We present a retrospective assessment of lung cancer mortality for 38 years of follow-up in a large cohort of railroad workers, finding elevated risk among engineers, firemen, conductors, and brakemen, job categories identified as diesel exposed. Disregarding exposure in the 5 years before death, the RR for these workers compared with workers without regular work in an exposed job was 1.40 (95% CI, 1.30–1.51). Unlike the original findings of greatest risk in younger workers, lung cancer mortality was elevated to a similar extent regardless of age at entry (in 1959) except in workers 60–64 years of age. Thus, excess risk was not limited to workers with the greatest opportunity for exposure because of their being younger at the start of the diesel era. Finally, there was no evidence of an increased risk with increasing years of work (the exposure surrogate) in a job with exposure to diesel exhaust.
Our observation of lung cancer risk is similar to the risk noted by others in the literature. In > 35 studies of workers with occupational exposure to diesel exhaust, excess risk of lung cancer is consistently elevated by 20–50% (reviewed in Bhatia et al. 1998
; Lipsett and Campleman 1999
). Most occupational studies rely on a single report of job title to define exposure. In this study, job title was available for each year of follow-up, and jobs with exposure to diesel emissions were defined by an industrial hygiene survey. These results are similar to smoking-adjusted RRs attributable to fine particulate air pollution on lung cancer in prospective population-based cohorts (Dockery et al. 1993
; Pope et al. 2002
) and risk of lung cancer attributable to vehicle exhausts in urban settings (Nyberg et al. 2000
). Effects for cardiovascular and respiratory disease mortality are also consistent with observations reported by population-based studies (Dockery et al. 1993
; Pope et al. 2002
Although we originally reported that lung cancer risk increased with increasing years of work in diesel-exposed jobs (Garshick et al. 1988
), subsequent reanalyses of these data, with adjustment for attained age, indicated decreased risk with more years worked (Crump 1999
; Health Effects Institute 1999
). This inverse association with exposure duration could be explained by a healthy worker survivor effect. Analysis in this updated cohort with longer follow-up also indicates that lung cancer mortality is inversely related to total years worked. The possibility that the healthy worker survivor effect influences the assessment of mortality had not been considered previously. Although methods for controlling for the healthy worker survivor effect have been proposed, it is uncertain whether full adjustment by statistical methods is possible. It was not possible to implement methods suggested by Robins (1987)
because there was little change in exposure status, and retirement patterns were stable. Other methods to adjust for healthy worker effects (Arrighi and Hertz-Picciotto 1993
) consider employment status and exposure lag models to exclude recent exposure. With overall employment duration and employment status considered, the relationship between lung cancer risk and years of work in a diesel-exposed job was elevated regardless of exposure duration (). Restriction of the cohort to subjects who survived beyond the last year worked and stratification on retirement time also gave similar results. We also conducted alternative survival analyses (compared with proportional hazards methodology) employing recently developed techniques in which time to an event is modeled using “first hitting time” methodology (Lee and Whitmore 2003
; Lee et al. 2004
). Using these methods, there was evidence of a healthy worker survivor effect, with an elevated risk of lung cancer mortality among train crews (Lee et al. 2004
Exposures before 1959 and changes in exposure patterns could also modify a relationship between years of work and lung cancer mortality. An expectation of increasing risk with years of exposure implicitly assumes that the exposure intensity is approximately constant across years. Diesel locomotive emissions changed throughout the follow-up period. Explicit exposure data are not available, but the first diesel engines (1940s through 1950s) were said to be “smokier” than later locomotives (Woskie et al. 1988b
). Cleaner locomotives were introduced in the early 1960s and the 1980s. Although diesel engines are known to produce mainly fine and ultrafine particles, similar information is not available on coal-fired locomotives. Temporal changes in diesel and other combustion-related emissions might contribute to the lack of an exposure–response relationship based on duration of exposure in the train crews.
Because all workers were employed in 1959 and had exposures in the previous 10–20 years, we could not assess whether work exclusively during the diesel or steam locomotive era or with early diesel locomotives differentially influenced mortality. However, in a case–control study using RRB records to determine deaths in 1981–1982, workers > 65 years at death were exposed mainly to steam engine emissions, and younger workers mainly to diesel engine emissions. In the older group, work in diesel-exposed jobs was not associated with lung cancer mortality, whereas the RR was significantly elevated for the younger group. In the present study, the oldest workers (60–64 years of age at study entry) had the fewest years of work after 1959 and the lowest mortality due to lung cancer. These results suggest that introduction of diesel locomotives significantly contributed to lung cancer mortality in the cohort.
Small RRs may be affected by uncontrolled confounding, such as differences in cigarette smoking habits in subjects with and without diesel exposure. In this retrospective cohort, individual data on smoking history are not available. To minimize the possible effect of uncontrolled confounding by smoking, efforts were made to include only workers of similar socioeconomic class, a known correlate of smoking habits (Brackbill et al. 1988
; Stellman et al. 1988
). Further, estimates were similar when the reference group was restricted to signal maintainers, potentially a more blue-collar unexposed group. Smoking rates vary by birth cohort (Burns et al. 1997
). However, all analyses are stratified by age; thus, birth cohort is controlled for.
In our previous case–control study using RRB records (Garshick et al. 1987a
), smoking history was obtained from next of kin, and crude and smoking-adjusted effects of exposure were similar. With the distribution of job-specific smoking habits from the case–control study and a survey of 514 white male workers employed by a small railroad in 1982 (Garshick et al. 1987b
), we calculated age- and job-specific smoking adjustment factors using Schlesselman and Axelson methods (Axelson and Steenland 1988
; Larkin et al. 2000
; Schlesselman 1978
). These factors, the ratio (diesel exposed:unexposed) of literature-based lung cancer risks weighted by job-specific smoking behavior generally ranged from 1.1 to 1.2 (Larkin et al. 2000
). Other investigators have reported similar factors (Blair et al. 1985
; Levin et al. 1990
; Siemiatycki et al. 1988
). Dividing the observed RR for lung cancer for the present study by these factors attenuated the RR to between 1.17 and 1.27. These estimates are consistent with other literature-based smoking-adjusted risks attributable to diesel exhaust, traffic emissions, and air pollution (Dockery et al. 1993
; Nyberg et al. 2000
; Pope et al. 2002
; Steenland et al. 1990
). This indirect method is limited in adjusting for smoking by assuming no interaction between diesel exposure and smoking, but there are insufficient data to assess this possibility.
Respiratory disease mortality, including from COPD and allied conditions, was also associated with exposure. The predominant cause of these diseases is cigarette smoking, possibly providing evidence of confounding by smoking in our lung cancer analyses. However, smoking-adjusted cohort studies show that occupational exposures to dusts and fumes are also associated with chronic respiratory symptoms and airflow obstruction (Garshick et al. 1996
; Hnizdo et al. 2002
). Studies of workers specifically exposed to diesel exhaust indicate that there is an increase in respiratory symptoms and a reduction in pulmonary function with exposures (U.S. EPA 2002
). Controlled studies of human exposures to diesel exhaust and to other fine particles results in pulmonary inflammatory changes (Ghio et al. 2000
; Salvi et al. 1999
). Therefore, exposure to diesel exhaust from operating trains may in fact lead to an increased risk of chronic respiratory disease mortality, independent of smoking.
Factors other than smoking that might modify the risk of lung cancer seem unlikely to contribute further uncertainty to these results (Alavanja et al. 2001
; Henley et al. 2002
; Olson et al. 2002
). These factors are much less significant than smoking and not expected to be related to exposure. Controlling for the healthy survivor effect by considering the ability to work (years of work) and live into retirement (time off work) in the regression models also may reduce uncontrolled confounding by other lifestyle-related factors and might further address adjustment for smoking behavior. Death certificates were used to identify causes of death. Death certificates may overascertain rather than underascertain primary lung cancer (Bauer and Robbins 1972
; Goldman et al. 1983
; Jimenez et al. 1975
; Kircher et al. 1985
; Percy et al. 1981
; Rosenblatt et al. 1971
). This type of misclassification is likely to be random with respect to exposure and would make the effect of exposure harder to detect.