Residential road traffic noise and incidence of depression in Denmark: a nationwide cohort study

Summary

Background

Evidence linking transportation noise to depression is limited. We investigated associations between road traffic noise at the most (LdenMax) and least (LdenMin) exposed façades and depression risk in Denmark.

Methods

We established a nationwide cohort of 2,941,456 adults aged 25–65 years and for sensitivity analyses a nested questionnaire-based cohort, the Danish National Health Survey (DNHS, n = 138,987). During follow-up from 2000 to 2017, 413,818 persons developed depression, defined as first hospital contact or ≥2 redeemed antidepressant prescriptions within one year. Road traffic noise (LdenMax and LdenMin) was estimated at all addresses and aggregated as 5-year means. Hazard ratios (HRs) were estimated using Cox models adjusted for individual and area-level socioeconomic factors, air pollution (PM_2.5, NO₂), and green space. DNHS analyses were additionally adjusted for important lifestyle factors.

Findings

In fully adjusted models, HRs per interquartile range were 1.04 (95% confidence intervals 1.04, 1.05) for LdenMax and 1.07 (1.06, 1.07) for LdenMin. Splines indicated that regardless of LdenMax level, having ≥10 dB difference between LdenMax and LdenMin (reflecting the presence of a quieter side of the dwelling) was associated with lower risk. HRs were similar in DNHS with slight attenuations after lifestyle adjustment. Associations were higher among low socioeconomic groups, defined by low education, low income, and blue-collar work.

Interpretation

Long-term exposure to road traffic noise was associated with increased depression risk. Having a quiet façade mitigated this association, pointing to the importance of strategies prioritizing silent facades.

Funding

European Union and the Swiss State Secretariat for Education, Research and Innovation.

Research in context.

Evidence before this study

Exposure to environmental noise has been linked to increased risk of depression, potentially by disrupting restorative environments and triggering stress responses. We searched PubMed for peer-reviewed articles published until 01/03/2026 using the search terms: (traffic noise OR transportation noise OR road traffic noise OR road noise) AND (depression OR depressive). Studies were included if they assessed residential exposure to road traffic noise and its association with depression. While systematic reviews were consulted, the most recent were published in 2020, excluding several newer studies. We further restricted our inclusion to prospective cohort studies. Evidence from cohort studies generally suggests a higher risk of depression among those exposed to higher levels of road traffic noise. However, large nationwide studies with robust outcome definitions remain scarce, and evidence on the potential effects of noise exposure at the least exposed façade of residences is very limited.

Added value of this study

This nationwide cohort study of almost 3 million adults in Denmark, followed for up to 18 years, provides one of the most comprehensive assessments of residential exposure to road traffic noise and risk of incident depression. Using validated address-level noise estimates linked to nationwide registry data with near-complete follow-up, we examined the association between road traffic noise and depression, including potential effect modification by socioeconomic status. Finally, we provide novel evidence on the mitigating role of noise exposure at the least exposed façade, highlighting the importance of low indoor noise levels.

Implications of all the available evidence

Consistent with findings from other populations, out results support an association between residential road traffic noise and increased depression risk, reinforcing environmental noise as a public health concern. The observed mitigating effect of a quieter façade suggests that policies should prioritize reducing indoor noise levels by ensuring silent rooms, through urban planning and building design. Greater susceptibility observed among lower socioeconomic groups indicates that road traffic noise may contribute to mental health inequalities, highlighting the need for targeted and equitable mitigation strategies. Future research should confirm the protective role of quiet façades and further investigate socioeconomic differences in vulnerability.

Introduction

Depression is a rising contributor to global disability, affecting one in seven individuals in high-income countries over their lifetime.¹ Established risk factors include younger age, female sex, chronic physical illness, psychosocial adversity, and socioeconomic disadvantage.1, 2, 3 Environmental determinants of mental health are receiving increasing attention, with environmental noise emerging as a potential modifiable risk factor for depression.²

Proposed mechanistic pathways linking road traffic noise to depression include stress, annoyance, and impaired sleep quality and duration.^4,5 Experimental and translational studies have shown that transportation noise can elicit a stress response, through activation of the hypothalamic-pituitary-adrenal axis and the sympathetic nervous system, resulting in the release of stress hormones.⁵ Such neuroendocrine activation has been associated with systemic and cerebral oxidative stress and inflammation, paralleling pathways involved in depression-like phenotypes observed in animal studies.^5,6 Chronic activation of these mechanisms may impair mental health through maladaptive coping, unhealthy lifestyles, or direct biological effects.⁶ Sleep disturbances and reduced sleep duration, which are common consequences of nighttime noise exposure, are themselves recognized as risk factors for depression.^6,7

A limited number of prospective cohort studies have investigated road traffic noise as a risk factor for depression, with most, but not all, reporting an increased risk.8, 9, 10, 11, 12, 13, 14 However, substantial methodological heterogeneity exists, mainly in outcome definitions, which have included hospital-diagnosed depression,^8,12 self-reported depression,^10,13,14 registry-based antidepressant use,^9,14 or combinations thereof.^11,13 Study populations also vary in age, with some restricted to students and young adults,^8,10 and other older adults.^11,12 In addition, most studies are based on relatively small cohorts with <500 incident cases.^10,12, 13, 14 Lastly, most studies assessed exposure only at the most exposed façade (LdenMax), rather than the quieter facade of residential buildings (a proxy for bedroom exposure), despite widespread recognition that sleep disturbance constitutes a principal pathway through which traffic noise may adversely affect mental health outcomes.⁶ To date, only one study has examined road traffic noise at the least exposed façade (LdenMin) in relation to depression, and this study was conducted exclusively among children and young adults.⁸ It reported a hazard ratio (HR) of 1.02 (95% confidence interval (CI): 0.98, 1.06) per 10 dB higher LdenMin.⁸ In summary, these inconsistencies limit comparability across studies and highlight the need for large cohorts with standardized depression definitions and assessment of both LdenMax and LdenMin to clarify the association between road traffic noise and depression.

There is growing evidence that the mental health effects of environmental exposures are not evenly distributed across populations.¹⁵ A prospective study of 3300 participants reported that road traffic noise was associated with an increased risk of depression only among individuals with ≤13 years of education, while no consistent association was found among those with higher educational attainment.¹⁴ This pattern suggests potential social inequalities in vulnerability to environmental noise, with more disadvantaged groups possibly bearing a disproportionate burden.

We aimed to investigate the association between long-term residential exposure to road traffic noise at the most and least exposed façades and risk for depression in a nationwide Danish cohort of adults aged 25–65 years. Furthermore, we aimed to examine whether this association differed by socioeconomic status to assess potential social inequalities in susceptibility to road traffic noise.

Methods

Study population

All Danish residents can be followed through nationwide registers using a unique personal identification number. We established a nationwide cohort of all residents aged 25–65 years without a history of depression at cohort entry, defined as 1 January 2000 or age 25 years, whichever occurred last. Information on vital status, emigration, and residential history was obtained from the Danish Civil Registration System.¹⁶ Participants were followed until incident depression, death, emigration, retirement, age 65 years, or 31 December 2017, whichever came first.

A second study population was nested within the nationwide cohort and comprised participants in the Danish National Health Survey (DNHS), a nationally representative survey of adults randomly sampled from the Danish population and linked to nationwide registers through the unique personal identification number. The DNHS included self-reported information on lifestyle factors and depressive mood collected in 2010 and 2013.¹⁷ For individuals participating in both survey waves, only data from the 2010 survey were included to ensure each participant was represented once in the analysis. Participants aged 25–65 years entered follow-up on questionnaire date and were censored according to the same criteria as the nationwide cohort.

Depression

Incident depression was defined as the first hospital contact for depression or ≥2 antidepressant prescriptions redeemed within one year, using the date of first redemption as the event date. Hospital contacts were identified in the National Patient Registry¹⁸ using International Classification of Diseases (ICD) version 10 codes F32–F33 from 1994 onward, supplemented with ICD-8 codes (296.0, 296.2, 296.8, 298, 300.4, 311, 313.1) from 1977 to 1993 to exclude prevalent cases. Antidepressant use was ascertained from the Danish National Prescription Registry (since 1995) using ATC code N06A, excluding N06AX12 and Yentreve® (N06AX21).¹⁹

In sensitivity analyses, we restricted the outcome to i) prescriptions with indication code 168 (“depression”; mandatory reporting since 2004) together with hospital contacts, ii) prescriptions only, and iii) hospital contacts only. Furthermore, a cross-sectional analysis in DNHS examined prevalent self-reported depression, defined as responding “yes, very bothered” to the item: “Have you, within the past 14 days, been bothered by feeling down, depressed, or unhappy?”, with the response options: no; yes, slightly; yes, very bothered.

Road traffic noise

Geocodes and floor information for all Danish addresses were obtained from the Building and Housing Register. Road traffic noise was modelled using the validated Nordic Prediction Method.20, 21, 22 Noise exposure was calculated as the equivalent A-weighted sound pressure level for day, evening, and night periods (Lden), incorporating 5 dB and 10 dB penalties for evening and nighttime noise, respectively.²²

Detailed noise maps were generated for 1995, 2000, 2005, 2010, and 2015 using three-dimensional building polygons, roads, and terrain data.²³ Traffic intensity, vehicle compositions, speeds, and road type (motorways, rural highways, roads wider than 6 m, roads 3- to 6-m wide, and other roads) for all Danish road links obtained from a national database²³ together with information on state noise barriers were subsequently incorporated into the models before noise levels were estimated and linked to all addresses.²² Reflection loss from façades was set to 1 dB, and both first- and second-order reflections were included in the calculations. Urban areas, road surfaces, and major water bodies were assumed fully reflective, whereas other surfaces were considered absorbent. Screening effects from buildings, terrain, embankments, and noise barriers were also taken into account.²²

For each residential building unit (including floor), noise levels were estimated at the centre point of all facades. The most exposed (LdenMax) and least exposed (LdenMin) façades were identified for each address, with values < 35 dB truncated at 35 dB. Annual address-specific noise levels for 1995–2017 were derived by linear interpolation between model years, and 1- and 5-year time-weighted averages were calculated across all residential addresses during follow-up.

Covariates

Yearly individual-level covariates (2000–2017) were obtained from national registers²⁴ and included sex, age, cohabitation status (married/cohabiting, other), highest attained education (mandatory, secondary/vocational, medium/long), personal income (sex- and year-standardized quintiles), occupation (white-collar, blue-collar, unemployed), country of birth (Denmark, other), and children living at home (0, 1, ≥2). A time-varying indicator of financial stress during the previous five years was based on ≥50% reduction in personal and/or household income between consecutive years, household income below the Danish relative poverty threshold, or job loss. Area-level covariates included parish-level socioeconomic indicators (proportions of residents with low income, basic education only, and with manual labour).

For DNHS, questionnaire-based lifestyle information included smoking status (never, former, current), weekly alcohol intake (0, 1–7, >7 drinks), physical activity level (low, moderate, high), and dietary habits (unhealthy, average, healthy) based on multiple questions.^17,25

Environmental covariates included residential exposure to PM_2.5 and NO₂, estimated for all addresses using the validated DEHM-UBM-AirGIS modelling system, which combines regional (5.6 km resolution), urban background (1 km resolution), and local street-level contributions (address-level resolution).²⁶ Monthly averages (1995–2017) were summarized as 1- and 5-year time-weighted running averages for each participant.

Green space exposure was assessed using a land use map of Denmark (10 m resolution), from which the proportion of recreational areas, forests, open nature, private gardens, and agricultural areas within 150 m radius of each residence were calculated.

Annual occupation data (DISCO-88 codes; 1995–2017) were obtained from Statistics Denmark. Occupational noise exposure (dB) was estimated using a Danish job exposure matrix based on full-shift noise measurements (LAeq during working hours) and information on job code, sex, age, collar type, and year.²⁷ Unemployment periods <2 years were assigned 35 dB. Individual 5-year running averages were calculated and categorized as <70, 70–75, and ≥75 dB.

Statistical analyses

Cox proportional hazards models with age as the underlying time scale were used to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) for incident depression per interquartile range (IQR) increase in 1- and 5-year time-weighted mean LdenMax and LdenMin exposure, and across noise categories. Age was used as the underlying time scale to provide optimal control for confounding by age, given its strong association with the risk of depression. All exposures and covariates were modelled as time-varying variables.

The proportional hazards assumption was assessed using Schoenfeld residuals. Graphical inspection indicated no evidence of non-proportionality; however, the diagnostic plots cannot be presented due to Danish data protection regulations governing microdata. Consistent findings were observed in age-stratified HRs (Supplementary Fig. S1). The exposure-response relationships for LdenMax, LdenMin, and ΔLden (LdenMax minus LdenMin) were estimated using natural cubic splines with three degrees of freedom. Because visual inspection of the spline functions indicated a non-linear association between ΔLden and depression (Supplementary Fig. S2), ΔLden was modelled as a spline in the statistical models.

We conducted analyses using three sequential adjustment models. Model 1 included sex and calendar year (2-year categories). Model 2 further adjusted for individual-level covariates (education, income, cohabiting, country of origin, and children living at home), and area-level socioeconomic indicators (proportions with basic education only, low income, and manual labour). Model 3 additionally adjusted for 1- and 5-year average PM_2.5 and NO₂ exposure and the proportion of green space. For LdenMax, all three models were run with and without adjustment for ΔLden to account for having a quieter facade.

Potential effect modification of the associations between LdenMax or LdenMin and incident depression by sex, cohabiting, education, income, occupation, financial stress, occupational noise, PM_2.5, and green space were evaluated by introducing an interaction term in Model 3 and tested by Wald’s test to calculate a global p-value.

We explored the exposure–response relationship between ΔLden and depression risk stratified by categories of LdenMax (≤50, 50–55, 55–60, 60–65, and >65 dB) using natural cubic splines with two degrees of freedom.

In sensitivity analyses, HRs for incident depression were estimated in the questionnaire-based DNHS using Models 1–3 (with statistical weights for non-response),¹⁷ and a fourth model additionally adjusting Model 3 for lifestyle factors: smoking, alcohol intake, physical activity, and dietary habits. We also estimated odds ratios (ORs) for self-reported depressive mood at baseline using logistic regression.

A directed acyclic graph (DAG) was constructed based on existing literature to guide covariate selection and clarify the assumed causal structure underlying the association between road traffic noise and depression (Supplementary Fig. S3). Socioeconomic and demographic factors were considered potential confounders, whereas lifestyle factors could potentially act as confounders, mediators, or correlates of early depressive symptoms. We analysed the biasing potential of lifestyle by analysing the DNHS with and without lifestyle adjustment We also used the DNHS to inform external adjustment of HRs from the nationwide cohort analysis. Additionally, we calculated E-values indicating the strength of an association an unknown confounder would need to have with both the exposure and outcome in order to fully explain the observed HRs.

Pearson correlations and variance inflation factors (VIF) were used to evaluate correlation and multi-covariance conditional on the full set of covariates. Spearman correlation coefficients were calculated to assess pairwise associations between road traffic noise, air pollution, and green space in the nationwide cohort.

Individuals with missing exposure, outcome, or covariate information were excluded at cohort construction, resulting in a complete-case analytic sample. The nationwide cohort was based entirely on nationwide registers with mandatory reporting, resulting in minimal missing data.

All statistical analyses were performed using R version 4.4.2 and SAS 9.4 (SAS Institute Inc., Cary, NC, USA). Reporting followed the STROBE guideline (checklist provided in the Supplement).

Ethics approval

Under Danish law (Section 14 of the Act on Research Ethics Review of Health Research Projects and Health Data Science Research Projects), ethics committee approval is not required for registry-based studies and questionnaire studies that do not involve biological samples.

Role of the funding source

The funders had no role in the study design, data collection, analysis, interpretation of results, manuscript preparation, or decision to submit the study for publication.

Results

Of 3,195,264 nationwide cohort participants, we excluded 219,386 with prevalent depression and 34,422 with missing covariates, leaving 2,941,456 participants (Supplementary Fig. S4). During a mean follow-up of 12 years, 413,818 participants developed depression. In DNHS (n = 176,856), exclusion of 27,993 with prevalent depression and 9876 with missing covariates yielded a final sample of 138,987 participants, of whom 5032 developed depression during a mean follow-up of five years, ascertained through national register data (Supplementary Fig. S4).

In the nationwide cohort, participants exposed to LdenMax ≥55 dB were generally younger and more likely to live alone, have lower income, and be employed in blue-collar occupations compared with those exposed to <55 dB (Table 1). They also lived in areas with higher air pollution and less green space. Similar patterns were observed for LdenMin exposure categories (≥45 vs. <45 dB; Supplementary Table S1) and in the DNHS cohort (Supplementary Table S2). Lifestyle factors in the DNHS differed only modestly across LdenMax exposure groups, although current smokers and high physical activity were slightly more common among participants exposed to ≥55 dB.

Table 1.

Baseline sociodemographic characteristics of the Danish nationwide cohort according to exposure to road traffic noise at the most exposed façade.

Baseline characteristics	Road traffic noise at the most exposed facade, 5-year mean
	<55 dB	≥55 dB
	n = 1,229,293	n = 1,712,163
Individual-level registry information
Men, %	50.9	51.7
Age, mean (SD), years	44.2 (11.7)	38.3 (12.4)
Country of origin, %
Danish	99.0	98.7
Non-Danish	1.0	1.3
Cohabiting status, %
Married/cohabiting	77.9	63.6
Living alone	22.1	36.4
Children living at home (number), %
0	50.1	63.0
1	20.2	17.4
≥2	29.7	19.6
Highest attained education, %
Mandatory	30.2	28.3
Secondary or vocational	52.5	57.2
Medium or long education	17.3	14.5
Personal income (quintiles),a %
Q1 (lowest)	15.6	26.5
Q2–Q4	61.7	58.6
Q5 (highest)	22.7	14.9
Financial stress (yes), %	16.7	19.8
Occupational status, %
Unemployed	17.3	16.1
Blue collar	50.4	56.7
White collar	32.3	27.2
Occupational noise exposure, %
<70 dB	18.0	21.3
70–75 dB	32.8	30.8
≥75 dB	49.2	47.8
Area-level registry information
Area-level variables (mean ± SD)
% with only basic education	12.9 (3.5)	11.8 (3.9)
% with low income	4.2 (2.0)	5.2 (2.6)
% with manual labour	15.8 (3.7)	13.8 (4.2)
Address-level information
Noise, 5-y, mean (SD)
LdenMax (dB)	48.3 (4.9)	61.8 (4.4)
LdenMin (dB)	42.4 (4.8)	47.1 (6.3)
ΔLden (dB)	5.8 (4.3)	14.7 (7.0)
Green space (%, 150 m), mean (SD)	59.0 (16.9)	49.6 (24.2)
Air pollution, 5-y, mean (SD)
PM2.5 (μg/m3)	16.5 (1.3)	17.0 (1.9)
NO2 (μg/m3)	18.7 (5.6)	24.3 (9.2)

N, number of individuals; SD, standard deviation; dB, A-weighted decibel; LdenMax, the most exposed façade; LdenMin, the least exposed façade; ΔLden, the difference between LdenMax and LdenMin.

^aStandardized by calendar year and sex.

LdenMax was moderately correlated with LdenMin (ρ = 0.46) and ΔLden (ρ = 0.68), while LdenMin was inversely correlated with ΔLden (ρ = −0.26). VIFs for noise were low to moderate (≤2.76) in all models (Table 2 and Supplementary Table S3). Noise exposures showed weak-to-moderate correlations with air pollution and inverse correlations with green space (Supplementary Table S3). In models adjusting separately for PM_2.5 and NO₂, adjustment for NO₂ resulted in higher HRs compared with adjustment for PM_2.5 (Supplementary Table S4). Distributions of LdenMax, LdenMin, and ΔLden are shown in Supplementary Fig. S5.

Table 2.

Associations between 1- and 5-y mean residential exposure to road traffic noise per interquartile range and risk of depression in the Danish nationwide cohort (n = 2,941,456; n cases = 413,818).

Residential road traffic noise	IQR (dB)	Model 1 HR (95% CI)	Model 2 HR (95% CI)	Model 2 VIF	Model 3 HR (95% CI)	Model 3 VIF
LdenMax
1-year mean	12.3	1.04 (1.04, 1.05)	1.04 (1.04, 1.05)	1.09	1.03 (1.03, 1.04)	1.34
5-year mean	11.7	1.05 (1.05, 1.06)	1.05 (1.05, 1.06)	1.09	1.04 (1.04, 1.05)	1.35
LdenMax adjusted for ΔLden
1-year mean	12.3	1.05 (1.04, 1.05)	1.06 (1.05, 1.06)	2.09	1.08 (1.07, 1.09)	2.72
5-year mean	11.7	1.06 (1.06, 1.07)	1.07 (1.06, 1.08)	2.06	1.10 (1.10, 1.11)	2.76
LdenMin
1-year mean	9.2	1.02 (1.02, 1.03)	1.02 (1.02, 1.03)	1.10	1.05 (1.05, 1.06)	1.33
5-year mean	8.9	1.03 (1.03, 1.04)	1.03 (1.03, 1.04)	1.11	1.07 (1.06, 1.07)	1.37

HR, Hazard ratio; CI, confidence interval; IQR, interquartile range; dB, A-weighted decibel; VIF: Variance Inflation Factor.

Model 1: Age as underlying timescale and adjustment for sex and year.

Model 2: Model 1 with further adjustment for country of origin, cohabiting, children living at home, education, income, and area-level variables.

Model 3: Model 2 with further adjustment for green space and time-weighted 1- and 5-years running means of PM_2.5 and NO₂.

Both LdenMax and LdenMin were associated with a higher depression risk across all adjustment models (Table 2). Associations were generally stronger for 5-year than 1-year mean exposures. Adjustment for individual and area-level factors resulted in only minor changes in risk estimates, whereas further adjustment for air pollution and green space attenuated estimates for LdenMax but increased estimates for LdenMin. For 5-year mean LdenMax, the fully adjusted HR was 1.04 (95% CI: 1.04, 1.05) per IQR of 11.7 dB. After additional adjustment for ΔLden, this increased to 1.10 (95% CI: 1.10, 1.11). The exposure-response curve showed increasing HRs with higher LdenMax, with upper-range levelling-off that disappeared after ΔLden adjustment (Fig. 1A).

Fig. 1.

Natural cubic splines visualizing the associations between 5-year mean exposure of road traffic noise and risk of depression in the Danish nationwide cohort for (a) the most exposed façade (LdenMax, with (orange) and without (blue) adjustment for ΔLden) and (b) least exposed façade (LdenMin). The Figure shows hazard ratios and corresponding 95% confidence intervals adjusted for age, sex, year, country of origin, cohabiting, children living at home, education, income, area-level variables, green space, PM_2.5 and NO₂ (Model 3).

For 5-year mean LdenMin, the fully adjusted HR was 1.07 (95% CI: 1.06, 1.07) per IQR increase (8.9 dB). The exposure-response curve showed increasing HRs with higher noise levels, with a tendency of a levelling-off at high exposure levels (Fig. 1B). Similar patterns were observed when LdenMax and LdenMin were analysed categorically (Supplementary Table S5).

Several sensitivity analyses were performed. Associations per 10 dB increase in noise are shown in Supplementary Table S6. Restricting the depression definition to antidepressant prescriptions with a depression indication code and hospital contacts only resulted in minimal changes in HRs (Supplementary Table S7). Similarly, HRs were largely unchanged when depression was defined solely by prescriptions with HRs of 1.10 (95% CI: 1.09, 1.11) for LdenMax adjusted for ΔLden and 1.06 (95% CI: 1.06, 1.07) for LdenMin (Supplementary Table S8). Stronger associations were observed when restricting the outcome to hospital contacts only (1.22 (95% CI: 1.18, 1.25) and 1.16 (95% CI: 1.14, 1.18), respectively; Supplementary Table S9).

Exposure-response curves for ΔLden across five strata of 5-year mean LdenMax are shown in Fig. 2. Across all strata, no change in risk was observed at ΔLden levels below ∼10 dB, whereas at higher ΔLden levels the risk of depression gradually decreased with increasing ΔLden.

Fig. 2.

Natural cubic splines visualizing the associations between ΔLden of residential road traffic noise (i.e., the difference between the most and least exposed façade based on 5-year mean exposures in dB) and risk of depression in the Danish nationwide cohort exposed to <50, 50–55, 55–60, 60–65 and ≥65 dB at the most exposed façade. The Figure shows hazard ratios and corresponding 95% confidence intervals adjusted for age, sex, year, country of origin, cohabiting, children living at home, education, income, area-level variables, green space, PM_2.5 and NO₂ (Model 3).

Sensitivity analyses in the DNHS yielded results comparable to those in the nationwide cohort. In Model 3, HRs for depression were 1.04 (95% CI: 0.99, 1.10) for 5-year mean LdenMax, 1.11 (95% CI: 1.03, 1.19) for LdenMax with additional ΔLden adjustment, and 1.08 (95% CI: 1.03, 1.15) for LdenMin (Supplementary Table S10). Additional adjustment for lifestyle factors had little effect on HRs and VIFs (Supplementary Table S10) and external adjustment using DNSH lifestyle variables resulted in only minimal changes to the estimated HRs, particularly in models including ΔLden (Supplementary Table S11). For models including ΔLden, E-values for Model 3 were 1.37 and 1.43 for 1-year and 5-year averaging periods, respectively. For other models, E-values ranged from 1.21 to 1.34.

In cross-sectional analyses of self-reported depressive mood, ORs were higher than the corresponding HRs for incident depression, and adjustment for lifestyle factors only slightly attenuated the ORs (Supplementary Table S12).

HRs for 5-year mean LdenMax and LdenMin across subgroups defined by sex, cohabiting, socioeconomic status (SES) and residential factors for the nationwide cohort are shown in Fig. 3 (p-values in Supplementary Table S13). For both LdenMax and LdenMin, HRs were slightly higher among women than men. Also, participants with higher SES, defined by higher education and income, white-collar employment, no financial stress, and low occupational noise, generally had lower HRs than those with lower SES. The LdenMax–depression association was stronger among participants living with a partner than among those living alone, and for those with PM_2.5 below the median. For LdenMin, a stronger association with depression was observed among participants with green space above-median compared with those below-median.

Fig. 3.

Associations between 5-year mean exposure to residential road traffic noise (dB) at (a) the most and (b) least exposed façades and the hazard ratios (HRs) for depression (Model 3), stratified by sex, cohabiting, education, income, occupation, financial stress, occupational noise level, PM_2.5, and green space among individuals aged 25–65 years in the Danish nationwide cohort. All p-values were significant at a 5% level, except for financial stress for LdenMax, and cohabitating and PM_2.5 for LdenMin (Supplementary Table S12).

Using HRs and the exposure distribution of individuals across six exposure categories (Supplementary Table S14), the population attributable fraction (PAF) for LdenMax was estimated. Applied to the national annual number of incident depression cases in Denmark (2017), this corresponded to an estimated 1277 incident depression cases per year attributable to LdenMax exposure.

Discussion

In the nationwide cohort, residential exposure to LdenMax and LdenMin were both associated with higher risk of depression. Adjustment for ΔLden strengthened the association for LdenMax. Regardless of residential LdenMax levels, ΔLden levels above approximately 10 dB were associated with lower depression risk. The effect of road traffic noise was more strongly associated with risk of depression among lower socioeconomic groups, defined by low education, low income, and unemployment or blue-collar work. In the DNHS, HRs were comparable to those observed in the nationwide cohort and were only slightly attenuated after adjustment for lifestyle factors.

Previous prospective cohort studies on LdenMax and depression have generally reported positive associations, with risk estimates ranging from 0.93 to 1.08 per 10 dB for studies using registry-based definitions of depression (hospital registry, prescription registry, or both).^8,9,11,12 These results are broadly consistent with our HR of 1.04 per 10 dB before adjustment for ΔLden. The relatively modest effect size may partly reflect the absence of metrics capturing within-dwelling noise contrasts, such as having a quiet façade. In our study, adjustment for ΔLden increased the HR for LdenMax to 1.09 per 10 dB, and shifted the exposure-response relationship from plateauing at levels >55 dB to a monotonic linear association across the exposure span. This suggests that ΔLden captures an additional exposure dimension not reflected by LdenMax (or LdenMin) alone. As none of the previous studies accounted for ΔLden, they may have underestimated the full mental health burden of road traffic noise. Our findings suggest that people living in residences with a quiet façade are better protected against depression, potentially by providing a restorative indoor environment that promotes rest and sleep. Supporting this, having a quiet façade has been associated with lower self-reported noise annoyance, particularly when bedrooms face the quieter façade.²⁸ Also, bedroom orientation away from the nearest street has been associated with reduced self-reported sleep disturbance.²⁹ These findings suggest that bedroom noise through its effect on annoyance and sleep disturbance, likely drives prioritization of a quiet façade for bedrooms, indicating that sleep-related mechanisms may be particularly important in explaining the health effects of a quiet façade. Interestingly, among people living in dwellings with façade-level noise contrast (ΔLden) exceeding ∼10 dB, higher ΔLden was associated with lower depression risk in an exposure-response manner. This suggests that the observed benefit may not reflect the least exposed façade per se, but rather what it represents: a quieter indoor environment, particularly for rest and sleep, which may mitigate the adverse effects of outdoor noise exposure. Accordingly, noise mitigation strategies may play a critical role in maintaining low indoor noise levels. The consistency of this pattern across five LdenMax strata (<50 dB to ≥65 dB) further supports the benefits of a quieter indoor environment even at relatively low LdenMax levels. The absence of an observed change in risk at ΔLden levels below ∼10 dB across all strata may reflect limitations in exposure assessment, as estimates are based on outdoor noise levels and may not fully capture individual indoor exposure, which depends on factors not available in our data, including façade insulation, bedroom orientation, and other noise sources. Consequently, ΔLden may only begin to capture a discernible and meaningful contrast in exposure once differences exceed 10 dB.

The present study is the first mental health study to investigate effects of ΔLden. One previous cohort study from Finland examined LdenMin in relation to hospital-diagnosed depression and reported a HR of 1.02 (0.98–1.06) per 10 dB, lower than the corresponding estimate for LdenMax (1.05, 1.02–1.09).⁸ This contrasts our findings, where the association was stronger for LdenMin than for LdenMax (without ΔLden adjustment), suggesting that the quiet façade may better capture aspects of noise exposure relevant to depression, potentially by reflecting sound levels during restorative periods.^4,6 However, direct comparison between the two studies is limited. The Finnish study had a mean baseline age of 14.3 years and, with nine years of follow-up, only a small proportion reached age 25, which was the lower age limit in the present study. As depression onset in adolescence and early adulthood is more strongly influenced by neurodevelopmental processes, family and school environments,³⁰ the effects of traffic noise on depression risk may differ in this age group. Moreover, the Finnish study relied solely on hospital diagnoses of depression, which likely captures more severe cases, whereas the present study also included antidepressant prescriptions, thereby also capturing milder cases treated in primary care. Further studies using comprehensive exposure assessment including both LdenMax and LdenMin are needed to assess the importance of a quiet façade.

HRs were slightly higher for 5-year than 1-year mean exposure to LdenMax (adjusted for ΔLden) and LdenMin, suggesting that longer-term exposure may be more relevant for depression risk. Another study examining different LdenMax exposure windows reported similar HRs for 1- and 5-year exposures, consistent with our findings for LdenMax without ΔLden adjustment.¹¹ Future studies should further explore the relevance of different exposure time windows.

We observed that road traffic noise was associated with a lower risk of depression among participants with long education compared with those with short or medium education. In support, one study reported that LdenMax >50 dB was associated with higher risk among participants with ≤13 years of education (relative risk (RR): 1.43) but not among those with >13 years (RR: 0.92).¹⁴ Similarly, in our study, LdenMax was associated with higher depression risk among participants with low and medium income compared to high income, and among blue-collar compared to white-collar workers. Higher risk of depression among lower SES groups may reflect reduced psychosocial resources and resilience, that confer protection against adverse mental health outcomes, such as depression. In addition, people with higher SES may be more likely to invest in residential noise-mitigating measures. High SES is also consistently associated with healthier lifestyle behaviours, stronger coping capacities, and more extensive social support networks,^1,3,31,32 all of which may mitigate emotional stress and are important determinants of mental health. Individuals with higher SES might therefore be better positioned to buffer the potential mental health effects of noise exposure.

A counterintuitive finding was the stronger association between noise exposure and depression among cohabiting individuals than among those living alone. This may reflect detection bias or differential healthcare utilization. While the literature generally indicates that individuals living alone are at higher risk of depression,¹ cohabiting individuals tend to have higher healthcare utilization³³ and may be more likely to have depressive symptoms recognized and treated, thereby increasing the likelihood of receiving a clinical diagnosis. Nevertheless, associations between noise exposure and depression were observed across both cohabitation groups, suggesting that differential help-seeking is unlikely to have substantially influenced our primary findings. Another unexpected finding was the stronger association among participants with higher levels of green space, particularly for LdenMin. This should be interpreted cautiously, as low green space often reflects greater urbanicity and thus higher noise levels and other urban environmental stressors. Consequently, the observed interaction may partly reflect spatial clustering of exposures rather than a true biological synergistic effect. This interpretation is consistent with Fig. 1B, where the depression risk appears to level off at higher noise levels, potentially contributing to the attenuated association observed at higher noise exposure in the interaction analysis.

A major strength of this study is its nationwide, population-based cohort of nearly three million adults followed for up to 18 years, minimizing the risk of selection bias and loss to follow-up. Another strength is the use of high-quality nationwide registry data, which enabled accurate assessment of residential history, individual- and area-level covariates, and a large number of depression cases. This likely makes our findings generalizable to other Western countries with comparable healthcare systems and sociocultural contexts. Lastly, residential road traffic noise and air pollution were estimated using validated, high spatial resolution models and high-quality input data.²⁶ A Nordic Prediction Method validation from 1997 observed a mean difference of 0.3 dB (SD 3.0 dB) between measured and estimated LdenMax across Nordic locations,²¹ and a subsequent validation of Nord2000 (the successor to the Nordic Prediction Method) reporting similarly small deviations.³⁴ These findings support the reliability of the noise exposure estimates used in the present study. Noise estimates at both the most and least exposed facades allowed a detailed assessment of the potential importance of the quiet façade, rarely examined in previous studies.

This study has some limitations. Some depression cases, particularly milder cases, may have remained undetected, as the main outcome was based on hospital and prescription registries. This is supported by a Danish validation study finding that these registers substantially underestimate the true prevalence of depression and primarily capture more severe cases.³⁵ Consistent with this, previous prospective studies based on self-reported depression have reported higher HRs for LdenMax (1.06–1.26)^10,13,14 than studies relying on registry-based definitions, suggesting that under-ascertained in register-based designs may bias estimates towards the null and that the true associations could be stronger than observed here. To explore this, we analysed self-reported depressive mood in the DNHS, although this measure was relatively crude, relying on a single questionnaire item. While ORs and HRs are not directly comparable, stronger associations were observed for LdenMax adjusted for ΔLden and for LdenMin when using self-reported depressive mood, supporting the possibility that register-based measures may under-ascertain milder cases, which may be more strongly related to road traffic noise exposure. Conversely, restricting the outcome to hospital contacts only resulted in higher HRs, suggesting that road traffic noise may also be associated with more severe depression, or that greater diagnostic specificity strengthens the observed associations.

Most antidepressants are prescribed for depression, although some are used for other indications. To assess whether this influenced our results, we restricted analyses to prescriptions with depression-specific identification codes. Results were consistent with the primary analyses, indicating that the observed associations were unlikely to be driven by antidepressants prescribed for other indications, such as anxiety disorders.

Road traffic noise has been associated with increased mortality, and death may therefore have acted as a competing event. However, the cohort consisted of adults aged 25–65 years, and incident depression was substantially more common than death during follow-up (14% vs. 3%), suggesting that competing risks are unlikely to have materially influenced the observed associations.

In the main nationwide cohort, information on lifestyle factors was unavailable. The appropriateness of adjusting for lifestyle is also uncertain, as these variables may function as confounders, mediators, and correlates of early depressive symptoms. In the DNHS, additional adjustment for lifestyle factors resulted in only minor changes in HRs. External adjustment analyses in the nationwide cohort using lifestyle distributions from the DNHS likewise suggested limited confounding potential by the lifestyle factors examined. E-values further indicated that an unmeasured confounder would need moderately strong associations with both noise exposure and depression to fully explain the observed associations in ΔLden models. Although residual confounding cannot be excluded due to the relatively crude categorization of lifestyle factors, the robustness of the estimates to lifestyle adjustment in the DNHS and in external adjustment analyses suggests limited potential for confounding or mediation by these factors. Finally, exposure misclassification may have influenced our results, as noise exposure was estimated using outdoor façade levels rather than indoor measurements. We also lacked information on bedroom orientation, sound insulation, and daily activity patterns. Such misclassification is likely to be largely non-differential with respect to the outcome; however, the direction of any resulting bias is uncertain.

Conclusion

Long-term residential exposure to road traffic noise was associated with a higher risk of depression. For LdenMax, the association strengthened after adjustment for noise contrast between facades (ΔLden), and a substantial façade contrast was associated with lower depression risk. This suggests that having a quiet façade may support restorative processes, underscoring the importance of urban planning and targeted noise mitigation strategies that ensure at least one side of the residence is characterized by low noise levels. Susceptibility to road traffic noise varied by SES, with higher depression risk observed among lower SES groups, indicating that road traffic noise may contribute to socioeconomic inequalities in mental health.

Contributors

AMK and MS conceptualized the study. AHP, LCT, ORN, MK, SSJ, and MS obtained the data, and AHP managed the data. AMK prepared the analytic datasets and performed the statistical analyses and visualization. MS supervised the study. AMK, AHP and MS had full access to the data, thus accessed and verified the underlying data. AMK and MS drafted the manuscript. All authors contributed to data interpretation, critically revised the manuscript for important intellectual content, and approved the final version. AMK and MS were responsible for the decision to submit the manuscript.

Data sharing statement

The study is based on data from the Danish national registers, which belong to the Danish Ministry of Health and Statistics Denmark. The authors are thus not allowed to share them in their raw form.

The use of AI and AI-assisted technologies in scientific writing

During the preparation of this manuscript, the authors used ChatGPT (version 5, OpenAI) to assist with grammar correction, language editing, and refinement of analytic code. The authors reviewed and edited all outputs generated by the tool and take full responsibility for the content of the manuscript.

Declaration of interests

All authors declare no competing interests and there was no funding source for this study.