Effects of short-term particulate air pollution and nitrogen dioxide on blood pressure in older women: Longitudinal data from the Women’s Health Initiative


 BackgroundShort-term variations in particulate matter (PM) and traffic-related air pollutants (e.g., nitrogen dioxide, NO2) have been associated with daily mortality and cardiovascular health outcomes in previous studies. We aimed to evaluate whether short-term changes in PM in three size fractions (PM2.5 , PM2.5-10 , and PM10) and NO2 were associated with systolic and diastolic arterial blood pressure (SBP and DBP, respectively) in the Women’s Health Initiative Observational Study (OS) and Clinical Trials (CT). MethodsWe used linear mixed-effect models to estimate the association between short-term air pollution concentrations and repeated measures of arterial blood pressure. ResultsWe found statistically significant positive associations between short-term measures (lag days 3-5) of PM2.5 as well as NO2 for both SBP and DBP in fully adjusted models when not controlling for calendar time. Also, in only the CT, PM10 and PM2.5-10 were associated with DBP but not SBP. In fully adjusted models controlling for calendar time, associations with PM2.5-10 and NO2 remained statistically significant for DBP (except for PM2.5-10 in the OS). Specifically, in the CT group, each IQR increase in lag 3-5 NO2 exposure (9.88 ppb) was associated with a 0.13 mm Hg increase in DBP. Also, each IQR increase in lag 3-5 PM2.5-10 exposure (8.46 µg m -3) was associated with a 0.05 mm Hg increase in DBP. Effect modification was found for body mass index (BMI), socioeconomic position (SEP), diabetes, dietary sodium intake, combined fruit and vegetable consumption, and long-term PM2.5 for PM2.5 , PM10 , and NO2 . Shorter lag periods (lag 0 through lag 2) typically exhibited lesser and, especially for SBP, sometimes negative associations. In two-pollutant models of exposures lagged over 3-5 days, NO2 associations with DBP were stronger (0.20 mm Hg per IQR), but those for PM2.5-10 were attenuated to null, as compared to single-pollutant models. ConclusionsOur findings are consistent with short-term (lag days 3-5) PM2.5-10 and NO2 levels as risk factors for acute cardiovascular outcomes and cardiovascular disease, though two-pollutant model results suggest NO2 is more likely responsible for the observed effects.

diastolic arterial blood pressure (SBP and DBP, respectively) in the Women's Health Initiative 23

Observational Study (OS) and Clinical Trials (CT). 24
Methods: We used linear mixed-effect models to estimate the association between short-term 25 air pollution concentrations and repeated measures of arterial blood pressure. 26 Results: We found statistically significant positive associations between short-term measures 27 (lag days 3-5) of PM 2.5 as well as NO 2 for both SBP and DBP in fully adjusted models when not 28 controlling for calendar time. Also, in only the CT, PM 10 and PM 2.5-10 were associated with DBP 29 but not SBP. In fully adjusted models controlling for calendar time, associations with PM 2.5-10 30 and NO 2 remained statistically significant for DBP (except for PM 2.5-10 in the OS). Specifically, in 31 the CT group, each IQR increase in lag 3-5 NO 2 exposure (9.88 ppb) was associated with a 32 0.13 mm Hg increase in DBP. Also, each IQR increase in lag 3-5 PM 2.5-10 exposure (8.46 µg m -33 Introduction 45 Short-term variations (from days to weeks) in particulate matter (PM) and traffic-related air 46 pollutants such as nitrogen dioxide (NO 2 ) have been associated with daily mortality and 47 cardiovascular health outcomes in previous studies [1][2][3]. PM<2.5 µm (PM 2.5 )-mediated arterial 48 blood pressure (BP) elevation may potentially be an important part of the causal mechanism 49 leading to acute cardiovascular outcomes [4,5]. One recent study from Women's Health 50 Initiative (WHI) suggests that long-term exposure to PM 2.5 and PM 10 may be essential modifiable 51 risk factors for hypertension in post-menopausal women [6]. 52 Findings from earlier studies of the effects of repeated short-term air pollutant exposures on BP 53 have been varied, though generally suggestive of positive associations [7-10]. One of these 54 studies conducted in California found that PM 2.5 (specifically the PM 2.5 component primary 55 organic carbon) was more strongly associated with BP measures than were gaseous pollutants 56 [7]. Additionally, traffic-related exposure measures have been identified as important modifiers 57 of the effect of PM 2.5 on arterial BP in a diverse population from the MESA study [11]. One 58 randomized controlled trial in humans showed that short-term exposure to traffic-related air 59 pollution (i.e., diesel exhaust) was significantly associated with increased systolic BP (SBP) but 60 not diastolic BP (DBP) [12]. Findings from many epidemiological studies of short-term air 61 pollution effects on arterial BP have been analyzed using meta-analysis [13]; this meta-analysis 62 showed overall significant positive though not robust short-term associations between several 63 air pollutants (PM 2.5 , PM<10 µm (PM 10 ), NO 2 , and SO 2 ) and increases in SBP and DBP, as well 64 as hypertension, an established risk factor for cardiovascular diseases. Blood pressure and air 65 pollutant levels (in many areas of the US) have both decreased over the past several decades 66 [14,15]. Adar et al. found significant associations for both SBP and DBP with PM 2.5 as well as 67 NO 2 for exposure averaging periods of seven days and longer in adjusted models that did not 68 control for calendar time. However, when calendar time was included, those associations were 69 attenuated to null [16]. 70 validated using 10-set cross-validation and had high predictive accuracy with a cross-validation 119 R 2 of 0.77. 120 Covariates 121 At baseline and during each annual follow-up visit, questionnaires were used to collect 122 demographic data. Covariates included in this analysis were age at visit, self-reported 123 First we evaluated unadjusted models between air pollutant concentrations (PM 2.5 , PM 2.5-10 , 150 PM 10 , and NO 2 ) and arterial blood pressure (SBP, DBP) in both OS and CT. For each pollutant, 151 the models included only a single lag period (lag days 0 to 6, in separate models) or a single 152 moving average. Next we evaluated the effect of lag 3-5 exposures in the models, then added 153 confounders, in sets, to form basic and adjusted models as described below. In basic models, 154 confounders included age, race/ethnicity, arm group (only for CT), census region, day of the 155 week, season, and a random slope for age. Next we added additional potential confounders to 156 the basic model; if the percent change in the air pollutant effect estimate was >10%, then the 157 variable was considered a confounder. If the percent change in the air pollutant effect estimate 158 was <10%, we then evaluated whether the Akaike information criterion (AIC) was lowered upon 159 inclusion of the variable and if so, the variable was considered a confounder. In adjusted 160 models, we further controlled for BMI, SEP, pack-years of smoking, and diabetes. Next, to 161 control for calendar time, we added the number of years since baseline exam to the fully 162 adjusted models. We evaluated fully-adjusted models with and without adjustment for calendar 163 time 1) to isolate the effect of controlling for calendar time and 2) to facilitate comparisons with 164 previous analyses which may or may not have controlled for calendar time. To evaluate effect 165 modification, we fit fully adjusted models with interaction terms for the lag 3-5 air pollutant and 166 each of BMI, SEP, diabetes, smoking, sodium intake, fruit and vegetable consumption, US 167 Census Region ("Northeast", "Midwest", "South", "West"), and long-term PM 2.5 concentrations. 168 When statistically significant effect modification was found (p-value for interaction term < 0.05), 169 stratified analyses were conducted. For pollutants which were found to have statistically 170 significant main effects in fully adjusted models controlling for calendar time, we further 171 examined these pollutants in two-pollutant models. 172

173
During the study period (1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005), a total of 136,008 participants (69,490 in OS and 66,518 in 174 CT) were included in the analysis. Approximately 1 in 3 participants came from the western 175 region of the US and most of participants were white (>80%). On average, participants were 176 overweight (i.e., BMI >= 25 kg m -2 ), with higher mean BMI among participants in CT than in the 177 OS, but this difference was not statistically significant. Characteristics of the study participants 178 and air pollutant concentrations are presented in Table 1. 179 Unadjusted models: 180 Regression coefficients (listed in the columns headed with "β" in the tables) represent the 181 change in BP per unit change in air pollution concentration (after transformation of the air 182 pollutant concentration to the IQR scale). Results from unadjusted models using lag 3-5 183 exposures were varied and not consistent in direction. For SBP, unadjusted models using lag 184 3-5 exposures showed that short-term exposure to PM 2.5-10 and PM 10 were significantly 185 negatively associated in the OS and CT (except for PM 2.5-10 in OS), while NO 2 was significantly 186 positively associated in both the OS and CT. For DBP, PM 2.5 , PM 2.5-10 , PM 10 and NO 2 were 187 significantly positively associated in both the OS and CT (except for PM 10 in the OS group and a 188 statistically significant negative association for PM 2.5-10 in the OS group) ( Table 2). 189

Fully adjusted models: 190
Results from multivariable models with each potential confounder added showed that BMI, SEP,191 pack-years of smoking, and diabetes were important confounders (in addition to those in the 192 basic models) among the three PM fractions as well as NO 2 and for both SBP and DBP, 193 especially for PM 2.5 and NO 2 (Supplemental Material Table S1). 194 Results from basic and fully adjusted models without controlling for calendar time using single 195 lag days (0-6) are presented in Supplemental Material Tables S2A and Table S2B. Shorter 196 single lag periods (lag 0 through lag 2) showed significantly negative or null associations for PM 197 (PM 2.5 , PM 2.5-10 and PM 10 ) with both SBP and DBP (except for PM 2.5 -DBP in OS group) as well 198 as NO 2 with SBP. For longer lag periods, effects were consistent and largest (i.e., most 199 positive) for single lag periods 3, 4, and 5. 200 Fully adjusted models not controlling for calendar time and using lag 3-5 exposures included the 201 following confounders: age, race/ethnicity, treatment arm (only for the CT group), region, day of 202 the week, season, BMI, SEP, pack-years of smoking, and diabetes. In these fully adjusted 203 models not controlling for calendar time, PM 2.5 and NO 2 were significantly associated with both 204 SBP as well as DBP (regression coefficients from the LME models without effect modification 205 are presented in Table 2). Effect sizes for NO 2 were largest among the pollutants considered 206 for SBP and DBP, except for PM 2.5 -SBP in the OS group. In the CT group, each IQR increase 207 in lag 3-5 PM 2.5 exposure (an increment of 7.66 µg m -3 ) was associated with a 0.07 mm Hg 208 increase in SBP and a 0.06 mm Hg increase in DBP. By comparison, each IQR increase in lag 209 3-5 NO 2 exposure (an increment of 9.88 ppb) was associated with a 0.45 mm Hg increase in 210 SBP and a 0.38 mm Hg increase in DBP. PM 10 and PM 2.5-10 were significantly associated with 211 DBP only in the CT group, again in fully adjusted models not controlling for calendar time. 212 In fully adjusted models controlling for calendar time, the association between NO 2 and DBP in 213 both the OS and CT groups remained statistically significant, as did the association between 214 PM 2.5-10 and DBP in only the CT group. Specifically, in the CT group, each IQR increase in lag 215 3-5 NO 2 exposure (9.88 ppb) was associated with a 0.13 mm Hg increase in DBP (Table 2). 216 Also, each IQR increase in lag 3-5 PM 2.5-10 exposure (8.46 µg m -3 ) was associated with a 0.05 217 mm Hg increase in DBP (Table 2). In the OS group, the effect size for lag 3-5 NO 2 exposure 218 was larger and was 0.32 mm Hg. 219 Effect modification: 220 We evaluated effect modification in either 1) fully adjusted models not controlling for calendar 221 time or 2) in fully adjusted models controlling for calendar time, depending on whether main 222 effects were statistically significant in Table 2. 223 In fully adjusted models not controlling for calendar time, we found effect modification by BMI, 224 SEP, diabetes, and long-term PM 2.5 levels for PM 2.5 , PM 10 , and NO 2 (p-value for interaction 225 terms <0.05; Table 3). 226 Because PM 2.5-10 -DBP and NO 2 -DBP associations were statistically significant in the fully 227 adjusted models controlling for calendar time, effect modification was evaluated in those 228 models. For NO2-DBP, we found effect modification by BMI, SEP, diabetes, dietary sodium 229 intake, and combined fruit and vegetable consumption (Table 4). No effect modification was 230 found between PM 2.5-10 and DBP in the CT group. 231 Effect modification by BMI: 232 BMI modified the effects of lag 3-5 PM 2.5 exposure as well as lag 3-5 NO 2 exposure for SBP in 233 only the CT group, in models not controlling for calendar time. Also, BMI modified the effects of 234 lag 3-5 NO 2 exposure for DBP in the CT group, again in models controlling for calendar time. 235 Stratified results showed both PM 2.5 -SBP associations and NO 2 -SBP associations were stronger 236 and more positive among participants with higher BMI; the PM 2.5 -SBP association was 237 statistically non-significant among those in the first (lowest) tertile of BMI (Table 3 & Table 4). 238 Effect modification by SEP: 239 SEP also significantly modified the effect of lag 3-5 PM 2.5 and NO 2 in both the CT and OS for 240 both SBP and DBP, in models not controlling for calendar time (Table 3). For both PM 2.5 -SBP 241 and PM 2.5 -DBP, stratified associations were lower among those with higher SEP. For NO 2 -SBP, 242 stratified associations were again lower among those with higher SEP. 243 For NO 2 -DBP associations in models controlling for calendar time (Table 4), stratified 244 associations also were lower among those with higher SEP in both the OS and CT groups. 245 Effect modification by other factors: 246 For PM 2.5 , PM 10 , and NO 2 lag 3-5 exposures, associations with SBP and DBP were stronger and 247 more positive among participants with diabetes compared to those without, in models not 248 controlling for calendar time (Table 3). Also, for lag 3-5 PM 2.5 exposures, stratified associations 249 with DBP were stronger and more positive for the second and third tertiles of long-term PM 2.5 250 level in the CT group. 251 For NO 2 -DBP associations in models controlling for calendar time, stratified associations were 252 lower among those with higher fruit and vegetable consumption in the OS group. Also, stratified 253 results by dietary sodium intake showed NO 2 -DBP associations were stronger and more 254 positive in the first and third tertile of sodium intake in the OS group, as compared to the second 255 tertile (Table 4). 256 Two-pollutant models controlling for calendar time: 257 In Table 2, only PM 2.5-10 and NO 2 had statistically significant main effects in fully adjusted models 258 controlling for calendar time and only for DBP, so only these two pollutants were used in two-259 pollutant models. Though PM 2.5-10 and NO 2 were significantly correlated in our analysis, 260 repeated measures correlation coefficients were low and ranged between 0.02 and 0.09 across 261 the seven lag periods lag 0 to lag 6. 262 Regression coefficients from single-pollutant and two-pollutant fully adjusted models controlling 263 for calendar time are presented in Table 5. For DBP, compared to single pollutant model 264 results, NO 2 associations were stronger and more positive in two-pollutant models. However, 265 those for PM 2.5-10 were attenuated to null. We also found effect modification by US Census 266 Region in the fully adjusted two-pollutant model controlling for calendar time (Supplemental 267 Material Table S3). 268

Discussion 269
Our findings indicate that short-term measures (lag days 3-5) of PM 2.5 as well as NO 2 are 270 consistently associated with changes in SBP and DBP in models not controlling for calendar 271 time. When evaluating PM 2.5 and NO 2 exposures averaged over lag days 3-5, patterns of 272 association were consistent and robust with respect to the direction of association. associations were found between lag day 1, 2-, and 3 PM 10 and SBP as well as lag day 2 PM 10 279 and DBP, and between lag day 2 and 3 NO 2 and SBP. In contrast, statistically significant 280 positive associations were found between lag day 1 and lag day 2 NO 2 and DBP among 281 nonsmoking adults [21]. Statistically non-significant associations were also found between lag 282 day 1 PM 2.5 with both SBP and DBP among elders with no anti-hypertensive medication use 283 [22]; similar results were found for PM 2.5 exposures immediately and 24-h after a 2 hr walk in 284 close proximity to traffic for both SBP an DBP among healthy adults [23]. In children, 285 statistically non-significant associations were found between lag day 0 (i.e., same day) 286 exposures to PM in three size fractions and SBP [24]; also statistically non-significant 287 associations were found between lag day 1 exposures to PM 2.5 as well as NO 2 and both SBP 288 and DBP in children [26]. Statistically significant negative associations were also found 289 between lag day 1-3 exposures to PM 2.5 with SBP and DBP among young adults [25]. 290 In our analysis, in models not controlling for calendar time, SEP, BMI, and diabetes were found 291 to be statistically significant effect modifiers. Two prior studies also reported effect modification 292 by BMI on the association between PM 2.5 and BP [7, 27]. We also found effect modification by 293 long-term PM 2.5 level on short-term exposure to PM 2.5 , a result consistent with earlier studies 294 which showed that the association between short-term PM 2.5 and SBP was stronger in areas 295 with higher long-term PM 2.5 levels [27,28]. 296 Our results for short-term (lag days 3-5) exposures, in models not controlling for calendar time, 297 were broadly consistent with those from previous studies. One panel study of 62 cardiac 298 rehabilitation patients showed a positive association between moving-average (over the 299 previous 5-days) PM 2.5 exposure and SBP, as well as moving-averages of the previous 4-, and 300 5-day PM 2.5 exposure levels and DBP [29]. Another panel study of 64 elderly subjects with 301 history of coronary heart diseases [7] found that multiday (3-day, 5-day, and 7-day) averaged air 302 pollution exposures were positively associated with increased SBP and DBP. In our single-303 pollutant models, lag 3-5 NO 2 had stronger effects on both SBP and DBP than did lag 3-5 PM 2.5 , 304 compared on an IQR basis. Stronger effects of NO 2 on BP than those of PM 2.5 have also been 305 shown in another study from Canada [30]. 306 In models controlling for calendar time, our results are broadly consistent with those in Adar et 307 al. [16]. However in that analysis associations attenuated to null after controlling for calendar 308 time (for exposure averaging periods of seven days, for example), whereas in the present study 309 associations for lag 3-5 PM 2.5 -10 and NO 2 remained statistically significant for DBP. Despite 310 remaining statistically significant, the effect sizes are small. However, they may still be clinically 311 relevant: Across the (somewhat skewed) distribution of exposure to NO 2 , comparing the most 312 highly-exposed individuals to the least, they may experience a change in exposure of 313 approximately three IQRs, and therefore the corresponding effect size would, using the larger 314 effect estimate from the OS, be 3*0.32=0.96 mm Hg in DBP. In these models, we found effect 315 modification by SEP, BMI, and diabetes in the NO 2 -DBP association, suggesting that 316 participants with high BMI, low SEP, and diabetes may be particularly susceptible to the effects 317 of short-term NO 2 on DBP. We also found that participants with more fruit and vegetable 318 consumption were less susceptible to the effects of NO 2 on DBP, suggesting potential dietary 319 intervention to mitigate air pollution-induced CVD risk [31]. We also noted effect modification by 320 dietary sodium intake, but the non-monotonic pattern in these associations does not support 321 meaningful causal interpretation. However, some results from our analysis differ from earlier studies. Our study showed 331 statistically significant negative or null associations between individual lag day 0-2 air pollutant 332 exposures and both SBP and DBP. These negative associations persisted even when 333 controlling for calendar time. In contrast, a panel study of 74 patients undergoing cardiac 334 rehabilitation [32] found a statistically significant positive association between 0-5 hour moving-335 average PM 2.5 exposure and SBP, with each IQR increase corresponding to an increase of 0.94 336 mm Hg (95%CI: 0.02-1.87). They also found statistically non-significant associations between 337 PM 2.5 in individual lag periods (evaluating lag days 0, 1, 2, 3, and 4 separately) and DBP, which 338 is inconsistent with our findings. In another study, Dvonch et al. [28] reported that short-term 339 exposure (i.e., lag day 2) to PM 2.5 was positively associated with SBP among all subjects, 340 suggesting a positive effect more acute than in our findings. These differences could be due to 341 differences in study population or to different exposure estimation procedures. A recent meta-342 analysis of air pollution exposure and blood pressure reported substantial heterogeneity in effect 343 estimates on blood pressure for PM 2.5 , PM 10 , and NO 2 ; also, NO 2 had a larger meta-estimated 344 association with DBP than PM 2.5 [13]. Their analysis also provided evidence of publication bias 345 for the association between NO 2 and DBP. Also, earlier studies have documented evidence of 346 spatial and temporal variability of particulate pollution with regard to sources and chemical 347 composition [33,34], and as such differences in PM composition, as discussed in Giorgini et al. 348 [5], could be another reason our findings differ from those in earlier studies. 349 This study has several strengths. One is the large sample size and recruitment from many 350 areas of the US which allowed us to perform stratified analysis with sufficient statistical power; 351 also the longitudinal study design using repeated measurements of blood pressure increased 352 statistical power to detect associations between air pollution exposures and BP measures. 353 Secondly, the estimates of air pollution exposure from daily lognormal kriging models contain 354 greater temporal (i.e., daily) and spatial (including urban-scale gradients) resolution than in 355 previous studies using conventional exposure assessment methods. Our study also has 356 several limitations. The first concerns the lack of PM2.5 monitoring before 1999. Second, we 357 were unable to control for other potential confounders such as physical activity and occupational 358 exposure. The third is exposure error; a small amount of spatial error is unavoidable when 359 performing spatial interpolation and kriging models did not include very local, micro-to 360 neighborhood-scale information (or their proxies) on air pollutant levels. Of course, where 361 monitoring was sparse, interpolation was based on distant measurements. The fourth is limited 362 generalizability. The findings from this study may not be generalizable to males, nor to younger, 363 pre-menopausal women in the U.S. 364 In conclusion, our findings are consistent with short-term (lag days 3-5) PM 2.5-10 and NO 2 levels 365 as risk factors for acute cardiovascular outcomes and cardiovascular disease, though two-366 pollutant model results suggest NO 2 is more likely responsible for the observed effects among 367 elderly women not taking anti-hypertensive medication. 368

Supplementary information 369
Additional file 1: Table S1. Effects of an IQR change in air pollutant concentration on systolic 370 blood pressure (SBP) and diastolic blood pressure (DBP) in basic models with varying levels of 371 adjustment in the WHI Observational Study (OS) and Clinical Trials (CT) components. Table  372 S2A. Associations between an IQR change in air pollutant concentration and systolic blood 373 pressure (SBP) and diastolic blood pressure (DBP) for single lag days (0-6) based on basic 374 models. Table S2B. Associations between an IQR change in air pollutant concentration and 375 systolic blood pressure (SBP) and diastolic blood pressure (DBP) for single lag days (0-6) 376 based on fully adjusted models.   **: PM is particulate matter; PM 2.5 is PM < 2.5 µm; PM 2.5-10 is 2.5 µm < PM < 10 µm; PM 10 is PM < 10 µm; NO 2 is nitrogen dioxide. Table 3. Effect modification of lag 3-5 day air pollutant exposures assessed using stratification by BMI, SEP, diabetes, and long-term PM 2.5 level based on fully adjusted models not controlling for calendar time (when interaction terms were significant in Table 2  Tertile 3 34,884 0.24 0.065 0.0002 *Fully adjusted + calendar time models included the main effect for each effect modifier (even if not identified as a confounder). **Categorized by tertile of SEP score, BMI, sodium intake, and fruit and vegetable consumption separately.     Table 3. Note: CI is confidence interval.  Table 5. Note: CI is confidence interval. group from single-pollutant and two-pollutant models. Note: CI is confidence interval.