Document rpOmxVO17gqngyQyj54Z2xn97

FILE NAME: Z< DATE: DOC#: Z< DOCUMENT DESCRIPTION :ZZ&ZZ >ZZtZ Inhalation Toxitology. 2011; 23(12): 681~ o 2011 Informa Healthcare USA-Ine. ISSN 0895-8378 printllSSN 1091-7691 online Dol: 103109/08958378.2011.580472 RESEARCH ARTICLE informa healthcare Asbestos fiber concentrations in the lungs ofbrake repair workers: commercial amphiholes levels are predictive of chrysotile levels Gary M. Marshl, Ada o. Youkl, and Victor L. Roggli2 IDepartment 0/Biostatistics, Graduate School o/Public Health, Genterfor Occupaticqud Biostatistics and Epidemiology, University o/Pittsburgh, Pittsburgh, PA, USA, and 2Department o/Pathology, Du/f;1tUniversity Medical Gente" Durham, Ne, USA P9i~~'andinftuential Abstract ' ,', " Objectives:To investigate the impact of extreme data points in An~steirts"~9 findings of no association between lung levels ofcommercial and non-commerdal amphiboles (pri.nerpall tr,~blite as Cl marker for chrysotile asbestos) in brake repair workers with mesothelioma. .i, Methods: We first identified potentialoutliers. high leverage points among lung levels of commercial amphiboles and tremolite among lS pel'$Ons"""f,onlyknqwn exposure to asbestos was through' brake repair work. We used sensitivity analysis and qua'rltil "io~tpaccountfor extreme data points and model commercial amphibole levels as a predictor of tre"wlite, We ~lsP,usedquantile regression to evaluate whether case-reported duration of employment as a brake rep~ , er p~lcted lung levels of commercial amphiboles or t r e m o l i t e . ' ", ' Results: We found lung levels of commer~;al amptlillr;iles are'a'stc!tistkally significant predictor of tremolite levels via sensitivity analysiS (f= 0.82. slope estimate P~'ae= O.()ot; R2=o.o8~and quahtile regression (slope estimate P-value <0.0001). Our data provide no evidente~qutadon:ofemployment as a brake repair worker was a predictorof lung levels of tremolite or commerciaUllnphjpOfes: . wark. ConclUSions: Our findings suggest t~f~lE!vatedlung leveJ$oftreJnOlite in the lungs of brake repair workers with elevated levels of amphiboles ar~cflpm concurrent exposures,to commercial amphibole and chrysotile asbestos in occupational settingScOthertttan'tmille repair These ~Ddings are supported by five new cases. The weight of the scientific eviden~edoesnqt~upport a rolefOr occupatj()rial exposure to brake dust and other friction products in the development g(ine$lJln~foma. Keywords: ~,arrit;hiboles, tremQlite, chrysqtiJe, amosite, crocidolite, occupational diseases, lung cancer. mesotheliom;i robust regressioni sensitivity analysis Introdudion Canadian chrysotile, which aceotinted for the vast majority of chrysotile used in the United States in the manufacture of asbestos products, is well-recognized to be contaminated with tremolite. It has been suggested that tremolite might be a reasonably good biomarker for exposure to Canadian chrysotile, as chrysotile tends to be degraded and removed from the lungs over time, whereas amphiboles like tremolite are more biopersistent (Churg, 1988). Indeed, analyses of lung tissue samples from Canadian chrysotile miners and millers have shown predominance of tremolite fibers (Churg et al, 1993). Furthermore, mesothelioma occurrence among Canadian miners and millers correlates best with the degree of tremolite contamination in the various mines (McDonald et al., 1997). In 2002, Roggli et al. reported the correlation between type of occupational asbestos exposure and asbestos fiber Address/or Correspondence: Gary M. Marsh, Department ofBiostatistics, Graduate School ofPubJic Health, University of Pittsburgh, Pittsburgh, PA 15261. USA. Thl: 412-624-3032. Pax: 412-624-9969. E-mail: gmarsh@CObe.pitt.edu (Received 19February2JJ11; revised OS April2011; accepted 08ApriI2JJl1) * 681 lA I .. HT. L N Kf} ,. ' f ' ," 682 G.M. Marsh et at. lungburden levels for 1445malignant mesothelioma cases. Butnor et al. (2003) reported results of lung liber analysJs for 10 malignant mesothelioma cases from Roggli et al. (2002) whose only known exposure to asbestos was occupational contact with friction products (e.g. brake dust as a brake repaIr worker). Commercial ampbiboles (amosite or crocidolite asbestos) were found in eight cases with excessive commercial amphibole libers found in five of the cases. The authors concluded that the presence of the elevated commercial ampbiboles in the lungs of some of these cases suggested exposures other than occupational exposure to brake dust, which contains small amounts (0.1-1%) ofmostlyshort(<l J.IDlin length) chrysotile asbestos fibers (Lynch, 1968; Hatch, 1970; Davis and Coniam, 1973; Williams and Muhlbaier, 1982; Wong, 2001). Finkelstein (2008) reported a reanalysis of the data in Butnor et al. (2003) using an alternative computation for mean fiber concentration, which resulted in mean concentrations of non-commercial asbestos expOsure (principally tremolite, with some actinolite and anthophyllite, measured as a marker of chrysotile exposure) that were higher than reported by Butnor et al. (2003). In a letter to the editor, Roggli et at (2009) considered this finding consistent with those ofButnor et al. (2003). In response to the Roggli et at (2009) letter, Finkelstein (2009) concluded from a regression analysis of the same 10 cases (Butnoretat, 2(03) thatlevels ofcommercial ampbiboles are not predictive of the levels oftremolite (as a marker of chrysotile exposure), thus implicating exposure to chrysotile fibers from friction products as a causative factor for these mesothelioma cases. We questioned the Validity of Finkelstein's 2009 conclusion due to the observation in his Figure 1 of at least two extreme data points that were not appropriately weighted in the regression analysis. The purpose of our reanalysis reported here was to investigate the robustness ofFmlcelstein's 2009 findings with respect to the presence of these extreme points. Our reanalysis also extends the work of Finkelstein (2009) by including amphibole and tremolite fiber concentration levels for five new cases of malignant mesothelioma from the Roggli et al. case series whose only known asbestos exposure was occupational contact as a brake repair worker. We also examined whether case-reported duration of employment as a brake repair worker was a predictor of lung levels of commercial amphiboles or tremolite. Methods We abstracted from Butnor et al. (2003) data on the levels of commercial amphibole fiber and tremolite fiber found in the lungs of 10 mesothelioma cases used in the Finkelstein (2008,2009) analyses. The 10 cases from Butnor et al. (2003) were selected on the basis of three criteria: (1) having mesothelioma, (2) only known asbestos exposure was occupational contact as a brake repair worker; and (3) lung tissue was available for analysis. In the process of preparing this manuscript, we identified and corrected three errors in Butnor et al. (2003) related to the lung fiber levels oftwo ofthe 10 cases. Specifically, for Case 2, the commercial amphiboles level changed from 4810 to 3940; for Case 3, commercial ampbiboles changed from 380 to 970 and tremolite changed from 4630 to 4110. We also included lung levels of commercial amphibole and treIDolite fibers for the five new mesothelioma cases from the Roggli et al. series who were observed since the publication of Butnor et al. (2003) and who met the same three inclusion criteria. We also included information on case-reported duration of employment as a brake repair worker for all 15 cases (Table I). 8.. Case 3 0 0 ~ en 1...J Q)O ~~ E !! l- -S . 0 0 Case 10 Case 2 2000 4000 Commercial Amphibole Levels Case 9 6000 Figure 1. The relationsbip between the levels of commercial ampblbole fibers and tremoUte ftbers In the lungs of brake repair workers: fitted simple linear regression model with 10 original data points (Model 1). From Butnor et al. (2003) and revised per Methods section. Inhalation ToxicolOQY ...1 ~.HT. L t N K4I) Fiber analyses and ascertainment of the diagnosis of mesothelioma in the five new cases were as described previously by Butnor et al. (2003) and are summarized in Table 2 along with other relevant case information. All were males with ages ranging from 40 to 70 years (median: 58 years). Four were pleural mesotheliomas and one was peritoneal. Four out offive were smokers or ex-smokers. The length of employment as a brake repair worker ranged from 5 to 32 years (median: 20 years). As in previous analyses (Roggli et al., 2002; Butnor et al., 2003; Finkelstein, 2009), we substituted a value of 1/2 the detection limit for fiber levels below the limit of detection. Cases with a less than (<) sign before the value are at the detection limit for that value. For example, as shown in Table 1, in Case 11 no asbestos fibers were Amphibole lung levels predict chrysotile levels 683 detected and the detection limit was 520. In Case 12, the asbestos body count is clearly elevated and a single crocidolite fiber was identified by SEMi the detection limit was 810 for that case. The detection limit varies from case to case because the amount of tissue available to analyze variedfrom case to case (i.e. theweight ofthe sample). For Case 15, one commerclal amphibole fiber and one noncommerclal amphlbole fiber were identified Chrysotlle was not detected (Table 2). We recognizedthat 15 cases representa relatively small sample from which to infer whether tremolite levels are correlated with commercial amphihole levels or whether either level is correlated with duration of employment Moreover, many data points for amphibole and tremolite levelswere belowthe detectionlimit (non-detects). These Table 1. Lung levels of commercial amphiboles and tremolite and reported duration ofemployment for 15 mesothelioma cases whose only known exposure to asbestos was occupational contact as a brake repair worker. Case Commercial amphibolesc.d TremoHtec.d Duration ofemployment (years) I" 3270 za 3940 3- 970 2180 37 440 27 4110 24 4" <720 (used 360) sa <580 (used 290) 720 40 1160 11 6" 490 490 7 7" 340 <340 (used 170) 15 8" 120 go 6000 240 40 3280 17 ID" 1440 2170 24 lIb <520 (used 260) 12b 810 <S20 (used 260) 5 <810 (used 405) 7 13b <480 (used 240) <480 (used 240) 25 14b <390 (used 195) <390 (used 195) 32 ISb 490 490 20 -Data abstracted from Butnor et al. (2003) and revised per Methods section. bNew cases from Roggll case series. eFibers per gram ofwet lung tissue for fibers 5 J.lm or greater in length. d(}ne-half the detection limit value used in analysis and shown in parentheses. Table 2. Demographic, pathologic, and occupational information, and results oflung tissue analysis for five new mesothelioma cases whose on!>' known exposure to asbestos was occupational contact as a brake repair worker. Case (from Age (years)/ Table 1) sex Tumor type/site Occupation Smoking (pack-year) Pleural plaque AB/If Chrysotileb 11 69/M BPI Brake and clutch 1-1.5 ppd (XS-S ND <7 <S2O repair(lH) years) 12 70/M EPI Auto mechanic Pipe (XS-manyyears) ND 90 <810 13 58/M EP) Automl!(;hanic NS NO 6.7 <480 14 58/M EPI Auto mechanic 1.5 ppd 45 years NO <29 <390 15 4O/M EPe QC parts inspector 0.5 ppd 23 years ND 2 <490 Median 58 <7 <490 Reference casesC 3 0.2-22) <600 (<100-1000) B, Biphasic; E, epithelial; IH, International Harvester; M, male; ND, not determined; NS, non-smoker; Pe, peritoneal; PI, pleural; ppd, packs per day; QC, quality control; XS, ex-smoker. AB, asbestos bodies; NAMF, non-asbestos mineral fiber. aAB/g, asbestOs bocUes/SW'et luns by light mkroscopy. hTotal coated (AB) and uncoated fibers ~5 JUIl (length)/g wet lung as determined by scanning electron microscopy and energy dispersive x-ray analysis. CMedian values and range (in parentheses) for 20 cases with normal lungs at autopsy, normal range tissue asbestos body counts, and no evidence of asbestos-related disease (Butnor et al., 2003). Cl 2011 Informa Healthcare USA,lnc. 684 G.M. Marsh et at. limitations notwithstanding, these 15 cases represent all eligible cases from the Butnor et al. (2003) and Roggli et a1. case series. Obtaining additional cases and/or using or developing methods that lower fiber detection levels were not practical options for our analysis. We first attempted to replicate the Finkelstein (2009) results by fitting a simple linear regression on the 10 data points from Butnoretal. (2003). Second, to assesswhether commercial amphiboJes are predictive oftremolite levels, we performed a simple linear regression and used regres~ sion diagnostics to locate any potential oudiers, high leverage, or lnfiuential points using all 15 data points. An oudier is defined as any point with a large residual (difference between observed and fitted values) value, a high leverage point is an observation with an extreme value in one of the predictor variables, and an influentIal point is one that substantially changes the estimates of the coefficients when removed. A standardized residual is the residual divided by the estimate of its standard error. Any standardized residual value over 2.0 is typically considered to be large. A standard marker for high leverage is anything greater than 3(k+1)/n, where k is the number of predictors in the model and n is the sample size. Criteria for a point to be considered a highly lnfiuential point is typically any value larger than 4/n (Chatterjee and Hadi, 2(06). We then performed a sensitivityanalysis by removing any points that were considered to be extreme (outliers, high leverage, or lnfiuential), refitting the regression model and comparing these results to the model based on all 15 points. We also fit a quantUe regression model to account for the undue influence ofthe extreme points without omitting them as in the sensitivity analysis. Unlike simple linear regression, which is based on a least squared deviation criterion, quantile regreSSion minimizes the sum of absolute residuals using a linear absolute value function. Because this function is symmetric, minimizing the sum of absolute residuals ensures that the same number of observations occur above and below the median. Because the quantile regression model estimates the relationship between a predictor variable(s) and specific quantiles ofthe outcome variable (rather than the mean of the outcome variable as in simple linear regression), the parameter estimates are robust with respect to large outliers or influential points (Koenker and Hallock, 2001j Hao and Naiman, 2007j Cameron and Travedi, 2009). We also used simple linear regression and quantile regression to evaluate the relationship between duration ofemployment as a brake repairworker and lunglevels of commercial amphiboles or tremolite. All statistical models were fit using the Stata Statistical Software Release 11 (StataCorp,2009). Results While data are not shown here, we replicated exactly the simple linear regression results ofFink.elstein (2009). Using the revJsed data for Cases2 and 3, we refitthe simple linear regression model as shown in Model 1 and Figure 1: Tremolite level = 981.01 + 0.29 (commercial amphiboles level) (Modell) As found using the original data, commercial amphiboles level was not a statistically significant predictor of tremolite (r =0.43, slope estimate P-value =0.21) and the J(l = 0.19 was low indicating a poor fit (only 19% of the variance explained by Modell). We refit the Finkelstein (2009) model using alll5 data points as shown in Model 2 and Figure 2: Tremolite level = 615.09 +0.38 (commercial amphiboles level) (Model 2) Unlike Model 1, commercial amphiboles level was a = = statistically sJgnificant predictor of tremolite (r = 0.53, slope estimate P-value 0.04) but the J(l 0.28 was low indicating a poor fit (only 28% of the variance explained by Model 2). Model 2 remains an inferior model because it does not properly account for extreme values. In Models 1 and 2, Case 2, Case 3, and possibly Cases 9 and 10 appear as extreme points that could affect the fitted line (Figures 1 and 2). We performed regression diagnostics to determine if these points were in fact causing undue influence to the fitted equation. Table 3 shows the computed residuals, standardized residuals, leverage, and inO.uence statistics. For these data, the cutofffor high leverage is 0.40 and for a hIghly lnfiuential point it is 0.27. In Table 3, the highlighted values are those that exceed the typical cutoffvalues. Case 3 is an outller and an infiuential point. Cases 2 and 10 also have the potential to be outliers as their residuals are much larger than the other residuals. Case 2 is also an influential point Case 9 is a high leverage point. We performed a sensitivity analysis by refitting the simple linear regreSSion of tremolite versus commercial amphiboles with the three extreme points removed (Cases 2,3, and 9). Case 10 was not removed because the cutoff was not exceeded for the standardized residual, leverage, and lnfiuence. The fitted line without these three points is shown in Model 3 and Figure 2: = Tremolite level 257.49 + 0.68 (commercial amphiboles level) (Model 3) The constant term in Model 3 is about 1/3 that of Model 2 and the slope estimate for commercial amphiboles increased almost 2.5-fold. Commercial amphibole concentration is now a statistically significant predictor of tremolite (r= 0.82, slope estimate P-value =0.001) and the Jl2=0.68 (68% of the variation is explained by this Model 3). e t Inhalation ToxicoloQV R. ~.H T I N K4} Thefitted quantile regression modelisshownln Model 4 and Figure 2: Tremolite level ::: 177.96 + 0.52 (commercial amphiboles level) {Model 4) As revealed by the slope estimates and the fitted lines In Figure 2, Model 4 falls between the model using all 15 points (Model 2) andthe model with the three extreme points omitted (Model 3). In our quantile regression model that properly accounts for extreme values without Table 3. Results ofregression diagnostics: residuaIs, standardized residuals, leverage and influence. Case Residuals Standardized residuals Leverage Influence I' 318.583 0.317 0.160 0.010 2" -1676.780 -1.745 0.234 0.465 3" 3125.201 2,951 0.069 0.322 4" 32.305 -0.031 0.087 0.0001 58 434.374 0.415 0.090 0.008 6" -311.853 -0.296 0.081 0.004 7" -574.682 -0.548 0.088 0.014 8" -420.832 -0.404 0.099 0.009 ga 378.076 0.540 0.594 0.213 10" 1006.066 lIb -454.191 0.949 -0.434 0.067 0.091 0.032 0.009 12" -518.827 l3b -466.569 -0.491 -0.446 0.072 0.092 0.009 0.010 14b -494.417 lSb -311.853 -0.473 -0.296 0.095 0.081 0.012 0.004 'Data abstracted from Butnor et al. (2003) and revised per Methods section. bNew cases from Roggli et al. case series. Amphibole lung levels predict chrysotile levels 685 omitting them (Model 4), commercial amphibole concentration is a statistically significantpredictor oftremo lite concentration (slope estimate P-value <0.0001). The correlation coefficient and R2 are not shown for the quantile regression as no accepted method exists for calculating these statistics. Figures 3 and 4 show the scatterplots of duration of employmentandlunglevels oftremoliteand commercial amphiboles, respectively, as well as the fitted lines based on the simple linear regression and quantile regression. Each scatterplot reveals evidence of influential points. The fitted quantile regression models are given by: Tremolite level:::359.23 +6.54 (duration of employment) Commercial amphiboles level = 517.58 - 3.94 (duration ofemployment) Neither model revealed any evidence that duration of employment as a brake repair worker was a predictor of lung fiber level (slope estimate Pvalues=O.90 for both models). Discussion Regression diagnostics, sensitivity analysis, and quantile regression are rigorous, well-accepted, and commonly used analytical procedures to ensure that the fit of a particular statistical model is not overly influenced by one or few observations (Koenker and Hallock, 2001; Chatterjee and Hadi, 2006; Hao and Naiman, 2007; Cameron and Travedi, 2009). Our use of these rigorous procedures clearly showed that four extreme data points (Cases 2, 3, 9, and 10) led Finkelstein (2009) to conclude incorrectly Case 3 Case 10 Case 2 o o ~---------------.---------------r--------------~ 2000 4000 6000 Commercial Amphibole Levels FigUre 2. The relationship between the levels of commercial amphibole fibers and tremolite ftbers in the lungs of brake repair workers: fitted simple linear regression model with all 15 data points(Model 2) and omittingtbree extreme points (Model 3). Also showing quantile regression model (Model 4). From Butnor et al. (2003) and revised per Methods section, and including five new cases from Roggli et al. case series. 2011 Informa Healthcare USA, Inc. " a tt T . L . .,. '~. ,<, ~ t N K4} " 686 G.M. Marsh et al. that lung levels of commercial ampbiboles are not predictive of the levels of tremolite as a marker of chrysotile asbestos. When we removed or properly accounted for these extreme observations using an appropriate regression model, we found that lung levels of commercial ampblboles are a statistically slgnlficant predictor oflung levels oftremolite. We performed several confirmatory analyses to determine if our results were sensitive to the method of data substitution or data analysis. For example, since quantile regression is only one of several statistical approaches for handling data sets that deviate from the usual assumptions of simple linear regression, such as extreme observations and absence of homoscedasticity, we fit two regression models as alternatives to Model 4. We also evaluated whether the results ofModel 4 and the alternative models were sensitive to the method used (1/2 detectable level (OL for substituting values for non-detectable lung levels of asbestos. Finally, we refit Models 2-4 using the original 10 data points reported by Fmkelstein (2009) and revised per the Methods section. The first alternative regression model considered was robust regression. This model first removes points with an influence statistic greater than 1.0 (none in our Case 3 Case9 Case 10 Case 1 Simple linear regression -----!---------------------------- Quantile regression model - 1 - - - - - - - -- - - -- - - - - - - - o o 10 20 30 40 Duration of Employment Figure 3. The relationship between duration ofemployment as a brake repair worker and the levels oftremolite fibers in the lungs ofbrake repair workers: fitted simple linear regression model with ail 15 data points and quantlle regression model Case 9 Case 2 Case 1 Simple linear regression ------------------!--------------- Quantile regression model -----.--~-----------. o ~----------_r----------_.-----------.----------__r o 10 20 40 Duration of Employment Figure 4. The relationship between duration ofemployment as a brake repair worker and the levels of commercial amphibole libers in the lungs of brake repair workers: fitted simple linear regression model with all 15 data points and quantUe regression model. Inhalation ToxicoiOQY 51 I " T . L N K.) . 1 . > I' ,. , ~ . . ~ . , < .. Amphibole lung levels predict chrysotile levels 687 analysis), then performs a weighted least squares regression using weights that give small orzero weight to points with large residuals (Berl<, 1990; Chatterjee and Hadi, 2006). In the second alternative regression model, an unweighted least squares criterion (simple linear regressIon) was used to fit logarithmic-transformed values of tremolite and commercial amphibole lung levels. Here, the logarithmic transformation gives less weight to the extreme values. Alternative methods considered for the substitution of non-detects were DL/2112 and 2DL/3 as proposed by Sanford et al. (1993) and discussed by Helsel (2006,2010). Although data are not shown here, both alternative regression models considered led to similar fitted models that corroborated the central result of Model 4, that is, lung levels of commercial amphiboles are a statistically significant predictor of lung levels of tremolite. Moreover, the choice of non-detect substitution method had negligible impact on the DL/2-based results of Model 4 and the two alternative regression models. We also found similar results when Models 2-4 were refit using the 10 original data points revised per the Methods section. Ourfinding thatlunglevels ofcommercial amphiboles predict lung levels of tremolite was supported by our related findings that case-reported duration of employment as a brake repair worker was not associated with lung levels of tremollte or commercial amphiboles. That is, one would expect lung levels of tremolite (chrysotUe) to increase with increasing duration ofemployment only ifthere was an association between friction products and mesothelioma. Also, because automotive brakes in the UnitedStatesinclude onlythechrysotileform ofasbestos, one would not expect lung levels of commercial amphiboles to increase with blcreasing duration of employment as a brake repair worke& This laclc of correlation between exposure duration and lung levels of tremolite or commercial amphiboles is an important and heretofore unreported observation, although the raw data were available in the original Butnor et aI:s (2003) work. Thus, our findings suggest that elevated lung levels of tremolite (as a marker ofchrysotile) in the lungs ofsome brake repair workers with elevated levels of amphiboles arose from concurrent exposures to commercial amphibole and chrysotUe asbestos in occupational settings other than brake repair work. Regarding the study by Butnor et at (2003), some investigators have questioned the validity of using scanning electron microscopy (SEM) for the analysis of lung tissue samples. Reanalyzing lung tissue from Case 1 of Butnor et al (2003), Dodson et al. (2008) claimed that transmission electron microscopy (TEM) is the superior instrument. As we have stated elsewhere, this is a moot poblt, becauselungfiberburdenstudiesofmesothelloma using TEM have come to similar conclusions as studies using SEM (Srebo et al, 1995; Roggli and Vollmer, 2008; Roggli et al., 2(08). Moreover, studies from two laboratories employing TEM to analyze a smaller number oflung tissue samples of brake mechanics with mesothelioma found similar results to those we reported. Dodson et al. (2005,2008) reported two cases ofmesothelioma (one in a brake mechanic and the other an auto machinist), both of which had elevated levels of commercial amphiboles (the auto machinist case was a reanalysis of Case 1 from Butnor et al., 2003). In fact, Hammar and Dodson have written elsewhere that "the question as to whether automotive mechanics .,. have an blcreased risk ofmalignant mesothelioma remains unresolved and contentious" (Hammar et al., 2008). Furthermore, Gordon and Dikman reported eight cases of mesothelioma in brake mechanics. All eight had only chrysotile in their lung tissue samples, and the chrysotile in each instance was within their reported background population range (Gonion and Dikman, 2009). Because the findings bl these two TEM studies are the same as what we have reported (either background levels of asbestos or elevated commercial amphiboles with orwithout elevated tremolite or chrysotile), there is no validity to the claim that TEM is superior to SEM for this purpose. As noted in the study of Butnor et al. (2003) and summarized here, our reanalysis and extension of the work of Finkelstein (2009) has some underlying limitations. First, our reanalysis is based on a case series of 15 medico-Iegal cases that may notbe representative ofall individuals exposed to friction products occupationally. Also, historical information obtained by patient interview is subject to recall bias for events occurring decades previously. This may explain the absence of reported exposures to commercial amphiboles among the cases considered. Additionally, many data points for amphibole and tremolite levels associated with the 15 cases were below the detection limit (non-detects), As non-detects can complicate statistical analysis, they reflect the very low levels of asbestos bl lung tissue, which in itself is indicative of a lack of a relationship between asbestos exposure and mesothelioma in brake repair workers. Conclusions Finkelstein's 2009 finding that lung levels of commercial amphiboles are not predictive of the levels of tremollte asbestos (chrysotiIe) is blvalid because the regression analysis did not properly account for extreme data poblts. Our regreSSion diagnostics, sensitivity analysis, quantile regression, and evaluation of reported duration ofemployment as a brake repair worker suggest that elevated lung levels of tremolite (chrysotile) in the lungs of brake repair workers with elevated levels of amphiboles arose from concurrent exposures to commercial amphibole and chrysotile asbestos in occupational settings other than brake repair work. These findings are supported by the five additionally reported cases in the current study. Despite suggestions to the contrary by Dodson et al (2008) and Finkelstein (2008, 2009), the weight of the scientific evidence does not support a role 2011 'nforma Healthcare USA.Inc. 688 G.M. Marsh et aJ. for occupational exposure to brake dust and other friction products in the development of mesothelioma. Declaration of interest Drs. Marsh and Youk performed this work (study and manuscript preparation) as a private consulting activity for Honeywell International, Inc. Dr. Marsh was on the Asbestos Panel of the Scientific Advisory Board of the us Environmental Protection Agency. He provided expert testimony for defendants in three asbestos litigation cases, including two involving friction products. Dr. Marsh also provided epidemiological and biostatistical consultation for defendants, including Honeywell, International, Inc., in cases involving friction products. Dr. Youk was a statistical consultant and performed analyses in preparation for the trial of a defendant in an asbestos litigation case. Dr. Roggli has served as an expert witness for both plaintiffs and defendants in asbestos litigation, including Honeywell in cases involving friction products. References Berk RA. 1990. A primer on robust regression. In Modem Methods of Data Analysis, Fox J, Long Js. eels. Newbury Park, CA: Sage Publications, pp. 292-324. Butnor KJ, Spom TA, Rogg6 VL. 2003. Exposure to brake dust and malignant mesothelioma: a study of 10 cases with mineral flber analyses. Ann OC(:Up Hyg 47:325-330. Cameron AC, Trivedi PK. 2009. Microeconometrics using Stata. College Station, TX: Stata Press. Chatterjee S, Hadi AS. 2001;' Regression Analysis by Example, 4th edition. Hoboken, NI: John WJley & Sons, Inc. Churg A. 1988. Chrysotile, tremolite, and malignant mesothelioma in man. Chest 3:621-628. Churg A, Wright IL, Veda) S. 1993. Fiber burden and patterns of asbestos-related disease in chrysotile miners and millers. Am Rev Respir Dis 148:25-31. Davis IM, Coniam SW. 1973. Experimental studies on the effects of heated chrysotile asbestos and automobile brake lining dust injected into the body cavities of mice. Exp Mol PatholI9:339-353. Dodson RI} Grae! R, Shepherd S, O'SUl1lvan M, Levin J. 2005. Asbestos burden In cases of mesothelioma from individuals from various regions of the United States. IDtrastruct Pathol 29:415-433. Dodson RP, Hammar sP, Poye LW. 2008. A technical comparison of evaluating asbestos concentration by phase-contrast mlcroscopy (PCM), scanning electron microscopy (SEM), and analytical transmission electron microscopy (ATEM) as Illustrated from data generated from a case report.lnhal Toxicol20:723-732. Finkelsteln MM 2008. Asbestos fibre concentrations in the lungs of brake workers: another look. Ann Occup HY852:455-461. Finkelstein MM. 2009. Asbestos fibre concentrations in the lungs of brake workers: reply to Roggli et al. Ann Occup Hyg 53:192-193. Gordon R, Dlkman S. 2009. Asbestos fiber burden analysis of lung and lymph nodes In 100 cases of mesothelioma. Am J Respir Crit Care Med 179:A5892. Hammar SP, Henderson DW, Klebe S. Dodson RF. 2008. Neoplasms of the Pleura, Chapter 43. In Dail and Hammar's Pulmonary Pathology, 3rd edition, Tomashefsld JP, Cagle PT, Farver CF, Fraire AE, eds. Springer-Veriag: New York, pp. 558-734. Hao L, Naiman DQ. 2007. Quantile Regression. Thousand Oaks, CA: Sage Publications. Hatch D. 1970. Possible alternatives to asbestos as a friction material. Ann Occup Hyg 13:25-29. Helsel DR. 2006. Fabricating data: how substituting values for nondetects can ruin results, and what can be done about it. Chemosphere 65:2434-2439. Helsel D. 2010. Much ado about next to nothing: incorporating nondetects In science. Ann Occup Hyg 54:257-262. Koenker R, Hallock KF. 2001. Quantiie regression. J Eeon Perspect 15:143-156. Lynch JR. 1968. Brake lining decomposition products. J AIr Pollut Control Assoc 18:824-826. McDonald AD, Case BW, ChuIg A, Dufresne A, Gibbs GW, Sebastien P, McDonald JC. 1997. Mesothelioma in Quebec chrysotiie miners and millers: epidemiology and aetiology. Ann Occup Hyg 41:707-719. Roggli VL, Sharma A, Butnor KJ, Spom T, Vollmer RT. 2002. Malignant mesothelioma and occupational exposure to asbestos: a clinico- pathological correlation of 1445 cases. IDtrastruct Pathol 26:55-65. RoggIi VL, Spom TA, Butnor KJ. 2008. Comments on Dodson et al. A technical comparison of evaluating asbestos concentration by phase contrast mIcroscopy (PCM), scanning electron microscopy (SEM), andanalyticaltransmIsslonelectronmIcroscopy(ATEM)asillustrated from data generated from a case report.lnhal Toxicol20:1113-Il14. RoggliVL, Sporn TA, Case BW, Butnor KJ. 2009. Comments on Asbestos fibre concentrations In the lungs of brake workers: another look. Ann Occup Hyg 53:191. RoggIi VL, VoUmer HT. 2008. Twenty-five years of fiber analysis: what have we leamed? Hum PathoI39:307-315. Sanford RT, Pierson CT, Crovelll RA. 1993. An objective replacement method for censored geochemical data. Math Geol25:59-80. Srebo SH, Roggli VL, Samsa GP. 1995. Malignant mesothelioma associatedwith lowpulmonarytissue asbestos burdens: alight and scanning electron microscopic analysis of 18 cases. Mod Pathol 8:614-621. StataCorp. 2009. Stata Statistical Software: Release 11. College Station, TX: StataCorp LP. WllHams RL, Muhlbaier JL. 1982. Asbestos brake emissions. Environ Res 29:70-82. WongO. 2001. Malignant mesothelioma and asbestos exposure among auto mechanics: appralsal of scientific evidence. Regul Toncol Pbarmacol34:17o-177. Inhalation ToxicoloQV R " T. L I fill Kt} .' I' 'It '" . .