Methods

The design of the Framingham Offspring Study and Third Generation Studies has been described in detail elsewhere. Participants who attended Examination 7 of the Offspring Study (1998 to 2001) or Examination 1 of the Third Generation Study (2002 to 2005) and also completed the brachial examination were eligible for inclusion in this analysis. All participants provided written informed consent for the Framingham Heart Study examinations, and both the Committee on Clinical Investigation at Beth Israel Deaconess Medical Center and the Institutional review board at the Boston Medical Center approved the study protocol. There were 5,719 participants for whom residential proximity to roadway and PM _2.5 data were available. We then excluded 37 addresses identified as zip-code centroid or street midpoint and 570 participants who smoked within 6 hours of the examination, leaving a total of 5,112 eligible participants, of whom 2,828 were participants in the Third Generation Study (55%). Participant primary addresses were geocoded using ArcGIS 10 (ESRI, Redlands, California).

Neighborhood-level socioeconomic characteristics, including median household income, were assigned at the census tract level from US Census 2000 data. Distance to nearest major road was determined by residential proximity to the nearest A1, A2, or A3 road (US Census Features Class) at the time of the brachial examination (Offspring Examination 7, Third Generation Examination 1). Based on previous work showing that particle levels diminish to background levels 100 to 300 m from major roads, we first examined associations using categories of distance: <50 m, 50 to <100 m, 100 to <200 m, 200 to <400 m, and 400 to <1,000 m. These categories were selected a priori to reflect the decay function of traffic pollution and noise as proximity to roadway decreases to background levels. We also tested the natural logarithm of the proximity to a major roadway and brachial function because it is associated linearly with mortality. Participants living farther than 1,000 m from a major road in rural areas beyond background were excluded from the roadway analyses (530 participants, 10%).

Our approach utilizes Moderate Resolution Imaging Spectroradiometer satellite-derived aerosol optical density (AOD) measurements to predict daily PM _2.5 concentration levels at a 10 × 10 km spatial resolution across New England. Starting in 2000, AOD was calibrated daily using ground PM _2.5 measurements from 78 monitoring stations, land use regression, and meteorological variables (temperature, wind speed, visibility, elevation, distance to major roads, percent of open space, point emissions, and area emissions). To estimate PM _2.5 daily concentrations in each grid cell, we first calibrated the AOD–PM _2.5 relation with data from grid cells including both monitor and AOD values using mixed models with random slopes for day and nested regions. Next, we estimated exposures on days when AOD measures were not available (e.g., due to cloud coverage or snow). A model was fit with a smooth function of latitude and longitude and a random intercept for each cell that takes advantage of associations between grid cell AOD values and PM _2.5 data from monitors located elsewhere and associations with available AOD values in neighboring grid cells. We averaged each daily concentration for the year 2001 in the 10 × 10 km grid cell for each home address. The year 2001 was selected because it falls between the Offspring Cycle 7 (1998 to 2001) and Third Generation Examination 1 (2002 to 2005) data collection and complete data were available. The first model calibrations resulted in high out-of-sample 10-fold cross-validated R ² (mean out-of-sample R ² = 0.83). Even with days without any AOD data, our model presented extremely high fits (R ² = 0.81). We also evaluated the spatial and temporal components of the model separately, which also resulted in very high fits (R ² of spatial component = 0.78 and temporal component = 0.84). We then used local land use terms (distance to primary highways, distance to point source emissions, population density, percent open spaces, elevation, and traffic density) to model the difference between the 10 × 10 km grid cell predictions and monitored values. We regressed the residuals for each monitor against local land use characteristics for each monitor and a smooth function of traffic density. The sum of the grid cell predictions and residuals from the land use model represents a measure of total PM _2.5 at a specific location within a 50 × 50 m grid. This further increased the overall model R ² by 1.9%.

Fasting brachial artery tracings were measured by 1 of the 3 experienced sonographers following rigorous standardized protocols and were acquired for participants in Framingham Offspring Examination 7 and Third Generation Examination 1. Baseline diameter, FMD% (percent change in diameter from baseline), baseline mean flow velocity, and mean hyperemic flow velocity were determined as the outcomes of interest. Hyperemic flow velocity was assessed using midartery pulse Doppler signal. FMD%, and reactive hyperemia as measured by hyperemic flow velocity may reflect, in part, decreased endothelial-dependent vasodilation mediated by NO in the conduit and forearm microvessels, respectively.

Investigators measured brachial artery diameter at baseline and 1 minute after reactive hyperemia induced by 5-minute forearm cuff occlusion using commercially available software (Brachial Analyzer v. 3.2.3, Medical Imaging Applications). Doppler flow was assessed at baseline and during reactive hyperemia. Mean baseline and hyperemic flow velocities were analyzed from digitized audio data with semi-automated signal averaging (Cardiovascular Engineering). Reproducibility of the approach is extremely high and has been described elsewhere. Flow measurements were available for a subset of participants because measures began after the Offspring 7th Examination cycle had started. Of the 5,112 eligible participants, there were 498 offspring participants (22%) in whom flow was not measured and 79 (3%) missing flow data in the Third Generation Cohort.

History of cardiovascular disease (coronary heart disease, intermittent leg claudication, heart failure or stroke, or transient ischemic attack) was determined by a panel of 3 investigators using previously published criteria. At Examination 7, participants were asked about antihypertension and lipid-lowering medication, and postmenopausal hormone use. In the Third Generation Examination 1, participants brought their medications, and these were coded according to their World Health Organization anatomical therapeutic chemical codes. We used the anatomical therapeutic chemical codes to classify antihypertensive, lipid-lowering drugs and postmenopausal hormone use. Prevalent diabetes was defined as having a fasting glucose ≥126 mg/dl or oral hypoglycemic or insulin use at an examination or any previous history of diabetes (excluding gestational diabetes). Systolic and diastolic seated blood pressures were calculated as the mean of 2 measurements taken by the physician administering the clinical exam.

Linear regression was used to evaluate associations between ambient pollutants (near roadway measures and PM _2.5 ) and brachial measures (baseline brachial diameter [mm], FMD (%), baseline mean flow velocity (cm/s), and hyperemic flow velocity [cm/s]). Associations were first modeled using a parsimonious approach (Model 1) adjusting for age, age ² , sex, body mass index, individual-level education (no high school diploma, high school degree, some college, technical school or associate’s degree, and bachelor’s degree or higher), date of examination, median household income from 2000 census tract (quartiles), cohort (Offspring or Third Generation), smoking status, and seasonality (sine and cosine of day of year). We next adjusted for additional covariates that have been associated with brachial measures in previous studies of this sample. Model 2 covariates added to those included in Model 1 were walk test administered before brachial examination, history of diabetes, lipid-lowering medications, antihypertensive medications, current postmenopausal hormone use, a history of cardiovascular disease, systolic blood pressure, diastolic blood pressure, heart rate, triglycerides, and total cholesterol/high-density lipoprotein (HDL) ratio. We then evaluated the linearity of exposure-response relations for the log of distance to roadway and PM _2.5 measures using restricted cubic splines with knots at the 5, 27.5, 50, 72.5, and 95 percentiles of the distribution.

We hypothesized that associations between exposure to ambient air pollution and vascular function may differ by factors related to sex, history of diabetes, current smoking status, and age >65 years (n = 776) and median household income at the census tract level below the first quartile ($48,973). We, therefore, tested for statistical interaction using cross-product terms and determined statistical significance from the p value from the Wald test of the cross-product. We also hypothesized that regular use of vasoactive medication would alter the association between long-term exposure to ambient air pollution and vascular response; therefore, we evaluated stratification by the use of antihypertensive and statin drug use.

We performed several sensitivity analyses. Two outliers were identified in the distribution of PM _2.5 exposures (≥19 μg/m ³ ), and analyses were conducted excluding these observations. We also analyzed the associations with PM _2.5 among participants who lived within 1,000 m of a major road. Previous studies have used different approaches to account for the fact that FMD is calculated as a ratio, and the denominator (baseline brachial artery diameter) may be associated with air pollution exposure. We therefore adjusted our FMD% results first by hyperemic flow velocity and then by 1/baseline diameter. To account for potential clustering of the observed brachial outcomes by neighborhood-level characteristics, analyses were also conducted using generalized estimating equations with exchangeable covariance structure to examine the associations after accounting for census tract.

For models in which we examined residential proximity to roadway in 5 categories, tests for linear trends were performed by assigning each exposure category the natural log of the median distance within each category and including the term as a continuous variable in the regression model. The p value obtained represents the log-linear component of trend (p trend). We present the results of log-linear residential proximity to major roadway analyses by contrasting participants who live 50 m from the nearest major roadway to those living 400 m away from the nearest major road to reflect the mean-adjusted difference in brachial measures between participants likely to have high levels of exposure to traffic-related pollution versus those exposed to near-background levels. This calculation was performed by scaling the beta coefficient such that ln(400 m) *β − ln (50 m) *β = ln(400/50) *β, where β denotes the beta coefficient from the linear regression model treating log of distance to roadway as a linear continuous variable. Results for models of PM _2.5 are scaled to an interquartile range (IQR) for the observed distribution in our data. All analyses were conducted in Stata v.12 (Statacorp, College Station, Texas) and R 2.13.2 (R Foundation for Statistical Computing, Vienna, Austria, MGCV, 1.7-22). Plots were created using the POSTRCSPLINE package in Stata. All statistical tests were 2 sided, and p <0.05 was considered statistically significant.

Results

Participant characteristics are listed in Table 1 . Among participants who reported their race (n = 4,790), there were 17 participants who reported a race other than white (<1%). Compared with other participants, participants in whom hyperemic flow was not measured did not differ by age, blood pressure, prevalence of diabetes mellitus, baseline diameter, or FMD%. The exposure distribution is listed in Table 2 . Analyses examining residential proximity to a major roadway as the exposure of interest excluded participants living >1,000 m from a major road (530 participants, 10%).

Neither residential proximity to a major roadway nor PM _2.5 was associated with baseline brachial diameter or mean flow velocity. Associations with categories of distance to roadway showed no consistent pattern ( Table 3 ). We observed consistent associations between PM _2.5 and both FMD% and hyperemic flow velocity. We also observed log-linear associations between residential proximity to a major roadway and FMD% ( Table 4 ).

Stay updated, free articles. Join our Telegram channel

Join