Trends of VOC exposures among a nationally representative sample: Analysis of the NHANES 1988 through 2004 data sets

Abstract

Exposures to volatile organic compounds (VOCs) are ubiquitous due to emissions from personal, commercial and industrial products, but quantitative and representative information regarding long term exposure trends is lacking. This study characterizes trends from1988 to 2004 for the 15 VOCs measured in blood in five cohorts of the National Health and Nutrition Examination Survey (NHANES), a large and representative sample of U.S. adults. Trends were evaluated at various percentiles using linear quantile regression (QR) models, which were adjusted for solvent-related occupations and cotinine levels. Most VOCs showed decreasing trends at all quantiles, e.g., median exposures declined by 2.5 (m, p-xylene) to 6.4 (tetrachloroethene) percent per year over the 15 year period. Trends varied by VOC and quantile, and were grouped into three patterns: similar decreases at all quantiles (including benzene, toluene); most rapid decreases at upper quantiles (ethylbenzene, m, p-xylene, o-xylene, styrene, chloroform, tetrachloroethene); and fastest declines at central quantiles (1,4-dichlorobenzene). These patterns reflect changes in exposure sources, e.g., upper-percentile exposures may result mostly from occupational exposure, while lower percentile exposures arise from general environmental sources. Both VOC emissions aggregated at the national level and VOC concentrations measured in ambient air also have declined substantially over the study period and are supportive of the exposure trends, although the NHANES data suggest the importance of indoor sources and personal activities on VOC exposures. While piecewise QR models suggest that exposures of several VOCs decreased little or any during the 1990’s, followed by more rapid decreases from 1999 to 2004, questions are raised concerning the reliability of VOC data in several of the NHANES cohorts and its applicability as an exposure indicator, as demonstrated by the modest correlation between VOC levels in blood and personal air collected in the 1999/2000 cohort. Despite some limitations, the NHANES data provides a unique, long term and direct measurement of VOC exposures and trends.

1. Introduction

All individuals are exposed to volatile organic compounds (VOCs) due to emissions from numerous emission sources. Exposures result from indoor sources such as cigarette smoke, building materials, adhesives, coatings and paint strippers, dry cleaning, moth repellents, byproducts of water chlorination, and many other personal, commercial and industrial products (Wallace, 1987; Wallace et al., 1989; ATSDR, 1997a, 2005; US EPA, 2010a). Outdoors, VOCs are emitted from point sources, e.g., power plants and refineries, area sources, e.g., gasoline stations and printing shops, and mobile sources, e.g., engine exhaust and fuel-related emissions (MDE, 2010). Indoor concentrations typically exceed outdoor levels by 2–5 times, and indoor levels can be increased by low ventilation, high temperature and high humidity, and personal exposures can increase sharply after activities such as smoking and cooking (US EPA, 2010a). Concentrations in occupational settings can considerably exceed levels in indoor and outdoor environments (Rappaport and Kupper, 2004), although environmental exposures may be briefly elevated in situations such as automobile refueling (Egeghy et al., 2000).

In the U.S., emissions of many VOCs have declined in recent years, motivated by concerns regarding both the direct health effects of VOCs and their role in forming tropospheric ozone. Emissions have been lowered by substituting low emitting materials and processes, using controls such as catalytic converters, and shifting away from manufacturing jobs where solvent use was common. Based on the U.S. National Emissions Inventory (NEI), VOC emissions have been reduced by 35% from 1990 to 2005, or 2.3% per year, mainly due to controls on industry and on-road mobile sources (US EPA, 2010b). These and most other estimates of emission trends primarily use empirical and engineering factors, not actual measurements.

Decreased emissions have lowered ambient concentrations. A comprehensive review of air toxics data collected from 1990 to 2005 in the U.S. EPA’s Air Quality System (AQS) showed that median levels of benzene, toluene, ethylbenzene, styrene, xylene and tetrachloroethyelene (PERC) have declined by about 5–7% per year; chloroform by 1–4% per year; and 1,4-dichlorobenzene (1,4-DCB) by 0–9% per year, depending on the period (McCarthy et al., 2007). Benzene trends have also been examined by Fortin et al. (2005), who estimated an average decrease of 6.2% per year from1993 to 2002 and 9.8% per year between 1994 and 1999, mainly using Photochemical Assessment Monitoring Stations (PAMS) data, and by US EPA (2003a; 2007; 2010d), which showed decreases in urban areas of 8% per year from 1994 to 2000, 3% per year from 2000 to 2005, and 4% per year from 1994 to 2009. PAMS data are collected in the warmest portion of the year (the “ozone season”), and do not represent annual averages. Somewhat faster declines (9.8% per year) have been shown for quarterly averages of benzene in California from 1990 to 1995 (Hammond, 1998), and by data in the Urban Air Toxics Monitoring Program (UATMP), which has operated year-round since 1987, and which includes several sites located near busy roadways, commercial or industrial facilities (US EPA, 2001). Ambient data are subject to variability from year-to-year changes in emissions, meteorology and sampling methodology, although long term declines across a number of periods are quite consistent and indicate the effectiveness of emission controls (McCarthy et al., 2007). However, ambient monitoring only partially explains exposure trends due to the little time most individuals spent outdoors and the strength of VOC sources in building and commuting environments.

Longitudinal studies collecting VOC data using indoor air or biomonitoring sampling, large populations, and long periods are uncommon. Since most VOC exposure arises from indoor sources (US EPA, 2010a), it is essential to monitor the indoor or personal environment. Decreased smoking rates and restrictions on tobacco smoking, for example, may have lowered indoor concentrations and exposures of some VOCs more than changes in outdoor concentrations.

Exposures to pollutants can be estimated in many ways, but biomarker measurements often are considered the best exposure indicator since they account for multiple settings (e.g., indoor, outdoor and commuting environments), sources and exposure pathways (Ashley and Prah, 1997). In urine, concentrations of VOCs strongly correlate to indoor levels (Wang et al., 2007). In blood, VOC concentrations have been associated with airborne levels, smoking and other activities, as well as individual characteristics such as gender and body mass index (Lin et al., 2008). Biomarkers have limitations, e.g., VOCs with rapid clearance (short biological half-lives) will reflect only recent exposures, thus observed relationships between airborne and biomarker concentrations depend on the variability of airborne levels, the duration of exposure and sampling periods, and clearance rates (Kwok and Atkinson, 1995; Heinrich-Ramm et al., 2000; Sexton et al., 2005; Lin et al., 2008). To date, quantitative and nationally representative trends using biomarkers have not been reported. Such analyses require the use of consistent methodologies, representative and large samples, and long study periods.

This study examines trends in VOC exposures using data from the National Health and Nutrition Examination Survey (NHANES), which examines a large and representative sample of the U.S. population. We combined the five NHANES cohorts in which VOC biomarkers were measured, which spanned the period from 1988 to 2004, and estimate trends using quantile regression and other techniques.

2. Materials and methods

2.1. Data sources

VOC data were obtained from two cohorts of NHANES III (1988–1991, 1991–1994), and three cohorts of “continuous NHANES” (1999/2000, 2001/2002 and 2003/2004). Initially, NHANES focused on health and nutrition issues and did not include contaminant measurements. Participants were selected to be nationally representative using a stratified, multistage, probability–based sampling design, e.g., elderly and minorities were over–sampled. VOCs were measured for a subsample of adults aged 20–59 years for each cohort studied between 1988 and 2004, with sample sizes from 605 to 1489 as shown in Appendices A and B, (NCHS, 2000; 2010d). To obtain nationally representative results and allow comparability between cohorts, each cohort used the same sampling and weighting scheme (NCHS, 2006). There are several differences between cohorts. NHANES III used a 6 year survey cycle, 81 primary sampling units (PSUs) from 1988 to 1994 (randomly divided into two groups for 1988–1991 and 1991–1994), and about 15,000 participants per cohort. Continuous NHANES used a 2 year survey cycle, 12 PSUs in 1999/2000 (3 PSUs were omitted due to delays in data collection), 15 PSUs in both 2001/2002 and 2003/2004 cycles, and approximately 10,000 participants per cohort (NCHS, 2010a; 2010b; 2010c). Thus, continuous NHANES encompassed fewer PSUs and obtained smaller samples, and consequently, standard errors may be larger than those in NHANES III (NCHS, 2006).

NHANES III and continuous NHANES used similar procedures to collect and analyze blood samples (NCHS, 2000; 2011). Participants arrived at a central location and designated time, and were then shepherded through four air conditioned trailers that comprised the mobile examination center (MEC) in visits that could require up to 4 h (NCHS, 2009). Blood samples were drawn in the third trailer. Whole blood samples were analyzed for 15 compounds: benzene, toluene, ethylbenzene, m, p-xylene, o-xylene, styrene, chloroform, bromodichloromethane (BDCM), dibromochloromethane (DBCM), bromoform, 1,4–dichlorobenzene (1,4–DCB), tetrachloroethene (perchloroethylene, PCE), methyl tert–butyl ether (MTBE), carbon tetrachloride, and trichloroethene). Analyses used purge–and–trap extraction or headspace solid–phase microextraction (SPME), and capillary gas chromatography/mass spectrometry. Consistent quality control and quality assurance protocols were maintained (NCHS, 2010e). In 1999/2000, participants in the VOC cohort (n = 851) also collected personal air samples for VOC analysis (NCHS, 2010f). In this case, blood samples were collected, then upon leaving the MEC, participants were given a passive sampler to wear for 48–72 h. Later, participants returned the sampler to the MEC for analysis.

Several data sets were reviewed to derive trends in nationwide emissions and ambient concentrations to compare to the NHANES measurements. Emission data were taken from the National–Scale Air Toxics Assessment (NATA), an ongoing program used to derive pollutant emissions and risks (US EPA, 1996; 1999; 2002). Trend analyses using emission inventory must account for changes in inventory methods, e.g., NATA included additional source types in 1999 (US EPA, 1999). We also used NATA’s dispersion model predictions for 1996, 1999 and 2002, which are based on the NATA emission data but which reflect effects of dispersion. NATA significantly under predicts concentrations of many VOCs, due to missing and underestimated emission sources, among other reasons (US EPA, 2010c). However, our analysis stressed relative changes, which may be less sensitive to these biases. Several ambient monitoring data sets were also reviewed, including the 1993 to 2004 aromatic concentrations in the Photochemical Assessment Monitoring Stations (PAMS) (US EPA, 2011), the 2001 to 2004 data from UATMP (US EPA, 2001), and the 1990 to 2004 data from AQS (US EPA, 2011). PAMS and AQS data cover or nearly cover the period spanned by the five NHANES cohorts. Site annual averages from the AQS were downloaded and national level annual averages were calculated. To obtain reliable and representative averages, only sites collecting 24–h samples were used, each site had to collect at least 24 measurements per year, and at least 20 sites meeting these criteria were required to compute the annual average. Trends were plotted and percent changes per year were calculated using simple linear regressions.

2.2. Statistical analyses

Descriptive analyses followed the NHANES analytic guidelines (NCHS, 2006) and used weights to account for NHANES’ hierarchical clustered sampling strategy. The detection frequency (DF), defined as the percentage of measurements exceeding the method detection limits (MDLs), excluding missing values, was calculated for each VOC. (Appendix A shows DFs and MDLs.) VOCs with very low (<5%) DFs across the five cohorts were excluded from further analyses. MTBE was only measured in continuous NHANES, and was excluded from certain analyses. To ensure a sufficient sample size, at least 300 observations per VOC per cohort were generally required. New variables formed to examine related groups of VOCs included BTEX (the sum of benzene, toluene, ethylbenzene, m, p– and o-xylene concentrations) and total trihalomethanes (ΣTHM, the sum of chloroform, BDCM, DBCM and bromoform). Spearman rank correlation coefficients were used to test associations among blood VOCs and among the air and blood measurements for the 1999/2000 cohort. Group differences in key demographic variables (age, gender, race, education levels, and income) among the cohorts were tested using ANOVA and Chi–square tests for continuous and categorical variables, respectively.

Concentration trends were examined using quantile regression (QR) models, which estimate changes in conditional quantiles of a response variable with changes in VOC levels (Koenker and Bassett, 1978). This semiparametric method makes no parametric distribution assumptions for random errors. Model coefficients are estimated by optimizing an objective function and the accompanying standard errors are derived using either parametric assumptions on the model coefficients or via resampling techniques, e.g., bootstrap analysis (Cade and Noon, 2003). Compared to ordinary regression models, QR models are more robust, e.g., resistant to effects of outliers, a special concern for skewed distributions, which have been observed even after log–transformation of VOC data, following the NHANES guidelines (Jia et al., 2008; NCHS, 2010g). Moreover, QR models indicate changes at different quantiles, e.g., allowing comparison of trends at median and upper percentiles, and exploration of exposure patterns. Linear QR models were fitted for 0.5, 0.75 and 0.95 quantiles (50th, 75th and 95th percentile concentrations). In a sensitivity analysis to allow changes in trend over the long interval (1994–1999) between the NHANES III and continuous NHANES cohorts, piecewise QR models were used with knots (locations where the slope changes) at several locations (e.g., 1991–1994, 1999/2000).

To facilitate interpretation, annual average percentage changes in untransformed (raw) concentrations were computed for each VOC and quantile, e.g., the change across the 15 year study period is 1/15 (C₅ – C₁)/C₁ 100%, where C₁ and C₅ are concentrations for a specific VOC and quantile in the first and fifth cohorts, respectively. Annual relative changes were calculated similarly for emissions and ambient concentrations.

Cigarette smoking is an important source of benzene and other aromatic compounds (Wallace et al.,1987), and cotinine is a reliable biomarker of tobacco smoke (Benowitz, 1999). Correlations between serum cotinine levels and blood VOCs were determined, and the QR models were adjusted for this parameter.

Results of trend analyses might be affected by shifts in the occupational mix, e.g., the declining number of workers in industries where solvent use may be common. To account for such effects, we identified occupational groups associated with VOC concentrations, and adjusted QR models using indicator variables for these groups. Because many of the 41 occupational groups in NHANES had small sample numbers, groups were consolidated into eight categories (managerial and professional specialty occupations; professional specialty occupations; technical, sales and administrative support occupations; service occupations; farming, forestry and fishing occupations; precision production, craft, and repair occupations; operators, fabricators, and laborers; military occupations) based on 1990 Census Industrial & Occupational Classification Codes. Due to the small number of military personnel (n = 7), this category was dropped. ANOVAs were used to test whether VOC levels were associated with these occupational categories, using the managerial and professional specialty category as a reference group.

While the QR models used cohort–specific weights to obtain population–weighted results, these models cannot account for NHANES’ cluster sampling. As a sensitivity analysis to evaluate the effect of clustering, trends in the mean were estimated using linear and piecewise models with the appropriate weights, and compared to regression results with and without adjustments for strata and clusters.

SAS 9.2 (SAS Institute, Cary, North Carolina, USA) was used for statistical testing and model development. Weighted analyses used Survey means and Surveyreg, and QRs used Quantreg. Other analyses were calculated using Excel (Microsoft, Redmond, WA).

3. Results and discussion

3.1. Descriptive analyses

Table 1 breaks out descriptive summary statistics for the NHANES III (1988–1994) and continuous NHANES (1999–2004) cohorts. (Appendix B gives cohort–specific statistics.) Carbon tetrachloride and trichloroethene had very low DFs (5.5 and 4.8%, respectively), and were excluded from further analyses. In NHANES III, 1,4–DCB had the highest mean level (1.11 ± 0.12 µg L⁻¹) among the 12 VOCs, over twice that seen for the next highest compound, toluene, while BDCM had the lowest mean (0.008 ± 0.001 µg L⁻¹) with 86% of measurements fell below the MDL. In continuous NHANES, 1,4–DCB levels decreased (0.87 ± 0.10 µg L⁻¹), although it remained the single highest VOC. Again, DBCM had the lowest concentration (0.002 ± 0.000 µg L⁻¹) with 43% of measurements below the MDL (which also decreased). VOC levels decreased over these two periods, and differences in high–end exposures were particularly striking (Table 1). Again examining 1,4–DCB, the maximum was 52 µg L⁻¹ and the 1988–1994 95th percentile concentration was 11 µg L⁻¹, well above any other VOC. As discussed later, products containing 1,4–DCB have been widely used indoors, and possible occupational exposure and low clearance rates for this VOC may increase exposures and concentrations in blood.

Table 1.

Statistics of VOC concentrations (mg L⁻¹) in blood measured for NHANES III and continuous NHANES.

VOCs	NHANES III: 1988–1994							Continuous NHANES: 1999–2004
	N	DF	Mean	SE	50th	90th	95th	N	DF	Mean	SE	50th	90th	95th
Aromatics
Benzene	796	66	0.132	0.008	0.062	0.323	0.476	2482	62	0.091	0.006	0.032	0.190	0.320
Toluene	575	56	0.596	0.008	0.281	1.081	1.478	2587	95	0.278	0.014	0.120	0.578	0.880
Ethylbenzene	606	56	0.125	0.004	0.061	0.183	0.245	2439	68	0.049	0.002	0.031	0.089	0.133
m,p-xylene	1018	62	0.246	0.033	0.117	0.414	0.607	2602	97	0.206	0.012	0.140	0.374	0.512
o-xylene	628	59	0.153	0.004	0.101	0.198	0.267	2654	41	0.054	0.002	0.035	0.087	0.116
BTEX	1018	NA	0.845	0.101	0.463	1.642	2.380	2703	NA	0.645	0.030	0.363	1.293	1.842
Styrene	624	54	0.094	0.001	0.041	0.129	0.177	2476	52	0.068	0.012	0.021	0.110	0.158
THMs
Chloroform	876	47	0.042	0.002	0.023	0.072	0.118	2216	95	0.027	0.003	0.014	0.053	0.079
BDCM	937	13	0.008	0.001	0.006	0.011	0.019	2461	86	0.003	0.000	0.002	0.007	0.011
DBCM	919	11	0.010	0.000	0.009	0.015	0.022	2464	64	0.002	0.000	0.001	0.005	0.008
Bromoform	579	4.5	0.021	0.000	0.019	0.019	0.034	2413	60	0.003	0.001	0.001	0.005	0.010
∑THM	1016	NA	0.065	0.003	0.049	0.107	0.147	2513	NA	0.032	0.002	0.018	0.066	0.100
Others
1,4–DCB	915	86	1.112	0.122	0.322	4.658	11.03	2409	57	0.872	0.102	0.140	1.900	5.300
PCE	566	41	0.219	0.005	0.061	0.347	0.617	2577	29	0.081	0.007	0.034	0.090	0.180
MTBE	NA	NA	NA	NA	NA	NA	NA	2263	85	0.041	0.005	0.013	0.110	0.159

Sample size N includes measurements below MDL, which were replaced by 1/2 MDLs. Statistical analyses only accounted for detectable measurements and measurements below MDLs, which were replaced by 1/2 MDLs.

DF = detection frequency (%); SE = standard error; NA = not available.

As expected, related VOCs were correlated. The five BTEX compounds in blood had Spearman correlation coefficients from 0.14 (benzene and m, p-xylene) to 0.81 (ethylbenzene and o-xylene) in NHANES III, and from 0.38 (benzene and m, p-xylene) to 0.89 (ethylbenzene and o-xylene) in continuous NHANES (Appendix C). The THM compounds were significantly correlated, except for chloroform and bromoform in NHANES III. In general, correlation coefficients were lower in the 1988–1994 cohorts, in part due to the higher MDLs obtained during this period.

Correlation coefficients between blood and personal air measurements in the 1999/2000 cohort were statistically significant for the nine VOCs available, and ranged from 0.24 to 0.38 for the BTEX compounds, to 0.62 for PCE and 0.65 for 1,4–DCB (Appendix D). Thus, the personal air measurements explained a modest portion of the blood measurements. The NHANES study design likely lowered these correlations since sequential, rather than simultaneous, measurements were utilized, i.e., higher agreement likely would have occurred if blood was sampled when the personal air samplers were returned. Also, correlations are lowered by clearance rates that differ among VOCs, exposure pathways other than inhalation (e.g., consumption of chlorinated water), and experimental errors. Nonetheless, the positive and significant correlation suggests that the blood measurements provide useful exposure information.

Due to relatively rapid clearance, VOCs measurements in blood reflect exposures over only the immediate period preceding the blood draw (e.g., 2 or 3 half–lives). If sampling was random, blood measurements can reflect chronic exposures, although some attenuation is expected since blood draws would not immediately follow high exposure events due to time needed for travel and processing in the MEC. Consequently, the sample variability may not reflect the true variability of chronic exposures.

The 1988–1991 cohort had an excessive fraction (63%) of values reported as “extreme or illogical values” for toluene, ethylbenzene, o-xylene, styrene, bromoform and PCE, which left fewer than 200 valid measurements. Also, compared to subsequent cohorts, available data for these VOCs and cohort tended to have lower correlation among related compounds, and means (and medians) appeared inconsistent (Appendix B). For example, m, p-xylene measurements in this cohort were very low and inconsistent with data in subsequent cohorts. Measurements of these seven VOCs in the 1988/1991 cohort were not considered to be reliable, and thus were omitted from subsequent analyses, along with the derived BTEX and ΣTHM variables. We have not identified other assessments of VOC data quality in the NHANES documentation or general literature.

3.2. VOC Trends

Potential covariates were identified before evaluating VOC trends. Several occupational groups were associated with VOC levels, although none achieved statistical significance in ANOVA tests, possibly because effects were small or diluted due to the broad occupational categories used. Nevertheless, trend analyses were adjusted for groups that seemed likely to have VOC exposure: service occupations (associated with elevated 1,4–DCB levels); precision production, craft and repair occupations (BTEX); and operators, fabricators, and laborers (BTEX). A variable combining these groups was used as a covariate in QR models. Additionally, all VOCs except PCE were associated with serum cotinine levels, which dropped from an average of 107 to 70 ng mL⁻¹ over the 1988–2004 period. Initially, all QR models were adjusted using log–transformed cotinine levels. However, this variable was not statistically significant for non–aromatic VOCs and parameter estimates changed little, thus cotinine was maintained in the final QR models for only aromatic VOCs. Among demographic variables, only age and education differed significantly between NHANES cohorts, and both age and college attainment increased with time. QR models including these variables showed insignificant changes in parameter estimates, and thus the demographic variables were not included in the final models.

The trend analysis focused on concentration quantiles exceeding 0.5 (50th percentile). Often, lower quantiles were at or near MDL concentrations. Linear QR models representing the entire study period (1988–2004) and adjusted for solvent–related occupations and cotinine levels (aromatic VOCs) showed statistically significant trends at 0.5, 0.75 and 0.95 quantiles for all VOCs except for PCE at the 0.5 quantile, and styrene and 1,4–DCB at the 0.95 quantile (Table 2). For most VOCs, these changes corresponded to an average decrease of 2.5–6.4% per year (Table 3). Graphical interpretations of results for benzene,1,4– DCB and PCE are presented in Figs.1 to 3. Panel A of each figure shows box plots for the five cohorts, superimposed with the estimated linear QR trend lines; panel B shows quantile plots of the linear QR estimate at 0.25, 0.5, 0.75 and 0.95 quantiles, along with 95% confidence intervals. Due to low DFs, the 0.25 quantile (left–most point) is not meaningful for 1,4–DCB and PCE, and only somewhat meaningful for benzene. These plots suggest that the rate of decline can depend on the quantile, and three patterns were discerned across the VOCs. Pattern 1 has similar decreases at all quantiles, shown by benzene (Fig. 1B). This pattern suggests uniform emission and/or exposure reductions from the sources that dominate population exposures, e.g., reduced exhaust and evaporative emissions from vehicles, the largest benzene exposure source. Pattern 2 shows more rapid decreases at upper quantiles and slower decreases at lower quantiles, as seen for PCE (Fig. 3B). In this case, the most exposed cohort might have a unique exposure source, which has been controlled, or that other measures have been taken to limit high exposures, while lower level exposures continue largely unabated among the general population, possibly due to other sources that have not been controlled as much. This pattern could be explained by controls on the leading occupational exposure sources of PCE, e.g., dry cleaning and metal–degreasing operations. Pattern 3 is a rapid decrease at central quantiles that exceeds upper quantiles decreases, as seen for 1,4–DCB (Fig. 2B). This may result from controls on sources that affect indoor and/or outdoor concentrations, without a commensurate reduction in high exposure cases. For 1,4–DCB, this might be explained by reduced use of mothballs and air fresheners, the major exposure sources for the general population, while the most exposed individuals either continue to experience a separate exposure source, e.g., industrial production of repellents, insecticides, resins, etc., or they remain intensive users of this chemical. Patterns and possible sources for individual VOCs are discussed in the next section.

Table 2.

Linear quantile regressions of log–transformed blood VOC concentrations for the 1988 to 2004 period.^a

VOCs	0.5 Quantile		0.75 Quantile		0.95 Quantile
	Slope	SE	Slope	SE	Slope	SE
Aromatics
Benzene	−0.054	0.003	−0.078	0.009	−0.043	0.025
Toluene	−0.099	0.009	−0.144	0.017	−0.118	0.024
Ethylbenzene	−0.060	0.005	−0.066	0.008	−0.103	0.023
m,p-xylene	−0.033	0.006	−0.057	0.008	−0.117	0.042
o-xylene	−0.069	0.004	−0.097	0.007	−0.122	0.028
BTEX	−0.066	0.006	−0.080	0.010	−0.071	0.027
Styrene	−0.036	0.004	−0.039	0.009	−0.061	0.033
THMs
Chloroform	−0.065	0.005	−0.064	0.006	−0.103	0.025
BDCM	−0.097	0.007	−0.043	0.003	−0.034	0.012
DBCM	−0.202	0.014	−0.149	0.005	−0.077	0.007
Bromoform	−0.241	0.001	−0.201	0.000	−0.128	0.022
∑THM	−0.115	0.007	−0.101	0.009	−0.115	0.030
Others
1,4–DCB	−0.063	0.001	−0.045	0.009	−0.032	0.025
PCE	0.001	0.001	−0.166	0.006	−0.177	0.042

Aromatic VOCs were adjusted for solvent–related occupations and serum cotinine levels; THMs and other VOCs were adjusted for solvent–related occupations only.

SE = standard error.

Bold type means statistically significant (p < 0.05); benzene at 0.95 quantile is borderline significant.

^aExcludes 1988–1991 data for toluene, ethylbenzene, m,p-xylene, o-xylene, BTEX, styrene, bromoform, ΣTHM and PCE.

Table 3.

Relative changes (%) per year in untransformed blood VOC concentrations (µg L⁻¹) at various quantiles.^a

VOCs	1988–1991 vs. 2003/2004			1988–1991 vs. 1999/2000			1999/2000 vs. 2003/2004
	0.5	0.75	0.95	0.5	0.75	0.95	0.5	0.75	0.95
Aromatics
Benzene	−3.8	−4.3	−3.3	5.2	−1.2	0.1	−18.2	−14.8	−12.4
Toluene	−5.6	−5.7	−4.7	−1.9	−2.9	−1.3	−15.3	−14.8	−12.9
Ethylbenzene	−4.2	−4.9	−4.9	−4.2	−4.2	−3.2	−6.5	−9.4	−11.1
m,p-xylene	−2.5	−3.5	−4.5	−0.8	−1.8	−2.4	−6.3	−7.9	−10.8
o-xylene	−5.5	−5.5	−5.6	−7.9	−6.8	−5.3	−2.1	−6.4	−10.9
BTEX	−4.4	−4.6	−4.2	−2.1	−1.6	−0.8	−10.8	−12.0	−11.9
Styrene	−3.9	−2.7	−3.1	0.5	0.1	0.6	−12.3	−8.2	−10.1
THMs
Chloroform	−3.9	−3.6	−3.9	3.5	6.3	2.5	−17.5	−18.1	−17.0
BDCM	−5.0	−5.6	−3.4	−6.5	−3.5	−3.0	−3.1	−2.6	−6.8
DBCM	−6.3	−3.0	−4.8	−8.1	−6.6	−5.8	−13.5	−10.8	−5.9
Bromoform	−7.9	−7.5	−7.0	−11.9	−11.2	−10.6	2.8	−1.0	1.1
∑THM	−5.9	−5.4	−3.8	−5.4	−2.1	2.7	−12.2	−14.4	−13.9
Others
1,4–DCB	−3.5	−3.7	−3.7	−2.3	−2.5	−4.8	−9.0	−9.8	−0.8
PCE	−3.2	−6.2	−6.4	−2.7	−4.8	−5.1	−5.3	−14.8	−15.4

Relative changes of VOC levels per year between study period at each percentile.

Bold type means statistically significant trend (p < 0.05).

^aExcludes 1988–1991 data for toluene, ethylbenzene, m,p-xylene, o-xylene, BTEX, styrene, bromoform, ΣTHM and PCE.

Fig. 1.

A) Box plot of benzene concentrations showing 0.05, 0.25, 0.50, 0.75 and 0.95 quantiles for each NHANES cohort. Linear QR trend lines for 0.5, 0.75 and 0.95 quantiles are shown as dashed, dashed and dotted, and solid lines, respectively. B) Quantile plot for linear QR model of benzene over entire study period (1988–2004). A solid line shows coefficients for linear QR models at various quantiles. A dashed horizontal line shows coefficients for linear regression model, and dotted horizontal lines show 95% confidence intervals.

Fig. 3.

Box plots and linear QR model results for PCE. Otherwise as Fig. 1.

Fig. 2.

Box plots and linear QR model results for 1,4–DCB. Otherwise as Fig. 1.

The trend analysis also raised questions regarding the veracity of the 1999/2000 VOC data, which had the highest levels of benzene (average of 0.184 − 0.015 µg L⁻¹) and chloroform (0.058 − 0.005 µg L⁻¹) across five NHANES cohorts. Moreover, using the 1999/2000 data as a baseline, subsequent cohorts showed very rapid declines (>15% per year to 2003/2004) in median and higher percentile concentrations of benzene, toluene and chloroform, far faster than earlier years (Table 3). As noted, previous discussions of the comparability of this or other cohorts in the VOC dataset have not been seen. To investigate the sensitivity of results to the 1999/2000 cohort, linear QR models were rerun without these data. While this lessened the rate of decrease, differences were generally small, e.g., slopes changed by less than 30% for all VOCs and quartiles except benzene and toluene (0.75 quantile), BTEX (0.5 quantile), styrene (0.75 and 0.95 quantiles), chloroform and ∑THM (0.95 quantile), and few coefficients differed statistically (based on Wald tests assuming nil covariance between the two slopes) except benzene, toluene, o-xylene, and BTEX (0.5 quantile), benzene, toluene, bromoform, and PCE (0.75 quantile) (Appendix E). Bromoform and PCE at the 0.5 quantile also showed differences, but these were attributable to low DFs and are not meaningful. In summary, long–term trends were not strongly dependent on the 1999/2000 data, and thus these data were kept in subsequent analyses.

A second sensitivity analysis was undertaken that used piece-wise linear QR models allowing changes in trend over the study period. As before, models were adjusted for solvent–related occupations and cotinine. QR model results using a knot at 1999/ 2000 are shown in Appendix F. (Knots at other locations provided poor fits.) This analysis indicates that for most VOCs, declines from 1988 through 2000 were either not statistically significant or considerably smaller than declines from 1999/2000 through 2004, and that several VOC increased over the 1988–2000 period (including benzene and chloroform at the 0.5 quantile, benzene, toluene, styrene and chloroform at the 0.75 quantile, and benzene, m, p-xylene, styrene, and chloroform at the 0.95 quantile). Declines in the second period (shown as Slope2 in Appendix F) were reasonably consistent for the aromatic VOCs and chloroform, and faster than those from the linear QR models that spanned the entire period (Table 2). Overall, the piecewise QR models are similar to results in Table 3, and likewise suggest that reductions in blood VOC levels were largely accomplished from 1999/2000 onward. However, the piecewise models are less robust than the linear QR model since slopes for each time period use only three cohorts (or time points), and sometimes only two in the first period (1988–2000) since portions of the 1988–1991 data were omitted, and since they depend strongly on the 1999/2000 cohort data, which have several anomalies as noted previously. Moreover, trends in ambient concentrations for most VOCs do not support this steeper decline, as discussed below.

The third sensitivity analysis compared both linear and piecewise regression models with and without adjustments for strata and clusters. This showed only small differences in most cases: standard errors were larger for most VOCs, however, differences were significant for only BTEX among the linear models, and for DBCM, bromoform and PCE among the piecewise models. Although we cannot account for NHANES’ cluster sampling protocol in the QR models, these results suggest that the QR model results are reliable.

In summary, VOC levels in the NHANES blood samples substantially declined over the 15 year period. While piecewise models suggest that exposures to some VOCs did not decrease in the 1990’s and then rapidly declined in the early 2000’s, this may be driven by anomalies in the NHANES data, as discussed below.

3.2.1. Interpretation and reliability of trends

Many factors can affect the interpretation and representativeness of the NHANES data. First, while each cohort was designed to be nationally representative, biases might result from unknowingly over–sampling populations that are more exposed, genetically special (e.g., unable to rapidly clear VOCs), or otherwise not representative. As noted earlier, only minor group differences were seen among the demographic variables, literature discussing biases has not been identified, and while genetic differences can affect results, the biomarker documentation does not specify any such factor that affects the interpretation of VOC measurements in blood (ACGIH, 2001). Second, statistical variation is inherent in any sampling program and some cohorts had smaller PSU and sample sizes, but considering the NHANES sample sizes, this should not cause systematic biases. Third, whether the NHANES blood measurements represent valid exposure measures could be questioned, and indeed the approximate nature of these biomarkers was indicated by only modest correlation with air samples and the rapid clearance in the blood (discussed earlier). In this case, however, a bias towards the null (no trend) would be the likely outcome, which was not seen. Fourth, changes in protocols, including the air sampling conducted in the 1999/2000 cohort, the shift from NHANES III to continuous NHANES, or some other unknown study element, could affect results. We did identify NHANES data that appears suspect, and either excluded it or used sensitivity analyses to obtain confirm interpretations. Nothing emerged that could explain the observed patterns.

Several independent findings support the long–term VOC exposure trends derived from NHANES. First, the NATA emission inventory, while including only a few of the VOCs in measured in NHANES, reports that emissions of several VOCs increased in the 1990’s, e.g., benzene increased from 337,000 to 410,000 T yr⁻¹ from 1996 to 2002, and chloroform increased very markedly from 3310 to 15,139 T yr⁻¹ from 1996 to 1999; Appendix G). Annual average ambient concentrations predicted by NATA, spatially averaged, show negligible movement from 1996 to 1999 for benzene, chloroform, PCE and 1–4 DCB, and decreases of 3.9–18% per year for benzene, toluene, xylene and PCE from 1999 to 2002. These data support some of the piecewise trends, and also the high levels of benzene and chloroform seen in NHANES in 1999/2000, however, exposure analyses using emission inventories have limitations, as discussed in the Introduction.

Ambient air monitoring provides a more direct exposure measure. PAMS data are summarized in Appendix H. For the 2001–4 period, annual mean concentrations of benzene, toluene, ethylbenzene and o-xylene in the UATMP network decreased by 11–20% per year, and by 7–11% per year in PAMS. Thus, recent UATMP and PAMS trends are roughly similar, though UATMP concentrations are lower. Considering the older (1993–1999) PAMS data, annual mean concentrations of aromatic VOCs decreased from 4.4% per year (toluene) to 11% per year (styrene), and for five of the six VOCs measured, the rate was half of that seen in the 1999–2004 period. Issues regarding the spatial and temporal coverage of PAMS data were discussed in the Introduction. The AQS data may be more revealing, and annual means of the nine VOCs common to NHANES are tabulated and plotted in Appendix I. Regression analyses show approximately linear decreases of 5–7% per year for benzene, toluene, ethylbenzene and styrene from about 1990 to 2004. Trend plots show comparable long–term decreases and hints of somewhat accelerated trends since 2000 for m, p-xylene, o-xylene and 1,4–DCB. Chloroform shows a dramatic 21% per year decrease from 1990 to 1994, and then a flat trend. PCE levels decrease by 6.7% per year, although the trend is erratic. While ambient measurements too have limitations as exposure indicators, the national level data show that ambient concentrations of many VOCs have declined in a linearly over 15 years, and the rate appears slightly faster than those based on the NHANES exposure data. For several VOCs, some evidence suggests swifter declines after 2000, however, the ambient data does not reflect the high levels of benzene and chloroform in the 1999/2000 NHANES blood data.

In summary, ambient and emission data for most VOCs show strong downward trends from about 1990 through 2004. Regarding indoor exposures, national–level corroborating evidence is unavailable, however, there is linkage with ambient data in that outdoor concentrations represent a “floor” for indoor levels, and because the emission controls on fuels and vehicles that lower ambient VOC concentrations will also reduce exposures while commuting and in buildings with attached garages (Batterman et al., 2006). We next examine trends of individual VOCs.

3.2.2. Benzene

Over the 15 year study period, benzene exposures in NHANES declined by 3.3–4.3% per year, depending on the quantile. As noted, benzene trends matched pattern 1, with relatively consistent decreases at all quantiles, which parallel some of the emission and airborne concentration trends. Benzene was listed as a hazardous air pollutant by U.S. EPA in 1977 and as a carcinogen in 1986, and many emissions have been inventoried and regulated. U.S. emissions fell from 493,000 to 386,000 T yr⁻¹ between 1990–1993 and 2005 (US EPA, 2009b), representing a 1.5% per year decrease. On–road vehicle emissions, the single largest source category, declined faster, from 312,000 to 143,000 T yr⁻¹ or 3.6% per year. Further restrictions of benzene content in gasoline were issued in 2007, and additional reductions in mobile source air toxics emissions (including benzene) are anticipated (US EPA, 2010d). Benzene is metabolized fairly rapidly with a half–life in blood of about 8 h (Brugnone et al., 1992).

Inhalation exposure to benzene has been extensively reviewed (ATSDR, 2007a). Ambient measurements declined by 4.5–4.9% per year from 1994 to 2008; medians dropped from 2.10 to 0.79 µgm⁻³; and 90th percentile levels fell from 5.03 to 1.59 µgm⁻³ (US EPA, 2009a). Urban concentrations fell faster, e.g., PAMS data show 8.4%, 7.2%, and 6.9% per year declines at 0.5, 0.75, and 0.95 quantiles from 1993 to 2004 (Appendix H). Since few indoor sources exist other than smoking, benzene concentrations in outdoor, indoor and personal air can be similar (Kinney et al., 2002), however, an attached garage can elevate residential levels (Batterman et al., 2006). Differences in biomarker and ambient trends are reflected by the relatively low correlation between blood and personal airborne levels (r = 0.24, Appendix D). Occupational exposures in many settings have substantially declined, e.g., median personal concentrations of laboratory technicians at a refinery dropped from 319 to <32 µg m⁻³ from 1977 to 2005 (Panko et al., 2009), however, national statistics on occupational exposures are unavailable. As mentioned, tobacco smoke is an important exposure source (Wallace et al., 1987), and about 50% of benzene exposure in the U.S. has been apportioned to active and passive smoking (ATSDR, 2007a). However, NHANES data continued to show declines in each quantile after cotinine adjustment. Overall, the trends suggest that reductions in population exposure, as reflected in NHANES, have been driven largely by reductions in gasoline– and vehicle–related emissions.

3.2.3. Toluene

Over the 1988 to 2004 period, toluene exposures decreased by 4.7–5.7% per year, depending on the quantile. Like benzene, toluene reductions fit pattern 1 (consistent decreases across quantiles), which indicates improved control of general exposures, e.g., vehicle exhaust, as well as high–concentration exposures, e.g., architectural paints, which are now limited in VOC contents to 250 and 500 g L⁻¹ for flat coatings and graphic arts paints, respectively (US EPA, 1998). Toluene is one of the more prevalent components associated with vehicles and, unlike benzene, many household products contain and emit toluene. NATA emissions decreased from 996,443 to 884,066 T yr⁻¹ between 1999 and 2002, or 3.8% per year, on–road emissions decreased from 460,240 to 428,672 T yr⁻¹, or 2.3% per year (Appendix G) (US EPA, 1999; 2002), and average ambient predictions declined from 3.0 to 2.5 µg m⁻³, or 5.2% per year (Appendix J). Ambient concentration at PAMS sites decreased by 6.4–8.5% per year, depending on quantile (Appendix H), while annual means in the AQS data declined by 5.7% per year (Appendix I). Like benzene, blood and airborne levels had only modest correlation (r = 0.26, Appendix D). Toluene’s half–life in blood is short, about 4.5 h (Brugnone et al., 1986), thus blood levels tend to reflect current exposures.

3.2.4. Other BTEX compounds

QR results for the remaining BTEX compounds for the 1988–2004 period showed significantly downward trends that tended to fit pattern 2 (rapid decreases at upper quantiles), even after adjustment for cotinine (Tables 2 and 3). Ethylbenzene, m, pxylene, o-xylene, and styrene concentrations in blood decreased by 2.5–5.6% per year at each quantile. The composite BTEX exposure showed consistent decreases across quantiles in the same period; benzene and toluene contribute disproportionately to this indicator. The half–life of ethylbenzene in blood is very short (<1 h) (Adams et al., 2005; ATSDR, 2007b); xylenes are reported to have biphasic half–lives: 0.5–1 h initially, followed by 20–30 h (US EPA, 2003b); and styrene has biphasic half–lives of 0.58 and 13 h in blood (ATSDR, 2007c). Thus, blood tends represent only recent exposures. Correlation coefficients between personal air and blood for ethylbenzene, m, p-xylene and o-xylene in the 1999/2000 NHANES cohort were 0.35, 0.38, and 0.36, respectively, higher than seen for benzene and toluene (Appendix D).

In the NATA database, nationwide emissions of o– and m, pxylene fell from 712,084 to 595,241 T yr⁻¹ between 1999 and 2002 (Appendix G), or 5.5% per year, and on–road vehicle emissions decreased from 269,500 to 247,765 T yr⁻¹, only 2.7% per year (US EPA, 1999; 2002). Ambient measurements fell faster, e.g., median levels of aromatic VOCs in PAMS fell by about 9% per year from 1993 to 2004 (Appendix H), and AQS means fell by 5.8–6.4% per year, with faster declines after 2000 (Appendix I). Thus, ambient levels fell more rapidly that the roughly 4% per year seen for NHANES blood VOC levels from 1988–2004 (Table 3), but less rapidly than the more recent (1999–2004) blood VOC data. The divergence suggests that reductions of indoor VOC sources trailed outdoor reductions by perhaps a decade.

3.2.5. THMs

Chloroform was the most prevalent THM. With the 1999/2000 data included, levels declined rapidly at upper quantiles (pattern 2), while comparable reductions of about 4% per year were seen across quantiles when comparing starting and ending cohorts (Tables 2 and 3). BDCM, DBCM and bromoform showed rapidly decreases at central quantiles over the study period. Due to low DFs, trends at lower percentiles could not be evaluated (Appendix A). Over the 15 year study period, concentrations decreased by 5.0–7.9% per year for the median, and by 3.0–7.5% per year for upper quantiles.

Exposures of individual THMs, including chloroform, are likely to be highly correlated, although this was not consistently shown in the NHANES blood measurements (Appendix C). This can be explained, in part, by the rapid clearance of THMs from blood, e.g., half–lives of about 0.5 h (Ashley and Prah, 1997), and a biphasic clearance pattern is reported for chloroform with half–lives of 9–21 min and then 86–96 h (ATSDR, 1997a). Given these rates, the blood data represent only recent exposures. Chloroform showed a moderate but significant correlation (r = 0.38) between blood and personal air concentrations (Appendix D).

NATA emissions of chloroform jumped from3,310 to 15,139 T yr−1 from 1996 to 1999, or 119% per year, followed by a decline in 2002 to 6,805 T yr⁻¹, or 18% per year (Appendix G) (US EPA, 1996; 1999; 2002). The dramatic increase from 1996 to 1999 is likely due to changes in inventory procedures (US EPA, 1999). Predicted ambient concentrations increased by 0.9% per year from 1996 to 1999, and then decreased by 1.7% per year (Appendix J). Interestingly but perhaps serendipitously, the period of highest chloroform emissions (1999) corresponded to the highest blood measurements in NHANES (Appendix B). In the mid–1990s, Maximum Achievable Control Technology standards limited emissions of halogenated solvents at industrial and waste treatment facilities (US EPA, 2000). About the same time, maximum contaminant levels on THMs in drinking water were imposed, which is probably the largest exposure source (both ingestion and inhalation) of THMs for the general population. (NATA estimates do not account for THM emissions in to drinking water, but the NHANES blood data do account for the ingestion pathway.) Lowering THMs in drinking water is expected to decrease levels at all quantiles (pattern 1). Ambient concentrations of chloroform show a trend unique among the VOCs: early decreases of nearly 21% per year for the 1990–1994 period, followed by a flat trend from 1995 onward (Appendix I). Exposures of the brominated THMs had inconsistent trends, which are attributed to analytical uncertainties resulting from low concentrations (generally 10 times lower than chloroform).

3.2.6. Other VOCs

Styrene exposures significantly decreased at 0.5 and 0.75 quantiles, e.g., median levels fell by 3.8% per year over the study period (Table 3), but much faster (18% per year) from 1999 to 2004. Serum cotinine and blood styrene levels in NHANES were correlated (r = 0.49), but QR models adjusted for cotinine levels continued to showed a declining trend (Table 2). (NATA only included styrene data in 2002.) Ambient concentrations of styrene in PAMS declined by about 8% per year, depending on quantile, over the 1993–2004 period, while AQS means declined by 5.5% per year, though the data showed considerable scatter (Appendices H and I). Styrene is used in reinforced plastics manufacturing, and indoor emissions can occur from building materials and tobacco smoke (ATSDR, 2007c). It has biphasic half–lives of 0.58 and 13 h in blood (ATSDR, 2007c).

1,4–DCB decreased by 3.5% per year over the 15 year study period (Table 3). Decreases were more rapid at median quantiles, (pattern 3), and the 0.95 quantile result was not significant (Table 2). 1,4–DCB is widely used in mothballs, other pest repellents and toilet–deodorizer blocks, and airborne levels in occupational settings occasionally reach very high levels, e.g., 4350 mg m⁻³ in amono– and dichlorobenzene manufacturing plant (IARC, 1982). In the US, mean and median indoor 1,4–DCB concentrations were 24 µg and 1.7 µg m⁻³, respectively (ATSDR, 2006); the large difference reflects the highly skewed distribution of this VOC. A Japanese study found high indoor levels (mean = 114 µg m⁻³), far above outdoor levels (3.4 µgm⁻³) (Azuma et al., 2007). 1,4–DCB’s half–life is estimated to be 7.1–8.1 h in rats (no human data are available (Hissink et al., 1997). NATA emission estimates of 1,4–DCB fell from 12,794 to 7244 T yr⁻¹ between 1999 and 2002, or 15% per year (Appendix G) (US EPA,1999; 2002). Ambient concentrations are low, and median concentrations among 11 sites declined by 5.0% per year from 1995 to 2005, and by 10% per year among 32 sites from 2000 to 2005 (McCarthy et al., 2007). Among the AQS VOCs, 1,4–DCB showed the strongest decrease after 2000 (Appendix I). As noted, 1,4–DCB had the highest air–to–blood correlation coefficient among the NHANES VOCs (r = 0.65, Appendix D), thus exposures tend to reflect personal air concentrations.

PCE exposures declined by 3.2–6.4% per year, depending on quantile, over the 15 year study period, and decreases at upper quantiles were faster (pattern 2) (Tables 2–3). PCE’s half–life in blood, 12–16 h (ATSDR, 1997b), is the longest among the VOCs, and its air–blood correlation was relatively high (r = 0.62, Appendix D). NATA emissions increased by 2.0% per year, from 44,100 to 46,793 T yr⁻¹, between 1996 and 1999, followed by a 8% per year decrease to 35,613 T yr⁻¹ in 2002 (Appendix G) (US EPA, 1996; 1999; 2002). However, predicted ambient concentrations decreased slightly, 1.5% per year, between 1996 and 1999, and then by 18% per year between 1999 and 2002 (Appendix J). Nationwide emission data before 1993 are not available. MACT standards for dry cleaners, perhaps the major urban source of PCE (US EPA, 2010e), were initiated in 1993. Although the AQS means show considerable variation (Appendix I), the long term decline of ambient concentrations nearly exactly corresponds to the rate seen in blood.

4. Conclusions

This study examined VOC exposures from 1988 to 2004 using concentrations measurements in blood drawn from five cohorts of NHANES, a large and nationally representative sample of U.S. adults. There is no question that VOC exposures decreased over this period, however, the rate of decrease depends on the both the VOC and the quantile. Using quantile regression models, three patterns were discerned: exposures of benzene, toluene, BTEX and, with less confidence, ΣTHMs and chloroform, had similar decreases at all quantiles (pattern 1); ethylbenzene, m,p-xylene, o-xylene, styrene and PCE levels decreased fastest at upper quantiles (pattern 2); and 1,4–DCB declined faster at central quantiles (pattern 3). Because the sample included participants with a wide range of occupations and exposures, upper quantile exposures may reflect occupational exposure, while lower quantiles arise from general environmental sources. There is less certainty regarding the nature of the exposure trends. Linear models yielded reductions of 2.5–6.4% per year for most VOCs, a robust result that is consistent with ambient trends, described below. Shorter term trends, evaluated using piecewise models and other analyses, suggest that several VOCs had smaller changes through the 1990’s, followed by swifter reductions in subsequent years, however, these trends may be driven by previously unreported anomalies in the NHANES data that affected the 1988 through 2000 cohorts.

VOC emissions and ambient concentrations were compared to the biomonitoring data. For most VOCs, reported emissions decreased more slowly (e.g., 4–6% per year for toluene and xylene from 1999 to 2004) than median exposures. However, for most VOCs, long term trends of ambient concentrations decreased more rapidly than the NHANES exposure data. Exposure, emission and concentration trends may diverge, especially for VOCs with strong indoor sources, e.g., chloroform and 1,4–DCB. These differences suggest the importance of indoor emission sources, smoking, occupation, personal activities and other factors on exposure, in addition to emissions and ambient concentrations.

Internal checks on the validity of the NHANES measurements were made by comparing blood and personal sampling measurements collected in the 1999/2000 cohort, and by comparing results across cohorts. The low to moderate correlation found can be explained by NHANE’s experimental design, the rapid clearance of most VOCs from blood, and other factors. It should be noted that data were insufficient to estimate trends for BDCM, DBCM and bromoform, and also that portions of the 1988–1991 through 1999/ 2000 VOC data appear unreliable. Still, the NHANES measurements are unique and valuable in providing a 15 year history of population exposure to VOCs in the U.S.

Appendix Supplementary material

Supplementary data associated with this article can be found, in the online version, at doi:10.1016/j.atmosenv.2011.06.016.