ResearchOpen Access

Long-Term Air Pollution Exposure and Blood Pressure in the Sister Study

    Published: by:36



    Exposure to air pollution has been consistently associated with cardiovascular morbidity and mortality, but mechanisms remain uncertain. Associations with blood pressure (BP) may help to explain the cardiovascular effects of air pollution.


    We examined the cross-sectional relationship between long-term (annual average) residential air pollution exposure and BP in the National Institute of Environmental Health Sciences’ Sister Study, a large U.S. cohort study investigating risk factors for breast cancer and other outcomes.


    This analysis included 43,629 women 35–76 years of age, enrolled 2003–2009, who had a sister with breast cancer. Geographic information systems contributed to satellite-based nitrogen dioxide (NO2) and fine particulate matter (≤ 2.5 μm; PM2.5) predictions at participant residences at study entry. Generalized additive models were used to examine the relationship between pollutants and measured BP at study entry, adjusting for cardiovascular disease risk factors and including thin plate splines for potential spatial confounding.


    A 10-μg/m3 increase in PM2.5 was associated with 1.4-mmHg higher systolic BP (95% CI: 0.6, 2.3; p < 0.001), 1.0-mmHg higher pulse pressure (95% CI: 0.4, 1.7; p = 0.001), 0.8-mmHg higher mean arterial pressure (95% CI: 0.2, 1.4; p = 0.01), and no significant association with diastolic BP. A 10-ppb increase in NO2 was associated with a 0.4-mmHg (95% CI: 0.2, 0.6; p < 0.001) higher pulse pressure.


    Long-term PM2.5 and NO2 exposures were associated with higher blood pressure. On a population scale, such air pollution–related increases in blood pressure could, in part, account for the increases in cardiovascular disease morbidity and mortality seen in prior studies.


    Chan SH, Van Hee VC, Bergen S, Szpiro AA, DeRoo LA, London SJ, Marshall JD, Kaufman JD, Sandler DP. 2015. Long-term air pollution exposure and blood pressure in the Sister Study. Environ Health Perspect 123:951–958;


    There is a well-established relationship between combustion-related air pollution exposure—especially particulate matter ≤ 2.5 μm in diameter (PM2.5)—and cardiovascular disease (CVD) morbidity and mortality (Brook et al. 2010). Although there have been numerous studies that demonstrate this relationship, the mechanisms are poorly understood.

    One potential mechanism is an effect of inhaled air pollution on blood pressure (BP), mediated through autonomic nervous system dysfunction and/or changes in inflammation and oxidative stress. Increased BP is a strong risk factor for CVD including increases in left ventricular mass, which have been associated with long-term air pollution exposures (Van Hee et al. 2009).

    Recent work has suggested that short-term (hours to days) particulate matter and traffic-related pollutant exposures may lead to transient increases in BP (Baccarelli et al. 2011; Baumgartner et al. 2011; Brook et al. 2011; Cosselman et al. 2012; Hoffmann et al. 2012; Langrish et al. 2012; Wu et al. 2013). In contrast, a study of 9,238 nonsmoking adults in Taiwan found reductions in systolic BP (SBP) and pulse pressure (PP) following short-term exposure to air pollution (Chen et al. 2012).

    The relationship between chronic, long-term (e.g., yearly average) air pollution exposure and BP is less well understood, with some studies demonstrating an increase in BP associated with PM2.5 (Chuang et al. 2011; Fuks et al. 2011) and black carbon (Schwartz et al. 2012) exposure. Additional studies have investigated associations of BP with oxides of nitrogen (NOx; a marker of traffic-related pollution) (Dong et al. 2013; Sørensen et al. 2012), or have investigated the associations between BP and long-term exposures to both PM2.5 and gaseous traffic-related pollution exposure (Chuang et al. 2011; Coogan et al. 2012).

    Developments in fine-scale spatial modeling of air pollution—using advanced statistical methods, geographic information systems, and both ground-based and satellite-based monitoring information—are now available. Together with large national cohorts, these exposure advances provide the opportunity for an improved analysis of this important research question.

    We conducted a cross-sectional study to evaluate the relationship between BP (systolic, diastolic, pulse pressure, and mean arterial pressure) and long-term (annual average) exposure to PM2.5 and nitrogen dioxide (NO2) in a large U.S. cohort of women.


    Study population. Study participants were selected from the Sister Study, a large nationwide, prospective women’s cohort study investigating environmental and genetic risk factors for breast cancer and other diseases. 50,884 sisters of women with breast cancer, 35–76 years of age, were enrolled into the cohort between 2003 and 2009, as described elsewhere (Weinberg et al. 2007). The Sister Study was approved by the Institutional Review Board (IRB) of the National Institute of Environmental Health Sciences, National Institutes of Health, and the Copernicus Group IRB; all participants provided informed consent. In this analysis, participants were excluded due to residence outside of the continental United States (2% of participants), invalid address information (6%), missing BP measurement (0.3%), missing modeled NO2 estimates (0.06%), or other missing key covariate data (6%). Therefore, this analysis includes 43,629 (86%) of the recruited participants residing in the conterminous United States at enrollment.

    Computer-assisted telephone interviews were administered by extensively trained staff, who collected information on participant demographics, socioeconomic status (SES) factors, residential history, occupational history, personal medical history (including self-reported diabetes, hypercholesterolemia, and hypertension), medication use, perceived stress (four-item perceived stress scale) (Cohen et al. 1983), and behavioral factors such as alcohol use and smoking. Participants were asked whether they had ever been diagnosed with diabetes, hypercholesterolemia, and hypertension by a medical professional. Responses were self-reported as no, yes, or “borderline,” with the last category added to accommodate participants who have been told that they had or nearly had the condition but did not require medications. Medication lists were coded using the Slone Drug Dictionary (Kelley et al. 2003), and anti-hypertensive medication use was defined as self-reporting one or more drugs in anti-hypertensive drug classes.

    Women were enrolled throughout the United States and completed telephone interviews as close to the time they volunteered as possible; participation was not geographically or seasonally clustered. Home visits were conducted by examiners from a national company that performs insurance physicals, and were not scheduled in a manner to maximize geographic efficiency. The home visits provided measurements of anthropometry, fasting phlebotomy, and BP.

    Approximately 10% of participants were sisters with one or more study participant, and the analyses do not account for familial clustering in the population because the most common cluster size was very small.

    Blood pressure ascertainment. During baseline home visits, following consent and review of self-completed forms, participants were instructed to sit and rest for a few minutes before BP ascertainment. Trained examiners made three consecutive measurements of BP using an aneroid sphygmomanometer (model 760 & 775X; American Diagnostic Corporation). Measurements were taken from alternating arms, starting with the left arm using a left-right-left protocol, approximately 2 min apart. Examinations were scheduled, whenever possible, in the morning, and participants were encouraged to fast before the visit (excluding medications) and record whether anything had been taken by mouth.

    For SBP and diastolic blood pressure (DBP) separately, the second and third measurements were averaged when three measurements were available. In some cases, examiners were unable to obtain three BP measurements. When only two BP measurements were available (n = 1,677), the two were averaged; and when only one BP measurement was recorded, the single value was used (n = 684).

    Because the mechanism through which air pollution exposure may affect BP is not well understood, we also examined PP and mean arterial pressure (MAP), as other studies have also done (Auchincloss et al. 2008; Chen et al. 2012). PP, representing stroke volume and vascular compliance (Dart and Kingwell 2001), was determined by subtracting DBP from SBP; and MAP, a function of ventricular contractility, resistance, elasticity, and heart rate (Sesso et al. 2000), was calculated by PP/3 + DBP.

    Exposure assessment. Participant home latitude and longitude at study entry was geocoded using ArcGIS 9.3.1 or 10.1 (ESRI, Redlands, CA) in conjunction with TeleAtlas Dynamap 2000 v16.1 road network (TeleAtlas, Boston, MA). Based on the residential geocodes, we assigned the census block.

    For PM2.5, we developed a national prediction model for the year 2006, using partial least squares to select relevant components for the mean regression and universal kriging for spatial smoothing (Sampson et al. 2013). Briefly, the PM2.5 prediction model included satellite-based land use/land cover, road network characteristics, population density, vegetative index, distance to selected geographic features, and annual average U.S. Environmental Protection Agency’s Air Quality System monitor concentrations (; see also Sampson et al. 2013). The model was fit using maximum likelihood, with each region having its own parameters (cross-validated R2 = 0.88). Individual PM2.5 concentrations were predicted for each residential geocode.

    National NO2 predictions were developed using a previously described satellite-based land-use regression model for the year 2006 (Novotny et al. 2011). In short, atmospheric NO2 surface concentrations were predicted using multivariable linear regression based on land-use characteristics (impervious surfaces, tree canopy, sum of road lengths, elevation, and distance to coast) and tropospheric NO2 column abundance measurements from the Aura satellite’s ozone monitoring instrument (Novotny et al. 2011) (cross-validated R2 = 0.78). Individual NO2 concentrations were assigned based on the census block of the subject’s residential address.

    Predicted annual average PM2.5 and NO2 concentrations were used to approximate long-term residential exposure at the time of baseline examination (2003–2009). The correlation between PM2.5 and NO2 for this population was 0.37, and although both exposure models contain similar terms, the modeling approaches are quite different.

    Other geographic covariate measurement. To describe the overall urbanicity of the county in which participants reside, we used the Rural–Urban Continuum Codes of the U.S. Department of Agriculture (2013). The socioeconomic environment of the participants’ neighborhoods was defined by using neighborhood-level SES z-score based on U.S. Census block groups, which has been used in other studies (Diez Roux et al. 2001). A higher SES z-score signifies higher socioeconomic advantage.

    Statistical analysis. For descriptive analyses, annual average air pollution exposure predictions (PM2.5 and NO2) and BP parameters (SBP, DBP, MAP, and PP) were divided into quartiles. Global F-tests (analysis of variance) were used to examine the differences in mean values of continuous variables (age, BP parameters, pollution measures) across quartiles of pollutants and BP parameters. The chi-square test was used to compare the frequencies of categorical variables across quartiles of exposure and outcomes. Categorical covariates were included in the main models and interactions as defined in Tables 1 and 2. To examine the overall spatial distribution of the exposures and outcomes, we plotted the mean BP parameters and air pollution exposure metrics for the participants by state, county, and census tract on U.S. maps.

    Table 1 Baseline demographic characteristics of participants (n, mean ± SD, or %).

    CharacteristicQuartile of exposure to PM2.5 (μg/m3)Quartile of exposure to NO2 (ppb)All participants
    2.2– 8.88.8– 10.810.8– 12.412.4– 17.41.0– 6.46.4– 9.29.2– 12.612.6– 34.2
    No. of participants (n)10,92910,92410,91510,86110,92710,91710,88410,90143,629
    Age (years)55.5 ± 8.955.1 ± 9.154.8 ± 8.954.5 ± 8.955.3 ± 8.855.1 ± 8.954.8 ± 9.054.7 ± 9.055.0 ± 8.9
    Race or ethnic group (%)
    Non-Hispanic white928986759188847885
    Household income (%)
    < $20,000262523252825232325
    $20,000 to < $50,000454544434645444144
    $50,000 to < $100,000262628272325283027
    ≥ $100,000445534464
    Education (%)
    ≤ High school141615141716141215
    Some college353532323735333134
    Bachelor’s or above514952534650535752
    Married (%)767272638074686071
    Working > 20 hrs/week (%)586061645960616461
    Perceived stress score (%)
    Low (0–2)605757555958575557
    Medium (3–6)343536363435353735
    High (> 6)788877888
    Stable residence (%)575962625957596560
    Neighborhood SES z-score tertile (%)
    Rural–Urban Continuum Code (%)
    Metro area ≥ 1 million395858722544679057
    Metro area < 1 million392931234241301031
    Non-metro county221212533143013
    Metro, metropolitan. Shown as annual neighborhood SES z-score tertile: The socioeconomic environment of the participants’ neighborhoods was defined by U.S. Census block group characteristics. A higher SES z-score signifies higher socioeconomic advantage.

    Table 2 Baseline health characteristics of participants (mean ± SD or %).

    CharacteristicQuartile of exposure to PM2.5 (μg/m3)Quartile of exposure to NO2 (ppb)All participants
    Systolic BP (mmHg)114.3 ± 13.7114.6 ± 13.2114.8 ± 13.6115.6 ± 14.1115.2 ± 13.5114.6 ± 13.6114.4 ± 13.6115.0 ± 13.9114.8 ± 13.6
    Diastolic BP (mmHg)72.0 ± 8.872.2 ± 8.672.3 ± 8.773.1 ± 972.5 ± 8.672.3 ± 8.872.2 ± 8.972.5 ± 8.972.4 ± 8.8
    Mean arterial (mmHg)86.1 ± 9.786.3 ± 9.386.4 ± 9.587.3 ± 9.986.8 ± 9.486.4 ± 9.686.3 ± 9.786.7 ± 9.786.5 ± 9.6
    Pulse pressure (mmHg)42.3 ± 9.742.4 ± 9.642.5 ± 9.942.4 ± 1042.7 ± 9.842.3 ± 9.842.2 ± 9.742.6 ± 9.942.4 ± 9.8
    Body mass index (kg/m2)
    Normal (< 25)423938343738393938
    Overweight (25 to < 30)313132323332323132
    Obese (≥ 30)273030343030303030
    Smoking status
    Alcohol use
    On BP medication283030333130303030
    Borderline, self-reported classification that the participant had or nearly had the condition but did not require medications.

    We then fit multivariable linear models to investigate the relationship between individual BP parameters and each of the two pollutants of interest, adjusted for potential confounders including space [using unpenalized thin-plate regression splines (TPRS) in the MGCV package] (Wood 2003). TPRS are a flexible way of adjusting for spatial confounding. Using singular value decomposition, they decompose the distance matrix of all participant locations into a set of basis functions, the first k of which are included as adjustment covariates in the health models (Wood 2003).

    Our final model included all covariates considered a priori as potential confounders. The a priori selection was based on a review of the literature before the analysis to avoid model selection bias. To evaluate the effect of groups of covariates, we added variables to successive models in series, with model 1 including age and race/ethnicity; model 2 also including SES variables (household income, education, marital status, working ≥ 20 hr per week outside the home, perceived stress score, and SES z-score); model 3 additionally including spatial features that are likely to vary both with pollution and BP (Rural–Urban Continuum code and TPRS for latitude and longitude); model 4 additionally including CVD risk factors [body mass index (BMI), waist-to-hip ratio, smoking status, alcohol use, history of diabetes, and history of hypercholesterolemia]; and the full model 5 additionally including BP medication use. For the categorical SES variables in model 2, we assume that collinearity does not exist because within the levels of each categorical variable there is some heterogeneity of the other categorical variables. Unpenalized TPRS for latitude and longitude were fit in two dimensions using 10 degrees of freedom (df). Statistical analyses were carried out using R 2.15.0 (R Core Team 2013) and Stata/IC 12.1 (StataCorp LP, College Station, TX). In all instances, a p-value of < 0.05 was considered significant.

    When we observed significant associations with exposure in the full model, we additionally explored interactions with race/ethnicity, age, BMI, smoking, diabetes, and anti-hypertensive medication use by adding product terms of these variables with the exposure variable, and we examined interactive effect sizes and 95% confidence intervals (CIs) within strata using linear combinations of terms from the regression models (using wald.test and svycontrast in R).

    Because there may be spatially varying characteristics that we were unable to account for, sensitivity analyses included varying the number of df for spatial adjustment and investigating the impact on main effect sizes and standard errors of alternate forms of the other independent and dependent variables (including nonlinear associations for the exposure metrics using penalized TPRS). To provide a complementary view, logistic regression was used to examine the hypertension as an outcome, defined as using anti-hypertensive medication or having an SBP ≥ 140 mmHg and DBP ≥ 90 mmHg. We also examined the effect of several subgroup analyses, restricting the full model analysis to individuals with stable residence (defined as the current address at the time of the examination representing their longest lived address) to account for potential exposure misclassification from characterizing current residence as a location of long-term exposure, and, separately, restricting the analysis to those with three valid, left-right-left arm, BP measurements to examine precision based on potential BP measurement error. Finally, we examined models including both air pollution exposure variables in a co-pollutant model.


    Participant characteristics. Table 1 presents baseline demographic characteristics and Table 2 shows baseline health characteristics of participants, overall and by quartile of pollutant exposure. Among the 43,629 women, the mean ± SD age was 55 ± 8.9 years; range, 35–76 years. Thirty-one percent had self-reported hypertension or “borderline” hypertension, and 30% were on anti-hypertensive medications. Participants lived at their current address for a median of 11 years [interquartile range (IQR) of 16 years], ranging from < 1 year to 75 years.

    Bivariate associations. Compared with the remainder of the sample, the highest quartile of both NO2 and PM2.5 exposure was significantly associated with younger participants, fewer non-Hispanic whites and more blacks, higher household income, fewer married women, more working > 20 hr/week, higher stress scores, greater residential stability, and with living in large metropolitan areas. Higher NO2 (but not PM2.5) quartile was associated with higher neighborhood SES, less overweightness, more former smokers, and more current alcohol users, whereas higher PM2.5 (but not NO2) was associated with significantly lower SES z-scores, more obesity, more current smokers, and fewer current alcohol users. NO2 was not associated with diabetes or anti-hypertensive medication use but was associated with self-reported hypertension and hypercholesterolemia, whereas higher PM2.5 was associated with more diabetes, higher anti-hypertensive use, and more self-reported hypertension but not hypercholesterolemia in these unadjusted univariate comparisons. All risk factors and other SES and geographic covariates were highly statistically significantly associated with quartiles of SBP, DBP, MAP, and PP (data not shown).

    Residential pollutant exposures. Figure 1 shows the distribution of participants’ geocoded residential locations, with numbers representing the number of participants per state. The distribution of participants generally corresponds to the distribution of population across the United States. Figure 2 presents boxplots of the distribution of exposure predictions for PM2.5 and NO2, by U.S. census division. See Supplemental Material, Figures S1 and S2, for maps of mean pollutant levels of participants by U.S. census tract. PM2.5 shows large-scale spatial structure across the United States. NO2 exhibits a different spatial pattern, with high levels in highly urbanized areas, reflecting the traffic-related nature of NO2. Thus, PM2.5 exhibits greater between-city variability, whereas NO2 exhibits more within-city variability.

    Figure 1 United States map of participant residential locations, with number of participants per state. Each participant is represented by an open blue circle.

    Figure 2 Boxplots of PM2.5 and NO2 participant annual average residential concentrations by U.S. census division. Boxes extend from the 25th to the 75th percentile, horizontal bars represent the median, whiskers extend 1.5 times the length of the interquartile range (IQR) above and below the 75th and 25th percentiles, respectively, and outliers are represented as points.

    Adjusted relationship between pollutants and BP. Figure 3 shows the results of adjusted linear models by pollutant. In the fully adjusted models (model 5) shown in Table 3, a 10-μg/m3 increment in PM2.5 was associated with a 1.4-mmHg higher SBP (95% CI: 0.6, 2.3; p < 0.001), a 1.0-mmHg higher PP (95% CI: 0.4, 1.7; p = 0.001), a 0.8-mmHg higher MAP (95% CI: 0.2, 1.4; p = 0.01), and a 0.4-mmHg higher DBP (95% CI: –0.2, 1.0; p = 0.15). A 10-ppb increase in NO2 was associated with a 0.4-mmHg (95% CI: 0.2, 0.6; p < 0.001) higher PP, a 0.2-mmHg higher SBP (95% CI: 0.0, 0.5; p = 0.10), a 0.2-mmHg lower DBP (95% CI: –0.4, 0.0; p = 0.05), and no difference in MAP (95% CI: –0.2, 0.1; p = 0.63).

    Figure 3 Relationship between blood pressure and annual average air pollution exposure for PM2.5 (left) and NO2 (right). Model 1: Included age and race/ethnicity. Model 2: model 1 + household income, education, marital status, working ≥ 20 hr per week outside the home, perceived stress score, and socioeconomic status z-score. Model 3: model 2 + Rural–Urban Continuum Codes and unpenalized thin-plate regression splines for latitude and longitude. Model 4: model 3 + body mass index, waist-to-hip ratio, smoking status, alcohol use, history of diabetes, and history of hypercholesterolemia. Model 5: model 4 + blood pressure medication use.

    Table 3 Estimated effect of PM2.5 and NO2 exposure on blood pressure (mmHg), estimate (95% CI).

    OutcomePer 10 μg/m3 PM2.5 exposurePer 10 ppb NO2 exposure
    mmHg (95% CI)p-ValuemmHg (95% CI)p-Value
    Systolic blood pressure
    Model 10.8 (0.3, 1.3)0.002–0.4 (–0.6, –0.1)0.003
    Model 20.9 (0.4, 1.4)< 0.0010.2 (–0.1, 0.4)0.17
    Model 31.9 (1.0, 2.8)< 0.0010.3 (0.0, 0.6)0.07
    Model 41.5 (0.7, 2.4)< 0.0010.2 (0.0, 0.5)0.09
    Model 51.4 (0.6, 2.3)< 0.0010.2 (0.0, 0.5)0.10
    Diastolic blood pressure
    Model 10.8 (0.4, 1.1)< 0.001–0.3 (–0.5, –0.2)< 0.001
    Model 20.7 (0.4, 1.1)< 0.001–0.1 (–0.3, 0.0)0.11
    Model 30.7 (0.1, 1.3)0.03–0.2 (–0.4, 0.0)0.10
    Model 40.5 (–0.1, 1.0)0.12–0.2 (–0.4, 0.0)0.06
    Model 50.4 (–0.2, 1.0)0.15–0.2 (–0.4, 0.0)0.05
    Mean arterial pressure
    Model 10.8 (0.4, 1.2)< 0.001–0.3 (–0.5, –0.2)< 0.001
    Model 20.8 (0.4, 1.2)< 0.0010.0 (–0.2, 0.1)0.72
    Model 31.1 (0.4, 1.7)0.0010.0 (–0.2, 0.2)0.84
    Model 40.8 (0.2, 1.4)0.010.0 (–0.2, 0.2)0.67
    Model 50.8 (0.2, 1.4)0.01–0.1 (–0.2, 0.2)0.63
    Pulse pressure
    Model 10.1 (–0.3, 0.4)0.73–0.1 (–0.2, 0.1)0.59
    Model 20.2 (–0.2, 0.5)0.420.3 (0.1, 0.5)< 0.001
    Model 31.2 (0.6, 1.9)< 0.0010.4 (0.2, 0.6)< 0.001
    Model 41.1 (0.4, 1.7)< 0.0010.4 (0.2, 0.6)< 0.001
    Model 51.0 (0.4, 1.7)0.0010.4 (0.2, 0.6)< 0.001
    Model 1: Included age and race/ethnicity. Model 2: Model 1 + household income, education, marital status, working ≥ 20 hr per week outside the home, perceived stress score, and socioeconomic status z-score. Model 3: Model 2 + Rural-Urban Continuum Codes and unpenalized thin-plate regression splines for latitude and longitude. Model 4: Model 3 + body mass index, waist-to-hip ratio, smoking status, alcohol use, history of diabetes, and history of hypercholesterolemia. Model 5: Model 4 + blood pressure medication use.

    For PM2.5, adjustment for spatial features (model 3 vs. model 2) had the largest impact on effect estimates reflecting the large-scale spatial structure in PM2.5, with an increase in the positive association with SBP, a slight decrease in the positive association with DBP, and a concomitant increase in the PP association after adjustment (Table 3). For NO2, adjustment for variables representing individual and neighborhood SES (model 2 vs. model 1) had the largest impact on effect estimates particularly for SBP, with the association changing from negative and statistically significant to positive and approaching statistical significance. The importance of adjusting for these variables reflects the within-city nature of NO2 variability. After full adjustment, associations with NO2 and SBP are positive and DBP are negative, leading to a significant positive association with total PP. In general, all other added potentially confounding variables showed little impact on effect estimates.

    Interactions. For our finding of an association between PM2.5 and SBP, there was no significant evidence of interaction with BMI, race/ethnicity, age, smoking, diabetes, or anti-hypertensive medication use (see Supplemental Material, Figure S3).

    Sensitivity analyses. The results of varying the number of df used for spatial adjustment are shown in Supplemental Material, Figures S4 and S5. For PM2.5 the estimated associations with BP were fairly stable with ≥ 8 df. Varying the df had little impact on the associations of BP with NO2. Using natural logarithmic transformations of the exposure and outcome variables produced no appreciable changes in the overall findings of the analysis (data not shown). When the analysis was restricted to participants with residential stability (n = 26,217), PM2.5 effect estimates for SBP and PP were somewhat stronger; a 10-μg/m3 increase in PM2.5 was associated with a 2.1-mmHg higher SBP (95% CI: 1.0, 3.2; p < 0.001) and a 1.6-mmHg higher PP (95% CI: 0.7, 2.4; p < 0.001), and no substantive changes in other effect estimates (data not shown). Restricting the analysis to participants with three valid BP measurements at the examination (n = 41,263) also produced no change in estimates (data not shown).

    The results of sensitivity analyses using penalized TPRS to assess nonlinearity of associations between BP and the exposures of interest were generally consistent with linearity, with some evidence of nonlinearity at the extremes of the exposure distributions (data not shown).

    Neither a 10-μg/m3 increase in PM2.5 nor a 10-ppb increase in NO2 exposure was associated with increased odds of hypertension in model 5 (OR: 0.95, 95% CI: 0.38, 2.36, p = 0.92; OR: 1.02, 95% CI: 0.75, 1.38, p = 0.91, respectively).

    Though not observed for SBP, PP, or MAP, we saw a quadratic association between DBP and age. Using a quadratic rather than linear adjustment for age in the DBP models yielded null results between DBP and both exposures (data not shown). Age range did not vary across quartiles of exposure (data not shown).

    Co-pollutant analysis. Results from the co-pollutant analysis are shown in Supplement Material, Table S1. In the models that included both NO2 and PM2.5, the positive association between PM2.5 and DBP became stronger and statistically significant whereas the association with PP became essentially null and insignificant. Specifically in fully adjusted models (model 5), a 10-μg/m3 increase in PM2.5 was associated with a 1.2-mmHg higher DBP (95% CI: 0.5, 1.9; p = 0.001) and a 0.4-mmHg higher PP (95% CI: –0.4, 1.2; p = 0.3). The negative association between NO2 and DBP became stronger and remained statistically significant in the co-pollutant analysis, whereas the association between NO2 and MAP became stronger and statistically significant. For NO2, a 10-ppb increase in NO2 was associated with a 0.4-mmHg lower DBP (95% CI: –0.6, –0.2; p < 0.001) and a 0.3-mmHg lower MAP (95% CI: –0.5, –0.1; p = 0.02). No other associations were meaningfully changed from the primary single-pollutant models.


    This is the first large national cohort studied with individual BP measurements and the use of advanced modeling methods to assess fine-scale intraurban gradients in major criteria air pollutants, PM2.5 and NO2. Prior studies have either used coarser-scale exposure assessment (e.g., nearest regulatory monitor) or administrative records (e.g., records of hypertension diagnoses) for outcome assessment. With exposures in the range currently experienced in the United States, these findings are interesting and important.

    Our study demonstrates an association between increases in long-term residential exposure to PM2.5 and NO2 and higher measures of blood pressure (SBP, PP, and MAP for PM2.5 and PP for NO2). These relationships were robust to adjustment for multiple potential confounders, including SES and spatial characteristics, and apparently without threshold. The study also found an inverse relationship between NO2 and DBP in the fully adjusted model (model 5). We saw little evidence of effect modification by age, race/ethnicity, smoking, diabetes, anti-hypertensive medication use, or BMI (see Supplemental Material, Figure S3). Evidence of a long-term impact of air pollution on BP in our study population provides support to the hypothesis that air pollution induces autonomic dysfunction that may ultimately lead to vascular remodeling, increased BP, and atherosclerosis (Brook et al. 2010).

    Although these associations are modest at the individual level, the potential public health consequences of population-level changes in BP of this magnitude are substantial (Whelton et al. 2002). The effect sizes estimated in this study are the same order of magnitude as other traditionally recommended behavioral health interventions (He and MacGregor 2004). Because air pollution exposure is experienced at a population level, even a small pro-hypertensive response to long-term air pollution exposures could contribute significantly to CVD.

    In this analysis, neither PM2.5 nor NO2 exposure was associated with increased odds of hypertension, consistent with findings elsewhere (Chen et al. 2014; Foraster et al. 2014; Fuks et al. 2011); this null finding may be attributable to misclassification of hypertension cases (many cases are unrecognized) or regional differences in diagnosis and treatment.

    Few studies have examined the relationship between long-term average exposure to both PM2.5 and NO2 and BP, and none have done so over a large, spatially dispersed population such as this one. Furthermore, the few studies that have examined PP and/or MAP as outcomes focused on short-term air pollution exposure (Auchincloss et al. 2008; Chen et al. 2012; Dvonch et al. 2009; Zanobetti et al. 2004). Long-term average PM2.5 was shown to be associated with increased arterial BP in a population-based cohort study (n = 4,291) in a single metropolitan area in western Germany (Fuks et al. 2011). In Taiwan, a study with large air pollution exposure contrasts (n = 1,023) and no ability to account for neighborhood-level confounding showed strong positive associations between BP and both annual average PM2.5 and NO2 (Chuang et al. 2011). A study in an Ontario cohort found an association between PM2.5 estimated using satellite-based methods and the incidence of a hypertension diagnosis in electronic medical records (Chen et al. 2014).

    In contrast, a Danish population-based cohort study (n = 57,053) found a small reduction in SBP with long-term average NOx exposure (Sørensen et al. 2012). A study of Chinese adults (n = 24,845) found no relationship between nearest monitor NO2 and BP, but did find small increases in SBP and DBP in men associated with changes in PM10, SO2, and O3 (Dong et al. 2013). The inverse relationship between NO2 and DBP found in this study has not been reported by others (Chuang et al. 2011; Dong et al. 2013; Foraster et al. 2014), but it is possible that the inverse results may have been related to residual confounding.

    Alternatively, differences in exposure metrics (NOx vs. NO2) or other modeling methods may have contributed to differences in findings among studies. In a study of 853 elderly men in the Veterans Administration Normative Aging Study (Schwartz et al. 2012), positive associations between traffic particles and BP were observed.

    Primary strengths of this study include its large size, high-quality measurements of BP, detailed characterization of potential confounders including individual and neighborhood-level SES and spatial features, its large geographic extent, and the use of estimates of exposure to both PM2.5 and NO2.

    The cross-sectional nature of this study is its primary limitation. The cohort consists only of women and, thus, results might not be generalizable to men. Given that the cohort is composed entirely of sisters of women with breast cancer, it might also not be representative of the general U.S. female population. The prevalence of hypertension in the study population (31%) is similar to that of U.S. women (31.7%; 95% CI: 29.9%, 33.5%) according to the 2005–2008 National Health and Nutrition Examination Survey (Centers for Disease Control and Prevention 2011). Mean SBP was slightly lower and DBP was slightly higher in the study population (115 mmHg and 72 mmHg, respectively) compared with women in the general U.S. population (121 mmHg and 70 mmHg, respectively) (Wright et al. 2011).

    PM2.5 and NO2 exposures were modeled for the year 2006, whereas BP was measured between 2003 and 2009. The air pollution measures linked to residence at time of study enrollment were chosen as generally representative of long-term air pollution exposure. When our analysis was restricted to participants with residential stability, effect estimates appeared somewhat larger, suggesting that bias in these reported associations resulting from this exposure measurement error may underestimate the true associations.

    The results may also have been affected by exposure misclassification. This study evaluated long-term residential air pollution exposure, and did not account for occupational, personal, or indoor air pollution exposure. There may be residual confounding by short-term exposure to air pollution that this study was unable to account for, which was associated with higher SBP and DBP in a study of young adults in Taiwan (Lin et al. 2009). Additionally, the analysis assessed the effects of a 10-μg/m3 change in PM2.5 (IQR, 3.58 μg/m3; 10th–90th percentile, 7.38–13.38 μg/m3) and a 10-ppb change in NO2 (IQR, 6.21 ppb; 10th–90th percentile, 4.11–16.41 ppb) which may be extrapolating beyond the data in some regions or comparing extremes of the exposure distributions. A moderate amount of correlation between PM2.5 and NO2 was observed (R = 0.37), suggesting that one exposure is not acting as a surrogate for the other, which is consistent with other studies that have reported differences in associations with BP based on multi-pollutant models compared with single-pollutant models (Chuang et al. 2011; Coogan et al. 2012).

    Despite the detailed characterization of potential confounders, most were self-reported, including medication lists used to determine anti-hypertensive medication use. Similarly, physical activity and diet were not included, which could affect validity of the results via residual confounding; it is possible that the spatial adjustments may capture some of the anticipated variation in physical activity and diet. Although anti-hypertensive treatment lowers blood pressure, there was not an ideal way to account for medication use in our analysis; it does not appear to behave as a confounder in this analysis (Foraster et al. 2014).

    BP ascertainment on a single day does not allow a precise measurement of the individual’s true BP levels. Whenever possible, BP was measured in the morning, but hour of measurement was not included in the analysis. Although seasonal trends in BP could contribute to nondifferential misclassification, no discernible patterns were observed when reviewing exam month by geographic region.

    Potential residual confounding by traffic noise is a possibility (Dratva et al. 2012; Sørensen et al. 2011). However, confounding by noise in this study might be limited given the wide area studied and the large sample size, as demonstrated elsewhere (Tétreault et al. 2013).


    Our findings suggest that chronic PM2.5 exposure may lead to increases in both SBP and PP, and that chronic NO2 exposure may increase PP. These findings are consistent with our hypothesis that air pollution leads to CVD through mechanisms involving increased BP, potentially via the long-term vascular remodeling that accompanies chronic autonomic dysfunction or inflammation and oxidative stress.

    Supplemental Material

    (1.9 MB) PDF Click here for additional data file.

    We thank the many participants and study staff comprising the Sister Study who contributed to making this study possible. We also thank A. Gassett and E. Spalt (University of Washington) for providing assistance with the analysis and figures, and also the Multi-Ethnic Study of Atherosclerosis (MESA) Air and University of Washington Clean Air Center, whose infrastructure enabled this work.

    This research was supported by the Intramural Research Program of the National Institute of Environmental Health Sciences (NIEHS), National Institutes of Health grant Z01 ES044005; by NIEHS grants K24ES013195, P50ES015915, P30ES07033, K23ES019575, and T32ES015459; and by the U.S. Environmental Protection Agency (EPA) grants R831697 and R834796.

    This manuscript has not been subjected to the U.S. EPA’s required peer and policy review and therefore does not necessarily reflect the views of the agency and no official endorsement should be inferred.

    The authors declare they have no actual or potential competing financial interests.


    • Auchincloss AH, Diez Roux AV, Dvonch JT, Brown PL, Barr RG, Daviglus MLet al.. 2008. Associations between recent exposure to ambient fine particulate matter and blood pressure in the Multi-Ethnic Study of Atherosclerosis (MESA).Environ Health Perspect 116:486-491; doi:10.1289/ehp.1089918414631. LinkGoogle Scholar
    • Baccarelli A, Barretta F, Dou C, Zhang X, McCracken JP, Díaz Aet al.. 2011. Effects of particulate air pollution on blood pressure in a highly exposed population in Beijing, China: a repeated-measure study.Environ Health 10:108; doi:10.1186/1476-069X-10-10822188661. Crossref, MedlineGoogle Scholar
    • Baumgartner J, Schauer JJ, Ezzati M, Lu L, Cheng C, Patz JAet al.. 2011. Indoor air pollution and blood pressure in adult women living in rural China.Environ Health Perspect 119:1390-1395; doi:10.1289/ehp.100337121724522. LinkGoogle Scholar
    • Brook RD, Bard RL, Burnett RT, Shin HH, Vette A, Croghan Cet al.. 2011. Differences in blood pressure and vascular responses associated with ambient fine particulate matter exposures measured at the personal versus community level.Occup Environ Med 68:224-23020935292. Crossref, MedlineGoogle Scholar
    • Brook RD, Rajagopalan S, Pope CA, Brook JR, Bhatnagar A, Diez-Roux AVet al.. 2010. Particulate matter air pollution and cardiovascular disease: an update to the scientific statement from the American Heart Association.Circulation 121:2331-237820458016. Crossref, MedlineGoogle Scholar
    • Centers for Disease Control and Prevention.2011. Vital signs: prevalence, treatment, and control of hypertension—United States, 1999–2002 and 2005–2008.MMWR Morb Mortal Wkly Rep 60:103-10821293325. MedlineGoogle Scholar
    • Chen H, Burnett RT, Kwong JC, Villeneuve PJ, Goldberg MS, Brook RDet al.. 2014. Spatial associations between ambient fine particulate matter and incident hypertension.Circulation 129:562-56924190962. Crossref, MedlineGoogle Scholar
    • Chen SY, Su TC, Lin YL, Chan CC. 2012. Short-term effects of air pollution on pulse pressure among nonsmoking adults.Epidemiology 23:341-34822249240. Crossref, MedlineGoogle Scholar
    • Chuang KJ, Yan YH, Chiu SY, Cheng TJ. 2011. Long-term air pollution exposure and risk factors for cardiovascular diseases among the elderly in Taiwan.Occup Environ Med 68:64-6820833756. Crossref, MedlineGoogle Scholar
    • Cohen S, Kamarck T, Mermelstein R. 1983. A global measure of perceived stress.J Health Soc Behav 24:385-3966668417. Crossref, MedlineGoogle Scholar
    • Coogan PF, White LF, Jerrett M, Brook RD, Su JG, Seto Eet al.. 2012. Air pollution and incidence of hypertension and diabetes mellitus in black women living in Los Angeles.Circulation 125:767-77222219348. Crossref, MedlineGoogle Scholar
    • Cosselman KE, Krishnan RM, Oron AP, Jansen K, Peretz A, Sullivan JHet al.. 2012. Blood pressure response to controlled diesel exhaust exposure in human subjects.Hypertension 59:943-94822431582. Crossref, MedlineGoogle Scholar
    • Dart AM, Kingwell BA. 2001. Pulse pressure—a review of mechanisms and clinical relevance.J Am Coll Cardiol 37:975-98411263624. Crossref, MedlineGoogle Scholar
    • Diez Roux AV, Merkin SS, Arnett D, Chambless L, Massing M, Nieto FJet al.. 2001. Neighborhood of residence and incidence of coronary heart disease.N Engl J Med 345:99-10611450679. Crossref, MedlineGoogle Scholar
    • Dong GH, Qian Z, Xaverius PK, Trevathan E, Maalouf S, Parker Jet al.. 2013. Association between long-term air pollution and increased blood pressure and hypertension in China.Hypertension 61:578-58423357184. Crossref, MedlineGoogle Scholar
    • Dratva J, Phuleria HC, Foraster M, Gaspoz JM, Keidel D, Künzli Net al.. 2012. Transportation noise and blood pressure in a population-based sample of adults.Environ Health Perspect 120:50-55; doi:10.1289/ehp.110344821885382. LinkGoogle Scholar
    • Dvonch JT, Kannan S, Schulz AL, Keeler GJ, Mentz G, House Jet al.. 2009. Acute effects of ambient particulate matter on blood pressure: differential effects across urban communities.Hypertension 53:853-85919273743. Crossref, MedlineGoogle Scholar
    • Foraster M, Basagaña X, Aguilera I, Rivera M, Agis D, Bouso Let al.. 2014. Association of long-term exposure to traffic-related air pollution with blood pressure and hypertension in an adult population-based cohort in Spain (the REGICOR study).Environ Health Perspect 122:404-411; doi:10.1289/ehp.130649724531056. LinkGoogle Scholar
    • Fuks K, Moebus S, Hertel S, Viehmann A, Nonnemacher M, Dragano Net al.. 2011. Long-term urban particulate air pollution, traffic noise, and arterial blood pressure.Environ Health Perspect 119:1706-1711; doi:10.1289/ehp.110356421827977. LinkGoogle Scholar
    • He FJ, MacGregor GA. 2004. Effect of longer-term modest salt reduction on blood pressure.Cochrane Database Syst Rev 3:CD004937; doi:10.1002/14651858.CD00493715266549. Crossref, MedlineGoogle Scholar
    • Hoffmann B, Luttmann-Gibson H, Cohen A, Zanobetti A, de Souza C, Foley Cet al.. 2012. Opposing effects of particle pollution, ozone, and ambient temperature on arterial blood pressure.Environ Health Perspect 120:241-246; doi:10.1289/ehp.110364722020729. LinkGoogle Scholar
    • Kelley KE, Kelley TP, Kaufman DW, Mitchell AA. 2003. The Slone Drug Dictionary: a research driven pharmacoepidemiology tool.Pharmacoepidemiol Drug Saf 12(suppl 1):S168-169. Google Scholar
    • Langrish JP, Li X, Wang S, Lee MM, Barnes GD, Miller MRet al.. 2012. Reducing personal exposure to particulate air pollution improves cardiovascular health in patients with coronary heart disease.Environ Health Perspect 120:367-372; doi:10.1289/ehp.110389822389220. LinkGoogle Scholar
    • Lin LY, Lin CY, Lin YC, Chuang KJ. 2009. The effects of indoor particles on blood pressure and heart rate among young adults in Taipei, Taiwan.Indoor Air 19:482-48819682103. Crossref, MedlineGoogle Scholar
    • Novotny EV, Bechle MJ, Millet DB, Marshall JD. 2011. National satellite-based land-use regression: NO2 in the United States.Environ Sci Technol 45:4407-441421520942. Crossref, MedlineGoogle Scholar
    • R Core Team.2013. R: A Language and Environment for Statistical Computing. Vienna, Austria:R Foundation for Statistical Computing.Available: [accessed 5 December 2013]. Google Scholar
    • Sampson PD, Richards M, Szpiro AA, Bergen S, Sheppard L, Larson TVet al.. 2013. A regionalized national universal kriging model using Partial Lease Squares regression for estimating annual PM2.5 concentrations in epidemiology.Atmos Environ (1994) 75:383-39224015108. Crossref, MedlineGoogle Scholar
    • Schwartz J, Alexeeff SE, Mordukhovich I, Gryparis A, Vokonas P, Suh Het al.. 2012. Association between long-term exposure to traffic particles and blood pressure in the Veterans Administration Normative Aging Study.Occup Environ Med 69:422-42722383587. Crossref, MedlineGoogle Scholar
    • Sesso HD, Stampfer MJ, Rosner B, Hennekens CH, Gaziano JM, Manson JEet al.. 2000. Systolic and diastolic blood pressure, pulse pressure, and mean arterial pressure as predictors of cardiovascular disease risk in men.Hypertension 36:801-80711082146. Crossref, MedlineGoogle Scholar
    • Sørensen M, Hoffmann B, Hvidberg M, Ketzel M, Jensen SS, Andersen ZJet al.. 2012. Long-term exposure to traffic-related air pollution associated with blood pressure and self-reported hypertension in a Danish cohort.Environ Health Perspect 120:418-424; doi:10.1289/ehp.110363122214647. LinkGoogle Scholar
    • Sørensen M, Hvidberg M, Hoffmann B, Andersen ZJ, Nordsborg RB, Lillelund KGet al.. 2011. Exposure to road traffic and railway noise and associations with blood pressure and self-reported hypertension: a cohort study.Environ Health 10:92; doi:10.1186/1476-069X-10-9222034939. Crossref, MedlineGoogle Scholar
    • Tétreault LF, Perron S, Smargiassi A. 2013. Cardiovascular health, traffic-related air pollution and noise: are associations mutually confounded? A systematic review.Int J Public Health 58:649-66623887610. Crossref, MedlineGoogle Scholar
    • U.S. Department of Agriculture.2013. Rural-Urban Continuum Codes. Economic Research Service.Available: [accessed 5 February 2015]. Google Scholar
    • Van Hee VC, Adar SD, Szpiro AA, Barr RG, Bluemke DA, Diez Roux AVet al.. 2009. Exposure to traffic and left ventricular mass and function: the Multi-Ethnic Study of Atherosclerosis.Am J Respir Crit Care Med 179:827-83419164703. Crossref, MedlineGoogle Scholar
    • Weinberg CR, Shore DL, Umbach DM, Sandler DP. 2007. Using risk-based sampling to enrich cohorts for endpoints, genes, and exposures.Am J Epidemiol 166:447-45517556763. Crossref, MedlineGoogle Scholar
    • Whelton PK, He J, Appel LJ, Cutler JA, Havas S, Kotchen TAet al.. 2002. Primary prevention of hypertension: clinical and public health advisory from The National High Blood Pressure Education Program.JAMA 288:1882-188812377087. Crossref, MedlineGoogle Scholar
    • Wood SN. 2003. Thin plate regression splines.J R Stat Soc Series B Stat Methodol 65:95-114. CrossrefGoogle Scholar
    • Wright JD, Hughes JP, Ostchega Y, Yoon SS, Nwankwo T. 2011. Mean systolic and diastolic blood pressure in adults aged 18 and over in the United States, 2001–2008. Hyattsville, MD:National Center for Health Statistics.Natl Health Stat Report 35Available: [accessed 24 March 2013]. Google Scholar
    • Wu S, Deng F, Huang J, Wang H, Shima M, Wang Xet al.. 2013. Blood pressure changes and chemical constituents of particulate air pollution: results from the Healthy Volunteer National Relocation (HVNR) study.Environ Health Perspect 121:66-72; doi:10.1289/ehp.110481223086577. LinkGoogle Scholar
    • Zanobetti A, Canner MJ, Stone PH, Schwartz J, Sher D, Eagan-Bengston Eet al.. 2004. Ambient pollution and blood pressure in cardiac rehabilitation patients.Circulation 110:2184-218915466639. Crossref, MedlineGoogle Scholar