Long-term Noise Exposures and Cardiovascular Diseases Mortality: A Study in 5 U.S. States

Abstract

Background:

Previous studies have linked noise exposure with adverse cardiovascular events. However, evidence remains inconsistent, and most previous studies only focused on traffic noise, excluding other anthropogenic sources like constructions, industrial process and commercial activities. Additionally, few studies have been conducted in the U.S. or evaluated the non-linear exposure-response relationships.

Methods:

We conducted a relative incidence analysis study using all cardiovascular diseases mortality as cases (n=936,019) and external causes mortality (n=232,491) as contrast outcomes. Mortality records geocoded at residential addresses were obtained from five U.S. states (Indiana 2007, Kansas 2007–2009, Missouri 2010–2019, Ohio 2007–2013, Texas 2007–2016). Time-invariant long-term noise exposure was obtained from a validated model developed based on acoustical measurements across 2000–2014. Noises from both natural sources (natural activities, including animals, insects, winds, water flows, thunder, etc.) and anthropogenic sources (human activities, including transportation, industrial activities, community facilities & infrastructures, commercial activities, entertainments, etc.) were included. We used daytime and nighttime total anthropogenic noise & day-night average sound pressure level combining natural and anthropogenic sources as exposures. Logistic regression models were fit controlling for Census tract-level & individual-level characteristics. We examined potential modification by sex by interaction terms and potential non-linear associations by thin plate spline terms.

Results:

We observed positive associations for daytime anthropogenic L₅₀ (sound level exceeded 50% of time) noise (10-dBA OR=1.047, 95%CI 1.025–1.069), nighttime anthropogenic L₅₀ noise (10-dBA OR=1.061, 95%CI 1.033–1.091) in a two-exposure-term model, and overall L_dn (day-night average) sound pressure level (10-dBA OR=1.064, 95%CI 1.040–1.089) in single-exposure-term model. Females were more susceptible to all three exposures. All exposures showed monotonic positive associations with cardiovascular mortality up to certain thresholds around 45–55 dBA, with a generally flattened or decreasing trend beyond those thresholds.

Conclusions:

Both daytime anthropogenic and nighttime anthropogenic noises were associated with cardiovascular disease mortality, and associations were stronger in females.

1. Introduction

Noise, or unwanted sound, is an important environmental health risk factor that has been linked with hearing loss, activity interference, and annoyance¹. In recent years, studies have also reported that long-term noise exposures could be associated with other adverse health outcomes, including cardiovascular diseases. Many previous studies have reported that traffic-related noise were associated with several different adverse cardiovascular health outcomes like myocardial infarction (MI), coronary heart disease (CHD), hypertension, ischemic heart disease (IHD), and cardiovascular disease mortality^2–10. Meanwhile, a World Health Organization (WHO) report published in 2011 has estimated that in European Region, the disability-adjusted life-years (DALYs) lost from environmental noise was 61,000 years for ischemic heart disease (IHD), supporting an association between noise exposure and adverse cardiovascular health outcomes¹¹. However, some studies have also reported null or inconsistent results regarding the associations between noise exposure and cardiovascular diseases^12–14. In addition, few studies have examined these effects of noise in the United States, where the housing stock is in general set back further from the road and made of different materials, making it difficult to extrapolate from other areas, like European studies.

Moreover, most previous studies only focused on transportation-related noises including road traffic, railway, and aircraft^{2,3,5,6,13–22}. Fewer studies have included noise from other anthropogenic sources (like industrial activities, construction, lawn mowing and leaf blowing, road repair, community facilities & infrastructures, entertainments, and commercial activities at the city centers) or considered a combination of anthropogenic and natural sources which is closer to the real-world exposure profiles perceived by people. Meanwhile, regarding the potential heterogeneity of associations for daytime and nighttime exposures, previous studies on this topic have also reported inconsistent results^17,19,23,24. Besides, most of the previous studies have only focused on small areas like cities or small countries, which could be significant limitations on study power and generalizability^{2–4,19,22,25}.

Therefore, we aimed to conduct a study addressing these knowledge gaps and limitations. In this study, we utilized a validated sound pressure model to capture long-term total anthropogenic noise, as well as the total sound pressure level combining both anthropogenic and natural sources (such as wind, water flow, wild animals, insects, thunder, etc.), which could give us a better picture of the actual exposure profiles. With mortality data geocoded to residential addresses across 5 different states in the U.S., we included a larger population across regions with more diverse geographic and demographic characteristics than many previous studies limited to small areas. Our study includes entire states including smaller cities, towns, and rural areas, in contrast to studies done predominantly in large cities. We also further examined the potential modifications by sex and the potential non-linear associations between noise exposure and cardiovascular disease mortality in this study.

2. Materials and Methods

2.1. Mortality Data & Study Design

We obtained the individual residential geocoded mortality records for all deaths from five states of the U.S.: Indiana (IN) for year 2007, Kansas (KS) for years 2007–2009, Missouri (MO) for years 2010–2019, Ohio (OH) for years 2007–2013, and Texas (TX) for years 2007–2016. The mortality data were obtained from the Indiana Department of Health; Bureau of Public Health Informatics, Kansas Department of Health and Environment; Missouri Department of Health and Senior Services (DHSS); Ohio Department of Health; and Texas Department of State Health Services (DSHS), Center for Health Statistics Health Information Resources Branch.

The mortality records include the information of date of death, cause of death (as ICD-10 codes), residential address at death (geocoded to longitude and latitude), age, sex, race, and education level. The outcome of interest for this study was all deaths due to cardiovascular diseases (ICD-10 code I00-I99), which we identified as cases. To conduct a relative incidence analysis, we used all deaths due to external causes (ICD-10 code V00-Y99), which we believed to be unrelated to the noise exposure, as the contrast^4,26. In the original dataset, there were 1,254,907 deaths from cardiovascular diseases and external causes, with 1,002,672 identified as cases and 252,235 identified for contrast. We dropped mortality records with missing age, residential address, sex, race, and education level. This procedure left 1,168,510 deaths (93.1%) remaining in the final dataset, of which 936,019 were cases (93.4% remaining) and 232,491 were contrast outcomes (92.2% remaining).

All mortality data were stored, managed, processed, and analyzed in a level 3 protected server in compliance with the Data Use Agreements. This study was approved by the institutional review board at Harvard T.H. Chan School of Public Health.

2.2. Exposure Data

We obtained the long-term noise exposure data from a validated sound model developed based on the measurement data conducted by U.S. National Park Service ^27,28. The original model was developed based on 270,000h of acoustical measurements from 190 sites located in National Parks across the contiguous U.S., which were mainly in natural areas. The geographic distribution of these measurement sites could be found in the paper describing this model²⁷. The explanatory variables were derived from national geospatial data layers including location, climatic, landcover, hydrological, anthropogenic, temporal, and equipment information²⁷. The model was developed based on a random forest algorithm with the usage of cross validation to evaluate performance, and generated A-weighted sound pressure level predictions based on the collected measurements and explanatory factors²⁷.

Based on this version, a revised model was then developed across the contiguous U.S. to include urban areas in addition to the natural areas which have already been incorporated when the original model was fit²⁸. This revised model was also based on a random forest algorithm and has included measurements from both urban and natural areas which encompassed 1.5 million hours of measurements from 492 site locations in total across the contiguous U.S. during 2000–2014²⁸. The natural measurements were collected from the sampling sites located in National Parks mentioned before, and the urban measurements were collected from the sampling sites located in 14 U.S. urbanized areas (San Antonio, TX, Austin, TX, Vicksburg, MS, Los Angeles, CA, Riverside, CA, Kill Devil Hills, NC, San Francisco, CA, Washington, DC, Denver, CO, Bridgeport, CT, New York, NY, Boston, MA, Milwaukee, WI and Seattle, WA)^28,29. The data from airport noise monitoring system were also included in the model. The explanatory variable included the spatial information of topography, climate, landcover, hydrology, traffic information (traffic volume, distance to major road, etc.), and anthropogenic constructions & activities (human modification from commercial land use, industrial land use, etc.) to capture different types of natural and anthropogenic noise sources²⁸. The natural noises were considered as noises from natural activities like wild animals, wind, water flows, etc. The anthropogenic noises were considered as noises from human activities, like transportation, industrial, community facilities & infrastructures, entertainment, construction, road repair, and commercial activities, etc. Based on this revised model, the time-invariant predicted A-weighted sound pressure level across the contiguous U.S. representing time period 2000–2014 has been generated at a 270m×270m geographical resolution²⁸. Both anthropogenic and natural sound levels were estimated through this model, and the anthropogenic ones encompassed all the anthropogenic activities not only the traffic source. The predicted value from this model showed excellent correlation with measured data (R²=0.98), and the predictive performances were good in both natural and urban areas through the figures in previous paper describing the model²⁸.

In this study, we included three variables estimated from this model as our exposure terms: (1) anthropogenic daytime A-weighted sound pressure level (7am-7pm) exceeded 50% of the time (daytime ant. L₅₀) to capture the daytime total anthropogenic noise; (2) anthropogenic nighttime A-weighted sound pressure level (7pm-7am) exceeded 50% of the time (nighttime ant. L₅₀) to capture the nighttime total anthropogenic noise; (3) average total (anthropogenic and natural) A-weighted day-night average sound level over a 24-hour period where sound from 10pm to 7am is upweighted by 10 dBA (L_dn) to capture the overall sound pressure level. We treated them as long-term noise exposure terms. The noise exposure was linked to the mortality data by residential addresses coded as latitude and longitude. Since the sound model only developed the aggregated measurements across 2000–2014, they were time-invariant terms to capture the noise exposure across the study period. That means people living in the same place but different years were assigned the same exposure values.

2.3. Covariates Data

We obtained the Census tract-level demographic and socioeconomic variables from the U.S. Decennial Censuses (2000 and 2010) and the American Community Survey (ACS) 5-year estimate data (2009–2019) retrieved from the IPUMS data service³⁰. The variables included proportions of the population self-reported as White or Black, separately; proportions of the population age ≥65 or ≤18, separately; proportions of workers age ≥16 who commute to work by automobile, motorcycle, bicycle, public transportation, and walking, separately; proportion of the adult population with at least a high school degree or equivalent; proportion of adult population with income below the poverty line; proportion of households receiving public assistance; proportion of housed individuals renting their property; proportion of residences with at most two housing units in the structure; and median household income. These variables were used together in trying to capture the potential confounding by area-level socioeconomic, development, and demographic status. Linear interpolation was used to fill in the missingness for 2001–2008.

To account for air pollution as a confounder, we also obtained the daily PM_2.5 (24-hour mean) level predictions at 1 km×1 km spatial resolution from 2000 to 2016 for the contiguous U.S. from validated spatial-temporal ensemble models integrating machine learning algorithms^31,32. We then calculated the annual average PM_2.5 concentrations for each 1 km×1 km grid and aggregated them to Census tract level. Linear extrapolation was used to fill the missingness for 2017–2019. Then, we linked the annual PM_2.5 data to mortality records by the residential address and mortality year.

The individual-level age, sex, race, education records, and year and state of death were directly obtained from mortality data. We also included them as potential confounders in analysis.

2.4. Statistical Analysis

We applied a relative incidence analysis method in this study, which has also been used in several previous studies^4,26. In this method, we considered all mortality due to cardiovascular disease as cases and coded them as 1. We considered all mortality due to external causes, which we believed were not related to noise exposure, as contrast outcomes and coded them as 0. We then fit the logistic regression models for statistical analyses:

$$Logit(PrCase=1|\mathit{E},\mathit{X})=\alpha +{\mathit{\beta }}_1^{\prime }\mathit{E}+{\mathit{\beta }}_2^{\prime }\{\mathit{X}\}$$

In this formula, Case was the response for case and contrast outcomes, $\mathit{E}$ was the vector for exposure term(s), $\mathit{X}$ was the vector for covariates for confounder control, which included all variables mentioned in previous section: the individual-level factors, the census tract-level factors, and the annual PM_2.5 at residential address. We also included indicator variables for year and state to account for potential residual confounding by different years and different states.

We conducted 3 sets of analyses. First, for the main analyses, we fit four different models with four different exposure terms combinations: model 1 with only daytime ant. L₅₀, model 2 with only nighttime ant. L₅₀, model 3 with both daytime and nighttime ant. L₅₀, and model 4 with only L_dn. Since daytime noise exposure and nighttime noise exposure could act as confounder to each other when assessing their associations with cardiovascular disease mortality and are required to be adjusted for, we considered model 3 as our main model. Moreover, we have also conducted a crude analysis without adjusting for any potential confounders.

Then, to explore potential modifications by sex, we ran the main models with additional interaction terms between each exposure term and sex. We have also repeated the main analyses and sex modifications analyses without controlling for PM_2.5.

Finally, we examined the non-linear associations by fitting thin plate spline terms for noise exposures in the model. Thin plate spline is a spline-based function to define smooth terms in generalized additive models. We fit the terms with fixed 3 degree of freedoms to estimate the non-linear associations. More details of thin plate splines has been elaborated by Wood, et al. previously^33,34. We fit the spline terms models for our main model (model 3) and the model for overall day-night average sound pressure level (model 4) to assess for non-linearity. The spline terms models for model 1 and model 2 have also been fit as secondary analyses.

All the analyses were conducted using R 4.2.3³⁵. The logistic regression models were applied through the glm function from the stats package, and the thin plate spline terms were applied through the gam function from the mgcv package^34,36.

3. Results

3.1. Summary Statistics

Table 1 summarizes the demographic characteristics of cases and contrast outcomes included in this study. Of the 1,168,510 deaths (936,019 cases and 232,491 contrast outcomes), 53.3% were male, 86.3% were White and 11.7% were Black. Meanwhile, 30.5% deaths had an education level higher than high school and 27.3% didn’t completed a high school degree or equivalent. The mean age of all deaths was 71.1 years old. The number of cases and contrast outcomes for each state were presented in Table S1 of supplemental materials. Each state included in our study had similar portions of cases numbers compared to contrast outcomes numbers.

Table 1.

Demographic Characteristics of Cases and Contrast Outcomes

Characteristics	Overall	Cases	Contrast Outcomes
Total, n	1,168,510	936,019	232,491
Sex, n (%)
Male	622,562 (53.3)	465,291 (49.7)	157,271 (67.6)
Female	545,948 (46.7)	470,728 (50.3)	75,220 (32.4)
Race, n (%)
White	1,007,851 (86.3)	811,778 (86.7)	196,073 (84.3)
Black	139,182 (11.9)	108,804 (11.6)	30,378 (13.1)
Others	21,477 (1.8)	15,437 (1.6)	6,040 (2.6)
Education Level, n (%)
Less than High School	319,456 (27.3)	257,269 (27.5)	62,187 (26.7)
High School or Equivalence	492,567 (42.2)	395,703 (42.3)	96,864 (41.7)
Higher than High School	356,487 (30.5)	283,047 (30.2)	73,440 (31.6)
Age, Mean (S.D.)	71.1 (20.2)	76.7 (14.9)	48.9 (23.2)

Abbreviation: S.D. for standard deviation.

Cases and contrast outcomes had similar race characteristics, but cases were much more balanced by sex, with 49.7% of the cases versus 67.6% of the contrast outcomes being male. Contrast outcomes had a slightly higher proportion with education level higher than high school (31.6% vs. 30.2%), but also a much younger profile than cases (mean age at death 48.9 vs. 76.7)

Table 2 summarizes the exposure profiles of cases and contrast outcomes. Regarding all 1,168,510 deaths included in this study, we had a mean daytime anthropogenic L₅₀ of 47.2 dBA and nighttime anthropogenic L₅₀ of 44.4 dBA, with a mean L_dn of 58.3 dBA. The interquartile ranges for the three exposure terms were 5.29 dBA, 3.54 dBA, and 4.80 dBA, respectively. Cases and contrast outcomes had very similar exposure profiles, with very close descriptive statistics regarding all three exposure terms. A more detailed description of exposure profiles for each of the 5 U.S. states included in this study was presented in Table S2 of supplemental materials. Generally, the exposure profiles for each state are also close with similar descriptive statistics.

Table 2.

Exposure Profiles of Cases and Contrast Outcomes

Exposure Terms	Groups	Mean	S.D.	IQR	Median	10th	25th	75th	90th
Daytime Ant. L50 (dBA)
	Overall	47.20	4.69	5.29	47.87	40.74	44.73	50.02	52.85
	Cases	47.18	4.64	5.22	47.84	40.84	44.74	49.96	52.74
	Contrast Outcomes	47.29	4.86	5.60	47.97	40.39	44.67	50.27	53.31
Nighttime Ant. L50 (dBA)
	Overall	44.42	3.26	3.54	44.50	41.09	42.88	46.41	47.83
	Cases	44.40	3.23	3.50	44.47	41.12	42.88	46.38	47.79
	Contrast Outcomes	44.50	3.39	3.67	44.61	40.96	42.86	46.53	48.01
Total Ldn (dBA)
	Overall	58.33	3.33	4.80	58.28	54.09	55.94	60.74	62.67
	Cases	58.30	3.30	4.76	58.23	54.11	55.94	60.70	62.59
	Contrast Outcomes	58.45	3.44	4.95	58.46	54.02	55.96	60.91	62.99

Abbreviation: S.D. for standard deviation; 10^th for 10^th percentile; 25^th for 25^th percentile; 75^th for 75^th percentile; 90^th for 90^th percentile.

3.2. Main Analyses & Modification by Sex

Table 3 presents the results from the main analyses. We observed positive associations between noise exposures and total cardiovascular mortality for daytime anthropogenic L₅₀ and nighttime anthropogenic L₅₀ in single exposure term model1 and model 2, as well as for total L_dn in model 4 with odds ratios (OR) per 10 dBA increase as 1.078 (95% CI: 1.061–1.096), 1.103 (95% CI: 1.078–1.126), and 1.064 (95% CI: 1.040–1.089), respectively. In model 3 including both daytime and nighttime anthropogenic L₅₀, we observed attenuated but still positive associations for daytime noise (OR per 10 dBA increase of 1.047 with 95% CI as 1.025–1.069) and nighttime noise (OR per 10 dBA increase of 1.061 with 95% CI of 1.033–1.091). We also observed that when the associations were measured per 10 dBA increase, nighttime noise had a higher odds ratio compared to daytime noise. However, when the associations were measured per IQR or standard deviation increase, nighttime noise had slightly lower odds ratio compared to daytime noise. While the daytime and nighttime noise could act as confounders to each other and therefore should be mutually adjusted, the correlation between the daytime and nighttime anthropogenic L₅₀ (r=0.758) is large may increase the confidence intervals in model 3.

Table 3.

Main Analysis Results^a

Modelb, c	Exposure Terms	Per 10 dBA	Per Overall S.D.	Per Overall IQR
Model 1	Daytime Ant. L50	1.0781 (1.0609, 1.0956)	1.0359 (1.0281, 1.0437)	1.0406 (1.0318, 1.0495)
Model 2	Nighttime Ant. L50	1.1025 (1.0797, 1.1257)	1.0323 (1.0253, 1.0394)	1.0351 (1.0275, 1.0428)
Model 3d	Daytime Ant. L50	1.0471 (1.0254, 1.0692)	1.0218 (1.0118, 1.0319)	1.0246 (1.0134, 1.0360)
	Nighttime Ant. L50	1.0613 (1.0328, 1.0905)	1.0196 (1.0106, 1.0286)	1.0212 (1.0115, 1.0311)
Model 4	Total Ldn	1.0639 (1.0395, 1.0888)	1.0208 (1.0130, 1.0287)	1.0302 (1.0188, 1.0417)

Abbreviations: S.D. for standard deviation; IQR for interquartile range.

^aResults presented as odds ratios and corresponding 95% confidence intervals.

^bModel 1: only include daytime anthropogenic L₅₀ as exposure term; Model 2: only include nighttime anthropogenic L₅₀ as exposure; Model 3: include both daytime and nighttime anthropogenic L₅₀ as exposure terms; Model 4: only include total L_dn as exposure term.

^cAdjusted for: individual-level age, sex, race, and education level; census tract-level proportions of the population self-reported as White or Black separately, age ≥65 or ≤18 separately, proportions of workers age ≥16 who commute to work by automobile, motorcycle, bicycle, public transportation, and walking separately, proportion of the adult population with at least a high school degree or equivalent, proportion of the adult population with income below the poverty line, proportion of households receiving public assistance, proportion of housed individuals renting their property, proportion of residences with at most two housing units in the structure, and median household income; residential address-specific annual average PM_2.5 concentration; indicator variables for year and state.

^dConsidered as main model.

The results of modifications by sex are presented in Table 4. We observed that generally, females were more susceptible to noise than males, especially for nighttime anthropogenic L₅₀ noise in both model 2 and model 3. This modification was also quite significant regarding total L_dn in model 4. However, for daytime anthropogenic L₅₀ noise in model 1 and model 3, the odds ratios for females were only slightly higher than for males, which were very close to each other.

Table 4.

Modifications by Sex Results^a

Modelb, c	Exposure Terms	Sex	Per 10 dBA	Per Overall S.D.	Per Overall IQR
Model 1	Daytime Ant. L50	Male	1.0779 (1.0585, 1.0976)	1.0358 (1.0270, 1.0446)	1.0405 (1.0305, 1.0505)
		Female	1.0786 (1.0544, 1.1034)	1.0361 (1.0251, 1.0472)	1.0409 (1.0284, 1.0535)
Model 2	Nighttime Ant. L50	Male	1.0852 (1.0593, 1.1118)	1.0270 (1.0189, 1.0352)	1.0293 (1.0206, 1.0382)
		Female	1.1353 (1.1008, 1.1709)	1.0422 (1.0318, 1.0528)	1.0459 (1.0345, 1.0573)
Model 3d	Daytime Ant. L50	Male	1.0467 (1.0234, 1.0706)	1.0216 (1.0109, 1.0325)	1.0244 (1.0123, 1.0367)
		Female	1.0478 (1.0206, 1.0757)	1.0221 (1.0096, 1.0348)	1.0250 (1.0108, 1.0394)
	Nighttime Ant. L50	Male	1.0440 (1.0132, 1.0757)	1.0141 (1.0043, 1.0241)	1.0153 (1.0047, 1.0261)
		Female	1.0932 (1.0553, 1.1325)	1.0295 (1.0177, 1.0414)	1.0320 (1.0192, 1.0450)
Model 4	Total Ldn	Male	1.0600 (1.0326, 1.0881)	1.0196 (1.0107, 1.0285)	1.0284 (1.0155, 1.0414)
		Female	1.0709 (1.0372, 1.1056)	1.0231 (1.0122, 1.0340)	1.0334 (1.0177, 1.0494)

Abbreviations: S.D. for standard deviation; IQR for interquartile range.

^aResults presented as odds ratios and corresponding 95% confidence intervals.

^bModel 1: only include daytime anthropogenic L₅₀ as exposure term; Model 2: only include nighttime anthropogenic L₅₀ as exposure; Model 3: include both daytime and nighttime anthropogenic L₅₀ as exposure terms; Model 4: only include total L_dn as exposure term.

^cAdjusted for: individual-level age, sex, race, and education level; census tract-level proportions of the population self-reported as White or Black separately, age ≥65 or ≤18 separately, proportions of workers age ≥16 who commute to work by automobile, motorcycle, bicycle, public transportation, and walking separately, proportion of the adult population with at least a high school degree or equivalent, proportion of the adult population with income below the poverty line, proportion of households receiving public assistance, proportion of housed individuals renting their property, proportion of residences with at most two housing units in the structure, and median household income; residential address-specific annual average PM_2.5 concentration; indicator variables for year and state.

^dConsidered as main model.

The crude analyses results without adjusting for any potential confounders are presented in Table S3 in supplemental materials. We could see that upon adjusting for potential confounders, the associations between noise exposure and cardiovascular mortality turned from strong negative to strong positive, indicating potential negative and even qualitative confounding by SES and other factors controlled by our set of potential confounders.

The results of repeated analyses without controlling for PM_2.5 are presented in Table S4 and Table S5 in supplemental materials. The pattern of associations excluding PM_2.5 were generally consistent with the results controlling for PM_2.5 presented in the main text. The associations excluding PM_2.5 were slightly attenuated compared to the results controlling for PM_2.5, indicating some signs of confounding by air pollution.

3.3. Non-linearity with Thin Plate Splines

Figure 1 presents the results examining potential non-linear associations for daytime and nighttime anthropogenic L₅₀ in model 3, measured by the odds ratio taking the lowest exposure level as reference. For daytime anthropogenic L₅₀, we observed a monotonic increasing trend for the association between cardiovascular disease mortality and daytime noise up to exposure around 50 dBA, with a decreasing trend beyond this point. We also noticed that the confidence intervals were much wider in the range below around 35 dBA. The confidence intervals were also a little bit wider in the range beyond around 55 dBA resulting from fewer observations at these regions.

Figure 1:

Non-linearity Results & Exposure Level Distribution for Daytime (a) and Nighttime (b) Anthropogenic L₅₀ in Two Exposure Model (Model 3). Presented as Odds Ratios and Corresponding 95% Confidence Intervals Taking Lowest Exposure Level as Reference & Exposure Level Frequency Histograms.

For nighttime anthropogenic L₅₀, we also observed a monotonic increasing trend for the association between cardiovascular disease mortality and nighttime noise up to around 45 dBA, with a flattened and slightly decreasing trend in the region beyond this point. Like daytime anthropogenic L₅₀, the confidence intervals were much wider in the lower exposure range (below around 30 dBA) and upper exposure range (beyond around 50 dBA), again due to fewer observations in these rages. The upper range where the confidence intervals were much wider was also where deviations from monotonic increasing trend was seen.

Figure 2 presents the result for total L_dn in model 4. We observed a generally monotonic increasing trend for the associations between cardiovascular diseases and total L_dn sound levels up to around 55 dBA. For the region between around 55 dBA and 60 dBA, there was a flattened and a little bit decreasing trend, and then there was another range with weak increasing trend beyond around 60 dBA. We also noticed that the confidence intervals at upper range beyond around 65 dBA were much wider than the other regions.

Figure 2:

Non-linearity Results & Exposure Level Distribution for Total L_dn in Model 4. Presented as Odds Ratios and Corresponding 95% 578 Confidence Intervals Taking Lowest Exposure Level as Reference & Exposure Level Frequency Histograms.

The results examining non-linearity of model 1 and model 2 (single exposure term models) are presented in Figure S1 of supplemental materials. The findings are similar to what we observed from the main model, except for nighttime anthropogenic L₅₀ at upper range. There was a more clear decreasing trend beyond around 45 dBA for nighttime noise in single exposure model, which was more flattened in the main model when the daytime noise was also controlled.

4. Discussion

In this study of all cardiovascular deaths in five U.S. states over the time periods mentioned for each, we observed positive associations of address-specific daytime anthropogenic noise, nighttime anthropogenic noise, and total day-night average sound pressure level with increased risks of cardiovascular mortality in contrast to mortality due to external causes in single exposure models. The magnitude of odds ratio measured in 10 dBA increase for nighttime anthropogenic L₅₀ is higher than for daytime anthropogenic L₅₀ in the two-exposure model, indicating nighttime noise may be a more important risk factor than daytime noise. However, we should also notice that the correlation between daytime and nighttime noises may also influence the statistical analysis results.

We observed that females were more susceptible to noise than males, especially for nighttime noise. The odds ratios for females were always higher than for females in all interaction models. For the non-linearity analyses from model 3 and model 4, we observed all three exposure terms with a monotonic increasing trend in associations with cardiovascular mortality in the exposure ranges with the bulk of data but with less clear associations above and sometimes below that range. For daytime anthropogenic L₅₀, the upper level of that range is around 50 dBA, with a decreasing trend with increasing exposure beyond this value; for nighttime anthropogenic L₅₀, this threshold is around 45 dBA, followed by a flattened and slightly decreasing trend beyond this value; for total L_dn, this threshold is around 55 dBA, followed by a flattened and slightly decreasing trend until around 60 dBA, and a weak increasing trend beyond 60 dBA. The curve for all three exposure terms had wider confidence intervals at lower and upper range, corresponding to fewer observations at those levels. The non-linearity analyses for daytime and nighttime noise from single exposure term models showed similar patterns, except that there were more clear decreasing trends beyond the thresholds mentioned above for daytime and nighttime anthropogenic L₅₀.

Most previous studies examining the associations between long-term noise exposures and human health have focused on transportation-related noises, including road traffic, railway, and aircraft^{7,8,10,12,21,37,38}. Some studies have provided evidence supporting the associations between noise and adverse cardiovascular health outcomes. A study conducted in France focusing on aircraft has reported associations between long-term noises and mortality from all cardiovascular diseases, CHD, and MI³. Two studies conducted in Spain have also reported associations between traffic noise and mortality from all cardiovascular diseases, MI, and hypertension^4,25. In Denmark, based on Danish Diet, Cancer and Health Cohort and Danish Nurse Cohort, there are also studies providing evidence of long-term road traffic exposures’ associations with mortality from all cardiovascular mortality and the incidence of MI and atrial fibrillation (AF)^5,6,15. Also in Denmark, a larger national cohort study has reported association between road traffic noise exposure with IHD, MI, angina pectoris, and heart failure. A pooled analysis integrating 9 Scandinavian cohorts has also concluded with associations between road traffic and railway noise and IHD.

However, a number of studies have also reported null or inconsistent associations between long-term noise exposures and adverse cardiovascular health outcomes, including stroke, IHD, and MI^12,13,39,40. Regarding cardiovascular diseases mortality, there is also mixed evidence. A study conducted in São Paulo, Brazil focusing on aircraft noise has concluded only suggestive evidence for noises’ associations with mortality from all cardiovascular diseases and CHD¹⁸. Another study based on the Danish Nurse Cohort has reported associations with road traffic exposures for all-cause mortality, but not for cardiovascular diseases mortality⁴¹.

The mechanisms behind the noise-cardiovascular disease relationship are also not well-illustrated. Some studies have pointed out that the long-term noise exposure could be linked with stress hormones release, endothelial dysfunction, and higher blood pressure levels^42–44. Some other proposed mechanism has linked noise exposure with adverse cardiovascular health events through increased stress-associated activity of amygdala and the heightened arterial inflammation following to it⁴⁵. More studies are still needed to further identify the mechanisms behind this noise-cardiovascular health association.

Air pollution has been regarded as an important potential confounder when examine the link between noise exposure and human health. Two previous studies conducted in Denmark have reported the associations between noise exposures and adverse cardiovascular health outcomes to be attenuated after controlling PM_2.5, PM₁₀, NO₂, and NO_X^46,47. Another cohort study conducted in Sweden has also reported a little bite attenuated association for noise and cardiovascular events after controlling for NO_X. Our study has found strong and positive associations for anthropogenic daytime and nighttime noises and total day-night sound pressure level with mortality from all cardiovascular diseases while controlling for air pollution, which provides evidence supporting an impact by noise exposures on human cardiovascular health. However, in our analyses, controlling PM_2.5 resulted in somewhat larger effect size estimates, which are very close to the results without controlling air pollution. One possible explanation is that the confounding of air pollution in our study population may be weaker than in European studies. Another possible explanation is that we were only able to capture the air pollution exposure at the year of death based on the data we have (mortality records data with only information of address at the year of death without any residential history record). The lack of exposure history may result in an inadequate capture of the potential confounding by air pollution.

Most previous studies examining noise’s health impact have used L_den, which is the day-evening-night average noise level imposing additional penalties for sound levels during evenings and nights (usually 5 dBA for evenings and 10 dBA for nights) to capture the exposure profiles^{3,5,15,20,39,41,46,47}. Some studies also used L_dn sound level measurements, which was also used in over study, averaging the sound levels for days and nights without separating evenings from them^2,18. Both L_den and L_dn incorporated daytime and nighttime noises into one exposure metric. In addition, there are studies examining daytime and nighttime noise separately. However, their results are mixed. A study conducted in London focusing on MI survivors has reported modest associations for daytime traffic noise with all-cause mortality and MI readmission⁴⁰. Another study also conducted in London has reported stronger effects for daytime road traffic noise on cardiovascular diseases mortality than nighttime road traffic noise²³. However, there are also studies reporting a more important role for nighttime noise than daytime noise. A report focusing on existing experimental and epidemiological studies published in 2020 has concluded that nighttime traffic noise was more important than daytime noise regarding cardiovascular effects²⁴. Another study conducted in Switzerland has reported stronger effects for nighttime traffic noise on arterial stiffness, which is an important determinant for cardiovascular diseases¹⁷. There are also two studies utilizing the same exposure metrics as our study for total anthropogenic noises: one has reported no association for either daytime or nighttime noise, while the other has reported an association for nighttime noise but not for daytime^48,49. Besides, a cohort study conducted in Pisa, Italy on vehicular traffic noise has reported almost same association for nighttime and daytime noise with cardiovascular disease mortality and hospitalization²².

In our study, the daytime and nighttime noise have been examined through both separate term and an overall day-night average L_dn sound pressure level. Our study has reported strong positive associations for mortality from all cardiovascular diseases with both daytime and nighttime anthropogenic noises separately, and also strong positive associations for cardiovascular mortality with total day-night average L_dn sound pressure level. For comparison between daytime and nighttime noise, our study suggested that nighttime noise might be a more important risk factor with higher odds ratio estimates per 10 dBA increase but given the current status of mixed results regarding daytime vs. nighttime noise, further research seems necessary.

Previous studies examining the potential modifications of the noise-health association by sex have also reported mixed results. Two studies based on the Danish Diet, Cancer and Health cohort have reported that males were more susceptible to noise regarding MI and blood pressure increase^6,20. The aforementioned study focusing on aircraft noise in France has also reported a higher susceptibility for males on cardiovascular disease mortality³. However, there are also studies in Switzerland and Sweden reporting higher risks for females on noise exposures’ associations with mortality, IHD, and stroke^12,19. There are also two cohort studies in Stockholm, Sweden and Pisa, Italy, and a review of researches in Austria have reported higher susceptibility for females to noise exposure regarding cardiovascular diseases^9,12,22. Besides, a study conducted in Barcelona, Spain has reported that males were more susceptible to noise regarding MI mortality, while females were more susceptible regarding hypertension mortality. Similarly, a systematic review conducted in 2021 has also concluded different susceptibilities for males and females regarding different noise types and different cardiovascular outcomes with mixed result pattern, with females more susceptible to community noise regarding hypertension, and male more susceptible regarding MI mortality and incidence, which has some form of inconsistency with our study results that females were suffering higher risks⁵⁰. Our study reported higher susceptibility to mortality from all cardiovascular diseases due to noise exposure for females than males regarding all three noise exposures: daytime anthropogenic, nighttime anthropogenic, and overall day-night average sound pressure level. These results are consistent with some previous studies, but there are also existing studies reporting higher susceptibility for male as mentioned above. Since existing studies’ results in this topic are still mixed and the potential biological mechanisms behind these differences are not clear yet, we still need more studies to further explore it.

Our non-linearity examination showed monotonic increasing trends between noise exposures and mortality from all cardiovascular diseases in almost all circumstances below certain thresholds. For daytime anthropogenic noise, there was a clear decreasing trends between noise exposures and cardiovascular mortality beyond the threshold of around 50 dBA, while for nighttime anthropogenic noise and overall day-night average L_dn, the trend beyond their threshold of around 45 and 55 dBA are more flattened, with another increasing trend for L_dn beyond over 65 dBA. Many previous studies examining the potential non-linearity of the noise-cardiovascular health associations have also found generally monotonic increasing trends, which are quite different from our results^{8,10,22,38,41,46,47}. However, some studies categorizing exposure into different groups have reported no consistent findings^18,23. There are also some studies that have reported more similar results to ours. The aforementioned cohort study in Stockholm, Sweden, has found that with increasing transportation L_den noise, the hazard ratios of IHD taking <45 dB as reference had a first increasing then decreasing trend, with decreasing happened at 50–54 dB group or 55–60 dB group¹². Another Swedish cohort study has also reported that compared to road traffic noise<53 dB group, the risk of cardiovascular diseases mortality would first decrease for 53–58 dB and 58–63 dB, and then increase beyond 63 dB, with the region of decreasing trend consistent with our findings²¹. Therefore, one possible explanation for our results could be the generally lower exposure profiles we get, which might only cover the exposure level up to the region where previous studies have reported decreasing trends. Another possible explanation might be due to our study design taking external causes mortality as contrast outcomes. We hold the assumption that external causes mortality is unrelated to noise exposures. However, at very high noise level, there could be situations where the noise might lead to annoyance and distraction and a higher probability of accidents like car crashes, resulting in higher external causes mortality risk.

There are some limitations in our study. First, we had a highly homogeneous study population with about 90% White participants, while this number in the whole U.S. according to the most recent data is only about 60%.⁵¹ Therefore, although our study already covered a wide range of geographic regions, there could still be vulnerable populations not well-represented, which may impose limitation to the results generalizability. Second, since we only had the residential address at the year of death, we were also not able to track the residential history to better capture the exposures years before the death. Therefore, for individuals who has moved prior to death, there could be measurement errors for their exposure assessments. However, we would expect these kinds of exposure measurement errors to be non-differential, and lead to a bias towards null.⁵² Third, due to the data limitation, we only had time-invariant exposure metrics to capture overall exposure levels across the study period but were not able to capture the varying exposures across different years. Thus, for individuals at different years, there might be source of exposure measurement errors. Meanwhile, since there is no information on the noise trend over years in this exposure model, we are not able to identify the exact direction of bias introduced by this type of exposure measurement error. Besides, the noise model utilized in our study has only separately modeled daytime and nighttime noise without a third time period for evening noise (typically when people are awake at home during 7pm to 10pm). Therefore, it may not be able to capture the different annoyance by people at home in the evening and in the night as well as previous studies also considered night as a separate time period^{3–6,10,12,20,21,37,39,41,46,53}. Fourth, the contrast approach in this study was based on the assumption that external causes mortality is unrelated to noise exposure. The validity of this assumption at different noise level would also influence the validity and precision of our results. The unbalanced case-contrast outcome ratio may also reduce study power. However, despite this issue, the 95% confidence intervals we get are still mostly not covering the null. Lastly, the correlation between daytime and nighttime anthropogenic noises may influence the results of the statistical analysis models including both daytime and nighttime noise at the same time.

There are also important advantages of this study. First is study power. Although the case-contrast outcome ratio was not ideal, we included nearly one million cases and over 230 thousand contrast outcomes across 5 different U.S. states and were able to estimate much more robust results than most previous studies, with almost all the 95% confidence intervals we got not covering the null values. Second, although our exposure measurements could not capture the time-varying exposures, this large-scale exposure model still allowed us to capture the anthropogenic noise and overall sound pressure level across much larger geographic regions and conduct a study including more areas with different geographic and demographic characteristics, like residents of smaller cities, towns, and other areas remote from urbanized areas in addition to the larger cities generally examined by previous studies. Third, this exposure model also incorporated all anthropogenic sources, not only the traffic like existing studies, which is a disadvantage when addressing the impact of traffic noise, but an advantage for examining all anthropogenic noises. For the urban planning and policy, the non-traffic noise, like from outdoor entertainments, construction, commercial activities, lawn mowers and leaf blowers, and the noises from all other human activities when people are crowded are also very important, which should be taken into consideration on how to reduce them. Lastly, our study controlled for PM_2.5, an important potential confounder which has not been widely considered by previous studies. Our study also provided more insight in U.S. to draw conclusions in this country, where homes are generally set back further from road and cities are generally more spread out than in Europe, as demonstrated by the lower noise levels we found.

5. Conclusions

We have observed positive associations for daytime anthropogenic noise, nighttime anthropogenic noise, and overall day-night sound pressure level combining both natural and anthropogenic sources with total cardiovascular disease mortality. The association for nighttime anthropogenic noise was stronger for daytime anthropogenic noise at 10 dBA increase scale. Females were found more susceptible to noises in this study, and generally monotonic increasing trends were found below certain thresholds at around 45–55 dBA for daytime noise, nighttime noise, and overall average day-night L_dn sound pressure levels.

Highlights.

• Validated model for noises from all human activities was used for exposures

• Mortality records from 5 U.S. states were collected to increase the study power

• Potential effect modifications by sex and non-linear associations were examined

• Both daytime and nighttime noises were found associated with cardiovascular death

• Males were found more susceptible to noises regarding cardiovascular mortality