The widespread use of diesel engines has long raised concerns regarding potential health effects from exposure to diesel exhaust (DE), especially with respect to lung cancer. More than 35 cohort and case–control studies of lung cancer incidence and mortality and DE have been published to date. A majority of studies have found an increased risk of lung cancer with surrogate measures of exposure (ie, job or tenure). However, few have measured workplace DE exposures, and fewer have used them directly or indirectly in their analysis [eg, the US railroad workers cohort (1–3), the US Teamster trucker study (4), German potash miner study (5), and a recent study of truckers (6)]. Criticisms of existing studies include small study size, lack of reliable historical quantitative exposure data, inadequate latency period, and potential confounding (7–10). Determinations by the International Agency for Research on Cancer (11) and the National Institute for Occupational Safety and Health (12), and reviews, meta-analyses, and one very large pooled analysis (13–16) have concluded that there is evidence that lung cancer is related to DE exposure, but other reviews have disagreed (9,17–20).

To provide additional information on lung cancer and other health outcomes possibly associated with DE exposure, and to address gaps and limitations in prior investigations, we conducted an epidemiological investigation of DE-exposed non-metal (ie, mineral) miners. The Diesel Exhaust in Miners Study (DEMS) consisted of a cohort mortality study (presented here), a nested case–control study of lung cancer mortality (21), and current and retrospective exposure assessments (22–26). The objectives of the cohort study were to evaluate total and cause-specific mortality and to assess lung cancer mortality in relation to quantitative estimates of DE exposure. The study was specifically designed to have adequate statistical power, to use extensive current and historical quantitative DE exposure data, and to investigate an environment having low levels of potentially confounding workplace lung carcinogens (27). For these reasons, and because of the wide range in DE levels seen in the study mines, this study largely addressed the limitations of previous investigations.

About 35% (12%–50% across facilities) of the person-years were accrued on the surface, either exclusively or before moving underground (Table 1). The average year of first exposure to DE in the study cohort varied from 1967 to 1976 across facilities with an overall mean of 1971 (the range for individuals was from 1947 to 1996). Mean tenure underground overall was 8.0 years, with a range across facilities from 6.6 to 9.1 years (Table 1). The range across individuals was from close to 0 to 44 years. A total of 2220 (18%) individuals were deceased at the study mortality cutoff date.

Average intensity of REC exposure, based on each individual’s full work history at the study facilities, was 87.0 μg/m³ for all workers, 128.2 μg/m³ for ever-underground workers, and 1.7 μg/m³ for surface-only workers (Table 2). These and the other exposure data in Table 2 pertain to an individual's complete work history; surface work exposures were truncated at the time the individual first went underground, whereas ever-underground exposures were computed from first time underground. There were substantial differences across facilities in levels of REC exposure underground (average REC intensity range: 78–216 μg/m³). In general, estimated REC levels rose from dieselization, typically peaking in the late 1970s or early 1980s and then falling to about half the peak levels by end of study (25). The exception was facility A, at which estimated underground REC levels rose continuously from the mid-1960s. Mean levels of all potential confounders, including silica, asbestos, non-DE PAHs, radon, and respirable dust, were low and, in contrast to REC, typically differed little between surface and underground work (Table 2).

The DE–lung cancer associations seen separately for the ever-underground and surface-only workers were not readily apparent in the complete cohort without adjustment for worker location (ever underground vs surface only), which may be attributable to the generally higher lung cancer mortality and relatively low exposures among the surface workers (Tables 3, ​,5,5, and ​and6).6). In the complete cohort, most of the surface-only workers fell into the reference category, leading to a concomitant reduction in the HRs at higher exposures. After adjustment for worker location, the exposure–response findings were similar to those reported separately for the two worker location groups. (See the companion case–control study (21) for further information on this observation, including the observation that the risk of lung cancer from cigarette smoking differed by worker location.)

Some differences possibly remain between the two worker locations. As noted earlier, a formal test of significance of the ever-underground vs surface-only HR slopes indicated a statistically significant difference between the two location groups at the 5% level for log average REC intensity (1.26 vs 2.60). The surface-only HR for untransformed cumulative REC exposure was also greater than that for ever-underground workers, although the difference was statistically nonsignificant. This result may be attributable to aging and transformation of DE by sunlight, ozone, and other factors. During these processes, some exhaust components become more toxic (31,32). For example, secondary and unique nitro-PAHs are produced by atmospheric reactions of DE (33).

We found that the HRs were generally greater after exclusion of workers with shorter tenures, which was true for both ever-underground and surface-only workers. Shorter-term employees include transient workers, who may take lower-paid more hazardous jobs, have less access to health care, smoke more, and generally have more lifestyle and occupational risk factors for disease than workers who work for extended periods with the same employer. Such factors have been shown to affect mortality (34,35). Younger short-term workers may go on to receive hazardous exposures elsewhere after leaving employment in the study facilities, whereas workers who enter the study at older ages may have had extensive prior experience in hazardous jobs. Information pertinent to prior or later workplace exposures was not available for most individuals in the cohort study, nor sufficiently reliable, and therefore, we could not take it into account formally in the cohort analysis. Lifetime occupational histories were obtained in the companion case–control study (21), in which a relationship between long-term employment in high-risk occupations for lung cancer was associated with increased lung cancer mortality.

We explored the short-term worker effect further by adding terms for tenure and a REC by tenure interaction to the proportional hazard models for ever-underground workers instead of excluding workers with less than 5 years tenure (data not shown). This analysis led to virtually the same findings, showing that those with the longer tenure had lower absolute risk but greater REC exposure–response slopes compared with short-term workers. In addition, we examined the effect of limiting the less than 5-year tenure worker exclusion by age of starting work at the study facilities. This analysis showed that individuals with short tenures who started work after age 40 had the largest REC exposure–response HRs. Because older workers would have had the greatest potential for prior work in other hazardous jobs, the findings suggest that previous exposures for some individuals may have affected the findings in this cohort and that exclusion of short-term workers may be a partial surrogate for adjustment for prior hazardous exposures. As noted above, lifetime work histories were acquired and used in the analyses for the companion case–control study.

It should be noted, however, that it was not necessary to restrict the analyses on tenure for statistically significant exposure–response findings to arise. For example, the cumulative REC exposure HR for ever-underground workers including all tenures and restricted to less than 1280 μg/m³-y was 2.79 (P < .001). Moreover, statistically significant findings were found for other models in which tenure was not restricted (Supplementary Table 7, available online). Furthermore, in proportional hazard models using time since entry into cohort and adjusting for age instead of using age as the underlying time variable, similar findings as shown in Table 4 were found without any tenure restriction (Supplementary Table 17, available online).

Among the ever-underground workers, the HRs rose fairly consistently over the REC exposure range but declined or reached a plateau at the higher levels of exposure [Table 4 and case–control findings (21)]. Given the clear evidence of increasing risk with increasing exposure for the lower part of the exposure range, we undertook analyses omitting the highest exposures to provide risk estimates pertinent to the lower range. However, given that the findings (Table 4) could also be interpreted as showing a plateau similar to that seen for average REC intensity (ie, apart from the HR of 5.01, the rest fall in the range 2.21–2.69 from 80 μg/m³-y and higher), we also fitted log cumulative REC exposure for the full exposure range. However, the log transformation model fitted the data less well than the restricted exposure model. Declines and plateaus have frequently been reported for other occupational exposures, and a variety of plausible explanations have been proposed, for example, misclassification at high exposures, worker selection effects, and enzyme saturation (36).

The overall findings were essentially unchanged after inclusion of silica, asbestos, non-diesel PAHS, and respirable dust in the models. Although radon was associated with lung cancer risk, the observed effect was driven by a small subset of older workers hired before 1947 at facility A. At that facility, the estimated radon exposure levels were about half of those at four other facilities (Table 2). The hazard ratio for cumulative radon exposure was 6.2, considerably greater than the 1.1 predicted from radon-exposed underground miners for the same exposure levels (37). We conclude that the radon finding is anomalous and probably arose from chance or other unknown factors affecting early older workers at that one facility.

Few studies of lung cancer and DE have been conducted in mining operations. A previous mortality study of US potash workers, based on mine tenure, did not find an excess of malignant neoplasms of the respiratory system (38). However, as noted by the authors, the latency period was inadequate to detect elevations in lung cancer mortality associated with DE (at study closure, only two mines used diesel engines, with follow-up of 10 and 18 years, respectively). In a study of German potash miners (5), a statistically significantly higher lung cancer relative risk (2.47, 95% CI = 1.02 to 6.02) was found for miners with 2.04 to less than 2.73 mg/m³-y total carbon (TC, which includes both EC and organic carbon) exposure, compared with the reference group (<1.29 mg/m³-y TC). A positive trend in increasing mortality with increasing TC exposure was observed (P = .09).

Apart from those mentioned above, there are no other existing cohort mortality studies of DE and lung cancer in miners, and none of the cohort studies in other industries used quantitative measurements of exposure directly in the epidemiological modeling. A recent study of truckers (6) reported higher hazard ratios (approximately 2) among long-haul drivers and pick-up/delivery drivers with 20 years of work and with adjustment for smoking. Spline analyses also showed evidence of increasing lung cancer risk with increasing tenure in truck driving jobs, with estimated hazard ratios ranging from 2.5 to 4.0 after 40 years exposure. These findings are fairly similar to those from an earlier study of truckers (39), in which an odds ratio of 1.89 was reported for diesel truck drivers with 35 or more years tenure. These hazard ratios, possibly applicable to reported arithmetic mean exposures of 3–5 μg/m³ for highway and local drivers (40), are similar to what we found for surface-only workers at equivalent levels of exposure (HR of 2.33 at the average REC level for the 2 to <4 μg/m³ category).

Among causes of death previously reported to be associated with DE exposure, few were higher in this standardized mortality ratio analysis. Standardized mortality ratios for nonmalignant respiratory disease (apart from pneumoconiosis), heart disease, and all-cause mortality were either less than or close to expected values. A statistically nonsignificant increase in the standardized mortality ratio for bladder cancer was seen for surface-only workers, but there was a statistically nonsignificant deficit for the ever-underground workers (who had much higher DE exposures). A similar pattern was observed for leukemia. However, the number of bladder cancer and leukemia deaths was too small to draw meaningful conclusions. The hazard ratios for esophageal cancer were higher in ever-underground workers but not in surface-only workers, and there was a suggestion of a relationship with level of DE exposure. However, the number of esophageal cancers was too small for reliable evaluation (16 for ever-underground and seven for surface-only workers).

We observed 17 deaths from pneumoconiosis, whereas only two were expected. These consisted of 10 coal workers’ pneumoconiosis, one asbestosis, one silicosis, and five unspecified. Of the 16 having past work information, 13 had worked in coal mining before employment at the study facilities. Three individuals had worked extensively in jobs with either likely or possible exposure to lung carcinogens, including a power plant (22 years; asbestosis), an auto mechanic (27 years; unspecified), and a mining mill operator (27 years; silicosis). Overall, the development of pneumoconiosis in these workers appears likely to be related to their previous work.

This investigation had limitations typical of cohort mortality studies, including the uncertainty commonly encountered in exposure assessment (eg, limited sampling data for some jobs and reliance on surrogate information for extrapolation to the past), incomplete information on potentially hazardous exposures received in other jobs held before or after employment in the study facilities, and lack of information on lifestyle factors (eg, smoking). These limitations pertain to the information for every individual in the study to a greater or lesser extent. Although the study lacked smoking data, the findings from the companion case–control study (21) showed that the REC effect was not attenuated by the inclusion of smoking in the models. Smoking information available from the case–control study and from a morbidity survey of underground metal and non-metal miners undertaken in the mid-1970s that included six of the mines in this mortality study [see (36) for results for the potash miners] did not show any evidence of increased smoking prevalence in the more highly DE-exposed jobs underground. In addition, Hein et al. (37) have pointed out that adjustment for birth year (as undertaken in this study) can partially adjust for confounding from smoking. Taken overall, this information implies that smoking was unlikely to have caused the observed relationship of lung cancer with DE exposure in the cohort study.

The case–control study also adjusted for other lifestyle factors as well as for employment associated with potentially confounding exposures from work outside of the study mines. As noted above, the exclusion of workers with less than 5 years tenure appears to have been a surrogate for adjustment of other exposures, although it should be noted that the main REC exposure–response findings were evident without any tenure exclusion (Figures 1 and ​and22).

This study also had strengths, including 90% statistical power to detect a doubling of risk of lung cancer in the highest DE exposed workers, time since first exposure sufficient to detect excess lung cancer mortality, and multiple study facilities in various geographical locations and mining different commodities. DE levels among underground workers in this study were considerably higher on average, and had a wider range, than in virtually all previous investigations, thus increasing the power to detect any DE effects. The study had extensive information on potential workplace confounders for lung cancer, and we chose workplaces largely devoid of known confounders. In addition, the exposure assessment relied on thousands of recent and historical measurements instead of relying solely on surrogate exposure information (23). The anchor measurements (the 1998–2001 DEMS environmental samples) consisted of job-specific means of multiple REC samples. They therefore provided reliable estimates of environmental conditions in 1998–2001. Because the samples were collected for epidemiological use, they were more likely to accurately represent the working environment than samples collected for compliance purposes (compliance measurements may be greater than typical levels because compliance samples are more likely to be taken when environmental levels are suspected to be high). Past exposure extrapolation was based on trends in measurements of CO (along with REC, a major emission from diesel engines) supported by models using indicators of diesel usage and ventilation rates obtained from records and supplemented by local knowledge from workers (25). Although the CO data were based on compliance measurements, we used them only for the development of relative trends, not to estimate absolute levels of exposure.

The exposure assessors (P. A. Stewart, R. Vermeulen, J. B. Coble) were blinded to the mortality outcome data during their development of the exposure estimates. Their estimates were evaluated by comparison with independent data (see “Methods”) and showed good agreement (22). Through the use of alternative metrics, we were able to explore the effect of exposure assessment assumptions and demonstrated that similar findings emerged regardless of the REC estimate used (Supplementary Table 13, available online). Importantly, although imprecision in our exposure estimates exists, we feel that it is non-differential (ie, not systematically biased to higher or lower exposure levels across the study), and as such, would lead only to the exposure–response coefficients being biased to the null in expectation (26). Finally, the consistency of the findings across multiple analyses using alternative exposure estimates and modeling approaches demonstrated that the results were robust to different choices of methodological strategy.

We believe that it is unlikely that the results are subject to healthy worker survivor selection effects (41) arising from individuals leaving work because of respiratory disease. We specifically chose the non-metal mining environment for its absence of known lung cancer health effects relating to dust exposures. The earlier morbidity study of underground potash miners, undertaken at most of the facilities participating in this study, showed no obvious severe respiratory problems at the mines (42). Moreover, the respiratory disease findings from this study do not indicate any excess mortality among causes that would lead to the suspicion that workers might have left work because of respiratory disease.