Secondary Logo

Journal Logo

Exposure to Ambient and Nonambient Components of Particulate Matter

A Comparison of Health Effects

Ebelt, Stefanie T.*; Wilson, William E.; Brauer, Michael

doi: 10.1097/01.ede.0000158918.57071.3e
Original Article

Background: Numerous epidemiologic studies report associations between outdoor concentrations of particles and adverse health effects. Because personal exposure to particles is frequently dominated by exposure to nonambient particles (those originating from indoor sources), we present an approach to evaluate the relative impacts of ambient and nonambient exposures.

Methods: We developed separate estimates of exposures to ambient and nonambient particles of different size ranges (PM2.5, PM10–2.5 and PM10) based on time-activity data and the use of particle sulfate measurements as a tracer for indoor infiltration of ambient particles. To illustrate the application of these estimates, associations between cardiopulmonary health outcomes and the estimated exposures were compared with associations computed using measurements of personal exposures and outdoor concentrations for a repeated-measures panel study of 16 patients with chronic obstructive pulmonary disease conducted in the summer of 1998 in Vancouver.

Results: Total personal fine particle exposures were dominated by exposures to nonambient particles, which were not correlated with ambient fine particle exposures or ambient concentrations. Although total and nonambient particle exposures were not associated with any of the health outcomes, ambient exposures (and to a lesser extent ambient concentrations) were associated with decreased lung function, decreased systolic blood pressure, increased heart rate, and increased supraventricular ectopic heartbeats. Measures of heart rate variability showed less consistent relationships among the various exposure metrics.

Conclusions: These results demonstrate the usefulness of separating total personal particle exposures into their ambient and nonambient components. The results support previous epidemiologic findings using ambient concentrations by demonstrating an association between health outcomes and ambient (outdoor origin) particle exposures but not with nonambient (indoor origin) particle exposures.

From the *Department of Environmental and Occupational Health, Rollins School of Public Health, Emory University, Atlanta, GA; the †National Center for Environmental Assessment, U.S. Environmental Protection Agency, Research Triangle Park, North Carolina; and the ‡School of Occupational and Environmental Hygiene, The University of British Columbia, Vancouver, BC, Canada.

Submitted 28 January 2004; final version accepted 24 January 2005.

Supported in part by a research grant from the British Columbia Lung Association, by an Infrastructure Program (UBC Centre for Health and Environment Research) award from the Michael Smith Foundation for Health Research, and by Contract 2C-R368-NASA from the U.S. Environmental Protection Agency.

Correspondence: Michael Brauer, School of Occupational and Environmental Hygiene, The University of British Columbia, 2206 East Mall, Vancouver BC V6T1Z3 Canada. E-mail:

Back to Top | Article Outline


Click on the links below to access all the ArticlePlus for this article.

Please note that ArticlePlus files may launch a viewer application outside of your web browser.

Adverse health effects have been repeatedly associated with outdoor concentrations of airborne particulate matter (PM) in epidemiologic studies.1,2 Nearly all studies of acute effects have used ambient monitoring data to estimate exposure. With individuals spending up to 90% of their time indoors, the validity of using ambient concentrations as an accurate estimate of exposure has raised concerns because exposure misclassification could bias epidemiologic results.3 A person's total personal exposure to PM (T) has an ambient exposure (A) component resulting from exposure to ambient PM while outdoors and to ambient PM that has infiltrated indoors while the person is indoors, plus a nonambient exposure component (N) resulting from exposure to indoor-generated PM while in various indoor environments.4 The ambient concentrations (C) and the total personal exposures (T) can be measured directly, whereas the ambient and nonambient exposures cannot be measured directly and must be estimated.

Many studies have demonstrated that individual personal exposures to PM are poorly correlated spatially with ambient concentrations.5 Longitudinal exposure assessment studies of PM and specific PM components with repeated measures have found higher correlations between personal exposures and ambient concentrations.6–8 However, such studies have also identified substantial between-subject variability in personal ambient correlation coefficients, including some individuals with low correlations.9–11 Indoor monitoring and activity data has indicated that much of this variability can be accounted for by indoor sources of particles and indoor–outdoor air exchange rates.8,9

Although evidence of high longitudinal correlations between personal exposures and ambient concentrations lends support to the use of ambient concentrations in time-series epidemiologic studies, the relative toxicities of particles of ambient and nonambient origin have not been systematically evaluated. Indoor sources can emit high concentrations of particles,12,13 including particles produced in combustion processes.14–16 However, ambient and nonambient particles differ in sources and sizes13 and likely differ in composition and biologic properties as well.17 Therefore, there is considerable interest in understanding the potential health impacts associated with ambient and nonambient particles.

This article presents an approach to address 2 hypotheses: 1) the reduced exposure misclassification resulting from the use of ambient exposures instead of ambient concentrations will provide more precise and stronger estimates of effect in epidemiologic analyses; and 2) ambient and nonambient exposure will demonstrate different associations with health outcomes. To illustrate the use of improved exposure estimation, we address these hypotheses in an extended analysis of a repeated-measures panel study conducted in Vancouver, Canada, with 16 persons who had chronic obstructive pulmonary disease (COPD).10,18 Relationships between total personal exposures (PM2.5 and sulfate) and ambient air concentrations (PM2.5, PM10, and sulfate) measured during the study have been reported.10 Associations of respiratory and cardiac outcomes with these measured exposure indicators have also been reported.18 In the previous analysis, weak associations were observed between several of the outcomes and ambient PM10, and to a lesser degree with ambient PM2.5, but total personal exposure was not associated with any of the health outcomes. Here, the earlier analysis is expanded by developing mixed-effect regression models and by evaluating additional exposure indicators.

A unique aspect of this analysis is the use of separate exposure indicators for the ambient and nonambient components of total personal exposure. Nonambient PM2.5 exposures and ambient PM2.5, PM10–2.5, and PM10 exposures were estimated based on time-activity data and the use of particle sulfate measurements as a tracer for indoor infiltration of ambient particles with a size distribution similar to that of sulfate.4,19,20 Associations of health outcomes with the various exposure indicators were determined using mixed models. The associations of health outcomes with estimated ambient and nonambient exposures were compared with those using measured total personal exposures and measured ambient concentrations.

Back to Top | Article Outline


Study Population

Details of primary data collection have been described previously.10,18 Briefly, 16 patients with COPD underwent 5 to 7 repeated 24-hour monitoring periods spaced approximately 1.5 weeks apart, during which exposures, respiratory and cardiovascular health endpoints were measured, for a total of 104 subject-days of data. The study was conducted in Vancouver, Canada, during the summer of 1998 and all subjects (mean age, 74 years; range, 54–86 years) were residents of the Greater Vancouver Regional District. All subjects were current nonsmokers. Only 1 subject lived with a smoker who agreed to smoke outdoors or not at all on each of the subject's sampling days. Ethical approval was obtained from the Clinical Research Ethics Board, University of British Columbia. Informed consent was obtained in writing from all study subjects.

Back to Top | Article Outline

Air Sampling

Subjects wore a PM2.5 sampler that provided 24-hour integrated personal PM2.5 exposure data. Subjects also completed time-activity diaries during each monitoring period. Daily ambient monitoring was conducted at 5 sites within the study area. Ambient PM10 was measured continuously using a tapered elemental oscillating microbalance and 24-hour ambient PM2.5 was collected using Harvard Impactors timed to overlap with the personal monitoring periods. All PM2.5 filters were analyzed for mass and sulfate (SO42−).

Back to Top | Article Outline

Health Measurements

In addition to personal exposure monitoring, subjects wore an ambulatory electrocardiogram monitor for 24-hour recording of heart rhythm data. Spirometry and systolic/diastolic blood pressure measurements were collected before and after each 24-hour monitoring period and symptom questionnaires were administered at the end of each session. Arrhythmia, heart rate, and heart rate variability data were obtained by analysis of the electrocardiogram recordings by a certified electrocardiogram scanning technologist.

Back to Top | Article Outline

Estimation of Ambient and Nonambient Exposures

Based on the collected data, we developed an approach to separately estimate exposures to PM of ambient and nonambient origin based on the mass balance model and methods introduced previously.4 Technical aspects of the modeling approach are described in detail in the Appendix. Briefly, total personal exposures (T) were directly measured for PM2.5 mass (T2.5) and for sulfate (S_T2.5) using personal monitors worn by study participants. These measured personal exposures incorporate contributions from ambient (A) and nonambient (N) PM encountered both indoors and outdoors. Time-activity data describing the amount of time spent outdoors by study subjects and measurements of sulfate as a tracer of PM infiltration were then used to estimate the fraction of the measured ambient concentration (C) that resulted in ambient exposure (A). Nonambient exposure (N) was estimated by subtracting the estimated ambient exposure (A), described previously, from measured total personal exposure (T).

First, we developed estimates for the PM2.5 size fraction. Estimates for the ambient (A10–2.5) component of the coarse (PM10–2.5) fraction were developed by using the sulfate tracer measurements to estimate air exchange rates for each subject on each day and then combining these quantities with estimated infiltration factors and ambient concentrations of the coarse particle fraction (C10–2.5). The ambient coarse fraction concentration was estimated by subtracting measured ambient concentrations of PM2.5 (C2.5) from the measured ambient concentrations of PM10 (C10). In addition, we estimated ambient PM10 exposures (A10) from the sum of the ambient coarse fraction (A10–2.5) and ambient PM2.5 (A2.5) exposures. Finally, the nonsulfate components of the ambient PM2.5 concentration (NS_C2.5) and ambient PM2.5 exposure (NS_A2.5) were calculated by subtracting the measured sulfate concentrations and exposures from measured ambient PM2.5 concentrations and estimated ambient PM2.5 exposures. With this procedure, 6 new parameters were derived from 6 variables that were measured or calculated (Table 1).



Back to Top | Article Outline

Data Analysis

The selection of outcome variables and identification of outliers has been described in detail previously.18 Consistent with the previous analysis of these data, we used forced expiratory volume (in milliliters) in the first second (FEV1) as a measure of respiratory health, considering both postsample FEV1 and the change in FEV1 (postsample FEV1 − presample FEV1) (ΔFEV1). Postsample blood pressure recordings (systolic and diastolic), which were highly correlated with presample recordings, were chosen for further analysis. From the electrocardiographic recordings, 24-hour mean values over each subject-day were abstracted. Supraventricular ectopy (SVE, in beats per hour [bph]) was chosen as an arrhythmia variable and indicates abnormal impulse formation. We used heart rate (HR, in beats per minute [bpm]) as a basic rhythm variable. Heart rate variability (HRV) was determined by time-domain analysis, which examines the standard deviation (SD) of sinus (ie, nonarrhythmic) R-R intervals (measured in milliseconds [ms]) over the sampling period in a variety of forms. Standard deviation of normal–normal beats (SDNN), derived from direct measurements of the beat-to-beat intervals, was used as a measure of overall HRV. R-MSSD is the square root of the mean squared differences of successive normal R-R intervals over the duration of the entire electrocardiogram and was used as a measure of short-term HRV.

Distributions of these 8 health parameters were evaluated; for SVE, a natural log transformation was used, whereas no transformations were used for the other variables. We reviewed histograms of the health outcome data for identification of outliers. These plots indicated 1 extreme value for FEV1. Review of questionnaire data indicated that the subject had not used his medications before spirometry on this 1 occasion, but he did use his medications before all other lung function testing. Therefore, this point was removed from the dataset and not included in subsequent data analyses. In addition, 7 diastolic blood pressure measurements that were recorded as zero by our field staff as a result of difficulties in determining the correct blood pressure level were excluded; this effectively excluded 1 subject from the analyses for this outcome and it affected 1 sample for another subject. After removing these samples, distributions of all health outcomes fell within 3 to 4 standard deviations of the mean value.

Associations between health outcomes and exposures were evaluated with mixed-effects models using the MIXED procedure in SAS V8 (SAS Institute Inc., Cary, NC). We modeled health outcomes as the dependent variable, exposures were modeled as fixed effects, and subjects were modeled as random effects. Because the aim was to compare effect estimates across the various exposure parameters for each outcome, regression analyses for each outcome were restricted to subject-days for which data was available for all exposure parameters. After this restriction, the total sample size per outcome ranged from 76 to 98 with 3 to 7 repeated measurements per subject. Distributions of the residuals were inspected graphically for each regression. These plots demonstrated rather uniform distribution of the residuals with no apparent influential observations. Histograms and time-series plots of the exposure metrics were also inspected for potential influential data points. Unusually high coarse PM concentrations were observed on the first day of the study (>10 standard deviations from mean coarse PM concentration of the remainder of the study), and therefore analyses of all exposure metrics were conducted both with and without the first study day. For the remainder of the study, no further outliers were identified with all data points falling within 3 to 6 standard deviations of each pollutant's mean concentration.

Back to Top | Article Outline


Exposure Estimation

Abbreviations and summary statistics of the 12 exposure metrics, including the 5 previously reported variables, are presented in Table 1. On average, ambient fine particles, indexed by PM2.5 (C2.5), contributed greater mass (11.4 μg/m3) to ambient PM10 (C10) than did the thoracic coarse fraction, indexed by PM10–2.5 (C10–2.5, mean = 5.6 μg/m3). This was observed to be the case on most days during the study period (May–September 1998; Electronic Appendix Fig. 1, available with the electronic version of this article). Additionally, the majority (82%) of the ambient PM2.5 mass was composed of nonsulfate components (NS_C2.5). Personal PM2.5 exposure (T2.5) (mean = 18.5 μg/m3) was largely composed of nonambient particle exposure (N2.5) (mean = 10.6 μg/m3), as expected given the high amount of time subjects spent indoors (up to 90% per 24-hour period). Similar to the findings for ambient concentrations, the nonsulfate component of exposure (NS_A2.5) formed 82% of the ambient PM2.5 exposure (A2.5) mass.

Back to Top | Article Outline

Correlations Between Exposure Metrics

Selected correlations between exposure measures are presented in a path diagram (Fig. 1). As reported previously,18 we found high correlations among measured ambient PM10, ambient PM2.5, ambient sulfate, and personal sulfate. Low correlations were found between personal PM2.5 and the other measured exposure variables. Among the noteworthy correlations with the new exposure estimates, total personal PM2.5 exposures (T2.5) and exposures to nonambient PM2.5 (N2.5) were highly correlated (r = 0.84), indicating that variations in total personal PM2.5 exposures were largely driven by nonambient factors; this finding coincides with the observation that nonambient PM contributed more than 50% to personal PM2.5 exposures. Exposures to ambient and nonambient PM2.5 were not correlated (r = −0.06), likely as a result of the different sources contributing to these exposures.



Ambient concentrations (C) and exposures to ambient PM (A) for each respective size class were highly correlated (r ≥0.71). However, between the size classes, ambient concentrations and ambient exposures of PM2.5 and sulfate were not highly correlated with their respective thoracic coarse particle (PM10–2.5) metrics. These observations are again likely the result of the differing sources that contribute to the 2 size fractions. Ambient PM10, composed of particles from both size classes, was highly correlated with both the fine and the thoracic coarse fractions (r ≥0.69). Whereas both ambient PM10 concentrations and exposures were more highly correlated with their corresponding PM2.5 metrics than with the PM10–2.5 metrics, the difference was greater for ambient exposures. For example, exposure to ambient PM10 (A10) was more highly correlated with exposure to ambient PM2.5 (A2.5) (r = 0.92) than with exposure to ambient coarse PM (A10–2.5) (r = 0.72). Exposure to ambient PM2.5 (A2.5) is expected to be a larger fraction of exposure to ambient PM10 (A10) than the ambient concentration of PM2.5 (C2.5) would be of ambient PM10 (C10) because the infiltration factor for PM2.5 is greater than that of PM10–2.5. Exposure to ambient PM2.5 (A2.5) was 77% of exposure to ambient PM10 (A10) compared with the ambient concentration of PM2.5 (C2.5) being 67% of the ambient concentration of PM10 (C10).

Back to Top | Article Outline

Relationships Between Exposure Metrics and Health Outcomes

Table 2 presents abbreviations and summary statistics for each of the health outcome measurements. Effect estimates for selected outcomes are presented graphically in Figure 2. Electronic Appendix Table 1 (available with the electronic version of this article) displays the effect estimates for all health outcomes for interquartile range increases in the various exposure indicators.





Back to Top | Article Outline

Transported Asian Dust

High concentrations of PM10 were observed in Vancouver on April 30, 1998 (the first day of the study) as a result of transport of wind-blown dust from deserts in Asia.21 On this day, PM10–2.5 was unusually high (32.2 μg/m3 compared with a mean concentration during the study period of 5.6 μg/m3) and exceeded PM2.5 by 13.6 μg/m3. As a result of the atypical nature of the first study day, analyses were conducted with and without this day's data. This exclusion did not affect regression results for any outcomes when using either PM2.5 or sulfate exposure metrics. The exclusion did, however, alter effect estimates for ambient PM10 and PM10–2.5 (especially for ΔFEV1, SVE, and HR outcomes) by increasing their magnitude in the previously seen directions. Because it is likely that the composition of the wind-blown dust from Asian deserts is different from ambient coarse PM in Vancouver, this observation (albeit with very limited data) suggests that chemical composition as well as mass are important properties in determining the relationship between PM and health effects. All results presented in the tables and figures exclude the Asian dust event.

Back to Top | Article Outline

Exposures to Particles of Ambient Versus Nonambient Origin

Except for models predicting post-FEV1, total personal PM2.5 (T2.5) exposures and exposures to nonambient PM2.5 (N2.5) were not associated with any of the health end points. For post-FEV1, total personal PM2.5 (T2.5) exposures and nonambient PM2.5 exposures (N2.5) both demonstrated effects opposite to those of the other exposure indicators and to prior hypotheses (increased FEV1 with increased exposure). Based on previous research, we expected particle exposure to be associated with decreased lung function, blood pressure, and HRV and with increased HR and SVE.22 In parallel with these hypotheses, for most outcomes (including post-FEV, ΔFEV1, systolic blood pressure, diastolic blood pressure, SVE, HR, and SDNN), exposure to ambient PM2.5 (A2.5) provided greater effect estimates in the expected directions than did exposure to nonambient PM2.5 (N2.5).

Back to Top | Article Outline

Ambient Concentration versus Exposure to Ambient Particles

For most outcomes, the effect estimates between PM indicators and health outcomes were similar in direction and magnitude for ambient concentrations (C10, C10–2.5, C2.5, and NS_C2.5) and their respective estimated ambient exposures (A10, A10–2.5, A2.5, and NS_A2.5). However, effect estimates for ambient exposure indicators (A) were generally equal to or larger than those for the respective ambient concentration indicators (C) for post-FEV, ΔFEV1, systolic blood pressure, SVE, and SDNN. Confidence intervals around the effect estimates for A indicators were also less likely to cross the null value, especially for ΔFEV1 (all 4), systolic blood pressure (exposure to ambient PM2.5, A2.5, and exposure to nonsulfate ambient PM2.5, NS_A2.5), and SVE (exposure to ambient PM10, A10; exposure to ambient PM2.5, A2.5; and exposure to nonsulfate ambient PM2.5, NS_A2.5). An exception to this trend was for R-MSSD, whose effect estimates for the ambient concentrations of PM10 (C10), the ambient concentration of PM2.5 (C2.5), and the ambient concentration of nonsulfate PM2.5 (NS_C2.5) were larger than those of their respective ambient exposures. Overall, however, these results support the hypothesis that the use of ambient exposures rather than ambient concentrations results in larger and more precise effect estimates in epidemiologic analyses of health outcomes.

Back to Top | Article Outline

Sulfate versus Nonsulfate Components of PM2.5

Effect estimates from regressions using exposure to ambient PM2.5 (A2.5) and exposure to nonsulfate ambient PM2.5 (NS_A2.5) were almost identical, as was the case for the ambient concentrations of PM2.5 (C2.5) and nonsulfate PM2.5 (NS_C2.5). This was expected considering the very high correlations (r = 0.98) between these exposure metrics. In comparison, however, for most health outcomes, the results using sulfate (either the ambient concentration, S_C2.5, or exposure to ambient sulfate, S_A2.5) differed from the comparable nonsulfate results with exposure to nonsulfate ambient PM2.5 (NS_A2.5), usually showing a larger effect in the expected direction than exposure to ambient sulfate (S_A2.5or S_T2.5). For lung function and SVE in particular, the results for sulfate were closer to the null, supporting the notion that the sulfate component of ambient particles is not associated with these specific health effects. This observation is also supported by the toxicologic literature, which indicates very low relative toxicity for sulfate particles.23 These results suggest that nonsulfate ambient particles, at least in this study region, may have higher relative toxicity than sulfate particles for these health outcomes. Because sulfate concentrations and exposures were very low (ambient concentration of sulfate, S_C2.5 = 2 ± 1.1 μg/m3; exposure to ambient sulfate, S_A2.5 = 1.5 ± 0.9), the lack of an association could also have resulted from measurement error or might indicate the existence of a threshold in the health effects of exposure to sulfate.

Back to Top | Article Outline


The primary purposes of this study were: 1) to demonstrate a methodology that permits epidemiologic analyses using separate indicators for ambient and nonambient exposure, instead of being limited to the use of ambient concentrations or total personal exposure; and 2) to use this methodology to test 2 hypotheses. We found that that for several health outcomes, analyses with ambient exposures resulted in larger effect estimates with smaller confidence intervals than did analyses with ambient concentrations; these results support the first hypothesis, that ambient exposures provide more precise and stronger estimates of effect in epidemiologic analyses than do ambient concentrations. Our finding that the effect estimates for nonambient exposure were close to the null (or, in 1 case, in the opposite direction of the effects of ambient exposure) supports the second hypothesis that ambient exposure and nonambient exposure will demonstrate different associations with health outcomes.

The feasibility and usefulness of separating total personal exposure into its ambient and nonambient components has been demonstrated and has provided some interesting and useful new information. However, the specific results obtained should be interpreted cautiously and require replication. This study included a small sample and a limited number of measurements per subject conducted during 1 season only, because the collection of such detailed individual measurements is laborious. In addition, the lack of association with specific outcomes measured in the study does not necessarily indicate a lack of toxicity for that exposure variable. This study was limited to patients with COPD; other subgroups may respond differently and may exhibit different exposure relationships. Furthermore, only a limited set of health outcomes were measured.

Finally, like in previous studies conducted in Vancouver24 and consistent with the known continental sulfate concentration distribution,25 sulfate levels were low during this study. Therefore, some values of the attenuation factor, α, for sulfate, used as a surrogate for α for PM2.5, may have had some error, which would therefore increase the overall error in the estimated exposures. It would be useful to conduct studies similar to this one in communities with higher sulfate concentrations and to use other components to estimate α. For example, elemental or black carbon (which can be measured continuously with an athelometer) can be used as an infiltration tracer for homes without internal combustion. The differences between sulfate and nonsulfate PM2.5 suggest that it would be useful to apply source apportionment techniques to determine source category contributions for both ambient and nonambient exposures.26

Our study demonstrates the feasibility and usefulness of determining associations between health outcomes and estimated values of both ambient exposures and nonambient exposures, in addition to associations with measured values of ambient concentrations and total personal exposures. It is important to treat ambient exposure and nonambient exposure as independent predictors of PM-related health effects, because the particulate matter responsible for these 2 types of exposure differ in size, sources, chemical composition, and temporal patterns. As our findings indicate, they probably also differ in type and degree of toxicity.

In an analysis of exposure error in community time-series epidemiology, Zeger et al27 point out that exposures to nonambient PM will not bias effect estimates calculated using ambient concentrations (as a surrogate for ambient exposures) provided that ambient concentrations and the corresponding exposures to nonambient PM are independent. In our study, the correlation coefficient between the ambient concentration of PM2.5 and exposure to nonambient PM2.5 was −0.1, indicating that the 2 may be considered independent. This independence of ambient concentrations and exposure to nonambient PM is an important assumption in epidemiologic studies that use ambient concentrations as a surrogate for exposure; this independence should be evaluated in other studies. Zeger et al also point out that effect estimates will be biased by the difference between ambient concentrations and ambient exposures. The use of ambient exposures instead of ambient concentrations, as we have done in this article, avoids this bias. In addition, developing separate estimates of exposure to the various components of PM exposure reduces the overall exposure measurement error such that more accurate risk estimates can be obtained with more specific estimates than with surrogates incorporating multiple components that may have different relationships with outcomes.

The results incorporating new exposure metrics have provided new insights into the analysis of these data. Specifically, the new results indicate that the measured effect estimates were of greater magnitude for ambient concentrations and ambient exposures, including the thoracic coarse fraction of PM, than with total personal exposures or nonambient exposures. Although the toxic components of PM have not been exhaustively evaluated, recent studies have shown clear differences in the composition and source contributions of ambient and nonambient PM.26,28 Should additional studies confirm our findings, this would provide direction to focus on specific particle constituents produced in ambient air. Furthermore, verification of our findings would suggest that attention should be directed toward ambient sources and on the differential impacts of ambient PM on exposure. Finally, our results add general support for the use of ambient monitoring data in time-series studies.

The use of sulfate as an indicator of PM2.5 infiltration to estimate ambient and nonambient exposure separately, as well as the determination of the association of health outcomes with each exposure indicator, demonstrates a novel and useful approach to the examination of epidemiologic relations that can be of benefit when trying to identify important factors associated with the health effects of particles.

Back to Top | Article Outline


This article has been reviewed by the U.S. Environmental Protection Agency and cleared for publication. However, the views expressed in this paper are those of the authors and do not necessarily reflect the views or policies of the EPA. We thank Melanie Noullett, Mary Ross, and Lucas Neas for their useful comments.

Back to Top | Article Outline


1. Pope CA III, Bates DV, Raizenne ME. Health effects of particulate air pollution: time for reassessment? Environ Health Perspect. 1995;103:472–480.
2. Pope CA III. What do epidemiologic findings tell us about health effects of environmental aerosols? J Aerosol Med. 2000;13:335–354.
3. Vedal S. Ambient particles and health: lines that divide. J Air Waste Manag Assoc. 1997;47:551–581.
4. Wilson WE, Mage DT, Grant LD. Estimating separately personal exposure to ambient and nonambient particulate matter for epidemiology and risk assessment: why and how. J Air Waste Manag Assoc. 2000;50:1167–1183.
5. US Environmental Protection Agency. Air Quality Criteria for Particulate Matter. Research Triangle Park, NC: National Center for Environmental Assessment-RTP Office, 1996; Report No. EPA/600/P-95/001aF-cF.3v.
6. Janssen NA, Hoek G, Brunekreef B, et al. Personal sampling of particles in adults: relation among personal, indoor, and outdoor air concentrations. Am J Epidemiol. 1998;147:537–547.
7. Janssen NA, Hoek G, Harssema H, et al. Personal exposure to fine particles in children correlates closely with ambient fine particles. Arch Environ Health. 1999;54:95–101.
8. Rojas-Bracho L, Suh HH, Koutrakis P. Relationships among personal, indoor, and outdoor fine and coarse particle concentrations for individuals with COPD. J Expo Anal Environ Epidemiol. 2000;10:294–306.
9. Sarnat JA, Koutrakis P, Suh HH. Assessing the relationship between personal particulate and gaseous exposures of senior citizens living in Baltimore, MD. J Air Waste Manag Assoc. 2000;50:1184–1198.
10. Ebelt ST, Petkau AJ, Vedal S, et al. Exposure of chronic obstructive pulmonary disease patients to particulate matter: relationships between personal and ambient air concentrations. J Air Waste Manag Assoc. 2000;50:1081–1094.
11. Lioy PJ, Waldman JM, Buckley T, et al. The personal, indoor and outdoor concentrations of PM-10 measured in an industrial community during the winter. Atmos Environ. 1990;24:57–66.
12. Brauer M, Hirtle R, Lang B, et al. Assessment of indoor fine aerosol contributions from environmental tobacco smoke and cooking with a portable nephelometer. J Expo Anal Environ Epidemiol. 2000;10:136–144.
13. Abt E, Suh HH, Catalano PJ, et al. Relative contribution of outdoor and indoor particle sources to indoor concentrations. Environ Sci Technol. 2000;34:3579–3587.
14. Long CM, Suh HH, Koutrakis P. Characterization of indoor particle sources using continuous mass and size monitors. J Air Waste Manag Assoc. 2000;50:1236–1250.
15. Abt E, Suh HH, Allen G, et al. Characterization of indoor particle sources: a study conducted in the metropolitan Boston area. Environ Health Perspect. 2000;108:35–44.
16. Wallace L. Indoor particles: a review. J Air Waste Manag Assoc. 1996;46:98–126.
17. Long CM, Suh HH, Kobzik L, et al. A pilot investigation of the relative toxicity of indoor and outdoor fine particles: in vitro effects of endotoxin and other particulate properties. Environ Health Perspect. 2001;109:1019–1026.
18. Brauer M, Ebelt ST, Fisher TV, et al. Exposure of chronic obstructive pulmonary disease patients to particles: respiratory and cardiovascular health effects. J Expo Anal Environ Epidemiol. 2001;11:490–500.
19. Whitby KR. The physical characteristics of sulphur aerosols. Atmos Environ. 1978;12:135–159.
20. Sarnat JE, Long CM, Koutrakis P, et al. Using sulfur as a tracer of outdoor fine particulate matter. Environ Sci Technol. 2002;36:5305–5314.
21. Husar RB, Tratt DM, Schichtel BA, et al. Asian dust events of April 1998. J Geophys Res. 2001;106:18317–18330.
22. Brook RD, Brook JR, Rajagopalan S. Air pollution: the ‘heart’ of the problem. Curr Hypertens Rep. 2003;5:32–39.
23. Schlesinger RB, Cassee F. Atmospheric secondary inorganic particulate matter: the toxicological perspective as a basis for health effects risk assessment. Inhal Toxicol. 2003;15:197–235.
24. Brauer M, Brook JR. Ozone personal exposures and health effects for selected groups residing in the Fraser Valley. Atmos Environ. 1997;31:2113–2121.
25. NARSTO (2004). Particulate matter assessment for policy makers: a NARSTO assessment. McMurray P, Shepherd M, Vickery J, eds. Cambridge: Cambridge University Press: 2004.
26. Yakovleva E, Hopke PK, Wallace L. Receptor modeling of total particle exposure assessment methodology data. Environ Sci Technol. 1999;33:3645–3652.
27. Zeger SL, Thomas D, Dominici F, et al. Exposure measurement error in time-series studies of air pollution: concepts and consequences. Environ Health Perspect. 2000;108:419–426.
28. Hopke PK, Ramadan Z, Paatero P, et al. Receptor modeling of ambient and personal exposure samples: 1998 Baltimore Particulate Matter Epidemiology-Exposure Study. Atmos Environ. 2003;37:3289–3302.
29. Ozkaynak H, Xue J, Spengler J, et al. Personal exposure to airborne particles and metals: results from the Particle TEAM study in Riverside, California. J Expo Anal Environ Epidemiol. 1996;6:57–78.
30. Long CM, Suh HH, Catalano PJ, et al. Using time- and size-resolved particulate data to quantify indoor penetration and deposition behavior. Environ Sci Technol. 2001;35:2089–2099.
Back to Top | Article Outline


Quantitative Expression of Exposure Variables

This section describes the methodology and rationale for developing separate estimates of ambient and nonambient exposure to PM2.5, PM10–2.5, and nonsulfate PM2.5 based on the measured or calculated ambient concentrations of PM2.5, sulfate, and PM10–2.5 and measured total exposures to PM2.5 and sulfate.

Total personal exposures (T) were directly measured for PM2.5 mass (T2.5) and for SO42− (S_T2.5) using personal monitors worn by study participants. T is the sum of ambient (A) and nonambient (N) PM encountered both indoors and outdoors:

According to the equilibrium mass balance model,4,5,20

where Cai is the concentration of ambient PM that has infiltrated indoors, C is the ambient PM concentration, FINF is the infiltration factor, P is the penetration factor, a is the air exchange rate, and k is the particle removal rate. In this article, C refers to the ambient PM concentration averaged across the 5 ambient monitoring sites for each study day, j. The fraction (referred to as the attenuation factor, α) of the ambient concentration (C) resulting in ambient exposure (A), may therefore be expressed as

where y is the fraction of time when the subject is outdoors. Using the collected time-activity diary data, we determined the fraction of time spent outdoors (yij) for each subject i on each monitoring day j (commuting time was included as time spent outdoors).

Although air exchange rate is independent of particle size, P and k, and therefore FINF and α are functions of particle size. Wilson et al4 discussed the use of FINF or α for a PM component with no indoor sources as an indicator of FINF or α for other components with the same particle size distribution. Sulfate was suggested as a candidate for such a tracer because the size distribution of atmospheric sulfate is similar to that of the accumulation mode of particles, which comprise the majority of PM2.5 mass.19 Sarnat et al20 showed good agreement between FINF for sulfate and of FINF for particle volume in the accumulation size range and concluded that sulfur compounds were primarily of ambient origin and behaved in a manner that was representative of PM2.5. Based on these observations, we assume that the infiltration and attenuation factors for sulfate may be used as a tracer of the infiltration factor for PM2.5. Thus,

It follows from equations 2 and 3 that

Individual daily values of αSO4 for the ith person and the jth day (αijSO4) were calculated using our daily individual measurements of S_T2.5/S_C2.5. For 2 subjects, S_T2.5/S_C2.5 was greater than 1 and the regression of S_T2.5 versus S_C2.5 gave a positive intercept, suggesting an indoor source of sulfate. We could not identify, however, any sources of indoor sulfate based on time activity data and housing characteristics information. For these 2 subjects, the average αSO4, as given by the slope of S_T2.5 versus S_C2.5 for each individual, was used. The daily, individual values of αSO4 were then multiplied by the corresponding daily C2.5 value to obtain AijPM2.5 (equation 7), followed by estimation of NijPM2.5 using equation 1.

To estimate ambient exposure to the thoracic coarse fraction of PM (A10–2.5), air exchange rates (aij) for each subject-day were first calculated using equation 3 and yij, αijS and estimates of P = 1 and k = 0.2 for sulfate,29 because air exchange rates do not depend on particle size, but do vary over time and across residences. The values of yij and aij were then input to equation 3 along with P = 1 and k = 1.030 for PM10–2.5 to estimate αij10–2.5. Finally, Aij10–2.5 is estimated from Aij10–2.5 = αij10–2.5 ·Cj10–2.5, where C10–2.5 was obtained by calculating the difference between the measured values of C10 and C2.5. In addition, ambient PM10 exposures (A10) were estimated using A10 = A10–2.5 + A2.5. The nonsulfate components of ambient PM2.5 concentration (NS_C2.5) and ambient PM2.5 exposure (NS_A2.5) were calculated from C2.5-S_C2.5 and A2.5- S_T2.5, respectively.

The values of P and k used were determined by statistical analysis of the PTEAM data base29 and are based on the infiltration of the fine and coarse modes. These values are in reasonable agreement with those obtained from nighttime measurements of indoor/outdoor ratios as a function of mobility particle size (and assuming no resuspension during nighttime). In this case, the summertime data analysis yielded P = 0.97 and k = 0.18 for the 0.1 to 0.5-μm size range.30

Supplemental Digital Content

Back to Top | Article Outline
Copyright © 2005 Wolters Kluwer Health, Inc. All rights reserved.