Association between Mortality and Short-Term Exposure to Particles, Ozone and Nitrogen Dioxide in Stockholm, Sweden

In this study, the effects on daily mortality in Stockholm associated with short-term exposure to ultrafine particles (measured as number of particles with a diameter larger than 4 nm, PNC4), black carbon (BC) and coarse particles (PM2.5–10) have been compared with the effects from more common traffic-pollution indicators (PM10, PM2.5 and NO2) and O3 during the period 2000–2016. Air pollution exposure was estimated from measurements at a 20 m high building in central Stockholm. The associations between daily mortality lagged up to two days (lag 02) and the different air pollutants were modelled by using Poisson regression. The pollutants with the strongest indications of an independent effect on daily mortality were O3, PM2.5–10 and PM10. In the single-pollutant model, an interquartile range (IQR) increase in O3 was associated with an increase in daily mortality of 2.0% (95% CI: 1.1–3.0) for lag 01 and 1.9% (95% CI: 1.0–2.9) for lag 02. An IQR increase in PM2.5–10 was associated with an increase in daily mortality of 0.8% (95% CI: 0.1–1.5) for lag 01 and 1.1% (95% CI: 0.4–1.8) for lag 02. PM10 was associated with a significant increase only at lag 02, with 0.8% (95% CI: 0.08–1.4) increase in daily mortality associated with an IQR increase in the concentration. NO2 exhibits negative associations with mortality. The significant excess risk associated with O3 remained significant in two-pollutant models after adjustments for PM2.5–10, BC and NO2. The significant excess risk associated with PM2.5–10 remained significant in a two-pollutant model after adjustment for NO2. The significantly negative associations for NO2 remained significant in two-pollutant models after adjustments for PM2.5–10, O3 and BC. A potential reason for these findings, where statistically significant excess risks were found for O3, PM2.5–10 and PM10, but not for NO2, PM2.5, PNC4 and BC, is behavioral factors that lead to misclassification in the exposure. The concentrations of O3 and PM2.5–10 are in general highest during sunny and dry days during the spring, when exposure to outdoor air tend to increase, while the opposite applies to NO2, PNC4 and BC, with the highest concentrations during the short winter days with cold weather, when people are less exposed to outdoor air.


Introduction
For PM 10, PM 2.5 , NO 2 and O 3 , the short-term health effects in terms of increased daily mortality have been investigated in many studies. Results from original studies have also been combined in meta-analyzes. For NO 2 , a significant association between short-term exposure and mortality, based on 60 studies from different parts of the world, was described in Mills et al. [1]. In this study, The purpose of this study was to compare the excess risks of daily mortality associated with potentially relevant traffic pollution indicators, namely UFP (PNC 4 ), BC and PM 2.5-10 , and compare these results with the excess risks of the more established indicators, namely PM 10 , PM 2.5 and NO 2 , as well as with O 3 . We have examined the associations between daily mortality and exposure to NO 2 , O 3 , PM 2.5 , PM 10 , PM 2.5-10 , PNC 4 and BC, based on data from an urban background measuring station in Stockholm during the period 2000-2016.

Materials and Methods
This study covers the city of Stockholm with a population of approximately 0.8-0.9 million people over the period 2000-2016. Population data were obtained from the Swedish Central Bureau of Statistics. Natural cause mortality data were obtained from the National Cause of Death Register. Natural cause mortality was defined on the basis of the underlying cause of death (ICD-10: A00-R99), and these data included the daily number of deaths from natural causes occurring among the registered population.
Air pollution exposure was estimated from central measurements made on the roof-top of a 20 m high building in central Stockholm. The monitoring station is part of the city's regulatory air pollution control network, and equipped with reference (or equivalent) instruments for regulated pollutants according to the EU air quality directive for NO 2 , O 3 , PM 2.5 and PM 10 (Table A1). The O 3 measurements are based on daily maximum 8-h mean. In addition, it includes measurements of unregulated black carbon (BC), total particle number concentrations (PNC 4 ), and the coarse fraction (PM 2.5-10 ), estimated by subtracting PM 2.5 from PM 10 . Different instruments have been used to measure BC. However, when the hourly mean values of two different BC measurements techniques in Stockholm were compared in 2006, the R-values were 0.87 and 0.95, respectively, at the two measurement sites [14]. For the measurements of ultrafine particles (PNC 4 and PNC 7 ), an instrument was used that registered all particles larger than 7 nm (PNC 7 ) during the period from May 2001 until November 2013. From March 2008 to the end of the period (December 2016), another instrument was used to record particles larger than 4 nm (PNC 4 ). By applying a linear regression between PNC 4 and PNC 7 during overlapping periods, measured PNC 7 was used to construct a more complete time-series of PNC 4 . The mean value of the measured concentrations of PNC 4 was 8650 cm −3 , with a root mean square error (RMSE) of 1355 cm −3 in relation to the modeled PNC 4 values, based on the linear regression with PNC 7 (see Figure A2 in Appendix A). Several plots illustrating the daily and monthly mean concentrations during 2000-2016, and the correlations between different pairs of the analyzed air pollutants, have been performed by using the "Openair" package [15].
The associations between different air pollutants and the daily mortality were modelled by using a quasi-Poisson regression model with a logistic link function. The model estimated the effect of an IQR increase of air pollutants on daily mortality for lag 01 and 02, while controlling for other time-varying factors by assuming a linear additive effect on a logarithmic scale: where APi is the concentration of a specific or a combination of air pollutants on day i, Wi is variables controlling for the weather on day i, more specifically maximum temperature and snowfall, DOWi is the day of week, and the long-time trend is a smooth function varying over time to capture any long-term and seasonal patterns in mortality. The smooth function used was a penalized regression spline restricted to 5 d.f. (degrees of freedom) per year. Snowfall has been included, since it is a risk factor for daily mortality, as described in Auger et al. (2017) [16]. All pollutants were modelled by assuming a linear relationship with daily mortality. Air pollutants were first modelled in single-pollutant models (Figure 4), and traffic-related pollutants with effect estimates with a p-value smaller than 0.2 were included in multi-pollutant models together with O 3 (Figures 5-7). We also investigated the correlation matrix ( Figure 3), and included pollutants that were negatively correlated, or positively correlated, but with opposite effects in the single-pollutant model ( Figure 4). Temperature were adjusted for by using two different smooth functions corresponding to the different lag-windows of 0-2 and 3-10. The model allowed for the use of 4 d.f. for each function. In the sensitivity analysis, we added models allowing for 8 d.f. in the smooth temperature functions as well as adding another temperature variable with temperatures from lag 11-20. In addition, we used an indicator variable to identify if the modelled PNC 4 data (see Figure A2) generated different risk estimates in comparison with the measured data.

Descriptive Data
The descriptive data are presented in Table 1. An overview of the temporal variation of the daily mean concentrations of air pollutants and daily mortality in Stockholm is given in Figure A1 in Appendix A. For most pollutants, data capture is high, >80% (Table 1); the exceptions are BC (53%) and PNC (70%).  Figure A1).
There were also pronounced seasonal variations in the concentrations, as shown in Figures 1 and 2. NO 2 , PNC and BC exhibited the highest concentrations during winter (October to March) and lowest in summer (June-July), whereas PM 10 , PM 2.5 , PM 2.5-10 and O 3 exhibited peak concentrations during late winter to early summer (March-May).
the sensitivity analysis, we added models allowing for 8 d.f. in the smooth temperature functions as well as adding another temperature variable with temperatures from lag 11-20. In addition, we used an indicator variable to identify if the modelled PNC4 data (see Figure A2) generated different risk estimates in comparison with the measured data.

Descriptive Data
The descriptive data are presented in Table 1. An overview of the temporal variation of the daily mean concentrations of air pollutants and daily mortality in Stockholm is given in Figure A1 in Appendix A. For most pollutants, data capture is high, >80% (Table 1); the exceptions are BC (53%) and PNC (70%).  Figure A1).
There were also pronounced seasonal variations in the concentrations, as shown in Figures 1  and 2. NO2, PNC and BC exhibited the highest concentrations during winter (October to March) and lowest in summer (June-July), whereas PM10, PM2.5, PM2.5-10 and O3 exhibited peak concentrations during late winter to early summer (March-May).        Figure 3 shows the correlations between all pairs of data for the measured air pollutants. High correlations (R > 0.6) were found between pairs of NO2, PNC7, PNC4 and PNC, reflecting their common origins. Likewise the pairs PM10-PM2.5, PM10-PM2.5-10 and BC-PM2.5 showed high correlations. O3 showed negative correlations with NO2, PNC and BC. Note that the R-values in Figure 3 are given in percent.    Table 1. Figure 4 shows single-pollutant models, where the excess risks for lag 01 and 02 associated with an IQR increase of the different pollutants are presented. Lag 01 and 02 represent a lagging effect of the same and the previous day, and the same and the previous two days, respectively. Figures 5-7 show multi-pollutant models, where the effects of the modeled pollutants are adjusted for each other. All multi-pollutant models are based on lag 02.
In the multi-pollutant models ( Figures 5-7), the risk estimates in the single-pollutant model have been adjusted for the effects of some other pollutants. Pollutants that were negatively correlated (see Figure 3), or positively correlated, but with opposite effects in the single-pollutant model ( Figure 4) have been used in the multi-pollutant models. For O 3 , the significant effect for lag02 in the single-pollutant model remained significant in two-pollutant models after adjustments for PM 2.5-10 , BC and NO 2 (Figures 4-7). For PM 2.5-10 , the significant effect for lag02 in the single-pollutant model remained significant in the two-pollutant after adjustment for NO 2 , and also after adjustments for both O 3 and NO 2 together ( Figure 5). The significantly negative effect associated with NO 2 for lag02 in the single-pollutant model remained significant in two-pollutant models after adjustments for PM 2.5-10 , BC and O 3 (Figures 5-7). The significantly negative effect associated with NO 2 also remained significant after adjustment for both PM 2.5-10 and O 3 together (Figure 7). Modelling the effect of PNC 4 , while adjusting for O 3 , increased the effect estimate by 23%, while remaining non-significant. The estimated effect of PM 2.5 was to some degree affected by the introduction of NO 2 in the model, where the negative, non-significant estimate for PM 2.5 (p-value: 0.95) changed to a small positive effect, while remaining non-significant (p-value: 0.65).  Figure 4 shows single-pollutant models, where the excess risks for lag 01 and 02 associated with an IQR increase of the different pollutants are presented. Lag 01 and 02 represent a lagging effect of the same and the previous day, and the same and the previous two days, respectively. Figures 5-7 show multi-pollutant models, where the effects of the modeled pollutants are adjusted for each other. All multi-pollutant models are based on lag 02.
In the multi-pollutant models ( Figures 5-7), the risk estimates in the single-pollutant model have been adjusted for the effects of some other pollutants. Pollutants that were negatively correlated (see Figure 3), or positively correlated, but with opposite effects in the single-pollutant model ( Figure 4) have been used in the multi-pollutant models. For O3, the significant effect for lag02 in the singlepollutant model remained significant in two-pollutant models after adjustments for PM2.5-10, BC and NO2 (Figures 4-7). For PM2.5-10, the significant effect for lag02 in the single-pollutant model remained significant in the two-pollutant after adjustment for NO2, and also after adjustments for both O3 and NO2 together ( Figure 5). The significantly negative effect associated with NO2 for lag02 in the singlepollutant model remained significant in two-pollutant models after adjustments for PM2.5-10, BC and O3 ( Figures 5-7). The significantly negative effect associated with NO2 also remained significant after adjustment for both PM2.5-10 and O3 together (Figure 7). Modelling the effect of PNC4, while adjusting for O3, increased the effect estimate by 23%, while remaining non-significant. The estimated effect of PM2.5 was to some degree affected by the introduction of NO2 in the model, where the negative, nonsignificant estimate for PM2.5 (p-value: 0.95) changed to a small positive effect, while remaining nonsignificant (p-value: 0.65).       Percentage change with 95% CI Multi-pollutant model with BC    Percentage change with 95% CI Multi-pollutant model with BC

Sensitivity Analysis
In the sensitivity analysis, we allowed for more flexible temperature associations. Increasing the number of d.f. allowed in the temperature function lowered the p-values for mainly PM10 and PM2.5. However, adding a temperature variable, adjusting for a longer time-frame of 11-20 days prior, rendered the PM variables insignificant. The other air pollution variables were essentially unaffected by inclusion of a longer temperature adjustment. The reasons for these patterns are unknown, but resuspension of road dust depends on complex changes in weather conditions, e.g. from periods with rain or snow to dry weather, which could also affect mortality.
The investigation of the modelled PNC4 data ( Figure A2) found that there was no difference between the estimates generated by the modelled and the measured data. Consequently, the lack of statistically significant excess risks associated with PNC4 is not caused by the use of the modelled PNC4 data, based on the linear regression with PNC7 ( Figure A2).

Local and Non-Local Sources
This study includes pollutants that are relatively good indicators of different local and non-local sources. NO2, PNC are all mainly influenced by local vehicle exhaust emissions [17,18], and are therefore highly correlated, as shown in Figure 3. BC is also emitted mainly from local vehicle exhaust, but is also influenced by long-range transport [14], making the correlation with NO2 and PNC somewhat smaller. The coarse particle fraction, PM2.5-10, is mainly due to local road-dust suspension [17], and since PM10 largely consists of coarse particles, it is highly correlated with PM2.5-10. PM2.5 is dominated by long-range transported secondary particles, and shows low correlations with pollutants like NO2, PNC and PM2.5-10, mainly influenced by local sources. However, PM2.5 shows higher correlations with BC and PM10 due to some influence of long-range transport on these compounds. And finally, the O3 concentrations in the city depend mainly on the long-distance transport, but are to some degree also influenced by the chemical reactions involving nitrogen monoxide (NO). The O3 concentrations are therefore reduced when primary exhaust concentrations are high during stagnant conditions.

Sensitivity Analysis
In the sensitivity analysis, we allowed for more flexible temperature associations. Increasing the number of d.f. allowed in the temperature function lowered the p-values for mainly PM 10 and PM 2.5 . However, adding a temperature variable, adjusting for a longer time-frame of 11-20 days prior, rendered the PM variables insignificant. The other air pollution variables were essentially unaffected by inclusion of a longer temperature adjustment. The reasons for these patterns are unknown, but resuspension of road dust depends on complex changes in weather conditions, e.g., from periods with rain or snow to dry weather, which could also affect mortality.
The investigation of the modelled PNC 4 data ( Figure A2) found that there was no difference between the estimates generated by the modelled and the measured data. Consequently, the lack of statistically significant excess risks associated with PNC 4 is not caused by the use of the modelled PNC 4 data, based on the linear regression with PNC 7 ( Figure A2).

Local and Non-Local Sources
This study includes pollutants that are relatively good indicators of different local and non-local sources. NO 2 , PNC are all mainly influenced by local vehicle exhaust emissions [17,18], and are therefore highly correlated, as shown in Figure 3. BC is also emitted mainly from local vehicle exhaust, but is also influenced by long-range transport [14], making the correlation with NO 2 and PNC somewhat smaller. The coarse particle fraction, PM 2.5-10 , is mainly due to local road-dust suspension [17], and since PM 10 largely consists of coarse particles, it is highly correlated with PM 2.5-10 . PM 2.5 is dominated by long-range transported secondary particles, and shows low correlations with pollutants like NO 2 , PNC and PM 2.5-10 , mainly influenced by local sources. However, PM 2.5 shows higher correlations with BC and PM 10 due to some influence of long-range transport on these compounds. And finally, the O 3 concentrations in the city depend mainly on the long-distance transport, but are to some degree also influenced by the chemical reactions involving nitrogen monoxide (NO). The O 3 concentrations are therefore reduced when primary exhaust concentrations are high during stagnant conditions. It should be noted that the influence on PM 10 of local vehicle generated road-dust suspension is relatively large in Stockholm compared to many other cities in Europe. Even though also NO 2 and PNC originate from local road traffic, the temporal correlations with road dust is very low (Figure 3). The reason for this is that road-dust suspension is highly influenced by the wetness of the road surfaces, as shown earlier by Johansson et al. [17]. So, in summary, the mix of pollutants included in this study are indicators of local road traffic emissions from vehicle exhaust and local non-exhaust particles, non-local secondary particulate matter and photochemical pollutants (O 3 ).

Representativeness of One Central Monitoring Station for Population Exposure
Clearly, the possibility to quantify any associations between mortality and different pollutants depends on how well the exposure can be quantified. In this study, we have used one single urban background site, assuming that the temporal variability at this site reflects the temporal variability of the exposure in the population. A high temporal correlation between ambient concentrations at different measuring sites within a city means that one centrally located measuring station should be enough in order to estimate the short-term variations in pollutant concentrations in time-series studies, even though it is inadequate when it comes to estimate the long-term health effects, due to the spatial gradients in exposure concentrations [17,19].
Since PM 2.5 are mainly influenced by non-local sources, spatial variations in ambient concentrations are small, and temporal variations will be very similar everywhere in the city. This has also been verified earlier for Stockholm in the TRAPCA study by Cyrys et al. [20], and it is also well known from other studies (see e.g., review by Monn [21]). O 3 concentrations in Stockholm are also mainly influenced by long-range transport, even though there is some impact of the titration by NO x (photochemical removal) close to densely trafficked roads with high emissions of NO x . There is no local photochemical production of O 3 in Stockholm.
For pollutants like PNC, BC and NO 2 , with road traffic emissions being the main source, the temporal variability in the concentrations may be expected to be quite similar everywhere in the city, as traffic intensities usually show similar temporal variations along different roads in the city. High temporal correlations (R ≈ 0.8) between traffic sites have been observed for PNC in Helsinki (Buzorius et al. [22]), and between 24-h mean concentrations of PNC at central sites and homes in Amsterdam, Athens, Birmingham and Helsinki, with a median correlation for PNC per city in the range of 0.67 and 0.76, as shown in Puustinen et al. [23]. Likewise, Cyrys et al. [19], found high correlations (R > 0.80) for PNC when they compared four different measurement stations in Augsburg. However, the variability of the concentrations of PNC in different cities may not be driven by the same emissions sources and atmospheric processes, and the PNC variability does not always indicate the impact of road traffic on air quality [24]. The variability in PNC depends on meteorology and on the size of the smallest particles considered due to the increasing influence of particle dynamics on particles smaller than 20 nm. Model calculations of PNC in Stockholm by Gidhagen et al. [18] showed that episodic losses of nanoparticles due to coagulation and dry deposition, some kilometers downwind of major roads, rise in connection with low wind speed and suppressed turbulent mixing. Similar results was found by Karl et al. [25], based on model calculations of PNC in Oslo, Helsinki and Rotterdam. Removal due to coagulation and deposition may thus lead to different temporal variations in different parts of a city. Moreover, for particles smaller than 100 nm, the formation is highly temperature dependent, as has been shown for Stockholm, where the number of particles, normalized by NO x , increases with decreasing temperature [26]. In addition, exposures in microenvironments, like indoors, might be very important for the daily exposure doses of PNC (Kumar et al. [27]). And for BC, the spatiotemporal variability was not found to be very high in Stockholm; different urban sites were poorly correlated even for daily averages (R < 0.70) [14], indicating that a single central measurement site would lead to misclassifications in the exposure.
Consequently, this means that the central monitor should reflect day-to-day variations in the exposure to ambient PM 2.5 and O 3 , but may be less good for BC and PNC.

The Estimated Excess Risks and Explanatory Factors
In both epidemiological and clinical studies, short-term exposure to air pollutants has been demonstrated to increase the mortality related to cardiovascular and respiratory diseases. Regarding short-term mortality related to PM, there are several potential biological mechanisms behind this relationship. Exposure to PM in both the coarse (PM 2.5-10 ) and the fine (PM 2.5 ) fraction induces oxidative stress and inflammation. Exposure to PM 2.5 can also affect the autonomic nervous system, and can thereby cause alterations in the autonomic control of the heart, which is also a risk factor for cardiovascular mortality [28].
The results presented in the single-pollutant model in Figure 4 show statistically significant positive excess risks for O 3 and PM 2.5-10 and PM 10 , but with no positive significant excess risks associated with exposure to the other pollutants. However, when comparing these coefficients with the results from similar studies, there are similarities regarding the results. In Meister et al. [29], a stronger effect on daily mortality (lag01) from PM 2.5-10 in comparison with PM 2.5 was found for Stockholm, based on data from 2000 to 2008.
In this study, statistically significant excess risks for O 3 were found for both lag 01 and lag 02 in the single-pollutant model, and the excess risks remained significant in all two-pollutant models. However, this phenomenon is in line with other studies that have done similar analyzes. In Raza et al. [30], where short-term effects of air pollution on out-of-hospital cardiac arrest in Stockholm were analyzed, significant effects were observed for O 3 , but not for PM 2.5 , PM 2.5-10 , NO x and NO 2 . In another study performed in Stockholm, where short-term exposure to ozone and mortality in subjects with and without previous cardiovascular disease was analyzed, significant associations were found for O 3 , and these associations remained basically unchanged in two-pollutant models with NO 2 and PM 10 [31].
The results in Figure 4 exhibit a tangible pattern, with significant excess risks for O 3 , PM 2.5-10 and PM 10 , but with non-significant excess risks for all the exhaust-related pollutants. In Stockholm, the mass of PM 10 consists largely of mechanically generated particles from road abrasion, and since PM 10 and PM 2.5-10 are highly correlated (R = 0.81), their sources are thus also largely the same.
There are some possible reasons for the robust associations between exposure to O 3 , PM 2.5-10 and PM 10 , and the daily mortality. Besides the established harmful effects, one possible reason may be that the measurement data for O 3 , PM 2.5-10 and PM 10 reflect the exposure better than for the other pollutants due to behavioral factors. Increased O 3 concentrations coincides with sunny and warm days in spring and summer when people spend more time outdoors and allow windows to be opened to a greater extent. This will increase the exposure and thereby contribute to the significant excess risks. The concentrations of PM 2.5-10 and PM 10 also tend to increase during sunny days in spring and early summer due to suspension from dry roadways, which will then cause a higher exposure during these days. Contrariwise, the exhaust-related emissions tend to be lowest when the outdoor activities tend to be highest and vice versa, which possibly can contribute to the absence of significant positive associations for all exhaust-related pollutants (NO 2 , PM 2.5 , PNC 4 and BC) in Figure 4. In this way, the exposure misclassification may work in different directions for different pollutants. The uncommon result for NO 2 could also partly be explained by the high contribution from local sources resulting in larger weather influences. Large spatial variation in combination with the use of one centrally located measuring station may also increase the exposure misclassification for NO 2 , if the average exposure levels vary differently than at the central monitoring station. Negative associations between NO 2 and cardiovascular mortality were also observed for the northernmost cities of Stockholm and Helsinki in the APHEA2 Study [32]. Even though the problem with behavioral factors has been discussed also for southern European cities (Chiusolo et al. [33]), it is probably not as pronounced as in the northern European cities, where the contrast in weather between summer and winter is much more pronounced, which can contribute to the negative associations observed in Stockholm and Helsinki.

Strengths and Limitations of This Study
A strength of this study is that we have an extensive dataset with continuous measurements of many pollutants that are good indicators of different sources: local vehicle exhaust and non-exhaust emissions, non-local secondary particles and an important gaseous oxidant (O 3 ). Another strength is that we use high quality mortality data from the National Cause of Death Register.
The most obvious limitation of this study is that the measured concentrations are obtained from one centrally located measuring station, which potentially creates exposure misclassifications among the population. Another limitation of the study is the poor data capture for PNC 4 and PNC 7 , where a simple linear regression during overlapping periods has been used to construct a complete time series of PNC 4 . However, the calculated RMSE between the measured and the modeled PNC 4 concentrations is relatively small in comparison with the mean value of PNC 4 during this overlapping period, and the investigation of the modelled PNC 4 data found that there was no difference between the estimates generated by the modelled and the measured data. Another limitation is that we have used particle counters that include nanoparticles, which may be subject to large variability due to dynamic processes like coagulation and condensation/evaporation, and also due to dry deposition. By including only particles larger than e.g., 20 nm, the influence of particle dynamics and deposition may be avoided and make the exposure estimate using a single site more representative for the population.

Conclusions
The conclusion of this study is that the excess risks associated with exposure to exhaust emissions (NO 2 , PM 2.5 , BC and PNC 4 ) exhibit much more uncertain relationships in comparison with O 3 , PM 2.5-10 and PM 10 . The results, where significant associations were found only for O 3 , PM 2.5-10 and PM 10 are, however, in line with other studies from Stockholm which have analyzed similar relationships. The spatial and temporal variations associated with pollutants of local origin can make it harder to estimate population exposure based on one centrally located measuring station. However, the potential reason for the result findings, where statistically significant positive excess risks were found for O 3 , PM 2.5-10 and PM 10 , but not for NO 2 , PM 2.5 , PNC 4 and BC, is probably to a large part caused by behavioral factors. The concentrations of O 3 and PM 2.5-10 are in general high during sunny days, when outdoor activities tend to increase, while the opposite applies to NO 2 , PM 2.5 , PNC 4 and BC, with the highest concentrations during the winter months.
Author Contributions: The study was planned by all authors. The draft manuscript has been written by H.O. and has been revised and approved by all authors. The figures performed in OpenAir have been made by C.J. The statistical analyses have been performed by C.Å.
Funding: This project was partly funded by ERA-PLANET (www.era-planet.eu), trans-national project SMURBS (www.smurbs.eu) (Grant Agreement n. 689443) under the EU Horizon 2020 Framework Programme. Bertil Forsberg and Christofer Åström were supported by the Swedish Clean Air and Climate research program (SCAC) funded by the Swedish Environmental Protection Agency.

Conflicts of Interest:
The authors declare no conflict of interest.