Study subjects were vermiculite miners, millers, and processors. Workers also may have been assigned jobs in the screening plant, railroad loading dock, expansion plants, or an office located in the town of Libby (several miles from the mine). The cohort was enumerated in May 1982, and study subjects were followed through December 2001. The design allowed a minimum 20 years of follow-up since first exposure, with > 65 years of follow-up for the earliest hired workers. Demographic and work history data were abstracted from company personnel and pay records. A database created by the National Institute for Occupational Safety and Health (NIOSH) in the 1980s contained demographic data, work history, and vital status at the end of 1981 for 1,881 workers. The data were validated against company records on microfilm at NIOSH, and work history data were reabstracted. One person was removed from the cohort because company records stated that he was hired but never worked. Nine workers with Social Security numbers listed in company records were excluded because demographic and work history data were not available, leaving 1,871 potential study subjects.

The NIOSH Institutional Review Board approved the research protocol; the study complied with all applicable U.S. requirements and regulations for studies involving human subjects. Vital status follow-up through 2001 used the National Death Index (NDI-Plus), the Social Security Administration, the Internet [Ancestry.com (2006), RootsWeb.com (2006), and electronic links to state death records], and a tracing service. Workers known to be alive on or after 1 January 1979 (the date NDI began tracking deaths nationwide), but not found in the NDI, were assumed to be alive at the study end date of 31 December 2001. Vital status follow-up was completed for 97.8% of the cohort (n = 1,830). Nearly 47% of these workers (n = 877) had died by 31 December 2001.

For 97% of those known to be deceased, cause of death was determined from death certificates and coded to the International Classification of Diseases (ICD) using the rubrics of the ICD revision in effect at the time of death {ICD-8 (National Center for Health Statistics 1967); ICD-9 [World Health Organization (WHO) 1977]; or ICD-10 (WHO 1992)}. Deaths before 1979 were coded by a single National Center for Health Statistics-trained nosologist; for 1979–2001, ICD codes were obtained from the NDI.

The mining, milling, and processing operations at Libby; conditions of exposure; and job-specific estimates of exposure intensity have been thoroughly described previously (Amandus and Wheeler 1987; Amandus et al. 1987, 1988; McDonald et al. 1986). Briefly, miners extracted vermiculite ore from an open-pit mine. The ore was processed in a dry mill (1935–1976) and/or two wet mills (1955–1990) that operated on Vermiculite Mountain. The resulting concentrate was shipped by railroad to processing plants where the vermiculite was expanded for use in loose-fill attic insulation. Before 1975, exposures in the mine ranged from 9–23 fibers per cubic centimeter (fibers/cc) of air for drillers; exposures in other mining jobs were estimated to be < 2 fibers/cc (Amandus et al. 1987). Early fiber exposures in the dry mill were as high as 182 fibers/cc during sweeping operations; by 1964, exposures in the mill had been reduced by 80%. Amandus et al. (1987) estimated that by 1972 exposures in all work areas were < 1 fiber/cc as an 8-hr time-weighted average, compared with today’s Occupational Safety and Health Administration (OSHA) asbestos standard of 0.1 fiber/cc (U.S. Department of Labor 2006).

The job-exposure matrix developed for this study was based on that used in the earlier NIOSH study (Amandus and Wheeler 1987), with some important exceptions. In the earlier study, workers with “common laborer” job assignments and some workers with unknown job assignments were assigned the relatively low exposure estimated for the mill yard. In the current analysis, these workers were assigned the average estimated exposure intensity for all unskilled jobs during the relevant calendar time period. This resulted in higher estimates of cumulative exposure, with likely attenuation of the effect estimators. In addition, reabstracting work histories for the current study identified several job assignments not mentioned in the earlier publications. Exposure estimates for the additional job/calendar time period combinations were extrapolated based on professional experience and review of exposure records from earlier studies of Libby workers (Amandus and Wheeler 1987; Amandus et al. 1987, 1988; McDonald et al. 1986).

The exposure index available for this and previous studies of Libby workers is based on fiber count data obtained using optical phase contrast microscopy (PCM). PCM measures fibers longer than 5 μm and wider than 0.25 μm [i.e., the fiber size regulated under the OSHA standard (U.S. Department of Labor 2006)]. Fiber count data obtained using PCM do not distinguish between the various amphiboles in the Libby vermiculite. Thus, fiber count estimates used in the exposure assessment include not only the regulated tremolite asbestos fibers but also the asbestiform amphibole fibers not mentioned in the regulations (winchite and richterite). Recall that Meeker et al. (2003) characterized the Libby amphibole as 6% tremolite. If the observed health effects are explained by tremolite asbestos alone, then exposure has been considerably overestimated, and the effect of each fiber per cubic centimeter-year increment in exposure has been substantially underestimated (McDonald et al. 2004).

Dates of termination were unknown for 58 of 640 workers (9%) who left employment before September 1953. These workers were assumed to have worked for 384 days, based on the mean duration of employment among all workers with known termination date before September 1953. U.S. Census Bureau data (2004) indicate that 95% of the local population identify themselves as white. Because workers at this facility were known to be primarily Caucasian, 935 workers with race unknown were assumed to be white (NIOSH 2001). Similarly, because 96% of the workforce was male, seven workers with sex unknown were assumed to be male after review of names.

Data were managed using ACCESS 2000 (Microsoft Corporation, Redmond, WA) and SAS, Version 8.0 (SAS Institute, Cary, NC). Descriptive analyses were performed using SAS, Version 9.1 (SAS Institute).

Because 95% of study subjects were white men, the analytic cohort was limited to 1,672 white males [excluding 95 men who died or were lost to follow-up before 1960, the first-year for which comparison rates for asbestosis are available in the NIOSH Life Table Analysis System (LTAS) (NIOSH 2001)]. Using LTAS software (NIOSH 2001; Steenland et al. 1990, 1998), SMRs were calculated to determine if study subjects experienced greater mortality from specific causes than was expected based on the U.S. population experience. SMRs were adjusted for age at risk and calendar year of follow-up (categorized into 5-year age and calendar time groups). SRRs evaluated exposure response across increasing categories of cumulative exposure and duration of employment, with workers in the lowest exposure group serving as the baseline for comparison. A formal test for a linear trend in the slope of the SRRs evaluated the hypothesis that β₁ = 0 against the alternative hypothesis that duration or cumulative exposure predicts mortality. The 95% confidence limits were calculated. If the 95% confidence interval (CI) for the slope did not include zero, the hypothesis was rejected in favor of the alternative at the p < 0.05 level.

Cumulative exposure data were categorized to achieve an approximately equal number of cases in each exposure category, a method previously found to be most efficient (Richardson and Loomis 2004; Sullivan et al. 1996), while maintaining sufficient person-years in each exposure category to obtain valid estimators. In some analyses, the lowest categorical cut point was set at 4.5 fibers/cc-years [i.e., a worker’s cumulative lifetime exposure if exposed to asbestos fibers at the current OSHA standard of 0.1 fibers/cc over a 45-year working life (U.S. Department of Labor 2006)]. Exposure duration was categorized to facilitate meaningful interpretation (i.e., < 1 year, 1–9.9 years, ≥ 10 years). The lag period was chosen to avoid excluding cases with disease assumed to be work-related. For the outcomes of interest, I found an exposure lag of 15 years to present a clear picture of exposure response.

Person-years at risk and observed deaths were accrued from the date comparison rates were available (1 January 1960), or from the date of first exposure (if later) until the time that each worker died or was lost to follow-up, or until the end of the study (31 December 2001). Ten workers lost to follow-up were considered to be alive, but person-years of observation were truncated on the date each of these workers was last observed alive (date of termination, or last date known to be alive in earlier vital status follow-up). One worker was excluded from the SMR analysis because he died in Canada; another was excluded because his date of death was unknown. Twenty-four workers known to have died, but with cause of death unknown, were added to the residual cause code (NIOSH 2001).

The analysis presented here focuses on occupational respiratory conditions potentially related to asbestos exposure: lung cancer, asbestosis, and other nonmalignant respiratory disease. Mesothelioma was not coded as a distinct cause of death under ICD coding rubrics until 1999, so the SMR for mesothelioma is based on only 3 years of data (1999–2001).

Table 3 presents the evaluation of exposure–response relationships for asbestos-related occupational respiratory disease mortality. The SMR for lung cancer rose from 1.5 (95% CI, 0.9–2.3) among workers with < 4.5 fibers/cc-years cumulative exposure to 1.9 (95% CI, 1.2–2.9) among workers exposed to at least 100 fibers/cc-years (allowing for a 15-year exposure lag).

There was significant excess mortality from nonmalignant respiratory disease even among workers with < 4.5 fibers/cc-years cumulative exposure (SMR = 1.8, 95% CI, 1.1–2.8). The SMR for nonmalignant respiratory disease rose to 4.8 (95% CI, 3.1–7.3) among workers exposed to ≥ 300 fibers/cc-years (Table 3).

Cumulative exposure was a significant predictor of nonmalignant respiratory disease mortality even among those who worked < 1 year, with the SMR rising from 1.9 (95% CI, 1.1–3.2) among those with < 3.5 fibers/cc-years exposure to 2.6 (95% CI, 1.5–4.3) among short-term workers exposed to ≥ 15 fibers/cc-years (not shown). The test for a linear trend in the SRRs was statistically significant with p < 0.001.

Although 40 white male workers died with asbestosis listed on their death certificates (Table 1), SMR analysis is based on the 22 workers with asbestosis listed as the underlying cause of death (currently, the LTAS software does not include multiple cause comparison rates for asbestosis). SMRs for asbestosis increased with increasing cumulative exposure (Table 3). The SMR rose from 37.3 (approximate 95% CI, 7.5–122.3) among workers with < 50 fibers/cc-years exposure, to 749.1 (95% CI, 373.0–1367.8) among those with ≥ 250 fibers/cc-years cumulative exposure, after allowing for a 15-year exposure lag.

Table 3 also provides SRRs for lung cancer, nonmalignant respiratory disease, and asbestosis over increasing categories of cumulative exposure. For each outcome, linear trend tests were statistically significant at the p < 0.01 level.

Those working < 1 year experienced a significant excess in lung cancer (SMR = 1.6; 95% CI, 1.1–2.1), with the SMR rising to 2.5 (95% CI, 1.4–4.3) among those working for ≥ 10 years (Table 4). The SMR for nonmalignant respiratory disease was 2.1 (95% CI, 1.6–2.8) among those who worked < 1 year, and rose to 3.6 (95% CI, 2.2–5.7) among those employed ≥ 10 years. As there was only one death attributed to asbestosis among those working < 1 year, 15 months was used as the cut point for the lowest category of exposure duration (providing a more stable estimator for comparison in the SRR analysis). Those working < 15 months were 38.2 (approximate 95% CI, 7.7–125.1) times more likely than expected to die from asbestosis; among those employed ≥ 10 years, the SMR was 628.6 (95% CI, 301.1–1185.1). The SRRs for lung cancer, nonmalignant respiratory disease, and asbestosis increased across increasing categories of exposure duration, and for each outcome, the test for a linear trend in the slope of the SRRs was statistically significant at the p < 0.05 level (Table 4).

Retrospective exposure estimates were developed using a combination of government inspection reports, company compliance-monitoring data (available from 1974), and professional judgment (Amandus et al. 1987; McDonald et al. 1986). These methods likely resulted in some measurement error. Assuming that any misclassification of exposure was not systematic, the most likely effect is bias toward the null (Checkoway et al. 2004; Mannetje et al. 2002). Further, there is insufficient sampling data to develop reliable exposure estimates for potential confounders for lung cancer such as workplace exposure to diesel particulate generated by mine machinery, or exposure to respirable crystalline silica dust.

Work history data were missing for 9% of workers who terminated before 1954. These workers were assumed to have worked about 1 year, based on the average employment duration among other workers who terminated between 1935 and 1953. Without the missing data, it is not possible to determine with certainty the impact of this approach, although the most likely effect is bias toward the null. Analysis deleting the 55 workers with missing termination dates did not result in appreciably different effect estimates than are reported here.

No minimum employment duration restriction was imposed in the present analysis. Thus, overall effect estimates are somewhat lower than those reported in previous studies of this cohort (Table 5). Elevated SMRs were observed for the occupational respiratory diseases of interest and among those employed < 1 year. The assumption that biologically significant exposures occurred immediately at hire is realistic in this workplace (particularly in early years), where new hires were frequently assigned to the labor pool and often rotated through the most heavily exposed jobs. Evaluation of job-assignment patterns suggests that some job tasks (i.e., “bin mucker”) were systematically assigned to transitory workers. Alternatively, these jobs may have been so onerous that newly hired workers quit after only a day or two.

Strengths of this study include the long period of vital status follow-up—> 65 years from 1935–2001. A minimum of 20 years since first exposure to the end of follow-up allowed sufficient latency for most cancers. Previously published analyses of data from this occupational cohort reported SMRs for nonmalignant respiratory disease (Amandus and Wheeler 1987; McDonald et al. 1986, 2004). In the analysis reported here I made use of comparison rates for asbestosis that have become widely available within the last 10 years. The resulting effect estimates for asbestosis provide a more accurate description of the magnitude of asbestos-related disease among this cohort of workers.