INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Epidemiologic evidence suggests that exposures to short-term ambient levels of respirable particle matter (PM) are associated with adverse health effects (1, 2). These include increase in respiratory symptoms (3, 4), an increase in hospital admission (5, 6), and excess daily mortality (7, 8). Particles less than 10 µm in diameter (PM₁₀), less than 2.5 microns in diameter (PM_2.5), and sulfate (SO₄²) or strong acid (H⁺) components of ambient aerosol are implicated, and the United States Environmental Protection Agency (U.S. EPA) has established a new particulate ambient air quality standard for PM_2.5 (9). Most existing studies of particulates and their association with human health outcomes used total suspended particulate (TSP) or PM₁₀ as the measurement for PM exposure, and little data exists in which PM_2.5 is used as the exposure measure.

Air contaminants may cause effects on airways, lung parenchyma, bronchial responsiveness, and cause acute transient bronchoconstriction. Recordings of peak expiratory flow (PEF)which may be reduced by airway narrowing or factors that limit maximal inspirationcan be used to monitor short-term effects of these contaminants (10). Decreased respiratory function, measured by PEF, is associated with ambient particulate air pollution (11), nonasthmatic symptomatic (3, 14- 16) and asthmatic (4, 17, 18) child populations as well as general (11), nonasthmatic symptomatic (19), and asthmatic (4, 18) adult populations. Prior studies have focused on compromised populations (3, 4, 16) and have not definitively distinguished particle components responsible for the observed particulate/effects.

We conducted a prospective cohort study of healthy nonsmoking women living in nonsmoking households and between the ages of 19 and 43 yr, with a newborn child, to assess the impact of exposure to ambient PM_2.5 on their upper and lower respiratory symptoms and on their respiratory function as measured by PEF. This study enabled us to investigate an association between ambient PM_2.5 and its components (SO₄², H⁺, and nitrate [NO₃]) and PEF in a nonsmoking, nonenvironmental tobacco smoke exposed, healthy population under relatively low air pollution conditions.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Study Population

The primary study investigates daily respiratory symptoms in infants and their mothers over a 1-yr period and variations in mothers' PEF during a winter and summer period. The subject population of the primary study is 918 women who recently delivered babies in Connecticut and Virginia. The subjects for this report consist of all 473 women from the primary study who were followed in the summer of 1995 or 1996 and recruited from Virginia.

Potential study subjects were selected from women delivering babies at five hospitals in southwest Virginia between November 1994 and November 1996. Mothers were eligible if they: (1) delivered a singleton child; (2) were  18 yr of age; (3) did not plan on moving out of the area; (4) spoke English; (5) did not smoke and did not have any smokers living in the home. Mothers of infants with birth weights less than 1000 grams were excluded.

Subjects enrolled in the study within 3 mo of giving birth. During a home interview, the interviewer obtained consent and verified there were no smokers in the household. The interviewer instructed the mother in use of the PEF meter, provided her with PEF recording forms, and administered a standardized questionnaire to obtain detailed information on the following: vented (wood-burning stove, fireplace, oil furnace, gas furnace) and unvented (gas stove, unvented kerosene heater) combustion sources and other potential domestic air contaminant exposures; demographic data; health history of mother; drug exposure (prescription and nonprescription); home building characteristics; occupation.

Site Description and Ambient Air Pollutant Monitoring

All exposure assessment data were collected at a central stationary ambient outdoor monitoring site (Virginia Stationary Air Monitoring Site) in Vinton, Virginia, located in the eastern area of greater Roanoke, Virginia. A regional aerosol is characteristic of the site, which is not impacted by any large stationary particulate or sulfate sources. All meteorological data were collected from Roanoke Airport, 6 miles northwest of VSAM. All five hospitals in central to southwest Virginia from which subjects were recruited were located within 115 miles of VSAM.

We measured 24-h integrated particulate mass (PM₁₀ and PM_2.5), inorganic fine particulate ions (SO₄², NO₃, H⁺, and NH₄⁺), and atmospheric pollutant gases (SO₂, HNO₃, HNO₂, and NH₃). Hourly O₃, NO, NO₂, and SO₂ were also collected at the same site by the Virginia Department of Environmental Quality (DEQ). Pollutants were measured from May 30 to September 15, 1995 and from May 14 to September 14, 1996 at VSAM. Air pollution sampling was performed daily for 24-h periods (6:00 P.M. to 6:00 P.M. in 1995; 8:00 A.M. to 8:00 A.M. in 1996). Samplers were placed 1.5 meters off the ground in a fenced lot adjacent to the DEQ monitoring site.

PM_2.5 and PM₁₀ samples were collected at 10 and 4 L/min, respectively, using Harvard-Marple impactors (Air Diagnostics, Harrison, ME). Coarse particles' (2.5 < aerodynamic diameter [d_a] < 10 µm) mass concentrations were calculated as the difference between measured PM₁₀ and PM_2.5 concentrations. Fine particle SO₄², H⁺, NO₃, and ammonium (NH₄⁺), and gaseous species nitrous acid (HNO₂), nitric acid (HNO₃), ammonia (NH₃), and sulfur dioxide (SO₂) were measured using Harvard Glass Honeycomb Denuder/Filter Pack Samplers. A full description of methods used in the exposure assessment protocol, including a full description of quality assurance and quality control, is provided elsewhere by Leaderer and coworkers (20). Daily meteorological data including temperature, relative humidity, visibility, barometric pressure, dew point, visibility, fog, and haze were obtained from the Virginia State Climatology Office and were summarized over daily 24-h periods (12:00 A.M. to 12:00 A.M.).

PEF Measurements

The lung function measurement used for the current study is PEF, which is a validated outcome for environmental epidemiology studies (10, 21). At the home interview, each mother was provided with a MiniWright PEF monitor (Armstrong Medical Industries, Lincolnshire, IL), written instructions on its use, and forms for recording readings. The interviewer trained the respondent in its use, utilizing a standardized protocol requiring the respondent to demonstrate her technique and ability to record values. The respondent recorded morning and evening PEF readings (three blows for each measurement) for the 2-wk period immediately after completion of the initial home interview. The initial 2-wk PEF measurements were collected solely for the purpose of training participants in PEF measurement technique; these data were not used in the present analyses. At two additional time periods, during the following year, one winter and one summer, the respondent was asked to record morning and evening PEF readings over a 2-wk period. This study presents and discusses results of daily peak flows over the follow-up 2-wk summer period for each woman.

Peak flow data were normalized to adjust for differences in subject height and age and to allow comparison of each subject to her own baseline. This enables one to investigate daily variations from each subject's baseline (relative PEF). Separate analyses were done on morning and evening PEF as well as on subject-specific daily difference between evening and morning PEF.

Peak flow data were normalized in three ways with each normalization procedure done separately for morning and evening. First, the 2-wk average of each subject's PEF was subtracted from each daily PEF reading to compute a daily PEF, according to the method used by Pope and Dockery (3). This is referred to as the delta normalization method. Second, the 2-wk average of each subject's PEF was subtracted from each daily PEF reading to derive a daily PEF. This value was divided by each subject's 2-wk average PEF and normalized for size by multiplying it by a factor of 460 L/min, which represents the average PEF measured in this study population. This method has been used by Neas and coworkers (15) and is referred to as the body size normalization method. Third, each subject's PEF reading was transformed into a Z-score by subtracting that subject's mean PEF and dividing the difference by the standard deviation of all PEF values for that subject. This method has been used by Lebowitz and coworkers (11) and is referred to as the Z-score normalization method. The Z-score parameters are unitless; those of the delta and body size normalizations are in L/min.

Statistical Methods

Air pollution and meteorological data preparation. Hourly continuous air pollution and meteorological data were formatted in the following ways: 24-h average, 24-h maximum, 24-h minimum, 8-h average calculated from 10:00 A.M. to 6:00 P.M., 12-h average calculated from 10:00 A.M. to 10:00 P.M., running 8-h average, and running 8-h maximum. For both air pollution and meteorological data, lag effects and cumulative lag effects were explored. For each parameter, data were entered into the model as the same day (lag 0), 1, 3, and 5 d prior (lag 1, 3, and 5, respectively) and previous 3-, 5-, and 7-d average (avg 3, 5, and 7, respectively). Contaminants considered as exposure variables were: 24-h average PM_2.5, PM₁₀, H⁺, SO₄², HNO₂, HNO₃, NO, NO₂, O₃, and SO₂; 24-h peak NO, NO₂, O₃ (both hourly and 8-h moving average), and SO₂; 12-h (10:00 A.M.-10:00 P.M.) O₃; and 8-h (10:00 A.M.-6:00 P.M.) O₃. Meteorological measures, considered as independent variables in the models, were 24-h average and 24-h maximal temperature, relative humidity, visibility, barometric pressure, dew point, visibility, fog, and haze.

PEF flow data preparation. There were 14 study days for each subject. Because subjects were often provided with their PEF meter after wake time on the first day of the monitoring period, morning PEF data for Day 1 are available from 128 of 473 (27%) subjects. Because of this, data from Day 1 were excluded from the modeling analyses. Subject-specific PEF summary statistics were reviewed to check for unusual measurements or trends (i.e., atypically high maximum, low minimum, or high variation). Further, subjects with three or fewer days (7 of 473) of combined PEF and air pollution data available were eliminated from analysis.

Because the home interview was followed by a 2-wk PEF collection training period, subjects were familiar with the MiniWright PEF meter prior to their summer time 2-wk measurement period. Thus, any learning effect occurring in the measurement of peak flows would not affect PEF data used in the present analysis. Separate regression models were run using morning and evening PEF as the dependent variable and study day (Days 1 to 14) as an independent continuous variable. Study day was not significantly associated with morning or evening PEF. We concluded that no learning effect was present so data from the second day of the PEF collection were analyzed. Further, normalized PEF was plotted as a function of day of study and LOESS (S-PLUS; StatSci Division, MathSoft, Inc., Seattle, WA) was used as a nonparametric smoothing function in order to investigate trends. A review of the pattern for study Days 2 to 14 did not reveal any evidence of a learning curve.

PEF and air pollution descriptive analyses and modeling. The distribution of the outcome variable (PEF), the primary exposure of interest (PM_2.5), other exposures (e.g., O₃, SO₄², H⁺), and other potential confounding variables (e.g., meteorological variables, presence of air conditioning, presence of dehumidifiers, socioeconomic status [SES], age) was examined for normality and outliers.

The data were further explored by plotting normalized daily average PEF for morning and evening against 24-h PM_2.5. Residuals of the regression of normalized morning PEF versus 24-h PM_2.5 were tabulated and plotted against study day (2 to 14) to investigate autocorrelations between study days for each subject.

A final model for describing the relationship between PEF and air pollutants assumed a linear relationship for each individual. The summary provided an overall description of the population from which these subjects were drawn. Normalization of the air pollutants modeled (e.g., PM_2.5, H⁺, O₃) and PEF results in values that have a defined mean of zero within subject. Hence, the model fitted to these data reflected this constraint by excluding an intercept term, which may overestimate the coefficients and their p values, if the intercept is not zero. However, we compared our main results to those from models run with an intercept and found little difference in the outcome. An unknown regression parameter for each individual was assumed to arise from a normal distribution, and the mean and standard deviation of this distribution was estimated using PROC MIXED in the SAS software package (22). From this model we obtain an estimate of the mean of these slopes for each individual, and the standard error of this parameter. The standard deviation of these slopes (_b) summarizes variability of the effect of the air pollutant on PEF among individuals. The error standard deviation (_e) reflects variability of each individual's observations about their trend line. To summarize the extent to which an air pollutant explained overall variance in outcome, we present the percent reduction in variance of PEF across individuals when that air pollutant is included in the analysis.

The above analyses were explored further by dividing the air contaminant data into quartiles and then comparing the PEF among the air pollution quartiles using PROC MIXED, which allowed tests for trend. Normalized PEF data were plotted against air pollution quartiles for a descriptive analysis of trends.

Modeling analyses of normalized PEF versus air pollution were repeated after stratifying by subject demographics, home characteristics, and subject health history, in order to test for PEF variability differences among different components of the subject population.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Cohort Characteristics and Air Pollutant Concentrations

Characteristics of the cohort are presented in Table 1. Mean summertime (1995 and 1996) air pollutant concentrations and meteorologic measures from the VSAM, as well as Pearson correlation coefficients, are summarized in Table 2. Continuous time series of 24-h PM_2.5, daily maximal 8-h average O₃, and daily averages of normalized morning PEF values for all subjects are presented in Figure 1. At no point during the study period was the U.S. EPA 24-h PM_2.5 National Ambient Air Quality Standard (NAAQS) (65 µg/m³) (23) or the World Health Organization (WHO) suggested lowest observed effect level for 24-h PM₁₀ (110 µg/m³) (24) exceeded. The U.S. EPA 8-h O₃ NAAQS of 80 ppb (25) was exceeded on only 2 d throughout the entire study period. PM₁₀, PM_2.5, SO₄², H⁺, and NH₄⁺ concentrations were strongly correlated with each other at the VSAM. PM_2.5 accounted for approximately 74% of the PM₁₀ (PM_2.5/PM₁₀ ratio of 0.74 ± 0.098) and SO₄² accounted for approximately 41% of the PM_2.5 (SO₄²/PM_2.5 ratio of 0.41 ± 0.10). PM₁₀ and SO₄² were strongly correlated with PM_2.5 (r = 0.96 and 0.92, respectively). H⁺ was strongly correlated with PM_2.5 (r = 0.81) and with SO₄² (r = 0.89), the latter suggesting the presence of acid sulfate. The strong correlation between NH₄⁺ and SO₄² suggests that a major form of sulfate in the region is ammonium or bisulfate. The ion balance for SO₄² aerosol (ratio of equivalents of cations to equivalents of anions) resulted in a ratio of one, indicating that sulfate could be accounted for by H⁺ and NH₄⁺. Sulfate appears to comprise the major portion of the PM_2.5 aerosol. SO₂ was not correlated with any pollutants, suggesting a regionally formed sulfate.

View larger version (40K): [in this window] [in a new window]   Figure 1.   24-h average PM2.5 and daily maximal 8-h average O3 at the Vinton stationary ambient monitoring site, and daily variations in morning PEF: summer 1995.

Mixed Linear Random Coefficients Modeling: PEF versus Air Pollutant Concentrations and Meteorological Parameters

Morning PEF. Modeling results for both morning and evening PEF are summarized in Table 3. For morning PEF using the delta normalization, same day measurements of H⁺ (p = 0.011), PM_2.5 (p = 0.030), PM₁₀ (p = 0.020), and coarse mode (p = 0.049) were the only significant parameters whereas SO₄² (p = 0.097) and precipitation (p = 0.085) were marginally significant. No other parameters, including temperature (p = 0.169) and humidity (p = 0.841) were related to morning PEF as main effects in models. For each increase of 10 µg/m³ of PM_2.5, a decrease of 0.73 L/min, which related to a 0.16% decrease in PEF (95% confidence interval [CI] = 0.02%, 0.31%), was observed. Similarly, for each increase of 50 nmol/ m³ of H⁺ (50 nmol H⁺ = 2.5 µg H₂SO₄), a decrease of 0.89 L/ min (0.20%; 95% CI = 0.05%, 0.35%) was observed. The percent reduction in variance of PEF across individuals (i.e., the extent to which the air contaminant explained the overall variancecalculated as residual variance in fitted model divided by total residual variance in unrestricted model) is 2.2% (398.2/407.0) for PM_2.5, 2.6% (397.5/408.2) for PM₁₀, and 2.1% (406.6/415.4) for H⁺. The same day exposure measurements for H⁺, PM_2.5, and PM₁₀ were more significant in models with morning PEF as the dependent variable than models using lag or cumulative lag exposure effects (Table 3). Of the various O₃ averaging times and lag effects considered, only O₃ average with a 1-d lag (p = 0.105) and O₃ average of the previous 5-d average (p = 0.081) were statistically significant.

When the main effects (H⁺ and PM_2.5 or PM₁₀) were analyzed as individual main effects in multivariate models related to morning PEF, none remained significant in the presence of any other. For example, when PM_2.5 and H⁺, which are highly correlated, were run together as two separate main effects in a multivariate model, neither was significant at p < 0.05. In this model, H⁺ was marginally significant (p = 0.102) and PM_2.5 did not approach significance (p = 0.647). Significant PM parameters (PM_2.5, PM₁₀, and coarse mass) were tested separately with H⁺ as an interaction term (e.g., PM*H⁺) to explore the possibility of a particle/particle-strong acid interaction; none of the interaction terms were significant.

Evening and daily change PEF. Ozone was the only air pollution and meteorological sample that remained significant as an independent variable in models with evening PEF as the dependent variable. Among all O₃ averaging times and lag effects considered, only O₃ average with 1- and 3-d lag (p = 0.051 and 0.024, respectively) and O₃ previous 3- and 5-d average (p = 0.027 and 0.006, respectively) were significant. The observed associations between O₃ and decrements in PEF were 3.05 (95% CI = 0.40, 5.69) and 7.65 L/min (95% CI = 2.25, 13.0), for 3-d lag and 5-d cumulative lag exposure, respectively, per 30 parts per billion (ppb) increase in O₃. These correspond to evening PEF decreases of 0.66% for O₃ 3-d lag effect, and 1.66% for O₃ 5-d cumulative lag effect, respectively (Table 3). In a model of previous 5-d average O₃ versus evening PEF, the percent reduction in variance of PEF across individuals explained by O₃ in the model is 5.2% (376.5/397.1). No air pollution or meteorological measures were significant as main effect independent variables in models with daily change in PEF (Evening PEF  Morning PEF) as the dependent variable.

Tests for trend, alternative PEF normalization methods, and stratified mixed models. Ambient air pollution data were stratified into quartiles and plotted against normalized PEF data. When exposure data (versus morning PEF: 24-h PM_2.5, PM₁₀, and H⁺; versus evening PEF: previous 5-d average O₃) are plotted against PEF, a near-linear trend is observed for each contaminant with lower PEF values corresponding to higher air pollution levels (Figure 2). For individual models with morning PEF as the dependent variable and previous day 24-h average PM_2.5, PM₁₀, H⁺, and SO₄², respectively, as independent variables, PM₁₀ (p = 0.030) and previous 5-d average O₃ (p = 0.016) were the only exposure effects that had a significant test of trend whereas PM_2.5 (p = 0.192), H⁺ (p = 0.249), and SO₄² (p = 0.258) were not statistically significant.

View larger version (25K): [in this window] [in a new window]   Figure 2.   Normalized morning PEF (versus PM2.5, PM10, H+) and evening PEF (versus O3) plotted by quartile of air contaminant.

When the models reviewed in Table 3 were run using other PEF normalization methods (body size normalization and Z-score normalization), the results were similar to those described in Table 3. For example, same day (lag 0) PM_2.5 when modeled against morning PEF normalized with the delta transformation found a 10 µg/m³ increase in PM_2.5 related to a PEF decrement of 0.73 L/min. Using body size and the Z-Score transformation, decreases of 0.66 and 0.50, respectively, were observed. Similar results are seen when results of the three transformations are compared for PM₁₀, coarse mass, SO₄², H⁺, and O₃.

When the models with significant main effects (e.g., same day PM_2.5 and H⁺ versus morning PEF and previous 5-d O₃ versus evening PEF) were reanalyzed with data stratified by subject demographics, SES, health history, and home characteristics, no strongly significant differences emerged.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Outdoor exposures to air pollution modify bronchial responsiveness and cause acute transient bronchoconstriction (11, 26). While the American Thoracic Society has not defined acute, largely reversible changes in lung function, such as those measured in this study, as an adverse health effect, both U.S. EPA (27) and WHO (24) have accepted acute changes in lung function as adverse indicators of effects of exposure to O₃ pollution (28).

Previous studies of particulate and lung function have focused on TSP and PM₁₀ in children (3, 12, 13, 15) or those with respiratory illnesses (3, 4, 16). In this study, women who had recently given birth were found to have peak flow changes associated with fine particulate fraction (PM_2.5) and particle strong acid (H⁺) exposures consistently below the NAAQS for Particulate Matter. These results remained largely unchanged when the 35 asthmatics were dropped from analysis, and could not be explained by smoking, environmental tobacco smoke, or other indoor air quality factors.

In this study, a 10 µg/m³ increment in 24-h PM_2.5 was associated with a 0.73 L/min (95% CI = 0.07, 1.38) decrease in morning PEF. These findings are consistent with previous studies of healthy and nonasthmatic symptomatic or asthmatic children. In a study of 71 asthmatic children in Mexico City, Romieu and coworkers (17) found a decrease of 0.86 L/min (95% CI = 0.17, 1.55) in evening PEF, and a decrease of 1.18 L/min (95% CI = 0.43, 1.93) in morning PEF, associated with each additional 10 µg/m³ 24-h PM_2.5. In separate studies, Neas and coworkers (14, 15) found a decrease of 0.55 L/min per 10 µg/m³ increment of 12-h PM_2.1 (95% CI = 0.01, 1.09) and 0.22 L/min per 10 µg/m³ increment in 24-h PM_2.1 (95% CI = 0.25, 0.69) in nonasthmatic symptomatic and asymptomatic children. For perspective, a normal PEF for a hypothetical 25-yr-old woman who is 5'6 tall is aproximately 500 L/min (29), which compares well with the mean morning average PEF of 457 L/min found in the present study.

A 50 nmol/m³ increment in 24-h fine particulate H⁺ was found associated with a decrease of 0.89 L/min (95% CI = 0.21, 1.57) in morning PEF in our study. These findings are consistent with previous studies of healthy and symptomatic or asthmatic children. Inverse variance weights were used by Neas and coworkers to combine the results of their studies in Uniontown (15) and State College, Pennsylvania (14). They found a 12-h daytime time-weighted exposure to an additional 50 nmol/m³ H⁺ associated with a 0.72 L/min (95% CI = 0.28, 1.16) decrease in evening PEF. Raizenne and coworkers (30) followed 96 Canadian children over three 13-d summer camp sessions in 1986. They found a 50 nmol/m³ increment in 12-h particle strong acidity associated with a 0.56 to 0.76 decrease in PEF for the first two periods. Particle strong acidity was associated with improved PEF during the third period.

A 10 µg/m³ increment in 24-h PM₁₀ was found to be associated with a decrease of 0.73 L/min (95% CI = 0.11, 1.35) in morning PEF in our study. As shown in Table 4, these findings from a healthy population of women are consistent with previous studies of asymptomatic (3, 12, 13, 15), nonasthmatic symptomatic (3, 15, 16), and asthmatic (4, 17, 18) children as well as adults with asthma (4, 18) or under medication for obstructive airway disease (19).

In this study, a 30 ppb increment in 24-h average O₃ was associated with a decrease of 2.49 L/min (95% CI = 0.01, 5.01) in evening PEF. These results are consistent with previous studies. Neas and coworkers (15) observed a decrease of 2.79 L/min in PEF for 12-h personal exposure or a 30-ppb increment in daytime O₃. Lioy and coworkers (31) reported that a 30 ppb increment in O₃ (1-h daily peak) was associated with a decreased PEF of 5.4 L/min (95% CI = 2.4, 8.3). Berry and coworkers (32) reported that a 30 ppb increment in O₃ (8-h average) was associated with a decreased PEF of 4.2 L/min (95% CI = 0.9, 9.3). Similarly, Spektor and associates (33) reported a decrease of 4.4 L/min (95% CI = 0.9, 7.8). More recently, Krzyzanowski and colleagues (34) reported that a 30 ppb increment in O₃ (1-h maximum) was associated with a decrease in PEF among nonasthmatic children of 3.57 L/min (95% CI = 0.1, 7.3). Thurston and colleagues (35) found a decrease of 6.6 L/min for each 30 ppb O₃.

The effects of cumulative lagged exposure to air pollution in this study are similar to those in earlier studies. Pope and Dockery (3) found a decrease in PEF associated with a 5-d mean that was four times larger than same day effect estimates for PM₁₀. Roemer and coworkers (16) found a decrease in PEF associated with 7-d mean that was two times larger than same day effect estimates for PM₁₀. Peters and coworkers (18) observed that the decrease in PEF with PM₁₀ became more pronounced when a 5-d mean was used to represent cumulative effects of exposure to ambient air pollution. Romieu and coworkers (17) reported a similar finding for both morning and evening PEF associations with same day versus 7-d average ambient air pollution exposures. In our study, same day and 5-d cumulative average PM_2.5 are associated with morning PEF decrements of 0.73 and 0.89 L/min per 10 µg/m³ PM_2.5.

The variability around subject-specific slope effects, reflecting PEF response to air pollution, is large in comparison to the mean, demonstrating that while many individuals may exhibit an effect, others will show no effect or even a positive association. For example, in a model of morning PEF versus 24-h PM_2.5, the slope estimate of 0.073 has a standard deviation of 0.329 for the population of random slope effects, which indicates substantial variability of the effect of PM_2.5 on PEF among individuals. These results (i.e., small negative parameter estimate with a comparatively large standard deviation for the population of estimates) suggest that some individuals may be more sensitive than others to the effects of air pollution on PEF.

Air pollution concentrations of PM_2.5 measured in this study are considerably lower than in previous studies (see Table 4). Despite this, the observed relationship between PEF and particulate air pollution is largely comparable to those found previously. A linear exposure response for PEF from very low PM_2.5 to very high PM_2.5 levels is suggested by a combination of results found at low concentrations in this study with those from other studies done at higher concentrations.

Despite differences in chemical characteristics of ambient air in prior studies, each reported similar decrements in PEF associated with exposure to ambient PM₁₀, suggesting H⁺ is not the sole contaminant responsible for observed effects. Panel studies by Pope and Dockery (3) and Pope and coworkers (4) were conducted in an environment with H⁺ levels that did not exceed a detection limit of 8 nmol/m³ and relatively high particulate concentrations, whereas Neas and coworkers (15, 14) conducted studies in an environment with lower particulate concentrations and high H⁺ concentrations (12-h averages of 102.0 [all observations] and 119.7 [daytime] nmol/m³, respectively). Because of high correlation of the air pollution measures (e.g., PM_2.5, PM₁₀, coarse mass, fine particulate H⁺, SO₄²), it is not possible to delineate the air pollution component responsible for PEF decrements observed in the present study.

Both the present study and Peters and coworkers (18) found small PEF decrements associated with SO₄² compared with those observed for PM_2.5, suggesting that it too is not the sole contaminant responsible for the observed effect. Regarding the size fraction of particulate matter, PM_2.5 and PM₁₀ had virtually the same effect on morning decrements in PEF. The results of our study were similar regardless of which method was used to normalize the PEF data.

It is not clear why morning PEF, which is lower on a diurnal basis and follows sleep for most subjects, is more responsive to PM_2.5 than evening PEF. In similar studies reviewed in Table 4, Dusseldorp and coworkers (19) and Roemer and coworkers (16) both found larger decrements in morning PEF than in evening PEF in relation to ambient PM₁₀. Conversely, Romieu and coworkers (17) found larger decrements in evening PEF than morning PEF in relation to ambient PM₁₀. A majority of studies found effects in evening PEF related to ambient PM₁₀ and did not consider morning PEF (Table 4).

Select results (e.g., 24-h PM_2.5 or H⁺ versus morning PEF; 5-d O₃ versus evening PEF) from the Mixed Linear Random Coefficients Models in our study were reanalyzed after stratification by subject health, home, and SES characteristics. Comparison of parameter estimates from the models revealed no notably significant differences in sensitivity among the stratified groups.

The present study has several limitations. It is likely that some incorrect reporting of PEF occurred owing to subject-specific error in measuring and recording. However, it is likely that any incorrectly measured data were nondifferential and not associated with corresponding daily levels of air pollution. Nondifferential misclassification would tend to bias the estimates from bivariate analyses toward the null, and would thus result in an underestimate of actual effects. However, in a more complex setting in which there are also nondifferential errors in the assessment of exposure, the effect on the estimates can be less predictable (36).

This study examined associations for several air quality measures as well as considering different lag times between level of exposure and PEF. Therefore, multiple comparisons are being made and some statistically significant ones might have occurred by chance. However, the number of observed significant associations is greater than would be expected by chance alone. Because there is considerable correlation among the measures of air quality, especially among the various lagged measurements, we chose not to introduce a correction factor for the p value. Nonetheless, ultimate conclusions about the effect of these associations would benefit from confirmation in further studies.

The current study used a fixed site to determine exposure for the subject population. Personal exposure may be quite different from concentrations measured at a central site (37). We found the central site adequately represented concentrations of PM_2.5 and SO₄² during summer months outside and inside homes within the region, even though a substantial number (89%) of the homes use air conditioning (20). Absent any systematic bias, this should not invalidate the findings of this study. In fact, it is because of difficulty measuring exposure that large epidemiologic studies such as this one are needed, rather than studies of much fewer subjects. In the event that exposure bias did occur in this study, it would probably bias results toward the null (14).

Because of high correlation of air pollution components, our study, like prior air pollution studies, is limited in its ability to delineate components of air pollution responsible for observed associations with PEF decrements. Similarly, because of high correlation of air pollution concentrations observed from day to day, our study is limited in its ability to delineate the role of exposures from days prior to observed PEF decrements.

The present study may not have adequately controlled for confounding by other respiratory function risk factors, such as aeroallergens, which were not measured. Because pollen levels tend to be relatively low in summer, it is less likely to be a confounder. Thurston and colleagues (35) considered pollens in their summer study and found no associations between PEF and pollen counts. Because people are reactive to so many different pollens, no one pollen has a detectable effect. Hence, it is unlikely that pollen would be a significant factor for this study. The exposure database was quite extensive and considered a complete spectrum of ambient air contaminants and meteorological data. Smokers and those exposed to environmental tobacco smoke in the home were excluded and home, health, and socioeconomic status data were considered in the analyses to control for potential confounding. The present study is restricted to women, who may have slightly greater airway reactivity than men (38). It is possible that dose-response relationships described here may differ for healthy adult males.