The study was approved by the Institutional Review Board of the Harvard School of Public Health (Boston, MA, USA). Written informed consent was obtained from each participant. The study population consisted of 36 boilermakers. Using a repeated measures panel study, participants were recruited and monitored at an apprentice welding school over four sampling periods: January/February 2004, February 2005, June 2005, and January/February 2006. In 2004, subjects were invited to participate as either welders or controls, whereas in 2006, participants who welded were invited to participate on a second weekend as controls. Participants who had participated in previous years also had the opportunity to participate again on subsequent study weekends. Of the 36 unique participants, 19 participated in more than one sampling period as a control or welder (13 participated twice, four participated three times and two participated in all four sampling periods). In all, 63 multiple measurements were taken in sets of pre- and postwelding blood samples. Forty-seven of these were collected before and after exposure to metal fume and airborne fine particulate matter through electric manual metal arc welding and metal inert gas welding on mild and stainless steel. The majority of the welding was performed on mild steel. Sixteen sets were from participants who served as controls and did not weld. Identical testing was performed on the controls, who spent the study period reading and watching movies in the union hall proper and were exposed primarily to background levels of particulate matter in the union hall. Of the 63 observation sets, 14 were collected from individuals who had welded the day prior to the study onset. A modified American Thoracic Society standardized respiratory questionnaire was used to collect information on medical history, smoking history, and occupational history of the participants.

The study took place in the union hall training facility of the Boilermakers Local 29 in Quincy, MA. Participants arrived on Saturday morning around 7:30 am at which time preshift health monitoring was performed. On welding days, the participants welded for about 6 hours, including a 30-minute lunch break. Participants returned the following morning for repeat health monitoring.

Personal particle samplers were placed in the breathing zone of participants on their lapels, outside of the helmet, during their work or control day to monitor exposure to the particles with an aerodynamic mass median diameter <2.5 μm (PM_2.5). The KTL cyclone (GK2.05SH, BGI Incorporated, Waltham, MA, USA) with a 50% aerodynamic diameter cutpoint of 2.5 μm was used in line with a Vortex Timer 2 personal sampling pump (Casella USA, Amherst, NH, USA) calibrated at a flow rate of 3.5 l/min. The air sample was collected on a 37 mm polytetrafluoroethylene membrane filter (Gelman Laboratories, Ann Arbor, MI, USA) encased in a cassette and placed downstream of the cyclone. The filters were weighed before and after sampling on a MT5 microbalance from Mettler-Toledo Incorporated (Columbus, OH, USA) after equilibrating for a minimum of 24 hours in a temperature and humidity controlled room. The mass collected on the filter was divided by the air volume sampled to calculate the gravimetric PM_2.5 concentration.

Fasting venous blood samples were collected on both mornings of the study period (preshift and the next morning) and were analyzed for fasting cholesterol, low-density lipoprotein, high-density lipoprotein, and triglycerides. Participants reported the time of their last meal at the time of blood collection. Samples were considered to be fasting if the reported time of the last meal or beverage was more than 9 hours prior to sample collection. In the few cases where the participant did not report the time of his last meal, nonfasting status was presumed. Of 126 total baseline and next morning blood draws, 96 samples were reported as fasting, 24 samples were nonfasting and eight did not identify the time of the last meal and were presumed to be nonfasting.

Statistical analyses were performed using SAS version 9.1 (SAS Institute Incorporated, Cary, NC, USA). Exposure was investigated as both a dichotomized variable (nonexposed controls and welders) as well as a continuous variable using PM_2.5 concentration. The mean (standard deviation, SD) and median values of the PM_2.5 concentrations were determined for controls and welders. A two-sample t test was performed to compare the PM_2.5 concentrations in controls and welders.

Baseline and next morning mean (standard deviation) levels of total cholesterol, LDL, HDL, and triglycerides were determined by exposure status. To account for repeated measurements, linear mixed effects models were used to investigate the effect of welding fume exposure and PM_2.5 concentration on next morning lipid levels after adjusting for baseline levels. In a repeated measures study design, each participant serves as his own control by providing baseline lipid levels. This design has the advantage of inherently controlling for factors that may be predictive of lipid levels such as exercise, alcohol consumption, and family history of hyperlipidemia. In the adjusted model, data was additionally adjusted for age, statin use, fasting, smoking status, and welding the previous day. A similar model was used to investigate the linear exposure-response association between PM_2.5 concentrations as a continuous variable and lipid levels. Cook’s Distance <0.5 was used as the criterion for assessing outlying residuals and influential data points. The level of significance for all analyses was set at α = 0.05.

The study population demographic data are presented in Table 1. The study population included 36 men, of whom 34 served at least once as a welder, 14 served at least once as a welder and later as a non-welding control, and two served once as a control only. Upon first entry into the study, 34 participated as welders and two as controls. Participant ages ranged from 24 to 63 years, with a mean age of 41 years (SD 11). Most participants (29 of 36) reported white race. Participants had a median of 5 years of boilermaker experience, ranging from 0 to 40 years. Twelve participants were current smokers, 10 were ex-smokers, and 14 had never smoked. In general, the study population consisted of relatively young, healthy workers, with few having underlying respiratory or cardiopulmonary diseases. Nine participants reported a history of asthma, hypertension, heart disease and/or diabetes (one reported both heart disease and hypertension, the others reported only one condition). Three participants were taking a cholesterol-lowering medication in the HMG-CoA reductase inhibitor (statin) class.

Particle samples were collected from all controls and welders during the workday. Personal particulate matter samples were collected for a mean sampling time of 6.1 hours (SD 1.5). Measured PM_2.5 concentration ranged from 0.01 mg/m³ to 2.97 mg/m³. The median PM_2.5 concentration was 0.05 mg/m³ in controls and 0.65 mg/m³ in welders. As expected, the PM_2.5 exposure of welders was significantly higher than controls (P = <0.001).

The composition of the PM_2.5 welding fumes has been reported in a previous study of this cohort during the same study period.²¹ Using x-ray fluorescence to analyze the elemental content of the PM_2.5, the highest exposures were to iron, manganese, copper, zinc, sodium, potassium, silica, calcium, and sulfur.

The mean effects (95% CI) of welding fume exposure on lipid levels the following morning are shown in Table 2. In the crude model, there was a trend (P = 0.08) toward decrease in HDL levels of 2.3 mg/dL associated with welding, after adjusting for baseline HDL (95% Confidence Interval [CI]: −4.9 to 0.3). This effect was strengthened to a 2.6 mg/dL significant decrease (95% CI: −5.3 to 0.0, P = 0.05) after additionally adjusting for age, fasting status, use of a statin, current smoking, and welding the previous day. Restricting the analysis to fasting samples resulted in a similar effect size on HDL but with a wider confidence interval. Only baseline HDL levels and statin use were additional predictors of next day HDL (P < 0.001) in the multivariable model. There were no significant changes in the other lipid levels associated with welding in the crude or adjusted model. Some of the subjects (n = 14) performed welding the day before participating in the study. In these subjects, baseline lipid measurements may have been affected by the prior day’s welding exposure and would be equivalent to an 18-hour postwelding lipid measurement. To account for this, we also performed a separate analysis on the 48 sets of blood samples from those who did not weld the day prior (one welder did not specify previous day welding status and was assumed to have welded the day before). The subgroup that welded the day before study onset was too small for a statistically meaningful subgroup analysis. In the subgroup (n = 48) who did not weld the previous day, there was a significant decrease in next morning HDL of 4.3 mg/dL (95% CI: −8.0 to −0.7, P = 0.02) after adjusting for age, fasting status, use of a statin, and current smoking. The results of the subgroup analysis are shown in Table 3.

The association between PM_2.5 concentrations and lipid levels was also examined. There were no statistically significant effects of PM_2.5 concentrations on any of the lipid levels (P ≥ 0.10).