Seventy-three companies potentially making or using BPA were identified from the 2010 (n = 66) and 2011 (n = 2) US Environmental Protection Agency (EPA) Toxic Release Inventory (USEPA, 2016a) and by referral (n = 5). Among the 73 companies, 15 did not respond; 15 no longer produced or used BPA; 37 either had few workers handling BPA, infrequent BPA use, or could not be scheduled within the study period; and 6 participated in the study. We visited participating companies to identify BPA-related jobs and invited workers performing these jobs to participate in the study. This study was approved by the NIOSH Institutional Review Board. Participants gave written informed consent and were reimbursed $70 for the time and inconvenience of providing samples.

Participants were sampled over two consecutive work days after having been scheduled to be off work for at least 24 h. On Day 1, participants collected pre-shift, mid-shift (±30 min of participant’s shift mid-point), endshift, and post-shift (4–6 h after leaving work) urine samples. On Day 2, participants collected pre-shift, mid shift, and end-shift samples for a total of seven samples (Time Points 1–7). To collect samples, participants were given an insulated bag containing a sterile, 120-ml polypropylene specimen cup (Samco™) in a ‘BPA-free’ Ziploc ® or Glad® plastic bag, a large Kimwipe™ towel (Kimtech Science) pre-screened to be BPA-free, and frozen refrigerant packs for post-shift samples. Participants were instructed to wash hands with water only (to avoid potential interferences), dry hands with the provided towel, place the cup cap in the plastic bag to prevent contamination, collect the urine, write the void date and time on the cup label, and place the sample in the plastic bag.

For sample processing, study staff donned nitrile gloves (Kimberly-Clark), swirled the cup to mix the urine, measured specific gravity (SG) using a handheld refractometer (Atago® USA, Inc.) calibrated with distilled water, and aliquoted the urine into 5-ml Nalgene® polypropylene cryovials (ThermoFisher Scientific, Inc.) using sterile 5.8-ml Fisherbrand™ polyethylene transfer pipets (ThermoFisher Scientific, Inc.).

Quality control (QC) field blanks (FB) (two types) and blind duplicates of participant samples were collected. For the first FB (FB1), study staff took a kit to the bathroom used by participants, followed collection instructions and filled the specimen cup with 60 ml of Optima® LC/MS water (ThermoFisher Scientific, Inc.). For the second FB (FB2), the cup was filled with 60 ml of Optima® LC/MS water where samples were processed. FB and duplicate samples were aliquoted as described above. Cryovials were placed and shipped on dry ice, then stored at −80°C. The laboratory was blind to all participant information. The mean (±SD) duration from sample void time to freeze time was 2.5 ± 3.6 h (range 0.07–18 h), within the stability period for BPA-G in urine at room temperature (Waechter et al., 2007; Ye et al., 2007).

We quantified urinary concentrations of free and total (free plus conjugated) BPA by online solid phase extraction-high performance liquid chromatography-isotope dilution tandem mass spectrometry (Zhou et al., 2014). Each analytical run included calibration standards, reagent blanks, and high and low concentration QC urine materials. The limit of detection (LOD) was 0.1 μg l⁻¹. The mean (±SD) relative difference in the blind duplicates (n = 26) was 6.7 ± 4.7% (total BPA) and 5.9 ± 6.7% (free BPA). The mean (±SD) coefficient of variation (CV) for total BPA in QC materials (n = 83 each) was 6.4 ± 2.1% (QC low) and 3.7 ± 0.9% (QC high). For FB1 (n = 17), we detected total and free BPA in one (but different) blank each; for FB2 (n = 26), we detected total BPA in two blanks. Blank concentrations were at or near the LOD; therefore, we did not FB-correct BPA concentrations. Urine samples were analyzed within 1–5 months of collection, within BPA-G’s stability period when frozen (Ye et al., 2007).

We measured urinary creatinine using a Vitros® 250 Chemistry Analyzer (Ortho-Clinical Diagnostics). Each run included calibration standards and high and low concentration pooled urine QC. The lower reportable limit was 1.7 mg dl⁻¹. The mean (±SD) relative difference for blind splits was 4.2 ± 7.0% (n = 26). The mean (±SD) CV across five runs was 2.7 ± 1.7% (QC low) and 3.2 ± 2.2% (QC high). To adjust for urine dilution, BPA concentrations were divided by creatinine (units = μg g⁻¹ creatinine) or multiplied by (1.024 − 1)/(SG − 1) (units = μg l⁻¹) (Elkins et al., 1974).

For each participant, we collected information on sex, age, race/ethnicity, self-reported weight and height, job, shift duration, current smoking (yes/no), hours since last worked, and jobs worked during days off. We asked participants about the number of canned beverages and canned foods consumed in the past 24 h, dental procedures in the past 3 days (Day 1) and past 24 h (Day 2), clothing and personal protective equipment worn during work, BPA-related tasks performed, hand washing frequency during work, BPA spills, hand-to-mouth behaviors during work, changing clothes at shift end, and showering/bathing between Day 1 and Day 2.

We grouped companies into five industry categories: (i) phenolic resin manufacturing (Companies 1 and 2), (ii) BPA and polycarbonate manufacturing (Company 3), (iii) BPA-filled wax manufacturing/reclaim (Company 4), (iv) BPA manufacturing (Company 5), and (v) BPA-filled wax manufacturing/investment casting (Company 6). We assigned each worker’s job into one of seven job categories: (i) making BPA, (ii) kettle/reaction/field operator making a BPA-based resin (hereafter referred to as ‘kettle operator’), (iii) operator flaking a BPA-based resin, (iv) maintenance work in a BPA or resin manufacturing area, (v) making BPA-filled wax, (vi) working with solid BPA-filled wax (e.g. wax injection, pattern and mold assembly), and (viii) working with molten BPA-filled wax (e.g. wax reclaim, melt/burnout of BPA-filled wax from shells/molds). While we attempted to create job categories having similar tasks, some task variation occurred within jobs, usually because we did not have sufficient sample size to create a separate job(s).

We detected total BPA in all urine samples; free BPA in 71% of them. We assigned LOD/2 to free-BPA concentrations <LOD (Hornung and Reed, 1990). We used the ratio of free to total BPA to indicate possible exogenous BPA contamination. We excluded one participant from analyses who had two samples with >20% free BPA, a percentage used previously to suggest BPA contamination (Guidry et al., 2015; Thayer et al., 2016).

BPA concentrations were skewed to the right; therefore, a natural log transformation was applied. We computed summary statistics for total and free BPA, and for the ratio of free to total BPA. We split the data into two groups, baseline (Time Point 1) and post-baseline (Time Points 2–7). For both groups, we compared geometric means (GM) of creatinine-adjusted total BPA (total BPA_CR) by industry and by job using the PROC MIXED procedure in SAS; for post-baseline samples, we used a first-order autoregressive covariance structure [AR(1)]. We used Tukey’s method to adjust P-values for multiple comparisons.

We conducted separate linear regression modeling for baseline and post-baseline samples using the natural log of total BPA_CR [ln(total BPA_CR)] as the dependent variable. For the baseline analysis, we initially examined one-at-a-time the effect of age, body mass index (BMI), total hours off work before collecting the first sample, number of canned beverages and canned foods consumed in the past 24 h (separately and combined), current smoking, job, and industry on ln(total BPA_CR). Covariates with a P-value ≤ 0.2 were evaluated in a multiple linear regression model.

For the post-baseline analysis, we evaluated 22 covariates as potential predictors of total BPA_CR. These covariates included job, industry, personal protective equipment and clothing worn (four covariates), hygiene behaviors (six covariates), work tasks (seven covariates), age, number of hours away from work before collecting the first sample, and current smoking. Where covariate responses could vary between Day 1 and Day 2, responses for Day 1 were assigned to Time Points 2–5 and responses for Day 2 to Time Points 6 and 7. We initially examined each covariate one-at-a-time after adjusting for time point, ln(total BPA_CR) at baseline, and BMI. Covariates with a P-value ≤ 0.2 were included in a stepwise forward selection regression model with worker treated as a random effect, covariates as fixed effects and an AR(1) covariance structure. Covariates were entered into the model until all remaining covariates had P-values > 0.05.

In separate linear regression models, we examined the effect of BMI on ln(total BPA_CR) at Time Points 2–7 (separately) after adjusting for age and ln(total BPA_CR) at baseline. To test for a difference in total BPA_CR between Day 1 and Day 2, we used each person’s averages of ln(total BPA_CR) at Time Points 2 and 3 and 6 and 7 in a mixed model with person as a random effect. These four time points were selected to obtain comparable data between the 2 days.

Baseline and post-baseline regression models were rerun with ln(creatinine) as an independent variable instead of correcting BPA concentrations for creatinine in the dependent variable. Statistical analyses were performed in SAS v. 9.3 (SAS Institute, Inc.). Significance testing was done at α = 0.05.

Of 199 eligible workers at six companies (average 33 workers/company, range 25–55), 125 consented to participate (average consent rate of 63%, range 51–75%/company). We were able to schedule 78 (62%) of consenting workers for sampling; 77 remained after excluding one participant with possible BPA sample contamination. Two days of sampling were completed for 74 workers and 1 day for 3 workers for a total of 151 worker-days. All but one of the 77 workers were male, and 89.6% were white (Table 1). Median age was 44 years (range 20–63 years). Median BMI was 29.8 kg m⁻² (range 21.0–44.3 kg m⁻²).

Companies 1 and 2 added BPA and other ingredients to kettles, solidified molten resin, and converted it into a flake product. Participant jobs were kettle and flaker operators. Company 3 made BPA from acetone and phenol, then reacted BPA with phosgene to make polycarbonate resin. Participant jobs included operators, shift leads, and maintenance. Company 4 added BPA and other ingredients to kettles, then solidified molten wax into pastilles, billets, or slabs. Company 4 also reclaimed the wax component from used wax using large hot boxes at ~100°C to melt the wax, followed by removal of water and non-wax solids. Participant jobs included warehouse, wax preparation, blending and packaging, wax reclaim, and QC. Company 5 made BPA from acetone and phenol and then transferred BPA to the epoxy resin manufacturing unit via a closed system. Participant jobs included BPA operators, flakers, and loaders. Company 6 added BPA and other ingredients to kettles, solidified molten wax into pastilles, and used the wax in investment casting. Participant jobs included making wax, engineer, shift lead, lift-truck driver, wax injection, wax pattern/mold assembly, and wax removal from shells/molds in heated Boilerclaves® at ~170°C followed by burnout in ovens at ~1000°C. All companies adding BPA to reaction or mixing kettles used a mix of manual/partly manual methods (e.g. emptying bags or bulk sacks into addition hoppers) and automated methods (e.g. transporting BPA through enclosed systems). Specific addition methods and BPA form (prill or flake) were generally proprietary.

The GM concentration of total BPA_CR in 525 samples over seven time points was 88.0 μg g⁻¹, ~70 times higher than in adults ≥20 years in the 2013–2014 US National Health and Nutrition Examination Survey (NHANES) (GM = 1.27 μg g⁻¹) (NHANES, 2016) (Table 2). At baseline, the GM total BPA_CR (26.6 μg g⁻¹) was 20 times higher for adults in NHANES 2013–2014.

On average, total BPA_CR increased from pre-shift to post-shift on Day 1 (pre-shift GM = 26.6 μg g⁻¹, midshift GM = 60.7 μg g⁻¹, end-shift GM = 115 μg g⁻¹, postshift GM = 127 μg g⁻¹), decreased between post-shift Day 1 and pre-shift Day 2 (GM = 84.4 μg g⁻¹) without returning to Day 1 pre-shift levels, and then increased during Day 2 (mid-shift GM = 125 μg g⁻¹, end-shift GM = 178 μg g⁻¹) (Table 2, Fig. 1). Pre-shift Day 1 GM total BPA_CR was significantly lower than each of the other time points (P < 0.0001 each, Table 2). Differences in total BPA_CR between many other time points were also statistically significant (Table 2). Although total BPA_CR post-shift Day 1 (GM = 127 μg g⁻¹) was higher than end-shift Day 1 (GM = 115 μg g⁻¹), the difference was not statistically significant. The highest concentration measured (18 900 μg g⁻¹) was more than three orders of magnitude higher than the 95th percentile of adults from NHANES 2013–2014 (5.09 μg g⁻¹) (NHANES, 2016). We averaged total BPA_CR for Time Points 2–5 for 74 participants with all four samples to obtain an approximate ‘24-h’ total BPA_CR concentration for Day 1 (GM = 96.6 μg g⁻¹, range 3.05–7890 μg g⁻¹).

Total BPA_CR by industry and job are presented in Table 3; Figs 2 and 3; and Supplementary Figs S1 and S2, available at Annals of Work Exposures and Health online, and by time point within industry and job in Supplementary Table S1, available at Annals of Work Exposures and Health online. All industries and jobs had GM total BPA_CR concentrations significantly higher than adults in NHANES 2013–2014 (Supplementary Table S2, available at Annals of Work Exposures and Health online). By industry, at baseline, total BPA_CR was highest in BPA-filled wax manufacturing/reclaim (GM = 111 μg g⁻¹), followed by BPA/polycarbonate manufacturing (GM = 69.4 μg g⁻¹) and BPA manufacturing (GM = 37.4 μg g⁻¹), and lowest in phenolic resin manufacturing (GM = 6.56 μg g⁻¹). Compared to phenolic resin manufacturing, total BPA_CR at baseline was significantly higher in BPA-filled wax manufacturing/reclaim (P < 0.0001), BPA/polycarbonate manufacturing (P < 0.0001), and BPA manufacturing (P = 0.0257). Post-baseline, total BPA_CR was also highest in BPA-filled wax manufacturing/reclaim (GM = 121 μg g⁻¹), followed by BPA/polycarbonate manufacturing (GM = 218 μg g⁻¹) and BPA manufacturing (GM = 121 μg g⁻¹), and lowest in phenolic resin manufacturing (GM = 33.8 μg g⁻¹). Compared to phenolic resin manufacturing, total BPA_CR post-baseline was significantly higher in BPA-filled wax manufacturing/reclaim (P < 0.0001) and BPA/polycarbonate manufacturing (P = 0.0003).

By job, total BPA_CR at baseline was highest for maintenance (all at Company 3, GM = 157 μg g⁻¹), followed by working with molten BPA-filled wax (GM = 94.9 μg g⁻¹), and lowest for flaking a BPA-based resin (GM = 4.81 μg g⁻¹). Total BPA_CR at baseline in maintenance was 30 times higher than flaking a BPA-based resin (P < 0.0001). Total BPA_CR post-baseline was highest for the job of working with molten BPA-filled wax (GM = 441 μg g⁻¹), followed by making BPA-filled wax (GM = 254 μg g⁻¹), and making BPA (GM = 192 μg g⁻¹), while lowest for flaking a BPA-based resin (GM = 23.2 μg g⁻¹). Compared to the BPA-based resin flaking job, total BPA_CR post-baseline was significantly higher in those working with molten BPA-filled wax (P = 0.0027), in those making BPA-filled wax (P = 0.0019), and in those making BPA (P = 0.013). The maintenance job showed little difference in total BPA_CR concentrations at baseline (GM = 157 μg g⁻¹) and post-baseline (GM = 156 μg g⁻¹); all other jobs showed an increase, although not always significant. Total BPA_CR in the maintenance job also changed minimally across time points as compared to other jobs (Supplementary Table S1, available at Annals of Work Exposures and Health online).

For free BPA_CR, the GM was 0.43 μg g⁻¹ (range <LOD–62.9 μg g⁻¹, n = 525), with GMs for each of the seven time points <1 μg g⁻¹ (Table 2). For percent free BPA, the GM was 0.49 (75th percentile = 0.9%) (Table 2). SG-adjusted and volume-based (unadjusted) total- and free-BPA concentrations are summarized in Supplementary Tables S3 and S4, available at Annals of Work Exposures and Health online.

Covariate responses are summarized in Tables 1 and 4. Approximately 84% of the participants were off work ≥24 h before collecting the first sample; 72% for ≥48 h (range 11.5–252 h). Twelve participants (15.6%) were off work <24 h due to schedule changes, 7 of these worked in maintenance at Company 3. On 133 (88%) out of 151 days, participants wore either chemical-resistant gloves (66 days) or fabric/leather gloves (67 days). Respirator use was less common, 29 days (19.2%). Tyvek® coveralls were worn on 19 days (12.6%); 15 of these days at Company 1. Among hand-to-mouth activities queried (Table 4), eating during work (excluding lunch and breaks) was most frequent (91 days, 60.3%). Participants reported washing their hands during work one to four times on 43 days (28.4%), five to seven times on 68 days (45%), and more than seven times on 40 days (26.5%). None of the participants had dental work performed or worked at jobs associated with BPA exposure during their time off before collecting samples. All but three participants showered/bathed between Day 1 and Day 2.

We created several cross-company variables (Table 4). On 80% of the days, participants worked >8 h; on 70% of the days participants spent ≥50% of their shift in production areas versus control rooms/offices. On 18 days (12%), participants handled one or more containers of BPA (bulk sacks, bags, drums, or buckets). On 15 days (9.9%), participants handled one or more empty BPA containers. Taking process or bulk samples was dichotomized into less than three samples (140 days, 92.7%) or more than three samples (11 days, 7.3%). Because of small numbers, we could not examine process sample types separately. Participants spilled BPA on 10 days (6.6%) or cleaned up a BPA spill on 12 days (8.0%).

In univariate analyses where ln(total BPA_CR) at baseline was the dependent variable, job (P < 0.0001), industry (P < 0.0001), total hours off work before collecting the first sample (P = 0.0269), and current smoker (P = 0.0813) had P-values ≤ 0.2 (Table 5). When these covariates were included in a multiple regression model, only industry remained significant (P = 0.0248, data not shown).

We had an a priori interest in the relationship between BMI and total BPA_CR. After adjusting for age and ln(total BPA_CR) at baseline, we observed significant positive associations between BMI and ln(total BPA_CR) at Time Points 3 (P = 0.0448), 4 (P = 0.0173), 5 (P = 0.0045), and 7 (P = 0.0075), and borderline significance at Time Point 6 (P = 0.0610) (Supplementary Table S5, available at Annals of Work Exposures and Health online). Therefore, we adjusted for BMI in post-baseline models.

When ln(total BPA_CR) post-baseline was the dependent variable, 10 covariates with P-values ≤ 0.2 in univariate analyses (Table 5) were included in a stepwise forward selection model (Table 6). After adjusting for ln(total BPA_CR) at baseline, time point and BMI, total BPA_CR was positively associated with handling containers of BPA (1.8 times, P = 0.0035), taking more than three process/bulk samples (2.33 times, P = 0.0002) and wearing a Tyvek® coverall (1.93-times, P = 0.0042). Stepwise results for SG-adjusted total BPA were generally similar (Supplementary Table S6, available at Annals of Work Exposures and Health online).

We found a significant 64% increase in total BPA_CR from Day 1 to Day 2 when comparing the average of Time Points 6 and 7 (GM = 152 μg g⁻¹) to the average of Time Points 2 and 3 (GM = 92.5 μg g⁻¹) (P < 0.0001, n = 72). We did not find a significant interaction between day and industry (P = 0.28) or between day and job (P = 0.60), although total BPA_CR increased from Day 1 to Day 2 in all industries (1.3–2.0 times) and jobs (1.2–1.9 times) except maintenance (1.0 times) (data not shown).

Regression models that included ln(creatinine) as a covariate gave similar results, including coefficients, standard errors, and P-values (Supplementary Tables S7–S9, available at Annals of Work Exposures and Health online).

Absent human data on BPA elimination following inhalation or dermal exposures, we applied a simplified approach to estimate a participant’s 24-h BPA intake on Day 1 [equation (1)]. Specifically, we used the average total BPA_CR concentrations at Time Points 2–5, the average creatinine concentration at the four times, imputed values for 24-h urine volume representing the mean (1.2 l) and range (0.6–2.0 l) for adults (Wallach, 2007), and self-reported body weight. The GM (range) estimated BPA intake for Day 1 was 0.88 μg kg⁻¹ day (0.035–73.9 μg kg⁻¹ day, n = 74) for a urine volume of 0.6 l, 1.77 μg kg⁻¹ day (0.069–148 μg kg⁻¹ day) for 1.2 l, and 2.95 μg kg⁻¹ day (0.12–246 μg kg⁻¹ day) for 2.0 l (Supplementary Table S10, available at Annals of Work Exposures and Health online). Based on these estimates, 1.4% of the participants at a urine volume of 0.6 l, 2.7% at 1.2 l, and 8.1% at 2.0 l exceeded the US EPA oral Reference Dose for BPA of 50 μg kg⁻¹ day (USEPA, 2016b). Exceedance fractions were higher when comparing estimates to the European Food Safety Authority temporary Tolerable Daily Intake of 4 μg kg⁻¹ day (EFSA, 2015), 20.3% at a urine volume of 0.6 l; 28.4% at 1.2 l, and 41.6% at 2.0 l. Single-day estimates may not represent a worker’s lifetime exposure. Also, different approaches were used by EFSA (forward modeling of external exposure), EPA [lowest observable adverse effect level], and our study (backward modeling from biomonitoring data) to derive reference levels and intake estimates.

where bw = body weight.