Association Between Transportation Noise Exposure and Poststroke Outcomes: Population‐Based Brain Attack Surveillance in Corpus Christi Project

Abstract

Background

Transportation noise remains understudied as a potential risk factor for poststroke outcomes in the United States.

Methods

We conducted a prospective cohort study in the Brain Attack Surveillance in Corpus Christi Project (2016–2023) among patients with acute incident ischemic stroke. Transportation noise exposure was derived from US Department of Transportation data (2016, 2018) and defined as the maximum area‐weighted noise level within a 100‐m buffer of the residential address. Outcomes included initial stroke severity measured by the National Institutes of Health Stroke Scale, stroke recurrence, and all‐cause mortality. Linear regression was used to evaluate stroke severity, and Cox proportional hazards models were used for recurrence and mortality. Estimates were scaled to the interquartile range (IQR) of noise exposure to compare high (75th percentile) versus low (25th percentile) exposure.

Results

Among 2441 patients, median National Institutes of Health Stroke Scale score was 3.0 (IQR, 1.0–7.0). During follow‐up, 176 patients (7.2%) experienced a recurrent stroke and 818 (33.5%) died. Median follow‐up was 2.2 years (IQR, 0.7–4.3) for stroke recurrence and 2.5 years (IQR, 0.8–4.6) for mortality. Median transportation noise was 45.5 (IQR, 0.0–56.5) decibels. Higher transportation noise was not associated with stroke severity (−5% [95% CI, −12% 2%]; P=0.14) and mortality (hazard ratio to [HR], 1.08 [95% CI, to 0.941.24]; P=0.26) but was associated with a higher rate of stroke recurrence (HR, 1.36 [95% CI, 1.01 to 1.82]; P=0.04) after adjustments.

Conclusions

Transportation noise may be an environmental risk factor for stroke recurrence, underscoring the need for policies addressing noise exposure.

Nonstandard Abbreviations and Acronyms

BASIC

Brain Attack Surveillance in Corpus Christi Project

dBA

area‐level decibels

DOT

Department of Transportation

NIHSS

National Institutes of Health Stroke Scale

Research Perspective.

What Is New?

• The relationship between transportation noise exposure and poststroke prognosis remains understudied in the US population.

• In this population‐based cohort of patients with incident ischemic stroke, higher residential transportation noise exposure was associated with a higher rate of stroke recurrence but not with stroke severity of all‐cause mortality.

What Question Should Be Addressed Next?

• Future studies should replicate these findings in diverse populations, examine potential mechanisms explaining these findings, and assess the role of transportation noise in other cardiovascular and cerebrovascular outcomes in the US population.

Advances in stroke prevention and clinical management have led to a substantial decrease in stroke incidence in the United States. ¹ However, stroke remains a leading cause of death and disability, ² , ³ with recurrent strokes posing a particularly high threat to poststroke survival. Some studies suggest that temporal trends for recurrence ⁴ , ⁵ , ⁶ and mortality ¹ , ⁷ have plateaued or even increased in the past decade, highlighting the need to identify additional modifiable risk factors that influence stroke outcomes. Transportation noise, a pervasive environmental exposure, affects nearly 100 million people in the United States on a daily basis. ⁸ , ⁹ It has been linked to adverse cardiovascular events through mechanisms such as stress, sleep disruption, cognitive decline, and elevated blood pressure. ¹⁰ , ¹¹ , ¹² , ¹³ Although European studies report associations between transportation noise and incident stroke with relative risk ranging from 1.01 to 1.33 per 10 dB increase in the exposure, ¹⁴ , ¹⁵ , ¹⁶ , ¹⁷ data from the United States are limited. Additionally, the impact of noise on poststroke stroke outcomes, including severity, recurrence, and mortality remains underexplored. In this study, we investigated these associations in a biethnic population‐based cohort of patients with incident ischemic stroke.

METHODS

Data Availability Statement

Access to the deidentified data supporting the findings of this study will be considered upon reasonable request sent to llisabet@umich.edu based on existing institutional review board approvals and data use agreements.

Study Population

The BASIC (Brain Attack Surveillance in Corpus Christi) Project is an ongoing population‐based stroke surveillance study in Nueces County, Texas. ¹⁸ , ¹⁹ Located in the Coastal Bend region, Nueces is a largely urban county with a population of ∼350 000 residents (2022); 63% are of Hispanic origin, predominantly Mexican American. ²⁰ Corpus Christi, a major city in Nueces County, is the driving force of the regional economy, housing major petrochemical refineries and other industries. ²¹ It has a strong transportation network comprising 2 major highways (Interstate 37 and US Highway 77), 2 railroads, and 1 international airport. ²¹ Consequently, it is an ideal setting to examine the impact of transportation noise exposure on health outcomes.

Methods of the BASIC Project have been described previously. ¹⁸ , ¹⁹ Briefly, a combination of active and passive surveillance strategies has been used to capture stroke events in Nueces County since 2000. Patients younger than 45 years, living outside of Nueces County, or with strokes related to trauma were excluded per study protocol. Ischemic strokes were defined as an acute focal neurological deficit in the cerebrovascular distribution lasting >24 hours. For the present analysis, we included patients with incident (defined as those with no documented history of stroke/transient ischemic attack in the medical record) ischemic strokes from the BASIC Project identified between 2016 and 2023. The BASIC Project was approved by the University of Michigan institutional review board and the local hospital systems. All participants provided informed consent. This study is reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology guidelines.

Outcomes

The severity of the initial stroke (ie, the index event) was defined by the National Institutes of Health Stroke Scale (NIHSS) score abstracted from medical records. Stroke recurrence was captured through the BASIC Project surveillance system from January 1, 2016 through December 31, 2023 and defined as the first ischemic or hemorrhagic stroke occurring after the index event. All‐cause mortality was captured through a linkage of BASIC participants with the Texas Department of State Health Services death certificate database. Participants were followed through December 31, 2023 for this analysis. Time to stroke recurrence was calculated from the date of the index event to the date of stroke recurrence. Participants were censored at the time of death or at the end of follow‐up, whichever occurred first. Time to mortality was calculated from the date of the index event to the date of death. Participants were censored end of follow‐up.

Transportation Noise

The US Department of Transportation (DOT) has generated transportation noise data using deterministic models to estimate long‐term average 24‐hour equivalent sound levels (area‐weighted maximum 24‐hour sound level, A‐weighted to reflect the frequencies of the human ear) for the years 2016, 2018, and 2020. ²² We assigned transportation noise exposure using the DOT noise map closest to each participant's index stroke event year: stroke events in 2016 to 2017 were linked to the 2016 estimates, and stroke events in 2018 to 2023 were linked to the 2018 estimates. In our study region, the 2016 and 2018 maps include roadway and aviation noise. We calculated the noise exposure as the area‐weighted maximum 24‐hour sound level within a 100‐m buffer around each participant's residential address at the time of stroke, noting that the DOT noise model has a floor ²² wherein all noise levels <45 area‐level decibels (dBA) are recorded as 0 dBA.

Covariates

Individual‐level covariates included demographics (age, sex, race, ethnicity, insurance status, education) and health behaviors (smoking and alcohol use). These data were collected through medical record review and patient interviews. We also adjusted for neighborhood characteristics using tract‐level median household income and average population size, airborne particulate matter <2.5 μm in diameter, and nitrogen dioxide. Median household income and average population size were obtained from the ACS (American Community Survey). ²⁰ For patients with strokes between 2016 and 2017, ACS data from 2015 to 2019 were mapped to the 2010 US Census geographies and for patients with strokes between 2018 and 2023, ACS data from 2016 to 2020 were mapped to the 2020 US Census geographies. Air pollution variables were obtained from the Center for Air, Climate, and Energy Solutions and averaged across 2016 to 2020. ²³

Statistical Analysis

We summarized patient characteristics overall and by categories of transportation noise defined using the median transportation noise level. Continuous variables were reported as means±SD or medians and interquartile ranges (IQR), and categorical variables as frequencies and percentages. We used chi‐square tests to compare categorical patient characteristics by transportation noise categories and t tests for continuous variables.

We used linear regression models to assess the association between transportation noise and initial stroke severity defined by the log of the NIHSS score plus 1 (due to skewness). We used a sequential modeling strategy to account for confounding. Model 1 represented the unadjusted association between transportation noise and the outcome. Model 2 additionally adjusted for individual‐level sociodemographic variables (age, sex, race, ethnicity, insurance, education). Model 3 added health behaviors (smoking, alcohol use). Model 4 was a linear mixed‐effects model adjusting for tract‐level characteristics (median income, airborne particulate matter <2.5 μm in diameter, nitrogen dioxide, tract‐level average population) incorporating a random intercept for tracts to account for clustering of participants within tracts. We ran Cox proportional hazards regression models to measure the association between transportation noise and stroke recurrence and mortality, confirming the proportional hazards assumptions using Schoenfeld residuals. We used a similar incremental modeling strategy for recurrence and mortality outcomes as described above, with Models 3 and 4 additionally adjusting for stroke severity (log‐transformed NIHSS scores plus 1) and Model 4 including a frailty term for census tract to account for shared risk within tracts. Of note, our models did not include clinical comorbidities (atrial fibrillation, coronary artery disease, diabetes, hypertension, high cholesterol) as covariates for adjustment. These conditions are not expected to be confounders of the association between transportation noise and poststroke outcomes because they are not determinants of residential transportation noise exposure. Additionally, studies show that these conditions may be on the causal pathway between transportation noise and stroke outcomes. ¹³ , ²⁴ , ²⁵ Including them as covariates in our models may lead to overadjustment and attenuation of the total effect.

All models used a linear term for transportation noise because we found no evidence of nonlinearity when using restricted cubic spline terms with 3 knots. We scaled our results to an IQR of transportation noise to reflect a comparison between higher (75th percentile) versus lower (25th percentile) levels of transportation noise.

We conducted several sensitivity analyses to assess the robustness of our findings. First, to examine the potential impact of the floor effect of the DOT noise model on our findings, we performed the following sensitivity analyses: (1) we redefined transportation noise as a binary indicator (<45 versus ≥45 dBA) and (2) we restricted our analyses to those with noise levels ≥45 dBA and modeled noise continuously. To assess the validity of our choice of the exposure year assignment, we conducted a sensitivity analysis in which only 2016 noise estimates were assigned to all participants. The DOT also produced noise estimates for 2020, but these were not used because transportation activity during the coronavirus disease 2019 pandemic was likely substantially reduced and therefore may not represent typical ambient noise conditions for subsequent years. However, to assess for the short‐term impact of the pandemic on the exposure levels, we conducted a sensitivity analysis by linking stroke events occurring in 2020 to 2021 to the mean of 2018 and 2020 (road and aviation) transportation noise levels, while retaining 2018 levels for strokes occurring in 2022 and 2023 to reflect noise levels more in line with the prepandemic traffic patterns. To account for death as a potential competing risk in the association between transportation noise and stroke recurrence, we conducted a sensitivity analysis using Fine–Gray models (more details in Data S1). ²⁶ , ²⁷ We also conducted sensitivity analyses excluding early stroke recurrence cases (within 30 and 90 days) to reduce the potential impact of acute care confounding on our estimate. Lastly, we explored models allowing tract‐specific random slopes for transportation noise as a sensitivity analysis; however, the estimated between‐tract variability in the noise slope was negligible.

Out of 2504 patients with incident strokes between 2016 and 2023, those with missing NIHSS score (N=33), tract‐level covariates (N=28), or missing geocode data (N=3) were excluded from the analysis. Missing data for individual‐level covariates (36% for education and 1% smoking) were handled through multiple imputation with chained equations, assuming that the data were missing at random. Thirty data sets were created using 5 iterations per data set. The imputation model included all variables used in the analytic models, including demographic, behavioral, clinical, environmental, and outcome variables, as well as auxiliary variables associated with missingness. Each data set was analyzed separately, and the results were pooled using Rubin's rules to obtain combined estimates, and CIs using the mice R package. All analysis was conducted using SAS, Version 9.4, and R Studio, Version 4.4.0.

RESULTS

A total of 2441 patients with incident ischemic strokes were included in the analysis. The average age of patients was 68.4±12.6 years, 47.5% of whom were female, and 58.7% were Mexican American. Median initial NIHSS score was 3.0 (IQR, 1.0–7.0) (Table 1). Median transportation noise exposure was 45.5 dBA (IQR, 0.0–56.5) (Figure S1). Individual‐level patient characteristics did not differ by transportation noise levels. However, patients in higher transportation noise areas lived in census tracts with lower median household income ($49.7K versus $65.5K; P<0.01) and were exposed to higher levels of air pollution, including airborne particulate matter <2.5 μm in diameter (8.4 versus 8.0 μg/m³; P < 0.01) and nitrogen dioxide (6.2 versus 4.7 parts per billion; P<0.01). During follow‐up, 176 patients experienced a recurrent stroke, and 818 patients died. Median follow‐up for stroke recurrence was 2.2 years (IQR, 0.7–4.3), and the median follow‐up for all‐cause mortality was 2.5 years (IQR, 0.8–4.6).

Table 1.

Characteristics of Patients With Incident Ischemic Strokes by Transportation Noise Exposure, Nueces County, Texas, 2016 to 2023

	Missing	Overall	High transportation noise	Low transportation noise	P value
No.		2441	1221	1220
Age, y, mean±SD		68.4±12.6	68.0±12.7	68.8±12.5	0.15
Female sex		1160 (47.5)	594 (48.7)	566 (46.4)	0.26
Race or ethnicity					0.52
Non‐Hispanic White		985 (40.4)	488 (40.0)	497 (40.7)
Mexican American		1433 (58.7)	724 (59.3)	709 (58.1)
Other*		23 (0.9)	9 (0.7)	14 (1.2)
Uninsured		376 (15.4)	204 (16.7)	172 (14.1)	0.07
Education	820				0.26
<High school		470 (29.0)	240 (29.7)	230 (28.3)
High school		496 (30.6)	258 (31.9)	238 (29.3)
>High school		655 (40.4)	311 (38.4)	344 (42.4)
Health behaviors
Current smoking	21	591 (24.4)	316 (26.1)	275 (22.7)	0.13
Excessive alcohol use		150 (6.2)	75 (6.1)	75 (6.2)	0.99
Clinical comorbidities
Atrial fibrillation		324 (13.3)	159 (13.0)	165 (13.5)	0.71
Coronary artery disease		552 (22.6)	267 (21.9)	285 (23.4)	0.38
Diabetes		1132 (46.4)	582 (47.7)	550 (45.1)	0.20
Hypertension	1	2017 (82.7)	1014 (83.1)	1003 (82.3)	0.62
High cholesterol		1190 (48.8)	601 (49.2)	589 (48.3)	0.64
Admission National Institutes of Health Stroke Scale score		3.0 (1.0, 7.0)	3.0 (1.0, 7.0)	3.0 (1.0, 7.0)	0.64
Tract‐level covariates
Median household income (in 1000)		54.9 (42.0, 75.2)	49.7 (40.3, 61.3)	65.5 (43.3, 82.7)	<0.01
Airborne particulate matter <2.5 μm in diameter, μg/m3		8.2±0.6	8.4±0.4	8.0 + 0.7	<0.01
Nitrogen dioxide, parts per billion		5.5±1.5	6.2±1.0	4.7±1.5	<0.01
Population density		4261 (3037, 5597)	4157 (3068, 5290)	4294 (3037, 5641)	0.55

Transportation noise levels categorized by the median (45.5 Aarea‐level decibels) maximum noise levels within 100‐meter buffer radius.

^*Other includes Black (Hispanic or non‐Hispanic), Asian (Hispanic/non‐Hispanic), non‐Hispanic Native American.

The association between transportation noise exposure and stroke severity was not statistically significant in the unadjusted or fully adjusted models (Table 2). We did, however, observe evidence that patients exposed to higher versus lower transportation noise levels had a 36% higher rate of stroke recurrence even after full covariate adjustment (hazard ratio [HR], 1.36 [95% CI, 1.01–1.82]; P=0.04). Transportation noise was not associated with mortality after covariate adjustment (HR, 1.08 [95% CI, 0.94–1.24]; P=0.26). Covariate estimates from the fully adjusted models are presented in Table S1. Model fit statistics (R² and pseudo R²) are provided in Table S2.

Table 2.

The Association Between Transportation Noise Exposure Within a 100‐m Buffer Radius and Stroke Outcomes, Nueces County, TX, 2016 to 2023

	Stroke severity*	Stroke recurrence†	Stroke mortality†
	% change (95% CI)	HR (95% CI)	HR (95% CI)
Model 1 (unadjusted)	0.95 (0.88–1.02)	1.35 (1.01–1.80)	1.05 (0.92–1.20)
Model 2	0.96 (0.89–1.03)	1.35 (1.00–1.78)	1.08 (0.95–1.23)
Model 3	0.96 (0.89–1.02)	1.33 (0.99–1.77)	1.09 (0.96–1.25)
Model 4 (fully adjusted)	0.95 (0.88–1.02)	1.36 (1.01–1.82)	1.08 (0.94–1.24)

Model 1: unadjusted. Model 2: adjusted for individual‐level demographic variables (age, sex, ethnicity, insurance, education). Model 3: Model 2+ individual‐level health behaviors (smoking, alcohol use)+log (NIHSS score +1) for stroke recurrence and mortality only. Model 4: Model 3+tract‐level characteristics (median income, PM_2.5, nitrogen dioxide, tract‐level average population). HR indicates, hazard ratio; and NIHSS National Institutes of Health Stroke Scale.

^*Stroke severity measured in log‐transformed NIHSS (NIHSS +1) score. Percent changes calculated between patients exposed to the 75th percentile and the 25th percentile of exposure.

^† Hazard ratios compare hazard rates for recurrence and mortality in patients exposed to 75th percentile of exposure and 25th percentile of exposure.

Results from the sensitivity analysis were consistent with the main findings. Modeling noise as a binary exposure (<45 dBA versus ≥45 dBA, Table S3) and restricting analyses to participants with measurable noise levels (≥45 dBA, Table S4) did not meaningfully change the association. Exposure assignment from the 2016 transportation noise data (Table S5) or an average of 2018 and 2020 transportation noise data to events occurring in 2020 and 2021 (Table S6) did not materially alter the results. The association between transportation noise and stroke recurrence was not statistically significant in the competing risk analysis but produced subdistribution HRs similar to our main findings (Table S7). The cumulative incidence functions for recurrence and mortality are presented in Figure S2. Similarly, after excluding early stroke recurrence cases, the effect estimates slightly attenuated but remained similar in magnitude (Table S8).

DISCUSSION

In this population‐based stroke surveillance study with physician‐validated stroke events, we found that residential exposure to transportation noise was not associated with stroke severity or all‐cause mortality. But higher exposure levels were associated with a higher rate of stroke recurrence. Research on transportation noise in relation to stroke epidemiology has historically been limited in the United States. This is reflected in the limited nationwide monitoring and the lack of federal guidelines addressing transportation noise as a potential health hazard. ²⁸ In contrast, European policies recognize transportation noise as a modifiable cardiovascular risk factor with consistent, science‐based monitoring and mitigation strategies in place. ²⁹ Generating evidence in US populations is therefore necessary to increase awareness and build a body of evidence supporting preventive policy changes.

A growing body of mechanistic evidence suggests that noise may affect the risk of stroke by causing sleep disturbances, annoyance, and emotional distress. ¹⁰ , ¹¹ These impairments lead to increased oxidative stress, high cortisol levels, inflammatory responses, and disorders of the circadian rhythm. ¹⁰ , ¹¹ Through these multiple biologic mechanisms, exposure to high levels of transportation noise may increase the risk of developing precursors to stroke. Observational studies support these mechanisms, reporting associations between high levels of transportation noise and hypertension, diabetes, and atrial fibrillation, ¹³ , ²⁴ , ²⁵ all of which are also known risk factors of stroke recurrence.

To our knowledge, no prior studies have examined the association between transportation noise and stroke recurrence. One study from Germany in patients with acute coronary syndrome reported estimates comparable to ours for the association between aircraft noise and recurrent nonfatal events (HR, 1.24 [95% CI, 0.97–1.58]) but found no association with stroke (HR, 0.74 [95% CI, 0.37–2.05]). ³⁰ However, this study differed from ours in its focus on patients with acute coronary syndrome and their focus on aircraft noise in relation to stroke. Our study expands this literature by assessing stroke recurrence in a population‐based cohort and assessing the impact of both road and aircraft noise within the US context. Notably, the magnitude of association for transportation noise and stroke recurrence in our study is comparable to the observed effect sizes in prior studies for well‐known modifiable risk factors, such as atrial fibrillation, diabetes, uncontrolled hypertension, and smoking. ⁵ , ³¹ , ³² , ³³ , ³⁴ Although more research on transportation noise is required to confirm the validity of our findings, the comparable effect sizes indicate the potential relevance of transportation noise in secondary stroke risk profile.

Two prior studies from Europe have demonstrated conflicting evidence, reporting a positive or null association between transportation noise and stroke severity. ³⁵ , ³⁶ Both studies used a categorical measure (severe strokes: NIHSS score >5) of stroke severity, whereas we used a continuous measure of stroke severity and observed a null association. Additionally, prior studies had overall higher stroke severity distributions (mean and median NIHSS score ~5.0) compared with our cohort (median NIHSS score 3.0). Differences in outcome definition and underlying severity distribution may influence the ability to detect associations. Use of a categorical threshold captures likelihood of experiencing moderate‐to‐severe strokes. In contrast, modeling NIHSS continuously assesses incremental changes across the full severity spectrum. Our findings align with the study from Sweden (odds ratio [OR], 0.97) reporting a null association ³⁶ and indicate that residential exposure to transportation noise is likely not associated with stroke severity. Studies have reported a positive association between transportation noise and poststroke mortality (HRs ranging from 1.01 to 1.15) ³⁷ , ³⁸ and cardiovascular disease mortality (HR, 1.02). ³⁹ Our findings, although not statistically significant, are consistent with prior evidence. We further contextualize this trend in a US‐based stroke surveillance study, whereas previous studies were conducted in community‐dwelling European populations.

The strengths of this study include its novel focus on examining poststroke outcomes in relation to transportation noise exposure, use of physician‐validated stroke events, and the representativeness of the population with 47.5% female and 58.7% Mexican American patients. However, this study has its limitations. Given the low event rate and the CIs close to the null, the estimate for the noise and stroke recurrence association should be interpreted as largely hypothesis‐generating. We assigned the exposure using data from 2016 and 2018, which may introduce measurement error and fail to capture the key windows of exposure. Prior research suggests that transportation noise is generally stable across time, partly mitigating this concern. ¹² Noise levels are also estimated at participant homes without accounting for mitigation of exposures by the indoor environment or time spent at other locations such as work. In addition, although we adjusted for several neighborhood‐level socioeconomic factors, residual confounding remains possible due to unmeasured social determinants of health (eg, housing quality, indoor insulation, psychosocial stress, access to primary health care) or other environmental stressors (eg, presence of water quality violations, access to parks, food deserts). Adjustment for air pollution variables was based on multiyear (2016–2020) averages of airborne particulate matter <2.5 μm in diameter and nitrogen dioxide, which may not fully capture short‐term variability in these measures arising from environmental events, also potentially resulting in residual confounding. Additionally, the DOT noise estimates were modeled, which may not perfectly capture this complex environmental exposure. Noise models for Nueces County in 2016 and 2018 also excluded estimates for railroad noise which may have underestimated the exposure in individuals living close to railroads. The final issue with the exposure assessment was that the DOT noise model has a floor effect wherein noise levels <45 dBA are recorded as 0 dBA leading to low variability in our exposure distribution. We performed sensitivity analyses to account for this limitation, the results of which were not remarkably different from our main findings. Lastly, Nueces County is a largely urban county with its unique transportation network and housing structures. Therefore, our results may not be generalizable to other geographic locations with very different soundscapes.

CONCLUSIONS

Transportation noise exposure was associated with an elevated risk of stroke recurrence but not initial stroke severity or mortality. These findings demonstrate a possible impact of transportation noise on stroke prognosis. Further research using larger, geographically diverse cohorts with a longer follow‐up is required to support our findings for stroke recurrence and explore the impact of transportation noise on other poststroke outcomes to support science‐based noise policy changes in the United States.

Sources of Funding

This study was supported by the National Institutes of Health (R01 NS038916 and U24NS107214).

Disclosures

Dr Adar reports grant funding from National Institutes of Health, Centers for Disease Control and Prevention, Health Effects Institute, and World Resources Institute. Dr Adar also receives consulting fees from the National Institutes of Health as a review committee member and from the Health Effects Institute as a study section member. Dr Morgenstern reports grant funding from the National Institutes of Health. Dr Lisabeth reports grant funding from the National Institutes of Health and consulting fees from Bayer HealthCare Pharmaceuticals Inc. All other authors declare no competing interests.

Supporting information

Data S1. Supplemental Methods

Tables S1–S8

Figures S1–S2

STROBE Statement—Checklist

Reference [40]