Document 15o3yqNMQRG8EjDBd8o2NnJaZ
FILE NAME: Z<
DATE:
DOC#: Z<
DOCUMENT DESCRIPTION :ZZ&ZZ >ZtZ
Ann OClUp H Ig, pp i-7 :1 The Author 2008 Published by Oxford Om\ ~rslt> PreS5-
on behalf of the Bnhsh OecupatiolwJ HygiL;ne SOC1t~ty
dOl 10 1093/dllll}'yg/nH:n036
Asbestos Fibre Concentrations in the Lungs of Brake Workers: Another Look
MURRAY M. FINKELSTEIN*
Family Medicine Centre, Mount Sinai Hospital, 60 Murray Street, 4th Floor, Toronto, Ontario, Canada M5T 3L9
Received 28 November 2007; in final form 29 April 2008
Objective: To reanalyse data on the lung content of asbestos fibres among brake mechanics. Methods: I re-analysed data published by Butnor, Roggli and colleagues on the lung content
of chrysotile and tremolite asbestos fibres among brake mechanics and controls. Statistics of the distributions were estimated by maximum likelihood to accommodate observations below the detection limit. Mean concentrations were compared by the t-test, bootstrap resampling and interval-censored survival methods.
Results: The mean concentrations of fibres were higher among the brake workers than the controls. The concentration of tremolite fibres was higher than the concentration of chrysotile, a pattern similar to that observed among Quebec chrysotile miners and millers.
Conclusions: Re-analysis of published data does not support the interpretation that, in automotive brake repair workers with malignant mesothelioma, asbestos content is within the normal range. The alternative interpretation that brake mechanics have a greater than background burden of asbestos fibres, attributable to occupational exposure to dusts from friction products manufactured from Canadian chrysotile, appears more credible. Tllis asbestos burden might be associated with an increased risk of asbestos-associated cancers.
Keywords: asbestos brakes; friction products; lung fihre content
INTRODUCTION
Mesothelioma is a rare tumour strongly associated with exposure to asbestos. During most of the 20th century, asbestos was an important constituent of vehicular braking systems (Paustenbach et al., 2004). In 1973, the US Environmental Protection Agency reported that brake linings contained 33-73% asbestos by weight and estimated that 32 million kg of asbestos were worn away from brake linings each year (Jacko and DuCharme, 1973). Automotive and brake mechanics have occasionally been diagnosed with mesothelioma, and the attributability of these tumours to asbestos exposure has become a contentious legal issue in the US. A number of reviews have been published by authors working for lawyers or involved with litigation. Lemen (2004) concluded that 'even the so called "controlled" use of asbestos containing brakes poses a health risk to workers, users, and their families'. Laden et al. (2004) concluded that 'the evidence did not ~upport an increase in risk of either
*Author to whom correspondence should be addressed.
Tel. +1-416-326-7879, fax: +1-416-326-7761; e-mail: murrayfinkelstem@utoronto.ca
lung cancer or mesothelioma among male automobile mechanics occupationally exposed to asbestos from brake repair'. Goodman et al. (2004) concluded that 'employment as a motor vehicle mechanic dues not increase the risk of developing mesothelioma'. I was asked by a defence-side law firm to examine the literature on mesothelioma and friction materials. Following review and discussion, I was not asked to prepare a written report. Other than this instance, I ha ve had no involvement with friction-materials litigatitln.
During my reading, I noticed that both Ladell et al. (2004) and Goodman et al. (2004) referred to ['. study of the analysis of asbestos fibres in the lungs of subjects with mesothelioma published by Roggli et al. (2002). Both sets of reviewers interpreted the data as being consistent with the opinion that 'brake du~t is unlikely to cause mesothelioma'. 'Individual' data for 10 cases from this series of subjects with mesothelioma were published in 2003 (Butnor el at., 2003 J. Commenting on these data, Butnor et at write 'Lung burden analyses in automotive brake repaJr workers with malignant mesothelioma in our selies reflect asbestos content within the normal range or elevated commercial amphiboles'. I have examined
lof7
20f7
M. M. Finke1stein
the brake worker data reported in Sutnor et al. (2003) and the data for 19 control subjects without occupational asbestos exposure published by Roggli et al. (2002) from the same laboratory. I believe that a different interpretation of these data is credible, and my analysis and interpretation are presented below.
METHODS
The data used in this study were abstracted from published reports. The study subjects were selected by Roggli and Butnor from the consultation files of Victor Roggli (Sutnor et al., 2003). The control subjects were a convenience sample of subjects without occupational asbestos exposure (Roggli et al., 2002). The counting methodology was described by Sutnor et al. (2003). 'Fibre analyses were performed on fonllalin-fixed or paraffin-embedded lung parenchyma using the sodium hypochlorite digestion procedure described (Roggli et al., 2004). Digested lung tissue was collected on 0.4 J.!m pore size Nuclepore filters. For scanning electron microscopic (SEM) analysis, the filter was mounted on a carbon disc with colloidal graphite and then sputter-coated with gold. Only fibres :;:'5 ~lm in length with a length to width ratio of at least 3:1 and approximately parallel sides were counted. Fibres meeting these criteria were quantified by examining 100 consecutive fields, with a total area of ~2.53 mm2, or until a 200 fibre count was reached. The limit of detection is ~400 fibres/g for a 0.3 g sample. For cases in which no asbestos fibres were detected, the value was repOtted as less than the detection limit for that case. The chemical composition of fibres was determined by energy dis-
persive X-ray analysis. Asbestos fibres were classified as commercial amphiboles, specifically amosite
+ crocidolite (AC), non-commercial amphiboles, in-
cluding tremolite, anthophyllite and actinolite (TAA), or chrysotile (Roggli et af., 1992b),.
I have made several assumptions in the analysis and interpretation of the Roggli and Butnor data. These are
1. I assume that most of the friction material products used in the North American market during the 20th century would have been manufactured using chrysotile ti'om the mines of Quebec (Virta, 2000).
2. Quebec chrysotile is known to be contaminated with fibrous tremolite (McDonald and McDonald, 1997). I assume that both chrysotile and tremolite fibres would be constituents of respirable brake dust.
3. Roggli and Butnor report counts of noncommercial amphiboles, including TAA as a single group. According to Roggli et al. ~1993), the vast majority of these fibres had the typical COlllposition of Si-Mg-Ca which is the chemical signature for tremolite. I have presumed that TAA is predominantly tremolite.
4. I assume that counts of asbestos fibre~ would be lognormally distributed in the lungs of both case and control subjects in the Roggli series. This assumption is confirmed by the Shapiw-Wilk test of normality on the log-transformed data for the tremolite content of the lungs of both the cases
(P > 0.75) and controls (P > 0.87). Figure 1
shows, for the control series, the fit of the tremolite data, above the detection limit, to ,he lognormal distribution. There were too few subjects
5271367
6617711
_7964055
'0
E :5
.~
0)1{)
3 l.0
55
6
65
7
75
Inverse Normal
Grid lines are 5, 10 25, 50. 75. 90, and 95 percentiles
Fig. 1. The fit of the observed tremolite data to a lognormal distribution.
Lung asbestos content of brake workers
3 of 7
with counts above the detection limit to examine in detail the statistical distribution of the chrysotile fibres.
tions. Case or control status was used as the independent variable in the survival regression model and the distribution of the 'survival times' (that is, the lung fibre concentrations) was taken to be lognormal.
One important feature of the fibre count data in the Roggli and Butnor publications is that manyobservations, particularly for chrysotile, were below the detection limit of the analytical method. Roggli and colleagues dealt with this issue by presenting the 'median' as the summary of the data distribution. Comparison of point estimates of the median, without consideration of its sampling distribution, is not an optimal method for comparing distributions between two series of subjects. It is possible, however, to estimate the parameters (mean and standard deviation) of the distributions using the method of maximum likelihood (ML) and to make quantitative comparisons of the distributions of fibre concentrations among the case and control subjects. In 2001, a colleague and I published a methodology, implemented in spreadsheet software, to perform the ML calculations (Finkelstein and Verma, 200 l). The ML method was used here to compute the parameters of the distributions of the counts of tremolite and chrysotile fibres in the lungs of the case and control subjects, assuming that the distributions were lognormal.
It is of interest to compare the distributions of fibre concentrations in the lungs of cases and controls. The standard method would be to use a Hest to compare the estimated means of the logarithmic distributions. However, because of the data points below the detection limit, it is not clear what are the correct degrees of freedom to be used in the t-test. The standard method was thus supplemented here by two other methods.
In the second method, the means of the (logtransformed) distributions among mesothelioma and control subjects were compared using the bootstrap method (Davison and Hinkley, 1997). Thirty bootstrap samples were drawn from each of the case and control series and the ML method was used to compute estimates of the means of the logtransformed data for each series. The difference between the means of the case and control series was then calculated for each bootstrap sample.
The third method was that of interval-censored survival analysis. In this method, we conceive of each fibre concentration data point as analogous to a 'length of survival' in a hypothetical survival analysis. Observations below the detection limit are treated as 'left censored'; that is, in the survival framework, they are thought to have occurred at an unknown time before entry to the study, where the
time of entry is here the analytic detection limit.
The program INTCENS (Griffin, 2005) written for the Stata statistical software package (StataCorp, 2005) was used to perform the regression calcula-
RESULTS
Table 1 shows the individual counts (fibres per gram wet lung tissue) of uncoated tremol:te and chrysotile fibres 2':5 Ilm in length, for the 10 brake workers and 19 control subjects, abstracted from the papers of Butnor et al. (2003) and Roggli et al. (2002). Table 2 shows the means and standard deviations of the distributions of fibre counts from the lungs of brake workers and control subjects computed by the method of ML. The concentrations of both chrysotile and tremolite were higher in the lungs
Table 1. Tremolite and chrysoule asbestos fibre countS (>5 /lm in length per gram wet lung) in the lung tissues of brake workers (n = 10, labelled C) and control subjects
(n = 19, labelled CL)
Subject number
Number of fibres: tremolite >5 /l
Number of fibres: chrysotite > S /l
Cl
2180
2180
C2
440
<440
C3
4630
<660
C4
720
<720
CS
1160
<580
C6
490
<490
C7
<340
<340
C8
240
<120
C9
3280
2180
CIO
2170
720
CLl
<990
<990
CL2
1770
<1770
CL3
210
<lOO
CIA
400
<400
CL5
<570
<570
CL6
2540
<2540
CL7
470
<470
CL8
<760
<760
CL9
890
<300
CLlO
<170
<170
CLlI
<1000
1000
CLl2
1310
<650
CLl3
960
<960
CLl4 CLl5
<790 <430
<790 <430
CLl6 CLl7
<510 370
510 <370
CLl8 CLl9
<600 <600
<600 <600
Data are obtained from the publications of Butnor et <If. (2003) and Roggh et al. (2002).
40f7
M. M. Finkelstein
of the brake workers than in the lungs of the control subjects.
Comparison of the fibre concentrations in the lungs of case and control subjects
The t-test comparison. The t-test of the hypothesis that the concentration of chrysotile was greater in the lungs of the brake workers than in the lungs of the control subjects had aP-value "" 0.13. The t-test ofthehypothesis that the concentration of tremolite was greater in the lungs of the brake workers than in the lungs of the control subjects had a P-value = 0.023.
Comparisons by the bootstrap method. In the bootstrap comparisons, I drew 30 bootstrap samples among the cases and controls. For each bootstrap sample, I computed the ML estimate of the mean of the log-transfomled data and subtracted the estimate of the mean (log concentration) for the controls from that of the estimate of the mean (log concentra-
Table 2. Means and standard deviations of the distributions of logarithmically transfonned fibre counts in the lungs of brake workers and control subjects
Brake workers (n =0 10)
Control subjects (n =0 19)
Chrysotile
Mean (standard deviation) of logarithmically transformed data
Tremolite
Mean (standard devIation) of logarithmically transfonned data
5 1 (1.9) 6.9 (1.1)
4.3 (l.4) 6.0 (1.0)
Statistics are computed by the method of ML.
tion) for the cases. Because only a few or' the chrysotile measurements exceeded the limit OT detection, among the 30 bootstrap draws for chrysotile were five (three case draws, two control draw~) in which none of the data points were greater than the detection limit. In those instances, the ML method did not converge to a positive value and those draws were omitted from the case-control comparisons.
Figure 2 shows the distributions of the means of the log-transformed tremolite fibre concentrations for the 30 bootstrap draws among the case and control subjects. For each bootstrap draw, the mean concentration among the controls was subtracted from that of the case subjects. Figure 3 shows, for chrysotile, the results of the subtraction of the estimates of the mean log(concentrations) for the controls i'om those of the cases for each of the 25 bootstrap samples. In 14 of 25 bootstrap samples, the mean concentration was higher among the cases and in 11 the mean concentration was higher among the controls. This small difference in proportions is consistent with no significant differences between the chrysotile concentrations among the case and control subjects. Figure 4 shows the results of the same computation for the 30 bootstrap draws for the tremolite counts. In all 30 bootstrap samples, the mean cO::1centration was higher among the cases than among the controls. This difference in proportions is consistent with a significant difference between the mean tremolite concentrations among the case and control subjects.
Comparisons by interval-censored sun [val analysis. The test of the hypothesis that the concentrations of chrysotile were greater in the lungs of the cases than in the lungsofthe controls had a P-value = 0.095. The test of the hypothesis that the concentrations of tremolite
Case
Control
O~--~----------~----------'-----------~-----------r
55
6
65
7
75
Mean (Log Concentration) Tremoilte Fibres
Graphs by CaseContlol
Fig. 2. Comparison, for brake workers and controls, of the distributions of ML-estimated means from 30 bootstrap results using tremolite data as source.
Lung asbestos content of brake workers
50f7
o
5
10
15
20
25
Boostrap Draw Number
Fig. 3. For chrysotile, the results of the subtraction of the estimates of the mean log(concentrations) for the controls from those of the cases for each of 25 bootstrap samples.
(/)
c
0 I/) :;:;
c[l!
ID U C 0
0
E c
0
0
ID
g25 i tD "-1
<: (I)
0 c
0
0
(I)
'"<1l
U 0
0
10
1 ..
20
~." r -
30
Boostrap Draw Number
Fig. 4. For tremolite, the results of the subtraction of the estimates of the mean log(concentrations) for the controls from those of the cases for each of 30 bootstrap samples.
were greater in the lungs ofthe cases than in the lungs of the controls had a P-value = 0.017.
The three comparison methods were thus in agreement; the mean of the log(concentrations) of tremolite, but not of chrysotile, was significantly higher in the lungs of the cases than the lungs of the control subjects.
DISCUSSJON
Measurements of asbestos fibre counts in the lungs of 10 brake workers were reported by Butnor et al.
(2003) and data for a series of 19 control subjects were reported by Roggli et al. (2002). After analysing their data, Butnor et al. wrote 'lung burden analyses in automotive brake repair workers with malignant mesothelioma in our series reflect asbestos content within the normal range or elevated commercial amphibo1es' . I believe that another interpretation of these data is credible.
Table 2 showed that, using the method of ML to deal with observations below the analytical detection limit, the mean concentrations of both chrysotlle and