Diesel Exhaust Exposure and the Risk of Lung Cancer—A Review of the Epidemiological Evidence

To critically evaluate the association between diesel exhaust (DE) exposure and the risk of lung cancer, we conducted a systematic review of published epidemiological evidences. To comprehensively identify original studies on the association between DE exposure and the risk of lung cancer, literature searches were performed in literature databases for the period between 1970 and 2013, including bibliographies and cross-referencing. In total, 42 cohort studies and 32 case-control studies were identified in which the association between DE exposures and lung cancer was examined. In general, previous studies suffer from a series of methodological limitations, including design, exposure assessment methods and statistical analysis used. A lack of objective exposure information appears to be the main problem in interpreting epidemiological evidence. To facilitate the interpretation and comparison of previous studies, a job-exposure matrix (JEM) of DE exposures was created based on around 4,000 historical industrial measurements. The values from the JEM were considered during interpretation and comparison of previous studies. Overall, neither cohort nor case-control studies indicate a clear exposure-response relationship between DE exposure and lung cancer. Epidemiological studies published to date do not allow a valid quantification of the association between DE and lung cancer.


Introduction
Diesel engines have been widely used for decades in various industrial sectors such as underground mining, construction, public transportation, ship loading in docks, agriculture, operation of machines and fire-fighting. Diesel exhaust (DE) emissions are composed of gases and a particulate phase containing thousands of chemicals. Their composition varies according to engine type, speed, air/fuel ratio, temperature, fuel and many other factors [1]. DE contains large quantities of carbonaceous particulates to which polynuclear aromatic hydrocarbons and other heterocyclic compounds are adsorbed. The latter are known to be mutagenic and carcinogenic in both animals and humans [2].
In June 2012, a working group of the International Agency for Research on Cancer concluded that there was sufficient evidence for the carcinogenicity of DE in humans [3]. However, these findings appear to be based upon selected epidemiological studies with certain important methodological limitations, particularly in the assessment of confounding effects and the assessment of DE exposures [4]. In order to evaluate critically the epidemiological evidence for the association between DE exposure and the risk of lung cancer, we conducted a systematic review of the international literature.

Literature Search
For comprehensive identification of original studies on the association between DE exposure and the incidence or mortality of lung cancer, searches were performed for the period between 1970 and 2013 in the following databases: MEDLINE, EMBASE, NIOSHTIC, CISDOC, Cochrane and the databases in TOXNET. Multipart search strategies were applied using "diesel" combined with the following search terms: "lung cancer", "lung neoplasm?", "work?", "occupation?", "epidemiol?", "case control", "cohort" or "risk". Bibliographies and cross-referencing including comparison with reviews were additionally used for literature searches.

Quantification of DE Exposures Using MEGA-JEM
Previous studies on the effect of DE exposure focus mainly on risk estimation for jobs supposed to involve high and prolonged exposure to DE, such as those of professional drivers, railroad workers, heavy equipment operators, and so on. Although a large number of studies have been published, few are able to provide any information on the level of DE exposures in these jobs.
To allow an objective impression to be gained of the level of DE exposures in commonly exposed jobs, we created a job-exposure matrix for DE exposures based upon historical industrial hygiene data from the MEGA (Measurement data relating to workplace exposure to hazardous substances) database (see Table 1). Table 1. DE exposures in common exposed jobs in Germany (MEGA-JEM).

Exposure as Elemental Carbon (mg/m 3 ) (2)
Before 1990 (3) 1990-1993 (4) After 1993 ( (1) Exposure data from MEGA are related to the listed job titles; (2) Exposure data are calculated from exposure data of total carbon (TC) using the known task related mean relation between EC and TC; (3) 90% percentile of the exposure data for the period 1990-1993; (4) 50% percentile of exposure data.
The MEGA database is a large industrial hygiene database forming part of the Measurement System for Exposure Assessment of the German Social Accident Insurance Institutions (MGU). The database was established in 1972 and contains more than 2.4 million historical measurements of around 1,380 industrial chemical and biological agents. In total, around 4,000 historical measurements of DE exposures were entered in the database for the period from 1990 to 2000.
In this review, MEGA-JEM was used directly to estimate the exposure levels of jobs given in the results of previous published studies. If information on exposure duration is available, cumulative doses of DE exposure were quantified as "exposure level (MEGA-JEM)  median exposure duration". Effect estimates published in previous studies were summarized in a scatter plot. Based on these values, exposure-response relationship between DE-exposure and lung cancer and their 95% CI were quantified by a linear regression analysis with the software package SigmaPlot 12.0. The inclusion of MEGA-JEM in this review will permit a direct comparison of previously published epidemiological evidence.

Results
In total, 42 cohort studies and 32 case-control studies were identified in which the association between DE exposure and lung cancer was examined.

Cohort Studies
In general, historical industrial hygiene data on DE exposure (based on the measurement of elemental carbon) were not available in published cohort studies. Therefore, exposure assessment was limited only to job titles in 37 of the 42 identified cohort studies. Five studies allow a quantitative assessment of DE exposure based on industrial hygiene measurement. Three studies [5][6][7] quantified the DE exposures based upon historical surrogate measurements of nitrogen dioxide, while two other studies were based either on current industrial hygiene measurement of total carbon [8] or on historical surrogate measurements of CO [9].
The effect of DE exposure upon lung cancer was evaluated with the focus primarily on the following job categories: professional drivers, highway maintenance workers, railroad workers, mechanics, workers at gasoline filling stations, heavy equipment operators, dock workers and miners (see Table 2).
The effect of DE exposure was evaluated in most studies by comparison of the lung cancer risk among workers in highly exposed jobs with an external population by use of the standardized mortality ratio (SMR), standardized incidence ratio (SIR) or proportional mortality ratio (PMR). Internal comparison was carried out in nine cohort studies [2,[5][6][7][8][9][10][11][12]. All studies have large sample sizes. The possible confounding effect of smoking was adjusted in most of these studies (except the study by Bergdahl [7] and the study by Attfield [9]). Boffetta et al. reported in an earlier study that railroad workers, heavy equipment operators, miners and truck drivers have higher mortality both for all causes and for lung cancer when compared with workers without exposure to DE [2]. Similar findings were also reported by Garshick et al. [11,13] and Larkin et al. [12]. However, a reanalysis of the US railroad study (originally published by Garshick [13]) indicates that the effect of DE exposure published in the early study appears to be unstable. The estimates of the effect vary strongly depending upon how the exposure was assessed and how confounders were considered in the analysis [14]. If the confounders were considered in a different manner, an exposure-response relationship between DE exposure and lung cancer is no longer observed. This early methodological disagreement in the US railroad study gives an example about how difficult previous evidence can be properly interpreted. This problem seems to be solved in a later published extended follow-up of this cohort [10]. Therefore, only the latest publication of this study [10] was considered in this review.        Among the three cohort studies employing historical measurements of nitro compounds as surrogate indicators of DE exposures [5][6][7], a weak association (OR = 1.23, 95% CI: 1.0-1.5) between DE exposure and lung cancer can be demonstrated only in the study by Johnston et al. [5]. In the other two cohort studies [6][7], no relationship between DE exposure and lung cancer could be observed. Main strengths of these studies are large sample size, quantitative exposure estimations and consideration of smoking as a confounder in the analysis. However, some important limitations make the interpretation of these studies difficult. These include the population based setting and incomplete assessment of work history in the study by Guo et al. [6], and the missing consideration of occupational cofounders (such as respirable silica) in the analysis of the other two mining cohorts [5,7]. Since it is generally questionable if nitro compounds can be used as surrogate to measure DE exposures, the evidences provided by these studies are rather limited.
The German potash miner study [8] is the first study which quantified DE exposures by measuring carbon compounds. This study has a sample size of 5,862 workers with a follow-up duration of 30 years. After adjustment for age and smoking, the study demonstrates a clear exposure-response relationship between DE exposures and lung cancer mortality. However, in a recent reanalysis of this study, Möhner et al. [50] pointed out that a part of cohort members in this study were previously employed as uranium miners. These workers may have had a high exposure to respirable silica and radon daughters in their work history. If these subjects were excluded from the data analysis, an exposure-response relationship between DE exposure and lung cancer can no longer be observed. This finding leads to a further reanalysis of this cohort in which employment in external mines or industries was controlled [51]. The final results give no evidence of an association between DE exposure and lung cancer. Strengths of this study are large sample size and extensive control of both occupational and non-occupational confounders in the analysis [50,51]. Historical DE exposures were estimated based on the current industrial hygiene measurements.
In contrast to the German potash miner study, the US Miners study demonstrates an extremely high effect of DE exposure (up to 5-fold), although the initial analysis of this cohort did not reveal a clear relationship between DE exposure and lung cancer [9]. Main strengths of this study are large sample size (more than 12,000 workers with an average follow-up duration of about 23 years), quantitative assessment of DE exposures by measuring carbon compounds and the adjustment of smoking as a confounder in a nested case-control analysis [52]. However, some findings reported in this study need more clarification. For example, it is unclear why "surface only workers" (SMR = 1.33) have the same risk as the "ever underground workers" (SMR = 1.21) in the initial analysis, although DE exposure among "underground workers" was about 500 times higher than "surface workers". This finding seems to be contradictory with the final reported high effect of DE exposures. Possible limitations of this study have been discussed by Morfeld [53] and Gamble et al. [54] regarding the completeness of follow-up, essential exposure misclassification, inadequate control of occupational confounder and improper statistical methods used.
In order to compare previously published cohort studies objectively and to allow an overall judgement of the association between DE exposure and lung cancer, we calculated the historical DE exposure in previous studies by means of the MEGA-JEM. Due to limited exposure information (limited information on job title or exposure duration), cumulative doses of DE exposures are only available for six cohort studies (Table S1, Supplementary Information). The results of these studies are summarized in Figure 1. Overall, no exposure-response relationship between DE exposure and lung cancer can be demonstrated.

Case-Control Studies
In total, 25 population or hospital-based case-control studies, six nested case-control studies and 1 industry-based case-referent study were identified (see Table 3). Most of these studies have large sample sizes and adjustment of the possible confounding effect of smoking in the analysis.
Assessments of DE exposures were limited in most of these studies on job title (with different definitions) or dichotomous categorization (ever/never exposed). Quantitative or semi-quantitative assessment of DE exposure was carried out in only six studies, with use of different exposure assessment methods [51,52,[55][56][57][58]. Overall, a consistently increased risk of lung cancer was reported for jobs supposed to have high DE exposures. An exposure-response relationship was also presented in most studies. However, due to the different exposure assessment methods used, direct comparison between these studies is difficult.

Discussion
The possible association between DE and lung cancer, which constitutes an important occupational health question, has long been the subject of debate. Interpretation of epidemiological evidence faces a series of methodological challenges.
Lack of exposure information appears to be the major problem in interpreting human epidemiological data. The low volume of data documenting past exposures is due to the fact that no standardized method of measuring diesel fumes existed before the late 1980s. From an industrial hygiene prospective, it was not clear which substance to measure during assessment of occupational exposure to DE. Diesel fumes are composed of gases (nitrogen oxides, carbon monoxide) and various hydrocarbons bound to a carbon core. Early studies have reported levels of particulate, but such particulates are generated by many sources other than diesel engines [84]. Attention has also been focused on polycyclic aromatic hydrocarbons (PAHs) and nitro-PAHs in the exhaust. However, there are no standard methods of measuring PAHs, and PAHs are also emitted by sources other than diesel engines [84].
In the late 1980s, a standardized method of measuring diesel fumes by quantifying elemental carbon was introduced. Since then, systematic industrial hygiene measurements have been begun in some industrialized countries. However, a long time is needed for sufficient measurement data to be collected for use in epidemiological research. Most of the epidemiological studies published to date therefore provide no fundamental basis for an objective assessment of DE exposures.
In this review, we identified only two recent studies containing industrial hygiene measurement data for carbon compounds. In all remaining studies, the exposure assessments are based on expert judgements. A given job may be classified as having high exposure by one expert, but low by another [14,85]. Previous studies indicate that the differences in expert opinion have a strong influence on the estimated exposure-response relationship between DE exposure and lung cancer [14,85]. This problem makes the interpretation and comparison of previously published epidemiological studies difficult.
To facilitate an objective comparison of previously published epidemiological studies, we created a JEM for DE exposures based upon a large number of standardized industrial hygiene measurements conducted since the late 1980s. Three calendar periods were considered in the JEM, since most of the technical changes occurred during the period between 1990 and 1993. The values in the MEGA-JEM were considered in the interpretation of the epidemiological studies published to date. We found that conflicting findings were reported not only between studies, but also within studies. It is very common for jobs associated with higher exposure (according to the exposure value given in Table 1) to be reported as having lower risks than jobs with lower exposure, even within the same study. Since many studies indicated only job titles without detailed information on the exposure duration, direct comparison of the effect estimates was limited. To solve this problem, we summarized only studies with complete exposure information (both job title and exposure duration) and presented the results in Figures 1 and 2. Overall, neither cohort nor case-control-studies show exposure-response relationship between DE exposure and lung cancer.
Caution should be exercised during interpretation of these studies. Previous cohort studies often compare workers in certain job categories with a standard population without adjustment for important confounders, while case-control studies generally employ a population-based design which is less suitable for detecting weak associations related to DE exposures. For some of the early epidemiological studies, latency may also be too short to attribute lung cancer to DE exposure. The use of different definitions of job titles in the analysis (longest job, ever employed jobs, census job, job in death certificates or at the time of medical examination, etc.) and the related cross-contamination with current and previous occupational history may also have a strong influence on the estimated effects. This problem was clearly demonstrated in the cohort of German potash miners, for which the study results were strongly dependent upon whether previous work history in the uranium mining industry was considered in the analysis [50]. The JEM-approach used in this review has also some weaknesses. First, the exposure duration in most studies is given only in categories. Therefore, the use of the center of such category gave only a very crude estimate for the mean or the median of exposure duration. Furthermore, the JEM used in this review is based on German industrial hygiene measurement data. The data collected in Germany may not be representative for all industrialized countries. Since diesel engines were introduced into the workplace at variable rates over time by industry and country, the use of MEGA-JEM in this review may lead to some uncertainty in the exposure assessment. However, despite the exposure-assessment methods used (expert judgement, measuring nitro compounds, measuring carbon compound, MEGA-JEM) no consistent findings of an association between DE exposures and lung cancer can be demonstrated.

Conclusions
Overall, the previously published epidemiological evidence did not clearly support an exposure-response relationship between DE exposure and lung cancer. In fact, the limited exposure information available in previous studies does not even allow a valid estimation of an association between DE exposure and lung cancer. However, such an association cannot be ruled out. Causality of weak association is often difficult to establish, since it is susceptible to all forms of possible design bias. Due to the limited epidemiological evidence to date, well designed studies in an industrial context are still needed, for which detailed exposure assessment methods and adequate control for confounders are recommended.

Author Contributions
All authors participate in drafting the article or revising it critically for important intellectual content; and give final approval of the version to be submitted and revised.