Long-Term Exposures to Urban Noise and Blood Pressure Levels and Control Among Older Adults

This analysis draws data from 2 prospective cohort studies of older adults: MESA and CHAP. MESA was designed to investigate subclinical cardiovascular disease and the risk factors for progression to clinical cardiovascular disease. To this end, MESA enrolled 6814 participants from 6 metropolitan areas between 2000 and 2002, who were between 45 and 84 years of age and free of clinical cardiovascular disease. Individuals were observed in clinical examinations at baseline and during 4 follow-up visits in 2002 to 2004, 2004 to 2005, 2005 to 2008, and 2010 to 2012.²⁰ Since we have a noise model in only Chicago, IL, for this study, we restricted our analysis to the 1164 Chicago-based MESA participants for whom we could estimate noise levels.

CHAP was initiated in 1993 in Chicago to study the risk factors for chronic conditions with an emphasis on cognition.^21,22 Investigators enrolled a total of 10 802 residents on a rolling basis who were aged 65 years or older and conducted in-home study visits triennially. To be consistent with the MESA cohort, we focused on participants observed after 2000. Study protocols from both studies were approved by the institutional review boards of each study site (Rush University Medical Center, Northwestern University, and the University of Washington), all participants provided written consent for inclusion in CHAP and MESA, and this research was approved by the institutional review board at the University of Michigan.

Systolic (SBP) and diastolic (DBP) BPs were measured in both cohorts during each participant contact. After participants rested for 5 minutes in a seated position, BP measurements were collected using a sphygmomanometer. To get a stable estimate of BP, we averaged 2 sequential BP measurements from each participant at each exam. We used the hypertension definition contemporaneous with the assessment which was SBP≥140 mm Hg or DBP≥90 mm Hg²³ or being on antihypertensive medication (self-reported), although the recent thresholds were also used in secondary analyses.²⁴

BP control was defined as a 3-level outcome: not hypertensive or controlled (measured SBP<140 and DBP<90 and on antihypertensive medication), uncontrolled hypertension (measured SBP≥140 or DBP≥90 mm Hg, regardless of medication), and apparent treatment-resistant hypertension (aTRH). aTRH was defined as being on >4 classes of antihypertensive medication or being on 3 or more classes of antihypertensive medication (including a diuretic) and remaining hypertensive.³ Data on medication adherence was not available, thus only aTRH could be ascertained.

We estimated each participant’s long-term noise exposure using a prediction model developed for Chicago based upon 2 intensive sampling campaigns conducted between 2006 and 2007.^25,26 Briefly, 5-minute grab samples of A-weighted noise levels in decibels (dBA) were taken at 136 locations across the Chicago area during nonrush hour periods to capture noise near participant residences and locations with different proximities to local noise sources (eg, highways, airports). A-weighting was used so that our noise estimates would reflect what is actually perceived by the human ear by reducing decibel values for frequencies that are imperceptible to the human ear. We constructed a land-use regression model in which geographic covariates (eg, land use, proximity to roadways, bus stops, and trains) were used to predict noise measurements. We then used this model to estimated individual-level, long-term noise exposure for all participants based on their 5-year residential history before each interview.

Our models showed strong predictive power with a leave-one-out cross-validated R² of nearly 0.7. In addition, an external validation data set collected in 2016 demonstrated strong stability in the spatial variation of noise levels across time. Samples collected from the same locations in 2016 and 2006/2007 had a correlation of 0.8, indicating that the spatial differences remained stable. Data from 2016 similarly showed stability in the key predictors of noise captured by the previous model (Table S1 in the Supplemental Material).

From participant interviews, we had self-reported data on each participant’s age, sex, race and ethnicity, income, education, smoking status, alcohol intake, physical activity, and medications. Height and weight were also measured. A subsample in CHAP and all MESA participants were also administered a food frequency questionnaire. At each participant address, we obtained additional environmental factors: air and light pollution levels. Annual average ambient concentrations of nitrogen oxides (NO_x) and fine particulate matter (PM_2.5) were estimated based on a spatiotemporal model derived from intensive monitoring data in the MESA Air Pollution Study.²⁷ Both NO_x and PM_2.5 are important as they act as surrogates of traffic-related air pollution, which may be a key source of confounding in the estimated noise-BP association. Light pollution is also associated with urbanization and cardiovascular outcomes.^28,29 Total brightness levels at participant addresses were obtained from publicly available data.³⁰ On the area level, we generated a neighborhood socioeconomic score using a principal components analysis of census block-level data, such as percentage of residents with a bachelor’s degree, median home value, and household income.³¹

First, we used multivariable linear mixed models with random person-specific intercepts to estimate associations of long-term residential noise exposure with SBP and DBP levels. For these regression analyses, and following the methodological literature,^32,33 we imputed BP levels separately by cohort to adjust for antihypertensive medication use. Conceptually this approach approximates what a participant’s BP would be if they were not on hypertensive medications, as a function of their age, sex, race, and class(es) of antihypertensive medication(s).

Next, we examined the association of noise levels with BP control by modeling the association with a 3-level categorical outcome: normal BP or controlled hypertension (comparison population), uncontrolled hypertension, and aTRH. We used a multinomial nominal logistic regression with generalized estimating equations to account for repeated measures for each participant. This generates separate estimands for each outcome-level (uncontrolled hypertension or aTRH), which represent the difference in odds of the outcome-level with an increase in noise levels, as compared to the normal/controlled hypertension group (referent level).

All models were run stratified by cohort and altogether since there was no evidence of effect modification as evidenced by an interaction term between noise and cohort in our pooled model. We adjusted for visit-specific variables (age, calendar time,³⁴ and smoking status) and time-invariant variables (sex, race, income, education, NO_x neighborhood socioeconomic score). Since we observed statistical differences in BP levels and relationships of BP with key covariates by cohort, we also included a cohort term in our combined models as well as interaction terms of cohort with age, race, and sex. All associations are scaled to a 10 dBA increase in noise.

In secondary analyses, we checked for nonlinearity of the associations between noise and BP using b-splines and categories of exposure overall and stratified by cohort. We further explored adjustment for additional potential sources of confounding (body mass index, current alcohol use, physical activity, and dietary sodium; PM_2.5 and light pollution) as well as adjustment for possible selection bias due to attrition, using an inverse probability weighting approach.³⁵ Although our primary hypothesis is that exposure to noise would result in a reversible shift in BP levels, we still tested whether noise was associated with altered aging trajectories by estimating the difference in the BP rate of change per year of age, using a noise by age interaction term. Finally, to check for consistency with our main findings, we explored associations with prevalent but not incident hypertension due to the very small number of CHAP participants who were not hypertensive at baseline.

Positive associations between noise and BP were present in both CHAP and MESA (Figure 1). Following adjustment for personal and neighborhood characteristics, a 10-dBA greater residential noise level was associated with 0.6 (95% CI, −0.6 to 1.8), 2.9 (95% CI, 0.6–5.3), and 1.2 (95% CI, 0.1–2.2) mm Hg higher SBP in CHAP, MESA, and both cohorts, respectively. Likewise, there was a 1.1 (95% CI, 0.6–1.7), 1.3 (95% CI, 0.1–2.5), and 1.1 mm Hg higher DBP (95% CI, 0.4–1.7) for a 10 dBA increase in noise for CHAP, MESA, and both cohorts, respectively. Although associations with SBP (but not DBP) were suggestively stronger in the MESA cohort, these results were not statistically different from one another. These associations were monotonic and largely linear (Figure 2), insensitive to different adjustments for antihypertensive medication use (Figure S2), adjustment for additional putative confounders, including coexposures (Figure S3), and incorporation of inverse probability weighting to adjust for possible selection bias from differential attrition (Figure S3). Our secondary analyses also indicated that noise was associated with prevalent hypertension (odds ratio, 1.2 [95% CI, 1.0–1.3]) but was not associated with the pace at which BPs changed as participants aged (Table S2).

Noise was associated with aTRH (odds ratio, 1.2 [95% CI, 1.0–1.4], P=0.04) and were consistent across both study cohorts (Figure 1; CHAP: 1.2 [95% CI, 1.0–1.4]; MESA: 1.4 [95% CI, 0.8–2.4]). These associations were not sensitive to changes in BP cutoffs and were robust to further adjustment for environmental factors (light pollution and PM_2.5), and body mass index, physical activity, alcohol intake, and sodium intake, although the latter slightly strengthened the association with aTRH in the subsample with data on these parameters (n=3047, Figure S3).

Our study had limitations that warrant mention. Our noise assessment, while an improvement over that of previous studies,¹⁷ was still an aggregate measure and did not allow us to disentangle the different aspects of noise exposure including timing of exposure, factors affecting exposure,⁶⁴ and originating source. We also used noise levels based on daytime and outdoor levels, which have not been associated with BP as strongly as nighttime⁴⁸ and indoor noise,⁵² and failed to capture any temporal characteristics of noise that can elicit strong stress reactions such as intermittent noise fluctuations,⁶⁵ peak nighttime noise, and the timing of the nighttime noise exposure,⁴³ which would help us determine whether noise is detrimental by disrupting sleep.^43,44 Similarly, we did not have data on factors that affect noise exposures, such as window-opening and room orientation. Collectively, this likely added greater imprecision to our exposure measure and may have reduced our ability to detect the true magnitude of the association between noise and BP. Notably although data on noise sources may be important from a noise mitigation perspective, a recent meta-analysis found no difference in associations between air, roadway, and railway noise with prevalent hypertension.¹⁶ Finally, our analysis that conditions on hypertension and hypertension medication use has the potential to bias observed associations if predictors of higher levels of noise, such as socioeconomic score, also predict treatment-related factors like participants’ adherence and suboptimal dosing. We suspect that this is not too problematic since our models were robust to adjustment for socioeconomic score.

In summary, this research showed that not only is noise associated with higher BPs but also newly associates noise with resistant hypertension. These findings and its consistency across 2 cohorts and with those results reported in Europe, altogether suggest that that body of literature has direct relevance to possible standard setting in the United States. Such standards could have wide impacts as 50% or 100 million of Americans experience noise at high levels⁶⁶ and thus may be an effective means of improving BP levels. Further research is warranted into the specific characteristics of noise that are harmful, especially nighttime noise.