Secondary Logo

Journal Logo


Carbon Black and Lung Cancer Mortality—A Meta-regression Analysis Based on Three Occupational Cohort Studies

Yong, Mei PhD; Anderle, Laura PhD; Levy, Len PhD; McCunney, Robert J. MD, MPH

Author Information
Journal of Occupational and Environmental Medicine: November 2019 - Volume 61 - Issue 11 - p 949–00
doi: 10.1097/JOM.0000000000001713


The effects of dust exposure on human health are once again the focus of research since the European Commission is currently assessing whether a classification of titanium dioxide (TiO2) of the regulation regarding classification, labeling, and packaging (CLP) as a “substance with a suspected carcinogenic effect in humans” through inhalation. It was made clear in the assessment that the hazard described for TiO2, the general particle effects, applies to more than 300 substances in powder form known as “poorly soluble particles with low acute toxicity (PSLTs).” A range of particles, such as carbon black (CB), titanium dioxide (TiO2), talc, and coal dusts have the biophysical properties of PSLTs. Generally, PSLTs are not known to have substance-specific toxicity. Almost 30 years ago, Morrow1 postulated possible mechanisms to explain the dust overloading of the lungs, characterized by progressive reduction of particle clearance deep in the lung primarily due to a breakdown of alveolar macrophage-mediated dust removal. Recent reviews elucidate consistently that the mechanism of PSLTs in inducing cancer in rats, is a secondary genotoxicity resulting from lung overload and a cascade of inflammatory responses due to prolonged exposure to inhaled particles at sufficiently high concentration.2,3 However, the relevance of the rat lung response to particle overload for human risk assessment remains a matter of continuing debate because of different particle lung translocation pattern in humans under overload conditions.2,4

Carbon black is an industrially produced particulate form of nearly pure elemental carbon. Its health risks have been extensively studied both in laboratory animal studies and in epidemiological studies. In 2006, the International Agency for Research on Cancer (IARC) classified carbon black as possibly carcinogenic to humans (Group 2B) based on lung cancer findings in rats.5 However, evidence in humans did not suggest an increased lung cancer risk and therefore was considered inadequate.5 Furthermore, the IARC working group6 identified lack of exposure–response analyses to be a gap for evaluation.

For evaluation of causality, there are several criterions in epidemiological research,7 such as consistency of findings and exposure–response relationship, along with the biological plausibility. For quantitative risk assessment, an appropriate use of well-conducted exposure–response evaluation is useful for regulatory risk assessments for the purposes of quantifying the excess risk estimate per unit increase of exposure. Furthermore, evaluation of an exposure–response relationships can be performed to project an exposure scenario that has not yet occurred, or a sufficient latency period since exposure.8 However, the three existing cohort studies of carbon black production workers, which represented different exposure levels, did not provide conclusive evidence of exposure–response relationship in these investigations.

Against this background, the current study aims to perform a meta-regression analysis to combine the three existing cohort studies in carbon black covering a wider range of exposures to determine the nature of any exposure–response relationship of carcinogenicity.


The present meta-regression is based on the mortality cohort studies of exposed workers of carbon black production in US, UK, and Germany. After the IARC evaluation in 2006, a re-analysis of the UK data,9 an up-dated follow-up study of the US cohort10 that aimed to address the shortcomings identified by the IARC working group in 2006. The studies have been described in a previous review11 and represent the only three mortality studies of carbon black workers, including the re-analysis of the UK cohort after the IARC evaluation in 2006.

Description of the Studies

The characteristics and major results of the studies are summarized in Table 1. Brief highlights of the studies follow.

Review of Cohort Studies on Occupational Exposure and Mortality Risks

The US cohort, the largest one among the existing cohorts, involved 5011 carbon black production workers, some employed as early as the 1930s in the United States. The initial evaluation published in 200612 indicated that no risk for any type of cancer or non-malignant pulmonary disease was significantly elevated.

An updated evaluation10 with follow-up through 2011, which was designed in follow up an IARC evaluation in 2006 in which a recommendation to conduct a dose–response assessment of this cohort was made by the IARC Working Group,5 indicated a significant deficit for all-cause mortality, all malignant diseases, lung cancer, and marginally significant deficit for non-malignant diseases.

The UK cohort involved 1147 male manual workers with follow-up from 1951 through 1996.13 An elevated standardized mortality ratio (SMR) of 1.13 (95% CI: 1.02 to 1.25) was noted for all-cause mortality, SMR of 1.42 (95% CI: 1.19 to 1.68) for all malignant diseases, SMR of 1.61 (95% CI: 1.29 to 2.00) for lung cancer, and SMR of 1.07 (95% CI: 0.75 to 1.49) for non-malignant diseases. The UK cohort was re-evaluated after the completion of the evaluation by IARC.5 Elevated lung cancer risk was shown in two of the five plants, but not in the other three combined. In their re-evaluation,9 the investigators focused on the potential risk of lung cancer due to CB exposure in the past 15 years of employment, which is the so-called “lugging” effect, in contrast to the lagging effect from the early exposure. Elevated lung cancer risks were limited to workers with employment in the most recent 15 years, while this “lugging” effect was not confirmed with the German cohort.14 The investigators concluded cautiously that their findings could be the results of misclassification and potential confounding effects, such as smoking and prior occupational exposures.

The German cohort involved 1522 production workers with a follow-up from 1976 through 1998. Using national reference rates, an initial evaluation showed an elevated SMRs for all-cause mortality and for lung cancer, based on 50 cases.15

Several detailed analyses followed to clarify the uncertainties resulting from the effect of smoking and different referent rates.16–18 The smoking prevalence of this cohort was particularly high, which may result in an overestimate of the SMR for lung cancer. A nested case-control study from this cohort was conducted to estimate the biasing effects due to selection and confounding from smoking, prior exposures, and participation and being a prisoner during World War II. No association to carbon black exposure was found based on the analyses of 50 lung cancer cases.18

SMRs were then calculated using national, and regional expected mortality rates.16 Using the derived correction factor, the approach developed by Axelson and Steenland19 was applied to adjust SMRs for smoking and prior exposures. The SMRs for lung cancer were not significant, and reduced to about 1.33 (95% CI: 0.98 to 1.77) when compared with the referent rates of West German. The SMRs for lung cancer were reduced to 1.27 (95% CI: 0.93 to 1.69) and 1.20 (95% CI: 0.88 to 1.59), compared with the more relevant local reference rates from North-Rhine Westphalia and from Cologne, respectively.16

Furthermore, Cox regression analyses were performed to estimate the impact of various exposure metrics based on modified job exposure matrices, respectively for the full cohort and inception cohort.17 No positive association was suggested with carbon black exposure indices.

Information on Exposure Assessment and Data Extraction

From the US cohort study by Dell et al,10 a job-exposure-matrix (JEM) incorporating five similar exposure groups (SEGs) was constructed using more than 8000 time-weighted average measurements collected from 1979 through 2007. For measurements in years, where no measurements were available, interpolation by means of linear regression models was applied. Measurements for total dust were converted to inhalable dust measurements with an empirical factor of 2.97:1 (inhalable to total).20 We used hazard ratios from Cox regression analyses, respectively with no, 10-year, and 20-year lag of cumulative CB exposure. Categorizing the exposure based on quartiles of cumulative CB exposure (less than 12, 12 to 34, 34 to 72, more than or equal to 72 mg/m3 yrs), numerically elevated lung cancer mortality risks were indicated, though with no statistically significant trend.

From the UK cohort study by Sorahan et al,13 retrospective exposure assessment was conducted. A job-exposure-matrix (JEM) for inhalable CB dust exposure was constructed, based on 12 job categories incorporated with job histories, documented by year. We used relative risk for cumulative CB exposure with no lag, and 20-year lag, derived from Poisson regression models.

Compared with cumulative exposure to CB less than 20 mg/m3 yrs (reference), higher exposure categories (20 to 49, 50 to 99, more than or equal to 100 mg/m3 yrs) did not indicate elevated lung cancer mortality (0.89, 0.61, 0.70). No trend was observed (P = 0.25). With 20-year lag, relative risks (RRs) decreased with increasing exposure.

The German cohort represented relatively lower exposure levels. For this cohort, we initially used the results from Wellmann et al15 for the primary meta-regression analysis, followed by a series of sensitivity analyses. Wellmann et al15 constructed a JEM with three levels of detail: plant section, subsection, and job title. To each job title, a CB exposure score for 1960 and for 1980 was assigned. From Poisson regression without adjustment for smoking, five categories of cumulative CB exposure were classified. Taking the category of 0–5 mg/m3 yrs as the reference group, the RR decreased with increasing cumulative CB exposure. The two higher exposure categories 20 to 40, and more than 40 mg/m3 yrs indicated statistically significant decreased lung cancer mortality risks. The inverse exposure risk relationship was not statistically significant (RR = 0.79; 95% CI: 0.68–0.91) per 10 units.

Morfeld et al17 revised the JEM of the German cohort with some changes based on experts’ opinion in respect to an assumption of a stepwise decline of exposure, and a further elevation of carbon black scores in past exposures. Missing job histories in Wellmann et al15 were completed in the re-analysis in Morfeld et al.17 Four different approaches for exposure assessment were then compared. After adjusting for birth date and age at hire, the hazard ratio for time-dependent cumulative carbon black exposure was non-significantly reduced in the full cohort and in the inception cohort.17

Meta-Regression Analysis

We combined the data of lung cancer mortality risk estimates and cumulative exposure to carbon black from three cohorts of CB production workers. Meta-regression is a useful approach to quantifying the likelihood and the magnitude of any relationship between exposure and disease; in this case, lung cancer. This method uses category-specific data instead of subject-level to assess the association.

From the three studies, we extracted study-specific RR or hazard ratio (HR) estimates for lung cancer mortality in association with different cumulative CB exposure levels relative to the lowest category of exposure for each study (refer to Table S1, We used the midpoint of the range of each exposure category as an estimate of the cumulative exposure for each RR. For the highest exposure category, we calculated the midpoint as 5/3 times the lower bound of the category, as proposed by the U.S. Environmental Protection Agency in 2008.21

To estimate a pooled exposure–response curve from summarized exposure–response data, a two-stage procedure is used. In the first stage, we estimate the exposure–response association between the adjusted log relative risks and the levels of a specific exposure in a particular study. In the second stage, we combine study-specific estimates using methods for a meta-regression analysis.22–24

A log linear model for random-effects exposure–response meta-regression was applied, which can be expressed as

  • lnRR = β(exposure) + ε, where
  • β = the common slope associated with CB exposure across studies,
  • ε = the random effect between the studies.

In the meta-regression models, the natural logarithm (ln) of each study RR was inversely weighted by its variance, and correlations among the category-specific RRs from each individual study are accounted for by estimating their covariance.24 To account for potential between-study heterogeneity, the regression models allowed for random study-specific exposure effects.

Sensitivity Analyses

The meta-regression was repeated in a series of sensitivity analyses. We conducted sensitivity analyses, including the results from Morfeld et al17 for the German cohort, in which the modified exposure assessment and different models of additional confounding factors are used. From the US cohorts, we included the results from the most recent follow-up.10 We also conducted sensitivity analyses based on those studies providing the estimates with lag time of 20 years.

All calculations were done with the R version 5.0 statistics package25 using the library “dosresmeta.”26 Stata Version 1327 was used for creating the figure. A value of 5% was applied throughout as statistical significance level.


Figure 1 provides the 13 extracted risk estimates from the three cohort studies.10,13,15 Classification of exposure was based on the midpoints of the categories of cumulative exposure, from 2.5 to 150 mg/m3 yrs.

Predicted exposure–response relationship based on a log-linear regression model using RR estimates from the three cohort studies of CB exposure and lung cancer mortality. CB, carbon black; RR, relative risk.

The combined slope estimate, that is, the parameter estimates for 10 mg/m3 yrs increase of CB was –0.0013, indicating a RR of 0.999 (95% CI: 0.868 to 1.127) for every 10 mg/m3 yrs increment of exposure, without statistical significance (Table 2). Figure 1 demonstrates that the lower limits of the slope estimates were always below unity (lnRR = 0) along the whole range of exposure.

Exposure–Response Estimates With RR for 10 mg/m3 yrs Increase in CB From Individual Studies With No Lag Time and the Primary Combined Estimates Based on a Log-Linear Model

For sensitivity analyses, combined slope estimates were generally consistent with the primary estimate. The meta-regression model included data from Morfeld et al,17 in which exposure assessment was based on a modified JEM, and additional confounding factors were considered, yielded a combined slope estimate of 0.003, indicating an RR of 1.003 (95% CI: 0.996 to 1.010), with “unlagged” exposure data (cf. Table 3). The sign of the point estimate was positive, while the result does not change qualitatively. Combining the results from data with 20 years lagged exposure; the slope estimate was close to null, without statistical significance (cf. Table 3).

Sensitivity Analysis of Exposure–Response Estimates With RR for 10 mg/m3 yrs Increase in CB from Individual Studies With 0 Or 5 Years Lag Time and the Primary Combined Estimates Based on a Log-Linear Model


Carbon black is primarily manufactured for tire and automotive rubber products. IARC classified CB as possibly carcinogenic to humans (Group 2B) based on the positive lung cancer findings in the rats, while epidemiologic studies of workers in CB production and in the rubber industry have not provided adequate evidence of increased risk of cancer.6 Furthermore, the IARC Working Group identified lack of human exposure–response analyses to be a knowledge gap, which was addressed also in the US cohort, which was conducted following the IARC evaluation. The present analysis investigates the presence of an exposure–response relationship through the use of meta-regression analysis.

Consideration on Causality

Since 1965, the Bradford Hill Criteria,7 strength of association, consistency, specificity, temporality, biological gradient, plausibility, coherence, experiment, and analogy, remain the most important concepts in health research for inferring causation when association is observed. Hill wrote that “if a dose response is seen, it is more likely that the association is causal.” According to the traditional interpretation of biological gradient, the presence of an exposure–response relationship supports the causal association between an exposure and an effect.28,29

We used the results from three large cohort studies of CB production workers, combining the risk estimates of the exposure categories to evaluate an exposure–response relationship by meta-regression analysis. The approach to evaluating an exposure–response relationship was based on linear models, which yielded the slope estimates implying an increased risk with every unit increment of exposure. Combining the data from Wellmann et al15 from the German cohort with the data from the UK13 and US cohort,10 a slightly negative trend was suggested, with a slope estimate of –0.0017. For sensitivity analyses, we used data from Morfeld et al17 for the German cohort, which addressed confounding and re-evaluated exposure with a modified JEM. With no lagged exposure, a minor positive slope estimate was found, though without statistical significance. With a 20-years lagging time for exposure, a slope close to null was found. Overall, no exposure–response relationship could be concluded.

The findings from the recent meta-analysis for titanium dioxide (TiO2),30 another substance considered a PSLT are consistent with our findings. Le et al30 initially conducted a qualitative assessment, in which all mortality cohort studies were reviewed, and a summarized standardized mortality ratio (SSMR) was derived. No elevated risks for mortality due to lung cancer, all causes, all cancers, or non-malignant respiratory disease were observed for production workers in the TiO2 industry. For a quantitative assessment, in which the exposure–response relationship was investigated, an inverse relationship was suggested.

Consideration on Mechanism and Biological Plausibility

Dust overloading of lung is the core conceptual explanation1 for lung cancer that develops in some laboratory rats from exposure to PSLTs. Experimental studies in rats, which represent a particularly sensitive species to inhaled particles, suggest a unique hazard in developing lung tumors under particle overload conditions.2,31 The relevance of the rat lung response to particle overload for human risk assessment has been extensively discussed. More recently, an update of the evidence on the relationship between the inhalation of PSLTs, including CB, has concluded that the lung tumors seen in rats from overload exposure in rats are not predictive of such a hazard for humans. This proposes that the interstitialization of inhaled particles, seen in humans (which is the major way particles are cleared from the alveolar region) is very different from the predominantly alveolar accumulation of particles in the rat and subsequent inflammatory cascade; finally leading to alveogenic tumors.4

The above supports the conceptual adverse outcome pathway (AOP) model of lung overload sequelae in rats following chronic exposure to poorly soluble particulates exposures,31 impaired alveolar clearance is responsible for the successive chronic inflammation, anti-oxidative stress, cell proliferation, and eventually lung tumors. Studies in rats showed a breakdown of alveolar clearance at relatively low concentrations due to this very marked toxicokinetic difference of handling inhaled PSLTs; interstitialization accumulation in humans versus alveolar accumulation in rats.4 Dosimetry studies in coal miners with substantially high exposure of coalmine dust were conducted to test a three-compartment model of lung burden, that is, deposition, competing processes of clearance, translocation, or irreversible sequestration. The model with no clearance breakdown fit the lung burden best, both in the US data32,33 and in the UK data.34 The divergent clearance pattern observed in rats and coal miners enforces the doubt about the relevance of the rat lung response to dust for human hazard and risk assessment, for either hazard assessment of carcinogenicity or for a derivation of occupational exposure limit values. More generally, despite a high degree of genomic and physiologic similarities between rats and humans, functional differences in metabolism have been described,35–37 which could influence whether or not a substance may induce toxicity or activate a carcinogenic pathway.38,39

Methodological Considerations

Meta-regression analysis is a useful quantitative approach that employs published categorical data to study an exposure–response relationship. This approach is particularly advantageous if the power of the primary studies is limited and it helps to understand and explain inconsistencies. One of the limitations of this approach is that the categories of exposure from different studies cannot be redefined retrospectively. Log-linear models were used, in which the trend is forced through the origin, that is, with no intercept. The method of Greenland and Longnecker23 was used to estimate the variance, assuming a correlation between the risk estimates from each single study. Orsini et al40 evaluated the approaches of meta-regression and pooled analysis using the original data for studying an exposure–response relationship between alcohol consumption and colorectal and lung cancer risks. Minor differences were found between the results using the two approaches.

Exposure categories included in the regression model were limited by those in the three cohort studies, resulting in inherent uncertainty in the results. Heterogeneity between the studies was observable. The number of studies with sufficient exposure categories plays an important role for a valid estimation, which is a major limitation of the present analysis. We calculated this measure according to an approximate approach,41 assuming a constant in age-at-risk instead of using complicated life-table computations.


Neither the individual cohort mortality studies nor the present meta-regression analysis provides a positive exposure–response relationship between exposure to carbon black and lung cancer risk. Therefore, a causative link of carbon black exposure and cancer risk in humans is not substantiated. This conclusion based on the strong epidemiological findings from a large number (upwards of 9000 in total) of workers exposed to CB over many decades needs to be seen alongside the recent discussions described above. The evidences in humans cast doubt on the predictability of the lung tumors seen in rats under conditions of lung overload from the inhalation of PSLTs, in particular CB.


1. Morrow PE. Possible mechanisms to explain dust overloading of the lungs. Fundam Appl Toxico 1988; 10:369–384.
2. Warheit DB, Kreiling R, Levy LS. Relevance of the rat lung tumor response to particle overload for human risk assessment-Update and interpretation of new data since ILS. Toxicol 2016; 374:42–59.
3. Chaudhuri I, Fruijtier-Polloth C, Ngiewih Y, Levy L. Evaluating the evidence on genotoxicity and reproductive toxicity of carbon black: a critical review. Crit Rev Toxicol 2018; 48:143–169.
4. Bevan RJ, Kreiling R, Levy LS, Warheit DB. Toxicity testing of poorly soluble particles, lung overload and lung cancer. Regul Toxicol Pharmacol 2018; 100:80–91.
5. Baan RA. Carcinogenic hazards from inhaled carbon black, titanium dioxide, and talc not containing asbestos or asbestiform fibers: recent evaluations by an IARC Monographs Working Group. Inhal Toxicol 2007; 19:213–228.
6. Baan R, Straif K, Grosse Y, Secretan B, El Ghissassi F, Cogliano V. Carcinogenicity of carbon black, titanium dioxide, and talc. Lancet Oncol 2006; 7:295–296.
7. Hill AB. The environment and disease: association or causation? Proc R Soc Med 1965; 58:295–300.
8. Rothman KJ, Greenland S. Modern Epidemiology. Philadelphia: Lippincott - Raven; 1998.
9. Sorahan T, Harrington JM. A “lugged” analysis of lung cancer risks in UK carbon black production workers, 1951–2004. Am J Ind Med 2007; 50:555–564.
10. Dell LD, Gallagher AE, Crawford L, Jones RM, Mundt KA. Cohort study of carbon black exposure and risk of malignant and nonmalignant respiratory disease mortality in the US carbon black industry. J Occup Environ Med 2015; 57:984–997.
11. McCunney RJ, Morfeld P, Levy L, Muranko H. Carbon black. Environ Health Perspect 2011; 119:A332–A333.
12. Dell LD, Mundt KA, Luippold RS, et al. A cohort mortality study of employees in the U.S. carbon black industry. J Occup Environ Med 2006; 48:1219–1229.
13. Sorahan T, Hamilton L, van Tongeren M, Gardiner K, Harrington JM. A cohort mortality study of U.K. carbon black workers, 1951–1996. Am J Ind Med 2001; 39:158–170.
14. Morfeld P, McCunney RJ. Carbon black and lung cancer: testing a new exposure metric in a German cohort. Am J Ind Med 2007; 50:565–567.
15. Wellmann J, Weiland SK, Neiteler G, Klein G, Straif K. Cancer mortality in German carbon black workers 1976–1998. Occup Environ Med 2006; 63:513–521.
16. Morfeld P, Büchte SF, McCunney RJ, Piekarski C. Lung cancer mortality and carbon black exposure: uncertainties of SMR analyses in a cohort study at a German carbon black production plant. J Occup Environ Med 2006; 48:1253–1264.
17. Morfeld P, Büchte SF, Wellmann J, McCunney RJ, Piekarski C. Lung cancer mortality and carbon black exposure: cox regression analysis of a cohort from a German carbon black production plant. J Occup Environ Med 2006; 48:1230–1241.
18. Büchte SF, Morfeld P, Wellmann J, Bolm-Audorff U, McCunney RJ, Piekarski C. Lung cancer mortality and carbon black exposure: a nested case-control study at a German carbon black production plant. J Occup Environ Med 2006; 48:1242–1252.
19. Axelson O, Steenland K. Indirect methods of assessing the effects of tobacco use in occupational studies. Am J Ind Med 1988; 13:105–118.
20. Kerr SM, Muranko HJ, Vincent JH. Personal sampling for inhalable aerosol exposures of carbon black manufacturing industry workers. Appl Occup Environ Hyg 2002; 17:681–692.
21. Lenters V, Vermeulen R, Dogger S, et al. A meta-analysis of asbestos and lung cancer: is better quality exposure assessment associated with steeper slopes of the exposure-response relationships? Environ Health Perspect 2011; 119:1547–1555.
22. Orsini N, Bellocco R, Greenland S. Generalized least squares for trend estimation of summarized dose-response data. Stata J 2006; 6:40–57.
23. Crippa A, Orsini N. Dose-response meta-analysis of differences in means. BMC Med Res Methodol 2016; 16:91.
24. Greenland S, Longnecker MP. Methods for trend estimation from summarized dose-response data, with applications to meta-analysis. Am J Epidemiol 1992; 135:1301–1309.
25. R Core Team. R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing; 2018. Available at: Accessed March 20, 2018.
26. Crippa A, Orsini N. Multivariate dose-response meta-analysis: the dosresmeta R Package. J Stat Softw 2016; 72:15.
27. StataCorp. Release 13. Statistical Software. College Station, TX: StataCorp LP; 2014.
28. Rothman KJ, Greenland S. Causation and causal inference in epidemiology. Am J Public Health 2005; 95:S144–S150.
29. Swaen G, van Amelsvoort L. A weight of evidence approach to causal inference. J Clin Epidemiol 2009; 62:270–277.
30. Le HQ, Tomenson JA, Warheit DB, Fryzek JP, Golden AP, Ellis ED. A review and meta-analysis of occupational titanium dioxide exposure and lung cancer mortality. J Occup Environ Med 2018; 60:e356–e367.
31. ECETOC. Poorly soluble particles /lung overload; 2013. Brussels, Belgium: European Center for Toxicology and Chemicals. Available at: Accessed November 17, 2014.
32. Kuempel ED, O’Flaherty EJ, Stayner LT, Smith RJ, Green FH, Vallyathan V. A biomathematical model of particle clearance and retention in the lungs of coal miners. I. Model development. Regul Toxicol Pharmacol 2001; 34:69–87.
33. Kuempel ED, Tran C-L, Smith RJ, Bailer AJ. A biomathematical model of particle clearance and retention in the lungs of coal miners. II. Evaluation of variability and uncertainty. Regul Toxicol Pharmacol 2001; 34:88–101.
34. Tran CL, Buchanan D. Development of a biomathematical lung model to describe the exposure-dose relationship for inhaled dust among U.K. coal miners; 2000. Edinburgh, U.K.: Institute of Occupational Medicine. Available at: Accessed November 17, 2014.
35. Shannahan JH, Schladweiler MCJ, Richards JH, Ledbetter AD, Ghio AJ, Kodavanti UP. Pulmonary oxidative stress, inflammation, and dysregulated iron homeostasis in rat models of cardiovascular disease. J Toxicol Environ Health A 2010; 73:641–656.
36. Luo L, Schomaker S, Houle C, Aubrecht J, Colangelo JL. Evaluation of serum bile acid profiles as biomarkers of liver injury in rodents. Toxicol Sci 2014; 137:12–25.
37. Bando K, Kunimatsu T, Sakai J, et al. GC-MS-based metabolomics reveals mechanism of action for hydrazine induced hepatotoxicity in rats. J Appl Toxicol 2011; 31:524–535.
38. Bracken MB. Why animal studies are often poor predictors of human reactions to exposure. J R Soc Med 2009; 102:120–122.
39. Sundstrom LE. Thinking inside the box. EMBO Rep 2007; 8:S40–S43.
40. Orsini N, Li R, Wolk A, Khudyakov P, Spiegelman D. Meta-analysis for linear and nonlinear dose-response relations: examples, an evaluation of approximations, and software. Am J Epidemiol 2012; 175:66–73.
41. Vaeth M, Pierce DA. Calculating excess lifetime risk in relative risk models. Environ Health Perspect 1990; 87:83–94.

carbon black; exposure–response relationship; lung cancer; lung particle overload; meta-regression; poorly soluble low toxic particulates; retrospective cohort study; systematic review

Supplemental Digital Content

Copyright © 2019 by the American College of Occupational and Environmental Medicine