All individuals ever employed at the facility at any time between 1 January 1989 and 1 July 2003 were eligible to be included in the study cohort. This start date was selected because medical data were known to be complete beginning in 1989, and computerisation of existing employment records began in the early 1990s. The study protocol was reviewed by a federally (US) registered institutional review board.
Complete work history, date of birth, gender and race were available for all individuals actively employed at the facility for at least one day between 1 January 1989 and 1 July 2003. Employment data were available since the plant opened in 1949, allowing complete reconstruction of employment histories for all cohort members hired before the study start date.
Starting in 1989, annual medical examinations were conducted, generally around the anniversary of an employee's date of hire, and nearly all medical records were located. Exams were also provided at entry, retirement and to those re‐hired after lay‐off. Many employees had medical records generated before 1989—some going back to 1957—but these early records were not uniformly available or complete. There is no documented rationale or systematic pattern identified to explain missing records prior to 1989.
Records were abstracted for height, weight, date of exam, and 32 clinical chemistry variables (calcium, phosphorus, sodium, potassium, chloride, alkaline phosphatase, total protein, albumin, globulin, creatinine, glucose, iron; A/G ratio, BUN, BUN/creatinine LDH, AST, ALT, GGT, total bilirubin, total cholesterol, triglycerides, HDL cholesterol, LDL cholesterol, VLDL cholesterol, total cholesterol/HDL ratio, LDL/HDL ratio, TSH, thyroxine (T4), T3 uptake, free thyroxine index, uric acid). Three laboratories had been used by the company since 1989, and a fourth lab was used from 1976–88. We used the reference ranges from the laboratory used for the greatest number of years, which was also the most recent laboratory used by the plant, to identify values that could be considered abnormal.
Complete work histories were available for all employees. Each job record provided: date of hire into the company, job title, start date for specific job, job and location codes. Working with plant staff, all job titles were reviewed to identify those with potential surfactant exposure, and to consolidate multiple job titles that occurred for the same job over time. Employees with jobs in maintenance or technical positions were considered exposed to surfactant, though at a lower level than the process workers and other workers in the building manufacturing the polymer. A non‐exposed group consisted of administrative jobs and jobs in various other production processes at the facility that did not use the PFNA blend and worked in different buildings at the plant site. The final classification consisted of three exposure categories: high exposure (process job titles); low exposure (finishing, supervisory, lab, clerk, maintenance, and technical job titles); and no exposure (other processes, administrative job titles).
A variable representing the highest exposure level achieved by a worker was created using the final three surfactant exposure categories. Assignment to the exposure categories was hierarchical and mutually exclusive at any particular point in time. That is, once an employee was in a “high” exposure process job, his or her exposure was considered “high” for that year, and subsequent years. It is possible for an employee assigned to “low” or “no exposure” to move to a higher exposure group over time in the analysis, but not the reverse.
Before the start of the epidemiological study, the company had obtained blood samples from a subset of current employees who worked in various areas of the plant, to ascertain whether PFNA levels could be detected in the blood. These limited biomonitoring results were used to validate the exposure categories generated based on the occupational history, but were insufficient to be used in any analyses.
Three main analyses were conducted. First, a cross‐sectional analysis was conducted to evaluate differences in average values of all 32 clinical laboratory measures at five points in time. Additional annual cross‐sectional analyses of mean lab values by exposure groups and longitudinal analysis accounting for multiple measurements per person were also conducted. Analyses used the three exposure groups for men and two groups for women, combining “high” and “low” because few women worked in the process area historically. All analyses were adjusted for age and body mass index (BMI), used as continuous variables. All analyses were conducted in SAS 8.0 (SAS Institute, Cary, NC, USA).
Five time periods were selected for cross‐sectional analysis based on the following: (a) the earliest year (1976) sufficient clinical chemistry data were available for analysis; (b) the year (1989) complete medical records became available from the facility; (c) the year (1995) a liquid solution started to replace the powder form of surfactant; (d) the year (1998) following complete conversion of surfactant from powder to solution; (e) the latest year (2001) with sufficient data for analysis. Each time period, or analysis “window” encompassed a two‐year period around the five key dates in order to capture the annual physical examination for each employee, and to reflect different exposure potential. For example, the 1989 period is believed to be indicative of the entire period when powdered surfactant blend (with potential for airborne dust exposure) was used, whereas the 1995 period would reflect lower potential airborne dust exposures. Many employees contribute data in multiple analysis windows. Laboratory test results dated closest to the mid‐point of each analysis window were evaluated by exposure groups. Pairwise comparisons of adjusted means of each lab test were made across exposure groups in each analysis window using PROC GLM in SAS.
Based on findings in rats that liver enzymes and blood lipids may be the most likely to show any effects from exposure, adjusted annual means for these tests results by exposure group were graphed separately for men and women, using all years with data available. As with the analysis windows, each consecutive year does not represent the exact same group of individuals, because of the dynamic workforce, with substantial overlap from year to year.
Data used in the longitudinal analysis included annual measures of liver enzymes and blood lipids, age at entry into the cohort, annual measures of BMI, a weighted cumulative intensity score (up to the month before the annual exam) and the exposure group (none or any) in the month before the annual exam. A variable reflecting cumulative exposure to surfactant was also created. Each subject was assigned an annual weighted cumulative intensity score, where exposure intensity was quantified as 0 while not exposed, 1 during low exposure, and 2 during high exposure. Weighting was based on the proportion of each year spent in each exposure category. In addition, an indicator variable was created for powder or liquid surfactant use at the time of the annual exam. The general approach to the modelling was a mixed or random effects model. All longitudinal analyses used PROC MIXED in SAS.