The wide range of reported residential-dust PAH concentrations probably reflects true geographical variability. Specifically, one possible explanation for the relatively low levels of PAH in California homes is the infrequent use of coal-tar for sealing pavement in the Western U.S. . Interestingly, several factors that have been related to residential-dust PAH levels in previous studies; i.e., smoking , vacuum use frequency , season , and urban location , were not important determinants in this analysis. However, some variables that were omitted from the optimal model , were correlated with residential-dust PAH concentrations in bivariate analyses. Specifically, the variables urban location , traffic density , and vacuum use frequency were correlated with PAH levels. Moreover, PAH levels were higher in residences where some household smoking was reported compared to residences with no household smoking . Still, these variables were not important predictors of PAH concentrations when more informative variables were included in the model . Conversely, variables describing cooking habits, fireplace use, and season did not appear to be correlated with residential-dust PAH concentrations in bivariate or multi-variable regression analyses . Unfortunately, the variables describing cooking habits were crude and no information was available for most of the NCCLS population . Notably, the case-control status was not an important determinant of PAH concentrations when more informative variables were included in the model. The potential importance of reporting bias in the optimal model can be discounted, because case and control parents would not be expected to differentially report the important predictor variables, namely, address, child’s age,indoor grow methods and residence construction date. These analyses of total PAH concentrations assume that the 9 PAHs would have similar characteristics. To examine differences across PAHs, the variable set selected for the total PAH model was used to create a model for each individual PAH.
The regression coefficients for each of the 9 individual PAH models were fairly consistent, with each individual regression coefficient falling within the 99% confidence interval of the regression coefficient from the total PAH model . The consistency of the regression results across individual PAH models and the correlation between individual PAHs, suggests that the 9 PAHs measured have similar determinants. In summary, these analyses identified several determinants of PAH concentrations in residential dust and confirmed that gas heating and elevated outdoor PAH concentrations were significant predictors of indoor PAH levels. Moreover, the regression results suggest that PAHs measured in residential dust could be used as long-term surrogates for residential exposures to PAHs. Nonetheless, despite the large number of dust measurements and the extensive questionnaire- and GIS-based data developed by the NCCLS, the optimal model was only able to explain a small portion of the overall variability in PAH levels in residential dust . Hence, it is important to directly measure PAH levels in epidemiologic studies. Residential dust can act as a reservoir for indoor chemical contamination and persistent organic chemicals like PCBs accumulate in carpets . As such, PCB concentrations measured in residential dust may be long-term predictors of indoor PCB exposures. Moreover, because inadvertent dust ingestion could be responsible for a substantial portion of total PCB exposure in some young children , levels of PCBs in residential dust may be particularly relevant to the uptake of PCBs in children . The health impact of PCB exposure has not been fully characterized. Recently, investigators have reported that ambient exposure to PCBs was associated with an increased risk of type-2 diabetes .Similarly, investigators from the NCCLS noted that elevated levels of PCBs in residential dust were associated with the development of childhood leukemia . Because timely collection of biological and environmental samples is particularly challenging in case-control studies, interview-based exposure assessment is commonly employed.
Investigators have shown that certain demographic and lifestyle factors, including country of origin, sex, parity, body mass index, age, breastfeeding, and educational level, can influence biological levels of persistent chemicals . However, less is known about the relationship between self-reports and levels of PCBs in residential dust . Notably, one previous study identified floor age as an important predictor of PCB concentrations in residential dust . Chapter 5 assesses the predictive value of self-reported data in estimating measured levels of 6 PCBs in residential dust from 583 households in California, and discusses the implications for using questionnaires to classify PCB exposures in epidemiological studies more generally. Depending upon the particular PCB congener, between 45 and 91% of PCB measurements were below analytical limits of detection . Since such high proportions of the dust samples had non-detectable levels of PCBs, multi-variable logistic regression models were used to predict the probability that a particular PCB congener would be detected based upon the self-reported explanatory variables. The deletion–substitution–addition algorithm, a cross-validation tool for model selection written in R , was used to choose optimal models from the list of candidate variables . Briefly, the DSA algorithm partitioned the data into 10 complementary subsets, fit a candidate logistic regression model based on 9/10 of the data, and validated that model by comparing predicted and measured values in the remaining 1/10 of the data . After iteratively evaluating combinations of different variables, the optimal model was the one that minimized the mean squared error between the predictions and the observations in 100 validation sets. The DSA algorithm was restricted so that it identified main effects and 2nd order interaction terms that predicted the detectable presence of PCBs. This process was repeated 6 times so that each PCB congener had an optimal logistic model. Finally, for all observations above detection limits, multi-variable linear regression were used to evaluate whether the variables that were selected by the DSA algorithm were also associated with PCB concentrations in residential dust. Since the PCB congeners had approximate log-normal distributions, the natural log of the PCB concentrations were used for linear regression.
The age of the residence was the strongest predictor of the detectable presence of PCB in residential dust. For example, PCB-153 was detected in 74% of residences built before 1980, but it was only detected in 32% of more recently constructed homes . Since U.S. production of PCBs was banned in 1979 , homes built before 1980 were expected to have more PCB contamination than those constructed more recently. Parental age was also useful in predicting the detectable presence of PCB in residential dust from older homes. For example, PCB-118 was detected in 55% of older homes occupied by older mothers, but it was only detected in 36% of older homes occupied by younger mothers . Perhaps this observation is attributable to older parents owning older items that contain PCBs. Alternatively,cannabis dryer this observation could be explained by the fact that older parents tended to have older carpets. As shown in Table 3, the PCB concentrations measured in the NCCLS residential-dust samples were generally lower than those previously reported for residences in the U.S. . However, one recent study reported similarly low median PCB concentrations in dust collected from residences in Michigan . Previous investigators have observed that PCB concentrations were elevated in dust from older residences and in dust from older floor surfaces . It was noted earlier that concentrations of PAHs and nicotine measured in dust from NCCLS households were also positively associated with residence age. Taken together, these findings suggest that chemical contaminants may persist in household carpets for decades, and that residential dust represents an excellent resource for investigations of long-term chemical exposures in the home. Previous investigators have also reported that some construction materials, such as wood-floor finishes and caulk can contain high concentrations of PCBs. However, in these analyses, recent construction activities, including re-flooring, were not predictive of PCB detection. In summary, the DSA algorithm identified few determinants of PCB levels in a large sample of residential dust . In fact, the age of the residence and the age of its occupants were the only determinants of the detectable presence and concentrations of PCBs in residential dust. The lack of other questionnaire-based determinants of PCB contamination underscores the importance of directly measuring PCB levels in epidemiological studies. The results from this chapter suggest that PCBs measured in dust could be used as indicators of long-term residential PCB contamination. Concentrations of PBDEs in 81 dust samples from NCCLS homes were relatively high compared to levels measured around the world . However, as shown in Figure 11, in comparison to homes from other studies in California , the NCCLS homes had relatively modest median concentrations of BDE-47, BDE-99, and BDE-100.
In roughly half of the residential-dust samples from the NCCLS, BDE-209 was the predominant BDE congener, suggesting extensive historical use of the commercial Deca-BDE mix in California. The NCCLS is the first study from California to report concentrations of BDE-209 in residential dust. Results from Spearman correlations and principal component analysis revealed clear patterns in PBDE contamination. Specifically, 22 PBDE congeners were resolved into three principal components that reflected distinct sources of PBDEs; i.e., the Penta-BDE, Octa-BDE, and Deca-BDE commercial mixtures. These findings suggest that relatively few indicator PBDE congeners could be used to describe PBDE contamination in homes. Although BDE-202 has not been reported at detectable levels in any PBDE commercial mixtures, the median concentration of BDE-202 from 81 NCCLS dust samples was 3 ng/g and concentrations were as high as 77 ng/g. The presence of BDE-202 in the NCCLS dust samples points to degradation of BDE-209 molecules in the environment. Likewise, the ratios of total nona-BDE to BDE- 209 in the 81 dust samples from NCCLS residences were much greater than the ratio typically found in the commercial Deca-BDE mix . Additionally, Table 33 shows that concentrations of each nona-brominated diphenyl ether were highly correlated with concentrations of BDE-209. Taken together, these findings suggest that BDE-209 can break down into nona-brominated and octa-brominated diphenyl ethers in the environment. Since lower-brominated congeners are thought to be more toxic than BDE-209 , debromination of BDE-209 could lead to more harmful indoor contamination. In summary, NCCLS residences had some of the highest median concentrations of BDE-47, BDE-99, and BDE-209 reported in North America . Two PBDE congeners, BDE-99 and BDE-209, were found to predominate in dust samples from 81 Californian homes. Additionally, there was suggestive evidence of BDE-209 debromination in the indoor environment. Results from these analyses can guide epidemiologists in developing sampling strategies for residential dust as a medium for estimating exposures to PAHs, PCBs, or nicotine in their studies. Generally, investigators can improve the precision of their exposure estimates and limit attenuation bias by making repeated exposure measurements in each residence. However, the analytical advantages of a repeated sampling design must be balanced with the practical concerns of a study’s schedule and budget. As shown in Table 39, calculations that employed estimated variance ratios from the Fresno Exposure Study suggest that three repeated dust measurements per residence would be sufficient to reduce the magnitude of attenuation bias to less than 20% for each chemical measured in the current study. Moreover, if repeated measurements would not be feasible, Table 38 indicates that for 10 of the 13 chemicals analyzed, the expected magnitude of attenuation bias would still be less than 30%. Because the results shown in this chapter are based on a limited sample size , the variance ratio estimates are somewhat imprecise . However, these findings should be externally valid and useful for other investigators measuring these same chemicals in residential dust. Notably, the concentrations of chemicals measured in dust from the Fresno Exposure Study residences were generally similar to the concentrations reported for the NCCLS homes with respect to both the medians and the ranges of concentrations . Unfortunately, it is difficult to compare the results from this chapter to those from two other studies that repeatedly sampled dust from the same residences over time and reported corresponding variance components , because these studies published estimates for different chemicals in dust . However, estimated variance ratios for the Fresno Exposure Study residences were quite similar to those estimated using unpublished data from Egeghy et al. for several PAHs that were measured in residential dust from both studies .