An evidence-based appraisal of global association between air pollution and risk of stroke

An evidence-based appraisal of global association between air pollution and risk of strokeWan-Shui Yang ⁎,1,Xin Wang 1,Qin Deng 1,Wen-Yan Fan 1,Wei-Ye Wang 1Department of Social Science and Public Health,School of Basic Medical Science,Jiujiang University,Jiujiang,China Jiangxi Province Key Laboratory of Systems Biomedicine,Jiujiang University,Jiujiang,Chinaa b s t r a c ta r t i c l e i n f o Article history:Received 7February 2014Received in revised form 11May 2014Accepted 12May 2014Available online 17May 2014Keywords:Air pollution StrokeMeta-analysisCase-crossover study Time series studyBackground:The aim of this study was to evaluate the transient effects of air pollutants on stroke morbidity and mortality using the meta-analytic approach.Methods:Three databases were searched for case-crossover and time series studies assessing associations be-tween daily increases in particles with diameter b 2.5μm (PM 2.5)and diameter b 10μm (PM 10),sulfur dioxide (SO 2),carbon monoxide (CO),nitrogen dioxide (NO 2),ozone,and risks of stroke hospitalizations and mortality.Risk estimates were combined using random-effects model.Results:A total of 34studies were included in the meta-analysis.Stroke hospitalizations or mortality increased 1.20%(95%CI:0.22–2.18)per 10μg/m 3increase in PM 2.5,0.58%(95%CI:0.31–0.86)per 10μg/m 3increase in PM 10,1.53%(95%CI:0.66–2.41)per 10parts per billion (ppb)increase in SO 2,2.96%(95%CI:0.70–5.27)per 1ppm increase in CO,and 2.24%(95%CI:1.16–3.33)per 10ppb increase in NO 2.These positive associations were the strongest on the same day of exposure,and appeared to be more apparent for ischemic stroke (for all 4gaseous pollutants)and among Asian countries (for all 6pollutants).In addition,an elevated risk (2.45%per 10ppb;95%CI:0.35–4.60)of ischemic stroke associated with ozone was found,but not for hemorrhagic stroke.Conclusion:Our study indicates that air pollution may transiently increase the risk of stroke hospitalizations and stroke mortality.Although with a weak association,these findings if validated may be of both clinical and public health importance given the great global burden of stroke and air pollution.©2014Elsevier Ireland Ltd.All rights reserved.1.IntroductionAccording to a 2004report released by World Health Organization,stroke is a leading cause of death and disability globally,which accounts for approximately 5.5million deaths every year representing nearly 10%of all deaths;while 44million disability-adjusted life-years are lost annually due to stroke.Therefore,the primary prevention efforts to-ward stroke should be explored given the great stroke burden in terms of mortality and disability worldwide [1].Currently,there is increasing evidence of an association between acute exposure to air pollution and elevated risk of cardiovascular disease morbidity as well as mortality [2,3].In particular,American Heart Association (AHA)concluded that the evidence of fine particles (diameter b 2.5μm,PM 2.5)exposure as a causal risk factor for cardiovas-cular morbidity and mortality is suf ficient in their 2010scienti ficstatement [4],in which the conclusion was drawn on the basis of a com-prehensive review of current evidence.However,the AHA statement is speci fically designed as hazard identi fication,and it does not quantify the magnitude of the stroke risk associated with particulate pollutants.In addition,results from observa-tional studies assessing transient effects of other pollutants have been inconclusive [2],varying from a positive to a null association,which mainly hampered by the lack of power of any individual study [5–38].Moreover,most previous air pollution studies only focused on stroke in general,few studies have distinguished between ischemic and hem-orrhagic strokes and have yielded inconsistent results [7,12,17,23,26,28,31,34],which also could be,at least in part,explained by the limited power for single study;however,this issue is important because there may be major differences in the underlying mechanisms that may trig-ger ischemic stroke compared to hemorrhagic stroke [12,15,39].On the other hand,given the limited number of studies that evaluat-ed the shape of concentration –response function between air pollution and stroke among different pollution settings [8,10,31,32],the open questions about the differences in ambient air pollution –stroke associa-tion between low and high pollution settings still remained.On a global scale,quantifying the evidence by different regions characterized by var-ious pollution levels,for example,most Asian countries such as China and Korea (characterized as high pollution settings)vs.EuropeanInternational Journal of Cardiology 175(2014)307–313⁎Corresponding author at:Department of Social Science and Public Health,School of Basic Medical Science,Jiujiang University,Jiujiang 332000,China.Tel./fax:+867928577050.E-mail address:wanshuiyang@ (W.-S.Yang).1This author takes responsibility for all aspects of the reliability and freedom from bias of the data presented and their discussedinterpretation./10.1016/j.ijcard.2014.05.0440167-5273/©2014Elsevier Ireland Ltd.All rightsreserved.Contents lists available at ScienceDirectInternational Journal of Cardiologyj o u r n a l h o m e p a g e :w w w.e l s e vi e r.c o m/l o c a t e /i j c a r dcountries and USA(characterized as low pollution settings),using the meta-analytic technique,would help us to better understand this issue.We therefore conduct a systematic review and meta-analysis of case-crossover and time series studies to quantitatively assess the tran-sient acute effects of air pollutants including PM2.5,inhalable particles (diameter b10μm,PM10),sulfur dioxide(SO2),carbon monoxide (CO),nitrogen dioxide(NO2),and ozone(O3)on stroke hospitalization and stroke mortality.Here,the“transient acute effect”means the imme-diate change in the risk of acute-onset stroke hospitalization or mortal-ity due to short-term exposure to air pollution[40].We also conducted a secondary analysis by stroke subtypes(ischemic vs.hemorrhagic strokes),geographic locations(Asian countries vs.European countries and USA),and other characteristics of interests.2.Methods2.1.Search strategyWe searched Medline(PubMed),Embase,and Web of Science from their inception to October2013and systematically identified case-crossover and time series studies that evaluated the transient effect of air pollution on the risk of stroke hospitalization and mor-tality.No language restriction was applied.The search strategy included terms for out-come(stroke,cerebrovascular diseases,ischemic stroke,and hemorrhagic stroke), exposure(air pollution,particulate matter,sulfur dioxide,carbon monoxide,nitrogen dioxide,and ozone),and study design(case-crossover studies and time series studies). The reference lists of the retrieved original peer-review articles as well as pertinent review articles were also scanned to identify any additional relevant studies.2.2.Study selection and data extractionA published article was included if it1)had a case-crossover or time series design,2) evaluated the transient acute association between gaseous(carbon monoxide,sulfur dioxide,nitrogen dioxide,ozone)or particulate(PM2.5or PM10)air pollutants and stroke hospitalization or mortality,and3)presented odds ratio(OR),relative risk(RR)with its 95%confidence interval(CI)or standard error.If an article was duplicated,or derived from the same population as previously published and presented risk estimates for the same pollutants,the most recent publication was included.Using a unified data form,two investigators(W.S.Y.and W.Y.W.)independently eval-uated study eligibility and conducted data extraction;discrepancies were settled by con-sensus or by involving a third reviewer(W.Y.F.)for adjudication.Relevant variables included in the data form were as follows:study name(together with thefirst author's name and year of publication),study region,study periods,study design,number of cases,outcome measurement,and adjustments.If any of the above-mentioned data was not available in the articles,thefirst or corresponding authors were contacted by email for additional information.2.3.Quality assessmentThere is,to our knowledge,no validated scale to evaluate methodological quality for studies with case-crossover or time series design,we thus adapted a quality scale from val-idated scales for other types of observational studies(e.g.cohort and case-control designs) and particularly selected several items from the Newcastle–Ottawa Scale(NOS)[41]and the Cochrane risk of bias tool[42],and this method was also suggested by Mustafic et al.[43].We created a6-point scoring system,in which a study was judged on4broad per-spectives as follows:1)the quality of air pollutant assessments,2)the validation of stroke data,3)the extent of adjustment for potential confounders,and4)the generalizability of thefindings.For the quality of air pollutant assessments[43],studies received1point if measure-ments were performed at least daily with b25%missing data;whereas those with≥25% missing data and/or with measurement frequency b1time per day received no point.For the validation of stroke data,studies were assigned one point if the outcome of interest was coded based on the International Classification of Diseases or according to medical records,while no point was given for the absence of the above2criteria.For the extent of adjustment for potential confounders[43],studies received no point if no adjustment has been made for long term trends,seasonality,or temperature;studies can be given one point if the above3adjustments had been done;those that also adjusted for humidity and/or day of week received an additional1point;those that adjusted for holidays and/or influenza epidemics together with the adjustments corresponding with a score of2,can be assigned a full score of3.For the generalizability of thefindings[41,42],we considered the results to be appli-cable and assigned one point if the stroke cases in the study should be all eligible stroke cases over a defined period of time,and in a defined catchment area or in a defined hospi-tal or clinic,group of hospitals,health maintenance organization,or an appropriate sample of those cases(e.g.randomly selected).No point was given if not satisfying the above re-quirements in part,or not stated.Studies were judged to be of good quality if they obtained the full score for all the4 components;studies were considered to be of low quality if any component from the above4components received zero point;all other studies were deemed to be of interme-diate quality[43].2.4.Data synthesisThe RRs were used as the common measure of association across studies,and the ORs from case-crossover studies were considered equivalent to RRs in time series studies[44, 45].Time series analysis is the most commonly used technique to assess what fraction of the daily variations in counts of hospital admissions/deaths due to the daily variations in air pollution of the preceding days through relative risk regression analysis accounting for variables that varied in time such as meteorological parameters,but are less effective con-trol for secular trends such as seasonal effects[46].Because the unit of observation in time series studies is the day but not the individual,usual risk factors for stroke(e.g.smoking, diabetes,or hypertension)do not vary in the short-term time window analyzed with air pollution daily variations,can thus be excluded as confounders.The case-crossover design is considered to be an alternative to time series analysis,in which cases serve as their own controls,and risk estimates are based on comparisons of exposure in a case period when the event occurred with exposure in specified control pe-riods through matched case–control methods[40].The case-crossover design can thus control for individual characteristics such as age,sex,socioeconomic status,smoking, and comorbidityfixed.Also,through choosing the control period within a few weeks of outcome,this approach decreased any potential confounding role of the long-term time trends,seasonality,and day of week.Overall,both time series and case-crossover designs can provide reasonable estimates of transient effect(i.e.an immediate change in risk)of short-term exposure to elevated concentrations of ambient air pollutants on an acute-onset disease outcome[45],although risk estimates that obtained from case-crossover approach is less precise(with wider confidence intervals)than those in time series design[45].Because most of the included studies used generalized linear models and assumed a linear relationship between air pollution and outcome,and the current available studies with exposure-response analysis also supported a linear shape for stroke[8,10,31,32], we thereforefirstly created a standardized increment in pollutant concentration as fol-lows:10μg/m3for PM2.5and PM10,10parts per billion(ppb)for NO2,SO2,and O3,and 1part per million(ppm)for CO.The reason for choosing the above values as the standard-ized increments to present risk estimates is that these levels were most frequently used in previous air pollution studies.Secondly,we recalculated the risk estimates for the standardized increment for each pollutant for every included study using the following formula:RR standardized¼eLn RR originðÞIncrement originÂIncrement standardizedwhere RR is the relative risk,and Ln is the log to base e.In the third stage,we combined the recalculated study-specific RR using random-effects model[47].Heterogeneity within the studies was evaluated using Cochran's Q and I2statistics,and the null hypothesis that the studies are homogeneous was rejected if the p value for heterogeneity was less than0.10 [48]or the I2value was N50%.Publication bias was evaluated using Begg's rank correlation method[49].We also performed a meta-regression analysis to investigate the sources of heterogeneity according to study level characteristics,including sex,study population, study design,stroke subtype,and adjustment for confounding factors.Subgroup analysis was conducted by study design(time series vs.case-crossover studies),geographical location(Asia vs.Europe and North America),and stroke subtype (ischemic vs.hemorrhagic stroke).Population-attributable risks(PARs)per pollutant were also estimated using our overall risk estimates and the following equation:PAR%=100×P e(RR−1)/(P e[RR−1]+1),for which P e is the prevalence of the exposure(air pollution) in the population and is assumed to be100%.We combined adjusted risk estimates that controlled for meteorological,temporal, and seasonal parameters for every included study.Most of the included studies provided multiple estimates for single lags(g0,lag1,and lag2).In this case,the shortest lag was used in our overall analysis.We also combined the risk estimates according to differ-ent lags including lag0,lag1,and lag2for each pollutant.Some studies separated risk es-timates according to season(cold vs.warm season)[20,21]or temperature[28],study region[14],and stroke subtype[17,20,22,23,26],and did not report overall risk estimates. In this case,the stratified estimates were included in our analysis.Several studies[8,9,16, 27–30,35]only provided cumulative lags such as lag0-1,lag0-2,and lag0-3.In this case, we only included these in the overall analysis,but not for the single lag analysis.All data analyses were carried out using R2.15.3(meta3.1-2)(R Development Core Team,R Foundation for Statistical Computing,Vienna,Austria).3.Results3.1.Literature search and study characteristicsAflow diagram of the literature search strategy employed in the present study is shown in Fig.1.A total of34studies that consisted of 20time series[8,9,11,15,18,19,21,24,25,27,29–38]and14case-crossover[5–7,10,12–14,16,17,20,22,23,26,28]studies were included in thefinal analysis.308W.-S.Yang et al./International Journal of Cardiology175(2014)307–313The characteristics of 34included articles are given in Table 1.All the studies were published between 1996and 2013and were conducted in Asia [China [6–8,11,25,27,28,35],Japan [13,15,20],and Korea [31,32]],Europe [9,14,17,21,22,29,30,33,37,38],and North America [USA [5,10,19,24,26,34,36]and Canada [12,16,18,23]].Of all selected studies,24studies used stroke hospital admissions (HA)as an outcome [5,6,9,10,12–14,16–19,22–26,28–30,33–35,37,38],while 9studies reported stroke mortality [7,8,11,15,20,21,27,31,32],and one for both stroke mortality and stroke hospitalizations [36].Although the study popula-tions overlapped in the two articles by Hong et al.[31,32],these two studies were included in the meta-analysis because they reported risk estimates by different air pollutants.The median daily concentrations of pollutants reported in some included studies are given in Table 2,and we did not present all individuals because the relevant data were not available among few studies.Most of the included studies (27of all 34studies)were judged to be of good or intermediate quality accord-ing to the 6-point scoring system (Table 1).3.2.Overall analysis and subgroup analysisThere was a positive association between stroke hospitalization or stroke mortality,and all gaseous and particulate air pollutants except ozone (Tables 3&4).Speci fically,stroke hospitalization or mortality in-creased 1.20%(95%CI:0.22–2.18)per 10μg/m 3increase in PM 2.5,0.58%(95%CI:0.31–0.86)per 10μg/m 3increase in PM 10,1.53%(95%CI:0.66–2.41)per 10parts per billion (ppb)increase in SO 2,2.96%(95%CI:0.70–5.27)per 1ppm increase in CO,and 2.24%(95%CI:1.16–3.33)per 10ppb increase in NO 2.The strongest associations were found on the same day of exposure (lag 0),with this effect diminishing at longer lag days.Signi ficant between-study heterogeneity was detected for all pollutants except PM 2.5.We found that the air pollution –stroke associa-tions signi ficantly differed by study population and stroke subtype (for all pollutants except PM 2.5);whereas the meta-regression analyses didnot provide any evidence of a substantial effect of differences by sex,ad-justment for confounding factors,or by study design (data not shown).Begg's rank correlation test provided no evidence of substantial publica-tion bias in our meta-analysis.In the subgroup analysis,the positive associations were more appar-ent and remained signi ficant for ischemic stroke for all 4gaseous pollut-ants (Table 4).Although there is a null association of ozone with the risk of overall stroke,the summary RR of ischemic stroke for an increase in 10ppb of ozone was 1.0245(95%CI:1.0035–1.0460).When strati fied by geographical locations,the positive associations were more evident among Asian countries compared to those among Europe and North America for all 6pollutants.The positive relations between air pollution and stroke remained but with narrower con fidence intervals in the time series study than those in the case-crossover study.Assuming linear concentration –response relationships without a threshold between air pollution and risk for stroke,we estimated that a total of 8.34%,15.35%,and 1.54%of acute stroke hospital admissions or stroke mortal-ity could be attributable to major air pollutant (PM 2.5,PM 10,SO 2,CO,and NO 2)exposure in the World,Asia,and Europe and North America,respectively (Tables 3&4).4.DiscussionOverall,relying on current evidence,we found an increased risk of stroke hospitalizations and mortality with a transient increase in major ambient air pollutants (PM 2.5,PM 10,SO 2,CO,and NO 2).We also found a positive association between exposure to ozone and ischemic stroke.Interestingly,these positive associations were more apparent for ischemic stroke (for all 4gaseous pollutants)compared to hemor-rhagic stroke,and were more evident among Asian countries than those in Europe and North America (for all pollutants).Assuming linear concentration –response relationships without a threshold between air pollution and risk for stroke,daily exposure to ambient airpollutionFig.1.Flow chart of references selection in the meta-analysis.309W.-S.Yang et al./International Journal of Cardiology 175(2014)307–313may contribute to approximately8%of acute stroke hospitalizations or stroke mortality worldwide.Given the great global burden of stroke and air pollution,ourfindings if validated may be of both clinical and public health importance given the great global burden of stroke and air pollution.The mechanism by which air pollution leads to stroke is poorly un-derstood[4].Hypothesized mechanisms for air pollution-related stroke include systemic inflammation,thrombosis,and vascular endothelial dysfunction[50,51].Air pollutants could induce an acute systemic inflammatory response with an increased number of circulating fibrinogen,C-reactive protein and white blood cell[50,51],which could be a trigger for inflammation and increase blood coagulation [52].In addition,air pollution,may harmfully influence measures of car-diovascular physiology including blood pressure and heart rate,which could be via sympathoexcitation and/or impairing vasomotor function [50,51,53].Considering the potential etiological heterogeneity in stroke[39],we differentiated the subtypes(ischemic and hemorrhagic)and found that the adverse effects of air pollution were more evident for ischemic stroke for all4gaseous pollutants.These differences,if true,may reflect the heterogeneous etiology between ischemic and hemorrhagic stroke and need immediate hospital attention;the underlying mechanisms warrant further elucidation in future studies.However,because of a smaller number of cases of hemorrhagic stroke and a consequence of a wider confidence interval for risk estimate as compared with those for ischemic stroke in our analysis,the discrepancies would have been accidental and the play of chance or the lack of power could not be ruled out,which need replication.The results according to different study design showed that time series study had similar risk estimates but with narrower confidence intervals compared with the case-crossover study,which was in line with a previous methodological study[45].We noted that the increased risk of stroke in Asian countries were al-most2–9times as high as those in Europe and North America,where ambient air concentrations are generally higher(see Table2).The rea-sons for such regional differences in the association remain unclear. We hypothesize that the differences may reflect the potential effect modifications by disproportionate risk factors for stroke such as com-paratively lower income and socioeconomic status,less education[54] and access to care that are borne by low to middle income countries,be-cause most of the included studies conducted in Asia(n=13)are from China(n=7)and Korea(n=3).Besides,these regional differences were more apparent when we excluded data from Japan(data not shown).Given the increasing trend of stroke incidence and mortality [1,55]and the higher air pollutant levels in those regions(Table2), key targets for control stroke among less developed countries such as China are urgently introduced.These observed geographic differences may also reflect that there could be a difference in the shape of concentration–response function for stroke between low and high pollution settings.To our knowledge, only2studies that were conducted in countries with high air pollution concentrations including China[8]and Korea[31],have evaluated the exposure–response relationships between air pollution and risk for stroke,and have demonstrated a liner shape without a threshold. Given the limited evidence,more epidemiologic studies at both low and high pollution settings are needed to confirm or refute ourfindings.Strengths of present study include larger sample size with increased statistical power compared to each individual study,and a highTable1Contextual details of studies included in the meta-analysis.Study Region Design Period Population Outcome Potentialconfoundersincluded a Exposuremeasurements(0–1point)Outcomemeasurements(0–1point)Adjustments(0–3points)Generalizability(0–1point)StudyqualityXu et al.(2013)USA CC1994–2000≥65y HA A,B,C,D1121Intermediate Xiang et al.(2013)China CC2006–2008All HA A,B,C,D1121Intermediate Qian et al.(2013)China CC2003–2008≥65y M A,B,C,D1121Intermediate Chen et al.(2013)China TS1996–2008All M A,B,C,D1121Intermediate Carlsen et al.(2013)Iceland TS2003–2009All HA A,B,C,D,E,F1131Good Wellenius et al.(2012)USA CC1999–2008All HA A,B,C,O1111Intermediate Wang et al.(2012)China TS2001–2009All M C,D,E,O1101Low Villeneuve et al.(2012)Canada CC2003–2009≥20y HA A,B,C,D1121Intermediate Turin et al.(2012)Japan CC1988–2004All HA A,B,C,D,O0121Low Bedada et al.(2012)UK CC2003–200727–93y HA A,B,C,D1120LowYorifuji et al.(2011)Japan TS2003–2008All M A,B,C,D,E,F,G,O1131GoodO'Donnell et al.(2011)Canada CC2003–2008All HA C1101Low Andersen et al.(2010)Denmark CC2003–2006All HA A,B,C,D,O1121Intermediate Szyszkowicz et al.(2008)Canada TS1992–2002All HA A,B,C,D1121Intermediate Lisabeth et al.(2008)USA TS2001–2005All HA A,B,C1111Intermediate Kettunen et al.(2007)Finland TS1998–2004≥65y M A,B,C,D,E,F1131Good Henrotin et al.(2007)France CC1994–2004All HA A,B,C,D,F,G1131Good Yamazaki et al.(2007)Japan CC1990–1994≥65y M A,B,C,D1121Intermediate Villeneuve et al.(2006)Canada CC1992–2002≥65y HA A,B,C,D1121Intermediate Low et al.(2006)USA TS1995–2003All HA C,D,F,G0101LowChan(2006)Taiwan TS1997–2002≥50y HA A,B,C,E1121Intermediate Wellenius et al.(2005)USA CC NA b≥65y HA A,B,C,O1111Intermediate Tsai et al.(2003)Taiwan CC1997–2000All HA A,B,C,D1121Intermediate Kan et al.(2004)China TS2001–2002All M A,B,C,E,O1121Intermediate Sunyer et al.(2003)Europe TS1990–1996All HA A,B,C,D,F,G1131GoodHong et al.(2002)Korea TS1991–1997All M A,B,C,D,E,O1121Intermediate Hong et al.(2002)Korea TS1995–1998All M A,B,C,D,E,O1121Intermediate Le Tertre et al.(2002)Europe TS1989–1996All HA A,B,C,D,E,F,G1131Good Ballester et al.(2001)Spain TS1994–1996All HA NA b1101LowLinn et al.(2000)USA TS1992–1995All HA A,B,C1111Intermediate Wong et al.(1999)China TS1994–1995All HA A,B,C,D,E,F1131Good Wordley et al.(1997)USA TS1992–1994All HA and M A,B,C,D1121Intermediate Poloniecki et al.(1997)England TS1987–1994All HA A,B,C,D,F,G1131Good Ponka et al.(1996)Finland TS1987–1989All HA A,C,D0101LowAbbreviations:CC=case-crossover study;HA=hospital admission;M=mortality;TS=time series study.a A=long term trends;B=seasonality;C=temperature;D=humidity;E=day of week;F=holidays;G=influenza epidemics;O=others.b The information is not available in the original article,and we attempted to contact the authors but without any response.310W.-S.Yang et al./International Journal of Cardiology175(2014)307–313proportion of studies with moderate-to-high quality included in themeta-analysis.There are,however,several possible sources of bias in our systematic review and meta-analysis.First,the use of regional mon-itoring sites to determine the personal air pollution levels may lead to misclassification bias of exposure.Second,due to the high correlation among pollution components,it is difficult to separate the independent effect for each pollutant,which may confound the observed associa-tions.In outdoor air,for example,NO2is often highly correlated with other combustion products notably PM2.5.Thus,in most cases,NO2 may serve as a surrogate for all traffic-related combustion products [56].However,the results for ozone could be less influenced given the weak correlation between ozone and other pollutant concentrations in ambient air[56].Although particulate pollutants are considered to be responsible for a large number of adverse cardiovascular outcomes[4] and have received by far the most attention,we also noted a greater strength of associations for all gaseous pollutants.Third,without consideration the relevance of cumulative effects may limit the ability for causal inference.Because of a few studies focus-ing on long-term exposure to air pollution in relation to stroke,we were unable to evaluate the long-term effects for each pollutant,which may underestimate the current associations.Fourth,ourfindings should be interpreted with cautions given an important limitation that substantial heterogeneity was detected within all selected studies except those for PM2.5.The meta-regression analysis indicated that the differences in study populations and stroke subtypes could partly explain such high heterogeneity.Fifth,although publication bias has not been detected in our analysis, publication bias cannot be ruled out given a low power among current standard detection methods for publication bias[57].Additionally,as outcomes in most of the included studies were from the vital statistics department,the time of stroke symptom onset was not available for most stroke cases,and thereby the misclassifications of time of event onset due to the assignment of the exposure to air pollution based on hospital admission date instead of time of symptom onset may bias the results toward the null[58].Finally,given the aforementioned uncertainty for the shape of the concentration–response functions and the observed geographic differ-ences,the estimated PARs,which were calculated on the basis of the as-sumption that the air pollutant levels is linearly associated with the risk of stroke,may thereby be problematic and should be interpreted with cautions if the shape of the concentration–response function is non-linear,although most included studies assumed linear relationships, and few available studies with exposure–response analysis also demon-strated a linear shape for stroke without a threshold[8,31].The variabil-ity of pollutant concentrations,however,does not influence the calculations because PARs were calculated for specific increments if the linear assumption is true.In conclusion,air pollution has a close temporal association with stroke hospitalizations and mortality.Future air pollution studies are warranted to clarify the life-time course of cumulative effects,the vul-nerable populations,the shape of concentration–response function, and the responsible pollution constituents as well as their potential syn-ergisms for stroke.Although the causality and physiological relevance remain for further elucidation,air pollution has now become a global public health issue with major cardiovascular consequences,especially in the Asian countries such as China,which calls for urgent cooperative actions of this disease at many levels from local to national to global. Funding sourcesThis work was supported by Jiangxi Provincial Health Department of China(grant number20083168).The funders had no role in the study design,data collection and analysis,decision to publish,or preparation of the manuscript.Table2Median daily concentrations of particulate and gaseous air pollutants by geographical locations.aAmbient air pollutants Median Median range Q1Q3PM2.5(μg/m3)Asian countries26.621.6–36.5NA NAEurope8.0– 5.511.7 North America7.3 6.9–8.7 4.610.6 PM10(μg/m3)Asian countries71.645.0–110.049.7106.9 Europe20.623.6–32.114.126.2 North America23.128.0–19.416.334.1 SO2(ppb)Asian countries11.4 3.9–18.38.418.6 Europe 5.20.8–1.00.9 2.3North America 4.2 1.5–9.0 2.67.6CO(ppm)Asian countries 1.00.8–1.20.8 1.3Europe0.50.2–0.90.30.5North America0.70.3–1.00.5 1.1NO2(ppb)Asian countries28.116.0–34.020.635.2 Europe b21.511.7–35.08.917.0 North America19.312.4–23.513.825.8 Ozone(ppb)Asian countries22.020.1–23.813.332.0 Europe24.811.2–64.823.339.2 North America21.8 3.0–35.716.930.7Abbreviations:CO=carbon monoxide;NO2=nitrogen dioxide;PM2.5=particles with size b2.5μm;PM10=particles with size b10μm;Q1=first quartile value;Q3=third quartile value;SO2=sulfur dioxide.a Median pollutant concentration together with Q1and Q3derived from the average daily pollutant concentrations reported per study.Range of the median pollutant concentrations across the studies from minimum to maximum.b As some studies only reported the median concentration but not the values of thefirst and third quartiles,the interquartile range may not include the median value.Table3Summary risk estimates,heterogeneity,population-attributable risk,and assessment for publication bias stratified by particulate air pollutants.aPM2.5PM10Number of studies821Percent increase in risk(95%CI)per10μg/m3increase for eachparticulate air pollutant bOverall analysis 1.20(0.22–2.18)⁎0.58(0.31–0.86)⁎Subgroup analysisLag0to2Lag0 1.27(0.28–2.27)⁎0.62(0.54–0.70)⁎Lag10.13(−0.82–1.08)0.45(0.24–0.66)⁎Lag2−0.17(−1.68–1.37)0.27(0.01–0.54)⁎Stroke typeIschemic stroke 1.04(−0.25–2.34)0.72(−0.06–1.50) Hemorrhagic stroke 1.22(−0.55–3.02)0.68(−0.91–2.29) OutcomeHospital admissions0.50(−0.19–2.93)0.71(0.10–1.33)⁎Mortality 1.34(0.27–2.42)⁎0.65(0.54–0.77)⁎Study regionEurope and North America 1.62(−0.73–4.03)0.20(−0.17–0.57) Asia 1.11(0.04–2.19)⁎0.66(0.37–0.96)⁎Study designCase–crossover study 1.26(0.23–2.30)⁎0.66(0.32–1.01)⁎Time series study 1.23(0.20–2.27)⁎0.46(0.27–0.66)⁎I2(p for heterogeneity)38.2(0.09)76.2(b0.01) Publication bias(p value)0.590.67PAR%(95%CI)cWorld 1.19(0.22–2.14)⁎0.58(0.31–0.70)⁎Europe and North America 1.59(−0.74–3.87)0.20(−0.17–0.57) Asia 1.10(0.04–2.14)⁎0.66(0.37–0.95)⁎Abbreviations:PM2.5=particles with size b2.5μm;PM10=particles with size b10μm.a The*symbol indicates p b0.05.b Percent increase in risk=(1−RR)×100%.c PAR%=100×Pe(RR−1)/(P e[RR−1]+1),where P e is the prevalence of the exposure in the population and is assumed to be100%.311W.-S.Yang et al./International Journal of Cardiology175(2014)307–313。