Personal air samples were collected from 47 automotive spray painters in this study, thereby providing estimates of BZCs of monomeric and polymeric HDI in the automotive refinishing industry. Using quantitative inhalation exposure and covariate data in LMM, we identified the primary determinants of BZCs of HDI, uretidone, biuret, and isocyanurate. However, there were several limitations in this study that were taken into consideration when we evaluated the results.
First, although IPDI may be an important constituent of some automotive coatings, we did not analyze for IPDI or its oligomers in this study. Thus, our estimates of exposure do not capture all polyisocyanates of concern. Second, because the one- and two-stage samples were collected at different times during the study, they are not directly comparable. Last, >40% of the uretidone and biuret data were below detection limits. Consequently, statistical power was reduced for the analyses of the uretidone and biuret data. However, multiple imputation was used for BZCs below the detection limits in an effort to obtain better estimates of the parameters of interest (Lubin et al., 2004
The mixed models developed in this study described more than half the variability in BZCs of uretidone and biuret (R2
50%) and lesser variability in BZCs of HDI and isocyanurate (R2
20%). Marginal R2
statistics calculated for models specific to each booth type and analyte (data not shown) were highly variable, ranging from 0.05 for isocyanurate exposure in semi-downdraft booths to 0.79 for biuret exposure in semi-downdraft booths. Low R2
values (i.e. <0.20) may partially reflect the lack of between-worker variability in the respective exposure distributions (between-worker variability is generally easier to characterize than within-worker variability). Nevertheless, the large range of marginal R2
values among the different booth types for analyte-specific models suggest that the processes governing BZCs are different for the different booth types. Thus, classification of the mixed models by analyte and booth type is appropriate.
Although unique models were built for each measured polyisocyanate, analyte-specific paint concentration, airflow, and sampler type were significant predictors in three or more of the models. In addition, the interaction between paint concentration and airflow was significant in the biuret and isocyanurate models, suggesting that the relationship between paint concentration and BZC depends on the airflow in the booth. Expectations are that higher analyte-specific paint concentrations will lead to higher BZCs while increased airflow will lead to lower BZCs of each polyisocyanate. This was observed in the model predictions in which changing analyte-specific paint concentration and airflow were evaluated. Unexpectedly, using the same analyte-specific paint concentration, the models predicted higher BZCs of isocyanurate than any of the other analytes (). The most likely reason for this finding is that the isocyanurate model overpredicted BZCs at lower levels of paint concentrations due to the fact that 90% of the paint concentrations used to generate the model were between 10
900 and 195
000 mg l−1
. In fact, significant differences between predicted means and actual values were observed when the actual values were below the 10% quantile (~270 μg m−3
), and it was evident that paint concentration <10
000 mg l−1
had a negligible effect on isocyanurate BZC (). Despite this limitation, the isocyanurate model performed well in terms of prediction (i.e. 90% of the predictions within ±2 scaled residuals) and, therefore, provides reasonable estimates of central tendency for most of the data collected in this study. We believe that the predictions of oligomer BZCs at respective paint concentrations of 10
000 mg l−1
by these models are valid and, as such, demonstrate the differences between isocyanurate and the other oligomer concentrations in these spray painting operations ().
Another possibility for these differences is that the analysis of paint samples underestimated the true concentration of isocyanurate in the paint. The interquartile range of isocyanurate paint concentration was 45
000 to 135
000 mg l−1
, representing ~3.0 to 8.5% of the paint formulation. According to the material safety data sheets (PPG, 2009a,b,c) of some of the most common hardeners used in the workplace and assuming a hardener to coating ratio of 4:1, the proportion of polymeric HDI is expected to range from 2.5 to 20%. Thus, measurements of isocyanurate in paint are within the expected range.
Assuming that the mixed models accurately represent conditions in the atmosphere, differences in reactivity could explain the differences in the predicted mean BZCs (). For example, HDI, uretidone, and biuret may polymerize more rapidly in the atmosphere than isocyanurate. In contrast, isocyanurate could react more rapidly and completely with the derivatizing agent than the other polyisocyanates. Moreover, isocyanurate may be formed during the polymerization process as other polyisocyanates react with each other. The significant effect of temperature in the uretidone model may be indicative of increasing reactivity with increasing temperature. In fact, uretidone may be the most reactive polyisocyanate measured in this study due to the strained configuration of its four-member ring.
Reactivity of polyisocyanates is probably the reason why the effect of sampler type was significant in the HDI, uretidone, and isocyanurate models. Two-stage samplers may underestimate BZCs of reactive polyisocyanates due to polymerization of polyisocyanates on the untreated pre-filter of the sampler. This problem may be avoided by using one-stage samplers onto which all polyisocyanates are simultaneously collected and derivatized on one filter. Thus, higher BZCs may be measured when one-stage samplers are used instead of two-stage samplers.
Because two-stage samplers were used for the first part of this study and one-stage samplers were used for the second part of this study, observed differences could be due to variable differences between the first and second parts of the study (i.e. selection bias). However, we controlled for many of these variables (airflow, temperature, etc.) by including them in the LMM. Furthermore, the significant effect of sampler type was corroborated by paired two-sample t
-tests comparing side-by-side sets of one- and two-stage samplers in which significant differences were observed for the analytes HDI (P
0.0363) and isocyanurate (P
0.0035). Thus, it is likely that the observed differences between one- and two-stage sampling results for HDI and isocyanurate are due to differences in the design of the samplers. Significant differences were not observed between one- and two-stage sampling for uretidone and biuret. It is possible that significant differences were not detected because of insufficient statistical power due to the large proportion of non-detectable measurements or because of reactivity differences among the different polyisocyanates. Further investigation is needed to evaluate the reactivity of the different polyisocyanates in the painting atmosphere and implications of this reactivity on human health endpoints such as tissue absorption and respiratory sensitization.
Comparison of one- and two-stage sampling results in NC () seems to contradict our findings that suggest a two-stage sampling bias. However, only 19 one-stage samples were collected in NC. In addition to the lack of statistical power, the three shops that were sampled with one-stage samplers may not be representative of all the automotive repair shops in the study. It is possible that the paint being used in these three shops was rapidly curing and thereby resulted in lower measured concentrations due to polymerization on the air filters. Even one-stage samplers can underestimate concentrations in atmospheres containing rapidly curing isocyanates (Streicher et al., 2000
). Furthermore, these three shops appeared to have been better controlled than the other shops in NC. Downdraft booths were used in 18 of the 20 paint tasks with an average airflow of 261 m3
compared to the other shops in NC where downdraft booths were used in 39 of 55 paint tasks with an average airflow of 204 m3
. Moreover, the one-stage samplers in question were collected simultaneously with the two-stage samplers and these two-stage samplers measured significantly lower concentrations of HDI and isocyanurate.
A time-dependent decrease in sampling efficiency has been shown for long-term filter sampling of atmospheres containing toluene diisocyanate (Sennbro et al., 2004
). The variables, paint time, and total time (or sampling time) were evaluated in the mixed models. Total time was not significant in any of the models and paint time was significant only in the uretidone model. If uretidone was polymerizing over time, we would expect paint time to have a negative parameter estimate, but instead, paint time had a positive parameter estimate. The significant effect of paint time may simply represent the accumulation of uretidone aerosols in the atmosphere over time. Because total time (or sampling time) was not significant in the models, a time-dependent decrease in sampling efficiency for these short-term samples either did not exist or was negligible. Another possibility is that the effect of sampling time was unobservable (i.e. polymerization happened immediately) for rapidly curing polyisocyanates.
Experience (i.e. years painting) was a significant variable in both the biuret and the isocyanurate models in which more experience was associated with lesser exposure. Flynn et al. (1999)
found that the painter orientation relative to the direction of the airflow played a significant role in affecting BZCs in crossdraft booths. It is likely that more experienced painters received less exposure to biuret and isocyanurate because they produce less overspray or position their bodies to avoid overspray. Consequently, training automotive spray painters on the best techniques for applying paint may help reduce personal exposures.
In comparison to our mixed models, Woskie et al. (2004)
developed a multiple regression model to predict BZC's of TRIG. Significant covariates in this model included: volume of polyisocyanates applied, volume of clear coat used per month, and type of paint booth. These general process-related variables described an estimated 39% of the variability in the BZCs of TRIG, which is within the range of variability (20–54%) described by our analyte-specific models. Because the models generated in our study used specific process- and task-related variables, it is difficult to compare our models to the model developed by Woskie et al. (2004)
. Nevertheless, our models may be more practical in terms of identifying practices and control technologies to reduce personal exposures.
In addition to statistical models, deterministic models have been used to understand exposures during spray painting. Among the most notable in the literature is the model developed by Flynn et al. (1999)
for predicting BZCs of general aerosols during spray painting in crossdraft booths. In addition to painter orientation, the most important parameters of this model were generation rate, momentum flux of air from gun, and momentum flux of air to worker's body. Although these parameters were not directly measured in our study, generation rate may depend on the concentrations of polyisocyanates in paint, momentum flux of air from gun likely depends on the type of spray gun being used, and momentum flux of air to painter's body may depend on the airflow in the booth. These variables, except for gun type, were significant in three or more of the mixed models, and it is probable that gun type would have been significant had there been more variability in gun type (i.e. high volume, low pressure guns were used in 92% of the paint tasks). It is important to note, however, that airflow had a protective effect even in crossdraft booths. Thus, airflow in this study generally functioned to draw overspray away from the painter's body rather than toward the painter's body. Flynn et al. (1999)
determined that momentum flux of air from the gun was important as it affected the amount of aerosol ‘bounce back’ into the breathing zone. The size of object being sprayed, angle of spray gun relative to the object, and distance between spray gun and object are also important factors involved in ‘bounce back’ but were not recorded in this study due to the difficulty in measuring such factors.
The random effect of visit day was significant in all the final mixed models. This suggests that BZCs are likely to vary from visit to visit due to factors other than those which were evaluated in this study. Varying work practices are probably a major cause of the intervisit variability. Work practices may depend on a number of factors such as the size and orientation of the objects being painted, the busyness of the shop, and the condition of the work equipment. The intervisit variability observed in this study underscores the importance of collecting samples at different times of the year in order to obtain the most representative exposure estimates.
The BZCs reported in this paper ( and ) represent task-based (≤30 min) TWAs. Thus, ceiling limits or short-term exposure limits (STELs) are more appropriate for comparison than work-shift (i.e. 8-h TWA) exposure limits. The National Institute for Occupational Safety and Health ceiling limit for HDI (i.e. 140 μg m−3
) was exceeded only once (i.e. 179 μg m−3
) during this exposure-assessment study. This is not surprising since HDI represented <1% of all polyisocyanates in the automotive paint. Oregon is the only government entity in the USA to promulgate a STEL for HDI-based polyisocyanates biuret and isocyanurate (i.e. 1 mg m−3
). The BZCs measured in this study are not directly comparable to the Oregon STEL because they were not time weighted over 15 min. Nevertheless, it is interesting to note that the Oregon STEL was exceeded by 71% of the task-based BZCs, with the highest isocyanurate BZC (18
700 μg m−3
) being over 18 times greater than the recommended limit.
In a 1980–1990 survey of Oregon automotive repair shops, Janko et al. (1992)
measured a GM of 14 μg m−3
for HDI and 1600 μg m−3
for HDI-based polyisocyanates, with respective peak concentrations of 340 and 18
400 μg m−3
. Similar levels of biuret and isocyanurate combined (GM = 1470, peak = 19
700 μg m−3
) and lower levels of HDI (GM = 4.1, peak = 179 μg m−3
) were measured in our study. It is important to note that painters in this study were protected by respirators of various types (i.e. half face, powered air purifying, supplied air, etc.). Over 70% of the painters wore half-face respirators equipped with organic vapor cartridges. The Occupational Safety and Health Administration (OSHA) assigned protection factor for half-face respirators is 10 (OSHA, 2006
). After accounting for the OSHA protection factor (i.e. dividing the BZCs by 10), we observed that >5% of the task-based BZCs exceeded the Oregon STEL. Liu et al. (2006)
found that the average workplace protection factor for half-face respirators equipped with organic vapor cartridges was 388 for polymeric HDI. Such protection would reduce the inhaled portion of the highest isocyanurate concentration to ~50 μg m−3
. Although well below the Oregon STEL, this level of exposure could still pose health risks to susceptible or sensitized individuals. This underscores the importance of reducing air concentrations inside the paint booths.
The analyte-specific mixed models we developed () share some important similarities. Three variables (i.e. analyte-specific paint concentration, airflow, and sampler type) were common to three or more of the mixed models. According to the model predictions, reducing analyte-specific paint concentrations and/or increasing airflow results in lower BZCs of polyisocyanates. In addition, lower BZCs of all polyisocyanates were measured in downdraft booths than crossdraft or semi-downdraft booths, which is consistent with previous findings of particulate levels in paint booths (Heitbrink et al. 1995
). Although painters and shop managers have limited control over polyisocyanate concentrations in the paint and the type of paint booth installed in the workplace, airflow inside the paint booth can be maximized by changing supply and return air filters on a regular basis and ensuring that plastic sheeting and masking tape are not obstructing the return ducts. These simple acts of maintenance and prevention could have tremendous implications on the health and safety of automotive spray painters.
A significant finding from this study was the effect of sampler type on measured BZCs of HDI and isocyanurate in short-term samples (<30 min). Because the two-stage samplers appeared to underestimate the air concentrations of HDI and isocyanurate in short-term samples in this study, investigators should carefully consider the type of sampler to use for task-based monitoring of reactive compounds like polyisocyanates. The mixed models we developed may provide a reasonable way of estimating worker exposure in retrospective studies where air-sampling data is lacking but where the other covariates can be adequately estimated. However, validation of these models is necessary to confirm their usefulness for exposure reconstruction.