Indicators of residential traffic exposure: Modelled NOX, traffic proximity, and self-reported exposure in RHINE III

Accepted Manuscript Indicators of residential traffic exposure: Modelled NOx, traffic proximity, and selfreported exposure in RHINE III Hanne Krage Carlsen, Erik Bäck, Kristina Eneroth, Thorarinn Gislason, Mathias Holm, Christer Janson, Steen Solvang Jensen, Ane Johannessen, Marko Kaasik, Lars Modig, David Segersson, Torben Sigsgaard, Bertil Forsberg, David Olsson, Hans Orru PII:

S1352-2310(17)30520-4

DOI:

10.1016/j.atmosenv.2017.08.015

Reference:

AEA 15485

To appear in:

Atmospheric Environment

Received Date: 27 April 2017 Revised Date:

28 July 2017

Accepted Date: 6 August 2017

Please cite this article as: Carlsen, H.K., Bäck, E., Eneroth, K., Gislason, T., Holm, M., Janson, C., Jensen, S.S., Johannessen, A., Kaasik, M., Modig, L., Segersson, D., Sigsgaard, T., Forsberg, B., Olsson, D., Orru, H., Indicators of residential traffic exposure: Modelled NOx, traffic proximity, and selfreported exposure in RHINE III, Atmospheric Environment (2017), doi: 10.1016/j.atmosenv.2017.08.015. This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

ACCEPTED MANUSCRIPT Indicators of residential traffic exposure: modelled NOx, traffic proximity, and self-reported exposure in RHINE III

Authors

RI PT

Hanne Krage Carlsen1,2,3*, Erik Bäck4, Kristina Eneroth5, Thorarinn Gislason6,7, Mathias Holm8, Christer Janson9, Steen Solvang Jensen10, Ane Johannessen11, Marko Kaasik12, Lars Modig13, David Segersson14, Torben Sigsgaard15, Bertil Forsberg1, David Olsson1, Hans Orru1 ,16 Affiliations:

1) Department of Public Health and Clinical Medicine, Occupational and Environmental Medicine, Umeå University, Umeå, Sweden

SC

2) Engineering and Natural Sciences, University of Iceland, Reykjavik, Iceland

M AN U

3) Section of Occupational and Environmental Medicine, Department of Public Health and Community Medicine, Institute of Medicine, Sahlgrenska Academy at University of Gothenburg, Gothenburg, Sweden 4) Environment Administration, City of Gothenburg, Sweden

5) Environment and Health Administration, City of Stockholm, Sweden 6) Faculty of Medicine, University of Iceland, Reykjavik, Iceland

7) Department of Respiratory Medicine and Sleep, Landspitali National University Hospital of Iceland, Reykjavik, Iceland

TE D

8) Section of Occupational and Environmental Medicine, Department of Public Health and Community Medicine, Institute of Medicine, Sahlgrenska Academy at University of Gothenburg, Gothenburg, Sweden

EP

9) Department of Medical Sciences, Respiratory, Allergy and Sleep Research, Uppsala University, Uppsala, Sweden 10) Department of Environmental Science, Aarhus University, Roskilde, Denmark

AC C

11) Department of Global Public Health and Primary Care, University of Bergen, Bergen, Norway 12) Institute of Physics, University of Tartu, Tartu, Estonia 13) Division of Occupational and Environmental Medicine, Department of Public Health and Clinical Medicine, Umeå University, Umeå, Sweden 14) Swedish Meteorological and Hydrological Institute, Norrköping, Sweden 15) Department of Public Health, Aarhus University, Aarhus, Denmark 16) Department of Family Medicine and Public Health, University of Tartu, Tartu, Estonia

*Corresponding author: Hanne Krage Carlsen, 1

ACCEPTED MANUSCRIPT Department of Public Health and Clinical Medicine, Umeå University, Umeå 90187, Sweden

RI PT

Email: [email protected]

Highlights

SC

M AN U TE D EP

•

Self-reported and dispersion modelled, and traffic proximity exposures were weakly correlated in low-pollution Northern European cities Self-reported traffic noise had higher agreement with modelled exposure than self-reported traffic intensity in most locations

AC C

•

2

ACCEPTED MANUSCRIPT

Abstract

Keywords:

TE D

M AN U

SC

RI PT

Few studies have investigated associations between self-reported and modelled exposure to traffic pollution. The objective of this study was to examine correlations between self-reported traffic exposure and modelled (a) NOx and (b) traffic proximity in seven different northern European cities; Aarhus (Denmark), Bergen (Norway), Gothenburg, Umeå, and Uppsala (Sweden), Reykjavik (Iceland), and Tartu (Estonia). We analysed data from the RHINE III (Respiratory Health in Northern Europe, www.rhine.nu) cohorts of the seven study cities. Traffic proximity (distance to the nearest road with >10 000 vehicles per day) was calculated and vehicle exhaust (NOx) was modelled using dispersion models and land-use regression (LUR) data from 2011. Participants were asked a question about self-reported traffic intensity near bedroom window and another about traffic noise exposure at the residence. The data were analysed using rank correlation (Kendall’s tau) and inter-rater agreement (Cohen’s Kappa) between tertiles of modelled NOx and traffic proximity tertile and traffic proximity categories (0-150 metres (m), 150-200 m, >300 m) in each centre. Data on variables of interest were available for 50–99 % of study participants per each cohort. Mean modelled NOx levels were between 6.5–16.0 µg/m3; median traffic intensity was between 303–10,750 m in each centre. In each centre, 7.7–18.7 % of respondents reported exposure to high traffic intensity and 3.6–16.3 % of respondents reported high exposure to traffic noise. Self-reported residential traffic exposure had low or no correlation with modelled exposure and traffic proximity in all centres, although results were statistically significant (tau=0.057–0.305). Self-reported residential traffic noise correlated weakly (tau=0.090–0.255), with modelled exposure in all centres except Reykjavik. Modelled NOx had the highest correlations between self-reported and modelled traffic exposure in five of seven centres, traffic noise exposure had the highest correlation with traffic proximity in tertiles in three centres. Self-reported exposure to high traffic intensity and traffic noise at each participant’s residence had low or weak although statistically significant correlations with modelled vehicle exhaust pollution levels and traffic proximity.

AC C

EP

Traffic exposure, noise exposure, dispersion models, land-use regression models, NOx, cohort study

3

ACCEPTED MANUSCRIPT

1 Introduction

RI PT

Different indicators of traffic-related air pollution and noise exposure have been used in epidemiological studies of the long-term air pollution and noise health effects (lately reviewed by WHO, 2013; Recio et al., 2016). Earlier studies often used ecological comparisons between regions to ascertain health effects (e.g. Ferris and Anderson, 1962; Lunn et al., 1970, Rudnik et al., 1977 Dockery et al., 1993), whereas later studies have used measured community-wide air pollution levels (e.g. Filleul et al., 2005, Beelen et al., 2008, Krewski et al., 2009, Heinrich et al., 2013; Dehbi et al., 2016). As air pollution monitoring networks are often scarce, different proximity air pollution modelling, interpolation, kriging, and similar approximation techniques have been applied.

AC C

EP

TE D

M AN U

SC

Proximity models are the most basic approach to measure the proximity (often the distance) of a study participant’s residence to a pollution source (often a busy road). Geographical Information Systems (GIS) have been widely employed, often to estimate exposure in epidemiological studies (e.g. Brauer et al., 2003; Andersson et al., 2009, Pindus et al., 2015). Among more enhanced models, air pollution dispersion modelling has been frequently applied (e.g. Raaschou-Nielssen et al., 2012; Cesaroni et al., 2013). These models use a deterministic approach to describe physical and chemical processes than affect air pollution and incorporate e.g., traffic data, pollution point sources, and meteorological data to estimate pollution concentrations in grid cells or streets. Dispersion modelling has been combined with personal and regional monitoring of traffic-related pollutants such as NOx, or with regional, urban, and local (street canyon) models in hybrid models which may perform better (Zou et al., 2009). Instead of dispersion models, some recent large epidemiological studies, such as ESCAPE, have applied land-use regression models (LUR), which use measurements of pollution as dependent variables and land-use, traffic, demographic, and geographic characteristics as predictor variables (Beelen et al., 2013). In general, correlations between LUR and dispersion models have been relatively good (de Hoogh et al., 2014). Nevertheless, several epidemiological studies have also used self-reported exposure variables. Self-reported traffic exposure in terms of perceived traffic density close to one’s home address, the presence of many large vehicles (heavy traffic), or traffic congestion (traffic jams), have been associated with several respiratory health outcomes, such as asthma symptoms (Brunekreef et al., 2009; Vlaski et al., 2014), wheezing, rhinitis, and coughing (Kuehni et al., 2006), allergic respiratory complaints (Shirinde et al., 2015), and quality of sleep (Gislason et al., 2016). Noise exposure is associated with cardiovascular outcomes (Fritschi., 2011) and low birth weigh (Ristovska et al., 2014). Doubts remain regarding the validity of selfreported exposure measures, as they are prone to bias; sensitive individuals might over-report exposure (Oiamo et al., 2015; Persson et al., 2007) and personal factors confound the association (Riedel et al., 2014).. Several studies have also used annoyance as an indicator of traffic exposure; however, due to high variance between areas, it is not always considered to be the best proxy of air pollution exposure (Jacquemin et al., 2008). Nevertheless, the agreement between air pollution- and noise-annoyance ratings has been shown to be good (Shepherd et al., 2016), and both air pollution and noise annoyance have been related to NO2 exposure (Fernández-Somoano et al., 2015). Few studies have investigated agreement between measured/modelled and self-reported exposure, and different correlations have been reported (e.g. Cesaroni et al., 2008; Heinrich et al., 2005). In the Respiratory Health in Northern Europe cohort (RHINE, www.rhine.nu), self-reported air pollution exposure measurements, and GIS and dispersion models of NOx have been shown to be effective markers of local traffic pollution (Madsen et al., 2007). The aim of the current study was to 4

ACCEPTED MANUSCRIPT

AC C

EP

TE D

M AN U

SC

RI PT

investigate potential correlations between self-reported traffic exposure and both modelled NOx exposure and traffic proximity (distance to a busy road) in the seven RHINE centres, mostly midsized cities with low or moderate air pollution levels.

5

ACCEPTED MANUSCRIPT

2 Material and methods 2.1 Study population

SC

RI PT

The study population is from RHINE (www.rhine.nu), with a cohort from seven Northern European cities; Aarhus (Denmark), Bergen (Norway), Gothenburg, Umeå and Uppsala (Sweden), Reykjavik (Iceland), and Tartu (Estonia). These cities all participated in the European Community Respiratory Health Survey (ECRHS, www.ecrhs.org) over 1989–1992, and the study cohorts are described in detail elsewhere (Johannessen et al., 2014). The participating centres in the ECRHS sent out a screening questionnaire to a large random sample of adults born in 1945–1973 (25,000 individuals) and living in the study area. Approximately 10 years later all participants who had answered the screening questionnaire in the Northern European cities were invited to do the RHINE follow-up study questionnaire. RHINE III, the second follow-up survey, was undertaken over 2010–2012 (see Johannessen et al. 2014).

2.2 Self-reported exposure

Self-reported residential traffic exposure was measured using the following two questions in RHINE

•

•

M AN U

III:

“Is your bedroom window towards a nearby street (<20 m)?” o Possible replies constituted four levels: “No”; “Yes, a street with a little traffic”; “Yes, a street with a moderate level of traffic”; “Yes, a street with a lot of traffic” “Can you hear traffic noise in your bedroom?” o Possible replies constituted four levels: “Not at all”; “A little’”; “Much”, “Very much.”

2.3 Assignment and geocoding of addresses

AC C

EP

TE D

The study participants’ address at the time of the 2nd follow-up survey were obtained from population registries. In Aarhus, addresses were retrieved from the Central Population Registry (CPR) and geocoded using the Danish national address database. In Bergen, addresses were retrieved from the National Registry (Folkeregisteret) and geocoordinates obtained from Statistics Norway. For participants in Umeå and Uppsala, Statistics Sweden (SCB) retrieved home addresses from the population register using National Individual Registration numbers; addresses were matched with GIS coordinates by Lantmäteriet – the Swedish National Land Survey – for Umeå and Uppsala, and by SCB for Gothenburg. In Reykjavik, participants were sampled from Registers Iceland (Þjóðskrá) and address geocoordinates retrieved from Statistics Iceland. In Tartu, the RHINE III questionnaire was sent out according to each participant’s address in the Estonian Population Register 2011. In addition, the questionnaire asked participants about their actual address, which led to several of the addresses in the Tartu RHINE III cohort being corrected.

2.4 NOX exposure dispersion modelling In each centre the annual average concentrations of NOX were modelled using data collected in 2009–2011, depending on the centre (Supplementary Table A). All models had high resolution (40– 150 m) and used Gaussian dispersion distribution; however, several enhancements were applied. Later the air quality grid cell values were then matched with each participant’s residential geocoordinates for the year of participation.

6

ACCEPTED MANUSCRIPT

TE D

M AN U

SC

RI PT

For Aarhus, the Danish AirGIS exposure modelling system (Jensen et al. 2001; 2009b) was applied based on GIS and air quality models. The system calculates air quality at a street location as the sum of three contributions: (a) local air pollution from street traffic, calculated using the Operational Street Pollution Model (OSPM) from input data on traffic (intensity, vehicle types, travel speed), emission factors for the vehicle fleet based on COPERT emission model, street and building geometry, background air pollution concentrations, and meteorology. The OSPM takes into account the street canyon effect, and photochemistry, to estimate NO2 (Berkowicz, 2000); (b) urban background concentrations, calculated from a simplified area source dispersion formula that takes into account urban vehicle emission density, city dimensions (transport distance), and building heights (initial dispersion height) (Berkowicz et al. 2008); and (c) regional background concentrations, estimated from trends at rural monitoring stations and national vehicle emissions (Jensen, 1998). Input data for the AirGIS system were established from various sources and integrated into the dispersion model. A Danish national GIS road network with traffic data for the period 1960–2005 has been developed (Jensen et al. 2009a), and a database on emission factors for the Danish car fleet, with data on light- and heavy-duty vehicles, was built and entered into the emission module of OSPM. The national GIS database of building footprints was supplemented with building heights from the National Building and Dwelling Register, which provides street and building geometry of a given address. The AirGIS system automatically generates street configuration and traffic data for the OSPM, including street orientation, street width, building heights in wind sectors, and traffic intensity, vehicle type and speed based on GIS layers of addresses, road networks with traffic data and building footprints including building heights. The system has been applied in numerous epidemiological air pollution studies in Denmark (e.g. Raaschou-Nielsen et al. 2012; Sørensen et al. 2012) and validated in several studies (e.g. Berkowicz et al. 2008; Jensen et al. 2009b; Kakosimos et al. 2010, Ketzel et al., 2011).

AC C

EP

For Bergen and Reykjavík, NOx dispersion was described using Gaussian dispersion modelling in the air quality management system AirViro (SMHI, 2015). Bottom-up emission inventories had been compiled based on information acquired from the municipalities. The main road network, as well as emissions from ships, were included. The road and ship emission inventory is based on information from the local municipalities and port statistics from MARINE Traffic and EU Stat. Due to a lack of detailed local information regarding the composition of the vehicle fleet and driving conditions, it was assumed that these conditions were similar to the situation in Uppsala, which allowed the same emission factors to be applied. As in Reykjavík and Bergen, there were no large point sources present within the modelled domain in Uppsala, thus the contribution from point sources was disregarded. Validation of the spatial variations was carried out in both Bergen and Reykjavík, comparing modelled results with available measurements from campaigns with passive samples. Also it was the first time the model was used in these cities. For Gothenburg, annual means of NOX —provided by the Environment Administration of the City of Gothenburg—were modelled. The NOX models were made using both historical and current emission databases (EDBs), and calculations done using the Enviman AQPlanner (OPSIS AB, Furulund, Sweden) that consists of a Gaussian model AERMOD (US EPA). There were approximately 6,700 sources in the EDBs, most of which were road traffic (line) sources (approximately 5,900) and shipping (line) sources (approximately 100). Industry and larger energy and heat producers accounted for approximately 500 point sources; small-scale heating and construction machinery 7

ACCEPTED MANUSCRIPT emissions were considered as area sources. The model results were validated with data from air quality monitoring stations in Gothenburg (Molnár et al., 2015).

M AN U

SC

RI PT

For Tartu, the Gaussian plume model AEROPOL 5 that enabled the use of point, line, and area sources, taking into account both dry and wet deposition, was applied. Traffic emissions data were based on annual traffic flow measurements and modelling ordered by Tartu City Government, and the CAR-FMI emission coefficients (Karppinen et al., 2000; Härkönen et al., 2001). Domestic heating emissions and the heating patterns were obtained from earlier studies (e.g. Kaasik et al., 2007); industrial emissions were obtained from a registry of industrial point sources. Rush-hour traffic was measured in 2011 during the evening (16.30–17.30) at 33 sites; at the remaining 910 road segments, traffic modelling using CUBE software was performed. The AEROPOL model has previously been applied to air quality modelling of Tartu for other modelling studies and exercises (e.g. Orru et al., 2008). We then applied the porosity concept of Genikhovich et al. (2002) during post-processing of modelled ground-level concentrations: the area under buildings was excluded from the dispersion volume of each grid cell, thus the concentration was divided to the fraction of porosity, i.e. the proportion of non-built-up area (Kaasik et al., 2014). The street canyon effect was not applied, as continuous building fronts higher than or nearly equal to the width of the street seldom occur in Tartu. The computed annual average concentrations of NO2 were validated against passive sampling results available since 1997 (two-week measuring campaigns, four times a year) and data of an urban monitoring station available since 2007. As the AEROPOL model does not include chemical transformation, NO2 concentrations were derived from modelled NOX concentrations based on the monitoring ratios of Estonian urban areas. Each respondent’s address was geocoded in ArcGIS, and the centre of each respondent´s house linked to the corresponding grid cell. The concentration in grid cell was used as the exposure level of all respondents in that cell.

AC C

EP

TE D

For Umeå city and Uppsala, dispersion modelling was performed using the Gaussian model included in the Airviro air quality management system. In both cases, emission factors were determined from the Artemis database of the Swedish Transport Administration (Boulter et al. 2007). The Artemis model includes scenarios of the composition of various types of vehicles and fuels, such as the share of diesel cars, as well as the composition of the vehicle fleet in terms of European emission standards (Euro classification) for different years. For Umeå, other emissions were obtained from the SIMAIR 2005 (based on SMED emissions). For Uppsala, only emissions from road traffic were included in the current study, which was the dominant source of NOX in the region. The modelled annual mean concentrations were validated using NO2 passive sampling results—available at four different monitoring sites—as well as continuous measurements of NOX and NO2 at an urban monitoring station in central Stockholm. Dispersion model calculation was performed using the SMHI-Airviro Gaussian model, which has been validated in a number of studies (e.g. Johansson et al. 1999; Eneroth et al. 2006; Johansson et al. 2008; de Hoogh et al. 2014). For Umeå, in addition to the dispersion model, a land-use regression (LUR) model of NOX concentration was available for a large part of Västerbotten county, which was developed for the ESCAPE study (Cyrys et al., 2012; Beelen et al., 2013). As the Umeå cohort was sampled from the south-eastern part of Västerbotten county a large number of the participants lived outside the area covered by the dispersion model, and were assigned NOX exposure levels from the LUR model. Landuse regression utilized NOX concentrations as the dependent variable and variables such as traffic, topography, and other geographic variables as independent variables in a multivariate regression 8

ACCEPTED MANUSCRIPT model (Gilliland et al., 2005). Since the population density in the county varied considerably, the results were adjusted according to population density.

2.5 Traffic proximity modelling

RI PT

First, large roads were defined. Large roads constituted those with traffic of >10 000 vehicles per day, except in Umeå, where they constituted those with traffic of >8 000 vehicles per day due to no road having more than 10 000 vehicles per day at the time of the study. Traffic intensity (vehicles per day) was either measured in large number of receptor points or modelled based on small number of receptor points (see details in Supplementary Table A). Subsequently in all centers the large roads were mapped in GIS, and distances from the centre of nearest large road to the centre of each participant’s residential geocoordinates as centre of the bulding calculated.

2.6 Statistical methods

SC

Individuals who had answered basic demographic questions, self-reported exposure items and modelled exposure were included in the study. Within each study centre modelled NOX values were divided into tertiles based on frequency distributions using the 33rd and 67th percentiles (tertiles: lowest, medium, and highest values). Traffic proximity was recalculated into categories based on

M AN U

tertiles (closest, medium, and furthest) and categories (cut-offs: <150m, 150–300m, and >300m). A th

th

sensitivity analysis was performed with exposure distribution cutoffs at 50 and 90 percentile. The

AC C

EP

TE D

relationship between the different exposure metrics were first analysed using Kendall’s τ (tau) rank correlation. Rank correlations are used to study bivariate non-linear associations. Kendall’s tau (B) is a non-parametric statistical test of concordances and discordances of pairs of values adjusting for ties. It takes values from -1 to 1 corresponding to “perfect” concordance and discordance, and values near 0 indicate no significant concordance. Kendall’s tau is preferable in cases of a large sample size with many values of the same score (Newson et al., 2002). Agreement between the selfreported and modelled exposure metrics were compared using Cohen’s Kappa inter-rater agreement test with quadratic weights (Hallgren, 2012, Banerjee et al, 1999). The test assesses agreement between alternative methods of categorical assessment, and is interpreted as having poor agreement if the score is ≤0.2; fair for a score of 0.21–0.40; and very good for a score of 0.81– 1.00. Sensitivity analysis was performed for rural areas in Umeå, the only centre where part of the study population was rural. All statistics were performed using R (R Core Development Team, 2015) using the “irr” package (Gamer et al., 2012).

9

ACCEPTED MANUSCRIPT

3 Results 3.1 Descriptive statistics

RI PT

A total of 13,550 individuals participated in RHINE III. Of those, 10,708 (79%) had information about the variables of interest. The most common causes of dropout were that the address could not be geocoded or was outside the dispersion model area. In the individual centres participation varied between 50% and 99%; women were in the majority (53%, range 50%–60% among the centres), and the mean age was 52 years old (range 50–54 years old) (Table 1). Table 1. Characteristics of the study participants

M AN U

SC

Inclusion rate* % Female (%) Mean age N/n (years) RHINE 10,708/13,550 79 53 52 Aarhus 2,269/2,299 99 53 50 Bergen 1,737/2,364 73 50 52 Gothenburg 848/1,712 50 53 54 Reykjavik 1,681/1,942 87 54 53 Tartu 863/1,366 63 60 50 Umeå 1,610 /1,934 83 52 54 Uppsala 1,700/1,933 88 51 53 *Proportion of RHINE III participants who reported the variables of interest in the questionnaires

TE D

Average modelled NOX was 9.8 µg/m3 ranging from 6.5 in Uppsala to 16.0 µg/m3 in Gothenburg. Within-centre range NOX exposure also varied highly: a small range in Aarhus of 8.9–16.1 µg/m3 and a high range in Bergen of 2.1–63.8 µg/m3. Median distance to the nearest large road was from 303– 10,750 meters (cohort median 435 m); maximum distance varied from 2,170–331,000 m (Table 2). Table 2. Modelled NOX levels and distance to the nearest large road* NOX (µg/m3)

0.5 8.8 2.1 4.1 0.5 2.8 2.8 3.3

Mean

7.5 9.1 9.4 11.1 4.4 9.8 5.9 4.4

9.8 9.5 12.0 16.0 6.7 13.0 9.3 6.5

AC C

RHINE III Aarhus Bergen

33rd %ile

EP

min

Gothenburg

Reykjavik Tartu Umeå* Uppsala

Traffic proximity (m)*

67th %ile

Max

Min

33rd %ile

Median

10.5 10.3 12.6 19.1 8.2 14.5 9.3 7.6

63.8 16.1 63.8 54.9 24.0 54.3 46.8 24.2

1.0 1.0 1.3 9.0 3.3 11.9 7.1 1.0

248 281 256 232 145 198 693 353

435 435 394 391 345 303 10,750 733

67th %ile 771 1,006 589 581 3,108 429 32,123 1996

Max 331,000 34,310 6,987 2,170 4,571 2,513 331,000 52,370

*Large roads had traffic of >10,000 vehicles per day, except in Umeå, where the threshold was 8,000.

Self-reported traffic exposure was reported as low (no street within 20 m of participants’ bedroom window) by 56.3 % of participants (range 36.8–65.9 % among the individual centres). High traffic exposure (moderate or high traffic intensity within 20 m of participants’ bedroom window) was reported by an average of 11.0% of the total cohort (range 7.7–18.7 %). Traffic noise exposure was reported to be low (no traffic noise heard from the bedroom) by 45.7 % (range 19.2–58.6 %) of participants, and high by 6.8 % (range 3.6–16.3 %) of participants. The highest proportion of participants to report high traffic intensity and traffic noise occurred in Tartu. The lowest proportion 10

ACCEPTED MANUSCRIPT of participants to report high traffic exposure in terms of proximity occurred in Aarhus; the lowest proportion of participants to report high traffic noise exposure occurred in Uppsala (Figure 1).

Can you hear traffic noise in your bedroom? Uppsala Umeå

37.8%

50.2%

Tartu

44.8%

19.2%

64.4% 53.6%

Gothenborg

54.2%

40.4%

20%

M AN U

10%

5.3% 6.0%

51.1%

45.7% 0%

11.9%

56.7%

44.3%

RHINE

16.3%

34.5%

37.2%

Aarhus

5.0%

SC

Reykjavik

Bergen

3.6%

RI PT

58.6%

4.6%

45.2%

30%

Not at all

40%

50%

A little

60%

6.8%

70%

80%

90%

100%

Much/Very much

TE D

Is your bedroom window towards a nearby street (<20 m)? Uppsala

60.1%

Umeå

29.5%

58.8%

Tartu

28.8%

Reykjavik

36.8%

AC C

Gothenborg

Aarhus

RHINE

0%

No

10%

20%

18.7%

52.8%

51.7%

Bergen

10.4%

35.6%

61.2%

12.7%

28.3%

65.9%

10.5%

26.4%

56.3% 30%

12.4%

30.0%

EP

51.3%

10.4%

32.7% 40%

50%

Yes a street with little traffic

60%

70%

7.7% 11.0%

80%

90%

Yes a street with moderate/much traffic

Figure 1. Percentage distribution of self-reported exposure categories.

11

100%

ACCEPTED MANUSCRIPT Figure 2. Modelled concentrations of NOX within self-reported exposure categories: (a) residential traffic exposure (Is your bedroom window towards a nearby street (<20 m)?); and (b) traffic noise exposure (Can you hear traffic noise in your bedroom?).

EP AC C

b)

TE D

M AN U

SC

RI PT

a)

12

ACCEPTED MANUSCRIPT 3.2 Correlations

M AN U

SC

RI PT

Self-reported exposure to traffic near the bedroom window had low correlation (Kendall’s tau) with modelled NOX(0.161; p<0.05) with modelled NOX and measured traffic proximity (distance to nearest busy road) in the whole cohort; and also in each centre with tau ranging from 0.056–0.305 (all p<0.05) - lowest correlation was found in Reykjavík and highest in Uppsala (Table 3). Traffic proximity tertiles and categories (<150 m; 150–300m; >300m) correlated weakly with self-reported exposure; correlation coefficients were lower than for modeled NOX at 0.134 and 0.130, respectively (p<0.05 for both). Of all correlations, self-reported traffic exposure and modelled NOX was highest in five of seven centres; the exceptions were Bergen and Reykjavík, where traffic proximity tertiles and traffic proximity categories correlated better with NOX, respectively. Self-reported traffic noise levels from each participant’s bedroom window correlated weakly with modelled NOX, and traffic proximity tertiles and categories, among the whole cohort with taus of 0.157, 0.133, and 0.150, respectively (p<0.05 for all); these correlations were highest in Uppsala. No exposure metric correlated significantly with self-reported noise exposure in Reykjavik; traffic proximity tertiles did not correlate with self-reported noise exposure in Umeå (0.032; p>0.05) (Figure 2, Table 3). Selfreported traffic noise exposure had the highest correlation with NOX in Tartu and Umeå, with traffic proximity tertiles in Bergen and Uppsala, and with traffic proximity categories in Aarhus and Gothenburg. Comparing the correlation coefficients of both subjective outcomes amond the centres, traffic noise correlated highest with traffic proximity tertiles (in Bergen) and modelled NOX (in Tartu). The correlation between the two self-reported measures was moderate (Supplementary Table C), and a reanalysis with percentiles cut-off at 50th and 90th percentile showed very similar results (Supplementary table D). There was no association between modelled NOX levels and the correlation between subjective and objective measures (Supplementary figure A).

TE D

Table 3. Rank correlations (Kendall’s tau) between self-reported residential exposurea, and modelled NOX and traffic proximity Traffic near bedrooma NOX e

Traffic proximity c (tertiles )

EP

(tertiles)

Traffic proximity d (categories )

Traffic noiseb NOX (tertiles)

Traffic proximity c (tertiles )

Traffic proxity d (categories

0.161* 0.134* 0.130* 0.157* 0.133* 0.150* 0.109* 0.126* 0.142* 0.175* 0.183* 0.209* 0.106* 0.122* 0.078* 0.152* 0.183* 0.178* 0.181* 0.171* 0.132* 0.164* 0.169* 0.173* 0.057* 0.058* 0.053* -0.021 -0.006 -0.007 0.133* 0.104* 0.107* 0.167* 0.090* 0.104* 0.257* 0.099* 0.160* 0.209* 0.032 0.121* 0.305* 0.263* 0.227* 0.254* 0.255* 0.237* a b “Is your bedroom window near a street (less than 20 m)?” “Can you hear traffic noise from your bedroom?” cHighest tertile, lowest exposure. d Categories: 0–150 m; 150–300 m; >300 m. e Weighted mean. * p<0.05.

AC C

RHINE Aarhus Bergen Gothenburg Reykjavik Tartu Umeå Uppsala

3.3 Inter-rater agreement test Agreement between self-reported traffic exposure and the objective exposure metrics among the whole cohort was 0.107, 0.162 and 0.180 for NOx, and traffic proximity tertiles and categories, respectively (Table 4). Agreements varied between 0.069–0.274 (Reykjavík and Uppsala); all 13

ACCEPTED MANUSCRIPT agreements were statistically significant (p<0.05). Comparing agreements between self-reported traffic exposure and the objective exposure metrics, NOx had the highest agreement in four of seven centres; Gothenburg, Tartu, Umeå, and Uppsala); traffic proximity tertiles and categories correlated best with self-reported traffic exposure in Bergen and Aarhus, respectively. ; NOX and traffic proximity tertiles correlated joint-highest self-reported traffic exposure in Reykjavík.

AC C

EP

TE D

M AN U

SC

RI PT

Agreement between self-reported noise exposure and the objective exposure metrics among the whole cohort was 0.156, 0.148, and 0.168 for NOX, and traffic proximity tertiles and categories, respectively. In Reykjavík there were no statistically significant agreements; traffic proximity tertiles did not significantly agree with self-reported noise for Umeå (Supplementary Table B). Comparing agreements between self-reported noise and the different objective exposures, traffic proximity in categories had the best agreement in four of seven centres (Aarhus, Bergen, Gothenburg, and Uppsala); modelled NOX had better agreement with self-reported noise in Tartu and Umeå (Supplementary Table B). Comparing Cohen’s Kappa values for all outcomes across all centres, the best agreements occurred between self-reported noise exposure and traffic proximity tertiles in Aarhus and Bergen, and between the former and NOX in Tartu.

14

ACCEPTED MANUSCRIPT

Table 4. Agreement between self-reported residential exposurea and modelled NOx exposure and traffic proximity NOX (tertiles)

Traffic proximity (tertiles)

Traffic proximity (categories)

AC C

EP

TE D

M AN U

SC

RI PT

Highest exposure Kb (p) Highest exposure K (p) Less than 150 m K (p) Traffic intensity near a No Yes No Yes No Yes bedroom RHINE (n=10708) None or low 61.8% 27.3% 57.6% 31.4.4% 73.8% 15.2% Moderate or high 4.9% 6.1% 0.107*** 9.1% 1.9% 0.162*** 6.6% 4.3% 0.180*** Aarhus (n=2269) None or low 70.4% 21.9% 60.5% 31.8% 78.2% 14.1% Moderate or high 3.9% 3.7% 0.103*** 6.2% 1.5% 0.108*** 4.6% 3.1% 0.171*** Bergen (n=1737) None or low 58.4% 31.1% 58.0% 31.5% 74.5% 15.0% Moderate or high 8.1% 2.4% 0.100*** 8.7% 1.8% 0.112*** 7.2% 3.2% 0.093*** Gothenburg (n=848) None or low 61.2% 26.1% 55.9% 31.4 71.7% 15.6% Moderate or high 5.4% 7.3% 0.187*** 10.8% 2.0% 0.174*** 7.0% 5.8% 0.17*** Reykjavik (n=1681) None or low 61.6% 28.0% 58.8% 30.8% 60.7% 28.9% Moderate or high 5.1% 5.3% 0.069** 7.9% 2.6% 0.069** 4.8% 5.7% 0.064** Tartu (n=863) None or low 58.3% 23.1% 52.4% 29.0% 64.5% 16.8% Moderate or high 8.3% 10.3% 0.154*** 14.3% 4.4% 0.117*** 11.1% 7.5% 0.133*** Umeå (n=1610) None or low 61.9% 25.7% 57.6% 30.1% 80.7% 5.7% Moderate or high 4.7% 7.6% 0.238*** 9.1% 3.3% 0.102*** 8.9% 3.2% 0.184*** Uppsala (n=1700) None or low 63.7% 25.9% 57.1% 32.5% 78.4% 11.2% Moderate or high 2.9% 7.4% 0.274*** 9.5% 0.9% 0.240*** 5.9% 4.5% 0.270*** “Is your bedroom window towards a nearby street (<20 m)?” Possible replies constituted four levels: None:“No”; Low: “Yes, a street with a little a) traffic”; Moderate: “Yes, a street with a moderate level of traffic”; High: “Yes, a street with a lot of traffic” bCohen’s Kappa. *p<0.05, **p<0.01, ***p<0.001. 15

ACCEPTED MANUSCRIPT

4 Discussion

M AN U

SC

RI PT

In the current study, we found weak or no statistically although significant correlations and agreements between self-reported residential traffic exposure and modelled traffic pollution and GIS-derived traffic proximity in seven study centres in Northern European cities, a very diverse settings in terms of an investigation of this type. Self-reported traffic, and traffic noise exposure were reported as none or moderate by the majority of study participants, and reports of high exposure varied in frequency between the centres. The range of modelled NOX varied among the study cities, reflecting in part choices made by the modellers due to differences in the available raw data and different approaches, but also different urban environment characteristics, such as the degree of urbanisation and traffic density, the latter being reflected in the variability in traffic proximity. Modelled NOx correlated better than traffic proximity with self-reported traffic exposure except in Reykjavík and Bergen, where traffic proximity tertiles and category correlated better with self-reported exposure, respectively. However, in terms of traffic noise exposure, traffic proximity measurement correlated best, except in Tartu and Umeå. In our study, agreement and correlations of self-reported traffic exposure and self-reported traffic noise exposure to measured variables were similar, although traffic exposure tended to correlate better with modelled NOx exposure than did noise exposure.

4.1 Correlation and agreement between self-reported and modelled exposure in other studies

AC C

EP

TE D

Previous studies of self-reported traffic exposure and modelled outcomes have been set in less diverse settings, e.g. Cesaroni and colleagues (2008) found that Rome residents’ self-reported traffic exposure leves were significantly correlated with objectives measures from r=0.32– 0.49. However, exposure (mean NO2 was 45 µg/m3) was higher than in the current study, where the mean of NOX was 9.8 µg/m3. It is possible that self-reported exposure correlate better with objective data in areas with larger pollution gradients. Heinrich and colleagues (2005) compared rates of self-reported high traffic exposure within categories of modelled NO2 exposure, and in general found rather lowagreements, although they tended to be higher in urban compared to rural settings. In the current study, correlations and agreements between urban and rural areas were tested with Umeå data, the only centre where part of the study population was rural. We found no significant difference between urban and rural areas per se. In a study of modelled NO2 and self-reported odour annoyance, Oiamo and colleagues (2015) found significant correlations (Spearman’s rho of 0.22), but the correlation between self-reported noise and modelled NO2 was not significant (r=0.06). In that study, there were similar correlations of odour annoyance in low and high noise exposure areas (0.15 and 0.13, respectively), and the authors concluded that absolute exposure and noise sensitivity confounded the results.

4.2 Factors affecting dispersion modelling Dispersion models provide objective estimates of exposure to large roads or NOX, and are becoming standard in modelling exposure. However, models are only as good as the raw data upon which they are based, and while traffic sources were effectively characterized for most of our study centres, the databases varied in age, therefore interpolation was used to compensate for this. The contribution of NOX from other pollution sources, such as shipping and industrial sources, were considered in some centres meaning that self-reported traffic or traffic noise exposure underestimate actual exposure, and would provide a lower estimate than modelled NOX when there are high contributions 16

ACCEPTED MANUSCRIPT from marine traffic or industry. As eventual marine traffic and industry sources were not well characterized in all studied cities, such contributions may have contributed to differences between study centres. In some models, coarser grid resolution was applied outside of urban centres, so individuals who lived outside Uppsala city centre could have high exposure (as they in fact lived in Stockholm) which was less accurately assigned. Also different dispersion models are heterogenous which might cause regional heterogeneity across the included centers.

M AN U

SC

RI PT

The original cohort was selected from urban centres (except Umeå), areas covered by dispersion models (DMs), so that participants were lost during follow-up if they had moved outside the modelled area. This is a potential source of bias, as those who move and those who stay are likely to be different in a number of ways, including lifestyles, respiratory health, and susceptibility to bad air quality (Jie et al., 2013). In Umeå, the original participants were a mixture of urban and rural dwellers, and LUR was used in non-urban areas that the DM did not cover. In a validation study of the Umeå LUR and DM, the two models were highly correlated in terms of NO2 (Spearman’s rho of r=0.78; de Hoogh et al., 2014), and although NOX was not tested, it would likely also be highly correlated. Correlations and agreements between self-reported and modelled traffic exposure were lowest in Reykjavik. The NOX dispersion models of Bergen and Reykjavik were developed during this study, therefore these models had a smaller degree of validation than the other—older—models.

4.3 Factors affecting self-reported exposure

AC C

EP

TE D

Our results indicate that in some study settings, self-reported traffic noise exposure was a weakly correlated proxy of measured traffic exposure variables. This may have been because it entailed exposure to traffic regardless of factors not covered by the models, such as the floor level of the building or building characteristics, topography, or tunnels. In the current study, self-reporting of exposure was done using two questions, one for proximity of traffic to and the other for traffic noise exposure in their bedroom. Similarly, Heinrich and colleagues (2005) compared objective and subjective exposure in two centres, however, more questions were used. In one centre three questions were used to assess traffic exposure at one’s home address, including questions about traffic jams and truck traffic on weekdays. In the other centre road type was assessed (low corresponding to side streets with 30 km/hour speed limits) and additional questions were asked about traffic intensity and traffic jams. In the study of Persson and colleagues (2007), ten items were assessed on a six-point scale; in contrast, Cesaroni and colleagues (2008) used only one self-reported question of traffic exposure. Traffic and traffic noise exposure are related, but not exactly the same; however, few studies have studied their correlation and found it moderate (Fecht et al., 2016). In the current study, the two subjective measurements had a moderate level correlation, but it is noteworthy that Reykjavík, which had the lowest correlation and agreement between subjective and objective exposure metrics, also had the lowest correlation between the two self-reported exposure metrics of traffic proximity and noise exposure (Supplementary Table C). It is therefore possible that other local factors, for example landscapes features, building types, and or characteristics of the car fleet, contributed to the poor correlations of that city. Whereas modelled traffic proximity and NOX provide objective measures of exposure, self-reported exposure has the advantage of reporting exposure where the person actually lives, and is more sensitive to issues that models do not always account for, such as height above street level, building characteristics, whether a person sleeps with open windows, and or which way the window faces. Both outdoor and indoor PM2.5 was higher at street-facing buildings side than the rear (Sajani et al., 2016). Neither do address misclassification issues pose a threat to the validity of self-reported exposure levels. Although self-reported exposure is subjective, and likely to be biased in terms of an 17

ACCEPTED MANUSCRIPT

RI PT

individual’s habitual and sensory perception both on a within-region, and within-city level (e.g. relatively speaking busy street in a middle-sized Northern Europe city is very different from a busy street in an international megapolis), and sensitivities has been shown to be affected by character (anxiety levels) and indoor noise reporting (Persson et al., 2007). In addition, self-reported exposure might be affected by health status, e.g. in a study by Heinrich and colleagues (2005), individuals with hay fever or asthma clearly over-reported traffic exposure in German—but not in Dutch—urban areas. But subjective exposure might be still very important in the case of low air pollution levels as a recent study in Estonia (Orru et al., 2017) showed that perceived pollution and health risks might play an even more essential role in predicting environmentally induced symptoms and diseases than actual exposure. Understanding such modification effects would help the interpretation of future field studies.

5 Conclusion

AC C

EP

TE D

M AN U

SC

We observed significant but weak or no correlations between subjective and objective pollution exposure metrics in most cases in a multi-centre cohort study in low in low- or moderate air pollution exposure settings. Previous health effects studies from this cohort have utilized selfreported air pollution and noise pollution measures. Thus, the availability of the current study data offers an opportunity to compare health outcome associations between those metrics, although their correlation and agreement is low. In our study setting within-city exposure ranges were larger than between-city ranges, which is a motivation to conduct multi-centre studies to study air pollution effects in different study settings.

18

ACCEPTED MANUSCRIPT

Acknowledgements

AC C

EP

TE D

M AN U

SC

RI PT

Air pollution modelling was largely financed by the FAS (Swedish Council for Working Life and Social Research) grant 2010–0442. This work was also supported by the NordicWelfAir project funded by NordForsk. The RHINE Study has over the years received funding from many research foundations. HO’s and MK’s work on the preparation of this article was supported by the Estonian Ministry of Education and Research grants IUT34-17 and IUT20-11, respectively. The authors would also like to acknowledge Thomas Becker and Matthias Ketzel from Aarhus University, and Christian Asker and Mattias Jakobsson from the Swedish Meteorological and Hydrological Institute for their contributions.

19

20

ACCEPTED MANUSCRIPT

References Andersson M, Modig L, Hedman L, Forsberg B, Rönmark E. Heavy vehicle traffic is related to wheeze among schoolchildren: a population-based study in an area with low traffic flows. Environ Health. 2011;10:91.

RI PT

Artemis/HBEFA ref: SVARTEMIS Implementering av ARTEMIS Road Model i Sverige. EMFO emissionsforskningsprogrammet, IVL rapport B1831, februari 2009. Banerjee M, Capozzoli M, McSweeney L, Sinha D. Beyond kappa: A review of interrater agreement measures. Can J Statistics. 1999;27(1):3–23.

SC

Beelen R, Hoek G, van den Brandt PA, Goldbohm RA, Fischer P, Schouten LJ, et al. Long-term effects of traffic-related air pollution on mortality in a Dutch cohort (NLCS-AIR study). Environ Health Perspect. 2008;116(2):196–202.Beelen R, Hoek G, Vienneau D, Eeftens M, Dimakopoulou K, Pedeli X, et al. Development of NO2 and NOx land use regression models for estimating air pollution exposure in 36 study areas in Europe – The ESCAPE project. Atmos Environ. 2013;72:10–23.

M AN U

Berkowicz R. OSPM – A parameterised street pollution model. Environmental Monitoring and Assessment. 2000;65 (1/2), 323–331. Berkowicz R, Ketzel M, Jensen SS, Hvidberg M, Raaschou-Nielsen O. Evaluation and application of OSPM for traffic pollution assessment for a large number of street locations. Environ modell softw. 2008;23(3):296–303.

TE D

Boulter PG, Mccrae IS, Joumard R, André M, Keller M, Sturm P, et. al. ARTEMIS: Assessment and Reliability of Transport Emission Models and Inventory Systems - final report. TRL Published Project Report [Internet]. 2007 Oct [cited 2017 Apr 20]; Available from: https://trid.trb.org/view.aspx?id=886443 Brauer M, Hoek G, van Vliet P, Meliefste K, Fischer P, Gehring U, et al. Estimating Long-Term Average Particulate Air Pollution Concentrations: Application of Traffic Indicators and Geographic Information Systems: Epidemiology. 2003;14(2):228–39.

AC C

EP

Brunekreef B, Stewart AW, Anderson HR, Lai CKW, Strachan DP, Pearce N. Self-Reported Truck Traffic on the Street of Residence and Symptoms of Asthma and Allergic Disease: A Global Relationship in ISAAC Phase 3. Environ Health Perspect. 2009;117(11):1791–8. Cesaroni G, Badaloni C, Porta D, Forastiere F, Perucci CA. Comparison between various indices of exposure to traffic-related air pollution and their impact on respiratory health in adults. Occup Environ Med. 2008;65(10):683–90. Cesaroni G, Badaloni C, Gariazzo C, Stafoggia M, Sozzi R, Davoli M, Forastiere F. Long-term exposure to urban air pollution and mortality in a cohort of more than a million adults in Rome. Environ Health Perspect. 2013;121(3):324–31.Cyrys J, Eeftens M, Heinrich J, Ampe C, Armengaud A, Beelen R, et al. Variation of NO2 and NOx concentrations between and within 36 European study areas: Results from the ESCAPE study. Atmos Environ. 2012;62:374–90. Dehbi H-M, Blangiardo M, Gulliver J, Fecht D, de Hoogh K, Al-Kanaani Z, et al. Air pollution and cardiovascular mortality with over 25 years follow-up: A combined analysis of two British cohorts. Environ Int. 2017;99:275–81.

21

ACCEPTED MANUSCRIPT

De Hoogh K, Korek M, Vienneau D, Keuken M, Kukkonen J, Nieuwenhuijsen MJ, et al. Comparing land use regression and dispersion modelling to assess residential exposure to ambient air pollution for epidemiological studies. Environ Int. 2014;73:382–92. Dockery DW, Pope CA 3rd, Xu X, Spengler JD, Ware JH, Fay ME, Ferris BG Jr, Speizer FE. An association between air pollution and mortality in six U.S. cities. N Engl J Med. 1993; 329(24):1753–9.

RI PT

Eneroth K, Johansson C, Bellander T. Exposure COMPARISON BETWEEN MEASUREMENTS AND CALCULATIONS BASED ON DISPERSION MODELLING (EXPOSE) [Internet]. Stockholm, Sweden: SLB Analys; 2006 [cited 2017 Apr 13]. Available from: http://slb.nu/slb/rapporter/pdf8/lvf2006_012.pdf

SC

Filleul L, Rondeau V, Vandentorren S, Le Moual N, Cantagrel A, Annesi-Maesano I, et al. Twenty five year mortality and air pollution: results from the French PAARC survey. Occup Environ Med. 2005;62(7):453–60.Fecht D, Hansell AL, Morley D, Dajnak D, Vienneau D, Beevers S, et al. Spatial and temporal associations of road traffic noise and air pollution in London: Implications for epidemiological studies. Environ Int. 2016;88:235–42.

M AN U

Fernández-Somoano A, Llop S, Aguilera I, Tamayo-Uria I, Martínez MD, Foraster M, et al. Annoyance Caused by Noise and Air Pollution during Pregnancy: Associated Factors and Correlation with Outdoor NO2 and Benzene Estimations. Int J Environ Res Public Health. 2015;18;12(6):7044-58. doi: 10.3390/ijerph120607044. Ferris BG Jr, Anderson D0. The prevalence of chronic respiratory disease in a New Hampshire town. Am Rev Resp Dis. 1962;86:165.

TE D

Gamer M, Lemon J, Puspendra Singh IF. irr: Various Coefficients of Interrater Reliability and Agreement [Internet]. 2012 [cited 2016 Apr 21]. Available from: https://cran.rproject.org/web/packages/irr/index.html Genikhovich E, Gracheva I, Filatova E (2002) Modelling of urban air pollution: principles and problems. In: Borrego C, Schayes G (eds) Air pollution modelling and its application, XV. Kluwer, New York, pp 275–283.

EP

Gilliland F, Avol E, Kinney P, Jerrett M, Dvonch T, Lurmann F, et al. Air Pollution Exposure Assessment for Epidemiologic Studies of Pregnant Women and Children: Lessons Learned from the Centers for Children’s Environmental Health and Disease Prevention Research. Environ Health Perspect. 2005;113(10):1447–54.

AC C

Gislason T, Bertelsen RJ, Real FG, Sigsgaard T, Franklin KA, Lindberg E, et al. Self-reported exposure to traffic pollution in relation to daytime sleepiness and habitual snoring: a questionnaire study in seven North-European cities. Sleep Med. 2016;24:93–9. Hallgren KA. Computing Inter-Rater Reliability for Observational Data: An Overview and Tutorial. Tutor Quant Methods Psychol. 2012;8(1):23–34. Heinrich J, Gehring U, Cyrys J, Brauer M, Hoek G, Fischer P, et al. Exposure to traffic related air pollutants: self-reported traffic intensity versus GIS modelled exposure. Occup Environ Med. 2005;62(8):517–23. Heinrich J, Thiering E, Rzehak P, Krämer U, Hochadel M, Rauchfuss KM, Gehring U, Wichmann HE. Long-term exposure to NO2 and PM10 and all-cause and cause-specific mortality in a prospective cohort of women. Occup Environ Med. 2013;70(3):179–86.

22

ACCEPTED MANUSCRIPT

Härkönen J, Nikmo J, Karppinen A, Kukkonen J. A refined modelling system for estimating the emissions, dispersion, chemical transformation and dry deposition of traffic-originated pollution from a road. In: Cuvelier C, editor. A Refined Modelling System for Estimating the Emissions, Dispersion, Chemical Transformation and Dry Deposition of Traffic-Originated Pollution from a Road. European Joint Research Centre; Belgirate, Italy: 2001. pp. 311–313.

RI PT

Jacquemin B, Sunyer J, Forsberg B, Aguilera I, Briggs D, Götschi T, et al. Association between annoyance and individuals' values of nitrogen dioxide in a European setting. J Epidemiol Community Health. 2008;62(5):e12. Jensen SS. (1998): Background Concentrations for Use in the Operational Street Pollution Model (OSPM), NERI Technical Report No. 234. 1998. 107 p. Jensen SS, Berkowicz R, Sten Hansen H, Hertel O. A Danish decision-support GIS tool for management of urban air quality and human exposures. TRANSPORT RES D-TR E. 2001;6(4):229–41.

M AN U

SC

Jensen SS, Hvidberg M, Petersen J, Storm L, Stausgaard L, Becker T et al. (2009a): GIS-based National Road and Traffic Database 1960-2005. National Environmental Reseach Institute, Aarhus University, Roskilde. 73 s. NERI Technical report No. 678, 2009a (In Danish). http://www2.dmu.dk/Pub/FR678.pdf Jensen SS, Larson T, Deepti KC, Kaufman JD. Modeling Traffic Air pollution in Street Canyons in New York City for Intra-urban Exposure Assessment in the US Multi-Ethnic Study of Atherosclerosis. Atmos Environ. 43 (2009b);4544–4556.

TE D

Jie Y, Isa ZM, Jie X, Ju ZL, Ismail NH. Urban vs. Rural Factors That Affect Adult Asthma. In: Whitacre DM, editor. Rev Environ Contam T. 226 [Internet]. Springer New York; 2013 [cited 2017 Apr 20]. p. 33–63. Johannessen A, Verlato G, Benediktsdottir B, Forsberg B, Franklin K, Gislason T, et al. Longterm follow-up in European respiratory health studies - patterns and implications. BMC Pulm Med. 2014;14:63.

EP

Johansson, C., A. Hadenius, P.-Å. Johansson and T. Jonson (1999). SHAPE: The Stockholm study on health effects of air pollution and their economic consequences. Part 1: NO2 and particulate matter in Stockholm VAEGVERKET. PUBLIKATION.1999: 41 Available at: http://www.ivl.se/webdav/files/Brapporter/B1830.pdf

AC C

Johansson C, Andersson C, Bergström R, Krecl P. Exposure to particles due to local and non-local sources in Stockholm: estimates based on modelling and measurements 1997-2006. ITM RAPPORT. 2008 (175). Available at: http://slb.nu/slb/rapporter/pdf8/lvf2008_175.pdf Kaasik, M, Lukk T, Kartau K, Dovnar T. (2007). Nowcasting and forecasting the street pollution dispersion for Tallinn metropolitan area. In: Air Pollution Modeling and its Application (744−746). Elsevier. Kaasik M, Pindus M, Tamm T, Orru H (2014). The Porosity Concept Applied to Urban Canopy Improves the Results of Gaussian Dispersion Modelling of Traffic-Dominated Emissions. In: Steyn D; Mathur R. Air Pollution Modelling and its Application XXII (417−420). Springer. Kakosimos KE, Hertel O, Ketzel M, Berkowicz R. Operational street pollution model (OSPM) – a review of performed application and validation studies, and future prospects. Environ Chem. 2010; 7:485–503.

23

ACCEPTED MANUSCRIPT

Ketzel, M, Berkowicz R, Hvidberg H, Jensen SS, Raaschou-Nielsen O. Evaluation of AirGIS - A GISBased Air Pollution And Human Exposure Modelling System. Int. J. of Environment and Pollution. 2011;47:Nos. 1/2/3/4. Karppinen A, Kukkonen J, Elolähde T, Konttinen M, Koskentalo T, Rantakrans E. A modelling system for predicting urban air pollution: model description and applications in the Helsinki metropolitan area. Atm Environ. 2000;34(22):3723–33.

RI PT

Krewski D, Jerrett M, Burnett RT, Ma R, Hughes E, Shi Y, Turner MC, Pope CA 3rd, Thurston G, Calle EE, Thun MJ, Beckerman B, DeLuca P, Finkelstein N, Ito K, Moore DK, Newbold KB, Ramsay T, Ross Z, Shin H, Tempalski B. Extended follow-up and spatial analysis of the American Cancer Society study linking particulate air pollution and mortality. Res Rep Health Eff Inst. 2009;140:5-114; discussion 115-36.

SC

Kuehni CE, Strippoli MP, Zwahlen M, Silverman M. Association between reported exposure to road traffic and respiratory symptoms in children: evidence of bias. Int J Epidemiol. 2006;35(3):779-86. Epub 2006

M AN U

Lunn JE, Knowelden J, Roe JW. Patterns of respiratory illness in Sheffield junior school-children: a follow-up study. Brit. J. Prev. Soc. Med. 1970;24:223. Madsen C, Carlsen KCL, Hoek G, Oftedal B, Nafstad P, Meliefste K, et al. Modeling the intra-urban variability of outdoor traffic pollution in Oslo, Norway—A GA2LEN project. Atmos Environ. 2007;41(35):7500–11.

TE D

Molnár P, Stockfelt L, Barregard L, Sallsten G. Residential NOx exposure in a 35-year cohort study. Changes of exposure, and comparison with back extrapolation for historical exposure assessment. Atmos Environ. 2015; 115:62–9.Newson, R. Parameters behind “nonparametric” statistics: Kendall’s tau, Somers’ D and median differences. The Stata Journal. 2002; 2: 45–64. Oiamo TH, Baxter J, Grgicak-Mannion A, Xu X, Luginaah IN. Place effects on noise annoyance: Cumulative exposures, odour annoyance and noise sensitivity as mediators of environmental context. Atmos Environ. 2015;116:183–93.

EP

Orru H, Kaasik M, Antov D, Forsberg B. Evolution of traffic flows and traffic induced air pollution due to structural changes and development during 1993-2006 in Tartu (Estonia). Baltic journal of road and bridge engineering. 2008;3(4):206–12.

AC C

Orru K, Nordin S, Harzia H, Orru H. The role of perceived air pollution and health risk perception in health symptoms and disease: a population-based study combined with modelled levels of PM10. Int Arch Occup Environ Health, 2017 (In review). Persson R, Björk J, Ardö J, Albin M, Jakobsson K. Trait anxiety and modelled exposure as determinants of self-reported annoyance to sound, air pollution and other environmental factors in the home. Int Arch Occup Environ Health. 2007;81(2):179–91. Pindus M, Orru H, Modig L. Close proximity to busy roads increases the prevalence and onset of cardiac disease--Results from RHINE Tartu. Public Health. 2015;129(10):1398-405. Recio A, Linares C, Banegas JR, Díaz J. Road traffic noise effects on cardiovascular, respiratory, and metabolic health: An integrative model of biological mechanisms. Environ Res. 2016;146:359–70.

24

ACCEPTED MANUSCRIPT

Raaschou-Nielsen O, Andersen ZJ, Jensen SS, Ketzel M, Sørensen M, Hansen J, et al. Traffic air pollution and mortality from cardiovascular disease and all causes: a Danish cohort study. Environ Health. 2012;11:60. Respiratory health in northern Europe (RHINE) 2017. Project website. Available at: http://rhine.nu/

RI PT

Riedel N, Scheiner J, Müller G, Köckler H. Assessing the relationship between objective and subjective indicators of residential exposure to road traffic noise in the context of environmental justice. Journal of Environmental Planning and Management. 2014 Sep 2;57(9):1398–421. Ristovska G, Laszlo HE, Hansell AL. Reproductive Outcomes Associated with Noise Exposure — A Systematic Review of the Literature. International Journal of Environmental Research and Public Health. 2014 Aug 6;11(8):7931–52.

SC

Rudnik J. Epidemiological study on long-term effects on health of air pollution. Probl. Med. Wieku Rozwojowego. 1977; 7a(Suppl.):1. Sajani SZ, Trentini A, Rovelli S, Ricciardelli I, Marchesi S, Maccone C, et al. Is particulate air pollution at the front door a good proxy of residential exposure? Environ. Pollut. 2016;213:347–58.

M AN U

SMHI, 2015. Airviro user’s reference, vol 2: Working With the Dispersion Module, version 4.0. http://www.smhi.se/airviro/download SMHI Airviro Dispersion ref:http://www.smhi.se/airviro/modules/dispersion/dispersion-1.6846 Shepherd D, Dirks K, Welch D, McBride D, Landon J. The Covariance between Air Pollution Annoyance and Noise Annoyance, and Its Relationship with Health-Related Quality of Life. Int J Environ Res Public Health. 2016;13(8):792.

TE D

Shirinde J, Wichmann J, Voyi K. Allergic rhinitis, rhinoconjunctivitis and hayfever symptoms among children are associated with frequency of truck traffic near residences: a cross sectional study. Environ Health. 2015;14:84.

EP

Sørensen M, Hoffmann B, Hvidberg M, Ketzel M, Jensen SS, Andersen ZJ, et al. Long-Term Exposure to Traffic-Related Air Pollution Associated with Blood Pressure and Self-Reported Hypertension in a Danish Cohort. Environ Health Perspect. 2012;120(3):418–24.

AC C

Vlaski E, Stavric K, Seckova L, Kimovska Hristova M, Isjanovska R. The self-reported density of truck traffic on residential streets and the impact on asthma, hay fever and eczema in young adolescents. Allergologia et Immunopathologia. 2014;42(3):224–9. Team RDC. R: A Language and Environment for Statistical Computing [Internet]. Vienna, Austria: The R Foundation for Statistical Computing; 2015. Available from: http://www.R-project.org/ WHO Regional Office for Europe. Air quality guidelines for Europe: WHO regional publications; 1987. Report No. 23. WHO, Copenhagen WHO Regional Office for Europe. Air Quality Guidelines for Europe: WHO Regional Publications; 2000. Report No. 91. WHO, Copenhagen. WHO Regional Office for Europe. Review of evidence on health aspects of air pollution – REVIHAAP Project Technical Report. Copenhagen: WHO; 2013 p. p67. Zou B, Wilson JG, Zhan FB, Zeng Y. Air pollution exposure assessment methods utilized in epidemiological studies. J Environ Monit. 2009;11(3):475-90.

ACCEPTED MANUSCRIPT

25

Supplementary material City

Model (name and type)

Aarhus

OSPM within AirGIS

Model resolution

Emission data source

RI PT

Supplementary Table A. Schematic overview of the dispersion modelling methods Traffic data source

AC C

EP

TE D

M AN U

SC

Street concentrations at Vehicle emissions were based on COPERT From a national road and address location integrated into OSPM. traffic database with traffic data on every street. Bergen AIRVIRO – Gaussian 150x150 m Road traffic with emission factors from Modelled traffic flow (source) model HBEFA. Shipping traffic with emission factors from ENTEC 2002. Reykjavik AIRVIRO – Gaussian 150x150 m Road traffic with emission factors from Municipal traffic counts model HBEFA. Shipping traffic with emission factors from ENTEC 2002. Gothenburg AERMOD – Gaussian 50x50 m Road and marine traffic, domestic Traffic counts made by the model heating, and industrial sources. Emission municipality and Swedish factors based on inventories made by the Transport Administration municipality and national authorities. Emission factors from HBEFA. Tartu AEROPOL – Gaussian 40x40 m Emissions from traffic, domestic heating, Modelled with CUBE software, model and industrial sources. Emission factors based on rush-hour from CAR-FMI measurements at receptor points Umeå In city centre AIRVIRO – 50x50 m Traffic and all other background Traffic counts made by the Gaussian model. Outside emissions. Traffic emission factors from municipality and modelled the city LUR – land-use ARTEMIS, others from SIMAIR/SMED traffic counts of the Swedish regression model Transport Administration Uppsala AIRVIRO – Gaussian Central Uppsala: 50x50 m Emissions from traffic. Emission factors Municipal traffic counts on model Greater Uppsala: 100x100 from ARTEMIS municipal roads and National m road database on state roads Uppsala and Stockholm region: 500x500 m

ACCEPTED MANUSCRIPT

26

Supplementary Table B. Agreement between self-reported traffic noise exposurea, and NOx and traffic proximity Traffic noise

Highest tertile No Yes

Traffic proximity (tertiles) Kb (p)

Highest tertile No Yes

K (p)

RI PT

NOX (tertiles)

Traffic proximity (categories) Less than 150 m No Yes

K (p)

AC C

EP

TE D

M AN U

SC

RHINE (N=10708) None or low 63.5% 29.6% 60.9% 32.3% 76.6% 16.5% Moderate or high 3.1% 3.7% 0.156*** 5.8% 1.1% 0.148*** 3.9% 3.0% 0.168*** Aarhus (n=2269) None or low 71.7% 23.7% 63.3% 32.0% 80.0% 15.3% Moderate or high 2.6% 2.0% 0.162*** 3.4% 1.2% 0.157*** 2.9% 1.8% 0.214*** Bergen (n=1737) 64.0% 29.9% 61.9% 32.0% 77.5% 16.4% None or low Moderate or high 3.2% 2.8% 0.142*** 4.7% 1.3% 0.168*** 4.1% 1.9% 0.178*** Gothenburg (n=843) None or low 64.4% 30.3% 62.3% 32.4% 75.6% 19.1% Moderate or high 2.2% 3.1% 0.144*** 4.4% 0.9% 0.144*** 3.1% 2.2% 0.176*** Reykjavik (n=1681) None or low 61.4% 26.7% 57.2% 30.9% 60.9% 27.2% Moderate or high 5.3% 6.6% 0.016 9.5% 2.4% 0.031 4.6% 7.3% 0.030 Tartu (n=863) None or low 59.3% 24.3% 55.0% 28.6% 65.6% 18.1% Moderate or high 7.3% 9.0% 0.176*** 11.6% 4.8% 0.095** 10.1% 6.3% 0.108*** Umeå (n=1610) None or low 64.7% 30.2% 63.0% 31.9% 87.3% 7.6% Moderate or high 1.9% 3.1% 0.183*** 3.6% 1.4% 0.035 3.6% 1.4% 0.141*** Uppsala (n=1700) None or low 65.5% 30.9% 63.5% 32.9% 82.5% 13.9% Moderate or high 1.1% 2.5% 0.197*** 3.2% 0.4% 0.197*** 1.8% 1.8% 0.253*** a “Can you hear traffic noise in your bedroom?” with possible replies constituted four levels: None: “Not at all”; Low: “A little’”; Moderate: “Much”, High: “Very much.” bCohen’s Kappa. *p<0.05, **p<0.01, ***p<0.001.

ACCEPTED MANUSCRIPT

27

Supplementary Table C. Correlations within self-reported exposure metrics and subjective exposure metrics Modelled NOx tertiles vs Traffic proximity categories tau P 0.481 <0.001

tau 0.401

p <0.001

Aarhus

0.434

<0.001

0.661

<0.001

0.525

<0.001

Bergen

0.421

<0.001

0.520

<0.001

0.471

<0.001

Gothenburg

0.394

<0.001

0.490

<0.001

0.428

<0.001

Reykjavik

0.284

<0.001

0.483

<0.001

Tartu

0.458

<0.001

0.430

<0.001

Umeå*

0.432

<0.001

0.526

<0.001

Uppsala

0.388

<0.001

0.817

<0.001

M AN U 0.488

<0.001

0.415

<0.001

0.581

<0.001

0.603

<0.001

TE D EP AC C

SC

RHINE

Traffic proximity tertiles tau P 0.554 <0.001

RI PT

Self-reported traffic exposure and noise exposure

28

ACCEPTED MANUSCRIPT

Supplemenary Table D Rank correlations (Kendall’s tau) between self-reported residential exposurea, and modelled NOX and traffic proximity with cutoff at 50th and 90th percentile Traffic near bedrooma

Traffic noiseb

AC C

EP

TE D

M AN U

SC

RI PT

Traffic proximity Traffic proximity NOX NOX RHINE 0.165* 0.124* 0.157* 0.115* Aarhus 0.124* 0.134* 0.146* 0.174* Bergen 0.116* 0.103* 0.175* 0.154* Gothenburg 0.192* 0.145* 0.180* 0.177* Reykjavik 0.056* 0.061* -0.033 -0.017 Tartu 0.135* 0.109* 0.170* 0.085* Umeå 0.262* 0.077* 0.217* 0.016* Uppsala 0.286* 0.238* 0.264* 0.207* a b “Is your bedroom window near a street (less than 20 m)?” “Can you hear traffic noise from your bedroom?” cHighest tertile, lowest exposure. * p<0.05.

ACCEPTED MANUSCRIPT

29

AC C

EP

TE D

M AN U

SC

RI PT

Supplementary figure A Correlation of self-reported and subjective exposure metrics presented in Table 3 by exposure levels of each RHINE

center