Published on

L’aria è elemento essenziale per la vita dell’uomo.La “mission” di questo blog è quello di soddisfare le esigenze di ricerca e di conoscenza delle tecnologie che possono permettere alle persone di respirare ogni giorno un’aria più pulita e sana, migliorando la qualità e la durata della loro vita.

Published in: Health & Medicine, Technology

    L’aria è elemento essenziale per la vita dell’uomo.La “mission” di questo blog è quello di soddisfare le esigenze di ricerca e di conoscenza delle tecnologie che possono permettere alle persone di respirare ogni giorno un’aria più pulita e sana, migliorando la qualità e la durata della loro vita.
    Are you sure you want to  Yes  No
    Your message goes here
  • Be the first to like this


  1. 1. ResearchAcute Effects of Ambient Particulate Matter on Mortality in Europe andNorth America: Results from the APHENA StudyEvangelia Samoli,1 Roger Peng,2 Tim Ramsay,3 Marina Pipikou,1 Giota Touloumi,1 Francesca Dominici,2Rick Burnett,3,4 Aaron Cohen,5 Daniel Krewski,3 Jon Samet,6 and Klea Katsouyanni 11Department of Hygiene and Epidemiology, University of Athens Medical School, Athens, Greece; 2Department of Biostatistics, JohnsHopkins Bloomberg School of Public Health, Baltimore, Maryland, USA; 3McLaughlin Centre for Population Health Risk Assessment,University of Ottawa, Ottawa, Ontario, Canada; 4Environmental Health and Consumer Products Branch, Health Canada, Ottawa, Ontario,Canada; 5Health Effects Institute, Boston, Massachusetts, USA; 6Department of Epidemiology, Johns Hopkins Bloomberg School ofPublic Health, Baltimore, Maryland, USA mortality and hospital admissions. The pro- BACKGROUND: The APHENA (Air Pollution and Health: A Combined European and North ject originated at a time when the results of American Approach) study is a collaborative analysis of multicity time-series data on the effect of the multicity analyses, including APHEA and air pollution on population health, bringing together data from the European APHEA (Air NMMAPS, were being reported and consid- Pollution and Health: A European Approach) and U.S. NMMAPS (National Morbidity, Mortality ered in the development of ambient air qual- and Air Pollution Study) projects, along with Canadian data. ity standards for PM (European Commission OBJECTIVES: The main objective of APHENA was to assess the coherence of the findings of the 1999; World Health Organization 2004, multicity studies carried out in Europe and North America, when analyzed with a common proto- 2006). The main objective of the project was col, and to explore sources of possible heterogeneity. We present APHENA results on the effects of to assess the coherence of the findings of the particulate matter (PM) ≤ 10 µm in aerodynamic diameter (PM10) on the daily number of deaths for all ages and for those < 75 and ≥ 75 years of age. We explored the impact of potential environ- multicity studies carried out in Europe and mental and socioeconomic factors that may modify this association. North America, when analyzed with a com- mon protocol, and to explore reasons for any METHODS: In the first stage of a two-stage analysis, we used Poisson regression models, with nat- ural and penalized splines, to adjust for seasonality, with various degrees of freedom. In the second observed differences in the size of the air stage, we used meta-regression approaches to combine time-series results across cites and to assess pollution relative rates. effect modification by selected ecologic covariates. In this article, we present the APHENA RESULTS: Air pollution risk estimates were relatively robust to different modeling approaches. Risk findings on the association between daily estimates from Europe and United States were similar, but those from Canada were substantially higher. The combined effect of PM10 on all-cause mortality across all ages for cities with daily air Address correspondence to E. Samoli, Department of Hygiene and Epidemiology, University of Athens pollution data ranged from 0.2% to 0.6% for a 10-µg/m3 increase in ambient PM10 concentration. Medical School, 75 Mikras Asias St., 115 27 Athens, Effect modification by other pollutants and climatic variables differed in Europe and the United Greece. Telephone: 30-210-7462085. Fax: 30-210- States. In both of these regions, a higher proportion of older people and higher unemployment 7462205. E-mail: were associated with increased air pollution risk. Supplemental Material is available online at CONCLUSIONS: Estimates of the increased mortality associated with PM air pollution based on the APHENA study were generally comparable with results of previous reports. Overall, risk estimates suppl.pdf were similar in Europe and in the United States but higher in Canada. However, PM10 effect modi- The APHENA core group consists of H.R. fication patterns were somewhat different in Europe and the United States. Anderson, R. Atkinson, R. Burnett, F. Dominici, K. Katsouyanni, D. Krewski, A. Le Tertre, S. Medina, KEY WORDS: air pollution, effect modification, heterogeneity, meta-regression, mortality, natural R. Peng, T. Ramsey, J. Samet, E. Samoli, J. Schwartz, splines, particulate matter, penalized splines, time-series analysis. Environ Health Perspect G. Touloumi, and A. Zanobetti. The core group 116:1480–1486 (2008). doi:10.1289/ehp.11345 available via [Online 26 June 2008] APHEA-2 data providers were H.E. Wichmann (Germany); J. Sunyer (Spain); M.A. Vigotti, L. Bisanti, and P. Michelozzi (Italy); D. Zmirou (Grenoble, France); J. Schouten (The Netherlands);Hundreds of time-series studies worldwide pro- methods and data characteristics differ, pre- J. Pekkanen (Finland); L. Clancy (Ireland); A. Gorenvide compelling evidence of the health effects cluding definitive conclusions about their (Israel); C. Schindler (Switzerland); B. Wojtyniakof short-term exposure to air pollution. These quantitative consistency and about the extent (Poland); B. Kriz (Prague, Czech Republic); A. Paldystudies also pose problems of interpretation due of and reasons for differences in the magnitude (Hungary); E. Niciu (Romania); M. Macarol-Hittito variation in analytic methods and reporting, of the effects of short-term exposure among (Slovenia); B. Forsberg (Sweden); F. Kotesovecand the possibility of publication and analytic regions of the world. (Teplice, Czech Republic); and M. Pavlovic (Croatia). Research described in this article was conductedbias. Meta-analyses of published results can APHENA (Air Pollution and Health: A under contracts with the European Commissionprovide information about patterns in the rela- Combined European and North American Climate Programme (contract QLK4-CT-2002-tive rates of mortality and morbidity and evi- Approach) is a collaborative study among 30226) and the Health Effects Institute (HEI; con-dence as to the causes of their spatial variation, investigators involved in the European tracts HEI 039-2 and 4737/RFPA98-6/05-11). HEIbut they inherit many of the same limitations APHEA (Air Pollution and Health: A is an organization jointly funded by the U.S.of the individual studies. Coordinated multicity European Approach) study (Atkinson et al. Environmental Protection Agency (EPA) (assistance agreement R82811201) and automotive manufac-studies, designed partly to address these issues, 2001; Gryparis et al. 2004; Katsouyanni et al. turers. The contents of this article do not necessarilyhave now been conducted in Europe and 1997, 2001) and the U.S. NMMAPS reflect the views of HEI, nor do they necessarilyNorth America (Atkinson et al. 2001; Bell et al. (National Morbidity, Mortality and Air reflect the views and policies of the U.S. EPA or of2004; Burnett and Goldberg 2003; Burnett Pollution Study) study (Bell et al. 2004; motor vehicle and engine manufacturers. Additionalet al. 1998, 2000; Gryparis et al. 2004; Samet et al. 2000a, 2000b, 2000c), as well as support was provided by a career award from theKatsouyanni et al. 1997, 2001; Samet et al. Canadian studies (Burnett and Goldberg Natural Sciences and Engineering Research Council of Canada and the Social Sciences and Humanities2000a, 2000b, 2000c) and currently provide 2003; Burnett et al. 1998, 2000). APHENA Research Council of Canada to D.K.the most valid epidemiologic evidence of the addresses the short-term health effects of The authors declare they have no competingeffects of short-term exposure. The results of particulate matter (PM) ≤ 10 µm in aero- financial interests.these studies appear broadly similar, but their dynamic diameter (PM10) and ozone on daily Received 8 February 2008; accepted 26 June 2008.1480 VOLUME 116 | NUMBER 11 | November 2008 • Environmental Health Perspectives
  2. 2. Effects of particulate matter on mortalitymeasurements of PM10 and mortality. The weather. City characteristics both within and variation. In the second analytic stage, weresults, spanning two continents with a wide between the three collaborating centers used meta-regression to obtain center-specificrange of sources of ambient air pollution, are (Europe, United States, and Canada) exhibit (Canada, Europe, United States) and overallrelevant to one of the key uncertainties in our substantial variability. We provide more details estimates of risk based on the city-specific riskcurrent understanding of the health effects of on the data in the Supplemental Material estimates, and to investigate potential city-PM: the identification of those characteristics (online at level effect modifiers.of PM that are associated with toxicity 2008/11345/suppl.pdf). Individual city analysis. We investigated(National Research Council 2004). Current We restricted analysis to days with PM10 the PM10–mortality associations for each cityregulatory standards are based on overall indi- concentrations < 150 µg/m3, because the rela- using Poisson regression models allowing forcators of airborne PM mass as concentration tionship between air pollution and mortality overdispersion. The city-specific model is ofmetrics, in the face of uncertainty as to the within this range is effectively linear (Dominici the formspecific physical and chemical characteristics et al. 2002a; Samoli et al. 2005). This restric-that determine toxicity. The present study tion led us to exclude < 2% of the available ⎣ ⎦ t ( log E ⎡Y t c ⎤ = βo c + b c × PMc + s c timet c ,k )permits exploration of heterogeneity in the days in each city, except for three European c ( c ) + ∑ si x it ,ki + [ others ] , [1]effect of PM10 on mortality across the broad cities (Erfurt, Prague, and Turin), in which irange of atmospheres included in the we excluded 3–6% of the total number of c] where E[Y t is the expected value of theAPHENA cities. available days. Poisson distributed variable Ytc indicating the Any assessment of heterogeneity needs to Methods. When APHENA was initiated, daily mortality count on day t at city c withaddress the potential consequences of using dif- there was ongoing debate about the use of var(Ytc) = ϕE[Ytc], with ϕ representing thefering analytic strategies to estimate air pollution time-series methods to describe the relation- overdispersion parameter; PMtc is the air pol-health risks, and the extent to which apparent ship between air pollution and health after lution concentration on day t in city c; and xitcheterogeneity across studies reflects the conse- Dominici et al. (2002b) and Ramsay et al. is the value of the xi meteorologic covariate onquences of different analytical methods. Based (2003) identified modeling issues related to day t in city c. The smooth functions s captureon past work by the APHENA investigators and nonparametric smoothing. At that time, ques- the nonlinear relationship between the time-extensive sensitivity analysis, we developed uni- tions were raised concerning the choice of varying covariates and calendar time and dailyform approaches for first-stage (within-city) smoothing method, the degree of smoothing, mortality. We used PS and NS as smoothanalyses of the time-series data used in previous and parametric versus nonparametric methods. functions, with k denoting the number ofreports. We then used the regression estimates The U.S. Environmental Protection Agency basis functions. We used NS as basis functionsin second-stage analyses directed at characteriz- (EPA) requested that the Health Effects for the PS. We also included dummy variablesing heterogeneity of the effect of PM10 across Institute (HEI) organize a systematic reanaly- for day of the week and holiday effects.the APHENA cities and identifying factors sis of selected studies to assess the sensitivity The smooth function of time serves as acontributing to heterogeneity. of the original estimates produced with non- proxy for any time-dependent outcome predic- parametric modeling strategies, using prespec- tors or confounders with long-term trends andMaterials and Methods ified alternative modeling approaches. The seasonal patterns not explicitly included in theData. APHENA was based on previously results were published in a special report (HEI model. Hence, we removed long-term trendsassembled databases in the first-stage analysis. 2003), which concluded that no particular and seasonal patterns from the data to guardThe database included the 90 U.S. cities in method could be recommended as optimal, against confounding by omitted covariates. WeNMMAPS (Samet et al. 2000a, 2000b); the and recommended that analysts should incor- used 3, 8, and 12 df for seasonality control.32 European cities in APHEA, of which 22 porate extensive sensitivity analyses to assess Additionally, for complete air pollution timehad PM10 data (Katsouyanni et al. 2001); and the adequacy of control for time-varying series (i.e., the series without systematic missing12 larger Canadian cities used in previous potential confounders. values), we used the minimization of the sum ofmulticity projects, selected on the basis of Consequently, the APHENA investigators the absolute values of the partial autocorrelationavailability of air pollution monitoring data decided to implement a new protocol for function (PACF) of the model’s residuals as(Burnett and Goldberg 2003; Burnett et al. reanalysis of the daily air pollution data from another criterion to select the optimal df, when1998, 2000). the European and North American cities. we applied the PS method. The minimum df The databases included daily counts of all- First, we fit regression models in each city allowed under the PACF was 3 df/year. Tocause mortality [excluding deaths from external separately to control for seasonal effects, control for weather, we included smooth termscauses, according to International Classification weather, and other potential confounders. of temperature on the day of death and the dayof Diseases, 9th Revision (World Health Given simulation results (Peng et al. 2006; before death in the time-series models. We usedOrganization 1975), codes > 800] for all ages Touloumi et al. 2006), we considered two these terms for temperature because in priorand by age group (≥ 75 years and < 75 years of methods to control for confounding: natural analyses we found that same-day temperatureage). We obtained air pollution measurements splines (NS), for parametric modeling of flexi- accounts for hot-weather effects, whereas previ-from fixed-site monitoring stations in each ble families of curves (McCullagh and Nelder ous-day temperature accounts for cold-weathercity. All Canadian cities and 75 U.S. cities had 1989), and penalized regression splines (PS), effects. We set the df for both temperaturePM10 measurements every 3 or 6 days. All as implemented by Wood (2000) in R. terms at 3, based on an exploratory sensitivityEuropean and 15 U.S. cities had a small num- Initial methodological exploration indi- analysis done within APHENA that indicatedber of (apparently random) days on which air cated that the number of degrees of freedom robust results with respect to differentpollution measurements were missing. The 15 (df) for control of seasonality was the most approaches for weather control. For modelsU.S. cities with full time-series data were repre- important parameter in model specification based on minimization of PACF, we intro-sentative of the total of 90 U.S. cities with with respect to the magnitude of regression duced autoregressive terms, if necessary, in caseregard to both mortality rates and air pollution coefficient reflecting the effect of air pollution significant autocorrelation remained in the finallevels. We used city-specific time-series data on on population health. We carried out exten- model’s residuals.daily temperature (°C, daily mean) to control sive sensitivity analyses in which we progres- We did not control for influenza epi-for the potential confounding effects of sively increased df to control for temporal demics because previously published resultsEnvironmental Health Perspectives • VOLUME 116 | NUMBER 11 | November 2008 1481
  3. 3. Samoli et al.have indicated that these do not affect the eight models (two smoothers and four sets of center. We repeated the analysis for all thresh-association between air pollution and mortal- df for seasonality control) in each city for lags old values. We set a possible threshold at theity (Braga and Zanobetti 2000; Touloumi 0 and 1, whereas we applied six models (two value that minimized the mean al. 2005). smoothers and three sets of df) in cities with Second-stage analysis. The epidemiologic We carried out two sets of analyses, systematically missing data for lag 1 analysis. objectives of the second-stage analysis were todepending on the availability of measurements To investigate potential confounding assess bias due, for example, to exposurein each city. We based the first on cities with effects by O 3 , we applied two-pollutant measurement error; and to assess effect modi-complete time-series data (i.e., full daily data, models that controlled for 1-hr maximum O3 fication of the air pollution relative rateswith only a few days missing at random), concentrations. across study regions. Potential effect modifierswhich encompassed all European and 15 U.S. We also carried out center-specific used in the analysis included variables describ-cities. For these cities, we used the average of (Canada, Europe, and United States) threshold ing a) the average air pollution level and mixthe same and previous day’s air pollution as a analyses to investigate the exposure–response in each city (specifically, the mean levels,predictor of increased mortality, as well as relationship between PM10 and all-cause mor- standard deviations, and coefficients of varia-unconstrained distributed lag models span- tality. We used models with NS for con- tion for PM10, nitrogen dioxide, O3, and sul-ning lags 0, 1, and 2. When fitting distributed founder control with 8 df per year for fur dioxide and the ratio of PM10 to NO2);lag models, we used the same distributed lag seasonality. We selected a grid of threshold val- b) air pollution level exposure (number ofterms for temperature as for PM10. The sec- ues, ranging from 0 to 75 µg/m3 in increments monitors and density of monitors relative toond set of analyses included all cities, regard- of 5 µg/m3 (i.e., 0, 5, 10, up to 75 µg/m3). For population size); c) the health status of theless of data availability, and assessed only the each threshold value h, we fit a threshold population (cardiorespiratory deaths as a per-effect of the previous day’s air pollution (lag model to the data for the available cities. In the centage of total mortality, crude mortality1). Because the PACF criterion is based on threshold model, we included a pollutant term rate, directly standardized mortality rate, agecontrolling the autocorrelation in time-series (x+) in the model of the form (pollutant-h)+, structure described as percentage of the popu-data, we did not apply it when we analyzed where x+ = x if x ≥ 0 and x+ = 0 if x < 0, where lation ≥ 65, ≥ 75, and < 15 years of age); andthe previous day’s air pollution, because this h is the threshold value. We then computed d) climatic conditions (mean and variance ofanalysis included cities with systematically the Akaike information criterion (AIC) value temperature and relative humidity levels, andmissing time-series data, for which case the of the fitted model for all cities in each center mean minimum and maximum daily temper-concept of autocorrelation is not straightfor- for a given threshold value, and then the aver- ature). There were few comparable socio-ward. For complete time series, we then fit age AIC for that threshold over all cities in the economic status (SES) indicators across the different countries in Europe at the city level;Table 1. Percent increase (95% CI) in the daily number of deaths (all ages and ≥ 75 and < 75 years of age) indeed, only unemployment rate (percent)associated with an increase of 10 μg/m3 in PM10 concentrations (estimated by using 8 df/year to control was available for 14 cities. Unemploymentfor seasonal patterns and PS) in the three centers. data were available for all U.S. cities. Total mortality In the second stage of the analysis, we Controlling Distributed lag assumed the city-specific effect estimates, bc,Age group/center Lag 1 for O3 (lag 1) Average of lags 0, 1 models (lags 0, 1, 2) to be normally distributed around an overallAll ages (years) estimate. To test whether variability in bc was Canada 0.84 (0.30 to 1.40) 0.76 (0.20 to 1.30) NA NA explained by city characteristics, we estimated Europe 0.33 (0.22 to 0.44) 0.32 (0.21 to 0.42) 0.29 (0.14 to 0.45) 0.20 (–0.01 to 0.42) fixed-effects pooled regression coefficients by United Statesa 0.29 (0.18 to 0.40) 0.24 (0.08 to 0.41) 0.14 (–0.12 to 0.40) 0.26 (–0.08 to 0.61) weighted regression of bc on potential effect≥ 75 years Canada 1.00 (0.25 to 1.80) 0.98 (0.18 to 1.80) NA NA modifiers (at the city level) with weights Europe 0.44 (0.29 to 0.58) 0.41 (0.27 to 0.54) 0.39 (0.19 to 0.59) 0.32 (0.04 to 0.60) inversely proportional to the variances of bc United Statesa 0.47 (0.31 to 0.63) 0.37 (0.16 to 0.59) 0.19 (–0.19 to 0.56) 0.33 (–0.16 to 0.82) (DerSimonian and Laird 1986). If we found< 75 years substantial heterogeneity across cities, beyond Canada 0.63 (–0.12 to 1.40) 0.51 (–0.26 to 1.30) NA NA the variation associated with the effect modi- Europe 0.25 (0.10 to 0.40) 0.23 (0.07 to 0.39) 0.25 (0.09 to 0.42) 0.11 (–0.20 to 0.43) fiers, we applied random-effects regression United Statesa 0.12 (–0.02 to 0.27) 0.10 (–0.13 to 0.34) 0.09 (–0.20 to 0.38) 0.20 (–0.24 to 0.63) models. These models assumed that bc was aAbbreviations: CI, confidence interval; NA, not applied because of systematically missing data. sample of independent observations from aaWe based the U.S. estimates for lag 1 on 90 cities, and the average of lags 0 and 1 and distributed lag models on 15 cities. normal distribution with the same mean and with variances equal to the between-city vari- 2.5 2.5 ance and the squared SE of bc. We estimated A B the random variance component by iteratively PS 2.0 NS 2.0 reweighted least squares (Berkey et al. 1995). Percent increase Percent increase 1.5 1.5 Based on exploratory analysis, we exam- ined potential effect modification patterns 1.0 1.0 only for cities with complete time-series data 0.5 0.5 and for the effects of the average of 2-day air pollution (lags 0 and 1) because these were 0.0 0.0 more heterogeneous and there were indica- –0.5 –0.5 tions that they were relatively insensitive to 3 8 12 3 8 12 3 8 12 3 8 12 PACF 3 8 12 PACF 3 8 12 PACF the choice of analytic method. Because there Canada (n = 12) Europe (n = 22) USA (n = 90) Europe (n = 22) USA (n = 15) Europe–USA were differences in the distribution of the df/year df/year effect modifiers between Europe and UnitedFigure 1. Percent increase in the daily number of deaths, for all ages, associated with a 10-μg/m3 increase in States, the cut points for establishing cate-PM10: lag 1 (A) and lags 0 and 1 (B) for all three centers. PACF indicates df based on minimization of PACF. gories for these variables were center specific.1482 VOLUME 116 | NUMBER 11 | November 2008 • Environmental Health Perspectives
  4. 4. Effects of particulate matter on mortalityResults combined increases in the total number of 0.29–0.73%) in the U.S. cities. The corre-Table 1 summarizes the center-specific percent deaths estimated were 0.25% with PS and sponding estimates for cardiovascular mortalityincreases in the daily number of deaths (all ages 0.18% with NS at 8 df/year, 0.21% with PS among people < 75 years of age were positiveand by age group) associated with an increase and 0.18% with NS at 12 df/year, and 0.42% but not significant. There were far fewer respi-of 10 µg/m3 in PM10 concentrations, with and with PS and 0.25% with NS using the PACF ratory deaths than cardiovascular deaths in allwithout control for O3, estimated by models criterion. On average, the PACF method three centers. The results for respiratory mortal-using 8 df/year and PS, by various lags. (Table resulted in the selection of 5–6 df per year for ity were less consistent. PM10 at lag 0 and 1 was1 presents results from the model using 8 df seasonality control. more consistently associated with increased res-per year for seasonality control, thus reporting Figure 2 shows the effects of PM air piratory mortality, again with larger effectsrelatively conservative estimates among those pollution on total mortality among persons among those ≥ 75 years of age.from the different modeling strategies applied.) ≥ 75 years of age. Air pollution risk estimates We investigated effect modification pat-For the Canadian cities, which have measure- were higher than those for all ages combined. terns only for cities with a complete PM10 timements for 1 of 6 days, only lag 1 PM10 expo- Because of the difference in effect size for series. For most of the analytic scenarios consid-sure could be considered. Similarly, most U.S. Canada compared with Europe and the ered, the time-series models produced statisti-cities had data for 1 of 6 days, so we based the United States, we do not provide combined cally significant effects of PM 10 on totalU.S. estimates for lag 1 and longer lags on dif- estimates for lag 1. Figure 3 gives the corre- mortality. However, there was significant het-ferent numbers of cities (90 and 15, respec- sponding estimates for those < 75 years of erogeneity in the city-specific estimates of thetively). Air pollution risk estimates for the age. Although the size of the PM10 effect is effects of PM10 on total mortality across all agesCanadian cities were about 2-fold higher than smaller, the combined effect for lag 0 and 1 is and among those ≥ 75 years of age. Increasingthose for Europe and the United States. We statistically significant. the df to control for seasonality decreased theestimated a lag 1 increase of 10 µg/m3 PM10 to The estimated effects of PM10 on cardio- magnitude of the air pollution effect and, con-increase the daily number of deaths by 0.84% vascular mortality (data not shown) were gen- sequently, the degree of observed heterogeneity.[95% confidence interval (CI), 0.30–1.40%] erally similar to those for total mortality. The first-stage results for the European citiesfor Canadian cities, 0.33% (95% CI, Among those ≥ 75 years of age, the effects on were more heterogeneous than those for the0.22–0.44%) for European cities, and 0.29% cardiovascular mortality were larger than those U.S. cities. Nevertheless, the European pooled(95% CI, 0.18–0.40%) for U.S. cities. These on total mortality. Specifically, we estimated results were more consistent across analyticestimates decreased slightly with adjustment lag 1 PM10 to increase the daily number of methods. A detailed presentation of thefor O3. The effect estimates for people ≥ 75 cardiovascular deaths among the elderly by APHENA second-stage analysis results is avail-years of age were consistently larger than those 1.30% (95% CI, 0.19–2.40%) in the Canadian able in the Supplemental Material (online atfor people < 75 years of age. The previous day’s cities, 0.47% (95% CI, 0.23–0.70%) in the for all ages and for the elderly were European cities, and 0.51% (95% CI, 11345/suppl.pdf).statistically significant in all three centers. When considering the average effect for A 2.5 B 2.5lags 0 and 1, we estimated an increase of PS 2.0 NS 2.0 Percent increase Percent increase0.29% (95% CI, 0.14–0.45%) in the dailynumber of deaths per 10 µg/m3 in PM10 for 1.5 1.5European cities and 0.14% (95% CI, –0.12% 1.0 1.0to 0.40%) for U.S. cities with daily PM 10 0.5 0.5measurements. The effects were higher for theolder age group compared with those 0.0 0.0< 75 years of age. The effects of cumulative –0.5 –0.5exposure, assessed with distributed lag models 3 8 12 3 8 12 3 8 12 3 8 12 PACF 3 8 12 PACF 3 8 12 PACFof lags 0–2, were somewhat lower for Canada (n = 12) Europe (n = 21) USA (n = 90) Europe (n = 21) USA (n = 15) Europe–USAEuropean cities than for U.S. cities. PM10 df/year df/yeareffect estimates did not change when con-trolled for O3 levels. When we analyzed the Figure 2. Percent increase in the daily number of deaths, among those ≥ 75 years of age, associated with a 10-μg/m3 increase in PM10: lag 1 (A) and lags 0 and 1 (B) for all three centers. PACF indicates df based oneffect of the previous day’s PM10 in the 15 minimization of PACF.U.S. cities with daily time-series data, the cor-responding estimates were comparable withthose obtained for all 90 U.S. cities, indicat- A 2.5 B 2.5 PSing that these 15 cities do not differ systemat- 2.0 NS 2.0 Percent increase Percent increaseically from the larger group of 90 cities. Figure 1 shows the sensitivity of findings to 1.5 1.5the analytic approach. Figure 1A gives mortal- 1.0 1.0ity risk estimates for lag 1 PM10 concentrations 0.5 0.5by center. We did not pool the substantiallyhigher estimates for the Canadian cities with 0.0 0.0those for the U.S. and European cities. There –0.5 –0.5is a tendency for lower estimates to be obtained 3 8 12 3 8 12 3 8 12 3 8 12 PACF 3 8 12 PACF 3 8 12 PACFwith greater values of df. Figure 1B shows the Canada (n = 12) Europe (n = 21) USA (n = 90) Europe (n = 21) USA (n = 15) Europe–USAresults for lags 0 and 1 for cities with daily data df/year df/year(Canadian cities had missing data). The pat- Figure 3. Percent increase in the daily number of deaths, among those < 75 years of age, associated with atern of variation in risk estimates with df was 10-μg/m3 increase in PM10: lag 1 (A) and lags 0 and 1 (B) for all three centers. PACF indicates df based onsimilar to that seen with the lag 1 data. The minimization of PACF.Environmental Health Perspectives • VOLUME 116 | NUMBER 11 | November 2008 1483
  5. 5. Samoli et al. Effect modification patterns were generally Europe and North America. Overall, using specific source mix differences among theconsistent across analytic methods, particularly this common protocol, PM10 was associated APHENA countries that might explain thisfor those variables having a significant modify- with increased total mortality, particularly difference. Alternatively, although no thresh-ing effect on the association between PM air among those ≥ 75 years of age, in all three old has been detected in the exposure–pollution and mortality. The effect modifica- centers (Europe, United States, and Canada), response association between ambient PM andtion patterns identified in Europe and the with the effect notably greater in Canadian mortality, there may be a log-linear associationUnited States were not always consistent. cities. Mortality risk estimates tended to between air pollution and mortality, for whichWith respect to characteristics of exposure, we decrease with increasing adjustment for lower pollution levels contribute larger risks;found that in Europe higher levels of NO2 unmeasured time-varying covariates and were under this hypothesis, the lower air pollutionand a larger NO2:PM10 ratio were associated generally lower for the average of lags 0 and 1, concentrations in Canadian cities would leadwith a greater PM10 effect on mortality. This compared with lag 1 alone. Distributed lag to higher risks. An additional explanation,pattern was also present in the United States models exhibited a different risk pattern which cannot be explored with the APHENAbut was less pronounced. In contrast, we saw a between the United States and Europe, with a data, would be that PM10 acts primarily as asmaller PM10 effect on mortality among the more prolonged effect of exposure to PM10 surrogate of the true causal pollutants and thatelderly in cities with higher O3 levels, a pat- seen in the United States. the relationship between PM10 and the toxictern mainly observed in the U.S. cities. Effect The effects of PM10 on total mortality in components differs in Canada compared withmodification by climate was evident in European and U.S. cities were quite similar. the other countries.Europe, where higher temperature and lower Based on different modeling approaches, results Several meta-analyses of the effect of PM10humidity were associated with larger PM10 from the same data sets have been previously on mortality have been reported (Andersoneffects. We found no consistent pattern of reported and were quite close, with the small et al. 2004, 2005; Pope and Dockery 2006;effect modification with temperature in the discrepancies noted possibly due to the differing Stieb et al. 2002, 2003). The combined esti-United States, and the association with modeling approaches. For the European cities, mates from single-city studies tend to behumidity tended to be inverse. When we one APHEA report (Katsouyanni et al. 2001) higher, partly because not all estimates haveinvestigated variables characterizing the age provided an estimate of 0.6% increase in the been revised subsequent to the identification ofstructure and health status of the population, daily total number of deaths per 10 µg/m3 the S-Plus convergence criteria issue (Andersonan increasing proportion of elderly people was PM10, whereas the reanalysis provided an esti- et al. 2005). Furthermore, aspects of city selec-associated with higher PM10 effects in both mate of 0.4% (HEI 2003). Similarly, original tion and model specification in the single-cityEurope and the United States. A larger pro- NMMAPS results reported in Samet et al. studies may have led to upwardly biased esti-portion of cardiorespiratory deaths among all (2000b) estimated an increase in the number mates; there is also some evidence of publica-deaths was associated with higher PM10 effects of deaths of 0.4%, and the reanalysis reported a tion bias, which would also tend to result in anonly in the United States, and there only 0.2% increase (HEI 2003) based on 90 U.S. upward bias (Anderson et al. 2005). Summaryamong the elderly. The corresponding pattern cities. Within the context of APHENA, we estimates from single-city studies range fromin Europe was nonsignificant and tended to be had the opportunity to expand the investiga- about 0.4% to 0.8% per 10 µg/m 3 PM 10the inverse. A higher crude mortality rate was tion of within-center heterogeneity previously (Pope and Dockery 2006). The European andassociated with a higher PM10 effect in the reported. The main effect modification pat- U.S. estimates in APHENA lie just below thisUnited States. The only socioeconomic factor terns identified by Katsouyanni et al. (2001) range, whereas the Canadian estimates are atavailable for all cities was the percentage of were replicated within APHENA, and several the upper end of the range.unemployed: a higher percentage of unem- new modifiers were identified as well—for European cities tend to have a higherployed was associated with greater PM air example, the modifying effect of the percent- prevalence of diesel vehicles, particularly pas-pollution effects on both continents. age of unemployed on the association between senger cars, than do cities in North America Investigation of the exposure–response PM air pollution and mortality. The signifi- (European Commission 2005; Green Carrelationship between PM10 and total mortal- cantly higher estimates observed in Canada Congress 2008); although not characterized,ity across all ages in APHENA did not sup- previously (0.8%; HEI 2003) persisted in the source inventories related to power generationport the presence of a threshold in any of the present APHENA analysis. and industry are also likely to vary, boththree centers. If a threshold were present, we Because we analyzed the data according to between and within continents. The compa-would expect to see a U-shaped curve when standardized criteria in APHENA, the higher rability of European and U.S. risk estimateswe plot the AIC values for the various thresh- values observed in Canada cannot be attributed suggests that underlying differences in PM airold models against the thresholds used, with to differences in analytic approaches. Never- pollution sources may not have substantiallythe minimum AIC value corresponding to the theless, city-specific estimates of the effect of affected the overall risk.threshold. In fact, within each center, the PM10 on mortality for Canadian cities such as One objective of APHENA was tocity-specific AIC plots were quite flat for most Toronto (in which mortality among the explore patterns of effect modification acrosscities (data not shown). elderly was increased by 1.4% for a 10-µg/m3 a wide range of geographic locations with air increase in PM10) and U.S. cities of similar pollution coming from differing source mix-Discussion population size and climate such as Detroit, tures and with populations differing inIn this article, we report the results of a com- Michigan (0.8% mortality increase), were sociodemographic characteristics.prehensive analysis of time-series data relating close. The trend toward higher estimates could In prior analyses of both single-city andPM air pollution to mortality in the general possibly be the result of more accurate expo- multicity data, a number of potential modifierspopulation in 124 cities in Europe (22 cities), sure and outcome data in Canada compared of associations between air pollution and mor-the United States (90 cities), and Canada with the European countries and the United tality have been identified (O’Neill et al. 2003;(12 cities). The analysis protocol used in the States. (At this point, we have no validation U.S. EPA 2004). Within APHEA, prior analy-APHENA study was informed by theoretical data available to explore this possibility.) ses identified modification of the effect of PM10developments, sensitivity analyses, and simula- Although the effect of PM10 on mortality may on both mortality and admissions outcomestions, which we then used to complete a com- be greater in Canada compared with the other (Aga et al. 2003; Analitis et al. 2006; Atkinsonprehensive reanalysis of time-series data from countries, we cannot readily identify any et al. 2001; Katsouyanni et al. 2001; Le Tertre1484 VOLUME 116 | NUMBER 11 | November 2008 • Environmental Health Perspectives
  6. 6. Effects of particulate matter on mortalityet al. 2002; Samoli et al. 2005). In NMMAPS, assessment audit, differing measurement error and Panel Studies of Particulate Matter (PM) and OzoneSamet et al. (2000a, 2000c) explored effect structures across the three sets of data remains a (O 3 ). Report of a WHO Task Group. Copenhagen:World Health Organization.modification extensively in the original analyses possible source of heterogeneity. Anderson HR, Atkinson RW, Peacock JL, Sweeting MJ,of the 90 cities’ mortality data and identified Although a number of potential effect Marston L. 2005. Ambient particulate matter and healthseveral potential modifiers. Both projects found modifiers have been identified, the exploration effects: publication bias in studies of short-term associa- tions. Epidemiology 16:155–163.evidence of variation by geographic region. of effect modification in APHENA was limited Atkinson RW, Anderson HR, Sunyer J, Ayres J, Baccini M,Similar two-stage analyses have not been carried by the restricted number of variables that Vonk JM, et al. 2001. Acute effects of particulate air pollu-out previously for the Canadian cities. extended across the full data set. Finally, the tion on respiratory admissions: results from APHEA 2 pro- ject. Air Pollution and Health: A European Approach. Am J We addressed effect modification in relatively small number of cities with daily data Respir Crit Care Med 164:1860–1866.APHENA in the second-stage analysis using and the large statistical uncertainty of the city- Bell ML, McDermott A, Zeger SL, Samet JM, Dominici F. variables indicative of characteristics specific estimates may have limited the power Ozone and short-term mortality in 95 U.S. urban communi- ties, 1987–2000. JAMA 292:2372–2378.of the air pollution mixture, climate, age struc- for detecting effect modification patterns. A Berkey CS, Hoaglin DC, Mosteller F, Colditz GA. 1995. A ran-ture and health status, and SES determinants. more thorough discussion on the limitations of dom-effects regression model for meta-analysis. Stat MedPM10 effect modification patterns, explored APHENA is available in the Supplemental 14:395–411.only for cities with daily data (21 European Material (online at http://www.ehponline. Braga A, Zanobetti A. 2000. Do respiratory epidemics confound the association between air pollution and daily deaths?and 15 U.S. cities), were not entirely consis- org/members/2008/11345/suppl.pdf). Eur Respir J 16:723–726.tent across centers and varied somewhat The APHENA study led to the develop- Burnett RT, Brook J, Dann T, Delocla C, Philips O, Cakmak S,depending on the underlying model and geo- ment of a standardized protocol for analyses of et al. 2000. Association between particulate- and gas- phase components of urban air pollution and daily mortal-graphic location. Pollutant levels demon- daily time-series data on air pollution and ity in eight Canadian cities. Inhal Toxicol 12(suppl 4):15–39.strated different modifying effects for cities in mortality. We pooled data from studies that Burnett RT, Cakmak S, Brook JR. 1998. The effect of the urbanEurope and the United States, which may be had been carried out in multiple cities in ambient air pollution mix on daily mortality rates in 11 Canadian cities. Can J Public Health 89:152–156.attributed to variation in the complex mix of Europe and North America. The findings Burnett RT, Goldberg MS. 2003. Size-fractionated particulateair pollutants in Europe and the United States. confirm the acute, adverse effects of PM10 on mass and daily mortality in eight Canadian cities. In:Climatic variables were important only in mortality. The use of the primary raw data to Revised Analyses of Time-Series Studies of Air Pollution and Health. Boston, MA:Health Effects Institute, 85–89.Europe. One explanation may be related to conduct pooled analyses in APHENA permits Craig L, Brook JR, Chiotti Q, Croes B, Gower S, Hedley A, et al.the lower prevalence of air conditioning in more informative analysis than can be 2008. Air pollution and public health: a guidance documentEurope, which would lead to a higher expo- achieved through a simpler meta-analysis of for risk managers. J Toxicol Environ Health A71:588–698. DerSimonian R, Laird N. 1986. Meta-analysis in clinical trials.sure of the population to outdoor air in indoor summary risk estimates taken from the pub- Control Clin Trials 7:177–188.environments. We found the most consistent lished literature. Despite the differences in Dominici F, Daniels M, Zeger SL, Samet JM. 2002a. Air pollutionevidence of effect modification for age, with data availability and comparability of effect and mortality: estimating regional and national dose-an increasing proportion ≥ 75 years associated modifiers in Europe and North America, it is response relationships. J Am Stat Assoc 97:100–111. Dominici F, McDermott A, Zeger SL, Samet JM. 2002b. On thewith a greater effect of PM 10 in both the possible to consolidate an ongoing collection use of generalized additive models in time-series studiesUnited States and Europe; higher percentage of time-series data into a single data set so that of air pollution and health. Am J Epidemiol 156:193–203.of unemployment was also associated with similar analyses can be carried out periodically European Commission. 1999. Council Directive 1999/30/EC relating to limit values for sulphur dioxide, oxides of nitro-greater risk in both continents. in the future. For the purpose of air pollution gen, particulate matter and lead in the ambient air. Offic J Larger proportions of persons ≥ 75 years control, a need exists for periodic syntheses of Eur Commun 42:41–61.of age were associated in both centers with the literature to serve as the basis for establish- European Commission. 2005. The CAFE Programme & the Thematic Strategy on Air Pollution. Impact Assessment.larger PM10 effects on mortality all ages and ing air quality guidelines at the national and Commission Staff Working Paper. Annex to: Thein this age group. This finding may be associ- international level (Craig et al. 2008); thus, Communication on Thematic Strategy on Air Pollution andated with lower baseline mortality, possibly there is a strong rationale for the ongoing col- the Directive on “Ambient Air Quality and Cleaner Air for Europe” {COM (2205)446final}. Available: http://ec.europa.leading to higher relative risks. Another plau- lection and synthesis of estimates of the health eu/environment/archives/air/cafe/index.htm [accessedsible explanation may be that in locations risks of air pollution. This rationale is particu- 12 May 2008].with larger proportions of those ≥ 75 years of larly compelling for regions of the world Gouveia N, Junger W, Ponce de Leon A, Miranda V, Hurtado M, Rojas L, et al. 2008. Air Pollution and Mortality in Latinage, the mean age of this group is also larger, where levels of air pollution are higher and America: Results from the ESCALA Project (Multi-cityleading to an excess effect. The positive effect population susceptibility may be increased, as Study of Air Pollution and Health Effects in Latin America)modification pattern associated with higher well (e.g., Asia and Latin America). APHENA [Abstract]. Available: suggests that populations with has already served as an example and a source AnnualConferenceProgram2008.pdf [accessed 12 May 2008].lower SES may be more susceptible to PM10, of methodological guidance for two ongoing Green Car Congress. 2008. European Automobile Productionas was noted by Krewski et al. (2000) in stud- coordinated multicity studies: PAPA (Public Grows by 5.3% in 2007; Diesel Accounts for 53.5% of Newies of the effects of long-term exposure to Health and Air Pollution in Asia) and Car Registrations. Available: http://www.greencarcongress. com/2008/02/european-automo.html [accessed 12 MayPM2.5 on mortality. ESCALA (Estudio de Salud y Contaminación 2008]. The proportion of cardiopulmonary del Aire en Latinoamérica) (Gouveia et al. Gryparis A, Forsberg B, Katsouyanni K, Analitis A, Touloumi G,deaths relative to the total number of deaths 2008; Wong 2008). Schwartz J, et al. 2004. Acute effects of ozone on mortality from the “Air Pollution and Health: A European Approach”displayed an opposing pattern in European project. Am J Respir Crit Care Med. 170:1080–1087.and U.S. cities. In Europe, each national REFERENCES HEI. 2003. Revised Analyses of Time-Series Studies of Airauthority was responsible for coding the Pollution and Health. Cambridge, MA:Health Effects Aga E, Samoli E, Touloumi G, Anderson HR, Cadum E, Forsberg Institute.causes of death, and that the comparability of Katsouyanni K, Touloumi G, Samoli E, Gryparis A, Le Tertre A, B, et al. 2003. Short-term effects of ambient particles onthis practice has not been evaluated. mortality in the elderly: results from 28 cities in the Monopolis Y, et al. 2001. Confounding and effect modifica- The principal limitations in interpreting the APHEA2 project. Eur Respir J 40(suppl):28s–33s. tion in the short-term effects of ambient particles on total Analitis A, Katsouyanni K, Dimakopoulou K, Samoli E, mortality: results from 29 European cities within theAPHENA findings lie with the data available to APHEA2 project. Epidemiology 12:521–531. Nikoloulopoulos AK, Petasakis Y, et al. 2006. Short-termthe investigators. The data came from multiple effects of ambient particles on cardiovascular and respi- Katsouyanni K, Touloumi G, Spix C, Schwartz J, Balducci F,countries and were not collected according to a ratory mortality. Epidemiology 17:230–233. Medina S, et al. 1997. Short-term effects of ambient sul-uniform protocol. Although we edited and ana- Anderson HR, Atkinson RW, Peacock JL, Marston L, phur dioxide and particulate matter on mortality in Konstantinou K. 2004. Meta-analysis of Time Series Studies 12 European cities: results from time series data from thelyzed all data extensively and used a quality APHEA project. BMJ 314:1658–1663.Environmental Health Perspectives • VOLUME 116 | NUMBER 11 | November 2008 1485
  7. 7. Samoli et al.Krewski D, Burnett RT, Goldberg MS, Hoover K, Siemiatycki J, concurvity in generalized additive models linking mortality and health time-series studies: effect on air pollution Jerrett M, et al. 2000. Reanalysis of the Harvard Six Cities to ambient particulate matter. Epidemiology 14:18–23. effect estimates. Stat Med 25:4164–4178. Study and the American Cancer Society Study of Samet JM, Dominici F, Curriero FC, Coursac I, Zeger SL. 2000a. Touloumi G, Samoli E, Quenel P, Paldy A, Anderson RH, Zmirou Particulate Air Pollution and Mortality, part II: sensitivity Fine particulate air pollution and mortality in 20 U.S. cities, D, et al. 2005. Short-term effects of air pollution on total analysis. In: Reanalysis of the Harvard Six Cities Study and 1987–1994. N Engl J Med 343:1742–1749. and cardiovascular mortality: the confounding effect of the American Cancer Society Study of Particulate Air Samet JM, Dominici F, Zeger SL, Schwartz J, Dockery DW. influenza epidemics. Epidemiology 16:49–57. Pollution and Mortality: A Special Report of the Institute’s 2000b. The National Morbidity, Mortality, and Air Pollution U.S. EPA. 2004. The Particle Pollution Report: Current Particle Epidemiology Reanalysis Project. Cambridge, Study. Part I: Methods and methodologic issues. Res Rep Understanding of Air Quality and Emissions through 2003. MA:Health Effects Institute, 129–240. Health Eff Inst 94(pt 1):5–14. Research Triangle Park, NC:U.S. Environmental ProtectionLe Tertre A, Medina S, Samoli E, Forsberg B, Michelozzi P, Samet JM, Zeger SL, Dominici F, Curriero F, Coursac I, Dockery Agency, Office of Air Quality Planning and Standards. Boumghar A, et al. 2002. Short-term effects of particulate DW, et al. 2000c. The National Morbidity, Mortality, and Air Wong CM, on behalf of PAPA teams. 2008. Public Health air pollution on cardiovascular diseases in eight European Pollution Study. Part II: Morbidity and mortality from air pollu- and Air Pollution in Asia (PAPA): a multi-city study for cities. J Epidemiol Commun Health 56:773–779. tion in the United States. Res Rep Health Eff Inst 94(pt 2):5–79. short-term effects of air pollution on mortalityMcCullagh P, Nelder JA. 1989. Generalized Linear Models. Samoli E, Analitis A, Touloumi G, Schwartz J, Anderson HR, [Abstract]. Available: New York:Chapman and Hall. Sunyer J, et al. 2005. Estimating the exposure-response AnnualConferenceProgram2008.pdf [accessed 12 May 2008].National Research Council. 2004. Research Priorities for relationships between particulate matter and mortality Wood SN. 2000. Modelling and smoothing parameter estima- Airborne Particulate Matter IV: Continuing Research within the APHEA multicity project. Environ Health tion with multiple quadratic penalties. J R Stat Soc [Ser B] Progress. Washington, DC:National Academy Press. Perspect 113:88–95. 62:413–428.O’Neill MS, Jerrett M, Kawachi I, Levy JI, Cohen AJ, Gouveia N, Stieb DM, Judek S, Burnett RT. 2002. Meta-analysis of time- World Health Organization. 1975. International Classification of et al. 2003. Health, wealth, and air pollution: advancing the- series studies of air pollution and mortality: effects of Diseases, 9th Revision. Geneva:World Health Organization. ory and methods. Environ Health Perspect 111:1861–1870. gases and particles and the influence of cause of death, World Health Organization. 2004. Air Quality Guidelines forPeng RD, Dominici F, Louis T. 2006. Model choice in multi-site age, and season. J Air Waste Manag Assoc 52:470–484. Europe. 2nd ed. WHO Regional Publications, European time series studies of air pollution and mortality. J R Stat Stieb DM, Judek S, Burnett RT. 2003. Meta-analysis of time- Series, No. 91. Copenhagen:WHO Regional Office for Soc [Ser A] 169:179–203. series studies of air pollution and mortality: update in rela- Europe Available: CA III, Dockery DW. 2006. Health effects of fine particu- tion to the use of generalized additive models. J Air Waste 20050223_4 [accessed 12 May 2008]. late air pollution: lines that connect. J Air Waste Manag Manag Assoc 53:258–261. World Health Organization. 2006. Air Quality Guidelines: Global Assoc 56:709–742. Touloumi G, Samoli E, Pipikou M, Le Tertre A, Atkinson R, Update 2005. Available: TO, Burnett RT, Krewski D. 2003. The effect of Katsouyanni K. 2006. Seasonal confounding in air pollution E90038.pdf [accessed 12 May 2008].1486 VOLUME 116 | NUMBER 11 | November 2008 • Environmental Health Perspectives