Methods

Description of cohort

The cohort was identified from among all employees who have ever worked at the Washington Works plant between 1 January 1948 (plant start-up) and 31 December 2002.12 The cohort was ascertained primarily through the DuPont Epidemiology Registry with administrative records obtained from the human resources department at the Washington Works plant to identify all eligible workers for the cohort regardless of current employment status.12 Social security numbers for cohort members were submitted to the Social Security Administration (SSA) for confirmation of vital status as of 31 December 2002.16 Causes of death were determined primarily through the DuPont Epidemiology Registry mortality database that ascertains employee cause of death through death certificates submitted with life insurance claims filed by beneficiaries of deceased employees and pensioners. Additionally, the US National Death Index (NDI Plus) was used to ascertain cause of death for all cohort members reported as deceased by the SSA and to seek mortality records for individuals whose vital status could not be determined via SSA. IHD cases included mortality events with International Classification of Diseases, Ninth Revision (ICD-9) codes 410.0 through 414.9, and Tenth Revision (ICD-10) codes I20.0 through I25.9 listed as the underlying cause of death.

Exposure assessment

Time dependent APFO exposures were estimated from detailed work histories for all employees using an exposure reconstruction model developed from occupational information and serum PFOA data (KH Kreckmann, DuPont Epidemiology Program, 2008, submitted manuscript). We opted to use serum biomonitoring as the indicator for exposure as opposed to environmental measurements for three reasons. First, serum PFOA is believed to be a stable biomarker. Second, it reflects exposure from all routes of entry into the body (inhalation, ingestion and dermal exposure). Finally, it has the advantage of better illustrating actual exposure since it incorporates both occupational as well as non-occupational exposures.

Briefly, we conducted a cross-sectional health survey in 2004 that included serum PFOA measurements for 1025 employees, representing 55% of eligible workers.10 Since PFOA has a reported median half-life of 3.4 years, we examined the length of time an employee spent in a job before sampling.9 We found little association between participants’ serum PFOA level and the length of time in the job held at the time of sampling (Pearson correlation coefficient between log PFOA and time in the job r = 0.082). Therefore, recent job changes or lengthy tenures within a job should not substantially contribute to misclassification of job titles within exposure categories.

To establish relative exposure categories for current job titles, we linked individually measured serum PFOA levels with the job title held by the individual at the time of sampling. The calculation of median, range and distribution of serum levels was used to determine the typical exposure for each job title. Then we assigned these job titles into relative exposure categories (low, medium and high). Afterwards, exposure categories were assigned intensity factors that corresponded to the mean serum levels of all the jobs that were included in that category. Subsequently, we applied these intensity factors to all historical job titles in the cohort based on their correspondence with current job titles.

We validated the resulting job exposure matrix with additional historical blood data collected from 1979 through 2002 from voluntary participants in a separate biomonitoring program at the Washington Works plant. These samples had been collected to ensure the effectiveness of workplace controls on exposure and were not used to develop the reconstruction model. We found a good correlation between the historical biomarker level and the estimated exposure intensity.

To calculate each worker’s estimated cumulative exposure to APFO, we multiplied the duration of time in a specific work assignment by the mean intensity estimate from the job exposure matrix. All resulting period exposure estimates were then summed.17 Workers were divided into four categories based on estimated cumulative exposure quartiles determined by the distribution of IHD mortality cases (case quartiles) or the entire cohort (cohort quartiles).

Statistical analysis

Relative mortality risks were estimated for IHD mortality outcomes assuming a log-linear exposure–response relationship with APFO. Risk sets were compiled for each IHD case that included all non-cases in the cohort of the same sex and with occupational experience through the age in years of the IHD case.18 Therefore, in each risk set, cases and non-cases were matched on sex and age (the age of death of the case who defined each risk set). Cox proportional hazards regression models were fitted using the PHREG procedure in SAS v 8.1 (SAS Institute, Cary, NC). The model outcome was a time variable measuring age at IHD mortality event or censor date for each worker in each risk set.19 Time-related changes in IHD mortality were adjusted for in regression models using the calendar year of the case event or the calendar year at which non-cases in a risk set attained the same age as the matched case. Race was also adjusted for in our models.

Time-dependent exposures were evaluated based on the estimated cumulative exposure attained by each worker at the age of the case event in a risk set. We categorised estimated cumulative exposure metrics into four groups using two a priori approaches.19 20 21 22 23 24 First, quartiles of cumulative exposure were determined by the distribution of exposures for case subjects. Second, quartiles were determined by the distribution of exposures for all workers in the cohort. In both analyses, employees in the lowest exposure category comprised the referent group. We conducted tests for linear trend for each categorical analysis of estimated cumulative exposure. No adjustment was made for multiple comparisons. Finally, exposure was lagged to address the concern that early symptoms of IHD may cause subjects to leave the workforce, thus ending their exposure. Further, lagged exposures allow for comparison of exposure–response associations among cohort members accounting for potential disease latency period.

Results

Our analyses were restricted to employees who were part of the risk sets compiled for the IHD mortality analysis. Our cohort included 105 females (102 white and three non-white) and 4642 males (4460 white and 182 non-white). There were 239 deaths due to IHD in this cohort: three cases occurred among women (all white), one in a non-white male, and 235 among white males. The final risk set analysis comprised 4747 male and female employees who were matched to at least one IHD case in the study cohort. Employment-related characteristics are described for this group in table 1. Over one half of these workers were hired prior to 1974. The median age at hire was 26.6 years, while the median duration of employment was 22.5 years.

Table 2 displays characteristics of the cohort by vital status category. A total of 773 deaths (16.3% of the study cohort) have occurred through 2002 with 239 deaths due to IHD (30.9% of all deaths) observed. Acute IHD diagnoses, ICD-9 codes 410 and 411, accounted for 63% of all IHD deaths.

Categorical relative risk estimates for IHD mortality associated with estimated cumulative exposure are shown in table 3. Workers were divided into four categories based on estimated cumulative exposure quartiles determined by the distribution of IHD mortality cases (case quartiles) or the entire cohort (cohort quartiles). Those in the lowest exposure category comprise the reference group for each analysis. Although no risk estimates are statistically significant for any higher exposure category relative to the referent group, the results from the 10-year lagged exposure categories determined by cohort quartiles suggest an increasing trend in the relative risks for the highest two exposure categories (p for trend = 0.06). These results are attenuated towards the null for categories determined by case quartiles of estimated cumulative exposure (p for trend = 0.16).

Discussion

We observed no significant increase in the mortality risk for IHD associated with increasing exposure to APFO. When relative risks were estimated for categorised exposures, results from models of estimated cumulative exposure lagged by 10 years indicated elevated relative risks for the two highest exposure categories. While neither category-specific estimate was statistically significant, there was an apparent linear trend when categories were determined by cohort quartiles of estimated cumulative exposure. Categorical risk estimates for a second set of analyses using case-determined quartiles were attenuated towards the null. Positive risk estimates at four other exposure lags were absent.

In this occupational cohort of workers, IHD deaths accounted for 31% of all deaths in the cohort with 63% of these due to acute myocardial infarction, one of the leading causes of death in West Virginia and the United States.25 Previous studies of DuPont workers had found a steady decline in the incidence of and mortality from myocardial infarction from 1957 through 1983, a period accounting for approximately 30% of the deaths in this cohort.26 Although rates for IHD among white males residing in the state have been decreasing over time, West Virginia ranks among the highest states for IHD incidence and mortality.27 28 29 To account for long-term temporal trends in IHD, we adjusted for calendar year in our analyses.

Previous studies undertaken at the Washington Works plant showed a positive association between serum PFOA and serum lipids.10 11 Because increased lipids levels are an established risk factor for IHD, we undertook a retrospective cohort mortality study to determine whether IHD mortality is associated with APFO exposure. An earlier plant-wide study of mortality indicated a non-significant elevation for IHD mortality (SMR 109, 95% CI 96 to 124) relative to expected deaths based on rates of IHD mortality for a regional DuPont employee population.12 Additionally, the mortality experience of 3537 male and female workers at a separate company’s PFOA production plant was described by Gilliland and Mandel.30 The authors reported no increase for mortality from cardiovascular diseases in the cohort. Among men, the SMR for coronary and atherosclerotic heart disease was significantly reduced (SMR 69, 95% CI 57 to 83) using Minnesota rates for comparison. This estimate remained relatively unchanged when analysis was restricted to men who were employed in the chemical division and had direct exposure to APFO (SMR 74, 95% CI 54 to 100).30

Occupational cohort mortality comparisons are often affected by the healthy worker effect.13 14 This effect results from the employment and retention of relatively healthy individuals when compared to the general population.31 It is more often observed for outcomes that directly influence workforce selection processes such as cardiovascular diseases, diabetes mellitus and non-malignant respiratory diseases.32 The current exposure–response study evaluates IHD mortality using quantitative estimates of occupational exposure to APFO internally for workers at a single plant, thereby reducing potential bias due to healthy worker comparisons. Further, we lagged estimated cumulative exposure measures in order to adjust for the possible confounding due to continued employment of healthy workers.33 34 Lag periods remove more recent exposures from a worker’s cumulative estimate and may reduce bias that results from the preferential retention of healthier workers in the workplace.35 36 Although lagging exposure might reduce healthy worker selection bias to a degree, it is only an indirect approach and residual selection bias may still be present.

An additional limitation to the findings of this study is that we relied on administrative records for work history. Therefore, for the majority of employees, no information was available for individual risk factors for cardiovascular diseases such as smoking, hypertension, diabetes, dyslipidaemia, family history, obesity and other life-style factors. Data on treatment interventions including the use of lipid-lowering agents or anti-hypertensive medications were also not available. Given the observed association between APFO exposure and serum cholesterol levels, this may potentially confound mortality risk estimates. For this analysis, we assume that key demographic characteristics are similar among employees at the plant. Additionally, previous research found that important risk factors such as diet, lack of exercise and smoking are widely prevalent among West Virginians.37

Further, we analysed the association between serum PFOA and BMI, smoking status, blood glucose level, systolic and diastolic blood pressure, the use of glucose lowering agents, the use of anti-hypertensive agents, and the use of lipid-lowering agents in a subset of the plant cohort, specifically all employees (243 females and 782 males) who participated in the cross-sectional survey (n = 1025).10 We found no association between serum PFOA levels and any of these potential confounders. Although we cannot completely eliminate potential residual confounding due to unmeasured individual factors, it is unlikely to bias the mortality risk estimates associated with APFO exposure in such a way that a true association is masked.

Another potential confounder is socio-economic status. Workers at the plant have similar access to health care services through company insurance programs. In addition, workers in the APFO area who would have potentially higher APFO exposure do not differ from other employees at the plant in terms of education, training and income. Therefore, we do not expect socio-economic status to be differential among exposure categories and therefore to confound the relationship between the exposure and outcome in our study.

The primary strengths of this study include the availability of biomonitoring data to support retrospective exposure classification for a large cohort of 4747 workers. Given that there were over 50 years of mortality follow-up, sufficient data were available for several types of analyses evaluating internal cohort comparisons of mortality potentially associated with increased exposure to APFO.

In conclusion, the results show no convincing evidence of increased mortality risk from IHD associated with APFO exposure for workers at this plant. We found a marginally significant trend associated with mortality risk for higher categories of exposure at a lag of 10 years. Further study of the incidence of IHD in this cohort is being conducted by other researchers.38 Incidence studies may reduce the temporal effects that early diagnosis, access to medical treatment, and temporal changes in risk factors may have for subsequent IHD mortality.