Has the risk of mortality related to short-term exposure to particles changed over the past years in Athens, Greece?

Environment International xxx (xxxx) xxx–xxx

Contents lists available at ScienceDirect

Environment International journal homepage: www.elsevier.com/locate/envint

Has the risk of mortality related to short-term exposure to particles changed over the past years in Athens, Greece? K. Tzimaa, A. Analitisa, K. Katsouyannia,b, E. Samolia,

⁎

a

Department of Hygiene, Epidemiology and Medical Statistics, Medical School, National and Kapodistrian University of Athens, 75 Mikras Asias Str, 115 27 Athens, Greece b Department of Primary Care & Public Health Sciences and Environmental Research Group, King's College London, 150 Stamford Street, SE1 9NH London, UK

A R T I C L E I N F O

A B S T R A C T

Handling Editor: Martí Nadal

Although the health effects of short-term exposure to ambient particles have been well documented, there is a need to update scientific knowledge due to the continuously changing profile of the air pollution mixture. Furthermore the effect of the severe economic crisis in Greece that started in 2008 on previously reported associations has not been studied. We assessed the change in mortality risk associated with short-term exposure to PM10 in Athens, Greece during 2001–12. Time-series data on the daily concentrations of regulated particles and all cause, cardiovascular and respiratory mortality were analyzed using overdispersed Poisson regression models, controlling for time-varying confounders such as seasonality, meteorology, influenza outbreaks, summer holidays and day of the week. We assessed changes in risk over time by inclusion of an interaction term between particles' levels and time or predefined periods, i.e. 2001–07 and 2008–12. While the related mortality risks increased over the analyzed period, the difference before and after 2008 was significant only for total mortality (p-value for interaction .03) and driven by the difference observed among those ≥75 years. An interquartile increase in PM10 before 2008 was associated with 1.51% increase in deaths among ≥75 years (95% Confidence interval (CI): 0.62%, 2.40%), while after 2008 with a 2.61% increase (95%CI: 1.72%, 3.51%) (p-value for interaction .01). Our results indicate that despite the decline in particles' concentration in Athens, Greece during 2001–12 the associated mortality risk has possibly increased, suggesting that the economic crisis initiated in 2008 may have led to changes in the particles' composition due to the ageing of the vehicular fleet and the increase in the use of biomass fuel for heating.

Keywords: Air pollution Mortality risk Particulate matter Short-term effects Trends

1. Introduction As the causal association between health effects and exposure to ambient particulate matter (PM) has been well documented over the past years (WHO, 2013) research in the field has moved towards the identification of harmful components and sources or investigation of associations with specific health endpoints. Nevertheless, since the profile of the air pollution mixture reflects a dynamic atmospheric process, there is a need to re-assess effects that have been identified over 10 years ago. Time-varying factors defining air pollution composition include technological interventions to the traffic fleet that consists the major source in the urban settings, related policy measures such as implementation of limit values and also climate change related effects such as the temperature increases. According to the U.S. Environmental Protection Agency (2017) a decrease of > 35% in PM

⁎

levels has been observed in the U.S. during the last two decades (EPA, 2017). Dominici et al. (2007) have investigated the trend in the air pollution associated mortality risk in the US and found a weak indication that the effects of particles with aerodynamic diameter < 10 μm (PM10) on mortality declined during 1987–2000 although shortterm exposure to ambient particles continued to be associated with increased mortality. The decline in particles' levels has also been observed in Europe. According to the European Environmental Agency 2016 report, the annual mean concentrations of PM10 displayed a significant downward trend from 2000 to 2014, observed at 75% of all stations in the fixed network in European cities, while < 1% of the stations registered a significant increasing trend. Karanasiou et al. (2014) have investigated the long-term trend in air pollution concentration levels in Mediterranean European cities and reported that PM and gaseous pollutants,

Corresponding author. E-mail address: [email protected] (E. Samoli).

https://doi.org/10.1016/j.envint.2018.01.002 Received 16 October 2017; Received in revised form 22 December 2017; Accepted 8 January 2018 0160-4120/ © 2018 Published by Elsevier Ltd.

Please cite this article as: Tzima, K., Environment International (2018), https://doi.org/10.1016/j.envint.2018.01.002

Environment International xxx (xxxx) xxx–xxx

K. Tzima et al.

We further excluded 13 days when PM10 levels exceeded 150 μg/m3, thus including only days with PM10 ≤ 150 μg/m3 in the analysis, because the relationship between particles and mortality within this range is effectively linear (Samoli et al., 2008; WHO, 2013). Time-series data on daily temperature (°C, daily mean) and relative humidity (%, daily mean) were provided by the National Observatory of Athens from one central monitor that was in continuous operation during the analysis period.

except nitrogen dioxide (NO2), revealed a decrease with the highest reduction in particles observed in Athens, Greece (−4 μgm−3 for PM2.5) followed by a decrease in Milan and Turin, Italy. The authors attributed this decrease to the effectiveness of the vehicular emission control strategies implemented in Europe and the improvement in motor engine characteristics as vehicle emissions in Europe have been regulated through performance and fuel standards. Few studies have investigated whether this reported decline in the air pollution levels has affected the risks previously reported. In Erfurt, Germany (Breitner et al., 2009; Peters et al., 2009) reported decreasing effects for short-term associations in 1991–2002, as air pollution decreased following pollution control measures implemented in Eastern Germany. Renzi et al. (2017) estimated constant effects for the associations between air pollutants and total mortality in Rome, Italy during the last two decades, which is in accordance with the conclusion on PM10 and NO2 effects on total mortality in Switzerland in 2001–10 (Perez et al., 2015). Finally, Carugno et al. (2017) reported that in 2003–06, PM10 levels were responsible for 343 annual deaths from natural causes in the district of Lombardy, Italy, that decreased to 254 in 2007–10 and to 208 in 2011–14. Athens, Greece provides an excellent opportunity to study the longterm trends of the mortality risk associated with short-term exposure to particles due to the sharp decrease in the levels previously described (Karanasiou et al., 2014). Linear associations between short-term particles' exposure and mortality have been identified in the most commonly observed concentration range, but there were indications of a supra linear curve that implies greater risk at lower levels (WHO, 2013). On the other hand, the economic crisis in Greece that started in 2008 has led to decreased volume of the vehicular fleet but also to changed composition of the air pollution mixture due to the ageing of the fleet and the increase in the use of biomass fuel for heating (Valavanidis et al., 2015). Hence we investigated the long-term trends in the mortality risk due to particles in Athens, Greece during 2001–12 in order to identify changes associated with the changing particles' profile and other factors attributed to the economic crisis.

2.2. Methods The PM-mortality associations were investigated using Poisson regression models allowing for overdispersion (Samoli et al., 2008, 2013) of the form:

log E [Yt ] = β0 + b1 ∗PMt + s(timet,k) + s (tempt , k1) + s (lag1(tempt ), k2) + [other confounders]

(1)

where E[Yt] is the expected value of the Poisson distributed variable Yt indicating the daily mortality count on day t with Var(Yt) = φE[Yt], φ being the over-dispersion parameter, timet is a continuous variable indicating the time of event, tempt is the value of mean temperature on day t, lag1(tempt) is the temperature level the day before the death and PMt is the particles' average levels of the same and previous day of death t (lags 0-1). The smooth functions s capture the non-linear relationship between the time-varying covariates and calendar time and daily mortality, while ki denotes the number of basis functions. The smooth function of time serves as a proxy for any time-dependent outcome predictors or confounders with long-term trends and seasonal patterns not explicitly included in the model. Hence we removed long-term trends and seasonal patterns from the data to guard against confounding by omitted variables. We used penalized regression splines as smoothing functions, as implemented by Wood in R (Wood, 2000). We chose the degrees of freedom (df) that minimized the sum of the absolute value of partial autocorrelations (PACF) of the residuals with a minimum of 3 df per year in all models. To control for potential weather confounding effects we included smooth terms for the same and previous day temperature using a natural spline with 3 df for each term and a linear term for relative humidity. We included dummy variables for the day of the week effect and holidays. As there were not available influenza data, we included a dummy variable to control for influenza epidemics that was assigned the value of 1 when the 7-day moving average of respiratory mortality was greater than the 90th percentile of its distribution. We did not adjust for influenza when we analyzed respiratory mortality, as the influenza indicator was based on the respiratory mortality distribution (Touloumi et al., 2005). Finally, we controlled for the summer decrease in population during the summer vacation period using a three-level ordinal variable assigned a value of 2 during the 2-week period around mid-August, 1 from July 16 to August 31 (with the exception of the 2-week period under value 2), and 0 (reference category) on the remaining days (Samoli et al., 2013). We estimated the effects of PM10 on mortality during the period 2001–12 by including a linear term in the model of the average levels of the same and previous day of death (lags 0-1) to address acute effects and the average over 6 days (lags 0–5) to address prolonged effects. When we assessed the effect of prolonged exposure we controlled for temperature using a corresponding lag structure namely we included a natural spline with 3 df for same day temperature (lag 0) and a natural spline with 3 df for lags 1–5. To investigate the long term trends in mortality risks we followed Dominici et al. (2007) and introduced a linear function of calendar time for the period 2001–12, i.e. β(t) = α0 + α1t; [2] where β(t) denotes the particles' effect estimate as a function of time t (years), α0 is the mortality risk associated with PM10 (lags 0-1) in the baseline year and α1 is the average slope, interpreted as the change in mortality risk associated with a change in time of one year. To assess the association with time-

2. Material and methods 2.1. Material The Athens area forms a basin surrounded by mountains in the north, east and northwest and by the sea on the southwest side. The topography favors atmospheric inversion and the concentrations of the pollutants measured are high. The population in the greater Athens area was over 3 million inhabitants in 2004 (Eurostat, 2008). We obtained daily counts of all-cause mortality excluding deaths from external causes (International Classification of Disease ICD-9 < 800), cardiovascular mortality (ICD-9: 390–459) and respiratory mortality (ICD9:460–519), for all ages, by age group (≥75 and < 75 years) for the greater Athens area over the period 2001–2012 from the Hellenic Statistical Authority. Daily air pollution measurements for PM10 (24-h mean) and gaseous pollutants (nitrogen dioxide (NO2, 1-h max), ozone (O3, 8-h max), sulfur dioxide (SO2, 24-h mean) and carbon monoxide (CO, 8-h max)) were provided by the monitoring network operated by the Ministry of Environment, Energy and Climate Change (www.minenv.gr). Data were obtained from urban or suburban background sites or, when representing exposure of nearby population, from fixed monitors located near traffic that provided data for at least 75% of the days in the analyzed period. Four sites fulfilled these criteria for PM10 measurements, while gaseous pollutants measurements were available from more sites (4–9 sites). Missing values in the station-specific time-series were replaced by a weighted average from the available stations (Katsouyanni et al., 2001). Consequently, monitor-specific concentrations were averaged to obtain the pollutant's daily time-series for 2001–12. The resulting time-series were almost complete with < 0.8% missing data. 2

Environment International xxx (xxxx) xxx–xxx

K. Tzima et al.

3. Results

varying risk we also allowed for a spline function for the interaction term and compared the linear vs the spline model using the Akaike Information Criterion (AIC). We further included the year of death in [2] as an ordered variable to get year-specific effect estimates. To assess the potential effect of the economic crisis in Greece on mortality risks associated with particles we defined a priori 2008 as the onset year of the crisis and investigated the risk before and after that year by introducing an indicator variable for 2008–12 vs 2001–2007 and its interaction with PM10. To investigate potential confounding effects from other pollutants, we also applied two pollutant models, i.e. in addition to PM10, we alternatively included SO2, NO2, O3 or CO in our model. We assessed the sensitivity of our findings to the definition of the year for the onset of the economic crisis using as cut off years either 2007 or 2009 instead of 2008. We further assessed the sensitivity of our findings for total mortality to the main model definition by using alternative measures for temperature control and dropping the influenza indicator variable. We applied two sensitivity models: 1) one using different metrics for temperature to decrease possible concurvity, hence instead of mean temperature we used the maximum temperature for lag0 and the minimum temperature for lag 1, while we removed the indicator variable for the influenza epidemics; and 2) one model extending the control for minimum temperature to reflect prolonged cold effects over three days by using lags 1–3 instead of lag1 in the previous sensitivity model. To further investigate the trend in the particles-mortality association we extrapolated measured PM2.5 data that were available only after 2007 to obtain a time-series for the whole period 2001–12. As the ratio PM2.5/PM10 over the period 2007–12, when both measurements were available, was rather constant (median range: 0.51 to 0.60) we back-extrapolated the PM2.5 series using the median over 2007–12 ratio of 0.54. All analyses were performed using the R v3.3.2 statistical package (R Foundation for Statistical computing 2016). We report our results as percent change in daily mortality associated either with 10 μg/m3 increase in PM10 levels or per interquartile range (IQR) increase in PM10 in order to account for possible long term change in the absolute levels of the particles during the period analyzed.

Table 1 presents the distribution of the pollutants, the meteorological variables and mortality outcomes in Athens, Greece for 2001–12. Median PM10 concentration was 40.25 μg/m3 with interquartile range 22.9 μg/m3. The correlations between PM10 and the other pollutants ranged from −0.09 with O3 to 0.64 with NO2. Supplementary Fig. S1 presents the distribution of PM10 and temperature per year during 2001–12 in Athens, Greece indicating a decrease both in the median value and the variance of the particles' levels over the years, while no trend in temperature is observed. The mean daily number of all deaths was 81 of which 53 were deaths among those aged ≥75 years. We report statistically significant associations between mortality outcomes and PM10 for the whole period 2001–12 (Supplementary Table S1), that were stronger for respiratory and cardiovascular deaths. A 10 μg/m3 increase was associated with a 0.73% increase (95% Confidence Ιnterval (CI): 0.45%, 1.01%) in all-cause mortality, and 0.86% (95% CI: 0.47%, 1.26%) and 1.76% (95% CI: 0.94%, 2.58%) increases in cardiovascular and respiratory mortality correspondingly. Our estimates were robust to co-pollutant adjustment. The linear trend slopes over time in the PM10 effect on all-cause, cardiovascular and respiratory mortality indicated significant changes over time for the first two outcomes (p-values 0.01, 0.04 and 0.13 correspondingly). The percent changes in the risk associated with 1 in equation [2]) were 0.09% (95% CI: 10 μg/m3 increase in particles (a 0.02%, 0.16%), 0.11% (95% CI: 0.01%, 0.21%) and 0.16% (95% CI: −0.05%, 0.37%) respectively, for one year change in calendar time, that support increasing risks over the period 2001–12 on average, especially for all-cause and cardiovascular mortality. The model using a spline function for the time-varying risk did not improve the AIC of the models (5421 in the linear vs 5438 in the spline model for total mortality all ages; 5154 vs 5167 and 4824 vs 4836 respectively in cardiovascular and respiratory mortality all ages). Fig. 1 presents the percent change in mortality associated with 10 μg/m3 increase in PM10 from 2001 to 2012. After 2008, there is indication for an increase in the mortality risks, consistent in all outcomes. Table 2 and Fig. 2 present the percent increase in mortality associated with an interquartile and a 10 μg/m3 increase in PM10 for 2001–12, 2001–07 and 2008–12, for all ages and by age group. While

Table 1 Descriptive statistics for the pollutants, meteorology and mortality outcomes in Athens, Greece during 2001–12. Mean (standard deviation)

Minimum

Percentile 25

Pollutants PM10 (μg/m3) NO2 (μg/m3) O3 (μg/m3) CO (mg/m3) SO2 (μg/m3) Meteorological variables Temperature (°C) Relative humidity (%) Daily number of deaths All deaths All ages < 75 years ≥75 years Cardiovascular All ages < 75 years ≥75 years Respiratory All ages < 75 years ≥75 years

Number of days 4,344 4,358 4,358 4,358 4,347

th

Maximum th

50

75

th

44.1 (19.5) 87.4 (30.1) 64.5 (26.3) 1.7 (1.0) 14.4 (11.0)

6.0 23.1 4.7 0.3 1.7

30.5 66.2 43.2 1.0 6.8

40.3 83.6 65.1 1.4 11.1

53.4 104.0 85.6 2.1 18.2

146.2 261.7 144.0 7.6 80.4

4,358 4,358 Sum of deaths

18.9 (7.6) 63.1 (15.1)

−6.9 20.3

12.9 51.6

18.3 64.3

25.5 74.3

36.4 100.0

349,595 117,156 232,439

81 (13) 27 (6) 53 (10)

40 10 21

71 23 46

79 27 53

88 31 60

128 55 94

161,824 42,943 118,881

37 (8) 10 (4) 27 (7)

14 1 9

32 7 23

36 9 27

42 12 31

74 26 57

35,224 7,569 27,655

8 (3) 2 (1) 6 (3)

0 0 0

6 1 4

8 2 6

10 3 8

25 8 20

3

Environment International xxx (xxxx) xxx–xxx

K. Tzima et al.

the magnitude of the effects slightly increased during the period following 2008, the difference between periods was significant only when we considered all-cause mortality (p-value for interaction 0.03 for allcause, 0.38 for cardiovascular and 0.11 for respiratory mortality), and was driven by the difference in the risks observed among those ≥75 years. The percent increase in mortality in this age group before 2008 was 1.51% per 10 μg/m3 particles' increase, while after 2.61% (pvalue for interaction 0.01). Moreover we found a significant change in risk in cardiovascular deaths for those ≥75 years. It is interesting that regarding respiratory mortality, the risks among those below 75 years increased between the two periods (4.88% in 2008–12 vs 0.66% in 2001–07, p-value for interaction 0.09) reaching the 10% statistical significance level despite the large confidence intervals observed due to the small counts. On the contrary there was a non-significant decrease in the risk for cardiovascular mortality among those < 75 years that contributed to essentially no change for the overall mortality in this age group. When we considered prolonged exposures, effect estimates for the whole period were similar to those following acute exposure with indications of higher effects for respiratory mortality among those < 75yeras. The comparison of effects between periods was in accordance with the findings for lags 0–1, although only total and cardiovascular mortality effects among ≥75 years were statistically significantly higher after 2008 at the 5% and 10% level correspondingly. Adjusting for other pollutants did not affect the effect estimates or the interaction estimates in any of the models (Supplementary Table S2). When we assessed the sensitivity of our findings to the definition of the start year for the economic crisis (Supplemental Table S3) our results related to total and cardiovascular mortality among those ≥75 years were robust with statistically significantly greater risks in the later period as defined each time. Respiratory mortality effect estimates were more variable with largely overlapping confidence intervals, but nevertheless they were comparable when 2007 instead of 2008 was used as a cut off year. Effect estimates for total mortality all ages or among those ≥75 years were almost identical when we used min or max temperature metrics instead of mean daily temperature (Supplemental Table S4), while they were slightly decreased when we allowed for prolonged effects of temperature (lags 1–3), although they remained higher after 2008 and statistical significant at least at the 10% level. Finally the analysis of the PM2.5 (Supplementary Table S5) estimated significant associations with all mortality outcomes for 2001–12 except for respiratory mortality for those < 75 years. All effect estimates were stronger in 2008–12 compared with 2001–07 except for cardiovascular mortality for those < 75 years. In particular an IQR increase in PM2.5 in 2001–07 was associated with 0.96% (95% CI: 0.26%, 1.66%) increase in total mortality all ages and 2.01% (95% CI: 1.21%, 2.83%) in 2008–12 (p-value for the interaction < 0.001). 4. Discussion We have investigated trends in mortality associated with short-term particles' exposure in Athens, Greece during 2001–12 and report increasing risks. Despite the decline in PM10 levels over the years, we report an increase in the associated total mortality risk, both when considering the risk for a 10 μg/m3 or an IQR increase in each period, that is driven mainly by effects following 2008–the start year of the severe economic crisis in Greece. The increased risk was more evident for all-cause and cardiovascular mortality among those aged ≥75 years, while there were indications of increased risk for respiratory mortality among the age group < 75 years. The findings that risks for those ≥75 years drive the results for all ages analysis are consistent with previous ones (Samoli et al., 2013), while the indication for a changing pattern in respiratory mortality should be further evaluated as its interpretation is hindered also by the small counts and the large confidence intervals. Although health effects of short-term particles' exposure have been

Fig. 1. Percent increase (95% confidence intervals (CI)) in A) total, B) cardiovascular and C) respiratory mortality associated with 10 μg/m3 increase in PM10 (lags 0–1) between 2001 and 2012 in Athens, Greece.

4

Environment International xxx (xxxx) xxx–xxx

K. Tzima et al.

Table 2 Percent increase (95% confidence intervals (CI)) in mortality outcomes associated with an interquartile increase (IQR) in PM10 in Athens, Greece in 2001–12 for acute (lags 0–1) and prolonged exposure (lags 0–5) for different time periods denoting the onset of the economic crisis.a Mortality

Lags 0–1

Lags 0–5

2001–12

2001–07

2008–12

p-value for interactionb

2001–12

2001–07

2008–12

p-value for interactionb

1.54 (0.95, 2.13) 1.60 (0.71, 2.50) 2.07 (1.36, 2.79)

1.19 (0.47, 1.92) 1.43 (0.33, 2.54) 1.51 (0.62, 2.40)

1.88 (1.14, 2.63) 1.67 (0.48, 2.87) 2.61 (1.72, 3.51)

.03

1.56 (0.90, 2.23) 1.33 (0.35, 2.33) 2.10 (1.30, 2.90)

1.11 (0.37, 1.85) 0.99 (−0.11, 2.09) 1.41 (0.52, 2.31)

1.57 (0.84, 2.30) 1.30 (0.16, 2.46) 2.16 (1.30, 3.03)

.10

Cardiovascular All ages 1.81 (0.98, 2.65) < 75 years 2.55 (1.13, 4.00) ≥75 years 2.08 (1.12, 3.05)

1.69 (0.66, 2.69) 3.29 (1.56, 5.04) 1.53 (0.36, 2.72)

1.85 (0.77, 2.95) 1.21 (−0.79, 3.26) 2.69 (1.45, 3.95)

.38

2.31 (1.37, 3.26) 2.51 (0.92, 4.12) 2.64 (1.56, 3.73)

1.82 (0.80, 2.86) 2.45 (0.72, 4.21) 1.81 (0.62, 3.01)

2.11 (1.05, 3.18) 1.49 (−0.43, 3.46) 2.75 (1.54, 3.97)

.27

2.85 (0.67, 4.85) 0.66 (−3.17, 4.65) 3.98 (1.47, 6.54)

4.26 (2.20, 6.37) 4.88 (0.76, 9.15) 4.52 (2.21, 6.88)

.11

3.48 (1.57, 5.42) 4.61 (0.95, 8.40) 3.47 (1.32, 5.66)

2.37 (0.18, 4.61) 4.00 (−0.16, 8.33) 2.12 (−0.36, 4.67)

3.39 (1.40, 5.41) 3.55 (−0.44, 7.69) 3.60 (1.37, 5.87)

.22

All deaths All ages < 75 years ≥75 years

Respiratory All ages < 75 years ≥75 years

3.72 (1.99, 5.49) 2.61 (−0.68, 6.01) 4.52 (2.57, 6.51)

.43 .01

.27 .04

.09 .31

.44 .04

.71 .06

.85 .17

a Results from Poisson models adjusted for seasonality, temperature, relative humidity, day of the week, holidays, influenza and summer population decrease. IQR in lags 0-1 PM10: 21.35 μg/m3, 21.52 μg/m3 and 16.63 μg/m3 for 2001–12, 2001–07 and 2008–12 respectively; IQR in lags 0-5 PM10: 18.31 μg/m3, 16.95 μg/m3 and 13.0.3 μg/m3 for 2001–2012, 2001–07 and 2008–12 respectively. b Models including interaction terms for the comparison of effects between 2001–07 and 2008–12.

2009), U.S. for 1987–2000 (Dominici et al., 2007) and Lombardy, Italy (Carugno et al., 2017) in 2003–06, although the associations remained significant in any case. Constant total mortality effects following shortterm exposure to particles were reported for Rome in 2001–14 (Renzi et al., 2017) and Switzerland (Perez et al., 2015) during 2001–10. Furthermore, Breitner et al. (2016) using data from four urban areas in Finland, Denmark, Germany, and Italy between 1999 and 2010 reported that particles' number concentration was associated with a small non-statistically significant increase in mortality risk between 1999 and 2002 and 2007–09, while there was no temporal change in the shortterm effects of PM2.5 on all-cause mortality. On the contrary, Kim et al. (2015) in Seoul, Korea reported that although the levels of the particles have declined there were indications for increases in total and causespecific mortality risks in 2007–11 vs 2002–06. Nevertheless, possible between areas differences in physical and chemical characteristics of particles, prohibit direct comparisons between studies. Furthermore, previous research has not investigated areas undergoing a severe economic crisis as in the case of Athens, Greece that had direct impacts to all aspects of life including environmental and public health ones.

well documented over the past years (WHO, 2013) there is always a need to update scientific knowledge due to the continuous changing profile of the air pollution mixture following policy interventions, as well as changes in climatic conditions and in emission sources. Public health policies are best guided when they are based on recent findings. As the changing physical and societal environment is dependent on associated policy measures, it is a constant necessity to assess scientific knowledge related to policy interventions. The severe economic crisis in Greece that started in 2008 has impacts on the urban air quality, as well. Gratsea et al. (2017) reported that emissions from biomass combustion have gained an increasing role in atmospheric pollution levels in Athens in 2000–15, while this finding was also supported by Paraskevopoulou et al. (2015) who analyzed composition particles' data from 2008 to 2013. Limited research has been conducted to investigate possible trends in particles' effects over the years that indicated either no temporal change or small decline in Europe and U.S. following declines in air pollution concentrations. Weak declines in effects were reported in Erfurt, Germany in 1991–2002 (Breitner et al., 2009; Peters et al.,

Fig. 2. Percent increase (95% confidence intervals (CI)) in mortality outcomes associated with 10 μg/m3 increase in PM10 (lags 0–1) for the periods 2001–07 and 2008–12 in Athens, Greece.

5

Environment International xxx (xxxx) xxx–xxx

K. Tzima et al.

the population's socio-economic status and the degradation of the public health services, through the deterioration of air quality.

Wang et al. (2017) estimated that mortality attributed to long term exposure to PM2.5 in East and South Asia increased by 21% and 85% respectively, while in developed regions (i.e., Europe and high-income North America) decreased substantially (by 67% and 58% respectively). These findings are not directly comparable to ours primarily due to the focus on long term exposures, but also to the fact that the authors based their analysis on 1990–2010 annual average PM2.5 concentrations obtained from simulations that could not incorporate local trends as those that were starting to be observed in Greece. Accountability studies have been used to explore “natural experiments” and evaluate public health benefits due to air pollution improvement or regulation (Boogaard et al., 2017; Rich, 2017). Most of these studies have focused on effects of relatively short-term, localscale, and sometimes temporary interventions, while most results have associated reduction in air pollution with improvement in health indices. For example the coal ban in Ireland (Dockery et al., 2013) has resulted to reductions in cause specific mortality, although there was heterogeneity between areas. Reviews of such designs (Boogaard et al., 2017; Rich, 2017) have indicated the limitations of these designs related mostly to data availability and statistical considerations, but have also proposed future developments to improve causal inference. The consistency of our findings for the whole period 2001–12 with previous results from Athens, Greece (Samoli et al., 2013) as well as the consistency, particularly among those ≥75 years, under several sensitivity analyses support the internal validity of our study. Nevertheless the absence of measured PM2.5 and chemical composition data for the analysis period limits the assessment of our hypothesis that the reported trend may be attributed to a more toxic mixture. Unfortunately, monitoring of PM2.5 in Athens started in 2007 providing a small time window to assess our hypothesis, while composition data were measured only as part of research campaigns starting from 2008 and onwards with targeted measuring during specific time periods. To address these limitations we used extrapolated PM2.5 data that provided consistent conclusions with our PM10 analysis, although part of the difference observed between the two periods may also be attributed to differential measurement error, as data before 2008 are a combination of measurements in 2007 and back extrapolations, while after 2008 all data were provided by monitors. To further test our hypothesis of increasing particles' toxicity over the years, we analyzed black smoke (BS) time series data available until 2004 from fixed sites to compare them with previously published results on black carbon (BC), that is highly correlated to BS, for Athens during 2008–09 (Ostro et al., 2015). An IQR increase in BS levels (lags 0–1) in 2001–04 was associated with 0.85% change (95% CI: −0.16%, 1.87%) in total mortality all ages and 0.94% (95% CI: −0.29%, 2.19%) among those ≥75 years; while corresponding changes for cardiovascular mortality were 0.86% (95% CI: −0.46%, 2.20%) and 0.53% (95% CI: −1.02%, 2.10%); and for respiratory mortality 0.49% (95% CI: −2.58%, 3.65%) and 1.05% (95% CI: −2.45%, 4.67%). For 2008–09, Ostro et al. (2015) have reported higher risks per IQR change in previous day levels (for total mortality 1.8% (95% CI: 0.3%, 3.3%) for all ages and 3.6% (95% CI: 0.4%, 7.0%) for ≥75 years; for cardiovascular 1.2% (95% CI: −1.0%, 3.4%) and 2.3% (95% CI: −2.1%, 7.0%) respectively; and for respiratory mortality 8.4% (95% CI: 4.3%, 12.7%) and 18.4% (95% CI: 9.1%, 28.4%)). Despite the high BS/BC correlation, reported differences, however, may be attributed to different measurement error by pollutant, statistical power associated with the length of the time-series, statistical model (Poisson vs case-cross over analysis) and different lag structure. Further limitations inherent in the time-series ecological design include measurement error, although there is no reason to expect that this would differ by period and drive our estimates. In conclusion, our results indicate that despite the decline in particles' concentration in Athens, Greece in the recent years the associated mortality risk has possibly increased, suggesting that the economic crisis has effects in public health, beyond those directly associated with the deterioration of

Acknowledgements We thank the Hellenic Statistical Authority for the provision of the health data. This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. Appendix A. Supplementary data Supplementary data to this article can be found online at https:// doi.org/10.1016/j.envint.2018.01.002. References Boogaard, H., van Erp, A.M., Walker, K.D., Shaikh, R., 2017. Accountability studies on air pollution and health: the HEI experience. Curr. Environ. Health Rep. 4 (4), 514–522 2017 Dec. Breitner, S., Stölzel, M., Cyrys, J., Pitz, M., Wölke, G., Kreyling, W., Küchenhoff, H., Heinrich, J., Wichmann, H.E., Peters, A., 2009. Short-term mortality rates during a decade of improved air quality in Erfurt, Germany. Environ. Health Perspect. 117 (3), 448–454. Breitner, S., Samoli, E., Schneider, A., Cyrys, J., Andersen, Z.J., Forastiere, F., Lanki, T., Pekkanen, J., Stafoggia, M., Peters, A., on behalf of the UF&HEALTH Study Group, 2016. Association between short-term exposure to ultrafine and fine particles and mortality in four European urban areas: have the effects changed over time? In: Abstracts of the 2016 Epidemiology (ISEE). Abstract [4276]. Environmental Health Perspectives, Research Triangle Park, NC. http://dx.doi.org/10.1289/ehp.isee2016. Carugno, M., Consonni, D., Bertazzi, P.A., Biggeri, A., Baccini, M., 2017. Temporal trends of PM10 and its impact on mortality in Lombardy, Italy. Environ. Pollut. 227, 280–286. Dockery, D.W., Rich, D.Q., Goodman, P.G., Clancy, L., Ohman-Strickland, P., George, P., Kotlov, T., HEI Health Review Committee, 2013. Effect of air pollution control on mortality and hospital admissions in Ireland. Res. Rep. Health Eff. Inst. 176, 3–109. Dominici, F., Peng, R.D., Zeger, S.L., White, R.H., Samet, J.M., 2007. Particulate air pollution and mortality in the United States: did the risks change from 1987 to 2000? Am. J. Epidemiol. 166 (8), 880–888. Gratsea, M., Liakakou, E., Mihalopoulos, N., Adamopoulos, A., Tsilibari, E., Gerasopoulos, E., 2017. The combined effect of reduced fossil fuel consumption and increasing biomass combustion on Athens' air quality, as inferred from long term CO measurements. Sci. Total Environ. 592, 115–123. Hellenic Statistical Authority (ELSTAT) www.statistics.gr, Accessed date: 23 May 2017. Karanasiou, A., Querol, X., Alastuey, A., Perez, N., Pey, J., Perrino, C., Berti, G., Gandini, M., Poluzzi, V., Ferrari, S., de la Rosa, J., Pascal, M., Samoli, E., Kelessis, A., Sunyer, J., Alessandrini, E., Stafoggia, M., Forastiere, F., MED-PARTICLES Study Group, 2014. Particulate matter and gaseous pollutants in the Mediterranean Basin: results from the MED-PARTICLES project. Sci. Total Environ. 488–489, 297–315. Katsouyanni, K., Touloumi, G., Samoli, E., Gryparis, A., Le Tertre, A., Monopolis Yet al., 2001. Confounding and effect modification in the short-term effects of ambient particles on total mortality: results from 29 European cities within the APHEA2 project. Epidemiology 12, 521–531. Kim, H., Kim, H., Lee, J.T., 2015. Effects of ambient air particles on mortality in Seoul: have the effects changed over time? Environ. Res. 140, 684–690. Ostro, B., Tobias, A., Karanasiou, A., Samoli, E., Querol, X., Rodopoulou, S., Basagaña, X., Eleftheriadis, K., Diapouli, E., Vratolis, S., Jacquemin, B., Katsouyanni, K., Sunyer, J., Forastiere, F., Stafoggia, M., MED-PARTICLES Study Group, 2015. The risks of acute exposure to black carbon in Southern Europe: results from the MED-PARTICLES project. Occup. Environ. Med. 72, 123–129. Paraskevopoulou, D., Liakakou, E., Gerasopoulos, E., Mihalopoulos, N., 2015. Sources of atmospheric aerosol from long-term measurements (5 years) of chemical composition in Athens, Greece. Sci. Total Environ. 527-528, 165–178. Perez, L., Grize, L., Infanger, D., Künzli, N., Sommer, H., Alt, G.M., Schindler, C., 2015. Associations of daily levels of PM10 and NO₂ with emergency hospital admissions and mortality in Switzerland: trends and missed prevention potential over the last decade. Environ. Res. 140, 554–561. Peters, A., Breitner, S., Cyrys, J., Stölzel, M., Pitz, M., Wölke, G., Heinrich, J., Kreyling, W., Küchenhoff, H., Wichmann, H.E., 2009. The influence of improved air quality on mortality risks in Erfurt, Germany. Res. Rep. Health Eff. Inst. 137, 5–77 (discussion 79-90). Renzi, M., Stafoggia, M., Faustini, A., Cesaroni, G., Cattani, G., Forastiere, F., 2017. Analysis of temporal variability in the short-term effects of ambient air pollutants on nonaccidental mortality in Rome, Italy (1998–2014). Environ. Health Perspect. 125 (6), 067019. Rich, D.Q., 2017. Accountability studies of air pollution and health effects: lessons learned and recommendations for future natural experiment opportunities. Environ. Int. 100, 62–78 2017 Mar. Samoli, E., Peng, R., Ramsay, T., Pipikou, M., Touloumi, G., Dominici, F., Burnett, R.,

6

Environment International xxx (xxxx) xxx–xxx

K. Tzima et al.

Department of Chemistry, National and Kapodistrian University of Athens, Athens, Greece, . http://www.chem.uoa.gr/wp-content/uploads/epistimonika_themata/ atmosph_pollut_greece.pdf, Accessed date: 30 August 2017. Wang, J., Xing, J., Mathur, R., Pleim, J.E., Wang, S., Hogrefe, C., Gan, C.M., Wong, D.C., Hao, J., 2017. Historical trends in PM2.5-related premature mortality during 1990–2010 across the Northern Hemisphere. Environ. Health Perspect. 125 (3), 400–408. WHO, 2013. Review of evidence on health aspects of air pollution — REVIHAAP Project. In: Technical Report. WHO Publications, WHO Regional Office for Europe, Copenhagen, Denmark. Wood, S.N., 2000. Modelling and smoothing parameter estimation with multiple quadratic penalties. J. Royal Stat. Soc. B 62, 413–428.

Cohen, A., Krewski, D., Samet, J., Katsouyanni, K., 2008. Acute effects of ambient particulate matter on mortality in Europe and North America: results from the APHENA study. Environ. Health Perspect. 116, 1480–1486. Samoli, E., Stafoggia, M., Rodopoulou, S., et al., 2013. Associations between fine and coarse particles and mortality in Mediterranean cities: results from the MEDPARTICLES project. Environ. Health Perspect. 121, 932–938. Touloumi, G., Samoli, E., Quenel, P., Paldy, A., Anderson, R.H., Zmirou, D., Galan, I., Forsberg, B., Schindler, C., Schwartz, J., Katsouyanni, K., 2005. Short-term effects of air pollution on total and cardiovascular mortality: the confounding effect of influenza epidemics. Epidemiology 16, 49–57. Valavanidis, A., Vlachogianni, T., Loridas, S., Fiotakis, C., 2015. Atmospheric pollution in urban areas of Greece and economic crisis. Trends in air quality and atmospheric pollution data, research and adverse health effects. In: Internal Report of the

7