Open Collections

UBC Faculty Research and Publications

Risk estimates of mortality attributed to low concentrations of ambient fine particulate matter in the… Pinault, Lauren; Tjepkema, Michael; Crouse, Daniel L.; Weichenthal, Scott; van Donkelaar, Aaron; Martin, Randall V.; Brauer, Michael; Chen, Hong; Burnett, Richard T. 2016

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata


52383-Pinault_L_et_al_Risk_estimates_mortality.pdf [ 2.37MB ]
JSON: 52383-1.0305740.json
JSON-LD: 52383-1.0305740-ld.json
RDF/XML (Pretty): 52383-1.0305740-rdf.xml
RDF/JSON: 52383-1.0305740-rdf.json
Turtle: 52383-1.0305740-turtle.txt
N-Triples: 52383-1.0305740-rdf-ntriples.txt
Original Record: 52383-1.0305740-source.json
Full Text

Full Text

RESEARCH Open AccessRisk estimates of mortality attributed tolow concentrations of ambient fineparticulate matter in the Canadiancommunity health survey cohortLauren Pinault1* , Michael Tjepkema1, Daniel L. Crouse2, Scott Weichenthal3, Aaron van Donkelaar4,Randall V. Martin4,5, Michael Brauer6, Hong Chen7 and Richard T. Burnett8AbstractBackground: Understanding the shape of the relationship between long-term exposure to ambient fine particulatematter (PM2.5) concentrations and health risks is critical for health impact and risk assessment. Studies evaluatingthe health risks of exposure to low concentrations of PM2.5 are limited. Further, many existing studies lackindividual-level information on potentially important behavioural confounding factors.Methods: A prospective cohort study was conducted among a subset of participants in a cohort that linkedrespondents of the Canadian Community Health Survey to mortality (n = 299,500) with satellite-derived ambientPM2.5 estimates. Participants enrolled between 2000 and 2008 were followed to date of death or December 31,2011. Cox proportional hazards models were used to estimate hazard ratios (HRs) for mortality attributed to PM2.5exposure, adjusted for individual-level and contextual covariates, including smoking behaviour and body massindex (BMI).Results: Approximately 26,300 non-accidental deaths, of which 32.5 % were due to circulatory disease and 9.1 %were due to respiratory disease, occurred during the follow-up period. Ambient PM2.5 exposures were relatively low(mean = 6.3 μg/m3), yet each 10 μg/m3 increase in exposure was associated with increased risks of non-accidental(HR = 1.26; 95 % CI: 1.19-1.34), circulatory disease (HR = 1.19; 95 % CI: 1.07–1.31), and respiratory disease mortality(HR = 1.52; 95 % CI: 1.26–1.84) in fully adjusted models. Higher hazard ratios were observed for respiratory mortalityamong respondents who never smoked (HR = 1.97; 95 % CI: 1.24–3.13 vs. HR = 1.45; 95 % CI: 1.17–1.79 for eversmokers), and among obese (BMI ≥ 30) respondents (HR = 1.76; 95 % CI: 1.15-2.69 vs. HR = 1.41; 95 % CI: 1.04–1.91for normal weight respondents), though differences between groups were not statistically significant. A thresholdanalysis for non-accidental mortality estimated a threshold concentration of 0 μg/m3 (+95 % CI = 4.5 μg/m3).Conclusions: Increased risks of non-accidental, circulatory, and respiratory mortality were observed even at verylow concentrations of ambient PM2.5. HRs were generally greater than most literature values, and adjusting forbehavioural covariates served to reduce HR estimates slightly.Keywords: PM2.5, Fine particulate matter, Air pollution, Cardiovascular mortality, Respiratory mortality* Correspondence: Lauren.Pinault@canada.ca1Health Analysis Division, Statistics Canada, 100 Tunney’s Pasture Driveway,Ottawa, ON K1A 0T6, CanadaFull list of author information is available at the end of the article© 2016 Pinault et al. Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0International License (, which permits unrestricted use, distribution, andreproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link tothe Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver( applies to the data made available in this article, unless otherwise stated.Pinault et al. Environmental Health  (2016) 15:18 DOI 10.1186/s12940-016-0111-6BackgroundAmbient fine particulate air pollution (PM2.5) is knownto contribute to cardiovascular and respiratory morbid-ity, and is recognized as an important contributor to glo-bal disease burden [1]. Recent estimates from GlobalBurden of Disease suggest that ambient air pollutionwas responsible for nearly 2.9 million deaths per year in2013 [2]. While ambient PM2.5 concentrations in Canadaare generally below national and international guidelines,analyses from the 1991 Canadian Census Health andEnvironment Cohort (CanCHEC) suggest that long-termexposure to PM2.5 in Canada (mean = 8.9 μg/m3) maycontribute to non-accidental and cardiovascular mortal-ity [3]. However, that study did not include individual-level information on potentially important confoundingfactors such as smoking and obesity and applied anindirect approach to adjust for these and other factors[3, 4]. Analysis of the United States Agricultural HealthStudy (AHS) cohort also supports an association be-tween cardiovascular mortality and long-term exposureto low concentrations of ambient PM2.5 (mean = 9.2 μg/m3)[5]. Moreover, a recent meta-analysis of studies conductedin North America and internationally supports an associ-ation between long-term exposure to PM2.5 and mortality,with the strongest association observed for cardiovascularmortality [6].The WHO PM2.5 guideline of 10 μg/m3 was based onthe lower end of the exposure distribution in previousstudies [1], though there are few studies that have evalu-ated concentration-response associations at very low ex-posures. The Global Burden of Disease 2010 study [1]developed a mortality risk model for PM2.5 over the glo-bal range of concentrations. This model incorporated acounterfactual uncertainty distribution, below which noexcess risk was assumed, and was specified by a uniformdistribution between 5.8 μg/m3 and 8.8 μg/m3. This un-certainty distribution was selected based on the lack ofempirical evidence of any statistical association betweenambient PM2.5 and mortality below their counterfactualdistribution. These concentrations represent the 48.9thand 79.9th percentiles of the exposure distribution in thisstudy, respectively. Therefore, it is of interest to examinethe shape of the concentration-mortality association atthese very low concentrations, as well as the statisticalstrength of evidence for such an association.In this study, we examine the relationship betweenlong-term exposure to ambient PM2.5 and non-accidental,respiratory, and cardiovascular mortality in the CanadianCommunity Health Survey (CCHS) cohort. Participants inthis cross-sectional survey were enrolled across Canadabetween 2000 and 2008 and provided detailed individual-level information on potentially important confoundingfactors (e.g. smoking, obesity) that were not availablefor the previous analysis of PM2.5 and mortality inthe CanCHEC study [3]. As such, the primary aim ofthis study was to examine the relationship betweenvery low concentrations of PM2.5 (mean = 6.3 μg/m3)and different causes of mortality in Canada and theimpacts of adjusting for potential confounding factors.Finally, an improved, finer-scale, satellite-derived ex-posure model for PM2.5 (i.e., a 1 km2 grid) was usedto reduce exposure misclassification.MethodsData sourcesThe CCHS is a national, cross-sectional survey providinginformation about the health, behaviours, and healthcare use of the non-institutional Canadian populationaged 12 or older. The survey excludes full-time membersof the Canadian Armed Forces and residents of Indianreserves and certain remote areas. Exclusions representless than 3 % of the target population of Canada [7]. Theannual component of the CCHS was conducted everytwo years from 2000/01 to 2007, after which the surveywas conducted on an annual basis. The CCHS responserates are as follows: 84.7 % in Cycle 1.1 (2000/01),80.7 % in Cycle 2.1 (2003), 78.9 % in Cycle 3.1 (2005),77.6 % in 2007, and 75.0 % in 2008 [7]. CCHS respon-dents were eligible for the CCHS-mortality cohort ifthey gave permission to share and link their informationwith other administrative datasets; 86.0 % of CCHS re-spondents agreed to the linkage.The Canadian Mortality Database (CMDB) is a na-tional database that contains all deaths registered inCanada since 1950. Deaths that occurred between Janu-ary 1, 2000 and December 31, 2011 were eligible forlinkage. The CMDB includes data on underlying causeof death and date of death.The Historical Tax Summary File (HTSF) is a databaseof annual tax returns that represent all individuals whoreceived a tax declaration in a given year. Tax yearsbetween 1996 and 2011 were eligible for linkage. TheHTSF includes postal codes, names, and dates of death(if applicable).Linkage methodologyThe creation of the CCHS-Mortality Cohort was con-ducted in two steps. First, using a probabilistic linkagemethodology based on the Fellegi-Sunter theory of rec-ord linkage [8], eligible CCHS respondents were linkedto the HTSF (using date of birth, sex, name, and postalcode), in order to capture these variables and date ofdeath, as reported on tax files between 1996 and 2011.Approximately 85 % of eligible CCHS respondents werelinked to the HTSF. Alternative postal codes and nameswere captured through this initial linkage and were usedin the subsequent linkage to the CMDB, to improvelinkage results. Second, all eligible CCHS respondentsPinault et al. Environmental Health  (2016) 15:18 Page 2 of 15(regardless of whether they were linked to the HTSF)were also linked to the CMDB (which included under-lying cause of death), using standard probabilistic linkagetechniques (as described above) and followed for mortal-ity from cohort entry (i.e., date of CCHS interview) toDecember 31, 2011.Data preparationA total of 457,300 eligible CCHS-mortality respondentswere included, with 117,800 respondents in Cycle 1,112,900 respondents in Cycle 2, 113,900 respondents inCycle 3, and 112,700 respondents in 2007/08. CCHS re-spondents who were first linked to the HTSF had agreater probability to be linked to the CMDB since add-itional data in the HTSF (e.g., alternate postal codes,name, and date of death), were used in the probabilisticlinkage. In order to reduce the probability of false-negative links, we excluded those CCHS respondentswho were not linked to the HTSF (n = 69,300 respon-dents excluded) (Additional file 1).Since the purpose of this analysis was to evaluatelong-term effects of air pollution exposure, the studypopulation was restricted to adults aged 25 to 90 yearsof age at enrollment (n = 72,000 respondents excluded).Adults older than 90 years of age were excluded fromthis study to ensure a sufficient sample size within allage strata. Similar to the CanCHEC study [3], immi-grants living in Canada for less than 20 years (i.e., thosewho had arrived in Canada less than 20 years before thestart date), were excluded from this study (n = 13,200additional respondents excluded) for the following rea-sons. Immigrants are known to have better health andlive longer than the Canadian-born population [9]. Im-migrants also more frequently live in areas of greaterambient air pollution (unpublished data), and theirexposure to air pollution prior to living in Canada islargely unknown. Cause-specific mortality analyses amongrecent immigrants were also not meaningful due to smallsample sizes in the CCHS cohort (i.e., < 250 deaths).Therefore, the use of a larger cohort would be necessaryto examine the health effects of air pollution on recent im-migrant populations. Finally, we excluded an additional3,400 respondents who were not linked to air pollu-tion estimates since they live beyond the boundariesof the air pollution models (i.e., in the northern Ter-ritories) (Additional file 1). The final analytical samplewas 299,500 respondents (note slight inconsistenciesdue to rounding). All research using human data wascarried out at Statistics Canada in accordance withthe Statistics Act to meet standards of privacy andconfidentiality associated with the internal use ofsurvey data. The record linkage project was approvedby the Executive Management Board at StatisticsCanada (ref. num. 003–2015).The place of residence of respondents at the date ofentry into the Cohort was mapped in Geographic In-formation Systems (ArcGIS v.10; ESRI 2010) throughthe use of Statistics Canada’s Postal Code ConversionFile plus (PCCF+) V.6B, which assigns geographiccoordinates to postal codes based on a population-weighted random allocation algorithm [10]. Respond-ent locations were then spatially linked to estimatesfrom a surface layer of PM2.5 concentration derivedby relating total column aerosol optical depth retrievalsfrom the Moderate Resolution Imaging Spectroradiometer(MODIS) instrument to near-surface PM2.5 using theGEOS-Chem chemical transport model. Geographicallyweighted regression, which includes ground monitoringdata and land use information, was subsequently appliedto these estimates to produce average PM2.5 concentra-tions at a 0.01° × 0.01° (approximately 1 km2) resolutionfrom 2004 to 2012 [11]. These models included coveragefor nearly all of mainland North America. These estimateswere extended to 1998 to 2003 using the inter-annualvariation of Boys et al. (2014) [12], who inferred globalPM2.5 trends at 0.1° × 0.1° resolution using satellites from1998 to 2012. Average PM2.5 levels were strongly cor-related with ground-level observations in NorthAmerica (R2 = 0.82, slope = 0.97; n = 1440) [11]. Outliersthat included PM2.5 values >20 ug/m3 were excluded fromanalysis (<1 % of respondents were excluded in this man-ner in any year). These outliers were likely due to inaccur-ate estimates of aerosol optical depth from satelliteretrievals. For each year in the cohort, respondents wereassigned a PM2.5 value corresponding to the mean of thethree previous years to the follow-up year; therefore, ex-posure always preceded response. For example, for thefollow-up year 2001, we assigned the mean PM2.5 esti-mates from 1998 to 2000.Covariates and statistical methodsStandard Cox proportional hazards models [13] wereused for survival analysis of non-accidental and cause-specific mortality within the cohort, from the date ofinterview for the CCHS to either the date of death re-corded in the CMDB or the final date of the linkage pro-ject (i.e., 31 December, 2011). All models were stratifiedby sex and age (5-year intervals). Socioeconomic covari-ates included: immigrant status, visible minority status,Aboriginal status, and marital status, educational attain-ment, income adequacy quintile, and employment status(Table 1). Visible minority status was defined as in theEmployment Equity Act, as “persons, other than Aboriginalpeoples, who are non-Caucasian in race or non-white incolour” [14]. Income adequacy quintiles were calculatedbased on the ratio of household income to the low-income cut-off for their household and community size.Low-income cut-offs represent families that spend morePinault et al. Environmental Health  (2016) 15:18 Page 3 of 15Table 1 Descriptive statistics of the study cohort and PM2.5 exposure, with Cox proportional HRs for each covariate95 % C.I. PM2.5Covariate Persons+ HRǂ Lower Upper Mean SDAll 299,500 − − − 6.32 2.54SexMale 137,800 − − − 6.28 2.54Female 161,700 − − − 6.36 2.54Age group†25–34 years 52,500 − − − 6.39 2.5435–44 years 59,400 − − − 6.29 2.5045–54 years 58,100 − − − 6.21 2.5155–64 years 54,900 − − − 6.20 2.5165–74 years 41,700 − − − 6.41 2.5875–90 years 32,900 − − − 6.58 2.64Immigrant statusNot an immigrant 270,300 1.000 − − 6.19 2.50Immigrant (in Canada≥ 20 years) 28,800 *0.863 0.834 0.894 7.57 2.52Visible minority statusWhite 281,000 1.000 − − 6.31 2.53Visible minority 17,700 0.938 0.877 1.004 6.49 2.67Aboriginal statusNot Aboriginal 289,600 1.000 − − 6.36 2.54Aboriginal 9,200 *1.390 1.267 1.525 5.12 2.21Marital statusMarried or common-law 183,500 1.000 − − 6.09 2.46Separated, divorced, widowed 69,500 *1.344 1.306 1.382 6.62 2.60Single, never married 46,400 *1.512 1.446 1.581 6.82 2.63Educational attainmentNot completed high school 71,700 1.000 − − 6.01 2.58High school diploma 113,500 *0.829 0.806 0.852 6.25 2.50Post-secondary diploma/certificate 64,900 *0.723 0.694 0.753 6.43 2.51University degree 47,100 *0.581 0.552 0.611 6.83 2.51Low income adequacy quintile1st quintile - lowest 56,200 1.000 − − 6.53 2.642nd quintile 54,500 *0.787 0.762 0.813 6.37 2.583rd quintile 53,000 *0.662 0.637 0.689 6.37 2.524th quintile 53,300 *0.583 0.557 0.610 6.34 2.495th quintile - highest 56,700 *0.483 0.458 0.509 6.17 2.43Employment statusEmployed 174,500 1.000 − − 6.31 2.50Not employed: looked for work‡ 7,300 *1.522 1.319 1.757 6.20 2.61Not employed: did not look for work‡ 78,100 *1.818 1.732 1.908 6.25 2.55Permanently unable to work 9,800 *4.533 4.274 4.808 6.43 2.64Body Mass IndexaUnderweight (<18.5) 3,700 *2.140 1.989 2.303 6.76 2.60Normal weight (18.5 - 25.0) 93,700 1.000 − − 6.54 2.55Pinault et al. Environmental Health  (2016) 15:18 Page 4 of 15than 20 % of their income on food, shelter and cloth-ing, and are adjusted for size of family and area ofresidence [14].Neighbourhood socioeconomic status, including bothsocial and material deprivation, contributes to increasedrisk of mortality in Canadian cities, although the pres-ence of immigrants can reduce mortality risk [15].Ecological (contextual) covariates were derived from thelong-form Canadian Census at the Census Division (CD)and Dissemination Area (DA) geographic scale, from the2001 Census for respondents interviewed between 2000and 2003, and the 2006 Census for respondents inter-viewed during or after 2004. Census Divisions are a sub-division of the provinces and territories that usuallyrepresent communities, regional districts, or severalneighbouring municipalities, and range in size from sev-eral thousand to a few million persons [14]. Dissemin-ation Areas are the smallest geographical unit used bythe Census and are delineated based on populationcounts based on the previous census, to target a popula-tion of 400–700 persons [14]. There were 288 CDs and54,623 DAs in Canada as of 2006 [14]. These contextualcovariates were then linked to individual respondentsthrough a common geographic identifier (i.e., a numericcode identifying the DA or CD). For each CD and DA,the proportion of recent immigrants (<5 years residencyin Canada), educational attainment (the proportion ofpersons aged 15 years or older who had not graduatedfrom high school) and low income (the proportion ofpersons below the low-income cut-off ) were derived forboth Census years [16]. The proportion of recent immi-grants in a region may provide a health benefit in theform of social inclusion if the resident is a member of aunified community, though it also may represent socialdeprivation, since recent immigrants also include per-sons of very low SES upon arrival in Canada (e.g., refu-gees or temporary workers). The other two ecologicalcovariates (educational attainment and low income) pro-vide a more direct estimate of neighbourhood socioeco-nomic status. Although broader geographic scales suchas Census Tracts (CTs) are more often used to deriveneighbourhood contextual variables [16], CTs were notavailable for rural areas. Neighbourhood covariates weretherefore calculated by taking the difference between CDTable 1 Descriptive statistics of the study cohort and PM2.5 exposure, with Cox proportional HRs for each covariate (Continued)Overweight (25.0 - 30.0) 114,900 *0.804 0.781 0.828 6.29 2.52Obese I (30.0 - 35.0) 54,700 *0.884 0.852 0.917 6.14 2.52Obese II (>35.0) 24,200 *1.270 1.209 1.334 6.06 2.53Fruit and vegetable consumption<5 servings per day 153,200 1.000 − − 6.38 2.56≥5 servings per day 101,100 *0.828 0.806 0.851 6.52 2.52SmokingNever smoked 84,100 1.000 − − 6.41 2.53Former smoker 139,200 *1.284 1.244 1.324 6.26 2.51Current daily or occasional smoker 75,900 *2.604 2.509 2.702 6.33 2.59AlcoholRegular drinker (≥1 drink per month) 141,700 1.000 − − 6.51 2.55Occasional or former drinker 80,800 *1.394 1.356 1.433 6.25 2.59Never drinker 11,000 *1.274 1.214 1.337 6.17 2.64Ecological covariatesb% recent immigrants (CD-DA) − *1.102 1.064 1.141 − −% recent immigrants (CD) − *0.713 0.680 0.747 − −% completed high school (CD-DA) − *0.928 0.919 0.938 − −% completed high school (CD) − *0.897 0.886 0.908 − −% in low income families (CD-DA) − *1.119 1.107 1.131 − −% in low income families (CD) − *1.100 1.070 1.131 − −+Numbers were rounded to the nearest 100 for confidentialityǂModels were stratified by age (5 year categories) and sex*Significant HR (p < 0.05)†At time of entry into the cohort‡(Did not) look for work in past 4 weeksaAfter adjusting for self-reporting bias in CCHS, as in [16]bHRs provided for 10 % increase in populationPinault et al. Environmental Health  (2016) 15:18 Page 5 of 15and DA estimates. It was expected a priori that the eco-logical covariates would attenuate risk estimates as inprevious work on CanCHEC [3].In addition to the socioeconomic and ecological covar-iates, this study included four health status/behaviouralcovariates. Body Mass Index (BMI) was derived from theself-reported height and weight of respondents, and ad-justed using correction factors that were developed forthe CCHS to account for self-reporting bias in BMI data[17]. The International Standard Classification was usedto categorize Body Mass Index [18], with obesity subdi-vided into two categories (i.e., BMI 30 – 34.9 and BMI ≥35) to further differentiate health risks among obese per-sons within the study. Smoking behaviour was catego-rized as never, former, or current smokers. Detailed dataon smoking behaviour (e.g., number of cigarettessmoked per day) were available only for daily smokers(ca. 21.3 % of respondents) and were therefore not in-cluded. Fruit and vegetable daily consumption and alco-hol consumption were also included, as in previousstudies [19] (Table 1).Survival models were examined in a sequential mannerby adding all of the socioeconomic covariates in a singlemodel, then adding in the ecological covariates to thesocioeconomic models, and finally by adding the behav-ioural covariates to create fully adjusted models for non-accidental mortality (ICD-10 codes A-R) and mortalityattributed to circulatory disease (ICD-10: I00–I99, withand without diabetes, E10–E14), including the sub-groups of ischemic heart disease (ICD-10: I20–I25), andcerebrovascular disease (ICD-10: I60–I69). We also con-sidered models for mortality due to respiratory disease(ICD-10: J00–J99), also including chronic obstructivepulmonary disease (ICD-10: J19–J46), and lung cancer(ICD-10: C33–C34). We also examined a model of so-cioeconomic and behavioural covariates, excluding eco-logical covariates. We added groups of variables in thismanner to specifically examine the influence of includ-ing the behavioural variables to a model which includedboth socioeconomic and ecological variables, as wereavailable in previous cohort studies in Canada [3]. Effectmodification by sex, smoking behaviour (ever smokedvs. never smoked), BMI (obese: BMI ≥ 30 and obese II:BMI ≥ 35 vs. normal weight: BMI = 18.5–25), fruit andvegetable consumption (<5 servings vs. ≥ 5 servings),alcohol consumption (regular drinker vs. occasional/never/former drinker), and age (<75 years vs. ≥ 75 years)were also evaluated in separate Cox proportional haz-ards models, and Cochran’s Q-statistic heterogeneitytests were used to evaluate significant differences in HRsamong groups [20]. These covariates were chosen for ef-fect modification analysis due to known physiologicaldifferences between these groups of respondents, andinterest in previous studies [6].To examine the shape of the relationship betweennon-accidental mortality hazard ratio (HR) and air pollu-tion exposure, we fitted spline-based HR curves usingthe smoothing method in the R package “smoothHR” onthe fully adjusted model [21]. The package uses a com-bination of AIC and BIC to determine the optimaldegrees of freedom to use in the model [21]. We alsoestimated the PM2.5 threshold concentration (T) byfitting Cox proportional hazards models to a series ofnewly defined PM2.5 based variables of the form:PM2.5 (T) = PM2.5 – T; if PMi > T and 0 otherwise, forT = 1 to 10. Our estimate of T is the concentrationcorresponding to the largest (−2) log-likelihood value(−2LL) obtained from the Cox model. Ninety-five per-cent confidence intervals on T were based on changesin -2LL of 3.84 units.All descriptive statistics reported from the survey wererounded to the nearest hundred for institutional confi-dentiality reasons.ResultsA total of 299,500 respondents were included in the studyafter excluding respondents who were not linked to a taxfile, respondents who were not within the 25 to 90 yearage range and were not recent immigrants (i.e., < 20 yearsin Canada), and respondents who were not linked to airpollution estimates. Respondents were followed formortality for up to 12 years after cohort entry (meanfollow-up period (± SD) was 7.6 ± 2.7 years). The meanexposure (± SD) of respondents to PM2.5 estimated fromthe 3-year moving average was 6.3 ± 2.5 μg/m3. The PM2.5person-year exposure percentiles within the final study co-hort were: minimum: 1.0 μg/m3, 5th: 3.0 μg/m3, 25th:4.2 μg/m3, median: 5.9 μg/m3, 75th: 8.3 μg/m3, 95th:11.3 μg/m3, and maximum: 13.0 μg/m3. In large cities(metropolitan pop. > 1 million), PM2.5 estimates were gen-erally greater than in surrounding areas, and there wereareas of the downtown core exceeding 8 ug/m3 in all ofthese cities (Fig. 1). Mean PM2.5 exposure increased incre-mentally by decreasing income quintile and was highestfor respondents in the poorest income quintile (Table 1).PM2.5 exposure was also greatest for the most highly edu-cated respondents (Table 1). Obese respondents were ex-posed to less air pollution than those of increasingly lowerweight classes, with the greatest exposure among respon-dents classified as underweight (Table 1). Hazard ratiosfor non-accidental mortality were calculated for all vari-ables and ecological covariates (Table 1). Among eco-logical covariates for DAs and CDs, the proportion ofrecent immigrants, high school graduates and low incomefamilies were positively correlated with average PM2.5 airpollution exposure (Table 2). The proportion of recent im-migrants was protective for mortality at the broader land-scape level (i.e., the CT), though increased HRs at thePinault et al. Environmental Health  (2016) 15:18 Page 6 of 15neighbourhood scale (i.e., DAs) (Table 1). Associations be-tween all combinations of the covariates are provided inAdditional file 2.Separate Cox proportional hazards models were runfor all covariates in the fully adjusted models. Immigrantstatus, greater educational attainment, higher income,being overweight or obese (type I), and increasing con-sumption of fruits and vegetables were all associatedwith a lower risk of non-accidental mortality (Table 1).Aboriginal status, being unmarried, being underweightor obese (type II), not employed, smoking, and not regu-larly drinking alcohol were associated with a greater riskof non-accidental mortality (Table 1).Covariates were added in a stepwise manner to a Coxproportional hazards model for non-accidental mortalityto assess their contribution to the model (Table 3). InFig. 1 Map of mean PM2.5 estimates in Canada from 1998–2010 derived from satellite. imagery at 1 km2 resolution. Cities with populationsgreater than 1 million (in the metropolitan area) are indicated. All of these large city PM2.5 exposures were >8 ug/m3. Insets: detailed PM2.5estimates in southern Ontario, Toronto, Ottawa, Montreal, Vancouver, Edmonton, and CalgaryTable 2 Descriptive statistics of ecological covariates derived from the 2001 and 2006 CensusaPercentile Correlation withmean PM2.5Variable Min 5th 25th 50th 75th 95th MaxAggregated by Dissemination Area% recent immigrants 0.0 0.0 0.0 0.0 1.7 9.0 69.0 0.303% completed high school 0.0 47.4 63.6 73.5 82.4 92.3 100.0 0.245% in low income families 0.0 1.5 5.9 10.9 18.4 35.1 100.0 0.235Aggregated by Census Division% recent immigrants 0.0 0.1 0.3 0.7 1.9 9.5 16.7 0.424% completed high school 31.2 52.3 65.8 72.7 78.6 85.1 88.6 0.462% in low income families 3.4 7.8 10.5 12.9 15.3 21.1 37.1 0.192aSource: 2001 or 2006 Census data were chosen based on the closest year to the Cohort entryPinault et al. Environmental Health  (2016) 15:18 Page 7 of 15general, the addition of socioeconomic covariates im-proved the model fit, and resulted in a significantlyincreased HR from the unadjusted model (Table 3;Cochran’s Q = 4.29; p = 0.04). The additional of behav-ioural covariates to the socioeconomic model reducedHRs somewhat, though not significantly (Table 3;Cochran’s Q = 0.23, p = 0.63). The addition of ecologicalcovariates to the socioeconomic model, particularly thepercentage of recent immigrants and high school gradu-ates, also improved model fit and significantly increasedHRs (Table 3; Cochran’s Q = 27.30, p < 0.01). Theaddition of behavioural covariates to create a fully ad-justed model also improved model fit, though the HRsdeclined non-significantly from the second adjustedmodel (Table 3; Cochran’s Q = 2.41, p = 0.12).Table 4 presents the HRs and 95 % CI for Cox propor-tional hazard models for non-accidental mortality andmortality due to circulatory or respiratory causes. In thefully adjusted model, HR estimates for non-accidentalmortality were 1.26 (95 % C.I.: 1.19–1.34) per 10 μg/m3increase in ambient PM2.5. The strongest associationwas observed for respiratory disease mortality, with anHR of 1.52 (95 % C.I.: 1.26–1.84) per 10 μg/m3 increasein ambient PM2.5. In fully adjusted models, HRs weresignificantly greater than one for all causes of death ex-cept cerebrovascular disease and lung cancer, though theHRs were significant in the models that did not includebehavioural covariates (Table 4). For all causes of death,HRs were greater in the fully adjusted model than in theunadjusted model, though were reduced after adding be-havioural covariates (Table 4).The results of effect modification by sex, age, BMI(i.e., obese vs. normal weight), fruit and vegetable con-sumption (i.e., < 5 or ≥ 5 daily servings), smoking (i.e.,ever smoked vs. never smoked) and alcohol consump-tion are presented in Table 5. In a fully adjusted model,the HR for non-accidental mortality among men was1.34 (95 % C.I.: 1.24–1.46) per 10 μg/m3 increase in am-bient PM2.5 and was significantly greater than that ofwomen (Cochran’s Q; Table 5). The HRs for circulatoryand respiratory disease mortality among men were alsogreater than among women, though the differences inHRs were not statistically significant (Cochran’s Q;Table 5). None of the other comparisons among groupsTable 3 Cox proportional HRs for non-accidental mortalitya in the cohort, with stepwise addition of covariates95 % CIHRb Lower Upper (−2) log lUnadjusted 1.028 0.981 1.077 447,246SES covariates added separatelyImmigrant status *1.069 1.019 1.120 447,165Visible minority status 1.031 0.984 1.080 447,237Aboriginal status 1.035 0.988 1.085 447,217Marital status 0.999 0.954 1.047 446,677Educational attainment *1.114 1.063 1.168 446,442Income adequacy quintiles 1.031 0.985 1.081 446,127Employment 1.032 0.985 1.081 445,050All socioeconomic covariates *1.103 1.052 1.157 443,829All SES + ecological covariates added separately% recent immigrants *1.253 1.190 1.320 440,157% completed high school *1.349 1.278 1.424 437,545% low income 1.045 0.994 1.099 433,397All SES + all ecological covariates *1.345 1.270 1.424 433,080All SES + all ecological + behavioural covariates added separatelySmoking *1.341 1.267 1.420 431,304Alcohol consumption *1.292 1.221 1.368 432,308Fruit and vegetable consumption *1.342 1.267 1.421 433,004Body Mass Index *1.345 1.270 1.424 432,338All SES + all ecological + all behavioural covariates *1.261 1.190 1.336 429,524aNumber of deaths = 26,300bModels are stratified by age (5 year categories) and sex*Significant HR (p < 0.05)SES SocioeconomicPinault et al. Environmental Health  (2016) 15:18 Page 8 of 15Table 4 Cox proportional HRs for mortality per 10 μg/m3 increase in ambient PM2.5 in the study cohort (n = 299,500)Unadjusted+ Adjusted: SES† Adjusted: SES†+ behavioural cov.§ Adjusted: SES†+ ecological cov.‡ Adjusted: SES†+ ecological cov.‡+ behavioural cov.§95 % CI 95 % CI 95 % CI 95 % CI 95 % CICause of mortality Deaths HR To From HR To From HR To From HR To From HR To FromNon-accidentala 26,300 1.028 0.981 1.077 *1.103 1.052 1.157 *1.085 1.034 1.139 *1.345 1.270 1.424 *1.261 1.190 1.336Circulatory diseaseb 8,600 0.940 0.866 1.020 1.014 0.932 1.102 0.997 0.917 1.085 *1.297 1.174 1.434 *1.187 1.073 1.313Circulatory-diabetesc 9,500 0.939 0.868 1.015 1.016 0.938 1.100 1.011 0.933 1.096 *1.313 1.194 1.444 *1.210 1.099 1.331Ischemic heart d. d 4,700 0.979 0.877 1.093 1.090 0.975 1.220 1.078 0.963 1.207 *1.408 1.232 1.610 *1.290 1.127 1.477Cerebrovascular d. e 1,500 1.064 0.879 1.288 1.082 0.890 1.316 1.063 0.872 1.295 *1.360 1.078 1.715 1.241 0.981 1.570Respiratory diseasef 2,400 1.133 0.970 1.324 *1.269 1.083 1.487 *1.214 1.034 1.425 *1.628 1.347 1.969 *1.522 1.257 1.843COPDg 1,400 1.032 0.839 1.268 1.191 0.966 1.469 1.109 0.897 1.370 *1.480 1.150 1.903 *1.398 1.085 1.801Lung cancerh 2,700 1.007 0.871 1.166 *1.170 1.008 1.357 1.088 0.937 1.263 *1.216 1.017 1.453 1.167 0.975 1.396+Unadjusted and all adjusted models were stratified by age (5 year categories) and sex†SES covariates: immigrant status, visible minority status, Aboriginal status, marital status, income adequacy quintile, educational attainment, and employment§Behavioural covariates: smoking, alcohol consumption, fruit and vegetable consumption, and BMI‡Ecological covariates: (CD-DA and CD) for % recent immigrants, % completed high school, and % low income household*Significant HR, p < 0.05aIncludes ICD-10 codes A-R. bIncludes ICD-10 codes I00-I99. cIncludes ICD-10 codes I00-I99 and E10-E14. dIncludes ICD-10 codes I20-I25. eIncludes ICD-10 codes I60-I69. fIncludes ICD-10 codes J00-J99. gIncludes ICD-10codes J19-J46. hIncludes ICD-10 codes C33-C34Pinaultetal.EnvironmentalHealth (2016) 15:18 Page9of15Table 5 Effect modification of Cox HRs† by sex, ageǂ, smoking, obesity, and fruit/vegetable and alcohol consumption95 % CI 95 % CI Cochran’s QCause of death Deaths HR Lower Upper Deaths HR Lower Upper Q pFemales (n = 161,700) Males (n = 137,800)Non-accidental 12,700 *1.181 1.088 1.282 13,000 *1.344 1.239 1.457 4.829 0.028Circulatory 4,100 1.109 0.959 1.282 4,300 *1.268 1.101 1.459 1.687 0.194Respiratory 1,100 1.323 0.998 1.754 1,300 *1.698 1.307 2.206 1.617 0.204<75 years oldǂ (n = 266,600) ≥75 years oldǂ (n = 32,900)Non-accidental 13,100 *1.248 1.151 1.353 12,600 *1.237 1.140 1.342 0.023 0.880Circulatory 3,500 *1.239 1.058 1.450 4,900 1.100 0.965 1.254 1.295 0.255Respiratory 1,000 *1.553 1.158 2.083 1,300 *1.461 1.136 1.878 0.096 0.757Ever Smoked (n = 215,100) Never Smoked (n = 84,100)Non-accidental 19,400 *1.231 1.152 1.315 6,300 *1.397 1.242 1.571 3.381 0.066Circulatory 6,000 *1.164 1.034 1.311 2,300 *1.287 1.060 1.563 0.749 0.387Respiratory 1,900 *1.449 1.174 1.788 400 *1.966 1.236 3.129 1.376 0.241Obese I and II (n = 78,900) Normal weight (n = 93,700)Non-accidental 6,200 *1.215 1.077 1.370 8,700 *1.264 1.147 1.394 0.250 0.617Circulatory 2,100 1.110 0.903 1.364 2,700 1.125 0.945 1.339 0.009 0.922Respiratory 500 *1.757 1.146 2.694 900 *1.408 1.041 1.905 0.688 0.407Obese II (n = 24,200)b Normal weight (n = 93,700)Non-accidental 1,900 1.142 0.919 1.419 8,700 *1.264 1.147 1.394 0.698 0.403Circulatory 700 0.888 0.609 1.294 2,700 1.125 0.945 1.339 1.247 0.264<5 fruit/veg servings (n = 153,200) ≥5 fruit/veg servings (n = 101,100)Non-accidental 12,900 *1.217 1.124 1.318 8,500 *1.199 1.087 1.322 0.054 0.817Circulatory 4,100 1.098 0.954 1.263 2,900 *1.322 1.117 1.563 2.764 0.096Respiratory 1,200 *1.421 1.091 1.852 700 *1.505 1.078 2.101 0.070 0.792Regular drinker (n = 141,700) Not regular drinkera (n = 91,800)Non-accidental 9,600 *1.280 1.168 1.403 13,300 *1.280 1.182 1.387 <0.001 1.000Circulatory 2,900 *1.257 1.065 1.483 4,600 *1.201 1.048 1.376 0.174 0.677Respiratory 800 *1.473 1.070 2.027 1,300 *1.449 1.120 1.875 0.006 0.938†All models are stratified by age (5 year categories) and sex, and adjusted for the following covariates: immigrant status, visible minority status, Aboriginal status, marital status, educational attainment, incomeadequacy quintile, employment, body mass index, fruit and vegetable consumption, smoking, and alcohol. For each comparison, the stratum or covariate being compared was not included as a stratum/covariate inthe model (i.e., smoking was not included as a covariate in the smoking comparison)ǂAge at entry into Cohort+Cochran’s Q test for significant difference of HR between groups*Significant HR (p < 0.05)aIncludes occasional, former, or never drinkerbRespiratory mortality not shown; mortality for obese II: n < 200Pinaultetal.EnvironmentalHealth (2016) 15:18 Page10of15were statistically significant (Table 5). However, the HRof respiratory disease mortality was particularly highamong never smokers (HR: 1.97; 95 % CI: 1.23–3.13 per10 μg/m3 increase in PM2.5) and among obese respon-dents (HR = 1.76, 95 % CI: 1.15–2.69 per 10 μg/m3 in-crease in PM2.5) (Table 5).We fitted a nonparametric smoothing (spline) toexamine the shape of the association between exposureand non-accidental mortality within the fully adjustedmodel. The relationship between the logarithm of thehazard function and PM2.5 is presented in Fig. 2 inaddition to its 95 % confidence intervals. We specified areference concentration of 1 μg/m3 which forces thepredicted log-hazard function to equal 0 at the referencelevel. The smoothed curve generally increased with in-creasing concentration, however the confidence intervalsare relatively wide making it difficult to speculate on aspecific shape of the concentration-mortality associationbased on this graphical evidence. Our estimate of thethreshold concentration was 0 μg/m3 with an upper95 % CI value of 4.5 μg/m3.DiscussionWithin our cohort, exposure to PM2.5 assigned atbaseline was associated with an increased risk of non-accidental mortality and mortality due to circulatoryand respiratory disease. Risks for all causes of deathexamined were greatest after adjusting for socioeco-nomic and ecological covariates, though were reducedafter adjusting for smoking, alcohol consumption,BMI, and fruit/vegetable consumption. The largesthazard ratios per 10 μg/m3 increase in PM2.5 wereobserved for respiratory mortality compared to theother cause-specific estimates. Elevated risk was ob-served for respiratory mortality associated with airpollution among obese respondents and never-smokers, though the differences between these andreference groups were not statistically significant. Wealso examined the shape of the exposure-responsecurve, and although the lowest measured concentration ofPM2.5 was 1 μg/m3, we found no lower threshold for re-sponse. Although this finding is potentially informative forburden assessment, it is worth noting that we did not dis-tinguish between anthropogenic and natural sources ofPM2.5 in this study.This study adds to previous work in Canada, which hasa generally lower mean PM2.5 exposure than other coun-tries, by providing direct adjustments for behavioural co-variates (i.e., smoking and obesity) that are knowncontributors to mortality. This study used similar method-ology to a previous study in Canada, the Canadian CensusHealth and Environment Cohort (CanCHEC) [3], but wasunable to directly evaluate the role of behavioural covari-ates. In general, our HR estimates for non-accidental mor-tality (HR = 1.26; 95 % CI: 1.19–1.34) were greater thanthose in CanCHEC (HR = 1.15; 95 % CI: 1.13–1.16;Cochran’s Q = 9.3, p < 0.01), though our estimates for cir-culatory death were similar (CCHS HR= 1.19; 95 % CI:1.07–1.31; CanCHEC HR= 1.16; 95 % C.I.: 1.13–1.18;Cochran’s Q = 0.1, p = 0.8) (all units per 10 μg/m3 increasein PM2.5) [3].The fact that we found stronger associations betweenmortality and PM2.5 here than were observed in theCanCHEC study [3] might be due to improvements inestimates of PM2.5. A new PM2.5 model developed at amuch finer scale (1 km2 grid rather than 10 km2 grid)allowed respondents to be assigned more accurate, finer-scale estimates exposure to of PM2.5. This improved ex-posure model may have a particularly strong effect onrespondents who live in mid-sized cities (e.g., Calgary,Edmonton) that would otherwise have been assigned alower, regional (i.e., rural) average (Fig. 1). However, thisimprovement is expected to be limited somewhat by thelimitations of location error in geocoding residencesbased on postal code, as well as respondent mobilitythroughout the study area, resulting in differences inpersonal exposure. Another strength of this study wasthat it assigned exposures to respondents in the threeyears preceding death, thereby ensuring that exposurealways preceded health effects rather than beingassigned concurrently. This method also takes long-termvariation of exposure into account.In our study, HR estimates increased after theaddition of ecological covariates, which differs from theFig. 2 Nonparametric estimates of the dependence of non-accidentalmortality on PM2.5 exposure among in-scope respondents in theCCHS-cohort linked to a PM2.5 dataset (log hazard ratio with 95 %confidence intervals). The model was stratified by age and sex, andadjusted for all covariates (Table 1). Model predictions were made upto the 99th percentile of the PM2.5 exposure distributionPinault et al. Environmental Health  (2016) 15:18 Page 11 of 15earlier CanCHEC study, in which the addition of eco-logical covariates served to decrease the HR estimates[3]. As described earlier, the ecological covariates usedhere were derived for smaller areas than in the CanCHECstudy due to the absence of Census Tracts in rural areas.The methodological differences in deriving ecologicalcovariates, particularly at a finer scale (i.e., assigningDA-level covariates rather than CT-level covariates),may also be one of the primary reasons why differ-ences in HR estimates were observed between thisstudy and CanCHEC, since fine scale covariates wouldbe more spatially variable and covariates would moreaccurately reflect local socioeconomic conditions. Indeed,when the ecological covariates were removed from theCox models of non-accidental mortality, the otherwisefully adjusted model provided an HR = 1.085 (Table 3),which is more consistent with the fully adjusted models inCanCHEC [3]. Ecological covariates included in this studywere all positively correlated with PM2.5 (Table 2). Giventhe much greater PM2.5 exposure in urban environments,this association for recent immigrants and persons of higheducational attainment is possibly due to a higherpopulation of both in cities. The correlation withPM2.5 was weaker for the proportion of low incomefamilies, which was consistent with the similar pro-portions of regional-adjusted low-income families inrural and urban environments [22].Our HRs for non-accidental mortality were greaterthan those reported for all-cause mortality in other,international studies that had considered the same be-havioural covariates, though were generally similar whenecological covariates were excluded from our estimates[6, 23–25]. For example, the American Cancer Societystudy, which included 1.2 million adults in the UnitedStates, estimated an HR for all-cause mortality of 1.06per 10 μg/m3 increase in PM2.5 (95 % C.I.: 1.02–1.11)after controlling for behavioural covariates, though thatstudy did not include ecological covariates [23]. Simi-larly, a global, pooled meta-analysis estimated an all-cause mortality HR of 1.06 (95 % C.I.: 1.04–1.08) per10 μg/m3 increase in PM2.5 [6]. However, our resultswith ecological covariates were not significantly differentfrom those of a large meta-analysis of European studies,where the pooled HR estimate for natural-cause mortalityadjusted for socioeconomic and behavioural covariates(though not large-scale socioeconomic covariates) was1.09 per 5 μg/m3 increase in PM2.5 (95 % C.I.: 1.03–1.14)[26]. This estimate was not significantly lower than in ourstudy (Cochran’s Q = 0.8, p = 0.4), where HR = 1.12 (95 %C.I.: 1.09–1.16) when scaled to a 5 μg/m3 increase inPM2.5.Hazard ratio estimates for mortality due to circulatorydisease (i.e., HR = 1.19) were generally consistent withthose reported in the international literature, includingthe Harvard Six Cities study extended follow-up, whichreported a HR of 1.28 per 10 μg/m3 increase in PM2.5 [6,27], and a study in the U.K., which reported an HR of1.05 per 1.9 μg/m3 increase in PM2.5 after adjustmentfor sex, age, BMI, and smoking (our study: HR = 1.03,95 % CI: 1.01–1.05 when scaled to a 1.9 μg/m3 increase)[28]. However, our estimate was much greater than thatreported from a study in Rome (HR = 1.06, 95 % CI:1.04–1.08), which adjusted for some individual andarea-based socioeconomic covariates [24], and theDutch Environmental Longitudinal Study (DUELS),which reported an HR of 1.09 (95 % CI: 1.06–1.12)per 10 μg/m3 increase in PM2.5 [25].Our hazard ratio estimates for respiratory disease(HR = 1.52) were generally greater than those in theliterature, though literature estimates for HRs varyamong studies. For example, one study in Rome thatused area-based socioeconomic covariates identified anon-significant HR of 1.03 for respiratory disease[24]. On the other hand, the California TeachersStudy identified an HR for respiratory mortality of1.21 [29], and the Dutch cohort (DUELS) estimatedan HR of 1.18 [25], which were similar to our HR es-timate for respiratory mortality prior to adjustmentfor ecological covariates (HR = 1.21). Another study inthe UK reported an HR of 1.17 (95 % CI: 1.12–1.22)per 1.9 μg/m3 increase in PM2.5 [28]. Our HR esti-mate after adjustment for ecological covariates waslower than this study (HR = 1.08, 95 % CI: 1.04–1.12)when scaled to a 1.9 μg/m3 increase in PM2.5.Our study also evaluated the role of effect modifica-tion by sex, age and behavioural covariates, and found asignificantly greater HR estimate for non-accidentalmortality among men than women. In a pooled Euro-pean analysis of multiple cohorts, HRs were elevatedamong men but not women [26]. Our results are overallsimilar, although our generally greater HR estimates fornon-accidental mortality might explain why HR was sig-nificant for both men and women. Men also had agreater HR than did women for circulatory disease mor-tality (though the differences were not significant), simi-lar to the AHS cohort [5]. This finding was inconsistentwith the results of a small (n = 3,239) cohort of white,non-smoking adults, where the relative risk of coronaryheart disease mortality was elevated among women butnot among men in a fully adjusted model [30]. Observeddifferences might be, at least in part, explained by rela-tively small cohort sizes.Our HR estimates for non-accidental and circulatorymortality among obese and normal weight groups werenot significantly different. Effect modification of cardio-vascular mortality by obesity had previously been evalu-ated elsewhere in two all-female cohorts. One studyidentified a significantly greater HR with increasingPinault et al. Environmental Health  (2016) 15:18 Page 12 of 15BMI, with an HR for obese women of 1.35 (95 % C.I.:1.12–1.64 per 10 μg/m3 increase in PM2.5) [31]. The otherstudy did not test differences statistically among groupsbut did report an HR of 1.99 (95 % C.I.: 1.23–3.22 per10 μg/m3 increase in PM2.5) for obese women [32]. TheACS also reported a greater HR among obese men [5]. Inour study, obese respondents also had a high risk of re-spiratory mortality (HR = 1.76; 95 % C.I.: 1.15–2.69),though possibly due to a small number of deaths (n =500), the HR estimate was not significantly different fromthe normal weight population.In our study, persons who had never smoked had aqualitatively greater risk of non-accidental and circu-latory mortality from fine particulate exposure thanthose who had smoked, though the difference be-tween groups was non-significant (Table 5). This find-ing was consistent with the literature, where amarginally greater risk of cardiovascular mortality wasobserved among never smokers than among currentor former smokers [6, 23, 33, 34]. In a Dutch cohort,respiratory mortality was qualitatively greater amongcurrent smokers than never smokers [34], a findingthat was not consistent with our study.There were several limitations with our study that maycontribute to uncertainty in our estimates. The cohortwas chosen because of the inclusion of various behav-ioural covariates, but it is generally much smaller thanthat of CanCHEC, which used the Census of population(i.e., 20 % of the population of Canada) [3]. Mean esti-mates of PM2.5 in Canada are generally lower than inother study countries [6], and the effect size is relativelysmall, requiring a large sample size to have adequatepower for HR estimation. As a result, in our study the95 % CIs were very wide in comparison to other studies[3], and we were also unable to adequately assess theshape of the concentration-response curves for othercauses of death. It is also worth mentioning that ourstudy relied on self-reported estimates for BMI andsmoking. Although we were able to mathematically ad-just BMI for self-reporting error based on measuredBMI from another survey, it is possible that estimates ofsmoking may underrepresent actual smoking rates. Add-itionally, the follow-up period in our study was relativelyshort, particularly for respondents who entered the co-hort in the final survey year (i.e., 2008, with a maximumof 4 years of follow-up). However, respondents enteringthe cohort in the first year of survey and who hadremained in the cohort for the entire period werefollowed for a maximum of 12 years, which is compar-able to the mean follow-up period (i.e., 12.6 years) in areview of other cohorts examining the same relationship[6]. The limitation of having a short follow-up periodwas mitigated somewhat by considering exposures thatpreceded the event.In creating the cohort, 69,300 CCHS respondents wereexcluded since they were not linked to the HTSF (tax)file and were therefore not candidates for the probabilis-tic linkage. The excluded population were those who didnot file a tax return, and the characteristics of this popu-lation differed somewhat from the cohort. In general,the excluded population was younger and had a lowereducational attainment than the final cohort. Therefore,the cohort might be slightly biased towards higher edu-cational attainment and those active in the labour mar-ket, though these same characteristics were used foradjustment in survival models.Estimates of PM2.5 exposure were assigned at baselineat the person’s place of residence. Accuracy in geocodingresidences was limited by the program PCCF+, whichassigns residences to postal code representative points.The size of postal codes is relatively small (i.e. typically afew city blocks) in urban centres; therefore the PCCF+program is highly accurate within these areas. However,estimates of PM2.5 exposure in rural areas are less likelyto have been assigned accurately since postal code areascan be quite large. We performed a sensitivity analysisthat considered only cohort members that lived withinurban areas (i.e., Census Metropolitan Areas), and des-pite exposures being much greater in urban areas, re-sults were not significantly different than those reportedabove (HR = 1.19, 95 % CI: 1.11–1.27, Cochran’s Q = 1.71,p = 0.19). Given the short follow-up period, we also didnot assess mobility in this study, making the assumptionthat respondents did not move. By not assigning air pollu-tion exposures based on changes to residential history, itis expected that there would be some degree of exposuremisclassification associated with this limitation. A previ-ous study using CanCHEC considered the assignment ofexposures at baseline vs. considering mobility during thefollow-up period on mortality risk attributed to PM2.5. Ingeneral, there was very little difference in HR estimates(i.e., HR = 1.03, 95 % CI: 1.02–1.03 from baseline expos-ure, vs. HR = 1.04, 95 % CI: 1.03–1.04 for exposure consid-ering mobility) [35]. Although about 41 % of Canadiansmoved within the five-year period of 2001 to 2006 [36],the majority of moves were within cities or regions ofsimilar PM2.5 exposures (not published). To assess thislimitation, we ran a sensitivity analysis where we includedonly persons who had at least 3 years of residence in thesame postal code. HRs for non-accidental mortality weresimilar to those for the entire cohort (HR = 1.28, 95 % CI:1.19–1.37).Finally, the cohort was developed based on a probabil-istic linkage methodology to assign deaths to CCHSmembers. We attempted to reduce the potential for link-age error by limiting our cohort to persons linked to atax file, since mortality rates among cohort members notlinked to a tax file were substantially lower due to fewerPinault et al. Environmental Health  (2016) 15:18 Page 13 of 15elements of respondent data that could be used forlinkage.ConclusionsIn general, this study documented an association be-tween non-accidental, circulatory, and respiratorymortality and fine particular matter in a cohort ad-justed for socioeconomic, ecological, and behaviouralcovariates and exposed to a relatively low exposuredistribution (mean = 6.3 μg/m3). Although our CIwere wide in the concentration-response curve, an in-creased risk of mortality was observed even at verylow concentrations of PM2.5 (Fig. 2), at values lowerthan the WHO guideline of 10 μg/m3 [2]. Furtherstudies on a larger cohort are needed to evaluate theshape of the concentration-response curve at theselower concentrations of PM2.5. We also updated theresults of previous Canadian studies by using an im-proved, finer-scale exposure model to assign PM2.5estimates to cohort members, which may have, inpart, caused observed increases in HR estimates rela-tive to CanCHEC [3]. Finally, this study indicates thatthe addition of fine-scale behavioural covariates servesto reduce the HR estimates compared to the other-wise fully adjusted survival models.Additional filesAdditional file 1:Figure S1. Selection of Study Cohort.(PDF 88 kb)Additional file 2: Table S1. Comparison of all variables (Pearson'scorrelation or ANOVA/T-Test). (XLSX 22 kb)AbbreviationsAHS: United States Agricultural Health Study; BMI: Body Mass Index;CanCHEC: Canadian Census Health and Environment Cohort; CCHS: CanadianCommunity Health Survey; CD: Census Division; CI: confidence interval;CMDB: Canadian Mortality Database; CT: census tract; DA: dissemination area;HR: hazard ratio; HTSF: Historical Tax Summary File; -2LL: (−2) Log-likelihoodratio; MODIS: moderate resolution imaging spectroradiometer; PCCF+: PostalCode Conversion File Plus; PM2.5: Fine particulate matter; WHO: World HealthOrganization.Competing interestsThe authors do not declare any competing interests.Authors’ contributionsLP designed the study, linked the cohort to the air pollution models,conducted the statistical and GIS analyses, and drafted the manuscript.MT participated in the study design, the cohort linkage, and providedfeedback on the cohort. DLC participated in the study design, andprovided assistance with statistical techniques and covariate preparation.SW conceived of the study and participated in the study design and indrafting the manuscript. AVD and RVM developed the air pollutionmodels and provided feedback on their use in an epidemiologiccontext. MB and HC provided early feedback to improve the analysesand the presentation of results. RTB coordinated the study andparticipated in the study design and the statistical analysis. All authorsactively edited the manuscript and approved the final manuscript.AcknowledgementsThe authors wish to acknowledge the contributions of Masoud Nasari(Health Canada) for his assistance with R software and Saeeda Khan(Statistics Canada) for her assistance with the PCCF+ program.Author details1Health Analysis Division, Statistics Canada, 100 Tunney’s Pasture Driveway,Ottawa, ON K1A 0T6, Canada. 2New Brunswick Institute for Research, Data,and Training (NB-IRDT) and Department of Sociology, University of NewBrunswick, Fredericton, NB, PO Box 4400, E3B 5A3, Canada. 3Air Health EffectsScience Division, Health Canada, 269 Laurier Avenue West, Ottawa, ON K1A0K9, Canada. 4Department of Physics and Atmospheric Science, DalhousieUniversity, 6310 Coburg Road, Halifax, NS, Canada. 5Harvard-SmithsonianCenter for Astrophysics, 60 Garden St, Cambridge, MA 02138, USA. 6Facultyof Medicine/School of Population and Public Health, University of BritishColumbia, 2206 East Mall, Vancouver, BC V6T 1Z3, Canada. 7Public HealthOntario, Suite 300, 480 University Avenue, Toronto, ON M5G 1V2, Canada.8Environmental Health Science and Research Bureau, Health Canada, 50Columbine Driveway, Ottawa, ON, Canada.Received: 30 October 2015 Accepted: 28 January 2016References1. Lin SS, Vos T, Flaxman AD, Danaei G, Shibuya K, Adair-Rohani H, et al. Acomparative risk assessment of burden of disease and injury attributable to67 risk factors and risk factor clusters in 21 regions, 1990–2010: A systematicanalysis for the Global Burden of Disease Study 2010. Lancet. 2012;380:2224–60.2. GBD 2013 Risk Factors Collaborators. Global, regional and nationalcomparative risk assessment of 79 behavioural, environmental/occupationaland metabolic risks or clusters of risks in 188 countries 1990–2013: asystematic analysis for the GBD 2013. Lancet. 2015; in press.3. Crouse DL, Peters PA, van Donkelaar A, Goldberg MS, Villeneuve PJ,Brion O, et al. Risk of nonaccidental and cardiovascular mortality inrelation to long-term exposure to low concentrations of fine particulatematter: A Canadian national-level cohort study. Environ Health Persp.2012;120:708–14.4. Shin HH, Cakmak S, Brian O, Villeneuve P, Turner MC, Goldberg MS, et al.Indirect adjustment for multiple missing variables applicable toenvironmental epidemiology. Environ Res. 2014;134:482–7.5. Weichenthal S, Villeneuve PJ, Burnett RT, van Donkelaar A, Martin RV, JonesRR, et al. Long-term exposure to fine particulate matter: association withnonaccidental and cardiovascular mortality in the Agricultural Health StudyCohort. Environ Health Persp. 2014;112:609–15.6. Hoek G, Krishnan RM, Beelen R, Peters A, Ostro B, Brunekreef B, et al. Long-term air pollution exposure and cardio-respiratory mortality: a review.Environ Health. 2013;12:43.7. Canada S. Canadian Community Health Survey (CCHS) annual component:User guide, 2007–2008 microdata files. Ottawa, ON, Canada: StatisticsCanada; 2009.8. Fellegi IP, Sunter AB. A theory for record linkage. J Am Stat Assoc.1969;64:1183–210.9. Wilkins R, Tjepkema M, Mustard C, Choinière R. The Canadian censusmortality follow-up study, 1991 through 2001. Health Rep. 2008;19:25–43.10. Wilkins R, Peters PA. PCCF + Version 5K* User’s Guide. Automatedgeographic coding based on the Statistics Canada Postal Code ConversionFiles including Postal Codes through May 2011. Statistics Canada: Ottawa,ON, Canada, 2012. Catalogue no. 82F0086-XDB.11. van Donkelaar A, Martin RV, Spurr RJD, Burnett RT. High-resolution satellite-derived PM2.5 from optimal estimation and geographically weightedregression over North America. Environ Sci Technol. 2015; in press(doi:10.1021/acs.est.5b02076).12. Boys BL, Martin RV, van Donkelaar A, MacDonell RJ, Hsu NC, Cooper MJ,et al. Fifteen-year global time series of satellite-derived fine particulatematter. Environ Sci Technol. 2014;48:11109–18.13. Cox DR. Regression models and life tables. J Royal Stat Soc B.1972;20:187–220.14. Statistics Canada. Census Dictionary 2006. Statistics Canada: Ottawa, ON,Canada, 2010. Catalogue no. 92-566-X.Pinault et al. Environmental Health  (2016) 15:18 Page 14 of 1515. Ross NA, Oliver LN, Villeneuve PJ. The contribution of neighbourhoodmaterial and social deprivation to survival: a 22-year follow-up of more than500,000 Canadians. Int J Environ Res Public Health. 2013;10:1378–91.16. Crouse DL, Peters PA, Villeneuve PJ, Proux M-O, Shin HH, Goldberg MS, et al.Within- and between-city contrasts in nitrogen dioxide and mortality in 10Canadian cities; a subset of the Canadian Census Health and EnvironmentCohort (CanCHEC). J Expo Sci Env Epi. 2014. doi:10.1038/jes.2014.89.17. Connor Gorber S, Shields M, Tremblay MS, McDowell I. The feasibility ofestablishing correction factors to adjust self-reported estimates of obesity inthe Canadian Community Health Survey. Health Rep. 2008;19:71–82.18. World Health Organization. BMI Classification: The international classificationof adult underweight, overweight, and obesity according to BMI. (2006). Accessed28 Jul 2015.19. Chen H, Burnett RT, Kwong JC, Villeneuve PJ, Goldberg MS, Brook RD, et al.Spatial associations between ambient fine particulate matter and incidenthypertension. Circulation. 2013;129:562–9.20. Conover W. Practical Nonparametric Statistics. 3rd ed. New York: Wiley;1999.21. Meira-Machado L, Cadarso-Suárez C, Gude F, Araújo A. smoothHR: An Rpackage for pointwise nonparametric estimation of hazard ratio curves ofcontinuous predictors. Comput Math Methods Med. 2013; doi:10.1155/2013/745742.22. Fortin, M. A comparison of rural and urban workers living in low-income.Rural and Small Town Canada Analysis Bulletin (Statistics Canada).2008;7:1–18. Cat. no. 21-006-XIE.23. Pope CA, Burnett RT, Thun MJ, Calle EE, Krewski D, Ito K, et al. Lung cancer,cardiopulmonary mortality and long-term exposure to fine particulate airpollution. JAMA. 2002;287:1132–41.24. Cesaroni G, Badaloni C, Gariazzo C, Staffogia M, Sozzi R, Davoli M, et al.Long-term exposure to urban air pollution and mortality in a cohort ofmore than a million adults in Rome. Environ Health Perspect.2013;121:324–31.25. Fischer PH, Marra M, Ameling CB, Hoek G, Beelen R, de Hoogh K, et al. Airpollution and mortality in seven million adults: The Dutch EnvironmentalLongitudinal Study (DUELS). Environ Health Perspect. 2015; doi: 10.1289/ehp.1408254.26. Beelen R, Raaschou-Nielsen O, Stafoggia M, Jovanovic Andersen Z,Weinmayr G, Hoffmann B, et al. Effects of long-term exposure to airpollution on natural-cause mortality: an analysis of 22 European cohortswithin the multicentre ESCAPE project. Lancet. 2014;383:785–95.27. Laden F, Schwartz J, Speizer FE, Dockery DW. Reduction in fine particulateair pollution and mortality: extended follow-up of the Harvard Six CitiesStudy. Am J Resp Crit Care. 2006;173:667–72.28. Carey IM, Atkinson RW, Kent AJ, van Staa T, Cook DG, Anderson HR.Mortality associations with long-term exposure to outdoor airpollution in a national English cohort. Am J Respir Crit Care Med.2013;187:1226–33.29. Ostro B, Lipsett M, Reynolds P, Goldberg D, Hertz A, Garcia C, et al. Longterm exposure to constituents of fine particulate air pollution and mortality:results from the California teachers study. Environ Health Persp.2010;118:363–9.30. Chen LH, Knutsen SF, Shavlik D, Beeson WL, Petersen F, Ghamsary M, et al.The association between fatal coronary heart disease and ambientparticulate air pollution: are females at greater risk? Environ Health Persp.2005;113:1723–9.31. Miller KA, Siscovick DS, Sheppard L, Shepherd K, Sullivan JH, Anderson GL,et al. Long-term exposure to air pollution and incidence of cardiovascularevents in women. New Engl J Med. 2007;356:447–58.32. Puett RC, Schwartz J, Hart JE, Yanosky JD, Speizer FE, Suh H, et al. Chronicparticulate exposure, mortality, and coronary heart disease in the nurses’health study. Am J Epidemiol. 2008;168:1161–8.33. Krewski D, Jerrett M, Burnett RT, Ma R, Hughes E, Shi Y, et al. 2009.Extended follow-up and spatial analysis of the American Cancer Societystudy linking particulate air pollution and mortality. Res Rep Health EffInst. 2009;140:5–136.34. Beelen R, Hoek G, van Den Brandt PA, Goldbohm RA, Fisher P,Shouten LJ, et al. Long-term effects of traffic-related air pollution onmortality in a Dutch cohort (NLCS-AIR study). Environ Health Persp.2008;116:196–202.35. Crouse DL, Peters PA, Hystad P, Brook JR, van Donkelaar A, Martin RV, et al.Ambient PM2.5, O3, and NO2 exposures and associations with mortality over16 years of follow-up in the Canadian Census Health and EnvironmentCohort (CanCHEC). Environ Health Persp. 2015;123:1180–6.36. Statistics Canada. 2006 Census data products: 2006 census trends. (2010). Accessed 12Dec 2015.•  We accept pre-submission inquiries •  Our selector tool helps you to find the most relevant journal•  We provide round the clock customer support •  Convenient online submission•  Thorough peer review•  Inclusion in PubMed and all major indexing services •  Maximum visibility for your researchSubmit your manuscript your next manuscript to BioMed Central and we will help you at every step:Pinault et al. Environmental Health  (2016) 15:18 Page 15 of 15


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items