Factors influencing relationships between personal and ambient concentrations of gaseous and particulate pollutants

Science of the Total Environment 407 (2009) 3754–3765

Contents lists available at ScienceDirect

Science of the Total Environment j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / s c i t o t e n v

Factors influencing relationships between personal and ambient concentrations of gaseous and particulate pollutants Kathleen Ward Brown a,⁎, Jeremy A. Sarnat b, Helen H. Suh a, Brent A. Coull c, Petros Koutrakis a a b c

Exposure, Epidemiology and Risk Program, Department of Environmental Health, Harvard School of Public Health, Boston, MA, United States Department of Environmental and Occupational Health, Rollins School of Public Health of Emory University, Atlanta, GA, United States Department of Biostatistics, Harvard School of Public Health, Boston, MA, United States

a r t i c l e

i n f o

Article history: Received 4 August 2008 Received in revised form 27 January 2009 Accepted 10 February 2009 Available online 13 March 2009 Keywords: PM2.5 Sulfate Nitrogen dioxide Ozone Exposure

a b s t r a c t Previous exposure studies have shown considerable inter-subject variability in personal–ambient associations. This paper investigates exposure factors that may be responsible for inter-subject variability in these personal–ambient associations. The personal and ambient data used in this paper were collected as part of a personal exposure study conducted in Boston, MA, during 1999–2000. This study was one of a group of personal exposure panel studies funded by the U.S. Environmental Protection Agency's National Exposure Research Laboratory to address areas of exposure assessment warranting further study, particularly associations between personal exposures and ambient concentrations of particulate matter and gaseous co-pollutants. Twenty-four-hour integrated personal, home indoor, home outdoor and ambient sulfate, elemental carbon (EC), PM2.5, ozone (O3), nitrogen dioxide (NO2) and sulfur dioxide were measured simultaneously each day. Fifteen homes in the Boston area were measured for 7 days during winter and summer. A previous paper explored the associations between personal–indoor, personal– outdoor, personal–ambient, indoor–outdoor, indoor–ambient and outdoor–ambient PM2.5, sulfate and EC concentrations. For the current paper, factors that may affect personal exposures were investigated, while controlling for ambient concentrations. The data were analyzed using mixed effects regression models. Overall personal–ambient associations were strong for sulfate during winter (p b 0.0001) and summer (p b 0.0001) and PM2.5 during summer (p b 0.0001). The personal–ambient mixed model slope for PM2.5 during winter but was not significant at p = 0.10. Personal exposures to most pollutants, with the exception of NO2, increased with ventilation and time spent outdoors. An opposite pattern was found for NO2 likely due to gas stoves. Personal exposures to PM2.5 and to traffic-related pollutants, EC and NO2, were higher for those individuals living close to a major road. Both personal and indoor sulfate and PM2.5 concentrations were higher for homes using humidifiers. The impact of outdoor sources on personal and indoor concentrations increased with ventilation, whereas an opposite effect was observed for the impact of indoor sources. © 2009 Elsevier B.V. All rights reserved.

1. Introduction Many previous epidemiological studies have shown fine particulate matter (PM2.5), including its key chemical components, sulfate, elemental and organic carbon, to be strongly associated with adverse health effects (Dockery et al., 1993; Schwartz et al., 1996; Hoek et al., 2000; Laden et al., 2000). Other studies have also found associations between gaseous pollutants and adverse health outcomes (Sunyer et al., 1997; Bell et al., 2005). There has been some variability in the associations by pollutant as well as by geographic region (Janssen et al., 2002; Dominici et al., 2002; Levy et al., 2005). Some of these differences may be due, in part, to differences in actual personal ⁎ Corresponding author. 1432 Knightsbridge Drive, Blue Bell, PA 19422, United States. Tel.: +1 267 613 8237; fax: +1 617 384 8859. E-mail address: [email protected] (K.W. Brown). 0048-9697/$ – see front matter © 2009 Elsevier B.V. All rights reserved. doi:10.1016/j.scitotenv.2009.02.016

exposures to indoor and outdoor-generated pollutants. Indeed, personal exposures have been shown to be strongly influenced by housing characteristics or time-activity patterns and indoor or localized outdoor sources. (Spengler et al., 1985; Suh et al., 1992; Clayton et al., 1993; Liu et al., 2003; Williams et al., 2003). Exposures to pollutants that are mostly associated with outdoor sources (e.g., SO2− 4 , EC and O3) are higher for individuals living in wellventilated homes (Sarnat et al., 2000; Sarnat et al., 2006). As housing factors are likely to vary geographically and even seasonally within a region, so would their influence on personal exposures. In contrast, for pollutants emitted indoors, greater ventilation would lead to reduced exposures, since pollutants may not have the opportunity to build up over time indoors. has been typically shown to be a strong While outdoor SO2− 4 in the Eastern US and Canada, exposure predictor of personal SO2− 4 studies have shown inter-subject variability in the associations

K.W. Brown et al. / Science of the Total Environment 407 (2009) 3754–3765

between personal and ambient or indoor and outdoor SO2− levels 4 (Brauer et al., 1989; Suh et al., 1992; Wallace and Williams, 2005). Housing conditions, particularly ventilation, and time-activity patterns have been shown to be important predictors of personal (Sarnat exposures to pollutants of ambient origin, such as SO2− 4 et al., 2000; Liu et al., 2003). Many of these previous exposure studies with more recent studies focused on exposures to PM2.5 and SO2− 4 including EC (Janssen et al., 2000; Landis et al., 2001; Noullett et al., 2006). Fewer studies have measured particle species and gaseous copollutant exposures simultaneously. Previous studies have shown different degrees of associations for these pollutants in different geographic regions, likely related to differences in absolute concentrations as well as housing characteristics (Liu et al., 1995; Sarnat et al., 2000, 2005). Some previous studies have shown personal–ambient NO2 and O3 associations to vary by ventilation status and season; however, the associations are generally weaker than those seen for particulate species (Levy et al., 1998; Zota et al., 2005; Sarnat et al., 2006). In addition, the effects of local traffic or indoor sources on personal exposures need to be better understood, as they may weaken apparent personal–ambient associations, while still adding to the overall exposure burden. This paper examines the effects of different factors on personal exposures to several particulate and gaseous pollutants (SO2− 4 , EC, PM2.5, O3, NO2 and SO2). Data were collected as part of a personal exposure study of non-smokers conducted in Boston during 1999– 2000. This study was one of a group of personal exposure panel studies funded by the U.S. Environmental Protection Agency's National Exposure Research Laboratory to address areas of exposure assessment warranting further study, particularly associations between personal exposures and ambient concentrations of particulate matter and gaseous co-pollutants (Landis et al., 2001; Williams et al., 2002; Liu et al., 2003; Allen et al., 2004). The current study included personal, home indoor, home outdoor and ambient particulate and gaseous measurements for 15 homes in the Boston area during winter and summer sampling seasons. This paper focuses on the associations between personal and ambient concentrations of the particulate and gaseous species measured. The goal of the current analysis is to determine if exposure factors can explain some of the inter-subject variability in personal–ambient associations in exposure studies. Given the complexity and difficulty of conducting large-scale personal exposure studies, a better understanding of how some housing factors, activities and sources affect personal exposures can facilitate better estimates of exposures in future health assessment studies. 2. Methods 2.1. Study design Simultaneous 24-hour integrated personal, home indoor, and home outdoor SO2− 4 , EC, PM2.5, O3, SO2 and NO2 concentrations were measured for 25 individuals living in metropolitan Boston, MA, during winter (November 1999–January 2000) and summer (June–July 2000). Fifteen participants were monitored during each season, with five of the 25 individuals participating in both seasons. Monitoring for each individual was conducted over seven consecutive days. During a given sampling session, three to four homes were measured simultaneously, and each home had one to two individuals take part in the personal monitoring. In total, 30 seven-day sampling sessions were conducted during the study, comprising 210 sample-days. The 24-h integrated samples were exchanged every morning of the study between approximately 7:00 and 11:00 AM. Personal samplers were worn on the shoulder strap of a backpack. Participants were asked to bring the sampler with them at all times but were allowed to put the sampler nearby during stationary activities. Indoor samplers were placed inside the main activity room of the home on a tripod, approximately 1 m from doors, windows and vents. Outdoor samplers

3755

were placed in the yard on tripods, as possible at least 1 m from the house, trees, etc. Inlets for both the indoor and outdoor samplers were placed approximately 4 ft above the ground. Corresponding 24-hour SO2− 4 , EC and PM2.5, samples were collected each day, beginning at 9:00 AM, at a stationary ambient monitoring (“ambient”) site located at the Harvard School of Public Health in Boston, MA. The ambient site was located approximately 20 m above ground level, which was a higher elevation than most of the homes in this study. The building height was comparable to adjacent buildings, and nearby buildings did not likely inhibit air flow to the samplers. Hourly O3, NO2 and SO2 data were obtained from Massachusetts Department of Environmental Protection (MADEP)-operated SAM sites. Data were collected from one site for O3, four sites for NO2 and four sites for SO2. Hourly gaseous pollutant concentrations were averaged over 24-hour periods, beginning at 9 AM, to coincide with integrated samples. For NO2, the 24-hour average concentrations were used for the site closest to each home monitored. For SO2, which was not used in models, the average across the sites is presented. Ambient concentrations of O3, NO2 and SO2 were measured using a UV photometric analyzer, a chemiluminescence monitor and a pulsed fluorescent monitor, respectively. A map showing the home and ambient monitoring sites used in the study is provided in Fig. 1. All ambient monitors were located in generally urban areas, including the central HSPH site. Although the HSPH site is located in an urban area, it is not adjacent to highways. Local traffic affected all of the home sites with the exclusion of one home located in a generally rural area. Fig. 1 excludes one home located in Lowell, MA approximately 40 km north of the HSPH site, and one home located in Middleboro, MA, a rural area approximately 60 km south of the HSPH site. In addition to pollutant measurements, 24-h air exchange rates (AERs) were measured in each participant's home each day using perfluorocarbon tracers (PFT) and passive capillary absorption tubes (CATs) (Dietz and Cote, 1982). AERs in units of air changes per hour (ach) were calculated using the concentration of PFT on each CAT, source emission rates and home volume. As 95% of the blank values were below the laboratory detection limit, PFT samples were not blank corrected. The 90th percentile value of the PFT blank samples, or 1.0 pl, was used as the limit of detection (LOD). Approximately one-third of the CAT samples were below the LOD of 1.0 pl, which corresponded to air exchange rates ranging between 1.5 and 6.6 ach/h, depending on home volume. Duplicate samples were deployed in each home during each day of the study, with the number of duplicate pairs collected totaling 101 during winter and 49 during summer. The relative precision of duplicate AER measurements was approximately 30%. All duplicates were collected side-by-side using wire to place them out 6 in. from the wall and at least 4 in. from each other. The field staff was instructed to locate samplers avoiding high traffic areas, heat sources, air vents and kitchens. The poor precision of this method is likely due to under deployment of PFT sources in the homes, resulting in a large fraction of samples below detection. Subjects were recruited through radio and newspaper advertisements, fliers in hospitals and doctors' offices, and direct recruiting at senior centers. For eight participants during winter and six during summer, personal exposure monitoring was also conducted for a partner living in the home. Subjects participating in this study completed informed consent that was obtained prior to sampling. Consent forms and study procedures were approved by the Human Subjects Committee of the Harvard School of Public Health. 2.2. Sampling methods All personal samples were collected using a multi-pollutant sampler, comprising Harvard Personal Environmental Monitors

3756

K.W. Brown et al. / Science of the Total Environment 407 (2009) 3754–3765

Fig. 1. Map of sampling locations.

(HPEMs) (Demokritou et al., 2001). The sampler was comprised of and EC, which were individual samplers to collect PM2.5, SO2− 4 connected to a single pump. Flows in each line were controlled using valves to achieve flows within 10% of the target flow rates. Sampler flows were measured both before and after sampling. PM2.5 samples were collected using HPEMs, small inertial impactors that collected PM2.5 on 37-mm Teflon filters. All Teflon filters were refrigerated immediately after collection to minimize semi-volatile losses. Barometric pressure corrections were applied to each of the pre- and post-sampling weights (Koistinen et al., 1999). SO2− 4 concentrations were subsequently determined by extracting the PM2.5 filters and analyzing the aqueous extract by ion chromatography. EC samples were collected using a HPEM with a single pre-fired quartz filter. Collected EC samples were subsequently analyzed using thermal optical reflectance by the Desert Research Institute (Chow et al., 1993). O3 was measured using a nitrite-coated badge (Koutrakis et al., 1993). NO2 and SO2 were measured using passive badges (Yanagisawa and Nishimura, 1982). 2.3. Questionnaires Home characteristics information was collected for each day of sampling using questionnaires that asked for information on household activities and conditions that may have affected indoor particle concentrations. Participants also completed daily time-activity diaries denoting their activities and location every 15 min. All questionnaires were developed and provided by the US Environmental Protection Agency (Williams et al., 2002). Home addresses were geocoded (MassGIS, 2002) and distance to a major road and traffic on the nearest

roadway were obtained from the Massachusetts Highway Department. Roadway Inventory (MassGIS, 2002), and local land use were determined from US Census data (Census, 2001). 2.4. Quality assurance Standard methods were used to measure precision and limits of detection (LODs). Data from at least 12 duplicate pairs were used to estimate precision of the sampling method for SO2− 4 , EC and PM2.5. Absolute precision was calculated using the root mean squared pffiffiffi difference of the duplicate samples divided by 2, and it is reported in µg/m3. The LODs for all pollutants were calculated as three times the method precision. The absolute precision was then divided by the mean of the duplicate samples yielding the relative precision. LODs for the ambient site integrated measurements were not available for the time period of the field study. PM2.5 and EC field blanks were shown to be statistically different from zero, samples were therefore corrected using the median blank correction value for PM2.5: 3.8 µg (winter), 8.7 µg (summer); and for was 0.6 µg/ EC: 0.1 µg (winter), 0.4 µg (summer). The LOD for SO2− 4 m3, and its relative precision was ±6 to 8% during both seasons. The LODs for EC were approximately 1 µg/m3 with approximately 20% of the EC samples observed below the LOD during both seasons. The relative precision for EC was ±30% during winter and ±22% during summer. The LODs for PM2.5 ranged from 3 to 4 µg/m3 and precision was ±10% in both seasons, but a larger fraction of the winter than summer PM2.5 samples were less than the LOD. Duplicate O3 and NO2 passive samplers were collocated with the MADEP automated instruments for 15 days during the summer sampling

K.W. Brown et al. / Science of the Total Environment 407 (2009) 3754–3765

session. For NO2, the ratio of the mean passive NO2 to mean reference NO2 was 1.15 (n = 30 based on 15 pairs of duplicate NO2 samples). This ratio was then used to correct the collection rate for the passive sampler data. O3 collocation with the reference method yielded a ratio of 0.79, which was used to adjust the collection rate of the O3 samples. The field LODs were not available for the MADEP's continuous gaseous pollutant analyzers. O3 samples were blank-corrected using the mean blank value of 4.4 and 3.4 ppb during winter and summer, respectively. The ozone LODs were 8.3 and 6.1 ppb in winter and summer, respectively, with precisions of 22 and 11%. Due to outlier blank values, the median blank values of 4.0 and 4.7 ppb were used to correct the NO2 samples during winter and summer, respectively. The LODs for NO2 were 21.9 ppb (39% precision) in winter and 4.5 ppb in summer (6.9% precision). The winter and summer LODs for SO2 were 7.3 and 4.2 ppb, respectively, with relative precisions of 37 and 124%. Only summer SO2 samples were blank corrected using the median of 1.2 ppb, as the winter median SO2 blank value was 0 ppb. The SO2 sampler performance was poor with most of the duplicate samples (which were collected outdoors) less than the LOD and many less than 0. 2.5. Data analysis 3 SO2− 4 , EC and PM2.5 concentrations are reported in µg/m , and O3, NO2 and SO2 concentrations are reported in parts per billion (ppb). Negative pollutant concentrations and values below the method detection limit were included in the analyses to minimize bias of statistical estimates (Gilbert, 1987). Linear mixed effects models (Fitzmaurice et al., 2004) were used with subject as a random effect, personal pollutant exposures as the dependent variable and ambient concentrations as an independent variable. Models examining the importance of housing, ventilation and activity factors incorporated each factor individually, both as a fixed effect and as an interaction term with ambient pollution (Fitzmaurice et al., 2004). These models were used: (1) to characterize the impact of ambient pollutant concentrations on corresponding personal pollutant exposures; and (2) to identify exposure factors that may modify or contribute to personal exposures to the measured pollutants. All factors were modeled as dichotomous variables: AER as high (N1 ach) and low (≤1 ach), any or no open windows, presence of central air conditioning (CAC), age of home as less than or greater than the median of 37 years, apartment versus house, and less than or greater than the median time spent outdoors (3.4%). While more detailed time-activity data were collected, only results for time spent outdoors were modeled. Living less than or greater than the median distance (86 m) from a major road was also tested. A major road was defined as any road not classified as a local road by the Massachusetts Highway Department (MassGIS, 2002). It should be noted that some of the comparison groups had unbalanced numbers of observations. For example, for the SO2− 4 models, the number of observations with open windows in winter was 17, while the number with all closed windows was 51. Similar unbalance was found for closed windows during summer. One concern is that the smaller group may contain mostly observations from one or two homes that significantly skew the results. However, during winter, six homes had open windows on 1 to 6 days of monitoring in that season. Similarly, during summer, there were approximately seven homes with all windows closed on 1 to 4 days of monitoring. For the other comparison groups tested, the minimum number of observations in a comparison group was 24, compared to 54 observations in the other group, indicating this did not likely affect results for all of the models presented. In addition, despite the smaller sample sizes for some of the comparison groups, the mixed model slopes were still significant, and the two degree-of-freedom test for differences between the models for the comparison groups were significant.

3757

In addition, most activity categories were too infrequent to model. For example, while candle or incense burning, burning food while cooking, use of a kitchen exhaust fan, and presence of more than two people in the home occurred throughout the study, their occurrences were infrequent and thus could not be modeled. Crude R2 values were also calculated using simple linear regressions for all models, stratified by season and factor. R2 values are referred to as “crude,” as they were calculated using simple linear regression methods, even though there were repeated measures on subjects, but this method provides some idea of the scatter in the models presented. The inclusion of crude R2 values stratified by season and factor allowed for differentiation in the strength of the associations by the factors tested. Regression parameters (e.g., slopes and intercepts) from the linear mixed effect models were compared using two-degree of freedom contrasts testing differences in slopes and intercepts simultaneously to determine the impact of a specific housing or activity factor. Plots of the regression lines were used to depict the impact of the factors on personal exposures graphically. All models were stratified by season. Since factors affecting personal exposures were evaluated individually (i.e., one factor at a time), some observations were excluded from models, when those observations obscured the effects being evaluated in the model. This occurred for twelve personal PM2.5 and nine SO2− 4 results from two participants due to ultrasonic humidifier use, which was a significant indoor source of these pollutants. These observations significantly skewed the regression results during winter when other model types were run, obscuring the effect of other factors being tested. The effect of humidifiers in these two homes was evaluated separately. In addition, one extended cooking event at a home during summer resulted in a personal PM2.5 value of 35.7 µg/m3, when the ambient level was 7.1 µg/m3. This one point dramatically affected all of the model results (Table A1 in Appendix A provides model estimates including this point). SO2 data were not included in any of the data analyses due to their extremely low levels. 3. Results 3.1. Home and participant characteristics Table 1 presents housing characteristics for the study homes during each season. As shown in Table 1, the median building age of the participants' homes was more than 30 years. AERs were higher in summer than winter with medians of 0.8 and 1.5 ach, respectively. More than one-third of the homes had AERs that exceeded 1 ach Table 1 Housing characteristics for Boston panel study. Housing characteristics (number of homes)

Winter

Summer

Home located near a major road Gas stove Electric stove House Apartment Central air conditioning (CAC) At least 1 window or CAC unit in home Median air exchange rate (mean ± SD) Median building age in years (mean ± SD) Median distance of house to a major road in metersa (mean ± SD) Median percent urban land use for home address (%)a (mean ± SD) Median average daily traffic on nearest main roada (mean ± SD) Median population per square mile for home addressb (mean ± SD)

10 7 8 8 7 4 7 0.8 (1.3 ± 1.7) 39 (49 ± 30) 10 (74 ± 77)

5 5 10 10 5 5 12 1.5 (2.6 ± 3.9) 32 (44 ± 35) 117 (193 ± 292)

78 (56 ± 42)

45 (45 ± 36)

8900 (13,061 7700 (12,983 ± 7224) ±12,859) 5682 (7466 ± 6612) 4235 (4907 ± 5661)

a MassGIS, Office of Geographic and Environmental Information, Commonwealth of Massachusetts Executive Office of Environmental Affairs. b Census 2000. Washington, DC: US Census Bureau; 2001.

3758

K.W. Brown et al. / Science of the Total Environment 407 (2009) 3754–3765

Table 2 Descriptive statistics for personal and ambient site SO2− 4 , EC, PM2.5, O3, NO2 and SO2. Pollutant

Season

Location

N

Mean (SD)

GM (GSD)

5%

95%

LOD

% below LOD

(µg/m3) SO2− 4

Winter

Personal Ambient Personal Ambient Personal Ambient Personal Ambient Personal Ambient Personal Ambient Personal Ambient Personal Ambient Personal Ambient Personal Ambient Personal Ambient Personal Ambient

80 25 100 32 88 25 99 33 75 27 82 35 88 28 103 35 88 28 103 35 88 28 103 35

1.4 (0.6) 2.6 (1.4) 3.1 (2.1) 3.6 (2.2) 1.6 (1.7) 1.1 (0.6) 1.4 (0.6) 1.3 (0.4) 12.0 (6.0) 9.9 (5.1) 10.0 (6.2) 11.8 (5.5) 0.8 (3.4) 11.8 (4.8) 6.6 (7.0) 25.2 (9.8) 12.9 (8.6) 26.8 (7.6) 17.4 (15.6) 22.8 (6.4) 1.8 (1.9) 11.3 (5.9) − 0.2 (1.1) 3.6 (1.1)

1.3 (1.6) 2.2 (1.9) 2.5 (2.1) 2.9 (2.1) 1.2 (1.9) 1.0 (1.7) 1.3 (1.4) 1.3 (1.4) 10.4 (1.8) 8.5 (1.8) 8.5 (1.7) 10.7 (1.6) 1.6 (3.4) 10.7 (1.6) 4.1 (4.0) 23.3 (1.5) 10.4 (2.0) 25.8 (1.3) 13.9 (1.9) 22.0 (1.3) 2.3 (1.6) 9.9 (1.7) 0.5 (2.5) 3.5 (1.4)

0.5 0.7 0.8 0.5 0.4 0.4 0.7 0.7 3.2 3.1 3.5 4.3 − 2.0 4.9 − 0.5 12.3 3.1 15.7 4.7 14.0 0.0 4.0 − 1.3 2.0

2.4 4.3 7.7 6.5 3.9 2.0 2.6 1.9 22.4 19.0 20.9 27.8 6.7 18.5 19.3 44.7 28.2 39.8 36.1 32.8 4.4 23.6 1.7 5.6

0.6 NA 0.6 NA 1.2 NA 1.1 NA 4.2 NA 3.6 NA 8.3 0.6a 6.1 0.6a 21.9 0.4b 4.5 0.4b 7.2 2.0c 18.6 2.0c

0% NA 0% NA 64% NA 80% NA 3% NA 6% NA 94% 0 64% 0 39% 0 68% 0 93% 0 99% 1

Summer EC (µg/m3)

Winter Summer 3

PM2.5 (µg/m )

Winter Summer

O3 (ppb)

Winter Summer

NO2 (ppb)

Winter Summer

SO2 (ppb)

Winter Summer

(SD = standard deviation; GM = geometric mean; GSD = geometric standard deviation; LOD = limit of detection, measured as 3⁎standard deviation of field blanks.) NA = Quality assurance data not available for ambient integrated samples. a Minimum detection limit for USEPA method EQOA-0992-087. b Lower detectable limit for TECO 42c instrument (http://www.thermo.com/com/cda/product/detail/1,1055,14309,00.html). c Lower detectable limit for TECO 43c instrument TECO.http://www.thermo.com/eThermo/CMA/PDFs/Product/productPDF_12267.pdf.

during winter and more than two-thirds during summer. AERs differed by housing characteristics, with the highest rates in homes with at least one open window in the home (p b 0.0001 in both seasons) and for homes that did not use or have CAC during summer (p = 0.003). AER did not vary by home age (p = 0.91 during winter and p = 0.70 during summer). All participants lived in non-smoking households, and little environmental tobacco smoke (ETS) exposure was reported. Ten instances of ETS exposure were reported with 7 of 10 lasting 10 min or less. The other three observations lasted 20, 30 and 60 min. Participants spent on average approximately 80% of their time indoors at home, with little difference by season, as shown in Fig. 2. Mixed models were used to compare time spent outdoors in the two seasons. Time spent outdoors was significantly greater in summer compared to

winter (p = 0.03); this difference corresponded to an average of less than 30 min during winter and 90 min during summer. 3.2. Personal exposures and ambient concentrations Table 2 provides a summary of the personal exposures and ambient 2− concentrations of SO2− 4 , EC, PM2.5, O3, NO2 and SO2. Personal SO4 and gaseous pollutant exposures were lower than corresponding ambient levels in both seasons. Geometric mean (GM) personal EC exposures were 1.2 and 1.3 µg/m3 in winter and summer, similar to ambient GMs during winter and summer (1.0 and 1.3 µg/m3, respectively). Personal PM2.5 exposures (GM = 10.4 µg/m3) were higher than ambient levels in winter (GM = 8.5 µg/m3) lower (personal GM = 8.5 µg/m3) than ambient levels in summer (GM = 10.7 μg/m3).

Fig. 2. Boxplots of percent time spent in six major microenvironments by all subjects in the study. The dotted lines refer to the means.

K.W. Brown et al. / Science of the Total Environment 407 (2009) 3754–3765

3.3. Exposure factors A number of exposure factors were included in mixed regression models to examine their potential influence on personal particulate and gaseous exposures. In general, factors related to home characteristics and activity patterns were found to be important modifiers of personal exposures for several pollutants. For most of the measured pollutants, ambient concentrations were better predictors of personal exposures under well-ventilated conditions, including open windows, high AERs, no CAC and older homes, which may be less tightly constructed. Table 3 presents the mixed model results for all pollutants, stratified by season for: (1) all of the data within a season; (2) low versus high AER (defined as less than or greater than the median value of 1 ACH); and (3) open windows versus no open windows. Significant contrasts (p b 0.1) testing differences in slopes and intercepts simultaneously are indicated by the bold typeface in Table 3. Significantly higher personal exposures with greater ventilation were found primarily in the summer months, which is consistent with the fact that well ventilated conditions were more prevalent in the summer. The high degree of home ventilation during the summer is likely due to the limited availability or use of air conditioning in the homes measured. Ambient concentrations also accounted for a greater proportion of the variability in summertime personal particulate exposures, when windows were open (R2 = 0.52 for SO2− 4 and 0.15 for and 0.54 PM2.5) than when windows were closed (R2 = 0.74 for SO2− 4 for PM2.5). Table 3 shows the regression model results for the models that included a factor for open or closed windows as well as the models for low versus high AER. To illustrate the differences in exposures between the open window and no open window groups, the regression estimates from the mixed models were plotted within the range of concentrations for each pollutant (Fig. 3). Fig. 3a through e graphically presents the open/closed window regression models for SO2− 4 , EC, PM2.5, O3 and NO2, respectively. In the summer, personal exposures to SO2− 4 , EC, PM2.5 and O3 were on average higher for the same ambient concentration when windows were open as compared

3759

to closed. These differences were significant only for SO2− 4 , PM2.5 and O3 with p = 0.005, 0.005 and 0.081, respectively. For PM2.5 during winter, personal exposures were weakly associated with ambient concentrations in the homes with all windows closed, based on both low slope (0.27 ± 0.33) and R2 (0.15) values. However, the intercept for the closed window homes was quite high, indicating a greater contribution from non-ambient sources or activities in the home. Contrary to the other pollutants, NO2 showed generally higher personal exposures with closed windows, but the differences between open and closed windows were statistically significant only during summer (p = 0.01 for summer and 0.31 for winter). and O3 showed significantly higher During summer only, SO2− 4 slopes for high versus low AER and a similar trend was also seen for open versus closed windows (Table 3). For other pollutants, e.g., EC, PM2.5 and NO2, models using AER did not show the same trends as those for window status. For example, during summer the slope for low AER was 0.20, while for high AER it was 0.46, While closed versus open windows would be expected to show the same trend, the slope was 0.60 for closed windows (comparable to “low” AER) and for open windows (“high AER”) it was 0.34, an opposite trend from that seen for AER. Similarly inconsistent results for the two ventilation metrics were also found for PM2.5. For NO2, the lower ventilation, using either factor, showed significantly higher exposures compared to the higher ventilation group only during summer. During winter, closed windows and “high AER” were associated with higher personal exposures compared to open windows or low AER, an opposite trend for the two ventilation factors. Inconsistent results using the two ventilation metrics were found for EC, PM2.5 and NO2, all of which have known indoor sources. As a result, the impact of ventilation on the personal exposure differed depending on the pollutant and ventilation measure. The mixed model results discussed above provided an indication of the overall impact of an ambient concentration on personal exposure. As SO2− 4 has few indoor sources, it has been shown to be a good tracer of ambient PM2.5 (Sarnat et al., 2002). Without an indoor SO2− 4 source, the indoor–outdoor SO2− 4 ratio provides an indication of the degree of

Table 3 Results for personal–ambient mixed regression models for all data stratified by season and by ventilation factors.

Bold typeface and cell border lines highlight significant differences between models at p b 0.1 between comparison groups within a season (e.g., closed versus open windows during winter). R2 values are estimates from standard linear regression.

3760

K.W. Brown et al. / Science of the Total Environment 407 (2009) 3754–3765

Fig. 3. Plots of regression models for SO2− 4 (a), EC (b), PM2.5 (c), O3 (d) and NO2 (e), including a factor for open or closed windows.

infiltration into study homes. In addition, participants spent approximately 80% of their time in their homes, warranting a more detailed assessment of the ventilation conditions within each home. Fig. 4 concentrations for provides scatter plots of the indoor–outdoor SO2− 4 each home during the four winter (4a) and five summer (4b) sampling sessions. These plots suggest a low intra-home variability in the daily indoor–outdoor SO2− 4 ratios. Homes B4 and D2 in Fig. 4a had all points above the 1:1 line due to the use of ultrasonic humidifiers on most study days. Ultrasonic humidifiers were found to have a substantial impact on home indoor concentrations and personal as well as PM2.5, likely as the result of exposures for SO2− 4 aerosolization of sulfates and other dissolved minerals present in the water. For the two subjects with ultrasonic humidifiers, personal and PM2.5 exposures were approximately two and five times SO2− 4

greater, respectively, than corresponding ambient concentrations. Finally, Fig. 3b shows the high degree of infiltration during summer in all homes with the exception of three homes with CAC. In the homes ratio was 0.9 during without CAC, the mean indoor–outdoor SO2− 4 summer, and in the three homes with CAC, it was 0.5. In addition to housing characteristics, results showed that time spent outdoors, was an important determinant of personal exposures for most of the pollutants. Fig. 5a through e are graphical representations of the modeling results for SO2− 4 , EC, PM2.5, NO2 and O3, respectively. The detailed model results are available online in a supplementary table (Table A2). Overall, individuals spending more time outdoors had higher personal exposures to SO2− 4 , EC and O3 for the same ambient concentration as compared to other individuals. Greater time spent outdoors was associated with significantly higher

K.W. Brown et al. / Science of the Total Environment 407 (2009) 3754–3765

3761

4. Discussion

Fig. 4. Scatterplots of indoor and outdoor SO42− concentrations for each home during winter (a) and summer (b). The scale for each plot is 0 to 12 µg/m3 of SO2− 4 . Homes marked with a ⁎ indicate those with central air conditioning. In (b), summer homes also measured in winter show their winter identification code.

personal exposures to SO2− 4 , EC and O3 during summer (p = 0.0804, 0.0991 and 0.0748, respectively). Similar patterns were seen for and EC, but these differences were not statistically wintertime SO2− 4 significant. NO2 exposures did not differ significantly for the individuals spending a large versus small fraction of their time outdoors. Living close to a major road was associated with increased summertime personal exposures to EC, PM2.5 and NO2 (p = 0.0257, 0.006 and 0.0732). At the median ambient EC value during summer of 1.4 µg/m3, personal EC exposures were approximately 0.2 µg/m3 higher for individuals living near a major road (Fig. 6a). During summer personal PM2.5 exposures at the median ambient PM2.5 level of 11.4 µg/m3 were 2.3 µg/m3 higher for individuals with homes located close to a major road (Fig. 6b). For NO2, the net effect of the higher slope and lower intercept for homes close to major roads was higher 24-h personal NO2 exposures at all observed ambient NO2 concentrations for our study population (Fig. 6c), with 3.5 ppb difference at the median ambient NO2 value of 22 ppb. While wintertime EC and PM2.5 exposures were greater for homes located close to a major road, these differences were not significant. Detailed model results are available online (Table A3). Other residence-specific location factors, including traffic density, population density, and percent urban land use, were not significant modifiers of the personal– ambient associations for any of the pollutants during either season.

It is likely that variability in health risks associated with particulate or gaseous pollutants reflects differences in housing conditions, source profiles and activity patterns both within and between geographic areas. As individuals spend a large fraction of their time in their homes, personal exposures to pollutants with primarily outdoor sources tend to vary with their ability to infiltrate into individuals' homes. Given that personal exposure measurements are often not possible to make, especially for large populations, improved understanding of how infiltration and pollutant sources affect personal exposures is critical to our ability to estimate exposures in future health assessment studies. The current analysis indicates that several factors, especially ventilation of the home, time spent outdoors, proximity to a major roadway and ultrasonic humidifier use may account for some of the variability in personal particulate and gaseous exposures. For all particle species and gases, there were strong seasonal differences in how exposure factors and sources affected personal exposures. Higher exposures were associated with open windows for all pollutants, with the exception of NO2, likely due to the influence of gas stoves; however, the effect of gas stoves could not be determined due to the small number of homes (n = 5) with gas stoves measured during summer. The greater infiltration into homes during summer and somewhat greater time spent outdoors during summer likely contributed to the seasonal differences in exposures for most pollutants. Additionally, stronger R2 values with greater ventilation for many of the pollutants indicates that seasonal and regional differences in ventilation may lead to differences in health effects estimates, if stratified by season and geographic region. A number of previous studies have also shown higher personal or indoor levels associated with greater ventilation (Rojas-Bracho et al., 2000; Sarnat et al., 2000; Williams et al., 2003). While Williams et al. (2003) found little seasonal variability in PM2.5 infiltration into homes in North Carolina, the less dramatic climatic differences between seasons in that study likely explains this finding. This is further supported by the significantly lower AERs in that study (mean of 0.72 ach with little difference by season) compared to those measured here (means N1 ach during both seasons). Sarnat et al. (2000) found personal–ambient associations not to be season-dependent, but rather ventilation-specific. In the current study a rather large seasonal effect in exposures was seen even when ventilation factors were included in the exposure models. However, during winter there was not a large difference in personal exposures between low and high ventilation status, because most homes had no windows open during the winter sampling period and those that did only had one or two open for a short period of time. Allen et al. (2004) also found a significant seasonal difference between the fractions of ambient particulate matter contributing to personal exposures for a panel measured in Seattle. Mixed model slopes during heating and non-heating seasons were 0.55 and 0.80, higher than those reported for the current study (0.35 and 0.72, respectively). This indicates that homes in the Seattle study likely had greater ventilation than those in the Boston study. Results for another panel study (Landis et al., 2001), conducted in Baltimore by the USEPA during summer 1998, found lower personal–outdoor mixed model slopes for SO2− 4 and PM2.5 than those reported for the current study. In and PM2.5 slopes in Baltimore (0.40 and 0.46, fact, the summer SO2− 4 respectively) resembled the slopes reported for winter in Boston (0.35 and 0.37, respectively). The contrast in results from these panel studies further indicates the potential variability in exposures by region and season, regardless of absolute outdoor pollutant levels. The positive association between personal–ambient O3 during summer reaffirms findings for another panel study in Boston (Sarnat et al., 2005) and one in Steubenville, OH (Sarnat et al., 2006). Additionally, the overall seasonal slopes for personal–ambient O3 are comparable to those reported for the previous panel study in Boston

3762

K.W. Brown et al. / Science of the Total Environment 407 (2009) 3754–3765

Fig. 5. Plots of regression models for SO2− (a), EC (b), PM2.5 (c), O3 (d) and NO2 (e), including a factor for time spent outdoors. 4

(Sarnat et al., 2005) and somewhat higher than those reported in Steubenville in summer. While results from a similar study in Baltimore did not find a strong personal–ambient O3 relationship, this is potentially due to greater ventilation in the Boston homes, as the effect of ventilation on the associations during summer was significant in our study. Furthermore, similar to results in the current study, Zota et al. (2005) also found higher indoor NO2 concentrations with decreasing ventilation in Boston homes, likely due to the presence of gas stoves in those homes. That study also saw stronger effects during winter compared to summer, as was seen in the current study. A limitation of the current analysis is that approximately half of the winter homes and only one-third of the summer homes had gas stoves, precluding a more detailed analysis of the effect of gas stoves on the personal exposures. The generally more consistent results across pollutants for window status compared to AER measured in this study indicates that for gross

assessments of ventilation over a 24-hour period, the status of open windows may provide sufficient information without the difficulty and expense of conducting AER measurements. As the AER method cannot distinguish between make-up air from adjacent apartments, for example, open windows may be a better estimate of air exchange with outdoors for multi-unit buildings. While recall questionnaires can provide unreliable results, a previous study showed that single homes can function as multiple zones in terms of AERs (Wallace et al., 2002), which can lead to inaccuracies in estimating air exchange in these homes. This may explain some of the discrepancy in the results using the two indicators of ventilation—open windows and AER. Several indicator variables, such as CAC in the home and age of the home, also likely served as proxies for ventilation, as they provided similar model results as open window status. Additionally, while the participants and homes measured in this study were not randomly

K.W. Brown et al. / Science of the Total Environment 407 (2009) 3754–3765

3763

Fig. 6. Plots of regression models for EC (a), PM2.5 (b) and NO2 (c), including a factor for proximity to a major roadway.

selected, our results compare favorably to those reported previously for a Boston panel study (Rojas-Bracho et al., 2000). Due to the large fraction of time individuals spend in their homes, housing characteristics have tended to be the focus of exposure studies in examining factors that may influence personal exposures. However, our results also showed that time spent outdoors also had a significant effect on personal exposures. This finding was especially striking given the fact that the median amount of time spent outdoors was the equivalent of less than 1 h per day. Higher exposures during time spent outdoors were also found in the USEPA's North Carolina panel study. Using continuous instruments and time-activity data, Wallace et al. (2006) found the highest average exposure to PM2.5 occurred during time spent outdoors near the home. In the current study, it may have been that this factor affected a disproportionate

group of individuals who were perhaps more active, especially during summer. Additionally, living close to a major road was associated with higher traffic-related pollutants—EC, PM2.5 and NO2. The finding that higher EC, PM2.5 and NO2 exposures were associated with homes located within 86 m from a major road is consistent with previous studies that have shown EC and NO2 to be good indicators of trafficrelated pollution (Fischer et al., 2000; Kinney et al., 2000; Janssen et al., 2001). and PM2.5 The contribution from ultrasonic humidifiers on SO2− 4 levels in this study was pronounced, even if limited to observations from two homes. Other researchers have also seen effects on indoor SO2− 4 and PM2.5 from using tap water in humidifiers (Highsmith et al., 1992; Wallace and Williams, 2005). Our study showed the highest

3764

K.W. Brown et al. / Science of the Total Environment 407 (2009) 3754–3765

indoor SO2− 4 and PM2.5 concentrations during winter were associated with ultrasonic humidifier use. This may be an area warranting further research, since humidifier use is often recommended by physicians for individuals with compromised lung function, such as those with COPD or children with respiratory infections. The imprecision of the EC and NO2 sampling methods may have obscured potential traffic impacts. While there were significant effects for homes located close to a major road, the R2 values were relatively weak for all of the EC and NO2 models, indicating that the measurement methods employed may have reduced our ability to discern differences in exposures. Comparisons between personal–ambient and personal– outdoor model results showed that spatial variability in these pollutants was evident. Use of local monitors increased R2 values two to three fold for these pollutants. It should also be noted that each factor was evaluated individually. As a result, differences in model results for a particular factor tested may actually be due to other factors not accounted for in the model. For example, if homes near busy roads were more likely to have gas stoves, higher NO2 exposures associated with living near a busy road may actually have been due to the gas stove. Additionally, for some of the model results, the magnitude of the differences between comparison groups may appear quite small; however, these differences are averaged over a 24-hour period, so sources could have much greater short-term effects that are not captured in a 24-hour integrated sample. The greater effectiveness of continuous data models compared to integrated data in estimating source effects was shown in Chang et al. (2003). Nevertheless, the collection of so many pollutants simultaneously adds to our understanding of some of the factors affecting the variability in personal– ambient associations, given the limits in instrumentation with personal monitoring. Given the strong associations between personal and ambient SO2− 4 , measurements combined with our results showed that ambient SO2− 4 questionnaires about ventilation in the home may be sufficient indicators of personal exposures to regional pollutants, for homes without ultrasonic humidifier use. However, for spatially varying pollutants, such as EC and NO2, and pollutants with indoor sources, it may not be sufficient to estimate exposures using ambient concentrations and ventilation. Additional housing information, such as proximity to a major road, or personal activities may be necessary to determine local or indoor source effects on personal exposures to these pollutants. Acknowledgements The authors wish to thank all of the participants and field staff for this study as well as Dr Lance Wallace. Thank you also to Steve Melly for preparing the site map and Jose Vallarino for database development. The U.S. EPA funded this study (grant 827159) with additional funding provided by the Harvard-EPA Center on Particle Health Effects (grant R827353 and R832416). Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at doi:10.1016/j.scitotenv.2009.02.016. References Allen R, Wallace L, Larson T, Sheppard L, Liu L. Estimated hourly personal exposures to ambient and nonambient particulate matter among sensitive populations in Seattle, Washington. J Air Waste Manage Assoc 2004;54:1197–211. Bell M, Dominici F, Samet J. A meta-analysis of time-series studies of ozone and mortality with comparison to the National Morbidity, Mortality, and Air Pollution Study. Epidemiology 2005;16:436–45. Brauer M, Koutrakis P, Spengler J. Personal exposures to acidic aerosols and gases. Environ Sci Technol 1989;23:1408–12.

Census. Census 2000. Washington, DC: US Census Bureau; 2001. Chang LT, Koutrakis P, Catalano PJ, Suh HH. Assessing the importance of different exposure metrics and time-activity data to predict 24-h personal PM2.5 exposures. J Toxicol Environ Health A 2003;66:1825–46. Chow J, Watson J, Pritchett L, Pierson W, Frazier C, Purcell R. The DRI thermal optical reflectance carbon analysis system – description, evaluation and applications in United States air quality studies. Atmos Environ Part A 1993;27:1185–201. Clayton C, Perritt R, Pellizzari E, Thomas K, Whitmore R, Wallace L, et al. Particle Total Exposure Assessment Methodology (PTEAM) study: distributions of aerosol and elemental concentrations in personal, indoor, and outdoor air samples in a southern California community. J Expos Anal Environ Epidemiol 1993;3:227–50. Demokritou P, Kavouras I, Ferguson S, Koutrakis P. Development and laboratory performance evaluation of a personal multipollutant sampler for simultaneous measurements of particulate and gaseous pollutants. Aerosol Sci Technol 2001;35: 741–52. Dietz R, Cote E. Air infiltration measurements in a home using a convenient perfluorocarbon tracer technique. Environ Int 1982;8:419–33. Dockery D, Pope C, Xu X, Spengler J, Ware JH, Fay M, et al. An association between air pollution and mortality in six U.S. cities. N Engl J Med 1993;329:1753–9. Dominici F, Daniels M, Zeger S, Samet J. Air pollution and mortality: estimating regional and national dose–response relationships. J Am Stat Assoc 2002;97:100–11. Fischer P, Hoek G, van Reeuwijk H, Briggs D, Lebret E, van Wijnen J, et al. Traffic-related differences in outdoor and indoor concentrations of particles and volatile organic compounds in Amsterdam. Atmos Environ 2000;34:3713–22. Fitzmaurice G, Laird N, Ware J. Applied longitudinal analysis. New York: John Wiley and Sons; 2004. Gilbert R. Statistical methods for environmental pollution monitoring. New York: John Wiley & Sons, Inc.; 1987. Highsmith V, Hardy R, Costa D, Germani M. Physical and chemical characterization of indoor aerosols resulting from the use of tap water in portable home humidifiers. Environ Sci Technol 1992;26:673–80. Hoek G, Brunekreef B, Verhoeff A, van Wijnen J, Fischer P. Daily mortality and air pollution in the Netherlands. J Air Waste Manage Assoc 2000;50:1380–9. Janssen N, de Hartog J, Hoek G, Brunekreef B, Lanki T, Timonen K, et al. Personal exposure to fine particulate matter in elderly subjects: relation between personal, indoor, and outdoor concentrations. J Air Waste Manage Assoc 2000;50:1133–43. Janssen N, van Vliet P, Aarts F, Harssema H, Brunekreef B. Assessment of exposure to traffic related air pollution of children attending schools near motorways. Atmos Environ 2001;35:3875–84. Janssen N, Schwartz J, Zanobetti A, Suh H. Air conditioning and source-specific particles as modifiers of the effect of PM10 on hospital admissions for heart and lung disease. Environ Health Perspect 2002;110:43–9. Kinney P, Aggarwal M, Northridge M, Janssen N, Shepard P. Airborne concentrations of PM2.5 and diesel exhaust particles on Harlem sidewalks: a community-based pilot study. Environ Health Perspect 2000;108:213–8. Koistinen K, Kousa A, Tenhola V, Hänninen O, Jantunen M, Oglesby L, et al. Fine particle (PM2.5) measurement methodology, quality assurance procedures, and pilot results of the EXPOLIS study. J Air Waste Manage Assoc 1999;49:1212–20. Koutrakis P, Wolfson JM, Bunyaviroch A, Froehlich SE, Hirano K, Mulik JD. Measurement of ambient ozone using a nitrite-coated filter. Anal Chem 1993;65:209–14. Laden F, Neas L, Dockery D, Schwartz J. Association of fine particulate matter from different sources with daily mortality in six U.S. cities. Environ Health Perspect 2000;108:941–7. Landis M, Norris G, Williams R, Weinstein J. Personal exposures to PM2.5 mass and trace elements in Baltimore, MD, USA. Atmos Environ 2001;35:6511–24. Levy J, Lee K, Spengler J, Yanagisawa Y. Impact of residential nitrogen dioxide exposure on personal exposure: an international study. J Air Waste Manage Assoc 1998;48: 553–60. Levy J, Chemerynski S, Sarnat J. Ozone exposure and mortality – an empiric Bayes metaregression analysis. Epidemiology 2005;16:458–68. Liu L, Koutrakis P, Leech J, Broder I. Assessment of ozone exposures in the greater metropolitan Toronto area. J Air Waste Manage Assoc 1995;45:223–34. Liu L, Box M, Kalman D, Kaufman J, Koenig J, Larson T, et al. Exposure assessment of particulate matter for susceptible populations in Seattle. Environ Health Perspect 2003;111:909–18. MassGIS. Massachusetts Highway Department roads. Boston, MA: Office of Geographic and Environmental Information, Commonwealth of Massachusetts Executive Office of Environmental Affairs; 2002. Noullett M, Jackson P, Brauer M. Winter measurements of children's personal exposure and ambient fine particle mass, sulphate and light absorbing components in a northern community. Atmos Environ 2006;40:1971–90. Rojas-Bracho L, Suh H, Koutrakis P. Relationships among personal, indoor, and outdoor fine and coarse particle concentrations for individuals with COPD. J Expos Anal Environ Epidemiol 2000;10:294–306. Sarnat J, Koutrakis P, Suh H. Assessing the relationship between personal particulate and gaseous exposures of senior citizens living in Baltimore, MD. J Air Waste Manage Assoc 2000;50:1184–98. Sarnat J, Long C, Koutrakis P, Coull B, Schwartz J, Suh H. Using sulfur as a tracer of outdoor fine particulate matter. Environ Sci Technol 2002;36:5305–14. Sarnat J, Brown K, Schwartz J, Coull B, Koutrakis P. Ambient gas concentrations and personal particulate matter exposures: implications for studying the health effects of particles. Epidemiology 2005;16:385–95. Sarnat S, Coull B, Schwartz J, Gold D, Suh H. Factors affecting the association between ambient concentrations and personal exposures to particles and gases. Environ Health Perspect 2006;114:649–54. Schwartz J, Dockery D, Neas L. Is daily mortality associated specifically with fine particles? J Air Waste Manage Assoc 1996;46:927–39.

K.W. Brown et al. / Science of the Total Environment 407 (2009) 3754–3765 Spengler J, Treitman R, Tosteson T, Mage D, Soczek M. Personal exposures to respirable particulates and implications for air pollution epidemiology. Environ Sci Technol 1985;19:700–7. Suh H, Spengler J, Koutrakis P. Personal exposure to acid aerosols and ammonia. Environ Sci Technol 1992;26:2507–17. Sunyer J, Spix C, Quénel P, Ponce-de-León A, Pönkä A, Barumandzadeh T, et al. Urban air pollution and emergency admissions for asthma in four European cities: The APHEA project. Thorax 1997;52:760–5. Wallace L, Williams R. Use of personal–indoor–outdoor sulfur concentrations to estimate the infiltration factor and outdoor exposure factor for individual homes and persons. Environ Sci Technol 2005;39:1707–14. Wallace L, Emmerich S, Howard-Reed C. Continuous measurements of air change rates in an occupied house for 1 year: the effect of temperature, wind, fans, and windows. J Expos Anal Environ Epidemiol 2002;12:296–306. Wallace L, Williams R, Rea A, Croghan C. Continuous weeklong measurements of personal exposures and indoor concentrations of fine particles for 37 health-

3765

impaired North Carolina residents for up to four seasons. Atmos Environ 2006;40: 399–414. Williams R, Wallace L, Suggs J, Evans G, Creason J, Highsmith R, et al. Preliminary particulate matter mass concentrations associated with longitudinal panel studies: assessing human exposures of high risk subpopulations to particulate matter, EPA/ 600/R-01/086. Research Triangle Park, NC: U.S.E.P.A; 2002. Williams R, Suggs J, Rea A, Sheldon L, Rodes C, Thornburg J. The Research Triangle Park Particulate Matter Panel Study: modeling ambient source contribution to personal and residential PM mass concentrations. Atmos Environ 2003;37:5365–78. Yanagisawa Y, Nishimura H. A badge-type personal sampler for measurements of personal exposure to NO2 and NO in ambient air. Environ Int 1982;8:235–42. Zota A, Adamkiewicz G, Levy J, Spengler J. Ventilation in public housing: implications for indoor nitrogen dioxide concentrations. Indoor Air 2005;15:393–401.