APPROACH TO ESTIMATING PARTICIPANT POLLUTANT 1 EXPOSURES IN THE MULTI-ETHNIC STUDY OF 2

ATHEROSCLEROSIS AND AIR POLLUTION (MESA AIR) 3 4 5 Martin A. Cohen1*, Sara D. Adar1, Ryan W. Allen1,2, Edward Avol3, Cynthia L. Curl1, 6 Timothy Gould4, David Hardie1, Anne Ho,1 Patrick Kinney5, Timothy V. Larson4, Paul 7 Sampson6, Lianne Sheppard7, Karen D. Stukovsky7, Susan S. Swan1,8, L-J Sally Liu1,9, 8 Joel D. Kaufman1 9 10 11 1Department of Environmental and Occupational Health Sciences, University of 12 Washington, Seattle, WA, USA 13 2Current address: Faculty of Health Sciences, Simon Fraser University, Burnaby, BC, 14 Canada 15 3Department of Preventive Medicine, University of Southern California, Los Angeles, 16 CA, USA 17 4Department of Civil and Environmental Engineering, University of Washington, Seattle, 18 WA, USA 19 5Department of Environmental Health Sciences, Columbia University, New York, NY, 20 USA 21 6Department of Statistics, University of Washington, Seattle, WA, USA 22 7Department of Biostatistics, University of Washington, Seattle, WA, USA 23 8Current address: MDS Pharma Services, Inc., Bothell, WA, USA 24 9Current address: Institute of Social and Preventive Medicine, University of Basel, Basel, 25 Switzerland 26 27

Submitted to Environmental Science & Technology 28 November, 2008 29

30 31 *Corresponding Author: Martin Cohen, mcohen@u.washington.edu, Phone: 206-616-32 1905 33 34

2/13/09 Revision 6.0 1

Abstract 35

Most published epidemiology studies of long-term air pollution health effects 36

have relied on central site monitoring data to understand the health impacts of regional-37

scale differences in exposure. Few cohort studies have had sufficient data to characterize 38

localized variations in pollution, despite the fact that large gradients can exist over small 39

spatial scales. Similarly, previous data have generally been limited to measurements of 40

particle mass or several of the criteria gases. The Multi-Ethnic Study of Atherosclerosis 41

and Air Pollution (MESA Air) is an innovative investigation undertaken to link 42

subclinical and clinical cardiovascular health effects with individual-level estimates of 43

personal exposure to ambient-origin pollution. This project improves on prior work by 44

implementing an extensive exposure assessment program to characterize long-term 45

average concentrations of ambient-generated PM2.5, specific PM2.5 chemical components, 46

and co-pollutants, with particular emphasis on capturing concentration gradients within 47

cities. 48

This paper describes exposure assessment in MESA Air, including questionnaires, 49

community sampling, home monitoring, and personal sampling. Summary statistics 50

describing the performance of the sampling methods are presented along with descriptive 51

statistics of the air pollution concentrations by city. 52

2/13/09 Revision 6.0 2

Introduction 53

Several epidemiologic studies have reported associations between long-term 54

exposure to fine particulate matter (PM2.5) and cardiovascular morbidity and mortality 55

[1]. One mechanism through which chronic exposures are hypothesized to affect cardiac 56

health is through acceleration of atherosclerosis. This hypothesis is supported by 57

evidence from animal toxicological investigations [2-4] and cross-sectional 58

epidemiologic studies of prevalent atherosclerosis. [5-8]. Since no published studies 59

have yet evaluated air pollution and the progression of atherosclerosis, the Multi-Ethnic 60

Study of Atherosclerosis and Air Pollution (henceforth “MESA Air”) was initiated in 61

2004 to address this question. The primary aims of the ten-year project are to examine 62

the impacts of chronic exposures to PM2.5 of ambient origin on the progression of 63

atherosclerosis and the incidence of cardiovascular disease. Ambient-origin PM2.5 is 64

specifically defined in this study as both outdoor PM2.5 and the fraction of outdoor PM2.5 65

that has infiltrated into indoor environments. Secondary aims of MESA Air and its 66

ancillary studies are to evaluate associations between specific gases and particulate 67

components and cardiovascular disease. 68

In order to meet these aims, MESA Air has developed an exposure assessment 69

methodology that will allow for participant-specific estimates of long-term exposure to 70

PM2.5 of ambient origin during the time period of interest. The concept that ambient 71

exposure should be considered a key metric in air pollution epidemiology is not new [9], 72

and ambient exposures are known to be highly correlated with ambient concentrations 73

[10] [11]. However, in comparison with ambient concentrations, ambient exposures have 74

shown larger health effect estimates with smaller confidence intervals [12]. 75

2/13/09 Revision 6.0 3

Previous chronic air pollution studies have predominantly relied on limited 76

regulatory monitoring data and have assigned one ambient concentration as an assumed 77

exposure to numerous participants. We designed an extensive monitoring campaign to 78

better capture within-city variations in pollution, allowing for the resolution of 79

concentrations at a relatively small spatial scale (i.e. 10’s of meters rather than 80

kilometers). Within-city differences are an important focus of our exposure assessment 81

since recent evidence suggests that local concentration gradients are linked to differential 82

cardiovascular outcomes [1, 13, 14]. 83

Another key feature of MESA Air is the inclusion of participant-specific 84

residential infiltration efficiencies and time-location patterns. These unique data allow us 85

to estimate exposures to ambient-origin air pollution rather than using ambient 86

concentrations as surrogates for ambient [11, 15]. This is a potentially important 87

refinement from past research, since individuals spend the vast majority of their time 88

indoors [16], ambient PM2.5 concentrations are attenuated as they move indoors, and 89

attenuations differ between homes and over time within homes [17-19]. 90

Since it was not feasible to measure exposures for all 6,200 MESA Air 91

participants, a modeling approach was selected to characterize each participant’s long-92

term exposure to ambient-origin air pollution. This paper describes the general exposure 93

assessment approach of MESA Air, with an emphasis on the air pollution measurements 94

collected, chemical analyses employed, and quality assurance/quality control (QA/QC) 95

procedures utilized. Brief information on the cohort and the spatiotemporal modeling 96

methods are presented, with more detailed information on the epidemiologic design 97

including the statistical methodology to be utilized for assessing health effects in 98

2/13/09 Revision 6.0 4

Kaufman et al. (in preparation). A description of the ambient concentration models is 99

presented in Szpiro et al. [20]. 100

Study Population and Geographic Areas 101

MESA Air is an ancillary study to the Multi-Ethnic Study of Atherosclerosis (MESA) 102

described by Bild and colleagues [21]. The primary purpose of MESA was to 103

characterize subclinical cardiovascular disease and its associated risk factors among 104

African American, Caucasian, Chinese, and Hispanic persons aged 45-84 and free of 105

clinical cardiovascular disease at recruitment. MESA recruited a population-based 106

sample of men and women from 6 U.S. metropolitan areas: Baltimore City and Baltimore 107

County, MD; Chicago, IL; Forsyth County (Winston-Salem), NC; Los Angeles County, 108

CA (centered in the San Gabriel Valley); Manhattan and the Bronx, NY; and St. Paul, 109

MN. All MESA participants were eligible for inclusion in MESA Air along with new 110

participants recruited from three additional areas: coastal Los Angeles County, CA; 111

Riverside County, CA; and Rockland County, NY. These additional participants were 112

recruited to enhance exposure gradients among the cohort. Since the new recruits in 113

coastal Los Angeles were within close proximity to the other Los Angeles MESA 114

participants (generally less than 15 km), these two areas were considered as one for our 115

exposure monitoring efforts. This resulted in a total of eight areas of interest for our 116

exposure characterization campaigns. Further details about subject recruitment and 117

demographics of the MESA Air cohort can be found in Kaufman et al. (in preparation). 118

119

General Approach to Exposure Assessment 120

2/13/09 Revision 6.0 5

Exposure assessment within MESA Air is a multi-step process, illustrated in 121

Figure 1. To characterize the long-term concentrations of ambient pollution, extensive 122

pollution measurements were made at the homes and in the communities of a subset of 123

participants in each of the MESA Air areas. These data, along with local geographic, 124

meteorological, and emission information, inform a hierarchical spatiotemporal model 125

that predicts long-term average concentrations outside of each participant’s home [20]. 126

Similarly, measurements of indoor and outdoor particulate sulfur are integrated with 127

housing characteristic data collected from each participant to predict home-specific 128

infiltration efficiencies (Finf) and personalized estimates of indoor concentrations of 129

ambient-generated PM2.5 [22]. The ambient concentration to which individual i within 130

area k during time period v is exposed ( AkivE ), is a function of fraction of time spent 131

outdoors (f o), the particle infiltration efficiency (Finf), and the ambient concentration 132

( AkivC ): 133

AkivE = [f o+(1-f o)Finf] A

kivC 134

The total personal exposure can be written as: 135

136 ( )[ ] N

kivAkiv

ooNkiv

Akiv

Pkiv ECFffEEE +−+=+= inf1 137

138 where EN, or exposure to non-ambient sources, is the sum of exposure to indoor-139

generated PM2.5 (EI) and exposure to personal activity PM2.5. EI is exposure to the 140

fraction of the indoor particles generated indoors, expressed as the difference between 141

total indoor concentration CI and infiltrated concentration: 142

143 EI = [1-f o][CI – (Finf x CA)] 144

145

2/13/09 Revision 6.0 6

Exposure to personal activity PM2.5 is the difference between the measured EP and the 146

sum of calculated EA and EI. [23] 147

Further detail pertaining to the modeling of the long-term average ambient 148

concentration at the subject’s home address ( AkivC ) can be found in Szpiro et al [20]. 149

Information regarding the data collection, measured parameters, analytic methods, and 150

quality control procedures are presented below. 151

Pollutants of Interest 152

The pollutant of primary interest in MESA Air is PM2.5. Due to the influence of 153

traffic-generated pollution on within-city spatial gradients and the potential importance of 154

those gradients on health [24], we also measured traffic pollution indicators including 155

light absorbing carbon (LAC), total oxides of nitrogen (NOx), and nitrogen dioxide 156

(NO2). In addition, we characterized the composition of PM2.5 by analyzing for 157

elemental composition, organic (OC) and elemental carbon (EC), and specific organic 158

compounds. We also measured other gaseous pollutants, ozone (O3) and sulfur dioxide 159

(SO2), in select situations. 160

Data Sources and Collection Methods 161

Community-Based Monitoring 162

Environmental Protection Agency (EPA) Air Quality System (AQS) Stations 163

For regulatory compliance purposes, the EPA contracts with local agencies to 164

maintain multiple monitoring stations in and around all of the MESA Air areas except 165

Rockland County, NY. These stations include at least one site per area where speciated 166

PM2.5 data are collected. Data from the EPA monitoring stations are collected on various 167

time scales, including continuous measurements which are reported hourly as well as 24-168

2/13/09 Revision 6.0 7

hr average concentrations measured every one, three, or six days. These data provide 169

temporally rich but spatially limited information regarding ambient pollutants. 170

MESA Air Fixed Sites 171

In order to expand our characterization of temporal variability to more spatial 172

locations, we operated between 1 and 5 fixed monitoring stations in each study area. 173

These “fixed sites” remained at set locations and collected two-week integrated samples 174

of PM2.5, LAC, NOx, NO2, EC, OC, elemental and organic composition of PM2.5, and 175

SO2 (Table 1) over numerous years while MESA Air participants were being observed 176

for sub-clinical and clinical disease. The locations for these sites included libraries, 177

schools, or other buildings that were in participant-dense areas underrepresented by the 178

AQS network. One monitor was also situated within 100 m of an interstate or state 179

highway in each area to identify and characterize any differences in temporal trends near 180

major roadways, where AQS PM monitors are generally not located. One fixed site per 181

area was also co-located with a local EPA speciation site (except in Rockland, NY) to 182

calibrate our measurements with the Federal Reference Method (FRM) measurements. 183

Community Saturation Monitoring 184

In each area, we collected approximately 100 simultaneous two-week samples of 185

NO2, NOx, and SO2 (Table 1) to create a spatially rich dataset that would identify 186

important geographic predictors of within-city variability. Since traffic is a major source 187

of small-scale spatial variation, the saturation samples focused on roadway concentration 188

gradients but also captured other local sources. 189

Saturation sampling occurred during three sampling periods (December through 190

February; May through August; and in October, November, March, or April) to capture 191

2/13/09 Revision 6.0 8

seasonal differences in spatial concentration patterns. All samples were collected using 192

passive Ogawa samplers attached to utility poles approximately 3 m above ground. 193

Based on our preliminary spatial models and past investigations [25-27], we chose a 194

factorial sampling design to sample near and far from major roadways and in areas of 195

high and low population density. To maximize variability while minimizing technician 196

driving time, we selected a roadway gradient design for most locations with six samples 197

collected along a trajectory perpendicular to a major roadway (i.e., interstate or state 198

highways or major arterials) as defined by the U.S. Census Bureau’s Census Feature 199

Class Code (CFCC) A1, A2, or A3. In each direction, one sampler was situated between 200

0 and 50, 50 and 100, and 100-350 meters, respectively, from the major roadway’s edge. 201

Care was taken to avoid intersections with a large fraction of “stop and go” traffic and 202

samplers were deployed both in North-South and in East-West orientations. 203

To ensure adequate capture of non-roadway conditions, gradient locations were 204

sited in areas of low, medium, and high population density (defined by tertiles of the 205

general population in our region of interest) and in areas of varied land-use. We also 206

collected 10% of all samples at random locations and added new locations based on 207

important predictors identified during the first round of sampling. 208

Home and Participant Monitoring 209

MESA Air Questionnaire 210

To extrapolate our measured data to every participant, all subjects completed the 211

MESA Air Questionnaire upon entry into the study. The questionnaire included questions 212

about residential predictors of Finf including building characteristics, heating and air 213

conditioning, and window opening as well as information on time-location patterns such 214

2/13/09 Revision 6.0 9

as secondary residences (e.g. winter homes), employment or volunteer positions, time 215

spent indoors and outdoors, and time spent in traffic. All questionnaires and diaries used 216

in this study were written in English and translated into Spanish and Chinese. 217

Questionnaires are re-administered during follow-up phone calls approximately every 218

nine months if a participant moves or has major life changes that might affect their time-219

location information (such as retirement or the death of a spouse). Questionnaires will 220

also be re-administered to all participants during the follow-up clinical exam in 2010-221

2012 when participants will be examined for the progression of subclinical 222

atherosclerosis. 223

Home and Personal Pollution Measurements 224

Between 25 and 95 homes per area were selected for home outdoor monitoring to 225

add broader spatial coverage to the existing AQS monitoring network. In addition to 226

selecting homes for good geographic coverage of our areas (implemented by forcing data 227

in each of 4 zones per area), we also aimed to capture more extreme within-city variation 228

in concentrations than the AQS. Therefore, we over-sampled homes “near” (<50 meters) 229

major roads (25% of samples) and homes “far” (> 300 meters) from roads (25% of 230

samples). 231

Outdoor sampling was typically conducted in a participant’s backyard, away from 232

all structures. When it was not possible to sample in a backyard, samplers were placed 233

approximately 1 meter out a window that was sealed with weather-stripping. All outdoor 234

monitoring at participant homes included analysis for PM2.5, LAC, NOx, NO2, and SO2 235

and a subset included EC, OC, elemental and organic composition (Table 1). Homes 236

2/13/09 Revision 6.0 10

were visited on a rolling basis with as many as 12 homes in each area monitored 237

concurrently during any given two-week period. 238

A portion of the homes (between 25 and 80 per area) selected for outdoor 239

monitoring also received indoor pollution measurements. This sampling allowed us to 240

obtain information to characterize the Finf rates of outdoor pollution into different non-241

smoking homes in each area. Indoor samplers were set up in the participant’s main 242

activity room away from potential pollutant sources or ventilation systems. These 243

samplers collected the same pollutants as were collected outdoors (Table 1). The ratio of 244

indoor elemental sulfur concentration to outdoor elemental sulfur concentration was used 245

to estimate the infiltration rate of PM2.5 [22] [11] and O3 was measured to characterize 246

the Finf of reactive gases. 247

At all homes with indoor measurements, technicians administered an Infiltration 248

Questionnaire to the participants. This questionnaire characterized the predictors of Finf 249

also assessed on the MESA Air Questionnaire, but during the nominal two-week period 250

when paired indoor-outdoor monitoring occurred. Questions were also asked about 251

indoor sources of sulfur that might bias our approach to estimate Finf. These data inform 252

city-specific models of Finf that will be used to predict Finf for all MESA Air homes. 253

A subset (between 4 and 16 per area) of participants whose homes underwent 254

indoor-outdoor monitoring was also selected for personal exposure monitoring. Personal 255

monitoring was conducted to better understand potential sources of measurement error in 256

our study, and assess the modeled estimates of ambient-generated PM2.5 and traffic-257

related gases. Participants carried active samplers to collect PM2.5 mass and elements, 258

including sulfur, and passive samplers to measure NO2, NOx, and SO2. By including 259

2/13/09 Revision 6.0 11

sulfur, the ratio of personal sulfur exposure to ambient sulfur concentration can be used 260

as an estimate of the ratio of ambient component of personal PM2.5 to the ambient PM2.5 261

concentration. This latter ratio is the “attenuation factor” [28], or the “ambient exposure 262

factor”, α [29], and is equal to f o+(1-f o)Finf,, described previously. 263

Due to the higher level of alert dictated by the US Department of Homeland 264

Security, sampling among participants who rode public transportation in New York City 265

was limited to passive measurements only. All samples were collected during the same 266

2-week periods that indoor and outdoor samples were collected at the participant’s home 267

(see Table 1). Only participants living in non-smoking households participated in 268

personal monitoring. 269

In addition to the MESA Air Questionnaire and the Infiltration Questionnaire, 270

personal monitoring participants completed a Time-Location Diary (TLD) to understand 271

the impact of activities and locations (e.g. commuting) on personal exposures to ambient-272

generated pollution and to provide a comparison (non-contemporaneous) with the time-273

activity information in the MESA Air Questionnaire. On the TLD, participants recorded 274

their hourly presence in seven microenvironments: home indoor or outdoor, work indoor 275

or outdoor, motor vehicle, or other indoor or outdoor. In addition, participants indicated 276

if they conducted or experienced any of the following: exposure to environmental 277

tobacco smoke (active or passive), cooking, dusting or vacuuming, close proximity to a 278

fire, proximity to a major road while not in a vehicle or building, heavy traffic while 279

driving, or other sources of smoke. 280

2/13/09 Revision 6.0 12

All home and participant-based sampling was designed to be conducted twice per 281

location/person with repeat samples targeted to capture two of three distinct seasons: 282

summer, winter, or a transitional season (spring or fall). 283

Measurement Methods 284

PM2.5 mass concentrations were measured gravimetrically; the Teflon filters used 285

to determine PM2.5 mass concentration were pre- and post-weighed at the UW in a 286

temperature and humidity controlled environment [30] using standard filter weighing 287

procedures [31]. The Teflon filters were also used to determine the amount of LAC, a 288

surrogate for EC, via reflectance measurements. The relationship between LAC and EC 289

is developed empirically in each area. Following post-sample gravimetric analysis and 290

reflectometry, the filters from homes selected for paired indoor-outdoor or personal 291

monitoring were analyzed for 48 elements by X-Ray Fluorescence (XRF). EC and OC 292

were also determined using pre-fired quartz fiber filters at these locations. 293

Ogawa passive samplers were used to measure NO2, NOx, SO2, and O3. Ion 294

chromatography was used to analyze the sample extracts for nitrite, nitrate, and sulfate 295

for the quantification of NO2, O3, and SO2, respectively. Ultraviolet spectroscopy was 296

used for analysis of NOx, and the mass of NO2 was subtracted from the mass of NOx to 297

estimate the net mass of NO. Ambient concentrations of each pollutant were calculated 298

using the equations provided by Ogawa & Co.[32] 299

A more in-depth description of the measurement methods can be found in the on-300

line supporting information. 301

Geographic Data 302

303

2/13/09 Revision 6.0 13

Geographic parameters such as proximity to roadway, population density, land 304

use, and elevation were calculated or assigned to each participant home and monitoring 305

location using ArcGIS 9.2 (ESRI Corporation, Redlands, CA). Meteorological dispersion 306

models (e.g., CalRoads) were also implemented in order to better predict pollutant 307

gradients around roadways. These geographic data and dispersion modeling results will 308

be used to inform our spatiotemporal concentration prediction models [20]. 309

All participant home locations were geocoded using Dynamap 2000 TeleAtlas 310

(Menlo Park, CA). Only those homes with a minimum match score of 80% were 311

geocoded in an automated fashion. All others were geocoded interactively with a 312

minimum match score of 90 to ensure high accuracy. In addition, at all monitored homes 313

our technicians collected curb-side GPS measurements using a handheld Etrex global 314

positioning system (GPS) (Garmin International Inc, Olathe, KS) at the centerline of the 315

property and at selected sampling locations to evaluate spatial accuracy and 316

reproducibility. 317

Geographic data originated from a wide variety of sources. For example, block 318

group information and population density data were obtained from the US Bureau of the 319

Census (http://www.census.gov/) while land use was obtained from the United States 320

Geological Survey (http://edc.usgs.gov/geodata/). Pollution emission data were 321

characterized using the Toxic Release Inventory database (http://www.epa.gov/tri/) and 322

traffic volumes were characterized by overlaying traffic demand models developed 323

between 2000 and 2006 by regional traffic authorities on the Dynamap 2000 TeleAtlas 324

road network. 325

Quality Control/Quality Assurance 326

2/13/09 Revision 6.0 14

Both field blanks and co-located duplicate samples were deployed at a rate of 327

10% at the fixed and home sites for all sampling media. Blanks and duplicate samples 328

were deployed during community saturation sampling at rates of 10% and 15%, 329

respectively. To reduce participant burden, no duplicate personal PM2.5 samples were 330

collected. Duplicate Ogawa samples were collected in personal sampling at a rate of 331

15%. As a result of artifact issues associated with quartz filter sampling [33], we used 332

both standard blanks (10% of samples) and dynamic blanks (20% of samples); the latter 333

are back up filters placed downstream of the sample filter to collect any off-gassed 334

organic carbon compounds. A more in-depth description of the measurement methods 335

can be found in the on-line supplement. 336

Preliminary Results 337

Data Summary and Design Goals 338

Air pollution monitoring began in July, 2005 and is planned to continue through 339

August, 2009. Sampling goals for the home, personal, and community saturation 340

sampling were met by August, 2008, with 620 unique homes and 793 unique community 341

locations monitored. Of the 620 homes with outdoor air samples, 443 were also sampled 342

for indoor air, and 88 of those homes with paired indoor-outdoor samplers had additional 343

personal monitoring. Fixed site locations, which will continue operating through August 344

of 2009, totaled 27 unique locations. Counts of the sampling locations by area are 345

presented in the on-line supporting information. Overall, we were generally successful in 346

meeting our design goals for siting of our samples with good variation in proximity to 347

roadways, population density, and land use. Details are presented in the on-line 348

supporting information. 349

2/13/09 Revision 6.0 15

Quality assurance demonstrated high data quality with all pre-defined data quality 350

objectives (DQOs) met. Precision, as assessed by the relative percent difference of 351

duplicate samples divided by the square-root of 2, was less than 10% for all pollutants. 352

Comparisons of co-located MESA Air and EPA AQS stations also showed good overall 353

correlation between methods, with R2 ranging between 0.56 and 0.93, with some scatter 354

explainable by the small number of days included in the 2-week averages from the AQS 355

sites. Further detail is presented in the on-line supporting information. 356

Preliminary Findings 357

Figure 2 presents preliminary monitoring results for PM2.5 from outdoor home 358

samples collected in each of the MESA Air areas. In order to reduce the impact of 359

temporal variations on the 2-week averages, we have normalized these concentrations by 360

dividing by the area-average concentration during the 2-week time period and 361

multiplying by the long-term average (2005-2008) concentration for that area. Area-wide 362

short- and long-term averages were calculated using AQS data. 363

These data demonstrate that there is intra- and inter-urban spatial variation in 364

concentrations. Riverside and Los Angeles communities consistently have the highest 365

PM2.5 concentrations while concentrations were consistently lowest in Rockland and St-366

Paul, resulting in a between-area range of approximately 15 µg/m3. Within-area 367

variability of PM2.5 was generally on the order of 5 and 10 µg/m3 (25-75th percentile 368

differences) with smaller variation documented in Baltimore and Winston-Salem and 369

larger variation found in Riverside and Los Angeles. Szpiro et al. have made more in-370

depth investigation into the structure of the spatio-temporal variability of the data. 371

2/13/09 Revision 6.0 16

Substantial variation in Finf was also observed between and within-areas. Figure 3 372

presents preliminary indoor-outdoor particulate sulfur ratios from a subset of homes that 373

were sampled simultaneously for indoor and outdoor air pollution. These data are 374

presented independent of season due to the limited homes with analyzed speciation data 375

at the time of publication. These ratios range on average from 55% in St Paul, MN to 376

approximately 85% in New York and Los Angeles. The indoor-outdoor ratios were also 377

quite variable with the 10th to 90th percentiles differing by as much as 60%, showing the 378

within-city variability. 379

Limitations 380

Despite its considerable strengths, some limitations of this exposure assessment 381

approach should be noted. First, one emphasis of this project was to characterize 382

residential exposures to ambient-origin pollution. Although information was collected on 383

commuting patterns and work exposures, we do not have comparable monitoring data or 384

infiltration information to accurately quantify concentrations in non-residential scenarios. 385

Additionally, MESA Air was not intended to fully characterize indoor sources of 386

exposure such as cooking, smoking or occupational emissions. The exposure assessment 387

methodology employed assumes that average non-ambient personal exposure neither 388

varies across communities nor with ambient exposure. However, this assumption is 389

consistent with previous research [28] [34], and we will also be able to explore its 390

validity by comparing questionnaire-derived data on non-ambient pollutant sources 391

across communities. 392

An additional limitation of this project is that our PM2.5 samples may not capture 393

volatile or semi-volatile components; therefore our measured concentrations may in some 394

2/13/09 Revision 6.0 17

cases be underestimates [35, 36]. This may be especially problematic in and around Los 395

Angeles due to the high concentrations of particulate nitrate in that region [36]. 396

Nevertheless, the FRM also underestimates the concentration of nitrate and other semi-397

volatile species, so our PM2.5 measurements are expected to be comparable to those 398

measured using the FRM [36]. This issue also has implications for our estimation of Finf 399

since the Finf of sulfur accurately represents Finf of non-volatile PM2.5 (assuming that 400

there are no indoor-generated sources of sulfur) but may overestimate the Finf of total 401

PM2.5 in regions with relatively high nitrate concentrations [37]. 402

Discussion 403

Links between ambient fine particulate air pollution and adverse cardiovascular 404

health effects are increasingly well established[1], though epidemiological studies to date 405

have used very simple exposure estimation methods. MESA Air was funded by EPA to 406

reduce uncertainty in the relationship between PM2.5 and cardiovascular disease in part 407

through improved exposure assessment. Unlike many past chronic epidemiology studies 408

that have been limited to assigning one exposure estimate to all participants within a city, 409

MESA Air aims to predict unique exposure estimates for ambient-generated PM2.5 for 410

every study participant using a statistical modeling-based approach. This approach 411

characterizes and incorporates spatial and temporal variations in ambient concentrations, 412

differences in residential Finf, and the general movement of study participants between 413

various indoor and outdoor microenvironments. To our knowledge, this is the first 414

epidemiological study of long-term air pollution effects to directly incorporate many of 415

these potentially important sources of exposure heterogeneity into an exposure 416

assessment. Based on past research by Ebelt and colleagues, [12] we anticipate that these 417

2/13/09 Revision 6.0 18

improvements in our exposures assessment will result in less biased and more precise 418

effect estimates. 419

In most urban areas, a major source of within-city variability in air pollution 420

concentrations is traffic, and evidence is accumulating that these within-city 421

concentration contrasts may play an important role on cardiovascular mortality and 422

morbidity [13, 14, 24, 38]. Due to a lack of routine pollution monitoring near roads it has 423

become common for chronic health studies to use residential proximity to major 424

roadways as a surrogate for traffic-generated air pollution exposures. Although this 425

approach is supported by data showing elevated concentrations within approximately 426

100-300 meters of the roadway’s edge [39], this approach is limited by uncertainties in 427

the proximity-exposure relationship (due to topography, wind direction, or other factors) 428

[40]. The extensive data collected in MESA Air should allow us to better characterize 429

small-scale spatial gradients and determine empirically the relationships between traffic-430

generated air pollutants and geographic predictors in each study area. 431

Another limitation of previous air pollution epidemiology studies is that they have 432

relied on outdoor monitoring data as a surrogate for personal exposure to ambient 433

pollution. This approach implicitly assumes that the attenuation of ambient 434

concentrations by buildings is the same for all participants. However, previous studies 435

have demonstrated differences in residential Finf between homes (both between and 436

within cities) and over time within homes [18, 19, 41]. Failure to account for this 437

exposure heterogeneity may lead to uncertainty and/or bias in health effect estimates 438

[41]. The ability of MESA Air to account for these differences in such a large cohort of 439

individuals in multiple locations is another unique feature of this investigation. 440

2/13/09 Revision 6.0 19

In summary, MESA Air aims to characterize important sources of between and 441

within-city heterogeneity in exposure to ambient-generated PM2.5 and co-pollutants. 442

Using data collected as detailed in this manuscript, in combination with a statistical 443

modeling-based approach, MESA Air will calculate exposure estimates for each study 444

participant that will incorporate personalized estimates of ambient concentrations outside 445

their home, individual-level Finf estimates for their homes, and weights for the duration of 446

time spent inside their homes. Together, these efforts should provide more accurate 447

estimates of ambient-source exposure than have previously been available on cohorts of 448

this size, with an expected result of improving our ability to assess relationships between 449

chronic exposures to air pollution and cardiovascular disease. 450

Acknowledgements 451

Although the research described in this manuscript has been funded wholly or in 452

part by the United States Environmental Protection Agency through RD831697 to the 453

University of Washington, it has not been subjected to the Agency's required peer and 454

policy review and therefore does not necessarily reflect the views of the Agency and no 455

official endorsement should be inferred. The work was also partially supported by 456

National Institutes of Health grant ES013195. The authors greatly appreciate the 457

dedication of the air pollution monitoring technicians, the field center and exposure 458

assessment center staff supporting these activities, and the local air quality agencies who 459

have given us access to their sites. We would also like to thank all of the study 460

participants who graciously participated in our air pollution monitoring. 461

Supporting Information 462

2/13/09 Revision 6.0 20

Supporting information is available on-line that presents more details of the 463

measurement methods and quality control/quality assurance procedures and outcomes. 464

2/13/09 Revision 6.0 21

References 465

1. Pope, C. A.; Dockery, D. W., Health effects of fine particulate air pollution: Lines 466 that connect. J. Air Waste Manage. Assoc. 2006, 56, (6), 709-742. 467 2. Sun, Q. H.; Wang, A. X.; Jin, X. M.; Natanzon, A.; Duquaine, D.; Brook, R. D.; 468 Aguinaldo, J. G. S.; Fayad, Z. A.; Fuster, V.; Lippmann, M.; Chen, L. C.; Rajagopalan, 469 S., Long-term air pollution exposure and acceleration of atherosclerosis and vascular 470 inflammation in an animal model. JAMA 2005, 294, (23), 3003-3010. 471 3. Chen, L. C.; Nadziejko, C., Effects of subchronic exposures to concentrated 472 ambient particles (CAPs) in mice: V. CAPs exacerbate aortic plaque development in 473 hyperlipidemic mice. Inhal. Toxicol. 2005, 17, (4-5), 217-224. 474 4. Suwa, T.; Hogg, J. C.; Quinlan, K. B.; Ohgami, A.; Vincent, R.; van Eeden, S. F., 475 Particulate air pollution induces progression of atherosclerosis. J. Am. Coll. Cardiol. 476 2002, 39, (6), 935-942. 477 5. Kunzli, N.; Jerrett, M.; Mack, W. J.; Beckerman, B.; LaBree, L.; Gilliland, F.; 478 Thomas, D.; Peters, J.; Hodis, H. N., Ambient air pollution and atherosclerosis in Los 479 Angeles. Environ Health Perspect 2005, 113, (2), 201-206. 480 6. Hoffmann, B.; Moebus, S.; Mohlenkamp, S.; Stang, A.; Lehmann, N.; Dragano, 481 N.; Schmermund, A.; Memmesheimer, M.; Mann, K.; Erbel, R.; Jockel, K. H., 482 Residential exposure to traffic is associated with coronary atherosclerosis. Circulation 483 2007, 116, (5), 489-496. 484 7. Diez Roux, A.; Auchincloss, A.; Franklin, T.; Raghunathan, T.; Barr, R.; 485 Kaufman, J.; Astor, B.; Keeler, J., Long-term exposure to ambient particulate matter and 486 prevalence of subclinical atherosclerosis in the multi-ethnic study of atherosclerosis. Am. 487 J. Epidemiol. 2008, e-publication ahead of print. 488 8. Allen, R. W.; Criqui, M. H.; Diez Roux, A. V.; Allison, M.; Shea, S.; Detrano, R.; 489 Sheppard, L.; Wong, N.; Stukovsky, K. D.; Kaufman, J. D., Fine particulate matter air 490 pollution, proximity to traffic, and aortic atherosclerosis: The multi-ethnic study of 491 atherosclerosis. Epidemiology, in press. . 492 9. Wilson, W. E.; Suh, H. H., Fine particles and coarse particles: Concentration 493 relationships relevant to epidemiologic studies. J. Air Waste Manage. Assoc. 1997, 47, 494 (12), 1238-1249. 495 10. Mage, D.; Wilson, W.; Hasselblad, V.; Grant, L., Assessment of human exposure 496 to ambient particulate matter. J. Air Waste Manage. Assoc. 1999, 49, (11), 1280-1291. 497 11. Wilson, W. E.; Mage, D. T.; Grant, L. D., Estimating separately personal 498 exposure to ambient and nonambient particulate matter for epidemiology and risk 499 assessment: Why and how. J. Air Waste Manage. Assoc. 2000, 50, (7), 1167-1183. 500 12. Ebelt, S. T.; Wilson, W. E.; Brauer, M., Exposure to ambient and nonambient 501 components of particulate matter: A comparison of health effects. Epidemiol. 2005, 16, 502 (3), 396-405. 503 13. Miller, K. A.; Siscovick, D. S.; Sheppard, L.; Shepherd, K.; Sullivan, J. H.; 504 Anderson, G. L.; Kaufman, J. D., Long-term exposure to air pollution and incidence of 505 cardiovascular events in women. New Engl. J. Med. 2007, 356, (5), 447-458. 506 14. Jerrett, M.; Burnett, R. T.; Ma, R. J.; Pope, C. A.; Krewski, D.; Newbold, K. B.; 507 Thurston, G.; Shi, Y. L.; Finkelstein, N.; Calle, E. E.; Thun, M. J., Spatial analysis of air 508 pollution and mortality in Los Angeles. Epidemiol. 2005, 16, (6), 727-736. 509

2/13/09 Revision 6.0 22

15. Wilson, W. E.; Brauer, M., Estimation of ambient and non-ambient components 510 of particulate matter exposure from a personal monitoring panel study. J. Expos. Sci. 511 Environ. Epidemiol. 2006, 16, (3), 264-274. 512 16. Klepeis, N. E.; Nelson, W. C.; Ott, W. R.; Robinson, J. P.; Tsang, A. M.; Switzer, 513 P.; Behar, J. V.; Hern, S. C.; Engelmann, W. H., The national human activity pattern 514 survey (NHAPS): A resource for assessing exposure to environmental pollutants. J. 515 Expos. Sci. Environ. Epidemiol. 2001, 11, (3), 231-252. 516 17. Allen, G. A.; Johnson, P. R. S. In Spatial and temporal assessment of a mobile 517 source aerosol indicator during winter in Boston, MA: A pilot study, International 518 Conference by the American Association for Aerosol Research, Pittsburgh, PA, 2003; 519 Pittsburgh, PA, 2003. 520 18. Wallace, L.; Williams, R., Use of personal-indoor-outdoor sulfur concentrations 521 to estimate the infiltration factor and outdoor exposure factor for individual homes and 522 persons. Environ. Sci. Technol. 2005, 39, (6), 1707-1714. 523 19. Allen, R.; Larson, T.; Sheppard, L.; Wallace, L.; Liu, L. J. S., Use of real-time 524 light scattering data to estimate the contribution of infiltrated and indoor-generated 525 particles to indoor air. Environ. Sci. Technol. 2003, 37, (16), 3484-3492. 526 20. Szpiro AA, S. P., Sheppard L, Lumley T, Adar SD, Kaufman, J. , Predicting intra-527 urban variation in air pollution concentrations with complex spatio-temporal interactions. 528 UW Biostatistics Working Paper Series 2008, Working Paper 337, (November), 529 http://www.bepress.com/uwbiostat/paper337. 530 21. Bild, D. E.; Bluemke, D. A.; Burke, G. L.; Detrano, R.; Roux, A. V. D.; Folsom, 531 A. R.; Greenland, P.; Jacobs, D. R.; Kronmal, R.; Liu, K.; Nelson, J. C.; O'Leary, D.; 532 Saad, M. F.; Shea, S.; Szklo, M.; Tracy, R. P., Multi-ethnic study of atherosclerosis: 533 Objectives and design. Am. J. Epidemiol. 2002, 156, (9), 871-881. 534 22. Sarnat, J. A.; Long, C. M.; Koutrakis, P.; Coull, B. A.; Schwartz, J.; Suh, H. H., 535 Using sulfur as a tracer of outdoor fine particulate matter. Environ Sci Technol 2002, 36, 536 (24), 5305-14. 537 23. Allen, R. W., Wallace, L., Larson, T., Sheppard, L., Liu, S.L.J., Estimated hourly 538 personal exposures to ambient and nonambient particulate matter among sensitive 539 populations in Seattle, Washington. J. Air & Waste Manage. Assoc. 2004, 54, 1197-1211. 540 24. Adar, S.; Kaufman, J., Cardiovascular disease and air pollutants: Evaluating and 541 improving epidemiological data implicating traffic exposure. Inhal. Toxicol. 2007, 19, 542 (1), 135-149. 543 25. Gilbert, N. L.; Goldberg, M. S.; Beckerman, B.; Brook, J. R.; Jerrett, M., 544 Assessing spatial variability of ambient nitrogen dioxide in Montreal, Canada, with a 545 land-use regression model. J. Air & Waste Manage. Assoc. 2005, 55, (8), 1059-1063. 546 26. Brauer, M.; Hoek, G.; van Vliet, P.; Meliefste, K.; Fischer, P.; Gehring, U.; 547 Heinrich, J.; Cyrys, J.; Bellander, T.; Lewne, M.; Brunekreef, B., Estimating long-term 548 average particulate air pollution concentrations: Application of traffic indicators and 549 geographic information systems. Epidemiol. 2003, 14, (2), 228-239. 550 27. Cyrys, J.; Hochadel, M.; Gehring, U.; Hoek, G.; Diegmann, V.; Brunekreef, B.; 551 Heinrich, J., GIS-based estimation of exposure to particulate matter and NO2 in an urban 552 area: Stochastic versus dispersion modeling. Environ Health Perspect 2005 113, (8), 987-553 992. 554

2/13/09 Revision 6.0 23

28. Ott, W.; Wallace, L.; Mage, D., Predicting particulate (PM10) personal exposure 555 distributions using a random component superposition statistical model. J. Air Waste 556 Manage. Assoc. 2000, 50, (8), 1390-1406. 557 29. Wallace, L.; Williams, R., Use of personal-indoor-outdoor sulfur concentrations 558 to estimate the infiltration factor and outdoor exposure factor for individual homes and 559 persons. Environ. Sci. Technol. 2005, 39, (6), 1707-1714. 560 30. Allen, R.; Box, M.; Liu, L. J.; Larson, T. V., A cost-effective weighing chamber 561 for particulate matter filters. J Air Waste Manag Assoc 2001, 51, (12), 1650-1653. 562 31. EPA, U. S., Quality assurance guidance document 2.12 monitoring PM2.5 in 563 ambient air using designated reference or class I equivalent methods. In U.S. 564 Environmental Protection Agency, N.E.R.L., Ed. 1998 565 32. Ogawa and Company, U., Inc.: Pompano Beach, FL 1998., NO, NO2, NOx, and 566 SO2 sampling protocol using the Ogawa sampler. 567 33. Olson, D.; Norris, G., Sampling artifacts in measurement of elemental and organic 568 carbon: Low-volume sampling in indoor and outdoor environments. Atmos. Environ. 569 2005, 39, (30), 5437-5445. 570 34. Dominici, F.; Zeger, S. L.; Samet, J. M., A measurement error model for time-571 series studies of air pollution and mortality. Biostatistics 2000, 1, (2), 157-75. 572 35. Chow, J.; Watson, J.; Lowenthal, D.; Magliano, K., Loss of PM2.5 nitrate from 573 filter samples in central California. J. Air Waste Manage. Assoc. 2005, 55, (8), 1158-574 1168. 575 36. Hering, S.; Cass, G., The magnitude of bias in the measurement of PM2.5 arising 576 from volatilization of particulate nitrate from teflon filters J. Air & Waste Manage. Assoc. 577 1999, 49, (6), 725-733. 578 37. Sarnat, S.; Coull, B.; Ruiz, P.; Koutrakis, P.; Suh, H., The influences of ambient 579 particle composition and size on particle infiltration in Los Angeles, CA, residences J. 580 Air & Waste Manage. Assoc. 2006, 56, 186–196. 581 38. Hoek, G.; Brunekreef, B.; Goldbohm, S.; Fischer, P.; van den Brandt, P., 582 Association between mortality and indicators of traffic-related air pollution in the 583 Netherlands: A cohort study. Lancet 2002, 360, (9341), 1203-1209. 584 39. Zhu, Y.; Hinds, W. C.; Kim, S.; Shen, S.; Sioutas, C., Study of ultrafine particles 585 near a major highway with heavy-duty diesel traffic. Atmos. Environ. 2002, 36, (27), 586 4323. 587 40. Jerrett, M.; Arain, A.; Kanaroglou, P.; Beckerman, B.; Potoglou, D.; 588 Sahsuvaroglu, T.; Morrison, J.; Giovis, C., A review and evaluation of intraurban air 589 pollution exposure models. J. Exposure Anal. Environ. Epidemiol. 2005, 15, (2), 185-590 204. 591 41. Meng, Q.; Turpin, B.; Polidori, A.; Lee, J.; Weisel, C.; Morandi, M.; Colome, S.; 592 Stock, T.; Winer, A.; Zhang, J., Pm2.5 of ambient origin: Estimates and exposure errors 593 relevant to PM epidemiology. Environ. Sci. Technol. 2005, 39, (14), 5105 - 5112. 594 595 596

2/13/09 Revision 6.0 24

Table 1 Sampling sites and parameters measured 597 Fixed Community

Saturation Home

Outdoor Home

Indoora Personalb

Nominal Sample Duration 2 weeks 2 weeks 2 weeks 2 weeks 2-3 days/ 2 weeksc

Number of Sites Per Area 1-5 44-152 25-80 25-80 4-16

Number of Sites Per Area Per 2 Week Session 1-5 0 or 100 3-5 2-3 1-2

Number of Repeat 2 Week Sampling Sessions at Each Sampling Sites

Sequential, ~26/yr 3 2 2 2

Questionnaired Infiltration Questionnaire

Time Location Diary

Pollutant Measurements (Method)

PM2.5 (Harvard PEM - Gravimetric)

Light-Absorbing Carbone (Harvard PEM - Reflectometry)

Elementse (X-ray fluorescence)

EC/OCf

(IMPROVE, Thermal Optical Reflectance)

Specific Organic Compoundsf (Gas chromatography mass spectroscopy)

NOx (Ogawa – UV Spectrometry)

NO2 (Ogawa – Ion Chromatography)

SO2

(Ogawa – Ion Chromatography)

O3 (Ogawa – Ion Chromatography)

aAll home indoor measurements are concurrent with home outdoor measurements. 598 bAll personal measurements are concurrent with home outdoor and indoor measurements. 599 cPM2.5 is collected with 2 or 3 day consecutive samples which are then composited over the ~2 week 600 period, NOx/NO2 samples are nominally 2 weeks. 601 dAll participants, including those not selected for home monitoring, complete the full MESA Air 602 questionnaire and the follow-up questionnaire. 603 eMeasured on PM2.5 Teflon filters. 604 fMeasured on PM2.5 Quartz fiber filters. 605 606

2/13/09 Revision 6.0 25

Outdoor Pollutant

Measurements

IndoorPollutant

MeasurementsGeographicData

Reported Housing

Characteristics

Observed Housing

CharacteristicsDeterministic Models

Spatio-temporalHierarchical

Modeling

Infiltration Modeling

PredictedOutdoor Concentrations

at Homes

Reported Time/LocationInformation

PredictedIndoor Concentrations

at Homes

WeightedAverage

Personal ExposurePredictions forEach Subject

Measurements

Questionnaires

Predictions

607

Figure 1. Conceptual Diagram of MESA Air Exposure Assignment. The wiggly 608 arrows indicate multiple sources of input. 609

610

2/13/09 Revision 6.0 26

0

5

10

15

20

25

30

35

Bal

timor

e

Chi

cago

Los

Ange

les

LA- C

oast

al

LA- R

iver

side

New

Yor

k

NY-

Roc

klan

d

St. P

aul

Win

ston

-Sal

em

City

Nor

mal

ized

PM

2.5

conc

entr

atio

ns m

easu

red

at

part

icip

ant h

omes

(µg/

m3)

611 Figure 2. Normalized outdoor PM2.5 (µg/m3) concentrations at participants’ homes. 612

2/13/09 Revision 6.0 27

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Baltimore Chicago LosAngeles

LA -Riverside

New York NY-Rockland

St. Paul

Rat

io o

f ind

oor t

o ou

tdoo

r sul

fur c

once

ntra

tion

613 Figure 3. Indoor to outdoor particulate sulfur concentration ratios by area. Winston-Salem 614 data not presented because samples had yet to be analyzed. 615 616 617