Environmental Research 138 (2015) 372–380

Contents lists available at ScienceDirect

Environmental Research

http://d0013-93

n CorrTropicalFax: þ4

E-m

journal homepage: www.elsevier.com/locate/envres

The relationship between transportation noise exposure and ischemicheart disease: A meta-analysis

Danielle Vienneau a,b,n, Christian Schindler a,b, Laura Perez a,b, Nicole Probst-Hensch a,b,Martin Röösli a,b

a Department of Epidemiology and Public Health, Swiss Tropical and Public Health Institute, Socinstrasse 57, CH-4051 Basel, Switzerlandb University of Basel, Basel, Switzerland

a r t i c l e i n f o

Article history:Received 17 August 2014Received in revised form22 January 2015Accepted 21 February 2015

Keywords:Transport noiseExposureIschemic heart diseaseMyocardial infarctionMeta-analysis

x.doi.org/10.1016/j.envres.2015.02.02351/& 2015 Elsevier Inc. All rights reserved.

esponding author at: Department of Epidemioand Public Health Institute, Socinstrasse 57,161 284 8105.ail address: danielle.vienneau@unibas.ch (D. V

a b s t r a c t

Background: There is a growing body of evidence that exposure to transportation related noise canadversely affect health and wellbeing. More recently, research on cardiovascular disease has specificallyexplored the hypothesis that exposure to transportation noise increases the risk for ischemic heartdisease (IHD). Our objective was to review and conduct a meta-analysis to obtain an overall exposure–response association.Methods and results: We conducted a systematic review and retained published studies on incident casesof IHD using sources of transportation noise as exposure. Study-specific results were transformed intorisk estimates per 10 dB increase in exposure. Subsequently we conducted a random effects meta-ana-lysis to pool the estimates. We identified 10 studies on road and aircraft noise exposure conducted sincethe mid-1990s, providing a total of 12 risk estimates. Pooled relative risk for IHD was 1.06 (1.03–1.09) per10 dB increase in noise exposure with the linear exposure–response starting at 50 dB. Based on a smallnumber of studies, subgroup analyses were suggestive of higher risk for IHD for males compared tofemales (p¼0.14), and for persons over 65 years of age compared to under (p¼0.22). Air pollution ad-justment, explored only in a subset of four studies, did not substantially attenuate the association be-tween noise exposure and IHD.Conclusions: The evidence for an effect of transportation noise with IHD necessitates further researchinto the threshold and the shape of the exposure–response association, potential sources of hetero-geneity and effect modification. Research in different cultural contexts is also important to derive re-gional and local estimates for the contribution of transportation noise to the global burden of disease.

& 2015 Elsevier Inc. All rights reserved.

1. Introduction

Noise exposure from transportation, especially in urban areas,is one of the most widespread sources of environmental stress inthe daily lives. There is much evidence supporting the relationshipbetween exposure to environmental noise and wellbeing. Basneret al. (2013) and Munzel et al. (2014) provide a concise review ofthe effects of noise, including environmental noise, on health. Inaddition to causing sleep disturbance and psychological effectssuch as annoyance, noise is postulated to induce biological stresson the cardiovascular system, leading to changes in blood pressureand to cause hypertension (EC, 2002; Laszlo et al., 2012; vanKempen and Babisch, 2012; Hume et al., 2012; van Kempen et al.,

logy and Public Health, SwissCH-4051, Basel, Switzerland.

ienneau).

2002; Babisch, 2008). Most studies have investigated these andother non-auditory health effects of noise from road and aircrafttraffic (e.g. cognitive impairment in children (Clark and Stansfeld,2007) and diabetes in adults (Sørensen et al., 2013)), althoughnoise from railways is also a concern. For example, Croy et al.(2013) demonstrated experimentally that night-time freight trainnoise and vibration can accelerate heart rate during sleep, whichmay in turn be linked to cardiovascular disease (CVD). Althoughless studied, recent research on the potential relationship betweentransportation noise and ischemic heart disease (IHD) has yieldedinconsistent results (Selander et al., 2009; Beelen et al., 2009; Ganet al., 2012; Huss et al., 2010; Willich et al., 2006).

Babisch previously performed meta-analyses on studies of roadtraffic noise exposure and IHD. The first included five studies onincident myocardial infarction (MI) and reported a relative risk of1.17 (95% CI 0.87–1.57) per 10 dB increase in daytime (Lday) noise(Babisch, 2008). Recently updated, the study by Babisch (Babisch,2014) included 17 studies (incidence or prevalence) suggesting a

D. Vienneau et al. / Environmental Research 138 (2015) 372–380 373

relative risk of 1.08 (95% CI 1.04–1.13) per 10 dB increase (Ldn) inroad noise. Other transportation sources, however, were not con-sidered. In this paper, we also perform a meta-analysis on avail-able studies on the association between exposure to any trans-portation noise and IHD. We expanded the previous meta-analysis(Babisch, 2014) to aircraft noise by giving both a main effect esti-mate and estimates by source. Further, we systematically evaluatethe threshold and the shape of the exposure–response association,as well as potential sources of heterogeneity and effect modifica-tion. Recent noise exposure studies also include evaluation of co-exposure to air pollution, an important consideration given thatboth exposures derive from the same sources and are further bothassociated with CVD (Davies et al., 2009; Foraster et al., 2011;Tétreault et al., 2013).

2. Methods

2.1. Study selection and data extraction

We conducted a systematic review to identify papers usingroad, rail or aircraft noise as exposure and myocardial infarction(MI) or coronary heart disease (referred to here as ischemic heartdisease [IHD]) as outcomes (ICD10 codes I20-I25). We includedboth non-fatal and fatal incident cases. Studies on prevalence (i.e.cross-sectional) were excluded. The search was conducted inPubMed and EMBASE, for the 20 year period prior to January 2014and reference lists of relevant articles including the WHO burdenof diseases report (WHO, 2011) were screened. No geographicconstraints were defined, however the search was conducted inand limited to publications in the English language. The searchstrings are provided in Appendix A. Further inclusion criteriawere: eligible studies had to quantify the association betweenmodelled or measured exposure to the transportation noisesource, and myocardial infarction (MI) or ischemic heart disease(IHD) had to be in the title and/or abstract. Studies which quan-tified this relationship in dB, either categorically or by a lineartrend (i.e. increase in risk is constant per exposure interval), in-cluding a measure of precision (e.g. 95% confidence intervals),were retained. We conducted a double data extraction of retainedstudies. Data extraction included recording the risk estimates bynoise exposure categories (including reference level value), noisemetric (e.g. Lday, Lden – see Appendix B for description), noisesource, study population by sex, study design, and whether therisk estimate was adjusted for air pollution. Where available, riskestimates for specific subsets of the study population were alsoextracted (e.g. age and sex stratified, and for continuous years atthe same residential address [referred to as years in residence]).

2.2. Linear exposure–response (trend) estimation

Risk estimates from individual studies based on categoricalnoise exposures were transformed into a linear exposure–re-sponse (per 10 dB increase in Lden). For each study, a log-normalmodel in SAS was fitted to the data to estimate the mean level ofexposure in each of the exposure intervals. Model fit was based onthe proportion of person years or number of cases within differentexposure intervals for cohort and case-control studies, respec-tively. If necessary, the study specific interval means were con-verted to Lden using approximations from the literature:L16hþ2 dB; Ldnþ0.3 dB; and LAeq,24þ1.5 dB (EEA, 2010) .

Trend per 10 dB noise increment, zeroed for the study specificreference level, was estimated using generalised least squares(STATA glst). If the covariance matrix could not be specified, thevariance-weighted least squares (STATA vwls) method was used(Orsini et al., 2006). This approach was tested in a sensitivity

analysis with studies providing risk estimates both categoricallyand as a linear exposure–response. We further explored individualstudies for departure of the exposure–response from linearity byincluding a quadratic term for exposure level, and also performeda meta-analysis of the respective estimates. As a final check, weperformed a meta-regression of the original effect estimates onthe study's mean exposure level since a non-linear exposure–re-sponse relationship might only be seen when comparing acrossstudies. A positive (negative) association between study-specificestimates and mean exposure levels would be suggestive ofan exposure–response relationship with positive (negative)curvature.

2.3. Meta-analysis

Random effects meta-analysis (using STATA metan (Harriset al., 2008)) was conducted based on the risk estimates per 10 dBincrease in Lden noise for the individual studies. The percent totalvariance due to between-study heterogeneity was assessed withthe I2 statistic. Between strata heterogeneity was assessed on thebasis of the p value of a Chi square test. To specify the startingpoint for the pooled linear exposure–response association, wepooled the study specific reference values using the derived meta-analysis weights of each study.

We used the following effect estimates in the main analysis:non-fatal IHD, for studies reporting separate estimates for non-fatal vs. fatal cases; both sexes combined, for studies reportingmales, females and both. Subsequent stratified analyses were alsoconducted to explore potential sources of heterogeneity due tomethodological considerations, including: outcome definition(non-fatal vs. fatal and MI specific vs. unspecified IHD); study date(r2005 vs. 42005) because studies after 2005 explored potentialconfounding due to air pollution; type of transportation noisesource (road, rail, aircraft); study design (case control, cohort,small area study [i.e. individual health data aggregated on the levelof census areas with exposure assigned on this group level]);method of linear trend estimation (estimated from categorical vs.linear model in the original study); adjustment for air pollution(no, yes); and noise reference level. Reference levels were definedon the basis of the computed mid-point of the reference exposurecategory. For studies providing linear exposure–response esti-mates, we used the reference value reported in the text. If notspecified we assigned the category “no threshold.” We also ex-plored potential effect modification using stratified analyses, dueto: age (o65 years, Z65 years); sex; and years in residence (notspecified, 410 years).

The influence of individual risk estimates was assessed byperforming repeated meta-analyses with one study left out eachtime (further referred to as leave-one-out meta-analysis). The ef-fect of leaving out sets of studies on the basis of methodologicalfeatures was also explored (i.e. small area studies, North Americanstudies, those potentially over-adjusted, studies not adjusted forair pollution, and those not adjusted for smoking). Analyses wereconducted in STATA 12.

3. Results

3.1. Selected studies

We have identified 10 studies, conducted mainly in Europe,focussing on road and aircraft noise and incident cases of IHD(Selander et al., 2009; Beelen et al., 2009; Gan et al., 2012; Husset al., 2010; Sørensen et al., 2012; Hansell et al., 2013; Correia et al.,2013; Babisch et al., 1994, 1999, 2005). No studies specificallyinvestigating effects of railway traffic on IHD were found.

Table 1Summary of retained studies.

Location Citation Sourcea Noise data Originalmetric

Reference(Lden)

Studydesign

Method for line-ar ER estimationb

Sample sizec Outcomed Sex Age atbaseline

Available stratified risk estimatese

Air pollutionadjustment

Age Sex Years inresidence

Berlin I Babisch et al.(1994)

Road Map Lday 62 Casecontrol

Derived 243 MI Male 41–70 – – – 15

Berlin II Babisch et al.(1994)

Road Map Lday 62 Casecontrol

Derived 4035 MI Male 31–70 – – – 15

Caerphilly &Speedwell, UK

Babisch et al.(1999)

Road Map andmeasures

Lday 57 Cohort Derived �23700 IHD Male 45–63 – – – 15

Berlin III Babisch et al.(2005)

Road Model Lday 62 Casecontrol

Derived 4115 MI MaleþFemale 20–69 – – Yes 10

NL Beelen et al.(2009)

Road Model Max dB 50 Cohort Derived �1.2 mil IHD (f) Both 58–67 Yes NA NA –

Stockholm Selanderet al. (2009)

Road Model LAeq, 24 h 51.5 Casecontrol

Derived 3518; 2320 MI (b) Both 45–70 NA NA NA –

Switzerland Huss et al.(2010)

Aircraft Model Ldn 45.3 Cohort Derived �22.5 mil MI (f) Both 430 Yes Yes Yes 15

Denmark Sørensenet al. (2012)

Road Model Lden 42 Cohort Original �496,000 MI (b) Both 50–64 Yes Yes Yes 5

Vancouver Gan et al.(2012)

Community Model Lden None Cohort Original �1.8 mil IHD (f) Both 45–85 Yes Yes Yes –

London, UK Hansell et al.(2013)

Aircraft Model Lday 52 smallarea

Derived �20 mil MI (b) Both All ages Yes – – –

USA Correia et al.(2013)

Aircraft Model Ldn 45.3 Smallarea

Original �6 mil IHD Both 465þ Yes – – –

a Community noise refers to noise from both road and aircraft traffic.b ER¼exposure–response; derived¼ linear trend estimated, original¼original studies reported linear ER.c N persons for case control studies; Person-years for cohort studies and Hansell et al. (2013).d MI¼myocardial infarction; IHD¼ ischemic heart disease; Risk estimates are for non-fatal cases unless indicated in brackets, where f¼ fatal cases only, b¼ individual estimates available for fatal and non-fatal cases.e Dash (–) indicates not analysed; NA indicates analysed in original studies, but not available for inclusion in meta-analysis.

D.V

ienneauet

al./Environm

entalResearch

138(2015)

372–380

374

Fig. 1. Association between noise exposure (Lden) and IHD reported in original studies.Dot size is proportional to 95% CI for studies reporting categorical relative risks; noise level based on midpoint of respective exposure category. Dashed lines representstudies reporting linear trend.

D. Vienneau et al. / Environmental Research 138 (2015) 372–380 375

A summary of retained studies is provided in Table 1 and anoverview of risk estimates per exposure category is shown in Fig. 1.

Six of the studies (Selander et al., 2009; Beelen et al., 2009;Sørensen et al., 2012; Babisch et al., 1994, 1999, 2005) related toroad traffic noise while three large population studies focussed onaircraft noise: 65 civil airports and airfields in Switzerland (Husset al., 2010), Heathrow airport in London UK (Hansell et al., 2013)and 89 airports in the USA (Correia et al., 2013). The Vancouverstudy (Gan et al., 2012) looked at the effects of community noise,defined as noise from both road and aircraft traffic only. Half of thestudies investigated associations in a subset of persons living longterm at a single address. Earlier studies, those conducted prior to

Fig. 2. Forest plot of effect estimates per 10 dBA increase in

the end of 2005, did not specifically explore potential confoundingdue to air pollution. All of the more recent studies, however, in-cluded models adjusted for air pollution, and some provided ageand sex stratified risk estimates.

Two papers reported the results of the NaRoMi (BerlinIII); weselected the original paper by Babisch et al. (2005) instead ofWillich et al. (2006) because the latter excluded participants livingat smaller streets with low traffic intensities likely introducingselection bias. Babisch et al. (2005) included these individuals,who are likely exposed o60 dB, in the corresponding referencecategory. To avoid double counting, we also excluded the paper bySelander et al. (2012) which used the same population as in

transportation noise (Lden) and association with IHD.

Table 2Association between transportation noise exposure and IHD, including stratified analyses to explore potential heterogeneity due to methodological differences.

Subgroup Number of estimates Risk estimate per 10 dB (95% CI) Heterogeneitya

Between strata (p value) Between studies within strata (I2) (%)

None 12 1.06 (1.03–1.09) 28.3Outcome definition 0.92MI specific 7 1.06 (1.02–1.09) 0.0Unspecified IHD 5 1.05 (1.01–1.10) 62.1Disease stateb 0.49Non-fatal IHD 9 1.07 (1.05–1.09) 0.0Fatal IHD 6 1.05 (1.01–1.09) 54.1Study date 0.98r2005 4 1.06 (0.99–1.13) 0.042005 8 1.06 (1.03–1.09) 48.3Noise reference level 0.08No threshold 1 1.13 (1.06–1.21) –

o55 dBA (Lden) 6 1.05 (1.02–1.08) 38.0Z55 dBA (Lden) 5 1.04 (0.98–1.11) 0.0Type of Noisec 0.10Road 8 1.04 (1.00–1.10) 26.4Aircraft 3 1.06 (1.04–1.08) 0.0Community 1 1.13 (1.06–1.21) –

Study design 0.93Case control 5 1.06 (1.00–1.13) 0.0Cohort 5 1.05 (1.00–1.12) 67.7Small area 2 1.07 (1.04–1.09) 0.0Linear ER estimation 0.05Original (linear) 3 1.10 (1.05–1.15) 15.0Derived 9 1.04 (1.02–1.07) 19.4

a p value of the Chi square test used to assess between-strata heterogeneity; I2 statistic used to assess between-study heterogeneity.b Study N exceeds 12 because cause-specific estimates used where available.c Community noise refers to noise from both road and aircraft traffic.

Fig. 3. Influence plot of association between transportation noise exposure (Lden)and IHD. Leave-one-out meta-analysis (random effects) where study name in-dicates the left-out study.

D. Vienneau et al. / Environmental Research 138 (2015) 372–380376

Selander et al. (2009). The paper by de Kluizenaar et al. (2013) wasnot included because separate effect estimates for IHD were notavailable (effect estimates were reported for cerebrovascular dis-ease which included stroke). We further excluded cross-sectionalstudies such as Floud et al. (2013) and Banerjee et al. (2014).

The GLST command was used to estimate linear trend for allstudies except Babisch et al. (2005) where the covariance matrixwould not compute without exact number of cases and controls byexposure category. For this study, we therefore used VLWS.

3.2. Main effect estimates

All studies combined, regardless of outcome definition andnoise source, resulted in an overall risk estimate of 1.06 (95% CI:1.03–1.09) per 10 dB increase in noise exposure (Fig. 2 and Ta-ble 2). We calculated a pooled reference level of 50 dB as thestarting point for the linear exposure–response association.

Stratified by disease state, non-fatal IHD was 1.07 (1.05–1.09)compared to 1.05 (1.01–1.09) for risk of death. The difference inthese risk estimates was not statistically significant (betweenstrata p value¼0.49).

3.3. Potential sources of heterogeneity

The effect of individual studies on the main effect estimate wasassessed via leave-one-out meta-analysis (Fig. 3). No substantialimpact of a specific study was found. The highest estimate (1.07[1.05–1.08]) was found when Beelen et al. (2009) was omitted, andlowest estimate (1.05 [1.03–1.08]) when Gan et al. (2012) wasomitted. We further explored the impact of leaving out sets ofstudies with specific methodological features (Fig. 4). Neitheromitting the two small area studies (Hansell et al., 2013; Correiaet al., 2013) nor those studies which did not adjust for air pollution(Babisch et al., 1994, 1999, 2005) had a noticeable impact on themain estimate. However, dropping the three studies with potential

over adjustments (e.g. adjusted for high blood pressure or bothtraffic and air pollution; Beelen et al., 2009; Correia et al., 2013;Babisch et al., 2005), slightly increased the relative risk and re-duced the 95% confidence intervals; while dropping studies con-ducted in North America (Gan et al., 2012; Correia et al., 2013) orthose which did not adjust for smoking slightly reduced the re-lative risk (Gan et al., 2012; Huss et al., 2010; Hansell et al., 2013;Correia et al., 2013). In terms of other methodological issues, therewas no evidence of heterogeneity due to outcome definition, studydate, nor study design (Table 2). Studies that specifically reportedMI risk did not differ from those that reported unspecified IHD(1.06 [1.02–1.09] vs. 1.05 [1.01–1.10]; p¼0.92). As shown in Table 3,only four studies reported estimates for models which were ad-justed and not adjusted for air pollution. Although adjustment

Fig. 4. Sensitivity of the effect estimate when leaving out sets of studies based onmethodological differences.

D. Vienneau et al. / Environmental Research 138 (2015) 372–380 377

very slightly attenuated the risk, there was no difference betweenthese strata (p¼0.77).

The method used for trend estimation in individual studiesintroduced some heterogeneity (p¼0.05). The relative risk was�6% higher based on studies where linear trend was reported asopposed to derived using our method. We tested and found goodagreement for our linearisation approach in a sensitivity analysisof two studies publishing risk estimates both categorically and aslinear exposure–response (1.10 [0.97–1.25] vs. reported 1.12 [0.90–1.35] per 10 dB (Selander et al., 2009); 1.12 [1.01–1.24] vs. reported1.13 [1.06–1.21] (Gan et al., 2012)). More recent studies also tendedto report linear trend (Gan et al., 2012; Sørensen et al., 2012;Correia et al., 2013). We further found that a meta-estimate of thequadratic exposure term, introduced in models for individualstudies, was not statistically significant except for in Babisch et al.(1999) where we found indications for a stronger than linear effectof noise. Excluding Babisch et al. (1999) the meta-estimate of thequadratic terms was close to 0 and not statistically significantwhich generally supports our decision to treat the exposure–re-sponse for all studies as linear. The meta-regression of study-specific effect estimates on mean exposure levels provided aslightly negative but not statistically significant slope furthersupporting the concept of a linear exposure–response relationship.

Fewer studies were available for aircraft noise compared toroad noise (Table 2). The results seem to suggest some hetero-geneity between the groups (p¼0.1), however, this is mainly at-tributed to the study by Gan et al. (2012) in which the noisesources (i.e. from road and aircraft) were already combined as ameasure of community noise.

Table 3Association between transportation noise exposure and IHD restricted to studies with e

Subgroup Number of estimates Risk Estimate (per 10dB

Ageo65 years 3 1.04 (0.98–1.10)Z65 years 3 1.09 (1.03–1.16)SexMales 4 1.09 (1.04–1.13)Females 4 1.02 (0.95–1.10)Years in residenceNot specified (full data) 6 1.04 (1.00–1.08)410 years (subset) 6 1.08 (1.03–1.14)Air pollution adjustmentNo 4 1.06 (1.00–1.12)Yes 4 1.05 (1.00–1.11)

a p value of the Chi square test used to assess between-strata heterogeneity; I2 stat

We did not find strong indications that the reference level usedin the individual studies had an impact on the slope of the linearexposure–response association, although the between strata pvalue was relatively low due to one study which did not define athreshold (p¼0.08) (Gan et al., 2012).

3.4. Effect modification

Table 3 shows results for potential effect modification due toage, sex and years in residence. Three studies looked at IHD risk byage, showing indications of higher risk in the older age group (1.09[1.03–1.16] in 65 years and older vs. 1.04 [0.98–1.10] for under 65;p¼0.22). Relative risk for males (1.09 [1.04–1.13]) tended to begreater than for females (1.02 [0.95–1.10]) (p¼0.14). Persons re-siding longer term at the same address also tended to have higherrelative risks ([1.08 [1.03–1.14] resident for 10þ years vs. 1.04[1.00–1.08] years in residence not specified; p¼0.15). These re-sults, however, are only suggestive given that the 95% CIs betweenstrata overlap.

4. Discussion

4.1. Comparison with the literature

We found indications for a linear exposure–response associa-tion between IHD and transportation noise starting as low as50 dB and increasing by 1.06 (95% CI: 1.03–1.09) per 10 dB Lden.Babisch (2014) recently reported a relative risk of 1.08 (95% CI1.04–1.13) per 10 dB increase in Ldn, though with a slightly higherstarting point (�52.3 dB Lden) for the linear exposure–responseassociation. There are, however, several differences between ourtwo meta-analyses. Babisch focussed on road noise exposure onlywhereas we have considered all type of transportation noisesources. A further key difference between ours and the recentmeta-analysis by Babisch (2014) is that we focussed only on stu-dies addressing incident cases of IHD whereas Babisch also in-cluded prevalence studies. By excluding studies on prevalence weavoid potential bias which would be introduced if fatality is re-lated to the exposure. In other words, we may see fewer prevalentcases in areas with greater exposure simply because the fatal IHDcases are removed from the prevalence pool. Further differencesbetween the two meta-analyses include methodological details inthe derivation of study specific linear exposure–response re-lationships. Of note, we used air pollution adjusted effect esti-mates, whenever available in a study, whereas Babisch relied onunadjusted estimates. Despite all these differences our pooled

stimates in both strata.

A) Heterogeneitya

Between strata (p value) Between studies within strata (I2) (%)

0.220.0

39.20.14

0.033.2

0.150.00.0

0.7768.257.0

istic used to assess between-study heterogeneity.

D. Vienneau et al. / Environmental Research 138 (2015) 372–380378

estimates are relatively similar, with both our studies suggestingthat noise induced risk of IHD starts to increase at a lower levelthan previously presumed (Babisch, 2008).

4.2. Assumption of linear trend

Whether the exposure–response association between trans-portation noise and IHD is linear is not yet known. In principle acategorical meta-analysis would enable the evaluation of a non-linear relationship. This is the approach taken in the original meta-analysis by Babisch (2008) where the relationship between IHDand road traffic noise was represented by a polynomial function(OR¼1.63–0.000613 (Lday,16h)2þ0.00000736 (Lday,16h)3) with arather high threshold (55 dB LdayE57 dB Lden) (WHO, 2011). Ourdecision is to assume a linear relationship for the exposure–re-sponse is supported by the more recent studies included in ourmeta-analysis. Sørensen et al. (2012) for example, visually con-firmed a linear relationship between MI and road traffic noiseacross an exposure range of 42–84 dB Lden for a cohort in Den-mark (although depicted on a log scale, the relationship is linear).Furthermore, Sørensen et al. (2011) report a similar finding for theassociation between road noise and stroke incidence.

Nevertheless, as illustrated in Fig. 1, assuming a linear ex-posure–response relationship may be an over simplification insome studies, potentially masking a non-linear association in oneor both of the noise sources (Basner et al., 2013). For the twoaircraft studies, we see indications that the linear association witha higher threshold (between 55 and 60 dB, Lden) is likely appro-priate. The relevant threshold level and shape of the exposure–response association is less obvious for road traffic noise.

While the categorical approach used by Babisch (2008) canaccount for non-linear associations and help determine possiblethreshold effects, the disadvantage is that definition of exposurecategories in individual studies must be the same. The diversity ofexposure categories and noise metrics in our studies would haveimplied too many assumptions about choice of appropriate cate-gories. The linear approach is also less sensitive to differences inexposure modelling across studies if the slopes are consistent andoffset of the modelled value is the only concern.

Our decision to first estimate linear trend for each study wasbetter justified, and further supported by our analyses testing thisassumption. We also confirmed our general approach to trendestimation in two studies, showing good agreement (Selanderet al., 2009; Gan et al., 2012).

4.3. Effect modification

We found some suggestion that vulnerability differs across thepopulation. The number of studies in each stratum, however, wassmall and the differences were not statistically significant (95% CIsfor between strata comparisons were overlapping (Tables 2 and 3)).As a general trend, studies found men to be at greater risk fornoise-related IHD. The source of noise (road vs. aircraft) or theoutcome (MI vs. unspecified IHD) did not seem to play a role inthis respect. Nevertheless, in experimental short-term studieswith physiological parameters as endpoints, evidence regardinggender differences in susceptibility to transportation noise re-mains inconclusive (Davies and Kamp, 2012). In an experimentalsetting, exposure to railway noise during sleep was associatedwith somewhat stronger heart rate acceleration among mencompared to women (Croy et al., 2013). Indications for more apronounced effect of noise exposure on objective sleep quality wasalso found in men compared to women in Basel, Switzerland (Freiet al., 2014).

We also saw a tendency for higher risk in persons living longterm at the same address, in line with the hypothesis that chronic

transportation noise exposure is needed to induce IHD (Babischet al., 2005). An alternative explanation is that long term residencyimplies poorer insulated houses. Housing stock was not con-sidered in our included studies which were all based on availablenoise maps or modelling. Some support for this hypothesis,however, comes from Huss et al. (2010) who found higher noiseexposure and risk of death by MI for people residing in oldercompared to newer constructions or recently renovated buildings.We also found a tendency for higher relative risk in the older agegroup compared to those o65 years. Older persons may be morevulnerable to noise; but it is also possible that risk may be higherbecause they are likely to stay longer at the same address or live inolder buildings with less noise insulation.

4.4. Methodological issues

Several methodological issues are potential limitations to ourstudy. First is our combination of noise sources into a pooled riskestimate, which cannot be justified based on the few aircraft andlack of railway studies on IHD risk. In previous studies annoyanceresponse curves for aircraft were above that for road noise ex-posures (Miedema and Vos, 1998). It is less clear, however, whe-ther this also affects the linear relationship (i.e. slope) or whetherthis is only an offset issue. Further, self-reported annoyance israther different from objectively measured outcomes and thesefindings may not be transferable to disease risk. Recent researchindicates that while the effect of noise exposure on health relatedquality of life is mediated by annoyance and sleep disturbance(Héritier et al., 2014), the effect of noise exposure on objectivesleep outcomes (e.g. measured sleep efficiency) is also observed inpeople who are not annoyed by noise (Frei et al., 2014). Some ofthe more recent studies have investigated combined communitynoise (Gan et al., 2012) or confounding by other transportationnoise sources (Sørensen et al., 2012). Many factors influence thedecibels of noise (Leq), including acoustical characteristics of thesource as well as non-acoustical characteristics (e.g. environ-mental setting and timing of event). Our stratified results, how-ever, do not indicate heterogeneity between the studies on roadversus aircraft noise (Table 2).

Newer and older studies alike adjusted for typical confounders;common covariates across studies were age, sex (if included) andsocio-economic status. Some also adjusted for factors such asemployment status, occupational noise exposure, BMI, familyhistory, pre-existing comorbidities, and smoking status. Only Ba-bisch et al. (1999), however, included information for windowopening behaviour ̶ a practise which would directly influence thenoise exposure. Studies before year 2000 derived exposure on thebasis of gridded road noise maps, and may be subject to greaterexposure misclassification than recent studies which are typicallybased on more sophisticated receptor noise models to assign ex-posure at the home address. All studies conducted after 2005 alsoadjusted for air pollution exposure. Our results show only slight, ifany at all, attenuation of risk due to air pollution adjustment. Thissupports the recent findings from a systematic review suggestingminimal confounding by air pollution on the relationship betweencardiovascular disease and noise (Tétreault et al., 2013). There is,however, also recent evidence suggesting that the correlationsbetween these two exposures changes spatially (Foraster et al.,2011).

We explored the effect of additional methodological differencesthrough leave-one-out meta-analysis (Fig. 3) and subgroup meta-analysis (Table 2, Table 3 and Fig. 4). In general, not adjusting forair pollution had little impact on the results (Fig. 4). Althoughpotential over adjustment may be a problem. For example ad-justment for high blood pressure, as was done in Babisch et al.(2005), is not appropriate if this is on the causal pathway. Beelen

D. Vienneau et al. / Environmental Research 138 (2015) 372–380 379

et al. (2009) and Correia et al. (2013) are also potentially over-adjusted as both air pollution and traffic intensity or road densityare included as confounders and thus some of the noise effectsmay have been attributed to these variables. The influence plot ofindividual studies (Fig. 3) shows that the risk estimate slightlyincreased from 1.06 to 1.07 when Beelen et al. (2009) was re-moved. The individual effect of Babisch et al. (2005) and Correiaet al. (2013) however, was less pronounced. Although slightly at-tenuated, the effect of noise persisted after removing studies notadjusting for smoking, which were also the large population stu-dies (1.05 [1.02–1.09]).

Based on the original meta-analysis by Babisch (2008) the an-nual burden of environmental noise in Western Europe is an es-timated loss of 1 million healthy life-years (DALYS; disability ad-justed life-years) (WHO, 2011). Although this used a higher riskestimate than ours, we expect this is an underestimation of therisk due to the higher threshold (57 dB Lden vs. our 50 dB Lden).The polynomial exposure–response function used for the WHOburden of disease assessment crosses our exposure–responsefunction at a noise level of 71 dB Lden with a relative risk of ap-proximately 1.13. This implies that people exposed between 50and 71 dB have a higher risk for ischemic heart disease accordingto our meta-analysis than previously assumed in the WHO as-sessment. In Switzerland, 73% of the population live in areaswithin this exposure range for road traffic; (Höin et al., 2009) al-though the relative risk is small, this translates into a substantialnumber of additional DALYs. The specific cut-off used to define thethreshold may be context specific. Exposure assessment shouldtherefore include a careful assessment of the noise exposure, in-cluding recognition of the uncertainties in the data, especiallythose based on regulatory noise maps.

5. Conclusions

Using a linear exposure–response relationship between trans-portation noise and IHD, our meta-analysis supports a relative 6%increase in IHD per 10 dB Lden increase in exposure. Based on thereference levels of included studies, we suggest that the associa-tion starts as low as 50 dB. More studies are needed to furthersupport research into the shape of this relationship, threshold ofeffect and susceptibilities of at-risk populations. More studies onaircraft and rail, and studies from different cultural contexts arealso needed to derive regional and local estimates for the con-tribution of transportation noise to the global burden of disease.

Table B1

Indicator Description Period

Lday Average sound level over all the day periodsof a year

12 h or16 h

Ldn (Lday-night) Average sound level over all 24 h periods ofa year, with a penalty of 10 dB added for the8 night hours

24 h

Funding sources

This work was supported by the Swiss Bundesamt für Rau-mentwicklung/Federal Office for Spatial Development, Grant no.557301.

Lden (Lday-evening-night) Average sound level over all 24 h periods ofa year, with a penalty of 5 dB added for the4 evening hours penalty of 10 dB added forthe 8 night hours

24 h

Adapted from EEA 2010 Good Practice Guide (EEA, 2010).

Disclosures

Nothing to declare.

Appendix A. PubMed Search String

(“noise exposure” [Title/Abstract] OR “traffic noise” [Title/Ab-stract] OR “community noise” [Title/Abstract] OR “traffic noiseexposure” [Title/Abstract] OR “road traffic noise” [Text Word] OR“road noise” [Text Word] OR “rail traffic noise” [Text Word] OR“rail noise” [Text Word] OR “rail traffic noise” [Text Word] OR“railway noise” [Text Word] OR “air traffic noise”[Text Word] OR“aircraft noise” [Text Word]) AND

(“etiology”[MeSH Subheading] OR “etiology”[Title/Abstract] OR“etiological”[Title/Abstract] OR “epidemiologic studies”[MeSHTerms] OR “risk factors”[MeSH Terms] OR “case control study”[-Title/Abstract] OR “case-control” [Title/Abstract] OR “cohortstudy”[Title/Abstract] NOT “occupational” [Title/Abstract] NOT“industrial” [Title/Abstract]) AND

(“incidence” [Title/Abstract] OR “mortality” [Title/Abstract] OR“risk” [Title/Abstract] NOT “prevalence” [Title/Abstract]) AND

(“myocardial infarction” [Title/Abstract] OR “MI” [Title/Ab-stract] OR “ischemic heart disease” [Title/Abstract] OR “IHD” [Title/Abstract] OR “cardiovascular” [Title/Abstract] OR “coronary heartdisease” [Title/Abstract])

EMBASE Search String

(‘noise exposure’:ti,ab OR ‘traffic noise’:ti,ab OR ‘communitynoise’:ti,ab OR ‘traffic noise exposure’:ti,ab OR ‘road traffic noise’OR ‘road noise’ OR ‘rail traffic noise’ OR ‘rail noise’

OR ‘rail traffic noise’ OR ‘railway noise’ OR ‘air traffic noise’OR ‘aircraft noise’) AND (‘etiology’/exp OR ‘etiology’:ti,ab OR‘etiological’:ti,ab OR ‘epidemiology’/exp OR ‘risk factor’/expOR ‘case control study’:ti,ab OR ‘case-control’:ti,ab OR‘cohort study’:ti,ab NOT ‘occupational’:ti,ab NOT ‘industrial’:ti,ab) AND

(‘incidence’:ti,ab OR ‘mortality’:ti,ab OR ‘risk’:ti,ab NOT ‘pre-valence’:ti,ab) AND

(‘myocardial infarction’:ti,ab OR ‘MI’:ti,ab OR ‘ischemic heartdisease’:ti,ab OR ‘IHD’:ti,ab OR ‘cardiovascular’:ti,ab OR ‘coronaryheart disease’:ti,ab)

Appendix B. Noise indicators

D. Vienneau et al. / Environmental Research 138 (2015) 372–380380

References

Basner, M., Babisch, W., Davis, A., Brink, M., Clark, C., Janssen, S., Stansfeld, S., 2013.Auditory and non-auditory effects of noise on health. Lancet.

Munzel, T., Gori, T., Babisch, W., Basner, M., 2014. Cardiovascular effects of en-vironmental noise exposure. Eur. Heart J.

EC, 2002. Position Paper on Dose Response Relationships Between TransportationNoise and Annoyance. EU's Future Noise Policy, WG2 – Dose/Effect.

Laszlo, H.E., McRobie, E.S., Stansfeld, S.A., Hansell, A.L., 2012. Annoyance and otherreaction measures to changes in noise exposure — a review. Sci. Total Environ.435–436, 551–562.

van Kempen, E., Babisch, W., 2012. The quantitative relationship between roadtraffic noise and hypertension: A meta-analysis. J. Hypertens. 30, 1075–1086.

Hume, K.I., Brink, M., Basner, M., 2012. Effects of environmental noise on sleep.Noise Health 14, 297–302.

van Kempen, E.E., Kruize, H., Boshuizen, H.C., Ameling, C.B., Staatsen, B.A., de Hol-lander, A.E., 2002. The association between noise exposure and blood pressureand ischemic heart disease: a meta-analysis. Environ. Health Perspect. 110,307–317.

Babisch, W., 2008. Road traffic noise and cardiovascular risk. Noise Health 10,27–33.

Clark, C., Stansfeld, S.A., 2007. The effect of transportation noise on health andcognitive development: a review of recent evidence. Int. J. Compar. Psychol. 20,145–158.

Sørensen, M., Andersen, Z.J., Nordsborg, R.B., Becker, T., Tjonneland, A., Overvad, K.,Raaschou-Nielsen, O., 2013. Long-term exposure to road traffic noise and in-cident diabetes: a cohort study. Environ. Health Perspect. 121, 217–222.

Croy, I., Smith, M.G., Waye, K.P., 2013. Effects of train noise and vibration on humanheart rate during sleep: an experimental study. Br. Med. J. Open, 3.

Selander, J., Nilsson, M.E., Bluhm, G., Rosenlund, M., Lindqvist, M., Nise, G., Per-shagen, G., 2009. Long-term exposure to road traffic noise and myocardial in-farction. Epidemiology 20, 272–279.

Beelen, R., Hoek, G., Houthuijs, D., van den Brandt, P.A., Goldbohm, R.A., Fischer, P.,Schouten, L.J., Armstrong, B., Brunekreef, B., 2009. The joint association of airpollution and noise from road traffic with cardiovascular mortality in a cohortstudy. Occup. Environ. Med. 66, 243–250.

Gan, W.Q., Davies, H.W., Koehoorn, M., Brauer, M., 2012. Association of long-termexposure to community noise and traffic-related air pollution with coronaryheart disease mortality. Am. J. Epidemiol. 175, 898–906.

Huss, A., Spoerri, A., Egger, M., Röösli, M., 2010. Group ftSNCS. Aircraft noise, airpollution, and mortality frommyocardial infarction. Epidemiology 21, 829–836.

Willich, S.N., Wegscheider, K., Stallmann, M., Keil, T., 2006. Noise burden and therisk of myocardial infarction. Eur. Heart J. 27, 276–282.

Babisch, W., 2014. Updated exposure–response relationship between road trafficnoise and coronary heart diseases: a meta-analysis. Noise Health 16, 1–9.

Davies, H.W., Vlaanderen, J.J., Henderson, S.B., Brauer, M., 2009. Correlation be-tween co-exposures to noise and air pollution from traffic sources. Occup.Environ. Med. 66, 347–350.

Foraster, M., Deltell, A., Basagaña, X., Medina-Ramón, M., Aguilera, I., Bouso, L.,Grau, M., Phuleria, H.C., Rivera, M., Slama, R., Sunyer, J., Targa, J., Künzli, N., 2011.Local determinants of road traffic noise levels versus determinants of air pol-lution levels in a mediterranean city. Environ. Res. 111, 177–183.

Tétreault, L.-F., Perron, S., Smargiassi, A., 2013. Cardiovascular health, traffic-relatedair pollution and noise: are associations mutually confounded? A systematicreview. Int. J. Public Health, 1–18.

WHO, 2011. Burden of Disease from Environmental Noise: Quantification of HealthyLife Years Lost in Europe.

EEA, 2010. Good Practice Guide on Noise Exposure and Potential Health Effects.Orsini, N., Bellocco, R., Greenland, S., 2006. Generalized least squares for trend

estimation of summarized dose–response data. Stata J. 6, 18.Harris, R., Bradburn, M., Deeks, J., Harbord, R., Altman, D., Sterne, J., 2008. Metan:

fixed-and random-effects meta-analysis. Stata J. 8, 3.Sørensen, M., Andersen, Z.J., Nordsborg, R.B., Jensen, S.S., Lillelund, K.G., Beelen, R.,

Schmidt, E.B., Tjønneland, A., Overvad, K., Raaschou-Nielsen, O., 2012. Roadtraffic noise and incident myocardial infarction: a prospective cohort study.PLoS ONE 7, e39283.

Hansell, A.L., Blangiardo, M., Fortunato, L., Floud, S., de Hoogh, K., Fecht, D., Ghosh,R.E., Laszlo, H.E., Pearson, C., Beale, L., Beevers, S., Gulliver, J., Best, N., Ri-chardson, S., Elliott, P., 2013. Aircraft noise and cardiovascular disease nearheathrow airport in london: small area study. Br. Med. J., 347.

Correia, A.W., Peters, J.L., Levy, J.I., Melly, S., Dominici, F., 2013. Residential exposureto aircraft noise and hospital admissions for cardiovascular diseases: multi-airport retrospective study. Br. Med. J., 347.

Babisch, W., Beule, B., Schust, M., Kersten, N., Ising, H., 2005. Traffic noise and risk ofmyocardial infarction. Epidemiology 16, 33–40.

Babisch, W., Ising, H., Gallacher, J.E., Sweetnam, P.M., Elwood, P.C., 1999. Trafficnoise and cardiovascular risk: The caerphilly and speedwell studies, third phase– 10-year follow up. Arch. Environ. Health 54, 210–216.

Babisch, W., Ising, H., Kruppa, B., Wiens, D., 1994. The incidence of myocardial in-farction and its relation to road traffic noise – the berlin case-control studies.Environ. Int. 20, 469–474.

Selander, J., Bluhm, G., Nilsson, M., Hallqvist, J., Theorell, T., Willix, P., Pershagen, G.,2012. Joint effects of job strain and road-traffic and occupational noise onmyocardial infarction. Scand. J. Work Environ. Health.

de Kluizenaar, Y., van Lenthe, F.J., Visschedijk, A.J., Zandveld, P.Y., Miedema, H.M.,Mackenbach, J.P., 2013. Road traffic noise, air pollution components and car-diovascular events. Noise Health 15, 388–397.

Floud, S., Blangiardo, M., Clark, C., de Hoogh, K., Babisch, W., Houthuijs, D., Swart,W., Pershagen, G., Katsouyanni, K., Velonakis, M., 2013. Exposure to aircraft androad traffic noise and associations with heart disease and stroke in six eur-opean countries: a cross-sectional study. Environ. Health 12, 89.

Banerjee, D., Das, P.P., Foujdar, A., 2014. Association between road traffic noise andprevalence of coronary heart disease. Environ. Monit. Assess. 186, 2885–2893.

Sørensen, M., Hvidberg, M., Andersen, Z.J., Nordsborg, R.B., Lillelund, K.G., Jakobsen,J., Tjønneland, A., Overvad, K., Raaschou-Nielsen, O., 2011. Road traffic noise andstroke: a prospective cohort study. Eur. Heart J. 32, 737–744.

Davies, H., Kamp, I.V., 2012. Noise and cardiovascular disease: a review of the lit-erature 2008–2011. Noise Health 14, 287–291.

Frei, P., Mohler, E., Roosli, M., 2014. Effect of nocturnal road traffic noise exposureand annoyance on objective and subjective sleep quality. Int. J. Hyg. Environ.Health 217, 188–195.

Miedema, H.M.E., Vos, H., 1998. Exposure–response relationships for transportationnoise. J. Acoust. Soc. Am. 104, 3432–3445.

Héritier, H., Vienneau, D., Frei, P., Eze, I., Probst-Hensch, N., Brink, M., Röösli, M.,2014. The association between road traffic noise exposure, annoyance andhealth-related quality of life (HRQOL). Int. J. Environ. Res. Public Health 11,12652–12667.

Höin, R., Ingold, K., Köpfli, M., Minder, T., 2009. Sonbase – the GIS noise database ofSWITZERLAND. Technical bases. Environmental Studies no. 0908, pp. 61.