 3 Asbestos: Mine and mill Factory SMOKING HABIT 1st tranche (411 cases: 1642 referenis) Non-smoker of cigarettes or ex-smoker Smoker of less than 20 cigarettes a day Smoker of 20 or more cigarettes a day 2nd tranche (77 eases: 299 referents) Non-smoker of cigarettes or ex-smoker Smoker of less than 20 cigarettes a day Smoker of 20 or more cigarettes a day All subjects Non-smoker of cigarettes or ex-smoker Smoker of less than 20 cigarettes a day Smoker of 20 or more cigarettes a day AGE STARTED WORK (years: average) 1st tranche 2nd tranche DATE OF BIRTH (average) 1st tranche 2nd tranche ERA OF DEATH OF CASE before 1975 1975-1984 1985-1992 243 206 39 88 88 235 11 18 48 99 106 283 22.56 23.51 1910.11 1904.86 146 199 143 970 824 147 350 352 940 44 134 121 394 486 1061 22.57 23.37 1909.95 1904.60 581 795 565 9 classified and, although their referents had been chosen as also having unreliable histories, there was no guarantee that the actual habits of case and refer ents were similar; these 49 were also eliminated*. There were now 488 cases left in the study and for all but six of them, we had found four referents each. For three of the others we had three referents each and for a fourth two referents, but for the last two eases we had only one referent each. Thus the total number of referents was 1941. Some men were included more than once as referents for different cases, but in what follows each referent was treated on each occasion as a separate subject. See also Table 1 and the first paragraph of the Discussion. Liddell et al. (1997) describe how an annual exposure record for each of the 10,918 men in the complete cohort had been compiled. This was pre pared by calendar year, from the year in which the man started work up to and including the year in which he finished. The annual record was ofhow many months in the year the man had worked, expressed as the fraction of the year and the dust level, in millions of particles per cubic foot (mpef). The dust level for the man's job in the specific year up to 1 November 1966 was taken from the equivalent of a 5783x63 In preliminary analyses, it was shown that inclusion of these 49 eases and their referents made no noticeable differ ence. matrix (job class x year) provided by Gibbs and Lachance (1972), who had placed each dust level on a 13-point scale. In each of the 13 dust categories, a representative value, approximating to the mean, had been allocated as follows: 0.5, 2, 7,12, 17, 22, 27, 32, 37,42.47,70 and 140 mpef. For the period 1967-1985, dust levels were estimated as explained by Liddell et al. (1997). The work-week was 60-72 hours before 1938, 48 hours 1938-1949 and 40 hours thereafter. As work week adjustments, we adopted the factors: 66/40 ** 1,65: 48/40 = 1.2; and 1, for these three periods. In the exposure records, it was indicated into which of these periods the specific year fell and so it was possible to adjust the fractions of years worked by multiplying the recorded fraction by the appropriate factor. For all 2429 subjects we calculated 15 measures of exposure as follows:-- net service (adjusted for work-week); the sum of the adjusted fractions of years. years (adjusted) worked in dust category i (i = 1, 2,.... 13): the sum ofthe adjusted fractions of those years in which the dust level had been in category i. accumulated dust exposure (mpef x years): the sum of (adjusted fraction of year) x (dust level). For a case, the calculations were carried out up to ten years before his death and, for a referent, up to ten kw;<sw;sks'Kts HWBUI0007918 10 F, D, K. Liddell aL years before he reached the age at which the relevant ease died. For study of the "tremolite effect', the detailed work histories of all cases at Company 3 and of one referent for each, selected strictly at random from those defined above, were examined. For every subject, per iods of service in the central and in the peripheral areas were calculated up to 10 years before the age of the case's death: the periods were obtained as calendar years (without adjustment for work week)*. Years spent in the central area were expressed as proportions of total years and each man was categorised according as the proportion was: 0; greater than 0. but less than 1/2; at least 1/2, but less than 1; or 1. Subsequently, the three higher categories were grouped to form a two-point C;P class: all years in peripheral areas; and some or all years in central areas. Smoking habits were classified as: non-smoker of cigarettes or ex-smoker; smoker of less than 20 ciga 20rettes a day; or smoker of or more cigarettes a day, but later the two classes of smoker were combined. Three eras of death of cases were defined as follows: before 1975; 1975-1984; and 1985-1992. Exposure-response relationships were examined by means of conditional logistic regression using the EGRET package (Statistics and Epidemiology Research Corporation. 1989) and assuming multi plicative relative riskst. The output from the program includes regression coefficients, which are logarithms of the odds ratios, themselves estimates of the relative risks. The program also gives the standard errors of the regression coefficients and these can be used to calculate approximate confidence intervals, on the assumption that the coefficients are normally dis tributed; the confidence coefficient was set at 90%. In the tables, each odds ratio is an estimate of the relative risk that would have arisen from employment for 10 years adjusted for the length of the working week; there are two exceptions: Table 5, last line--see note (g)--and Table 5--see footnote. The EGRET program also provides the deviance before and after the fit of any model, the difference being treated as a likelihood ratio (LR) statistic with degrees of freedom (df) equal to the number of vari ables in the model; when a mode! is extended by the inclusion of one or more variables, the reduction in deviance is also an LR statistic w ith the same df as the number of additional variables (Armitage and Berry. 'f994)_A close estimate of the LR statistic for a single variable is provided by the Wald statistic, i.e. [(regression coefficient) (standard error)]:. For all subjects and for each of the three places of employment, analysis of the model with 13 exposure "The great effort involved in calculations for more than one referent per case, and especially for work week adjustmem. was not considered justified, + It would hate been preferable to assume additive relative risks, but convergence could not be obtained with these models. measures led to at least one negative regression coefficient, which taken at lace value would imply a protective effect of exposure. Negative coefficients were eliminated by pooling years in adjacent dust categories as described in the Appendix, Each of the variables smoking habit, era of death of case and C. P class was analyzed, where appropriate, as a factor so that its interaction with exposure measures could be evaluated. Even with the original variables replaced by the dichotomies explained above, most of the inter actions were minuscule and therefore deleted, A rough guide to statistical significance may be obtained by referring any LR statistic to the jf dis tribution with the same degrees of freedom. However, the P value found in this way is only nominal and it is essential that removal of negative and minuscule effects and the fact of simultaneous inference are taken into account in any attempt to evaluate sig nificance, Also, where the lower 90% confidence limit on an odds ratio is greater than 1, the P value is less than 0.10, but. for the same reasons, only nominally so. RESULTS The 2429 subjects in the enquiry are described in Table 1 according to the criteria by which the referents were selected, taking account of the tranche of selec tion. Also shown is the distribution of subjects by era of death of cases. In Table 2. we present all the exposure measures, incorporating adjustment for the length of the work-week, for the 488 cases and the 1941 referents. The material consists of means, stan dard deviations (sds) and coefficients of variation (CVs), together with the differences between the means and the corresponding t-siatistics. The matrix of the 78 coefficients of correlation between the 13 measures relating to the different dust categories is extremely complex; because of the very large number of degrees of freedom (2427). any value greater than 0. 04 would be considered significant at the 0.05 level and values above, say. 0.1 of enormous significance, and there are several such values. This means that all variables have to be analysed simultaneously. Likelihood Ratio statistics, each with I degree of freedom, from three different forms of matched analy sis of the complete data set are given in Table 3. Those in column (a) are quite close to the squares of the tstatistics of Table 2, which correspond to the LR statistics in unmatched analysis. The initial deviance. 1. e. before fitting any model, of 1564.79 was reduced to 1552.02 by including only net service in the analysis, yielding the LR statistic of 12.77 shown in column (a). In the analysis of the model including years in each of the 13 dust categories, there were two negative regression coefficients (indicated by brackets round the LR statistics). Pooling was carried out as outlined above and explained in the Appendix and this led to a reduced model with nine variables; regression coefficients are in column (d) of Table 3. When this Net service Iv ears: adjusts Years (adjuster 1 (0.5 mpef) 2 (2 mpef) ? (7 mpcfi 4(12 mpcfi 5 (I7mpcf) 6 (22 mpcfi 7 <27mpcf) 8 <32 mpef) 9 (37 mpef) 10 (42 mpef) 11 (47 mpef) 12 (70mpef) 13 (140 mpef) Accumulated . (mpef x years) All years h Table 3.1 Net service (years: adju Years (adjusts 1 (0.5 mpef) 2 (2 mpef) 3 (7 mpef) 4 (!2mpcf) 5 (17 mpcfi 6 (22 mpef) 7 (27 mpcfi S (32 mpef i 9 (37 mpef i 10 (42 mpcfi 1! (47 mpcfi 12 (70 mpcfi 13 (140 mpef i Accumulated Exposure (mi (a) Each v; (b) Figure? (c) As give (d) Model variables tha (e) Esiimui (0 Indepei (g) Esiima model was 1533.94. yit statistic, a iniproveme HWBUI0007919 Dust exposure and lung cancer in Quebec cbrysotile miners and millers 11 Table 2, Exposure measures for 4SS eases and 1941 referents 4^ CASES Mean (sd) cv 1941 REFERENTS Mean (sd) CV Net service nears: adjusted *) 18.56 Years (adjusted) in dust category: 1 (0,5 mpef) M3 3 12 mpef) 7.86 t (7 mpcl'i 3.06 4 (12 mpef) 1.63 5 il7mpc!) 0.54 6 (23 mpef) 0.37 7 (27 mpef) 0.58 8 (32 mpef) 0.51 9 (37 mpef) 0.81 j0 (42 mpef) 0.29 11 (47 mpef) 0.81 12(70 mpef) 0.39 13 (140 mpef) 0.58 Accumulated dust exposure (mpef x years) 293.6 (16.92) (3.78) (10.85) (5.51) 0.12) (LSD (1.20) (2.19) (1.53) (2.21) (0.99) (2.38) (1.45) (2.46) (492.4) 91.2% 335.9% 138.0% 180.1% 228.9% 277.5% 322.0% 377.0% 299.1% 272.0% 346.3% 293.2% 374.9% 427,9% 167.7% 15.76 (16.06) 101.9% 1.13 (4.39) 389.1% 7.08 (10,39) 146.8% 2.54 (5.11) 201.5% 1.44 (3.67) 254.2% 0.41 (1.52) 369.6% 0.24 (0.84) 349.7% 0.29 (114) 391.4% 0.36 (1.34) 368.3% 0.63 (2.05) 323.3% 0,34 (148) 428.1% 0.52 (1.77) 341.7% 0.34 (1.39) 402.4% 0.43 (1-95) 452.7% 227.3 (402.6) 177.2% All years have been adjusted for length of working week; see text. DIFFERENCE of means 1-statistic 2.80 -0.001 0.78 0.53 0.18 0.13 0.13 0.29 0.15 0.18 -0.06 0.29 0,04 0.14 66.4 3.40 -0.004 1.47 2.00 0,99 1.72 2.80 4.06 2.16 169 -0.83 3.04 0.58 1.38 3.10 Table 3. Likelihood ratio statistics from different forms of analysis, with odds ratios; 488 cases and 1941 r fereni Likelihood Ratio statistic Dependent: 13-variable Dependent; Independent model S-variable model (a) (b) (c) (b) (c) (d) Odds Ratio (e) Estimate 90% Confidence limits Net service (years; adjusted) 12.77 Years (adjusted) in dust category -- 1 (0,5 mpef) (0.00) ; (2 mpef) 2.10 3 (7 mpef) 4.55 4 <12 mpef) 0.99 f 1)7 mpef) 2.83 6 (22 mpef) 7.59 " i27mpef) 15.59 s i?2mpef) 4.34 9137 mpef) 3.27 HI (42 mpef) <0.74i 11 (47 mpef) 8.73 12 (70 mpef) 0,31 13 (140 mpef) 1.86 Accumulated dust Exposure (mpef x years) 9.47 -- 0,00 1.23 2.S2 0.04 0.14 2.15 9.85 2.67 0.26 (3.07) 4.70 (0.05) 0.63 --- -- -- 1.13 2.72 0.04 0.23 1.73 10.16 0.05 4.60 0.30 -- 1.125(f) 1 1.055 1.189 1.028 1.180 2.058 3.121 1.039 1.728 1.111 1.036 (g) 1.066 0.971 1.001 0.812 0.670 0.836 1.735 0.787 1,136 0.809 1.017 1.188 -- 1.146 1.413 1.300 2.078 5.069 5.614 1.372 2.628 1.526 1.055 (a) Each variable analyzed separately. tbi Figures in brackets indicate negative regression coefficients and hence odds ratios less than unity. (c) As given by the Wald statistic, i.e, [(regression coefficient) '(standard error)]3. <d) Model reduced by setting odds ratio in relation to dust category 1 to unity and by the least degree of pooling of variables that ensured no other estimated odds ratio was less than unity. (e! Estimate of relative risk arising from 10 (adjusted) years of employment. (D Independent analysis. (g) Estimate of relative risk arising from exposure of lOOmpcf x years; independent analysis. model was fitted, the deviance was further reduced to i 533.94, yielding the LR statistic 18.08, with 8 df. This statistic, associated with P =s 0.02, relates to the improvement in fit arising from subdividing net service into nine components and indicates that such sub division is meaningful. When the nine exposure variables were analysed in a single step, i.e. w ithout the intervening stage of fit- F. D. K, Liddell el al. sing nei sen ice, she LR statistic, with 9 df, became (^^79-1533.94 = ) 30.85 and the nine relevant odds tether with 90% confidence limits, are shown ifPrable 3t column (e); the odds ratios for net service and accumulated dust exposure were found in inde pendent analyses. These last two variables were omit ted from the analyses for Tables 4, 6, 7 and 8. In the fit of the 13-variable model to the data for the 243 cases of Company 3 and their 970 referents, four of the 13 regression coefficients were negative and the necessary pooling resulted in a 5-variabie model. Part (A) of Table 4 shows regression coefficients, with standard errors*; odds ratios are not given because they were all modified to some extent by the inclusion of other variables. In fact, all the interactions between smoking habit and/or era of death of case on the one hand and exposure on the other hand were very slight, except that between smoking habit and years spent in dust categories 5-7. Regression coefficients and odds ratios for the 6-variable model are in Part B of the table. Deviances were as follows;-- initial 781.29; reduced to 764.95 after fitting the five exposure vari ables (LR statistic 16.34:5 df); and finally to 762.03 (LR statistic 2.91:1 df) when the interaction was included. For 240 of the casest at Company 3 and for a single referent each, it was known how long each man had worked in central and peripheral areas up to ten years before the death of the case and to the corresponding u ' "brents: regression coefficients are presented i, together with odds ratios in relation to 10 caieffSYr years of net service. The deviances before and after analysis of net service alone were 332.71 and 319.43 and the effect of differentiating between service in the two areas was to reduce the deviance to 311.54, giving the LR statistic (1 df) as 7.89. I Because part of this undoubted statistical sig nificance could have arisen from differences in dust I levels in the tw o areas, the data for the same 240 cases and their referents have been re-analyzed; see Table I6, which is on the same lines as Table 4, Part (B), with the same pooling of dust categories, except that category 1 was retained because the relevant coefficient was not negative. After fitting the six txposure variibies, the deviance was 313.71, which vas seduced to 300.04 (LR statistic 13.67: 2df) when he two interactions were included. Regression coefficients and .Qdds^ratios For all eight variables are isvcn in this table, buFonly those for the interactions .re relevant; see note (c). Tables 7 and 8 present the analyses of the data from Asbestos, the former concerning the 206 cases from JUSMM fF*6\ Qm** --* d -- t doo'eed i **** SO fx| i s d ei c 5 e 2 J .S J5 OS w %> JS "i r 34 GO 8SS odd IS! .1 fr r rs 4 Regression coefficients, and their standard errors, are resented with 4-5 decimal places only so that odds ratios lose to unity can be calculated with reasonable precision for 3-vear periods of employment. tTh r three eases had to be eliminated, two because t" fret, .unrecorded moves between the two areas, the Sher because of a missing work history. 3 M Si f S nw >z Dust. Table 5. Lung cancer risks in n Set service (calendar years, with adjustment for length of workirr Without differentiation by area In central areas In peripheral areas (a) Estimate of relative risk an the mine and mill, with 824 re cases and 147 referents from i the necessary pooling of dus variable model: regression co> errors, are in Part A. All but with smoking habit and era of small. Part B gives regressiot ratios, with 90% confidence I account the interaction beiwt years in dust categories 7-13. lows:-- initial 663.09; reducer the four exposure variables ( ami then to 655.87 (LR statist! the one non-trivial interactioi reduced the number of dust c; three and all interactions wen before and after the fit of the tl were 120.40 and 118.81 and th (3 df). The odds ratios, with confide 4(B), 6,7(B) and 8 are summar D1SCUSSIO Methodology We obtained four referents f: work for all cases, except two a1 Table 6. Lung cancer risks in retain Years (adjusted) in dust category(ies):-- ! (0,5 mpcl) 2 (2 mpef) 3-4 (ca. 9tnpd) 5-7 (ca. 20mpcf) 8-10 (ca. 36mpcO 11-13 (ca. 92 mpef) (b) (i) (b) (It) Some or all > All years in p (bHiiil Some or all y, All years in p (b) (iv) 'See text. (a) Estimate of relative risk arising (b) The interaction with C P class (tit) 0.72 (negative coefficient]; (ivj 0 ; (c) These estimates are given only which are more reliable, being bawd in Table 6 are based on a subset of rc *r 1 ,*`j|^a*~s^1iirTJir4si"|out nflga y HWBUI0007921 Dust exposure and lung cancer in Quebec chrysolite miners and millers 13 Table 5. tune cancer risks in relation to years worked in central and peripheral areas of Thetford Mines. 240 cases. 240 referents Net service (calendar years, without adjustment for length of working week) Regression coefficient Estimate Standard error Odds Ratio (a) Estimate 90% Confidence limits Without differentiation by area in central areas In peripheral areas 0.02454 0.03423 0.00396 0.0070 0.G0S2 0.0102 1.278 1.408 1.040 1.139 1.23! 0.880 1.434 1.610 1.230 (a) Estimate of relative risk arising from 10 years of service. the mine and mill, with 824 referents, the latter the 39 cases and 147 referents From the factory. For Table 7, the necessary pooling of dust categories led to a 4variable model: regression coefficients, with standard errors, are in Part A. All but one of the interactions with smoking habit and era of death ofcase were very small. Part B gives regression coefficients and odds ratios, with 90% confidence limits, after taking into account the interaction between smoking habit and years in dust categories 7-13. Deviances were as fol lows:-- initial 663.09; reduced to 658.27 after fitting ihe four exposure variables (LR statistic 4.81:4df); and then to 655.87 (LR statistic 2.40:1 df) by the fit of the one non-trivial interaction. Pooling for Table 8 reduced the number of dust categories analyzable to ihree and all interactions were small. The deviances before and after the fit of the three exposure variables were 120.40 and 118.81 and the LR statistic was 1.59 (3 df). The odds ratios, with confidence limits, from Tables 4(B), 6, 7(B) and 8 are summarised in Table 9. DISCUSSION Methodology We obtained four referents from the same place of work for all cases, except two at Company 3 and four at the factory in Asbestos. Matching by smoking habit was nearly perfect in the first tranche of selection and remained so for non- and ex-smokers in the second tranche. Overall, there was a shortfall of 71 referents who had smoked 20 or more cigarettes a day, with an excess of 62 lighter smokers. Matching by age started work had been within eight bands mainly two or three years wide in the first tranche and in the second within four bands of average width 61/2 years. The outcome was extremely close agreement on average between cases and referents. Similarly, matching by date of birth had been first within quinquennium and then within decade, leading to very close agreement on average. It had been decided from prior considerations to truncate all exposure measures 10 years before the case's death and to adjust years worked to allow for the length ofthe working week. In a preliminary inves tigation, LR statistics were obtained for net service, both adjusted for work-week and not adjusted, and accumulated dust exposure, each with four amounts of truncation: 15, 10, 5 and 0 years before the death of the case. The LR statistics for adjusted service were higher than for unadjusted; those for 15- and 10-year truncation higher than for the other amounts of trunc ation. There was therefore no contra-indication to the use of the work-week adjustments and no strong contra-indication to truncation at 10 years before the Table 6. Lung cancer risks in relation to years worked in central and peripheral areas of Thetford Mines, by dust category: 240 cases; 240 referents Regression coefficient Odds Ratio (a) Years (adjusted) in dust category(ies);-- C/P class * Estimate Standard error 90% Estimate Confidence limits 1 (O.Smpcf) 2 (2mpcf) 3 4 <ea. 9mpcf) 5-7 (ca. 20mpcf) 8-10 (ca. 36mpcf) 11-13 (ca. 92mpcf) (b) (i) (b) (ii) Some or all years in central area All years in peripheral area (b) (iii) Some or all years in central area All years in peripheral area (b) (iv) {0.00921 {0.01014 0.04014 -0.01669 {0.01387 0.1188 -0.03861 {0.05028 0.0242 0.0112 0.0143 0.0224 0.0359 0.0622 0,0472 0.0232 <0 1.096 (c) 1.107 1.494 0.846 (0 1.149 3.281 0.680 (c) 1.653 0.736 0.921 1.181 0.585 0.636 1.179 0.313 1.129 1.633} 1.331} 1.890 1.223 2,073} 9.126 1.477 2.422} `See text. (a) Estimate of relative risk arising from 10 (adjusted) years of employment. !b) The interaction with C/P class was associated with a small LR statistic, namely: (i) 1.19; (ii) 0.03 [negative coefficient]; (in) 0.72 [negative coefficient]; (iv) 0.37. ic) These estimates are given only for the sake of completeness and are in no way substitutes for estimates in Table 4, which are more reliable, being based on four limes as many referents: the two sets are entirely compatible given that those m Table 6 are based on a subset of referents. 14 F. D K. Liddell ei al. H Taking account of interaction between smoking habit and years in dust categories 7 17 vO~j rnC-i r^-i X S' X rra--~c. i -- O rf -- fN IN X t/7 O V7 -- O -- r- T3 oC -CZ : 1O/ V5 --r- -acr xvn cr-~ cooco o E ^--*-, orf-- Orsi'--P**. wVTn| UJ o" Ec ooa"y roI-s-. Wrl ?^1/1 oooc UJ I O C O c; -33 cases death: both these device-- have been used throughout. An important finding (Table 2) is the extremely high coefficients of variation of all the exposure measures: most subjects spent no time in most specific dust cat egories; and very few men spent more than five years in any one of the dust categories 5-13. Some impli cations are discussed in the Appendix; in particular, a different (chance) selection of referents might have led to rather different odds ratios for the higher dust categories. The differences between the first two columns of Table 3 arise from the correlations between the exposure measures, which require these measures to be considered simultaneously, i.e. in dependent analysis. Accumulated dust exposure was omitted from all fur ther analyses and net service from all except for Table 5. The odds ratios in Table 3 for these two variables correspond to column (a); the others correspond to column (c). The materials to be analyzed are essentially the exposure measures, rather the differences, between cases and referents, in the mean measures, taking into account the correlations between them. In Table 2, the 13 differences of mean years in specific dust cat egories are seen to fluctuate greatly without dis cernible pattern and the correlations between the 13 differences of means, although mainly quite small, include several that have major influence. As a result, the corresponding odds ratios in Table 3 also exhibit great fluctuations, leading to major difficulties of interpretation, which are not decreased in the later tables. In the only earlier study on broadly similar lines. Vacek and McDonald (1990) demonstrated a steady increase in differences (obtained from their Table 1) between cases and 'controls' in 'durations' as average `dust values' increased through six steps from 0.6 to 105.0mpef; the odds ratios (calculated from Table II) for 10 years work in the six dust ranges were 0.71, 1.14. 1.02, 1.26, 2.02 and 1.68. Comparison between the two sets of results is, however, vitiated by inter alia two major differences in methodology: how vari ation in the length of the working week was treated (see Appendix); and consideration of the age at which employment started. In the earlier report, the referents were on average 5.2 years older than the cases when they started work; their initial dust exposure would thus tend to be less severe because of the continuing improvement in conditions and. compared with cases who died only a few years after last employment, their duration of employment could not be as long*. Indeed, although matching by age at which employ ment commenced might appear 'over-matching-, it is Tabic x Lung c.! V cars iadjusted I ir. categorytiesi - ] (0 5 mpcfl 2 (2 mpef) 3 -6 tea. 11 mpef) ". 13 tea. 42 mpef! Ui Estimate ot r (b) All interacne Table 9 Odds rai Dust category ties) (with average dust Category 1: (0.5 m Factory in Asbi Mine and mill i Company 3 at 2 Category 2: t2mp Factory m AsK Mine and null ; Company 3 at Categoric 1 -i `< r' -ompun.v 3 at all years in peri some c -> ! " categories 3-6: F Factory in Ash Mine and mill Categories 5-7: F Company 3 at cigarette smok non- and ex-srt Categories 8-10: Company 3 ai all years in pci some or al) ye. Categories 7-13: Factory in Ad Mine and mill cigarette smol non- and ex-s: Categories 11-1 Company 3 a la) Estimate i (b) Nominal , (c) Table 4B. (d) There is n (c) Estimate i If) Based on . A Without regard to interactions * If only for these reasons, cases would accumulate more u< 7? ic dust exposure than controls: Vacek and McDonald (1990) reported cumulative dust exposure for cases as 2 3/4 limes that of controls, whereas we found cases to have had only 30o more exposure (Table 2). considered es example. Lidi c.tn be riisast i risks (Sluts-C ................................................. T HWBUI0007923 Dust exposure and lung cancer in Quebec chrysolite miners and millers 15 Table 8. Lung cancer risks in relation to years workeS, by dust category, at factory in Asbestos: 39 eases; 14" referents 1 ears taJjusted) tn dust taicgoryUes):- Regression coefficient Estimate Standard error Estimate OJda Ratio <u) 90% Confidence limits J (0.5 mpcf) 2 12 mpcf) 3-6 (ca, 11 mpcf) 7-13 (ca. 42mpeO 0.01 S68 0.0044 ! 0.00384 0.0150 0.0513 0.0433 1.205 1.045 1.039 0,942 0.449 0.510 1.543 2.430 2.118 (a) Estimate of relative risk arising from 10 (adjusted) years of employment. lb) All interactions with smoking habit and era of death of case were very small. Table 9. Odds ratios (a), with 90% confidence limits, by dust category and place of employment, taking account of interactions, and classified by Likelihood Ratio statistic. Dust category(ies) (with average dust level) Likelihood Ratio statistic class (b) <1.64 >0.2} 1.65, <2.70 {0.2>/*>0.1} >2.71 (0.1 >P) Source (C) LR (Wald) statistic (1 degree of freedom) Category 1: (0.5 mpcf) Factory m Asbestos Mine and mill in Asbestos Company 3 at Thetford Mines Category 2: (2 mpcf) Factory in Asbestos Mine and mill in Asbestos Company 3 at Thetford Mines Categories 3-4: (ca. 9 mpcf) Company 3 at Thetford Mines all years in peripheral area some or all years m central area Categories 3-6: (ca. 11 mpcf) Factory in Asbestos Mine and mill in Asbestos Categories 5-7: (ca. 20mpcf) Company 3 at Thetford Mines cigarette smokers non- and ex-smokers Categories 8-10: (ca. 36 mpcf) Company 3 at Thetford Mines .ill years in peripheral area some or all years in central area Categories 7-13: (ca. 40 mpcf) Factory in Asbestos Mine and mill in Asbestos cigarette smokers non- and ex-smokers Categories 11-13: (ca. 92 mpcf) Company 3 at Thetford Mines 1 (d) 1.1 (0.8, 1.6) 1(d) 1.2 (0.9, 1.5) 1.1 (1.0, 1.2) 1.0 (0.9, 1.1) 0.8 (0.6, 1.2) 1.0 (0.4. 2,4) 1.2 (0.9. 1.6) 0.7 (0.3. 1.5) 1.0 (0.5. 2,1) 1.1 (0.8. 1.4) [8] [7B] [4B] 18] PB] I4B] 1,5 (l .2.1.9) [61(e) [6] (e) [8] [7B] 1.4 (0,9. 2.2) 6.1 (1.6. 23.6) [4B] [4B] 1.3 (1.0. 1.6) 3.3 (1.2. 9.1) 1.7 (1.1. 2.6) [6]!ei [6] (el [8] [7B] [7B] [4B] 0.32 1.55 1.18 0.00 0.56 (0 7.88 0.01 0.60 1.75 4.87 0.67(f) 3.65 0.01 0.08 4.24 2.69 (a) Estimate of relative risk arising from 10 (adjusted) years of employment, lb) Nominal P values are given in curly brackets. (c) Table 48,6. 7B or 8, as indicated in square brackets. (d) There is neither confidence interval nor LR statistic, because the odds ratio was `forced' to unity. ie) Estimate of relative risk arising from 10 calendar years of service; based on only one referent per case, (f) Based on a negative regression coefficient. considered essential for valid comparisons (see, for example. Liddell. 1988) and its lack--until rectified-- can be disastrous in leading to impossibly high relative risks (Sluis-Cremer et at.. 1992). Epidemiology Accumulated dust exposure, which was a useful measure in the subject-years analyses of the cohort (Liddell et al,, 1997), discriminated reasonably well HWBUI0007924 !e F. D. K. Liddell ct al. between the 488 cases and 1441 rei'erems*. bui not as well as net service and the inclusion of nine dustrelated measures improved the discrimination greatly (Table 3). Although these measures allow some exam ination of response to different intensities of exposure, the pattern of exposure-response was quite irregular, even after the pooling described above; for example, the most significant' odds ratios were for years in dust categories 7 and 11 (averaging 27 and 47 mpef), but there was little sign of `significance' in relation to the adjacent categories 8-10 (ca. 36 mpef) and 12-13 (ca. 92 mpef). It is not surprising, in the light of the material in the Appendix, that the relationships between odds ratio and dust category were nowhere clear-cut. The hypothesis we would have liked to test was that the odds ratio was related to intensity of exposure mea sured in fibres ml. but the lack of a reliable fibre dust ratio rendered such an approach impossible, forcing the use of mpef as a surrogatef, Further, it must be emphasised that dust exposure in any extractive industry is likely to be highly variable, with little dis cernible pattern; this is well illustrated in Table 10, w hich shows not only large differences between the places of employment but also several sporadic peaks at all three places, probably associated with changes in work practices. Although interactions between dust measures and era of death of case were very small at all three places of employment, the pattern of exposure-response was affected at least slightly by smoking habit both at Company 3 and at the mine and mill in Asbestos. However, considerably the most important inter actions related to whether work was in the central or peripheral area of Company 3. Odds ratios in relation to 10 years of employment are summarised in Table 9. in three columns according to the value of the LR statistic (with 1 df). which is quoted in the final column. The columns of odds ratios are also headed by nominal P values, but it should be borne in mind that these are considerably lower than if due regard were paid to the selection of effects 'tested' and simultaneous inference. Thus the odds j ratios in the first column must be considered as differ ing from unity only trivially and those in the central column cannot be thought ofas statistically significant at any level normally adopted. fn-thejjght of the information in the Introduction, inferred from the data of Liddell el al. (1997), about where the excesses of deaths from lung cancer, as a result of exposure in the industry, must have occurred, it was to be expected that risks would be elevated at Company 3 and. to a lesser extent, at the mine and mill in Asbestos, but probably not at the factory. This is dearly borne out by the findings in Table 9. Accumulated dust exposure also differentiated cases and referents at Company 3. but not at the other two places of employment + 5ee Dust-fibre relationships in the Appendix. It is also important to appreciate that excess lung cancer was apparently confined to men with sub stantial exposure, of the order of 300 mpef x years (Liddell et ah. 1997). The longest net service recorded in the current study was less than 68 years (after adjustment) and to accumulate 300 mpef x years even in such a long period would have required the dust level to average over 4 mpef throughout. It was thus to be anticipated that even long employment in dust categories I or 2 (averaging 0.5 and 2 mpef) would not appear hazardous. This also is clearly borne out in the first six lines of Table 9: the lowest value of P for category 1 was greater than 0.5 and even that for category 2 was about 0.25. Much has been written on the interaction of smok ing and asbestos exposure in the causation of lung cancer, summarised by Berry a al. (1985), who defined a 'Relative Asbestos Effect" (RAE) which would be unity if the interaction were multiplicative, i.e. if smo kers and non-smokers had closely similar relative risks from exposure. They found the RAE averaged over six studies as 1.8. indicating that the relative risk of lung cancer due to exposure to asbestos is greater for non-smokers than for smokers. One of those studies was of the 245 eases in the Quebec cohort before 1976. with 735 referents (Liddell el al,, 1984) and it had the same RAE (1.8) as the average. In subject-years analysis of deaths 1976-1988 in the Quebec cohort (McDonald el al., 1993), RAEs could be found in several ways but all were much greater than unity. In the current study, the direct effects of cigarette smok ing on the risks of lung cancer were effectively elim inated by the matching in the selection of referents. However, the interactions of smoking and asbestos exposure could be examined; the only two that were not trivial indicated higher lung cancer relative risks for non- and ex-smokers than for cigarette smokers. One of our most important findings is that the odds ratios relating to years worked in dust categories 3-4 and 8-10. but only in the peripheral area, were negli gible. whereas for men some or al! of whose years were spent in the central areas the corresponding odds ratios were substantial. It has been shown elsewhere (Sebastien ei al., 1989) that contamination of chryso lite by fibrous tremolite was much more serious in the group of five central mines of Company 3 than in the ten peripheral mines and no other reason has been found for the differences in odds ratios. There were great improvements over the years in dust levels everywhere; from 1974, only about 4% of men in the complete cohort were working in more than 5 mpef (Table 10); a few years later, con centrations had been brought to quite low levels vir tually throughout the industry. However, it must be borne in mind that the `average' case started work in 1929 when the mean dust levels were very high (see the 1904-1937 column in Table 10) and even in 1953. when the last of the 488 cases was first employed, the average dust level in the industry was still about 50 mpef (Gibbs and Lachance. 1972). roughly equi- Tabic COMPANY 3 AT Number of record' Percentage by dusi I. <0.5 mpef) 2: (2 mpef) : i7 mpef i J: < 12 mpef) 5:(17mpcO b: (22 mpef) 7: (27 mpefS 8: (32 mpef) 9: (37 mpef) JO: (42 mpef) 11; (47 mpef) !2: (7(impef) IV 1140 mpef) Total Mean dust level!( MINE AND Mil Number of recorc percentage by dw ! (i!.5mpcf| 2. (2 mpef) 3: (7 mpef) 4. <I2mpcf) 5: (! 7 mpef) 6: (22 mpef) 7: (27 mpef) x: (32 mpef) 9. (37 mpef) 10; (42 mpef) 11; (47 mpef) 12: (70 mpef) 13: (140 mpef) Total Mean dust level i I UTORY IN Number of rceo Percentage by d I: (0,5mpef| 2; (2 mpef) 3; (7 mpef) 4: (12 mpef) 5; (17 mpef) 6: (22 mpef) 7: (27mpcf) X; (32mpef) V (37 mpef) 10; (42 mpef) II (47 mpef) 12: (70mpef) 13: (140 mpef) Total Mean dust !eve HWBUI0007925 ,i! excess lung .on with sub- i)m5ife&r"Ci,rs Srviy^ellBPPfardfteedr f < years even aired the dust it. It was thus yment in dust icf) would not >rne out in the alue of P For even that for tion of smok- ation of lung I, who defined ich would be ve. i.e. if smo- r relative risks iveraaed over dative risk of ; is greater for ( those studies i before 1976. It) and it had i subject-years \ tiehec cohort : be found in ban unity. In surer .-tnok- jecti1 ,'im- of reierents. and asbestos : wo that were relative risks jtte smokers, that the odds ijategories 5-4 |a. were negli- ; whose years bonding odds ,\vn elsewhere on of chryso- serious in the . 3 than in the !t!3 son has been .. the years in 'about 4% of ting in more *. later, con:>w levels virr. it must be j rted work in ;Jery high (see ifeven in 1953. ' jsi employed, hs sr'.: sout |c jui- 1 ' Dust exposure and lung cancer in Quebec chrysolite miners and milters 17 Table 10. Distribution of 13 60*) annual work records by dust categories and place of employment 1904-1937 1938-1949 1950-1961 1962-1973 1974-1985 1904-1985 COMPANY 3 AT THETFORD MINES Number of records 1794 1864 1686 1107 334 6785 Percentage by dust category i: (0,5 mpef) 2: <2 mpef) V (7 mpef) 4: (12 mpef) j; <17mpcf) 6: (22 mpef) 7: (27 mpef) 8: (32 mpef) 9: (37 mpef) (0: (42 mpef) II: (47mpef) 12: (70 mpef) 13: (140 mpef) 4.3 5.3 11.3 13.0 38.9 9.4 30,0 25.3 36.7 46.8 54.8 34.3 20.5 19.0 24.7 33.2 5.4 22.4 14.2 12.3 10.6 6.2 0.9 10.S 4,7 3.8 3.7 0,4 -- 3.3 1.6 2.7 3.1 0.1 -- 1.9 0.9 1.6 1.4 0.2 -- 1.0 3.0 4.7 3.0 -- -- 0.6 2.1 2,5 0.1 -- 2.8 1.4 1.2 2.3 1.0 -- -- 1.2 1.6 15.2 0.5 ------- -- 4.7 1.8 2.5 l.S -- -- 1.5 15.8 3.3 0.5 -- 5.2 Total 100 100 100 100 100 100 Mean dust level (mpef) 30.8 21.8 9.5 4.2 1.7 17.3 MINE AND MILL AT ASBESTOS Number of records Percentage by dust category 1: (0,5 mpef) 2: (2mpcf) 3: (7 mpef) 4: (12 mpef) 5: (17mpef) 6: (22 mpef) 7; (27 mpef) 8: (32 mpef) 9: (37 mpef) 10: (42 mpef) 11: (47 mpef) 12: (70 mpef) 13: (140 mpef) Total Mean dust level (mpef) 2002 2.5 41.5 6.3 1.1 2.3 1.8 2.3 1.7 17.7 5.9 4.9 5.3 -- 100 19.1 1493 4.6 49.6 8.5 10.0 2.2 1.7 7.0 5.5 6.5 3.9 0.2 o.i 0.1 100 11.6 1273 723 10.4 8.7 69.0 87.6 9.0 3.5 4.6 0.1 1.7 0.1 2.7 -- 1.7 -- 0.7 -- ---- ---- --- .-- ---- ---- 100 100 4.2 1.9 237 5728 84.0 9.0 16.0 54.5 -- 6.9 --- 6.4 -- 1.8 -- 1.6 3.0 -- 2.2 -- 7.9 -- 3.1 1.8 -- 1.9 -- 0,0 100 100 0.7 10.9 FACTORY IN ASBESTOS Number of records Percentage by dust category 1: (0.5 mpef) 2: (2mpef) 3: (7 mpef) 4: (12 mpef) 5: (17 mpef) 6: (22 mpef) 7: (27 mpef) 8: (32 mpef) 9: (37 mpef) 10. (42 mpef) II: (47mpef) 12, (70mpef) 13: (!40mpcf) Total Mean dust level (mpef) 293 343 245 172 43 1096 9.2 9.3 15.5 16.9 83.7 14.8 58.4 53.4 65.7 80.8 16.3 60.3 11.6 25.1 14.3 2.3 -- 14.5 4.8 4.4 3.3 ------- -- 3.4 2.4 0.9 0.4 1.0 1.7 1.7 -- -- -- 1.0 0.7 2.0 -- ----- - _ 0.8 0.7 1.2 0.8 -- -- 0.7 6.1 0.6 --- -- -- 1.8 0.7 1.2 -- -- 0.5 1.7 0.3 -- -- 0.5 1.7 -- ___ _ ____ 0.5 0.3 -- -- -- -- O.i 100 100 100 100 100 100 10.4 5.8 3.1 1.7 0.7 5.1 HWBUI0007926 18 F. D& . K. Liddell M. valent to 175 fibres'ml. The interval between first exposure and death averaged 47 years, ranging from 15 to 71 years: thus many of even the latest cases would have been exposed before the industry had done very much to `clean up its act*. The corollary is that there may be a continuing, though small and dimin ishing, excess of lung cancer at Company 3 arising from conditions in the 1950s and 1960s. On the other hand, modern dust conditions are well below the average even ofdust category ! and so there can be considerable confidence that the risk of lung cancer as a result of such exposure has become van ishingly small. ENVOY In this paper, the last of a long series since 1971 on mortality in Quebec chrysotile miners and millers and the fourth of a recent quartet (the others being: Liddell et a!., 1997; McDonald and McDonald, 1997; and McDonald et al., 1997), it is appropriate to present the main conclusions relating to both lung cancer and mesothelioma. It is abundantly dear that the risks of excess lung cancer in this industry, except after extremely high dust exposure, were very low. The overall excess, 1904-1992, in the cohort (which included the great majority of men recruited to the industry over several decades) is estimated as no more than 65 deaths (0.8% of 8009) and little or no excess was seen below 300mpcf x years. To accumulate such exposure, a man would have had to spend his whole working life in dust levels of 5-6 mpef. very roughly equivalent to 20fibresiml. orders of magnitude higher than per mitted today--or of course a shorter period in even more severe conditions. Among miners and millers, there were just 33 mesothelioma deaths in a total of 7456 from all causes (0.4%) and none were in men employed less than two years. What has now become evident is that both these modest excesses were con centrated in men employed in the central area ofThetford Mines, with relatively little either in the periphery there or in the large mining and milling industry in Asbestos. This geographical pattern points to some important and clear-cut geological difference between areas of high and low risk; the distribution of amphibole fibres in the tremolile series offers a reasonable, biologically plausible and sufficiently specific expla nation. While we accept that associations dem onstrated epidemiologically cannot prove cause and effect, we believe the answer to the fundamental ques tion posed by Bradford Hill (Hill, 1966)--"is there any other way of explaining the set of Facts before us, is there any other answer which is equally, or more, likely than cause and effect?"--is "No", Acknowledgements--Sincere thanks are due to Ben Armstrong. Marielle Olivier and Paul Wilkinson. REFERENCES Alderson. M. (19761 An Introduction to Epidemiology, pp. 27-29. Macmillan. London. Armitage. P. and Berry. G. (1994) Statistical Methods in Medical Research. 3rd Edn. p, 428. Blackwell. Oxford. Berry. G.. Newhouse. M. L. and Antonis. P. (1985) Com bined effects of asbestos and smoking on mortality from lime cancer and mesothelioma in factory workers. Br. J. ind Med. 42. 12-18. Gibbs. G. W. and Lachance. M. (1972) Dust exposure in the chrysotile asbestos mines and mills of Quebec. Arch. Em iron, Health 24, 189-197. Gibbs. G. W. and Lachance. M- (1974) Dust-fibre relation ships in the Quebec chrysotile industry. Arch. Environ. Health 28,69-71. Hill. A. B. (1966) Principles of Medical Statistics. 8th Edn. pp. 305-313. The Lancet Ltd., London. Liddell, D. (1988) Maintaining progress in the analysis of cohort studies. In: Progress in Occupational Epidemiology (Edited by Hogstcdt. C. and Rcuterwall. C.j, pp. 71-74. Elsevier Science Publications, Amsterdam. Liddell. F. D. K... McDonald, A. D. and McDonald. J. C. (1997) The 1891-1920 birth cohort of Quebec chrysotile miners and millers: development from 1904 and mortality to 1992. Ann. Occup. Hyg. 41,13-36. Liddell, F. D. K., Thomas, D. C,, Gibbs, G. W. and McDon ald, J. C. (1984) Fibre exposure and mortality from pne umoconiosis, respiratory and abdominal malignancies in chrysotile production in Quebec. 1926-75. Ann. Acad. Med. Singapore 130' `siippl.). 340-344. McDonald, A. D,, Case, B,, Churg, A,, Dufresne, A., Gibbs. G. W.. Sebastien, P. and McDonald. J. C. (1997) Meso thelioma in Quebec chrysotile miners and millers: epi demiology and aetiology. Ann. occup. Hyg. 41,707-719. McDonald. J. C.. Liddell. F. D. K.. Dufresne, A. and McDonald. A. D. (1993) The 1891-1920 birth cohort of Quebec chrysotile miners and millers: mortality 1976-88. Br. J. Ind. Med. 50, 1073-1081. McDonald. J. C. and McDonald. A. D. (1995) Chrysotile. tremolite and mesothelioma. Science 267, 775-776. McDonald, J. C. and McDonald. A, D. (1997) Chrysotile, tremolite and carcinogenicity. Ann. occup. Hvg. 41, 699705. Sebastien. P,, McDonald. J, C,, McDonald. A. D.. Case. B. and Harley. R, (1989) Respiratory cancer in chrysotile textile and mining industries: exposure inference from lung analysis. Br. J. Ind. Med. 46, 180-18?. Sluis-Crcmer. G. K.. Liddell. F. D. K., Logan. W. P, D., Bezuidenhout. B. N. (1992) The mortality of amphibole miners in South Africa, 1946-80, Br. J. Ind. Med. 49,566575. Statistics and Epidemiology Research Corporation (1989) EGRET. Reference Manual. Seattle. Vacek, P. M. and McDonald, J. C. (1990) Effect of intensity in asbestos cohort exposure-response analysis. In: Occu pational Epidemiology (Edited by Sakurai, H. et at), pp. 189-193. World Health Organisation (1977) Manual of the Inter national Statistical Classification of Diseases, Injuries and Causes of Death. Vol. I. World Health Organisation, Geneva. APPENDIX Dust levels by period and by place ofemployment Liddell et al. (1997) explain how an exposure file had been created containing, for each of the nearly 11,000 men in the cohort, an annual record consisting of: (a) the fraction of the specific year worked; (b) an indicator of the length of the work week in that year (taken as 66 hours before 1938. 48 hours 1938 -1949 and 40 hours thereafter): and (c) an estimate of the level of dust, in millions of particles per cubic fool, or HWBUI0007927 Dust exposure and lung cancer in Quebec chrysolite miners and millers 19 fable II. Distributions of 243 cases and 970 referents St Company 3 in Thetford Mines according to years worked m 13 dust categories Years* worked in specific dust category 0 >0. <| 1, <5 5. <20 20- Total Dust category 1: % of 243 cases % of 970 referents Dust category 2; 0 o of 243 cases % of 970 referents Dust category 3: % of 243 cases % of 970 referents Dust category 4: % of 243 cases % of 970 referents Dust category 5: % of 243 cases % of 970 referents Dust category 6: % of 243 cases % of 970 referents Dust category 7: % of 243 cases % of 970 referents Dust category 8: % of 243 cases % of 970 referents Dust category 9: % of 243 cases % of 970 referents Dust category 10: % of 243 cases */ of 970 referents Dust category 11: % of 243 cases % of 970 referents Dust category 12: % of 243 cases % of 970 referents Dust category 13: % of 243 cases 0 o of 970 referents 82.7 2.9 7.4 6.2 0.8 100 82.0 4.4 7.3 4.3 2.0 100 29.2 14.4 20.6 26.3 9.5 37.0 12.4 14.8 25.3 10.5 100 100 32.5 9.1 23.9 30.0 4.5 100 38.4 9.8 24.5 22.4 4.9 100 55.1 6.2 20.6 16.0 2.1 100 62,5 5.6 17.5 13.1 1.3 100 70.8 5.8 17.3 77.0 4.8 15.3 6.2 2.9 -- 100 100 82.7 1.2 13.6 2.5 -- - 100 88.1 0.6 10.0 1.2 -- 100 86.8 i.6 10.3 1.2 -- 100 90.2 0.7 8.4 0.7 -- 100 73.7 2.9 19.8 3.7 -- too 84.1 3.1 10,9 1.9 -- 100 85.2 12.8 2.1 -- 100 88.6 1.2 9.4 0.8 -- 100 86.8 1.2 11.5 0.4 -- 100 91.2 1.6 6.4 0.5 0.2 100 75.3 QJ 10.7 13.2 -- 100 79.7 1.8 12.0 6.5 0.1 100 85.6 0.8 11.9 1.6 -- 100 87.7 1.9 8.7 1.8 100 73.7 6.6 12,3 6.2 1.2 100 79.0 4.3 11.1 5.3 0.3 100 adjusted for length of working week. Table 12. Regression coefficients at various stages of pooling dust categories: 488 cases and 1941 referents Years in dust category:-- 1 (0.5mpcf) 2 (2mpcf) 3 (7mpcf) 4 (12mpcf) 5(17 mpef) 6 (22mpcf) 7 (27mpcf) 8 (32 mpef) 9 (37 mpef) 10(42 mpef) 11 (47 mpef) 12 (70 mpef) 13 (140 mpef) Stage of pooling Stage 1(13 variables) Stage 2(12 variables) Stage 4 (10 variables) Stage 5 (9 variables) 0,00006 0.00561 0.01682 0.00290 0.01310 0.0809 S 0.1136 0.05931 0.01407 -0.08373 0.05594 -0.00866 0.01933 1 J -0.00002 0.00554 0.01668 0.00298 0.01340 0.08071 0.1134 0.05638 0.01220 -0.08730 0.05754 " 0.01064 -0.00093 0.00535 0.01732 0.00271 0.01636 0.07223 0.1138 0.00378 0.05476 0.01051 0.00534 0.01733 0.00272 0.01645 0.07217 0.1138 0.00381 0.05468 0.01054 HWBUI0007928 iiateaiwiivr.imi-*----- ry IieM, F. D. K. Liddell el a/. mpef. in 13 categories (Gibbs and Lachance, 1972). m which the man was employed in that year. Each man's record was compiled from the (calendar) year in which he started work to the year in which he finished and for each year m between. The longest period of service was found to be 58 years and, as first and last years were incomplete, the file had to allow for 59 annual entries; to retain ted record length, blank records were inserted for the years after each man finished work. As the file was enormous (5,9 Megabytes), a systematic sample was taken as follows. First, we selected for every man the record for the year after that in which he started work, i.e, usually his first complete year of work. Then, we added the records for every tenth succeeding year, i.e. the 12th. 22nd, 32nd. 42nd and 52nd. giving six records for each man. As expected, most records were blank (71%). usually because service had been completed earlier but some because ofa gap in service; a further 5% referred to less than a full year's service, while 350 were for men first employed in mines 49. The calendar year was entered by programme into the remaining 13.609 records, which were then classified into five periods, namely 1904-1937*. 1938-1949. 1950-1961. 19621973 and 1964-1985. Each of the small proportion of records of work in different dust categories within the year was allocated to the category the mean of which was closest to the (monthly weighted) average dust level which had been used in the basic 5.9 M-byte file. As there is no good reason for believing the members of the cohort were employed in selected areas of the industry', it seems appropriate to accept the 1 '* ^06 records as reasonably representative of dust conditions to the three places of employment. 1904-1985. Table 10 shows the distribution of these records according to dust category, by period, for each place of employment. Also given are the mean dust levels calculated from the levels extracted from the basic file. i.e. before re-categorisation. Distributions ofexposure measures Extremely high coefficients of variation of all the exposure measures were observed not only overall (Table 2) but also for both eases and referents at all three place ofemployment. Table 11 gives the distributions of the 1213 subjects first employed at Company 3 by (adjusted) years worked in each dust category. There were only four categories (2-4 and 11) jn which substantial numbers of cases were employed for 5 years or more and there were six categories (6-10 and 12) which had been worked in for such periods by less than 10 cases each and by no more than 18 referents. Only slight v ariaiions in the numbers of referents, which would probably arise from different random selections, would have had sub stantial effects on the LR statistics. Correlations between exposure measures I One important source of correlations between exposure measures was from men who were employed for long periods; they tended to stay in the same job. but the job itself would often be allocated to lowering dust categories as the environ ment-unprosed This led to correlations between the mea sures for adjacent categories, which would also have major effects on the LR statistics. Adjustments for length of ike working week There are many reasons why the mean years spent at specific dust levels in Table 1 ofVacek and McDonald (1990) are quite unlike what might at first sight appear to he cor responding means in our Table 2. much the most important lies in the treatment of the adjustment for length of work week. The earlier authors incorporated this adjustment into the 'dust value', so that, for example, a man who had been employed over a period 10 calendar years before 1938 in dust category 8 (averaging 32 mpef) would have been allocated the dust value (32 x 1.65 =) 52,8 mpef (i.e. in 'level' 7. the highest; SO or more mpef) and 'duration' of 10 years, so that the cumulative exposure would be 528 mpef x years. In the current paper, this example--with the same cumulative exposure of 528 mpef x years--would have been treated as (10 x 1.65 * ) 16.5 adjusted years at 32 mpef. Dust-fibre relationships Liddell et at, (1984) estimated a factor to convert dust counts to fibre counts as about 3.5 (fibres/m!)'mpef. but suited this would be quite unreliable except applied to mean dust levels for substantial groups of Quebec asbestos work ers. For the many jobs in which the 2217 men included in their study had worked, the fibre dust ratios had ranged from 0. 3 to 30 (fibres.-ml).'mpcf, virtually independently ofthe dust level; in the current study ratios job by job must have varied similarly, so that the classification of jobs by dust category would not be a reliable classification by fibre count. The two important reports by Gibbs and Lachance (1972); Gibbs and Lachance (1974) give some indication of the inherent complexity; a simple example is that work on the tailings dump in 1968 was extremely dusty but, as most of the fibre would have been eviracted. the fibre dust ratio must have been quite low. Elimination ofnegative regression coefficients In all the conditional regression analyses of the full model. 1. e. with 13 exposure measures, there was at least one negative regression coefficient, which taken at face value would imply a protective effect of exposure. Years in the highest relevant dust category were pooled with those in the adjacent category and the analysis was repeated. This process was iterated until either all coefficients had become positive, when it was terminated, or until the only negative coefficient was for category 1; in that circumstance, category 1 was eliminated from the model, which was equivalent to setting the coefficient to zero and the odds ratio to unity. The process is illustrated by Table 12, which was analyzed in five stages as follows:-- Stage Pooling I Cats. 12 & 13 3 Cats. 12 & 13 and Cats. 9 & 10 4 Cats. 12 & 13 and Cats. 8-10 5 Cats. 12 & 13 and Cats. 8-10 and eliminating Cat, 1 Dust categories Number of with negative variables coefficients 13 Cats. 10 and 12 12 Cats. 10 and 1 11 Cats. 9-10 and 1 10 Cat. 1 9 Not surprisingly in view of the methods of evaluation, a three-way split of this period was not revealing. Admittedly, there was a degree of arbitrariness in some of the pooling carried out. but every effort was made to retain any 'significant' effects. Perpair. A Comp a Therms Contrast GARRY BU Health and SaJ< The publ crocidolit with the good fort those use and anal fibrous d microsco 1938. Samp! showed t tnagnific current resolving Expo* 20fm!~ venttlati been sor The incidence o mid peritoneal the L.K. as a; related cancer? are on average cancers for eve factoring indu? thelioma regist for mesothdie and is current! 1995; Hodgsoi latency assocni (typically sonbased on hisio of exposure lc there is an atl pling and ana from the cum edge about f It ,cited III M. mmmmm .m.mmmvW mm HWBUI0007929