Relationship Between Lung Asbestos Fiber Type and Concentration and Relative Risk of Mesothelioma

Transcription

1 Reprinted from CANCER, Vol. 67, No.7, April I, Copyright 1991, by the American Cancer Society, Inc. J. B. Lippincott Company. Printed in U.SA Relationship Between Lung Asbestos Fiber Type and Concentration and Relative Risk of A Case-Control Study A. J. Rogers, MSc, * J. Leigh, MB, BS, MSc, MA, PhD, * G. Berry, MA, PhD,t D. A. Ferguson, MD,* H. B. Mulder, MSc,* and M. Ackad, MSc* Lung tissue from 221 definite and probable of malignant mesothelioma reported to the Australian Surveillance Program from January 1980 through December 1985 and from an age-sex frequency matched control series of 359 postmortem were examined by light microscopic (LM)and analytical transmission electron microscopic (TEM) analysis and energy dispersive x-ray analysis (EDAX).Concentrations of total fibers (coated and uncoated) (LM), crocidolite, amosite, chrysotile, and unidentified amphibole (TEM) (fibersjg dry lung tissue) were measured. Fiber concentrations < 10 JLm in length and :2: 10 JLm in length were separately quantified. By comparing (221) and controls (359 LM, 103 TEM), odds ratios for increasing fiber concentrations compared with < fibersjg (LM)and < fibersjg (TEM)(the respective detection limits) were calculated. Univariate analyses showed statistically significant dose-response relationships between odds ratio and fiber concentration for all fiber concentration measures. The relationship between log(odds ratio) and log(fiber concentration) was linear. Multiple logistic regression analysis showed that a model containing crocidolite :2: 10 JLm, amosite < 10 JLm, and chrysotile < 10 JLm as explanatory variables best described the data. The odds ratios for a XI0 increase in fiber concentration (fibersj JLg) were as follows: crocidolite :2: 10 JLm, 29.4 (95% confidence interval [CI], 3.6 to 241); chrysotile < 10 JLm, 15.7 (95% CI, 6.1 to 40); amosite < 10 JLm, 2.3 (95% CI, 1.0 to 5.3). An additive risk model gave similar results. In a subgroup of and controls with only chrysotile in the lungs, a significant trend in odds ratio with increasing fiber content was found. Cancer 67: , A ITER at.,l subsequent THE ORIGINAL research OBSERVATIONS in animals of andwagner humanset has now shown a clear causal link between exposure to certain types of asbestos fiber and malignant mesothelioma of the pleura and peritoneum. In reviewing the literature, Seaton 2 noticed an increasing gradient of potential to cause mesothelioma in humans from chrysotile (least) through amosite to crocidolite (greatest). Animal studies suggest that fiber geometry rather than chemistry is the significant factor in carcinogenic activity, with fibers greater than 8 JLm in length and less than 0.25,urn in diameter being the most carcinogenic.3,4 From the *National Institute of Occupational Health and Safety, and the tdepartment of Public Health, University of Sydney, Sydney, Australia. Supported by the National Health and Medical Research Council, the NSW Workers' Compensation (Dust Diseases) Board, and various other donors, and fully funded by the National Occupational Health and Safety Commission since The authors thank the Pathology Panel of the Royal College of Pathologists of Australasia, and all the colleagues throughout Australia who participated in the Australian Surveillance Program, particularly the state occupational health division medical officers, the state Department of Health community nurses, and the pathologists, for their invaluable assistance. Address for reprints: James Leigh, MB, BS, MSc, MA, PhD, National Institute of Occupational Health and Safety, G.P.O. Box 58, Sydney NSW 2001, Australia. Accepted for publication August 29,

2 1913 CANCER April Vol. 67 One gap in our knowledge, however, is the dose-response relationship for asbestos mesothelioma in humans, which is still not well characterized. The available evidence points to a direct relationship between some combination of duration and intensity of exposure, as well as fiber type, but this evidence is still tenuous. 5 Dose-response relationships are best determined by means of a retrospective or prospective cohort study but, in diseases of low incidence like mesothelioma, accurate exposure data are not often available for the higher levels of exposure prevailing in the past which have generated most current. Also, exposure at current levels is likely to be too low to produce sufficient for statistical analysis in a prospective cohort study. As an alternative to a cohort study, a case-control approach can be used, with either history of general environmental or occupational asbestos exposure (quantified or categorized) or lung asbestos fiber content in postmortem or resected lung specimens as the exposure variable. By comparing odds ratios for different exposure levels compared with zero exposure, an estimate of the dose effect on relative risk of mesothelioma can be obtained. Furthermore, the relative effects of different fiber types and sizes on mesothelioma risk can be separately quantified, even when, as is usually the case, exposure to multiple fiber types is found. This quantification can be made by multiple logistic regression analysis applied to the casecontrol study. Fiber content in the lung depends on both the amount of fiber deposited and amount cleared. The amount deposited depends on duration and intensity of exposure in the occupational or general environment. Clearance rate is thought to be dependent on the amount deposited at any point in time, i.e., clearance is exponential.6 Thus, the same fiber content in the lung at death or time of resection may be achieved from a high initial deposition, followed by absence of deposition and absence of clearance over a long period of time, or by a continuous deposition at a lower level, with or without clearance. Since detailed mechanisms of mesothelioma initiation and promotion are not known, "dose" as estimated by final lung fiber content may not relate to the "dose" required to produce mesothelioma. It is thus possible that a high lung fiber content in a mesothelioma case may represent continuing accumulation of fibers after a lower level of fibers had produced malignant change. It is more likely, however, that the malignant change did not occur until the fiber content reached a sufficiently high level. When comparisons are made with controls in whom no history of exposure is available, one must regard the "dose" of interest as the final lung fiber content, irrespective of the dynamics of its accumulation. In spite of these difficulties in interpretation, lung fiber content is a more certain index of exposure than retrospectively obtained work histories and estimates of past occupational and general environmental levels. Lung fiber counts have been shown to be reasonably well correlated with history of asbestos exposure.7,8 Deposition and clearance rates are also dependent on type and length of fiber. It has been suggested that long chrysotile fibers, being curly, are not deposited in the lung parenchyma.9,lo However, shorter chrysotile has a deposition fraction similar to amphiboley,12 Short-term animal studies also show that chrysotile is cleared more rapidly than crocidolite or amosite and that longer thinner chrysotile fibers are relatively less well cleared.4,13-15 Thus, any case-control study relating lung fiber content at death or resection to mesothelioma incidence must take fiber size and type into account. Studies of this type have been carried out by McDonald et ai.,16 Wagner et ai.,17 Mowe et ai.,s Gaudichet et ai., 18 and McDonald et ai.19 However, except for the last,19 none of these studies has been of sufficient size to quantify the dose-response relationship for fiber content, type, and size. The latter study unfortunately included 20 possible, doubtful, and unclassifiable in the 78 analyzed, with a resulting possible bias toward the null. esothelioma (yr) (15.0) (25.0) (30.0) (24.9)90 (10.9)12 (36.2)97 (2.9)0 (2.2) (4.1) (0.7)2 (32.4)18 (36.8) (25.9) (18.4) (4.5)0 (11.7)52 (35.3)32 (0.7)0 (20.9) (1.0)0 (2.3) (1.1) (0.6) (1.0)3 (10.4)30 (2\.3)47 (17.1)19 (0.9) (21.7) (14.5)42 (22.9)66 Percent2 (2.4)4 (0) (38.6) (35.9)42 (30.1)74 (9.9)3 201 Age-Sex No No Distribution Female Male Total of Cases andtable Controls 1. Controls

3 No.7 MESOTHELIOMA AND LUNG ASBESTOS FIBER CONTENT Rogers et at TABLE 2. Distribution of Fiber Concentration by Fiber Length: Light Microscopic Study, Uncoated Fibers lioma Percent No % ( )26 ( ) 0-<15, ( ) ( ) 93 Cornfield Controls Odds No. 48 ratio CI <10 I'm No ~lol'mpercent % 4.8 ( ) ( ) ( ) ( ) Cornfield Odds ratio CI Controls x', ~ (P < ) x', ~ (P < ) The current study further quantifies the dose-response relationship between risk of mesothelioma and asbestos exposure by comparing lung fiber content, by type, and length, in a large series of collected through a very complete national surveillance program,20 with that in a control group of lung tissue samples collected over the same period as the. The method is similar to that used in previously reported studies but a much larger series of (221) is analyzed and only definite or probable mesothelioma are accepted. Cases Materials and Methods Cases were obtained from notifications to the Australian Surveillance Program.20 This program operated between January 1, 1980 and December 31,1985. Formal voluntary notifications were sought from clinicians and pathologists throughout Australia. Cross-checks were made with registrations in statutory cancer registries in each state of Australia. Diagnosis of mesothelioma was based on histologic grounds. A panel of five experienced pathologists appointed by the Royal College of Pathologists of Australasia made a histologic classification of definite, probable, possible, and not mesothelioma, scored as 1,0.75,0.50, and 0, respectively. The definitive diagnosis was regarded as the category with score nearest to the mean score. Only definite or probable mesotheliomas were included in the current study. The total number of definite or probable mesothelioma diagnosed in the period was 697. In 221 of these (209 definite, 12 probable) formalin preserved postmortem lung tissue material was available for fiber analysis. Postmortem material from five possible was also available but was excluded from the current study. Two hundred seven were pleural and 14 were TABLE 3. Distribution of Fiber Concentration by Fiber Length: Light Microscopic Study, Coated Fibers Percent No <15, Controls No. 75 <10 I'm 28 7 Odds ratio Percent No. 95% ( ) ( ) ( ) ( ) ( ) Cornfield Controls Odds ~lol'm ratio CI X', = (P < ) 95% Cornfield CI 11.9 ( ) 20.0 ( ) 87.9 ( ) X', = (P < )

4 1915 CANCER April Vol. 67 thelioma TABLE 4. Distribution of Fiber Concentration by Fiber Length: Transmission Electron Microscopic Analysis, Chrysotile Percent No. 95% ( ) ( ) ( ) 0-<200,000 Cornfield Controls 2Odds ratio CI <10 I'm Percent I No. 95% ( ) ( ) Cornfield Controls Odds ;,,10 ratio cr I'm No. 181 X2, = 19.9 (P < ) X2, ~ (P < ) cr: confidence interval; fig: fibers per gram of dried lung tissue. peritoneal. In general, 5 X 5 X 5 cm blocks from the lower lobe of the uninvolved lung were analyzed. Controls As part of another study, formalin-preserved lung tissue was obtained from 391 consecutive necropsy specimens or lung resection specimens performed at the Royal Prince Alfred Hospital (Sydney, Australia) within the same period. Cases with pneumoconiosis, emphysema, pneumonia, or gastrointestinal cancer were excluded. This left 359 control tissue samples available for study (276 men, 83 women). The design was thus an unmatched case-control study with reasonable frequency matching for age and sex as seen from Table 1.The light microscopic control group included 23% women compared with 9% women among the. Fiber Content Analysis Lung fiber content was measured in 0.5-g samples of lung tissue after digestion in sodium hypochlorite. Digests were filtered onto Millipore 0.8 tlm (Millipore, Bedford, MA) filters for light microscopic study and Nuclepore 0.4 tlm filters (Nuclepore, Pleasantown, CA), carbon coated, for transmission electron microscopic (TEM) study using the methods of Rogers.7 For light microscopic (LM) analysis, 200 fields of view at LM X500 were counted using differential interference contrast. Fiber lengths were estimated with a Walton Beckett graticule. For TEM (JEOL 100CX [JEOL, Tokyo, Japan] and Phillips CM12 [Phillips, Eindhoven, The Netherlands]), 40 grid areas (ten grid areas from four separate grids) were examined at X8300. Fiber length was estimated from the proportions of the photographic screen or the Nuclepore hole replicas. Fiber type was identified by comparing the energy dispersive x-ray analysis (EDAX) spectrum of the fiber with those of the International Vnion Against Cancer (VICC) asbestos specimens prepared in the same manner. Only fibers> 2 tlm (TEM), >5 tlm (LM) in length were counted. Analytical sensitivity was 15,000 fibers per gram of dried lung (fibers/g) for light microscopic study and 200,000 fibers/g for TEM (corresponding to one fiber counted). helioma I TABLE 5. Distribution of Fiber Concentration by Fiber Length: Transmission Electron Microscopic Analysis, Amosite Percent No <200, Controls 7 <10 I'm I No. 2 Odds ratio -- Percent I No. Controls ;,,10 I'm X2, ~ 15.3 (P < ) Odds ratio 95% Cornfield cr 95% Cornfield cr 2.3 ( ) 5.3 ( ) 5.4 ( ) 6.9 ( ) 3.2 ( ) 3.8 ( ) 6.34 ( ) cr: confidence interval; fig: fibers per gram of dried lung tissue. X2, = 29.6 (P < )

5 No.7 MESOTHELIOMA AND LUNG ASBESTOS FIBER CONTENT Rogers et al ioma TABLE6. Distribution of Fiber Concentration by Fiber Length: Transmission Electron Microscopic Analysis, Crocidolite Percent ] No. 95% <200,000 ( ) ( ) ( ) Cornfield Controls Odds ratio CI <IO/Lm No. 132 II Percent No. 95% ( ) ( ) ( ) Cornfield Controls Odds."IO/Lm ratio CI X'l ~ (P < ) X2l = (P < ) Light microscopic fiber counts were made on all 221 and 359 controls. Electron microscopic fiber counts, EDAX, and fiber length measurements were made on all 221 and 103 male controls, drawn randomly from the total group of 276. Results Distributions of fiber content determined by light microscopic study and TEM (total and by fiber type and length) are shown for and controls in Tables 2 through 8. Dose-response analysis of relative risk was carried out by grouping fiber content levels into strata and calculating the relative risk (estimated by odds ratio) for each stratum of exposure compared with zero exposure (defined as < fibersjg by light microscopic study, or < 200,000 fibersjg by TEM). Where no controls were exposed, categories were pooled, as otherwise odds ratios are infinite. The Mantel chi-square test, with one degree of freedom for trend, was used to test the significance of linear trends of relative risk with dose. Tables 2 through 8 show univariate analyses of relative risk as a function of fiber content, type, and length. A progressive increase in relative risk with increasing fiber content can be seen for all fiber content measures. Tests for trend are highly significant in all. The increase in relative risk is apparent at the lowest dose levels but relative risk increased more rapidly for dosage greater than 3 X 105 (105.5) fibersjg (light microscopic study) and 106 to fibersjg (TEM). At the higher dose levels, there were often no exposed controls and many exposed. The relative risks for ~ 10,urn chrysotile, crocidolite, and total amphibole fibers are greater than the corresponding risks for fibers < 10,urn. When and controls are compared in relation to exposure to single-fiber types only, the number of and controls for analysis is much lower. However, increased relative risks for lung fiber contents greater than 200,000 fibersjg (all lengths) are still seen for single-fiber type lung contents of chrysotile and crocidolite. Fifteen 7-8 TABLE7. Distribution of Fiber Concentration by Fiber Length: Transmission Electron Microscopic Analysis, Unidentified Amphibole «10!Lm) 0.5 Percent I 9.0 II No. 95% ( ) ( ) ( ) Cornfield I Odds CIratioControls X2j = (P < ) fjg IOglO (fjg) CI: confidence interval; fjg: fibers per gram of dried lung tissue.

6 1917 CANCER April] 1991 Vol. 67 TABLE 8. Distribution of Fiber Concentration by Fiber Length: Transmission Electron Microscopic Analysis, Total Amphibole <1O,um Percent I No. 95% ( ) ( ) Cornfield Controls Odds ;.10 ratio CI,urn ( ) Percent IS No <200, Controls No. 100 Odds ratio X'I ~ 48.3 (P < ) 95% Cornfield CI 0.71 ( ) 1.02 ( ) 1.1 ( ) 3.5 ( ) 8.6 ( ) X'I = 41.9 (P < ) of 30 had ~ fibers/g crocidolite only compared with four controls of 39. In and controls where amosite only was found in the lungs six of 20 had ~ fibers/g in the lung compared with one of 36 controls. In and controls where only chrysotile was found in the lungs, (i.e., zero amphibole concentration) seven of 25 (two peritoneal) and three of 31 controls had ~ fibers/g in the lung (Table 9). Of the 25 with zero concentrations of amphibole found in the lung, 11 had a history of mixed exposure, two had a definite history of exposure only to chrysotile, and ten had no ascertainable history of exposure to asbestos. No history was available in two. Eleven had no fibers (chrysotile or amphibole) detected in the lung. Occupational histories of the controls were not available. Inspection of Tables 2 through 8 show a generally linear relationship between log odds ratio and log fiber content. This relationship is shown graphically for total uncoated fibers (all lengths, light microscopic study) in Figure 1. Accordingly, a series of multiple logistic regression models of the form: (P) n 10&(1 _ P) = bi + ~ bi 10glO(fibrecontent) + bn+1age where P = probability of having mesothelioma, and n = no. of different fiber types and lengths, were fitted using SAS 6.03 CATMOD (SAS Institute, Cary, NC). Only TEM fiber contents were used in this model. Age was included as a potential confounding variable for which and controls were approximately frequency matched. Since a log transformation was used, zero fiber contents were set to 0.2 fibers/,ug (200,000 fibers/g, the TEM detection limit). Table 10 shows that there are some significant correlations between the explanatory variables (loglofiber contents, age). As would be expected fiber contents ~ 10,urn and < 10,urn are highly correlated for all fiber types. Croci dolite and amosite are correlated and chrysotile < 10,um is correlated with amosite and crocidolite < 10,um. Age is significantly correlated only with chrysotile < 10,um (negative correlation, r = P < 0.02). Unidentified amphiboles are correlated with all other fibers. TABLE 9. Distribution of Fiber Concentration: Transmission Electron Microscopic Analysis, Chrysotile Only (All Lengths) o I Percent No. 95% 3 ( ) ( ) Cornfield Odds CI ratiocontrols fig loglo (fig) 0-200, X21 = 9.80 (P < )

7 No.7 MESOTHELIOMA AND LUNG ASBESTOS FIBER CONTENT Rogers et at (loge scale) 4'5 6' '5 (odds ratio) ':J LoglO(fibres/g) light (uncoated) FIG. 1. Relationship of lo&, (odds ratio) to 10glO(fiber concentration in lung) for total uncoated fibers by light microscopic analysis. Inclusion of highly correlated variables in a multiple logistic regression can lead to misleading results, so a judicious choice of models was made. Model parameter estimates are shown in Table 11. Inclusion of age and unidentified amphiboles ~ 10 Jim and < 10 Jim in any combination produced no significant improvement in fit to any model and were excluded from all models (chi-square} = 2.9 P > 0.05). The final model chosen to represent the data included crocidolite ~ 10 Jim, amosite < 10 Jim, and chrysotile < 10 Jim. Models containing other subsets of explanatory variables were difficult to interpret because of correlations and are not reported. The regression coefficients in the final model give the log(odds ratio) for an increase of one logarithmic unit of fiber dose (i.e., X 10 fibers/jig).the log.,(odds ratio) (10g., relative risk) for X 10 fibers/jig crocidolite ~ 10 Jim is thus 3.38 ± 1.96 X 1.07 (95% confidence interval [CI]). The multiplicative increase in relative risk is thus exp(3.38 ± 1.96 X 1.07) = 29.4 (95% CI ). Corresponding effects of amosite < 10.um and chrysotile < 10.um are as follows: amosite < 10.um 2.3( ), chrysotile < lo.um 15.7 (6.1-40). The logistic regression model is inherently a multiplicative risk model. Inspection of Tables 2 through 8 show that, although the risk of all fiber type concentrations is more than additive of those of single fiber types, it is not completely multiplicative. Accordingly, a series of models of the form n = 1 + L bixi _P 1 - P i~l were also fitted using GUM 3.77 (Numerical Algorithms Group, Oxford, UK). Regression coefficients of the two best models are shown in Table 12. Ninety-five percent CI for regression coefficients were obtained by finding values at which -2 log L increased by 3.84 (the likelihood profile method.) It can be seen that the relative importance of the various fiber types and lengths remain similar, with crocidolite ~ 10.um and chrysotile < 10 Jim being most important. Amosite ~ 10 Jim has a slightly larger relative effect. The results are thus consistent with the hypotheses that amphiboles are primarily responsible for mesothelioma and that longer fibers have a relatively greater effect. The dose-response relationships suggest that risk is related to dose. However, analysis of a small group of and controls with only chrysotile found in the lungs also showed a statistically significant trend (P < 0.005) of an increasing relative risk of mesothelioma with increasing fiber content in lungs (Table 9). Discussion Previous workers have estimated the relative risk (as approximated by odds ratio) for total amphibole lung content. In comparing their results with ours, we found an odds ratio of 2.36 (Cornfield 95% CI 1.38 to 4.05) for ~ 106 fibers/g compared with < 106 fibers/g for fibers < 10.um and (3.86 to 100.0) for fibers ~ 10.um. McDonald et at. 16 compared lung fiber contents in 99 of mesothelioma with available tissue, incident in Canada and the US in 1972, with 100 age-sex matched controls. The numbers of and controls at differing lung fiber content levels were tabulated but odds ratios AM> t Correlation 0.79* 0.27* 0.39* 0.45* UA 0.22* 0.35* t 0.09 AM< Croc> 0.30* 0.17* t 0.16t 0.31* 0.82* O.llt 0.50* Chry> 0.25* 0.41* UA> t Age< 10 < TABLE Matrix 10. of LoglO (Fibers/g), Age AM> 10!Lm AM < 10!Lm Croc> lo!lm Croc<IO!Lm Chry> 10!Lm Chry < 10!Lm UA> 10!Lm UA < lo!lm Age AM: amosite; Croc: crocidolite; Chry: chrysotile; UA: unidentified amphibole. * p < tp<o.oi. t P < 0.05.

8 b Modell TABLE 11. Multiple bsep Logistic Regression Coefficients CANCER 0.40 April Vol. 67 Predictor variable (all fiber concentrations) IOglO (f/ }J.g) Model 2 SE p Crocidolite Amosite Chrysotile Unidentified Amphibole Age (yr) -2 log L ~IO}J.m <10}J.m ~IO}J.m <lo}j.m ~IO}J.m <IO}J.m ~lo}j.m <lo}j.m SE: standard error; f/ }J.g: fibers per microgram of dried lung tissue. were not reported, nor were tests for trend carried out. Reanalysis of their data,6 in fact, shows an odds ratio of 3.8 (95% CI 1.8 to 8.0) for> 106 fibers/g compared with s106 fibers/g amosite and crocidolite combined and an increasing gradient of odds ratio with fiber content for amosite, crocidolite, and anthophyllite in men and women. This gradient was apparent for chrysotile only in the high (> 106 fibers/g) amphibole subgroup of those exposed to both chrysotile and amphibole. The dose-response relationship for tremolite was V-shaped. Similarly, reanalysis6 of the data of Wagner et al.17 (a study of 86 mesothelioma and 56 age-matched controls) gave an odds ratio of 7.4 (95% CI 3.5 to 16) for> 106 versus s 106 fibers/g combined amosite and crocidolite. A doseresponse gradient was not reported or recalculated. Mowe et at} in a study of 14 mesothelioma and 28 matched controls in Norway, showed an odds ratio of 8.5 (95% CI 2.3 to 31) for lung content> 106 fibers/g compared with s 106 fibers/g. Gaudichet et al.18 compared fiber content in 20 mesothelioma to four different age-sex matched control sets (squamous cell carcinoma of lung, adenocarcinoma oflung, metastatic carcinoma in lung, cardiovascular disease). Increased amphibole content was found in compared with controls for each control comparison, particularly with crocidolite and amosite and especially crocidolite and amosite > 8,urn in length. No difference in chrysotile content between and any control series was found. There were increased numbers of coated fibers in compared with controls. Dose-response relationships were not reported. In a more recent one-one matched case-control study on 78 identified by pathologists in Canada over the period 1979 through 1984, McDonald et al. 19 showed a clear dose-response relationship between mesothelioma risk and lung concentrations of long (~ 8,urn) and short «8,urn) crocidolite, amosite, and tremolite. However, multiple binary regression analysis showed that the relative risk was related only to fibers ~ 8,urn in length. From their Table 1, it can be computed that the relative risk (odds ratio) for total amphiboles in concentration ~ 1 fiber/,ug (106 fibers/g) compared with concentration less than 1 fiber/,ug is 2.2 (95% CI 1.1 to 4.4) for fibers < 8,urn and 6.6 (3.0 to 14.6) for fibers ~ 8,urn. Although a detailed matching procedure was used in this study, it is unfortunate that the criterion for inclusion of a case in the analysis on pathologic grounds was not more stringent. Cases classed as definite, probable, possible, doubtful, and unclassifiable were all included. Among these only 33 were definite and 25 probable. This wide inclusion criterion may have resulted in tumors which were not mesotheliomas, and with low lung fiber counts, being included, with a resulting bias in the results against any asbestos effects, thus underestimating relative risks. In making comparisons with the slightly different analysis of McDonald et al.,19 in which linear multivariate TABLE 12. Coefficients in Model of Form: R = 1 + 2: bixi Predictor variable ~lo}j.m % < < < < (all b CI CI P5.2 b Model 3Model CI: confidence interval. in ~ber/}j.g)

9 NO.7 MESOTHELIOMA AND LUNG ASBESTOS FIBER CONTENT Rogers et al risk models for long (~ 8!lm) and short «8!lm) fibers were separately fitted, it can be seen that we found lower coefficients for crocidolite and amosite and a higher coefficient for chrysotile. However, our coefficients are estimated with greatly shortened confidence intervals. We confirm that long amphibole fibers are more important than short, but differ somewhat in the relative effects of crocidolite, amosite and chrysotile. There is a clear dose response relationship in the current study. Although the negative correlation of chrysotile < 10!lm with age is consistent with a clearance effect, there was a significant dose response effect for short chrysotile, independently oflong amosite and long crocidolite. This is consistent with the analysis of the small group of and controls with only chrysotile found in the lungs which showed a statistically significant trend (P < 0.005) of relative risk of mesothelioma with lung fiber content (Table 9). In the current study, we were also able to examine a much larger (light microscope) case-control analysis of coated fibers (asbestos bodies), and to examine long (~ 10!lm) coated fibers separately. There was a very clear gradient of relative risk with coated fiber lung content, clearer still with fibers ~ 10!lm (Table 3). This gradient was of the same order as for amosite and for crocidolite. There are several possible selection biases operating in the current study. First, not all mesothelioma came to postmortem. The distribution by state of postmortem was closely comparable to the distribution by state of notification but there may have been other selection factors giving an increased chance of postmortem, e.g., likelihood of compensation for a relative. Secondly, the control group were patients coming to postmortem in one large referral hospital serving the whole of one state, New South Wales (NSW). New South Wales is the most populated state, with the largest number of mesothelioma notifications (426 in the study period). Whereas all controls would have been included as in the study, had they developed mesothelioma, their exposure background is more likely to have been that pertaining to NSW rather than to Australia as a whole. However, since most Australians live and work in cities with a similar mixture of industries and environments where asbestos exposure may occur, any bias due to this effect is likely to be small. Indeed, there is more opportunity for industrial and environmental exposure in NSW, so the restriction of controls to NSW would tend to underestimate relative risks. In conclusion, this study has shown that, for Australia as a whole, the greatest risk of mesothelioma is associated with exposure to crocidolite ~ 10!lm in length. It is difficult to assess the risks associated with chrysotile alone, due to almost universal mixed exposure to amphibole and chrysotile, and the possibility of differential clearance effects. In the current study there were two with only chrysotile found in the lungs and with an occupational history of exposure to chrysotile only. The results are generally consistent with the results of McDonald et al., 19 except that a considerably greater relative effect of crocidolite ~ 10!lm is apparent. This effect is very likely to be due to the wide distribution of Wittenoom crocidolite throughout Australia. REFERENCES I. Wagner JC, SleggsCA, Marchand P. Diffuse pleural mesothelioma and asbestos exposure in the North Western Cape Province. Br J Ind /vied 1960; 17: Seaton A. Asbestos-related diseases. In: Morgan WKC, Seaton A, eds. Occupational Lung Diseases. Philadelphia: WB Saunders, 1984; Stanton MF, Layard M, Tegeris A et al. Relation of particle dimension to carcinogenicity in amphibole asbestoses and other fibrous minerals. J Natl Cancer Inst 1981; 67: Davis JMG, Addison J, Bolton RE, Donaldson K, Jones AD, Smith T. The pathogenicity oflong versus short fibre samples of amosite asbestos administered to rats by inhalation and intraperitoneal injection. Br J Exp Patho/1986; 67: Davis JMG, McDonald Jc. Low level exposure to asbestos: Is there a cancer risk? Br J Ind /vied 1988; 45: Berry G, Rogers AJ, Pooley FD. s: Asbestos exposure and lung burden. In: Bignon J, Peto J, Saracci R, eds. Non-Occupational Exposure to Mineral Fibres. IARC, Lyon, France, 1989; 90: Rogers AJ. Determination of mineral fibre in human lung tissue by light microscopy and transmission electron microscopy. Ann Occup Hyg 1984; 28: Mowe G, Gylseth B, Hartveit F, Skang V. Fiber concentration in lung tissue of patients with malignant mesothelioma. Cancer 1985; 56: Middleton AP, Beckett ST, Davis JMG. Further observations on the short-term retention and clearance of asbestos by rats, using UICC reference samples. Ann Occup Hyg 1979; 22: Timbrell V. Deposition and retention of fibres in the human lung. Ann Occup Hyg 1982; 26: II. Roggli VL, Brady AR. Changes in numbers and dimensions of chrysotile asbestos fibers in lungs of rats following short-term exposure. Exp Lung Res 1984; 7: Roggli VL, George MH, Brody AR. Clearance and dimensional changes of crocidolite asbestos fiber isolated from lungs of rats following short term exposure. Environ Res 1987; 42: Abraham JL, Smith CM, Mossman B. Chrysotile and crocidolite asbestos pulmonary fibre concentrations and dimensions after inhalation and clearance in Fischer 344 rats. Ann Occup Hyg 1988; (Suppl) 32: Wagner JC, Berry G, Skidmore JW, Timbrell V. The effects of the inhalation of asbestos in rats. Br J Cancer 1974; 29: Middleton AP, Beckett ST, Davis JMG. A study of the short term retention and clearance of inhaled asbestos using UICC standard reference samples. In: Walton WH, ed. Inhaled Particles IV. Oxford: Pergamon Press, 1977; McDonald AD, McDonald JC, Pooley FD. Mineral fibre content oflungs in mesothelioma in North America. Ann Occup Hyg 1982; 26: Wagner JC, Pooley FD, Berry G et al. A pathological and mineralogical study of asbestos-related deaths in the United Kingdom in Ann Occup Hyg 1982; 26: Gaudichet A, Janson X, Monchaux G et at. Assessment by analytical microscopy of the total lung fibreburden in mesothelioma patients matched with four other pathological series.ann Occup Hyg 1988;(Suppl) 32: McDonald JC, Armstrong B, Case B et al. and asbestos fiber type: Evidence from lung tissue analysis. Cancer 1989; 63: Ferguson DA, Berry G, Jelihovsky T et at. The Australian Surveillance Program Med J Aust 1987; 147: