The Relationship between Averaged Sulfate Exposures and

Jun 1, 2009 - Harvard School of Public Health. , §. University of California. , ∥. Health Canada. Cite this:Environ. Sci. Technol. 43, 13, 5028-503...
0 downloads 0 Views 674KB Size
Environ. Sci. Technol. 2009, 43, 5028–5034

The Relationship between Averaged Sulfate Exposures and Concentrations: Results from Exposure Assessment Panel Studies in Four U.S. Cities J E R E M Y A . S A R N A T , * ,† KATHLEEN WARD BROWN,‡ SCOTT M. BARTELL,§ STEFANIE E. SARNAT,† AMANDA J. WHEELER,| HELEN H. SUH,‡ AND PETROS KOUTRAKIS‡ Department of Environmental and Occupational Health, Rollins School of Public Health, Emory University, Atlanta, GA, Exposure, Epidemiology and Risk Program, Harvard School of Public Health, Boston, MA, Program in Public Health, University of California, Irvine, CA, and Health Canada, Ottawa, Ontario, Canada

Received February 10, 2009. Revised manuscript received May 6, 2009. Accepted May 7, 2009.

This analysis examines differences between measured ambient, indoor, and personal sulfate concentrations across cities, seasons, and individuals to elucidate how these differences may impact PM2.5 exposure measurement error. Data were analyzed from four panel studies conducted in Atlanta, Baltimore, Boston, and Steubenville (OH). Among the study locations, 1912 person-days of personal sulfate data were collected over 396 days involving 245 individual sampling sessions. Longterm differences in ambient and personal levels averaged over time are examined. Differences between averaged ambient and personal sulfate among and within cities were observed, driven by between subject and city differences in sulfate infiltration, Finf, from outdoors to indoors. Neglecting this source of variability in associations may introduce bias in studies examining long-term exposures and chronic health. Indoor sulfate was highly correlated with and similar in magnitude to personal sulfate, suggesting indoor PM monitoring may be another means of characterizing true exposure variability.

Introduction Exposure assessment panel studies have investigated relationships between ambient fine particulate matter (PM2.5) concentrations and corresponding personal PM2.5 exposures with the goal of evaluating the validity of methods used in PM epidemiologic studies (1-4). These studies generally have repeated measure designs, characterizing personal PM2.5 exposures for panels of subjects over one to four weeks. Results have shown strong daily correlations between ambient PM2.5 and personal exposures to ambient PM2.5 (2, 5). * Corresponding author phone: 404-712-9725; fax: 401-727-8744; e-mail: [email protected]. † Emory University. ‡ Harvard School of Public Health. § University of California. | Health Canada. 5028

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 13, 2009

The strength of these daily correlations is the basis for the assertion that results from time-series studies examining associations between PM2.5 levels and acute health outcomes are unbiased (6, 7). For studies associating chronic health effects with PM2.5 exposures within or between cities, however, exposure measurement error introduced by using averaged ambient PM2.5 concentrations as surrogates for average personal exposures may comprise a primary component of bias for the observed risk estimates. Bias may be present if a personalambient relationship varies differentially by city or geographic locale within a city. In these situations, a single ambient PM2.5 concentration averaged over time represents variable amounts of personal exposure to that ambient concentration. A high annual ambient concentration in City A compared to City B, for example, does not necessarily entail that average annual personal exposures are similarly higher in City A. Despite the potential impact of this source of exposure measurement error, the degree to which ambient PM2.5 contributes to and is associated with personal PM2.5, and how this association varies over averaging times is still poorly understood. This is due, in part, to the difficulty of independently apportioning the fraction of ambient source PM2.5 from total personal PM2.5 exposures (8). To address this difficulty, particulate sulfate (SO42-) has been used as a surrogate of ambient source PM2.5 and ambient PM2.5 infiltration indoors (1, 9, 10), since it is a component of PM2.5 that (a) is largely ambient in origin; (b) comprises a substantial (25-50%) fraction of PM2.5 mass in many U.S. locations (11); and (c) exhibits infiltration behavior to indoor microenvironments similar to that of other stable, ambient PM2.5 components in the accumulation mode (particles with aerodynamic diameters between 0.1 to 1.0 µm) (9, 12). This last characteristic is pertinent for understanding human exposure since most individuals spend ∼90% of their time indoors (13). The current analyses examine differences between measured ambient, indoor, and personal SO42- concentrations, across different cities, seasons, and individuals, with the goal of elucidating how these differences impact exposure measurement error for PM2.5. Data have been analyzed from four large exposure assessment panel studies conducted in Atlanta (GA), Baltimore (MD), Boston (MA), and Steubenville (OH) by researchers from the Harvard School of Public Health (HSPH). Data from these individual locations have been used previously to examine longitudinal associations between personal particulate and gaseous exposures and associations between the pollutants and/or acute health outcomes (3, 5, 14-19). This analysis is the first effort comparing data among all four studies, briefly presenting the longitudinal associations while focusing on long-term, average associations. Questions addressed in this analysis include sources of variability in exposure to ambient particles and potential approaches for characterizing this variability in health effect studies.

Materials and Methods SO42- concentrations and exposures used in the analyses were collected from four exposure assessment panel studies conducted in Atlanta, Baltimore, Boston, and Steubenville between 1998 and 2000 (Table 1). Subjects were recruited at senior and community centers, through doctors’ referrals, and/or newspaper or radio advertisements. Subject selection was not random; however, the subjects lived in residences that were widely distributed throughout the study locations 10.1021/es900419n CCC: $40.75

 2009 American Chemical Society

Published on Web 06/01/2009

TABLE 1. Studies Included in the Analysis location Atlanta

sampling periods Fall 1999 Spring 2000

Baltimore

Summer 1998 Winter 1999

Boston

Summer 1999 Winter 1999-2000

Summer 2000 Steubenville

Summer 2000 Fall 2000

Na 15 9 13 9 15 10 15 15 15 15 15 15 15 4 11 6 5 10 7 25 25

cohorts measured b

COPD MIc COPDb MIc senior adults (>65 yrs) children (9 - 13 yrs)d senior adults (>65 yrs) children (9 - 13 yrs)d COPDb senior adults (>65 yrs) children (9 -13 yrs) senior adults (>65 yrs) children (9 - 13 yrs) COPDb CHDe healthy adult partners COPDb CHDe healthy adult partners senior adults senior adults

Microenvironments measured ambient, ambient, ambient, ambient, ambient, ambient, ambient, ambient, ambient, ambient, ambient, ambient, ambient, ambient, ambient, ambient, ambient, ambient, ambient, ambient, ambient,

outdoor, indoor, personal outdoor, indoor, personal outdoor, indoor, personal outdoor, indoor, personal personal personal personal personal personal personal personal personal personal outdoor, indoor, personal outdoor, indoor, personal outdoor, indoor, personal outdoor, indoor, personal outdoor, indoor, personal outdoor, indoor, personal indoor, personal indoor, personal

a N indicates number of subjects. b COPD ) individuals with chronic obstructive pulmonary disease. c MI ) individuals with a previous myocardial infarction. d Data from the children’s cohort was not included in the current analyses. e CHD ) individuals with chronic heart disease.

(15, 20) with the exception of Steubenville (18). With the exception of two subjects who lived in single-family residences, subjects from Steubenville lived in one of three independent living facilities that were located ∼2 km apart (18). Since the studies were designed to address varying hypotheses, differences in specific design components among the studies exist, such as specific cohorts recruited, pollutants measured, sampling durations, and microenvironments (MEs). Each of the studies followed a similar longitudinal sampling design, with subjects measured for 24 h integrated periods. Subjects from all locations were nonsmokers, living in homes with nonsmokers. Additionally, quality assurance results were comparable among the locations (5, 15, 18-21). In all locations, personal PM2.5 exposures were measured using personal samplers attached to pumps placed in a carrying bag (22). Small inertial impactors were designed specifically for personal monitoring to collect PM2.5 on Teflon filters. The Teflon filters were extracted and analyzed for SO42- using ion chromatography (23). Measurements in other MEs varied by study location, with each location having at least an ambient (i.e., measured at a centrally located ambient site), outside residence (“outdoor”) or inside residence (“indoor”) SO42- measurement for each personal exposure measurement. Quantitative measures of air exchange rate were also collected only for homes in the Atlanta and Boston studies and were, therefore, not included in this analysis. Study Locations and Designs. Sampling in Atlanta was conducted during fall 1999 and spring of 2000 (20). Subjects included those with either physician-diagnosed chronic obstructive pulmonary disease (COPD) or a previous myocardial infarction (MI). Twenty-four subjects participated in the fall and 22 in the spring. Indoor, outdoor, and personal exposure measurements were conducted on 7 consecutive days for each subject during each season. Concurrent ambient pollutant concentrations were measured at a stationary ambient site located in downtown Atlanta. A total of 245 valid person-days of SO42- exposure data were collected in Atlanta. The Baltimore exposure assessment study was conducted during summer 1998 and winter 1999 (5). The summer sampling season included 15 senior adults, and the winter sampling season included 15 senior adults, 15 individuals

with physician-diagnosed COPD, and 15 children between the ages of 9 and 13 years. Each subject underwent 12 consecutive days of personal exposure measurements during each of the seasons (“sampling session”). Since SO42- was not analyzed in the children’s PM2.5 samples, data from the children’s cohort was not included in the current analyses. For the other cohorts, concurrent ambient concentrations were available from a centrally located stationary site in Baltimore. A total of 468 valid person-days of SO42- exposure data were collected in Baltimore. Sampling in Boston was conducted during the summers of 1999 and 2000 and the winter of 1999-2000 (15, 24). Summer 1999 sampling included 15 children and 15 healthy senior adults. Winter 1999-2000 sampling included 4 individuals with physician-diagnosed COPD, 11 with physician-diagnosed chronic heart disease (CHD) and 8 healthy adults. The summer 2000 sampling included 5 individuals with COPD, 10 individuals with CHD and 6 healthy adults. Five of the summer 2000 subjects were also measured in winter 1999-2000. Subjects in the healthy senior adult and children’s cohorts were measured for 12 consecutive days during each season, whereas subjects in the other cohorts underwent 7 consecutive days of measurements. Concurrent ambient SO42- measurements were available from a centrally located site. Concurrent indoor and outdoor SO42- measurements were also available for the COPD, CHD, and healthy adult cohorts. A total of 923 valid person-days of SO42exposure data were collected in Boston. The Steubenville field study was conducted during summer and fall 2000 (19). Twenty-five healthy senior adults participated in each season, with all subjects participating in indoor monitoring and a subset of 10 subjects also participating in personal monitoring. Each subject underwent two consecutive days of indoor (and personal) measurements on a weekly basis for 12 weeks each season. Concurrent ambient measurements were collected at a monitoring site located within 1.5 km of all subjects’ residences. A total of 337 valid person-days of SO42- exposure data were collected in Steubenville. Data Analysis. Associations among ambient, indoor and personal SO42- levels were examined using descriptive statistics, correlation analyses, analyses of time-averaged VOL. 43, NO. 13, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

5029

TABLE 2. Descriptive Statistics Averaged by Date for Ambient Concentrations, Personal Exposures and Indoor Concentrations of SO42- a ambient

Atlanta fall spring Baltimore summer winter Boston summer winter Bostond summer winter Steubenville summer fall

personal

indoor

Nb

mean

med.

fifth

95th

person-daysc

mean

med.

fifth

95th

mean

med.

fifth

95th

36 33

4.5 4.4

3.9 4.1

1.8 1.4

9.7 8.7

132 131

2.8 2.7

2.7 2.5

1.3 0.8

5.1 4.9

3.5 2.5

3.1 2.4

1.4 1.0

6.8 4.7

35 34

10.5 4.0

8.0 3.7

1.9 1.5

27.3 7.7

160 308

5.8 1.8

5.5 1.8

1.4 0.6

12.9 3.4

N/A N/A

N/A N/A

N/A N/A

N/A N/A

31 38

5.7 3.1

3.8 2.3

1.3 1.0

17.0 6.5

351 301

4.2 2.0

3.0 1.5

0.8 0.7

14.8 4.7

N/A N/A

N/A N/A

N/A N/A

N/A N/A

32 25

3.6 2.6

3.2 2.5

0.8 0.7

6.3 4.3

139 132

3.1 1.4

2.3 1.3

0.6 0.5

7.7 2.3

3.1 1.6

2.2 1.4

0.5 0.5

8.7 2.9

60 72

7.8 6.2

7.3 5.3

2.5 1.4

16.2 15.4

147 188

5.8 4.4

5.1 3.3

1.8 1.0

13.4 11.8

3.0 1.4

2.7 1.4

0.5 0.7

5.6 2.3

a N/A indicates data not available for those studies. b N represents total number of days compared. represent total number of observations contributing to averaged daily values. d CHD and COPD panels only.

SO42- ratios, and linear mixed model regression analyses. Pearson’s subject-specific correlations (rp) were calculated to examine daily associations among the SO42- measurements. Because daily associations may differ from similar associations averaged over longer exposure periods, personal/ ambient (P/A), personal/indoor (P/I), and indoor/outdoor (I/O) ratios averaged over an individual’s sampling session were also examined. To compute these values, SO42- levels were averaged, by subject, and defined as follows: jk X jk Y

(1)

where Xj k is the mean personal SO42- exposure or indoor SO42concentration for subject k during their sampling session; and Yj k is either a mean indoor, outdoor, or ambient SO42concentration corresponding to a person-day of sampling for subject k during the corresponding sampling session with dates matched to those of the personal or indoor SO42-. Ratios had at least five valid measurements. P/A ratios have been denoted as “R” and referred to as either “attenuation” (10) or “ambient contribution” factors (24) previously. As frequently noted, R describes the amount of an ambient pollutant that contributes to a personal exposure and, for PM2.5, is broadly a function of particle infiltration efficiency and time spent both indoors and outdoors (10, 25). Since SO42- has minimal indoor sources, I/O ratios for SO42- can be interpreted as an estimate of the fraction of ambient PM2.5 infiltrating indoors (Finf), the infiltration efficiency. P/I ratios, correspondingly, can be interpreted as the fraction of ambient PM2.5 to which a person is exposed that has already infiltrated indoors, which is, thus, not affected by building-specific differences in infiltration. Outdoor-ambient ratios, a measure of the spatial distribution for SO42-, were not examined. Recent (24) and previous findings (26) have shown SO42- to be spatially homogeneous over metropolitan areas. Ratios greater than one, indicating either noise in the measurement method or presence of exposure to unreported indoor SO42- sources, were uncommon in this data set. Two subjects from the Boston winter sampling season, however, indicated prolonged ultrasonic humidifier usage during sampling and were excluded, since these contributions obscured the analysis objective of examining the contributions of ambient particles to personal exposures and indoor concentrations. None of the homes used kerosene heaters. Exposures to high-sulfur diesel engine 5030

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 13, 2009

c

Person-days

emissions were not reported but may comprise an additional source of personal exposure to SO42-. For this analysis, we used SO42- as a broad tracer of ambient accumulation mode PM2.5. It should be emphasized that other size fractions of ambient PM2.5 are likely subject to more effective atmospheric removal processes than SO42(9). Therefore, the SO42--derived values for R and Finf typically provide an overestimate of corresponding values for these other PM2.5 size fractions (8). Similarly, since SO42- is relatively nonvolatile, these estimates are likely overestimates compared to volatile PM components, including nitrate (17, 18). Variability in the time-averaged personal-ambient and personal-indoor associations by, e.g., season and subject, was assessed using mixed models given by Yik ) βik X ik+εik with βik ) γ + φ1+δ(i)k

(2)

where Yik is an averaged personal SO42- exposure for city i and subject k; Xik is either the averaged ambient or averaged indoor SO42- concentration for the same city and subject; βik is the slope for averaged personal versus ambient concentration or averaged personal versus indoor concentration for that same city and subject; and ik is the random error term. βik is modeled here as a sum of three statistically estimated parameters: γ is the average slope (a fixed effect); φi is the random effect of season on the slope; and δ(i)k is the random effect of subject on the slope, nested within season. Relative contributions of time-averaged ambient or indoor SO42- concentration, season, and subject to the variance in time-averaged personal SO42- were assessed using coefficients of determination based on eq 2. A more detailed explanation of the variance component analysis method can be found in the Supporting Information. Data analysis was conducted using SAS (v. 9.1, Cary, NC) and R (R Core Development Team, 2007).

Results Among the four study locations, 1912 person-days of SO42personal exposures were collected over 396 days of sampling involving 245 individual sampling sessions. Ambient SO42concentrations and corresponding personal exposures varied by city and season (Table 2). Ambient SO42- (expressed as (NH4)2SO4) comprised a substantial fraction of total ambient PM2.5 mass (Atlanta, spring ) 38%, fall ) 43%; Baltimore, summer ) 52%, winter ) 29%; Boston, summer ) 42%, winter

FIGURE 1. Subject specific Pearson’s correlation coefficients (rp). Diamonds represent median correlation coefficient. N represents number of subjects.

FIGURE 2. Mean SO42- ratios by location and season. Whiskers represent 95% confidence intervals. ) 37%; Steubenville, summer ) 53%; fall ) 43%). Correspondingly, in each of the sampling locations and seasons, ambient SO42- levels were highly correlated with total ambient PM2.5 (rp: Atlanta, spring ) 0.70; fall ) 0.71; Baltimore, summer ) 0.93, winter ) 0.65; Boston, summer ) 0.96, winter ) 0.91; Steubenville, summer ) 0.92; fall ) 0.92). Subject-specific correlation analysis showed that approximately half of the subjects (98 of 206) in the four locations had personal-ambient correlations greater than 0.90, with 70% of the subjects (145 of 206) having correlations greater than 0.80 (Figure 1). Subject-specific correlations were higher for both indoor-outdoor and personal-indoor as compared to those shown for personal-ambient. The median rp between personal and indoor SO42- in the three locations where indoor sampling was conducted (i.e., Boston, Steubenville and Atlanta) was 0.97, with 80% of all subjects having rp greater than 0.90. Averaging exposures and concentrations by subject over a complete sampling session provided information about

associations occurring over longer exposure windows, as well as potential long-term differences among absolute levels of ambient, indoor, and personal SO42-. Mean, time-averaged P/A ratios were less than 1 and varied by city and season (Figure 2). Stratifying by city across seasons, mean ratios ranged from 0.53 in Baltimore to 0.76 in Steubenville. Stratifying by city and season, more pronounced differences existed with values ranging from 0.47 in Baltimore during the winter to 0.83 in Boston during the summer. For both Baltimore and Boston, differences in the mean P/A ratios differed considerably between seasons, with significantly higher ratios observed during summer compared to winter. In contrast, differences between summer and fall P/A ratios in Steubenville and spring and fall ratios in Atlanta were negligible. Similar to the P/A ratio results, I/O ratios varied substantially by city and season (Figure 2). The lowest mean I/O ratio (0.57) was found in Boston during winter and the highest mean ratio (0.86) was found in Steubenville during fall. City VOL. 43, NO. 13, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

5031

TABLE 3. Mixed Model Results Examining Time-Averaged Ambient SO42- As a Predictor of Time-Averaged Personal SO42-(A); and Time-Averaged Indoor SO42- as a Predictor of Time-Averaged Personal SO42- (B) variability in time-averaged personal SO42- attributable to slope (LCL, UCL)

ambient SO42-

season

Atlanta Baltimore Boston Steubenville

0.63 (0.59, 0.68) 0.49 (0.40, 0.58) 0.74 (0.57, 0.90) 0.72 (0.67, 0.77)

0.76 0.69 0.66 0.88