The process of estimating retrospective exposure levels for this study was complex and time consuming, but quantitative levels of exposure were crucial to the epidemiologic study to allow investigation of exposure–response relationships between DE exposure and mortality outcomes. Given the large differences in exposure levels between the facilities' underground operations (Coble et al., 2010
) and the existence of pronounced trends in the estimated REC exposure levels over time (Vermeulen et al., 2010a
), quantifying exposures minimizes misclassification of exposure that would otherwise occur using cruder estimates of exposure levels. In this study, the types and amount of data available for estimation of past exposure levels varied across facilities, across jobs within the facilities, and over time. To maximize the usefulness of the data to accurately estimate REC exposure levels, we developed methods for integrating the disparate types of data, identified the assumptions being made, and instituted numerous quality control checks.
Wherever possible, supplemental data were used to evaluate the exposure assessment methods. Some investigators have described this type of effort as validation. That term, however, implies the availability of a gold standard, a standard that is rarely found in epidemiologic studies. We did not have a gold standard; however, we did have several different types of information that could be used for comparisons. Thus, we made a series of comparisons of the estimates or components of the estimates with sources of measurement data or exposure surrogate information not used in the estimation process, and we conducted several sensitivity analyses to understand the effect of different assumptions (). These evaluations were made to estimate the accuracy and the reliability of the exposure estimates and will be presented elsewhere.
Summary of the method evaluations
The goal of the exposure assessment effort was to develop quantitative estimates of historical exposure levels to DE. DE is a complex mixture of gases and particulates. Of all the DE constituents, REC currently is considered the best surrogate of DPM (Bunn et al., 2002
) because it can be measured easily and has acceptable measurement variability (Birch and Cary 1996
; Birch and Noll, 2004
). It is not known if REC is the best surrogate for determining the possible carcinogenicity of DE, but REC correlated moderately to strongly with most other components of DE in the DEMS surveys (Vermeulen et al., 2010b
). REC could not be used, however, to estimate historical exposure levels because there were no measurements before the 1990s. NO2
and CO, two other components of DE, have often been used historically as surrogates for DE (Pronk et al., 2009
). There were many fewer NO2
measurements than CO (~7000 versus 11
000), and for three of the facilities, the percentage of nondetectable NO2
measurements was very high (i.e. up to 90%). For these reasons, CO was selected to estimate relative differences in DE concentrations over time. The fact that CO measurements correlated moderately with REC measurements, they were associated with elemental carbon, NO, CO2
, and NO2
measurements in a factor analysis and they were approximately linear in log-log space with REC measurements supported our decision (Vermeulen et al., 2010b
The DEMS survey data were very comparable to measurements taken concurrently in some of the facilities by different investigators (Cohen et al., 2002
) (). A statistical difference between the means of three types of jobs (production, maintenance, and surface) in these two studies was found for only one of nine comparisons and there, only at the surface which may have been an artifact of how the jobs were defined in the facility. Comparability of the two data sets suggests that the DEMS data used to estimate 1998–2001 exposure levels were reliable.
Within any given facility, the underground operating sections or faces were at different distances from the intake and exhaust shafts, and the faces may have had different types of equipment being used and differing exhaust airflow rates. The DE air concentrations at the production faces, therefore, may have varied. The work histories, however, did not indicate at which face the study subjects worked, so we assigned the same facility-specific exposure estimate to all underground subjects with the same job in the same year. Examination of the measurement data suggests that this approach is not likely to have had a substantial impact on the estimates. The average REC levels of both the underground personal and the area measurements typically were characterized by geometric standard deviations (GSDs) of <3 within each facility (Coble et al., 2010
), even though these measurements were taken at different sections of the underground operations. Furthermore, for any given year within a facility, operating faces generally used similar types and numbers of equipment to extract the ore, so that the HP, and therefore the DE level, was not likely to have differed substantially between different operating faces of a facility. The homogeneity in the REC measurements between mine faces and in the equipment likely used at the faces suggests that exposure levels to DE were likely to have been similar across faces. The approach taken here has been used by others studying underground miners (e.g. Seixas et al., 1991
Exposure assessment in epidemiologic studies typically requires a strategy for developing exposure groups for the purpose of estimating exposure levels for jobs lacking measurements for all or part of the study time period. The goal of grouping is to reduce within-group variability and maximize between-group variability. The first set of underground job groups (U1) was highly specific and generally associated with low variability in the REC measurements (Coble et al., 2010
). U2 groups were based on the proportion of time jobs were worked in four particular areas of the underground operation. This approach was taken because the DEMS REC area measurements suggested that air concentrations in these four areas differed systematically (Coble et al., 2010
). Within each area of the operation, however, exposure levels were relatively homogeneous. About 70% of the underground area means had GSDs <3 and about half had GSDs <2. Both U2 and, indirectly, U3 groups used estimates of time spent in each of these areas based on data collected from the interviews. To determine the validity of the time estimates, we used these estimates with the DEMS area measurements to calculate 8-h time-weighted REC averages for various jobs and compared these to the corresponding jobs' means of the DEMS full-shift personal REC measurements. We found a low overall median difference of −19% and a high correlation of 0.83 between the two types of estimates (). Thus, the time estimates we used appeared to be representative of the true percentages of time spent by workers in these areas when monitored.
Pooling of measurements from different jobs into a single group provides more reliable estimates than keeping the measurements of individual job titles separate because the number of samples is increased, but this advantage is at the potential cost of greater (nondifferential) misclassification among the jobs, resulting in a reduction in the range of exposure levels between jobs. Our preference was to assign the most specific mean with a minimum of five measurements, which maximized the reliability of the job-specific estimates. We were unable to evaluate between and within worker variability because personal identifying information was not collected on the measured workers. After assigning the estimates, however, we conducted an analysis of variance of the DEMS REC measurements in the seven facilities to determine how well the underground grouping strategy performed. We found that the standardized job title approach (U1) explained more variability than the other two types of groups (U2 or U3) in three facilities (B, E, and G) (), and for these facilities, more of the underground exposure-years were based on the U1 groups than the other groups (although for Facility E there was little difference between U1 and U2 groups). In addition, there was little between job variance for Facilities A, H, and I. For these facilities, >90% of the actual underground exposure-years were based on U1–U3 groups. Thus, it appears that the adopted grouping strategy was successful in explaining the variability of the measurements. This approach assumed that the estimates of time remained valid historically, as did the calculation of exposure levels of mixed workers who worked both underground and on the surface within a work shift. The U2 and, indirectly, U3 groups were based on broad estimates of time (<30, 30–59, and >59%), which was likely to minimize misclassification. In addition, the primary jobs that spent time on the surface and underground were management and technical (e.g. engineers), but the number of exposure-years contributed by these jobs and the generally substantially lower exposure levels assigned to these jobs suggested that small variations in the time spent underground would have had minor effects on the overall study results. In all cases, when discussing jobs or diesel equipment use, the interviewers were asked to focus on historical conditions, not current conditions, which also should have minimized errors.
The facility-specific prediction models were based primarily on ADJ HP and the total airflow rates exhausted from the underground operations. The parameter estimates observed for these exposure variables varied slightly between facilities (Vermeulen et al., 2010a
), in spite of the operations having different physical configurations, levels of HP and exhaust airflow rates, and mining methods, suggesting that these factors were robust and valid predictors of exposure levels. In addition, several evaluations were made of the estimation process. First and foremost, in 1976–1977, the MESA/BoM surveys designed to obtain representative exposure levels for an epidemiologic study were conducted in six of the facilities in the study (Sutton et al., 1979
). Among the agents monitored was CO. We compared the underground CO estimates developed from the primary prediction models for 1976–1977 to the measured underground CO levels from those surveys. There was an overall median relative difference of only 29% (−25 to +49% across facilities) () (Vermeulen et al., 2010a
). We also had REC measurements from our feasibility study conducted in Facility B in 1994 () (Stanevich et al., 1997
). We were able to compare, for two of the four underground jobs monitored during this survey, the average measured REC levels with the 1994 REC estimates and found differences of only −6% and −10% for the two jobs.
We also developed two alternative sets of historical prediction models to evaluate the sensitivity of the primary models to the assumptions used in the models. One set of models used a power relationship of REC
for all facilities, based on the relationship between REC and CO in the cross-sectional DEMS surveys in 1998–2001, rather than the 1:1 ratio used in the primary models () (Vermeulen et al., 2010b
). The second set used 5-year averages of the MIDAS CO measurements. These two sets of relative trends were applied in sensitivity analyses to the underground jobs' 1998–2001 REC estimates, as had been done for the primary set of estimates, to develop historical estimates of REC levels. Cumulative exposure was calculated for each underground study subject using the three sets of estimates. Overall correlations between the subjects' cumulative exposure estimates based on the primary set of models and their cumulative exposure estimates derived from the two alternative set of models were ~0.9 for both models (0.95–0.99 for both the REC
and the 5-year average CO models across the facilities) (Vermeulen et al., 2010a
). These results support the robustness of the primary estimates.
No adjustment to the REC estimates was made to account for respirator use for several reasons. First, when respirators are used, there are substantial requirements of a respiratory protection program to ensure that wearers are adequately protected (US CFR 30, Part 57, 2001
). Second, as of 2005 no air purifying filter respirator had been certified by NIOSH for protection against DPM (US CFR 70, Part 57, 2005
). Third, respiratory use was optional in the study mines. Thus, for these reasons, it is unlikely that historical use of respirators was effective for protection against the higher levels of DE seen historically in this study. Fourth, no information was available as to which cohort members used respiratory protection. Information was available on subjects from the case–control study, with 45% of the controls reporting ever using a respirator or other respiratory protective equipment, such as a mask, at the study facilities. Reports of use of such protective equipment did not, however, impact the findings of the case–control study.
Estimation of exposure levels for surface jobs used a strategy similar to that used for underground jobs, i.e. we preferred increased specificity, based on at least five REC measurements from the DEMS surveys, over greater reliability of the estimates. Comparison of the AMs of the categories within each facility typically found increasing trends from the expected low to the expected high surface category, and there was generally an inverse monotonic trend for the percentage of nondetectable measurements (). Surface REC exposure levels were estimated to be 4–11 μg m−3
when operating heavy diesel equipment, driving a forklift indoors, or repairing diesel equipment; 2–4 μg m−3
when working near heavy equipment or operating light equipment; and 1–2 μg m−3
for the remaining, generally unexposed, workers. In other studies, measured REC exposure levels of construction workers operating heavy equipment were 8–15 μg m−3
(Blute et al., 1999
; Woskie et al., 2002
; Lewne et al., 2007
) and two studies of dockworkers found mean exposure levels of 1 and 24 μg m−3
(Zaebst et al., 1991
; Davis et al., 2007
). Mechanics repairing diesel equipment have been reported as having mean REC exposure levels of 1–36 μg m−3
(Zaebst et al., 1991
; Sauvain et al., 2003
; Seshagiri and Burton, 2003
; Davis et al., 2007
; Lewne et al., 2007
). Outdoor workers near diesel-powered equipment had a geometric mean exposure level of 4 μg m−3
(Lewne et al., 2007
), and the average background level near highways was 3 μg m−3
(Zaebst et al., 1991
). Finally, residential and industrial background air levels have been reported to be 1–2 μg m−3
(Zaebst et al., 1991
; Davis et al., 2007
). Thus, the study estimates for the surface REC exposure levels are reasonably consistent with the published literature.
Several of the comparisons conducted with independent data to evaluate the study estimates were described above. We conducted several other evaluations as well (). Each of these exercises has its limitations but overall, the results showed moderate to high agreement. Moreover, the most important comparison, that of the 29% difference between the 1976–1977 CO observed measurement means to the predicted 1976–1977 CO estimates, is in the range that investigators of other studies have found who have been able to evaluate their exposure assessment methods (Hornung et al., 1994
; Burstyn et al., 2002
; Stewart et al., 2003
; Astrakianakis et al., 2006
) and is similar to differences seen in two studies of side-by-side measurements of acrylonitrile (Zey et al., 2002
) and of REC (Cohen et al., 2002
). These findings, with those of other evaluations, indicate that the estimates were likely to have been accurate. This is not to say that we believe that the estimated exposure levels are without error; bias and imprecision in the estimates is inherent in any estimation procedure.
In conclusion, the goal of the study was to develop quantitative estimates of DE. We accomplished this for all 1164 facility/department/job combinations for every year from 1997 back to the first year of dieselization in each of the eight facilities by using REC as the surrogate for DE. The process was done first by calculating REC means of underground and of surface jobs or groups of jobs using the 1998–2001 REC measurement data and then back extrapolating these means using prediction models. Exposures to RD, silica, radon, asbestos, and PAHs also were assessed. The exposure assessment process was complex because the data available typically varied across facilities, across jobs within the facilities, and over time. We integrated these differing sources of information, explicitly defining the assumptions made, and performed a substantial number of quality control checks. We compared the estimates to several different types of independent data using various approaches. The cumulative evidence of the many evaluations indicated moderate to high agreement. In particular, the comparison with the 1976–1977 CO air concentrations found differences close to what others have found in epidemiologic and monitoring studies. For these reasons, we believe that the estimates are credible and can be used in the epidemiologic analysis with confidence.