Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 7-Day Trial for You or Your Team.

Learn More →

A Comparison of Short-term and Long-term Air Pollution Exposure Associations with Mortality in Two Cohorts in Scotland

A Comparison of Short-term and Long-term Air Pollution Exposure Associations with Mortality in... Research A Comparison of Short-term and Long-term Air Pollution Exposure Associations with Mortality in Two Cohorts in Scotland 1 2 3 4 1* 5,6,7 Iain J. Beverland, Geoffrey R. Cohen, Mathew R. Heal, Melanie Carder, Christina Yap, Chris Robertson, 8 4 Carole L. Hart, and Raymond M. Agius 1 2 3 Department of Civil Engineering, University of Strathclyde, Glasgow, United Kingdom; Edinburgh, United Kingdom; School of Chemistry, University of Edinburgh, Edinburgh, United Kingdom; Centre for Occupational and Environmental Health, University of Manchester, Manchester, United Kingdom; Department of Mathematics and Statistics, University of Strathclyde, Glasgow, United 6 7 8 Kingdom; Health Protection Scotland, Glasgow, United Kingdom; International Prevention Research Institute, Lyon, France; Institute of Health and Wellbeing, Public Health, University of Glasgow, Glasgow, United Kingdom cumulative effects that increase the sensitivity Background : Air pollution– mortality risk estimates are generally larger at longer-term, compared of highly exposed population subgroups. with short-term, exposure time scales. However, there have been few studies oBjective : We compared associations between short-term exposure to black smoke (BS) and mor- comparing the estimated effects of the same tality with long-term exposure–mortality associations in cohort participants and with short-term pollutant on the same population at both exposure–mortality associations in the general population from which the cohorts were selected. short- and long-term time scales. In the pres- M e t h o d s : We assessed short-to-medium–term exposure–mortality associations in the Renfrew– ent study, we present such a comparison, using Paisley and Collaborative cohorts (using nested case–control data sets), and compared them with mortality data from approximately 25 years of long-term exposure–mortality associations (using a multilevel spatiotemporal exposure model and follow-up on two Scottish cohorts with a com- survival analyses) and short-to-medium–term exposure–mortality associations in the general popu- bined size of nearly 20,000 participants and lation (using time-series analyses). estimates of temporal and spatial variations results : For the Renfrew–Paisley cohort (15,331 participants), BS exposure–mortality associa- in black smoke (BS) air pollution in the con- tions were observed in nested case–control analyses that accounted for spatial variations in pollution tiguous urban area of Glasgow, Paisley, and exposure and individual-level risk factors. e Th se cohort-based associations were consistently greater Renfrew in central Scotland, United Kingdom than associations estimated in time-series analyses using a single monitoring site to represent gen- (the Glasgow conurbation). Short-term effects eral population exposure {e.g., 1.8% [95% confidence interval (CI): 0.1, 3.4%] vs. 0.2% (95% CI: were estimated using a nested case–control 0.0, 0.4%) increases in mortality associated with 10-μg/m increases in 3-day lag BS, respectively}. approach in which the pollution experienced Exposure–mortality associations were of larger magnitude for longer exposure periods [e.g., 3.4% (95% CI: –0.7, 7.7%) and 0.9% (95% CI: 0.3, 1.5%) increases in all-cause mortality associated by each case immediately before death was with 10-μg/m increases in 31-day BS in case–control and time-series analyses, respectively; and compared with that experienced by controls 10% (95% CI: 4, 17%) increase in all-cause mortality associated with a 10-μg/m increase in geo- of the same age in approximately the same metic mean BS for 1970–1979, in survival analysis]. time period and at approximately the same c onclusions : After adjusting for individual-level exposure and potential confounders, short-term time of year. The effects of long-term pollu - exposure–mortality associations in cohort participants were of greater magnitude than in compara- tion exposure were estimated from a detailed ble general population time-series study analyses. However, short-term exposure–mortality associa- spatio temporal exposure model. Additionally, tions were substantially lower than equivalent long-term associations, which is consistent with the short-term exposure–mortality associations possibility of larger, more persistent cumulative effects from long-term exposures. were estimated using conventional Poisson k ey words : air, associations, cohort, exposure–mortality, long term, pollution, short term, time- regression time-series models in the urban series. Environ Health Perspect 120:1280–1285 (2012). http://dx.doi.org/10.1289/ehp.1104509 population from which the majority of cohort [Online 6 June 2012] participants were selected. Methods The magnitude of adverse health effects effects have examined relationships between associated with air pollution has been the daily counts of deaths in a given popula- Study population. The Renfrew–Paisley cohort subject of extensive research (Brook et al. tion and daily pollutant and meteoro logical was recruited from residents 45–64 years of 2010; Pope and Dockery 2006). An aspect varia bles recorded at central monitoring sites Address correspondence to I. Beverland, University of epidemiological research of importance to using Poisson regression modeling and gen- of Strathclyde, Department of Civil Engineering, policy makers is the time scale over which eralizations thereof. This type of time-series John Anderson Building, 107 Rottenrow, Glasgow adverse effects are most apparent (Künzli study has limited capacity to take account G4 0NG UK. Telephone: 0141 548 3202. Fax: 0141 2005). Discrepancies between the magnitudes of spatial variation in individual exposures 553 2066. E-mail: [email protected] of short-to-medium–term exposure effects or to investigate the possible modification *Current address: MRC Hub for Trials Methodology estimated at time scales of days to weeks, and of pollution effects by individual risk factors Research, University of Birmingham, Birmingham, UK. We are grateful for encouragement and support from of long-term exposure effects estimated by such as age, smoking history, and previous the late D. Hole, former professor of Epidemiology following cohorts over years to decades, has disease, although analysis using case-crossover and Biostatistics and head of West of Scotland Cancer been a particular concern (Pope 2007). methods (Carracedo-Martinez et al. 2010) Surveillance Unit, for his enthusiastic role in planning In longitudinal cohort studies, long-term may reduce some of these difficulties. The and implementing this research. average pollution exposures of individual estimates of air pollution effects from long- We gratefully acknowledge funding from the cohort participants are estimated, and the term exposure studies involving cohorts have Department of Health (England) Policy Research Programme as part of the Initiative on Air Pollution contribution of the geographical variation generally been considerably larger than those (grant 0020015). among these exposures to the risk of adverse from short-term time-series studies even The views expressed here are those of the authors and events over the entire follow-up period is after adjusting for other individual risk fac- not necessarily the Department of Health (England). determined by survival analysis, adjusting for tors (Pope 2007). It is plausible that larger e Th authors declare they have no actual or potential characteristics that may affect disease risk. In magnitudes of association between long-term competing financial interests. contrast, many studies of short-term exposure exposure and mortality may be attributable to Received 18 September 2011; accepted 6 June 2012. | | 1280 volume 120 number 9 September 2012 • Environmental Health Perspectives Short and long-term exposure–mortality associations age from Renfrew and Paisley in west cen- was considered the most appropriate site to use were randomly selected from the full set avail- tral Scotland, which are contiguous to each to estimate temporal variations in background able for each case. For all-cause deaths, 96% other and to the city of Glasgow. The cohort air pollution. Three averaging periods were (8,342/8,700) of cases in the Renfrew–Paisley comprised 78% of the target population, with used for assessing short- and medium-term cohort and 76% (1,160/1,524) of cases in the 15,402 participants screened between 1972 exposure: a 3-day average over the 3 days pre- Collaborative cohort had the full complement and 1976 (Hawthorne et al. 1995). The ceding the day of death; a 7-day average over of 9 controls and 99% and 92%, respectively, Collaborative cohort is an occupational cohort the day of death and the preceding 6 days; and had at least 4 controls. Six nested case–control of 7,028 participants recruited from 27 work- a 31-day average over the day of death and the data sets of this type were created for three case places in central Scotland between 1970 and preceding 30 days. outcomes (all-cause, cardiovascular, and respi- 1973 (Davey Smith et al. 1998). These two Analysi s of long-term pollution expo- ratory mortality) in the two cohorts. We then cohorts were designed for study together, sure for the decade 1970–1979 was based on used conditional logistic regression to compare and in this context are referred to as the records from 181 monitoring sites distributed the pollution experienced by the case immedi- Midspan study (Hart et al. 2005). In the pres- widely over central Scotland. Many of these ately before death and that experienced by con- ent study, we examined short- and medium- sites had long periods of missing data. After trols of the same age in approximately the same term exposure–mortality associations in 1979, the number of sites was greatly reduced. time period and at approximately the same 15,331 participants in the Renfrew–Paisley The use of the BS metric (derived from time of year, with adjustment for potential con- cohort with complete postal code information the darkness of particulate matter collected on founding variables measured at baseline: smok- and 3,818 Collaborative cohort participants a filter through which air has been sampled) ing history (six levels), social class (six levels), (35–64 years of age at recruitment) residing in is well established in historical UK monitor- body mass index (five quintiles), marital status the Glasgow urban area with complete postal ing networks and is considered to be a good (four levels), systolic blood pressure (linear), code information. To provide an indication marker for traffic and other combustion- and total cholesterol (linear). This conditional of geographical scale the contiguous conurba- related urban air pollution through similar logistic regression model does not correspond tion of Glasgow, Paisley, and Renfrew can be measurements of filter reflectance (Hochadel to a survival model because generally different encompassed within a radius of 12 km, with et al. 2006; Janssen et al. 2011). controls are in each case–control set. However, Renfrew and Paisley encompassed by radii of Temperature data. We obtained daily aver- because of the correspondence between the 1.5 and 3.5 km, respectively, within this 12-km ages of hourly temperatures from 0700–2300 likelihoods for proportional hazards and con- radius. Participants in both cohorts underwent hours at the UK meteorological office site at ditional logistic regression modeling (Collett physical examinations at recruitment and com- Glasgow airport, and calculated 3-, 7-, and 1994), the BS effect parameter, representing pleted similar detailed questionnaires. Baseline 31-day temperature averages as for BS. We the ratio between the odds of being a case ver- variables used in our analyses included marital modeled temperature effects using a bilinear sus a control for a given increment in exposure, status (married, single, widowed, or other), model which assumed a “knot” at 11°C with may be interpreted as an approximate estimate smoking status (never; ex-smoker; current differential linear effects of temperature change of the hazard ratio in a proportional hazards smoker of 1–14, 15–24, or > 25 cigarettes/ below and above the knot, based on a previ- survival model where the level of BS exposure day; or pipe/cigar smoker), occupational social ous analy sis of temperature effects on mor- is age dependent. class [categorized as I, II, III non-manual, III tality in the general population of Glasgow Models for long-term pollution effects. manual, IV, or V according to the Registrar (Carder et al. 2005). We did not adjust for Long-term exposures to BS were estimated for General’s Classification (General Register spatial variation in temperature when estimat- 1970–1979 using a multilevel spatio temporal Office 1966)], body mass index, systolic blood ing long-term effects because we assumed that model that used a combination of time-series pressure, and total plasma cholesterol. relatively shallow spatial gradients in long- and spatial smoothing techniques to model Follow-up of mortality using linkage to the term average temperature would not confound monthly BS at 181 monitoring sites (across National Health Service Central Register was air pollution– mortality associations. the central part of Scotland including the available up to April 1998 for both cohorts. Models for short-term pollution effects using Glasgow conurbation) simultaneously taking In addition to deaths from any cause, mortal- centrally estimated exposures. Because there was into account seasonal effects and local air qual - ity from cardio vascular causes (ICD-9 codes only one monitoring site with adequate data ity predictors including altitude (A), household 410–414, 426–429, 434–440) and respira- for the whole study period, a conventional sur- density within a 250-m radius (HD), distance tory causes (ICD-9 codes 480–487, 490–496) vival analy sis model could not be used to study to nearest major road (MR), and distance to an were considered [coded to the International short-term effects as all cohort participants urban boundary (UB) (Beverland et al. 2012): Classification of Diseases, Revision 9 (ICD-9; would have been assigned the same values for yt =fg^^hh + t World Health Organization 1977). short-term pollution exposure at a given time. ij ij iij Pollution exposure. Analysis of short-term We therefore constructed nested case– tt ij ij aakk ++aa CosSin pollution exposure was based on records of control data sets as follows. For each death 11 c s 12 12 daily BS concentration between 1974 and in a cohort, we considered controls selected ++aa A HD +a MR 23ii 4 i 1998 at a single monitoring site close to the randomly from among the cohort partici- center of the Glasgow conurbation, which pants who lived at least as long as the case. ++af UB . 5 iij had few missing values and was situated in Each control had an associated date, namely a residential area with medium- to high- the date when the control reached the exact Here y denotes monthly mean ln(BS + 0.5) ij density housing interspersed with some indus- age at which the case died. If the control date at site i at time t (where i = 1…s indexes sites, ij trial undertakings [UK classification A2/B2 was outside the follow-up period of the cohort, and j = 1…n indexes observations within a (Department for Environment Food and Rural that control could not be used. We restricted site); f (t ) is the BS temporal trend averaged ij Affairs 2005)]. From a review of approxi- controls to persons of the same sex who lived over the population of all sites; g (t ) is the i ij mately 10 potential monitoring sites in the longer than the case, whose control date was in deviation of the ith site from the population conurbation, taking into account prevailing the same calendar month as the case death, and mean at time t [f (t ) and g (t ) were estimated ij ij i ij winds, population distribution, site classifica - whose date of birth was within 1 calendar year using penalized linear splines]; sine and cosine tion and data capture during 1974–1998, this of the date of birth of the case. Up to 9 controls terms model seasonal effects with α , α as 1c 1s | | Environmental Health Perspectives • volume 120 number 9 September 2012 1281 Beverland et al. fixed-effect parameters; and α …α are local geometric mean BS concentrations at these at address postal codes) to the 1970–1979 2 5 air quality predictor fixed-effect parameters. sites ranged from 8.9 to 48.2 μg/m . We then geometric mean BS at the central monitoring The “within-site” error term, ε , representing created a “test data set” from 19 of these sites site (27.9 μg/m ). We applied these individual ij the deviation of predicted mean ln(BS) from selected at random and a “training data set” multiplying factors to the series of daily val- observed ln(BS), is assumed to be normally from the remaining 20 sites together with ues at the central site to estimate individual- distributed. The multilevel model allowed the 142 sites with < 80% data coverage. The specific time-series over the period 1974–1998 estimation of coefficients between BS and model was fitted to the training data set and reflecting both spatial and temporal variation. local air quality predictors in the presence then used to predict BS in the test data set. From these series, we calculated individual of missing data and hence was not depen- This cross-validation procedure was repeated daily values of the 3-, 7-, and 31-day aver- dent on imputation techniques (Beverland 10 times with different random selections ages of BS exposure for the conditional logistic et al. 2012). The model allowed imputation from the 39 complete data sites forming the regressions applied to the same nested case– of missing data at monitoring sites and sub- test set. The average difference (± SE) between control data sets as described above. sequent estimation of long-term average BS predicted and observed concentrations for test Time-series modeling. BS effects were also pollution exposure for individual cohort par- sites was 1.2 ± 1.10 μg/m and the root mean estimated using Poisson regression analy sis of ticipants based on geographical coordinates square difference was 6.8 μg/m . time-series of daily mortality at ≥ 50 years of and the smoothed residual effects of estimated We estimated the effects of long-term age (to provide an approximate match to the random intercepts and temporal trends at BS exposure on mortality up to 1998 using aging cohort during this period) from 1974 monitoring sites around each address loca- Cox proportional hazards models, with base- to 1998 in the general population (approxi- tion (with estimated individual exposures for line hazard functions stratified by 1-year age mately 1 million persons) of the contiguous 1970–1979 ranging from 6.4 to 55.3 μg/m ) groups and sex while incorporating the same Glasgow plus Renfrew–Paisley conurbation. (Yap et al. 2012). individual baseline covariates as in the short- Briefly, the model incorporated smooth func - We evaluated the multilevel model in a term effects models described above (Yap tions of time (natural cubic splines, with seven cross-validation study (Beverland et al. 2012). et al. 2012). degrees of freedom per year, to capture sea- First we identified BS monitoring sites with Models for short-term pollution effects sonal and other long-term effects), indicator > 80% data (39 sites) and imputed missing using individually weighted exposures. We variables for day of the week, and measures data with a site-specific time-series model with calculated the ratios of individual participants’ of BS (at single central monitoring site) and a flexible trend and month and day effects to geometric mean long-term exposures to BS temperature at lags ≤ 30 days grouped into give 39 sites with “complete” data. Ten-year (multilevel model estimates for 1970–1979 6-day periods (Carder et al. 2008). A simple Poisson model was assumed because there was 3 a Table 1. Mean BS concentrations (μg/m ) over preceding 3 days for cases and controls. no evidence of overdispersion. Cohort/mortality/ Results b c year of case death n Cases Controls Cases – controls Table 1 shows the numbers of cases with Renfrew–Paisley known exposure data available by decade, and All-cause 1971–1979 1,046 37.36 37.86 –0.50 compares mean values of 3-day mean BS over 1980–1989 2,823 19.90 18.63 1.27 cases and controls in each of the analy sis data 1990–1998 4,186 11.38 11.02 0.36 sets. In the Renfrew–Paisley area, the mean All 8,055 17.74 17.17 0.57 exposure for cases exceeded that for controls Cardio vascular in all comparisons except for all-cause deaths 1971–1979 446 38.59 38.08 0.51 in the 1970s. This suggests, prima facie, that 1980–1989 1,347 19.25 18.21 1.04 there was an association between short-term 1990–1998 1,850 11.64 11.23 0.41 All 3,643 17.75 17.10 0.65 BS exposure and mortality. We observed no Respiratory clear trend between decades, nor systematic 1971–1979 68 44.51 43.38 1.13 differences by cause of death. There was an 1980–1989 187 22.74 20.35 2.39 overall positive difference between case and 1990–1998 396 11.05 10.60 0.45 control exposure for each cause of death for All 651 17.90 16.83 1.07 the Collaborative cohort, but the differences Collaborative by decade were quite variable. All-cause Table 2 presents the estimated short-term 1971–1979 242 42.51 38.06 4.45 1980–1989 603 18.98 19.05 –0.07 effects for each cohort and mortality outcome, 1990–1998 571 12.65 12.59 0.06 using three averaging periods of exposure, and All 1,416 20.45 19.70 0.75 either using or not using the individual spa- Cardio vascular tial weights. In each case, the pollution effect 1971–1979 120 43.22 40.63 2.59 (percent increase in odds of mortality for a 1980–1989 278 19.11 18.52 0.59 10-μg/m increment in average BS concentra- 1990–1998 226 11.52 12.77 –1.25 tion) was estimated with and without adjust- All 624 21.00 20.69 0.31 Respiratory ing for the risk factors. The table illustrates 1971–1979 7 41.43 48.07 –6.64 how the estimates changed with increasing 1980–1989 50 22.75 21.43 1.32 degrees of adjustment for risk factors, spatial 1990–1998 54 12.87 13.09 –0.22 weighting, and bilinear tempera ture. All 111 19.12 19.05 0.07 Estimated effects on all-cause mortality. Each case–control set was given equal weight (i.e., the mean control exposure was calculated in each set and then Generally positive associations between short- these means were averaged). Number of deaths (cases); about 7% of deaths were excluded because the 3-day mean c term BS exposure and all-cause mortality were of BS was missing. Controls of same sex born within 1 year of case and reaching control date in the same calendar month as case death. estimated for both cohorts. Significant effects | | 1282 volume 120 number 9 September 2012 • Environmental Health Perspectives Short and long-term exposure–mortality associations were found in the Renfrew–Paisley cohort, par- respectively. Exposure–mortality associations BS in the Renfrew–Paisley cohort, but reduced ticularly when the individually weighted expo- were generally smaller and nonsignificant in those of the 31-day mean. Adjusting for tem- sures were used. Adjusting for risk factors and the Collaborative cohort, with some nonsig- perature made a substantial difference, reduc - temperature, the effect of 3-day mean BS was nificant negative associations for central site ing estimates by about 33% for the 3- and estimated as a 1.0% [95% confidence interval exposure over the 31-day period. Apart from 7-day means and considerably more for the (CI): –0.2, 2.3%] increase in hazard using cen- the smaller number of Collaborative cohort 31-day mean at the central site. trally estimated exposures and a 1.8% (95% participants included in our analyses, possi- Estimated effects on cardio vascular mortal- CI: 0.1, 3.5%) increase using individually ble reasons for the different results in the two ity. The pattern of associations with cardio- weighted exposures (Table 2). Corresponding cohorts are that they differed with regard to vascular mortality was similar to that for estimates of the effect of a 31-day mean BS age, social class, employment, geographical all-cause mortality, with higher estimates for were higher: 1.1% (95% CI: –2.3, 4.7%) and distribution, and election. the longer exposure periods and when indi- 3.4% (95% CI: –0.7, 7.7%) for centrally esti- Adjusting for risk factors did not greatly vidually weighted exposures were used. For the mated and individually weighted exposures change effect estimates of 3- and 7-day mean Renfrew–Paisley cohort the 31-day BS mean a 3 Table 2. Estimated percent change in mortality with a 10-μg/m increase in BS based on nested case–control analyses according to exposure time period (BS averaging period), cohort, and model. Days –1 to –3 Days 0 to –6 Days 0 to –30 Percent change Percent change Percent change (95% CI) p-Value (95% CI) p-Value (95% CI) p-Value All-cause Renfrew–Paisley CSBS 1.4 (0.3, 2.6) 0.011 1.7 (0.2, 3.2) 0.028 4.1 (1.2, 7.0) 0.005 CSBS + risk factors 1.5 (0.4, 2.6) 0.009 1.7 (0.2, 3.3) 0.027 3.7 (0.9, 6.7) 0.011 CSBS + risk factors + temperature 1.0 (–0.2, 2.3) 0.100 1.1 (–0.7, 2.8) 0.231 1.1 (–2.3, 4.7) 0.516 IWBS 2.3 (0.7, 3.8) 0.003 2.9 (0.9, 5.0) 0.005 6.3 (2.7, 10.0) 0.001 IWBS + risk factors 2.3 (0.8, 3.9) 0.003 3.0 (0.9, 5.1) 0.004 5.8 (2.2, 9.6) 0.001 IWBS + risk factors + temperature 1.8 (0.1, 3.5) 0.040 2.3 (0.0, 4.6) 0.051 3.4 (–0.7, 7.7) 0.106 Collaborative CSBS 1.3 (–1.0, 3.7) 0.270 1.7 (–1.6, 5.1) 0.305 –2.0 (–8.0, 4.4) 0.531 CSBS + risk factors 0.6 (–1.9, 3.2) 0.636 0.5 (–2.9, 4.1) 0.771 –3.1 (–9.2, 3.4) 0.345 CSBS + risk factors + temperature –0.1 (–2.9, 2.7) 0.921 –0.7 (–4.5, 3.3) 0.736 –5.0 (–12, 2.6) 0.194 IWBS 2.6 (0.3, 4.8) 0.024 3.4 (0.4, 6.5) 0.027 3.8 (–1.1, 8.9) 0.128 IWBS + risk factors 1.6 (–0.8, 4.1) 0.188 1.8 (–1.4, 5.1) 0.276 2.0 (–3.0, 7.2) 0.446 IWBS + risk factors + temperature 1.1 (–1.4, 3.8) 0.390 1.1 (–2.3, 4.7) 0.528 2.0 (–3.4, 7.6) 0.482 Cardio vascular Renfrew–Paisley CSBS 1.8 (0.1, 3.5) 0.042 2.9 (0.6, 5.3) 0.013 7.4 (3.1, 11.9) 0.001 CSBS + risk factors 1.4 (–0.3, 3.1) 0.109 2.6 (0.2, 5.0) 0.031 6.5 (2.1, 11.1) 0.003 CSBS + risk factors + temperature 0.9 (–1.0, 2.8) 0.351 1.5 (–1.1, 4.2) 0.271 2.9 (–2.3, 8.4) 0.276 IWBS 2.4 (0.1, 4.7) 0.043 3.8 (0.7, 7.0) 0.016 8.7 (3.2, 14.5) 0.002 IWBS + risk factors 2.0 (–0.3, 4.4) 0.089 3.5 (0.4, 6.8) 0.027 8.2 (2.6, 14.1) 0.004 IWBS + risk factors + temperature 1.4 (–1.2, 4.0) 0.287 2.2 (–1.3, 5.8) 0.225 4.1 (–2.2, 10.7) 0.208 Collaborative CSBS 0.3 (–3.2, 3.9) 0.875 1.7 (–3.2, 6.9) 0.507 –3.5 (–12, 6.0) 0.454 CSBS + risk factors –1.3 (–4.8, 2.4) 0.484 –0.3 (–5.1, 4.9) 0.919 –3.9 (–13, 6.0) 0.424 CSBS + risk factors + temperature –2.1 (–6.0, 1.9) 0.303 –1.1 (–6.5, 4.7) 0.712 –5.3 (–16, 6.1) 0.349 IWBS 1.5 (–1.9, 4.9) 0.394 3.0 (–1.6, 7.8) 0.202 1.3 (–5.8, 8.8) 0.735 IWBS + risk factors –0.2 (–3.6, 3.4) 0.930 1.0 (–3.6, 5.8) 0.681 0.3 (–7.0, 8.1) 0.944 IWBS + risk factors + temperature –0.6 (–4.3, 3.2) 0.747 0.6 (–4.4, 5.9) 0.820 0.4 (–7.5, 8.9) 0.930 Respiratory Renfrew–Paisley CSBS 2.8 (–1.3, 7.1) 0.181 3.1 (–2.3, 8.9) 0.270 10.1 (–0.6, 21.9) 0.065 CSBS + risk factors 1.7 (–2.6, 6.1) 0.449 1.9 (–3.6, 7.7) 0.503 5.6 (–4.8, 17.2) 0.304 CSBS + risk factors + temperature –0.2 (–5.0, 4.7) 0.920 –0.6 (–6.7, 6.0) 0.857 –0.6 (–12, 12.9) 0.932 IWBS 3.8 (–1.5, 9.4) 0.167 5.1 (–2.0, 12.7) 0.167 19.4 (5.2, 35.5) 0.006 IWBS + risk factors 2.0 (–3.6, 7.8) 0.495 3.0 (–4.2, 10.6) 0.425 12.1 (–1.4, 27.4) 0.082 IWBS + risk factors + temperature –0.4 (–6.4, 6.1) 0.912 0.2 (–7.6, 8.6) 0.964 7.2 (–7.5, 24.2) 0.358 Collaborative CSBS 0.0 (–9.6, 10.7) 0.998 0.7 (–12, 15.0) 0.922 –24 (–44, 3.2) 0.079 CSBS + risk factors 0.9 (–8.9, 11.6) 0.869 0.7 (–12, 14.9) 0.923 –23 (–43, 4.1) 0.089 CSBS + risk factors + temperature –1.2 (–12, 10.4) 0.826 –2.8 (–16, 12.8) 0.704 –29 (–50, 0.8) 0.055 IWBS 2.6 (–5.8, 11.8) 0.556 4.1 (–6.6, 16.1) 0.463 –16 (–34, 6.5) 0.151 IWBS + risk factors 2.4 (–6.1, 11.7) 0.589 2.5 (–8.2, 14.5) 0.656 –17 (–34, 4.6) 0.114 IWBS + risk factors + temperature 1.1 (–7.8, 10.9) 0.817 0.7 (–11, 13.4) 0.914 –19 (–38, 4.0) 0.097 Abbreviations: CI, confidence interval; CSBS, BS at (single) central monitoring site; IWBS, individually weighted BS from ratio of individual participant's long-term geometric mean exposure to BS, estimated by multilevel model, to geometric mean BS at the central monitoring site (time-series of daily values at central site were multiplied by individual ratios to produce individual-specific time-series for 1974–1998 reflecting both spatial and temporal variation); risk factors, adjustment for potential confounding variables measured at baseline—smoking history, social class, body mass index, marital status, systolic blood pressure, and total cholesterol; temperature, linear effects of temperature > and < 11°C. Estimates in this table are odds ratios (ORs) from conditional logistic regression models comparing the pollution experienced by the case immediately before death and that experienced by controls of the same age in approximately the same time period and at approximately the same time of year. | | Environmental Health Perspectives • volume 120 number 9 September 2012 1283 Beverland et al. was associated with larger mortality increases Table 3 also displays estimates of short- medium-term exposure on mortality within than the 3- and 7-day means [after adjusting term associations derived from time-series cohort participants directly to estimated effects for risk factors, temperature and spatial vari- modeling of the daily variation in mortality for of long-term exposure on mortality within the ation, a 10-μg/m increment in average BS the entire population of the Glasgow conurba- same cohorts (Yap et al. 2012). We also com- exposure over the previous 31 days was asso- tion that was > 50 years of age (Beverland pared the estimated effects of short- and medi - ciated with a 4.1% (95% CI: –2.2, 10.7%) et al. 2007). The time-series estimates were um-term exposure with rate–ratio estimates increase in cardiovascular mortality for this smaller than the corresponding short-term obtained from Poisson regression analysis of cohort]. In the Collaborative cohort there were estimates in the two cohorts, but they had nar- daily mortality within the general population no statistically significant associations and the rower confidence intervals, being based on a from which the majority of cohort participants estimated effect sizes were small and, in some much larger population. were selected (Beverland et al. 2007). instances, negative. For the Renfrew–Paisley cohort, larger Discussion Estimated effects on respiratory mortality. exposure–mortality associations were noted in There were very few statistically significant There have been a limited number of studies the nested case–control analyses than in the associations and estimates generally had very comparing short- and long-term effects of air time-series analyses using a single monitoring wide confidence intervals, making any infer - pollution in related study populations [e.g., site (Table 3); the difference being consistent ences extremely tentative. In the Collaborative Klemm and Mason (2003), Laden et al. for both all-cause and cardio vascular mortality. cohort, the 31-day exposures were associated (2006), and Schwartz et al. (1996) as reviewed This may be a result of more realistic expo- with strongly negative, although nonsignifi - by Pope (2007)]. However, to the best of our sure classification in the cohort-based analyses cant, effects; whereas in the Renfrew–Paisley knowledge, direct estimation of the short-term (Armstrong 1998; Sheppard et al. 2012). In cohort, the 31-day exposures were associated effects of air pollution within a cohort study both nested case–control and time-series analy- with large positive, but generally nonsig- has rarely been attempted. In our case–control ses, the pollution– mortality associations were nificant, effects. For 3- and 7-day mean BS, analysis, pollution on the days preceding the larger for medium-term (31-day) compared to the estimates were generally positive before death of a cohort participant was compared short-term (3-day) averaging/lag periods. adjusting for temperature. with the pollution preceding the days when However, the short- and medium-term Comparison with long-term effect the controls reached the age at which the case effect estimates from the nested case– control estimates. Table 3 compares selected estimates died. We matched for sex, month of death, analysis of the Renfrew–Paisley cohort were of the short and medium-term effects of BS on and within 1 year of birth and adjusted for substantially smaller than the equivalent effects mortality from Table 2 with estimated long- several individual-level risk factors strongly estimated for long-term exposure (Beverland term effects in the same cohorts, obtained associated with mortality. We also adjusted et al. 2007; Yap et al. 2012), which is consis- from proportional hazards modeling (Yap et al. exposure estimates for spatial variation in per- tent with a review of exposure–outcome asso- 2012). In the Renfrew–Paisley cohort long- sonal exposure, and adjusted for temperature. ciations at different time scales (Pope 2007). term pollution exposure–mortality associations This methodology might be expected to pro - For example, Yap et al. (2012) observed that were substantially greater in magnitude than duce estimates of short-term effects that are a 10-μg/m increase in average BS over the short-term exposure–mortality associations more realistic than those available from time- decade 1970–1979 was associated with a 10% for all three categories of mortality. Long-term series modeling and more comparable with (95% CI: 4, 17%) increase in the hazard of effects were much smaller in the Collaborative the estimates available from survival analy- all-cause mortality as compared with our esti- cohort than in the Renfrew–Paisley cohort and sis of long-term follow-up in cohorts. We mate of a 1.8% (95% CI: 0.1, 3.4%) increase were nonsignificant. compared the estimated effects of short- and in hazard for a 10-μg/m increase in average BS over 3 days. Estimated associations for the Table 3. Comparison of estimated magnitudes of associations [percent change (95% CI)] between short- Collaborative cohort were generally positive and long-term exposure to BS and mortality in the Renfrew–Paisley and Collaborative cohorts and in the but much less consistent in magnitude, with no population > 50 years of age of Glasgow, Renfrew, and Paisley conurbation with follow-up to 1998. significant pollution effects observed in separate a,b a,b c Mortality/population group Short-term (3-day) Medium-term (31-day) Long-term (1970–1979) analyses of this cohort. All-cause Our analyses of the effects of long-term Time-series 0.2 (0.0, 0.4) 0.9 (0.3, 1.5) — and individually weighted short-term BS expo- Renfrew–Paisley cohort 1.8 (0.1, 3.5) 3.4 (–0.7, 7.7) 10 (4, 17) sure assumed that the pattern of intraurban b,d Collaborative cohort 1.1 (–1.4, 3.8) 2.0 (–3.4, 7.6) 1 (–4, 6) spatial variations in long-term average BS Combined cohort 1.6 (0.2, 3.0) 2.9 (–0.5, 6.2) 5 (1, 9) exposure in 1970–1979 was largely sustained Cardio vascular Time-series 0.1 (–0.2, 0.4) 0.3 (–0.7, 1.2) — over the subsequent 1980–1998 period, which Renfrew–Paisley cohort 1.4 (–1.2, 4.0) 4.1 (–2.2, 10.7) 11 (1, 22) could not be verified because of the substan - b,d Collaborative cohort –0.6 (–4.3, 3.2) 0.4 (–7.5, 8.9) 3 (–5, 12) tial reduction in the BS monitoring network Combined cohort 0.8 (–1.4, 2.9) 2.7 (–2.4, 7.8) 7 (0, 13) during this later period. Similar assumptions Respiratory about relative invariance of spatial contrasts Time-series 0.3 (–0.2, 0.8) 3.1 (1.4, 4.9) — b in long-term air pollution exposure have been Renfrew–Paisley cohort –0.4 (–6.4, 6.1) 7.2 (–7.5, 24.2) 26 (2, 55) b,d made, by necessity, in almost all epidemiologi- Collaborative cohort 1.1 (–7.8, 10.9) –19.5 (–37.7, 4.0) –3 (–21, 18) cal studies of long-term intra urban air pollu- Combined cohort 0.1 (–5.1, 5.3) –2.6 (–15.2, 10.0) 11 (–3, 28) tion exposure effects. These assumptions have Table details percent increases in mortality associated with 10-μg/m increments in average BS. Rate ratios estimated by Poisson regression modeling (adjusted for temperature) for population > 50 years of age in the been partly supported by observations of “sta- contiguous Glasgow, Renfrew, and Paisley conurbation (Beverland et al. 2007). Odds ratios estimated by conditional bility” in spatial contrasts among measurement logistic regression modeling (adjusted for temperature) on matched case–control sets and adjusted for baseline risk sites in a small number of studies (Eeftens factors (smoking history, social class, body mass index, marital status, systolic blood pressure, and total cholesterol) and spatial variation (Table 2). Hazard ratios estimated by Cox regression modeling using long-term exposures esti- et al. 2011; Hoek et al. 2008). An additional mated from the spatio temporal model and adjusted for baseline risk factors listed in footnote b. Short-term effects were limitation is that exposure misclassification estimated for the Glasgow conurbation subset of the Collaborative cohort (n = 3,818); long-term effects were estimated e may have resulted from a lack of information for the full Collaborative cohort (n = 6,680). Combined cohort risk estimates computed from individual cohort estimates about participant mobility. (as outlined in above footnotes) using weights proportional to the inverse variance of risk estimates in individual cohorts. | | 1284 volume 120 number 9 September 2012 • Environmental Health Perspectives Short and long-term exposure–mortality associations Samoli et al. (2001) used time-series of air pollution, e.g., fine or ultrafine particles General Register Office. 1966. Classification of Occupations 1966. London:HMSO. methods to estimate an average effect on all- or specific transition metals (Heal et al. 2005; Goodman PG, Dockery DW, Clancy L. 2004. Cause-specific cause mortality of 3.1% (95% CI: 2.4, 3.9%) Hochadel et al. 2006; Janssen et al. 2011). mortality and the extended effects of particulate pollu- for a 50-μg/m increase in BS averaged over tion and temperature exposure. Environ Health Perspect Conclusions 112:179–185. the day of death and the previous day in four Hart CL, MacKinnon PL, Watt GC, Upton MN, McConnachie A, western European cities. A study using case– After adjusting for individual-level risk fac- Hole DJ, et al. 2005. The Midspan studies. Int J Epidemiol crossover analysis (Zeka et al. 2005), which tors, temperature, and geographical variation 34(1):28–34. Hawthorne VM, Watt GCM, Hart CL, Hole DJ, Smith GD, Gillis CR. has similarities to our approach, was applied in BS pollution, short and medium-term BS 1995. Cardiorespiratory disease in men and women in urban to PM data for 20 U.S. cities between 1989 exposure–mortality associations estimated Scotland: Baseline characteristics of the Renfrew/Paisley and 2000 and found that a 10-μg/m increase using cohort-based nested case–control analy- (Midspan) study population. Scott Med J 40:102–107. Heal MR, Hibbs LR, Agius RM, Beverland LJ. 2005. Total and in PM averaged over the day of death ses were of greater magnitude than associa- water-soluble trace metal content of urban background and the previous 2 days was associated with tions estimated for similar geographical areas PM , PM and black smoke in Edinburgh, UK. Atmos 10 2.5 0.45% (95% CI: 0.25, 0.65%), 0.50% (95% using single-pollution-site time-series analyses. Environ 39(8):1417–1430. CI: 0.25, 0.75%) and 0.87% (95% CI: However, short and medium-term exposure– Hochadel M, Heinrich J, Gehring U, Morgenstern V, Kuhlbusch T, Link E, et al. 2006. Predicting long-term average concentra- 0.38, 1.36%) increases in all-cause, cardio- mortality associations were of substantially tions of traffic-related air pollutants using GIS-based infor - vascular, and respiratory mortality, respec- lower magnitude than long-term exposure- mation. Atmos Environ 40(3):542–553. tively. Using data from Dublin for 1980–1996, mortality associations observed in the same Hoek G, Beelen R, de Hoogh K, Vienneau D, Gulliver J, Fischer P, et al. 2008. A review of land-use regression models to Goodman et al. (2004) estimated the effects of cohorts using survival analysis. These observa - assess spatial variation of outdoor air pollution. Atmos a 10-μg/m increase in 3-day average BS as tions indicate the importance of intraurban Environ 42(33):7561–7578. 0.4% (95% CI: 0.3, 0.6%), 0.4% (95% CI: variations in long-term pollution climates Janssen NAH, Hoek G, Simic-Lawson M, Fischer P, van Bree L, ten Brink H, et al. 2011. Black carbon as an additional indi- 0.2, 0.7%), and 0.9% (95% CI: 0.5, 1.2%) when estimating associations between expo- cator of the adverse health effects of airborne particles on all-cause, cardiovascular, and respiratory sure and mortality and suggest that public compared with PM and PM . Environ Health Perspect 10 2.5 mortality, respectively. Although none of these health impacts of air pollution may be domi- 119:1691–1699. Klemm RJ, Mason R. 2003. Replication of reanalysis of studies used adjustments for spatial variation nated by long-term exposure determined by Harvard Six-City Mortality study. In: Revised Analyses of that were comparable with our approach, and geographical differences in pollution climates. Time-Series of Air Pollution and Health. Special Report. they involved slightly different lag periods, the Boston:Health Effects Institute, 165–172. Available: http:// pubs.healtheffects.org/getfile.php?u=21 [accessed 8 May order of magnitude of the short-term effects Refe Rences 2012]. is not inconsistent with ours, especially given Künzli N. 2005. Unifying susceptibility, exposure, and time: Armstrong BG. 1998. Effect of measurement error on epide- the relatively wide confidence intervals associ - Discussion of unifying analytic approaches and future miological studies of environmental and occupational ated with our estimates. However, while the directions. J Toxicol Environ Health A 68(13–14):1263–1271. exposures. Occup Environ Med 55(10):651–656. Laden F, Schwartz J, Speizer FE, Dockery DW. 2006. Reduction use of spatially weighted exposures increased Beverland IJ, Robertson C, Yap C, Heal MR, Cohen GR, in fine particulate air pollution and mortality—extended Henderson DEJ, et al. 2012. Comparison of models for our effect estimates, even when using centrally follow-up of the Harvard Six Cities study. Am J Respir Crit estimation of long-term exposure to air pollution in cohort measured exposures our case–control estimates Care 173(6):667–672. studies. Atmos Environ; doi:10.1016/j.atmosenv.2012.08.001. Pope CA III. 2007. Mortality effects of longer term exposures to for 3-day BS were generally higher than those Beverland IJ, Yap C, Robertson C, Agius RM, Hole DJ, Cohen GR, fine particulate air pollution: Review of recent epidemio - et al. 2007. Department of Health research project (Ref: observed in general population time-series and logical evidence. Inhal Toxicol 19:33–38. 0020015). Health Effects of Long-term Exposure to Air case–crossover analyses. Pope CA III, Dockery DW. 2006. Health effects of fine particu- Pollution in Scotland. Available: http://www.dh.gov.uk/prod_ late air pollution: Lines that connect. J Air Waste Manag Another European multicity study using consum_dh/groups/dh_digitalassets/@dh/@en/documents/ Assoc 56(6):709–742. digitalasset/dh_092831.pdf [accessed 8 May 2012]. time-series methods reported city-specific all- Samoli E, Schwartz J, Wojtyniak B, Touloumi G, Spix C, Brook RD, Rajagopalan S, Pope CA III, Brook JR, Bhatnagar A, cause mortality effects of a 10-μg/m increase Balducci F, et al. 2001. Investigating regional differences Diez-Roux AV, et al. 2010. Particulate matter air pollution in short-term effects of air pollution on daily mortality in in PM over a 40-day period ranging from 10 and cardiovascular disease an update to the scientific the APHEA project: a sensitivity analy sis for controlling statement from the American Heart Association. Circulation –0.9% to 4.0% (Zanobetti et al. 2002). In our long-term trends and seasonality. Environ Health Perspect 121(21):2331–2378. case–control analy sis, we found associations of 109:349–353. Carder M, McNamee R, Beverland I, Elton R, Cohen GR, Boyd J, Schwartz J, Dockery DW, Neas LM. 1996. Is daily mortality all-cause mortality with 10-μg/m increases in et al. 2005. The lagged effect of cold temperature and wind associated specifically with fine particles? J Air Waste average BS over the day of death and the pre- chill on cardiorespiratory mortality in Scotland. Occup Manag Assoc 46(10):927–939. Environ Med 62(10):702–710. ceding 30 days of 3.4% (95% CI: –0.7, 7.7%) Sheppard L, Burnett R, Szpiro A, Kim S-Y, Jerrett M, Pope C III, Carder M, McNamee R, Beverland I, Elton R, Van Tongeren M, (Renfrew–Paisley after full adjustments) and et al. 2012. Confounding and exposure measurement Cohen GR, et al. 2008. Interacting effects of particulate error in air pollution epidemiology. Air Qual Atmos Health 2.0% (95% CI: –3.4, 7.6%) (Collaborative), pollution and cold temperature on cardiorespiratory mor- 5(2):203–216. tality in Scotland. Occup Environ Med 65(3):197–204. which are within this range and slightly greater World Health Organization. 1977. Manual of the International Carracedo-Martinez E, Taracido M, Tobias A, Saez M, Figueiras A. than the equivalent effect of 1.1% (95% CI: Statistical Classification of Diseases, Injuries and Causes of 2010. Case–crossover analysis of air pollution health effects: Death. Ninth Revision. Geneva:World Health Organization. 0.8, 1.3%) noted over a 40-day period in time- a systematic review of methodology and application. Environ Yap C, Beverland IJ, Robertson C, Heal MR, Cohen GR, Health Perspect 118:1173–1182. series analyses of BS and all-cause mortality in Henderson DEJ, et al. 2012. Association between long- Collett D. 1994. Modelling Survival Data in Medical Research. Dublin (Goodman et al. 2004). term exposure to air pollution and specific causes of London:Chapman and Hall. mortality in Scotland. Occup Environ Med; doi:101136/ The multicity studies cited above empha- Davey Smith G, Hart CL, Hole DJ, MacKinnon PL, Gillis CR, oemed-2011-100600. Watt GCM, et al. 1998. Education and occupational social size marked unexplained heterogeneity between Zanobetti A, Schwartz J, Samoli E, Gryparis A, Touloumi G, class: Which is the more important indicator of mortality city-specific estimates, so it is conceivable that Atkinson R, et al. 2002. The temporal pattern of mortality risk? J Epidemiol Community Health 52:153–160. responses to air pollution: A multicity assessment of mor- there are specific conditions in the Glasgow Department for Environment Food and Rural Affairs. 2005. tality displacement. Epidemiology 13(1):87–93. National Air Quality Data Archive. Available: http://uk-air. conurbation that might explain differences Zeka A, Zanobetti A, Schwartz J. 2005. Short term effects of defra.gov.uk/data/ [accessed 1 October 2005]. between our results and those of others. It is particulate matter on cause specific mortality: Effects Eeftens M, Beelen R, Fischer P, Brunekreef B, Meliefste K, Hoek G. of lags and modification by city characteristics. Occup also possible that BS gives a better measure 2011. Stability of measured and modelled spatial contrasts in Environ Med 62(10):718–725. than PM for the more damaging elements NO over time. Occup Environ Med 68(10):765–770. | | Environmental Health Perspectives • volume 120 number 9 September 2012 1285 http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Environmental Health Perspectives Unpaywall

A Comparison of Short-term and Long-term Air Pollution Exposure Associations with Mortality in Two Cohorts in Scotland

Download PDF
6 pages

Loading next page...
 
/lp/unpaywall/a-comparison-of-short-term-and-long-term-air-pollution-exposure-hYg9fhy0Do

References

References for this paper are not available at this time. We will be adding them shortly, thank you for your patience.

Publisher
Unpaywall
ISSN
0091-6765
DOI
10.1289/ehp.1104509
Publisher site
See Article on Publisher Site

Abstract

Research A Comparison of Short-term and Long-term Air Pollution Exposure Associations with Mortality in Two Cohorts in Scotland 1 2 3 4 1* 5,6,7 Iain J. Beverland, Geoffrey R. Cohen, Mathew R. Heal, Melanie Carder, Christina Yap, Chris Robertson, 8 4 Carole L. Hart, and Raymond M. Agius 1 2 3 Department of Civil Engineering, University of Strathclyde, Glasgow, United Kingdom; Edinburgh, United Kingdom; School of Chemistry, University of Edinburgh, Edinburgh, United Kingdom; Centre for Occupational and Environmental Health, University of Manchester, Manchester, United Kingdom; Department of Mathematics and Statistics, University of Strathclyde, Glasgow, United 6 7 8 Kingdom; Health Protection Scotland, Glasgow, United Kingdom; International Prevention Research Institute, Lyon, France; Institute of Health and Wellbeing, Public Health, University of Glasgow, Glasgow, United Kingdom cumulative effects that increase the sensitivity Background : Air pollution– mortality risk estimates are generally larger at longer-term, compared of highly exposed population subgroups. with short-term, exposure time scales. However, there have been few studies oBjective : We compared associations between short-term exposure to black smoke (BS) and mor- comparing the estimated effects of the same tality with long-term exposure–mortality associations in cohort participants and with short-term pollutant on the same population at both exposure–mortality associations in the general population from which the cohorts were selected. short- and long-term time scales. In the pres- M e t h o d s : We assessed short-to-medium–term exposure–mortality associations in the Renfrew– ent study, we present such a comparison, using Paisley and Collaborative cohorts (using nested case–control data sets), and compared them with mortality data from approximately 25 years of long-term exposure–mortality associations (using a multilevel spatiotemporal exposure model and follow-up on two Scottish cohorts with a com- survival analyses) and short-to-medium–term exposure–mortality associations in the general popu- bined size of nearly 20,000 participants and lation (using time-series analyses). estimates of temporal and spatial variations results : For the Renfrew–Paisley cohort (15,331 participants), BS exposure–mortality associa- in black smoke (BS) air pollution in the con- tions were observed in nested case–control analyses that accounted for spatial variations in pollution tiguous urban area of Glasgow, Paisley, and exposure and individual-level risk factors. e Th se cohort-based associations were consistently greater Renfrew in central Scotland, United Kingdom than associations estimated in time-series analyses using a single monitoring site to represent gen- (the Glasgow conurbation). Short-term effects eral population exposure {e.g., 1.8% [95% confidence interval (CI): 0.1, 3.4%] vs. 0.2% (95% CI: were estimated using a nested case–control 0.0, 0.4%) increases in mortality associated with 10-μg/m increases in 3-day lag BS, respectively}. approach in which the pollution experienced Exposure–mortality associations were of larger magnitude for longer exposure periods [e.g., 3.4% (95% CI: –0.7, 7.7%) and 0.9% (95% CI: 0.3, 1.5%) increases in all-cause mortality associated by each case immediately before death was with 10-μg/m increases in 31-day BS in case–control and time-series analyses, respectively; and compared with that experienced by controls 10% (95% CI: 4, 17%) increase in all-cause mortality associated with a 10-μg/m increase in geo- of the same age in approximately the same metic mean BS for 1970–1979, in survival analysis]. time period and at approximately the same c onclusions : After adjusting for individual-level exposure and potential confounders, short-term time of year. The effects of long-term pollu - exposure–mortality associations in cohort participants were of greater magnitude than in compara- tion exposure were estimated from a detailed ble general population time-series study analyses. However, short-term exposure–mortality associa- spatio temporal exposure model. Additionally, tions were substantially lower than equivalent long-term associations, which is consistent with the short-term exposure–mortality associations possibility of larger, more persistent cumulative effects from long-term exposures. were estimated using conventional Poisson k ey words : air, associations, cohort, exposure–mortality, long term, pollution, short term, time- regression time-series models in the urban series. Environ Health Perspect 120:1280–1285 (2012). http://dx.doi.org/10.1289/ehp.1104509 population from which the majority of cohort [Online 6 June 2012] participants were selected. Methods The magnitude of adverse health effects effects have examined relationships between associated with air pollution has been the daily counts of deaths in a given popula- Study population. The Renfrew–Paisley cohort subject of extensive research (Brook et al. tion and daily pollutant and meteoro logical was recruited from residents 45–64 years of 2010; Pope and Dockery 2006). An aspect varia bles recorded at central monitoring sites Address correspondence to I. Beverland, University of epidemiological research of importance to using Poisson regression modeling and gen- of Strathclyde, Department of Civil Engineering, policy makers is the time scale over which eralizations thereof. This type of time-series John Anderson Building, 107 Rottenrow, Glasgow adverse effects are most apparent (Künzli study has limited capacity to take account G4 0NG UK. Telephone: 0141 548 3202. Fax: 0141 2005). Discrepancies between the magnitudes of spatial variation in individual exposures 553 2066. E-mail: [email protected] of short-to-medium–term exposure effects or to investigate the possible modification *Current address: MRC Hub for Trials Methodology estimated at time scales of days to weeks, and of pollution effects by individual risk factors Research, University of Birmingham, Birmingham, UK. We are grateful for encouragement and support from of long-term exposure effects estimated by such as age, smoking history, and previous the late D. Hole, former professor of Epidemiology following cohorts over years to decades, has disease, although analysis using case-crossover and Biostatistics and head of West of Scotland Cancer been a particular concern (Pope 2007). methods (Carracedo-Martinez et al. 2010) Surveillance Unit, for his enthusiastic role in planning In longitudinal cohort studies, long-term may reduce some of these difficulties. The and implementing this research. average pollution exposures of individual estimates of air pollution effects from long- We gratefully acknowledge funding from the cohort participants are estimated, and the term exposure studies involving cohorts have Department of Health (England) Policy Research Programme as part of the Initiative on Air Pollution contribution of the geographical variation generally been considerably larger than those (grant 0020015). among these exposures to the risk of adverse from short-term time-series studies even The views expressed here are those of the authors and events over the entire follow-up period is after adjusting for other individual risk fac- not necessarily the Department of Health (England). determined by survival analysis, adjusting for tors (Pope 2007). It is plausible that larger e Th authors declare they have no actual or potential characteristics that may affect disease risk. In magnitudes of association between long-term competing financial interests. contrast, many studies of short-term exposure exposure and mortality may be attributable to Received 18 September 2011; accepted 6 June 2012. | | 1280 volume 120 number 9 September 2012 • Environmental Health Perspectives Short and long-term exposure–mortality associations age from Renfrew and Paisley in west cen- was considered the most appropriate site to use were randomly selected from the full set avail- tral Scotland, which are contiguous to each to estimate temporal variations in background able for each case. For all-cause deaths, 96% other and to the city of Glasgow. The cohort air pollution. Three averaging periods were (8,342/8,700) of cases in the Renfrew–Paisley comprised 78% of the target population, with used for assessing short- and medium-term cohort and 76% (1,160/1,524) of cases in the 15,402 participants screened between 1972 exposure: a 3-day average over the 3 days pre- Collaborative cohort had the full complement and 1976 (Hawthorne et al. 1995). The ceding the day of death; a 7-day average over of 9 controls and 99% and 92%, respectively, Collaborative cohort is an occupational cohort the day of death and the preceding 6 days; and had at least 4 controls. Six nested case–control of 7,028 participants recruited from 27 work- a 31-day average over the day of death and the data sets of this type were created for three case places in central Scotland between 1970 and preceding 30 days. outcomes (all-cause, cardiovascular, and respi- 1973 (Davey Smith et al. 1998). These two Analysi s of long-term pollution expo- ratory mortality) in the two cohorts. We then cohorts were designed for study together, sure for the decade 1970–1979 was based on used conditional logistic regression to compare and in this context are referred to as the records from 181 monitoring sites distributed the pollution experienced by the case immedi- Midspan study (Hart et al. 2005). In the pres- widely over central Scotland. Many of these ately before death and that experienced by con- ent study, we examined short- and medium- sites had long periods of missing data. After trols of the same age in approximately the same term exposure–mortality associations in 1979, the number of sites was greatly reduced. time period and at approximately the same 15,331 participants in the Renfrew–Paisley The use of the BS metric (derived from time of year, with adjustment for potential con- cohort with complete postal code information the darkness of particulate matter collected on founding variables measured at baseline: smok- and 3,818 Collaborative cohort participants a filter through which air has been sampled) ing history (six levels), social class (six levels), (35–64 years of age at recruitment) residing in is well established in historical UK monitor- body mass index (five quintiles), marital status the Glasgow urban area with complete postal ing networks and is considered to be a good (four levels), systolic blood pressure (linear), code information. To provide an indication marker for traffic and other combustion- and total cholesterol (linear). This conditional of geographical scale the contiguous conurba- related urban air pollution through similar logistic regression model does not correspond tion of Glasgow, Paisley, and Renfrew can be measurements of filter reflectance (Hochadel to a survival model because generally different encompassed within a radius of 12 km, with et al. 2006; Janssen et al. 2011). controls are in each case–control set. However, Renfrew and Paisley encompassed by radii of Temperature data. We obtained daily aver- because of the correspondence between the 1.5 and 3.5 km, respectively, within this 12-km ages of hourly temperatures from 0700–2300 likelihoods for proportional hazards and con- radius. Participants in both cohorts underwent hours at the UK meteorological office site at ditional logistic regression modeling (Collett physical examinations at recruitment and com- Glasgow airport, and calculated 3-, 7-, and 1994), the BS effect parameter, representing pleted similar detailed questionnaires. Baseline 31-day temperature averages as for BS. We the ratio between the odds of being a case ver- variables used in our analyses included marital modeled temperature effects using a bilinear sus a control for a given increment in exposure, status (married, single, widowed, or other), model which assumed a “knot” at 11°C with may be interpreted as an approximate estimate smoking status (never; ex-smoker; current differential linear effects of temperature change of the hazard ratio in a proportional hazards smoker of 1–14, 15–24, or > 25 cigarettes/ below and above the knot, based on a previ- survival model where the level of BS exposure day; or pipe/cigar smoker), occupational social ous analy sis of temperature effects on mor- is age dependent. class [categorized as I, II, III non-manual, III tality in the general population of Glasgow Models for long-term pollution effects. manual, IV, or V according to the Registrar (Carder et al. 2005). We did not adjust for Long-term exposures to BS were estimated for General’s Classification (General Register spatial variation in temperature when estimat- 1970–1979 using a multilevel spatio temporal Office 1966)], body mass index, systolic blood ing long-term effects because we assumed that model that used a combination of time-series pressure, and total plasma cholesterol. relatively shallow spatial gradients in long- and spatial smoothing techniques to model Follow-up of mortality using linkage to the term average temperature would not confound monthly BS at 181 monitoring sites (across National Health Service Central Register was air pollution– mortality associations. the central part of Scotland including the available up to April 1998 for both cohorts. Models for short-term pollution effects using Glasgow conurbation) simultaneously taking In addition to deaths from any cause, mortal- centrally estimated exposures. Because there was into account seasonal effects and local air qual - ity from cardio vascular causes (ICD-9 codes only one monitoring site with adequate data ity predictors including altitude (A), household 410–414, 426–429, 434–440) and respira- for the whole study period, a conventional sur- density within a 250-m radius (HD), distance tory causes (ICD-9 codes 480–487, 490–496) vival analy sis model could not be used to study to nearest major road (MR), and distance to an were considered [coded to the International short-term effects as all cohort participants urban boundary (UB) (Beverland et al. 2012): Classification of Diseases, Revision 9 (ICD-9; would have been assigned the same values for yt =fg^^hh + t World Health Organization 1977). short-term pollution exposure at a given time. ij ij iij Pollution exposure. Analysis of short-term We therefore constructed nested case– tt ij ij aakk ++aa CosSin pollution exposure was based on records of control data sets as follows. For each death 11 c s 12 12 daily BS concentration between 1974 and in a cohort, we considered controls selected ++aa A HD +a MR 23ii 4 i 1998 at a single monitoring site close to the randomly from among the cohort partici- center of the Glasgow conurbation, which pants who lived at least as long as the case. ++af UB . 5 iij had few missing values and was situated in Each control had an associated date, namely a residential area with medium- to high- the date when the control reached the exact Here y denotes monthly mean ln(BS + 0.5) ij density housing interspersed with some indus- age at which the case died. If the control date at site i at time t (where i = 1…s indexes sites, ij trial undertakings [UK classification A2/B2 was outside the follow-up period of the cohort, and j = 1…n indexes observations within a (Department for Environment Food and Rural that control could not be used. We restricted site); f (t ) is the BS temporal trend averaged ij Affairs 2005)]. From a review of approxi- controls to persons of the same sex who lived over the population of all sites; g (t ) is the i ij mately 10 potential monitoring sites in the longer than the case, whose control date was in deviation of the ith site from the population conurbation, taking into account prevailing the same calendar month as the case death, and mean at time t [f (t ) and g (t ) were estimated ij ij i ij winds, population distribution, site classifica - whose date of birth was within 1 calendar year using penalized linear splines]; sine and cosine tion and data capture during 1974–1998, this of the date of birth of the case. Up to 9 controls terms model seasonal effects with α , α as 1c 1s | | Environmental Health Perspectives • volume 120 number 9 September 2012 1281 Beverland et al. fixed-effect parameters; and α …α are local geometric mean BS concentrations at these at address postal codes) to the 1970–1979 2 5 air quality predictor fixed-effect parameters. sites ranged from 8.9 to 48.2 μg/m . We then geometric mean BS at the central monitoring The “within-site” error term, ε , representing created a “test data set” from 19 of these sites site (27.9 μg/m ). We applied these individual ij the deviation of predicted mean ln(BS) from selected at random and a “training data set” multiplying factors to the series of daily val- observed ln(BS), is assumed to be normally from the remaining 20 sites together with ues at the central site to estimate individual- distributed. The multilevel model allowed the 142 sites with < 80% data coverage. The specific time-series over the period 1974–1998 estimation of coefficients between BS and model was fitted to the training data set and reflecting both spatial and temporal variation. local air quality predictors in the presence then used to predict BS in the test data set. From these series, we calculated individual of missing data and hence was not depen- This cross-validation procedure was repeated daily values of the 3-, 7-, and 31-day aver- dent on imputation techniques (Beverland 10 times with different random selections ages of BS exposure for the conditional logistic et al. 2012). The model allowed imputation from the 39 complete data sites forming the regressions applied to the same nested case– of missing data at monitoring sites and sub- test set. The average difference (± SE) between control data sets as described above. sequent estimation of long-term average BS predicted and observed concentrations for test Time-series modeling. BS effects were also pollution exposure for individual cohort par- sites was 1.2 ± 1.10 μg/m and the root mean estimated using Poisson regression analy sis of ticipants based on geographical coordinates square difference was 6.8 μg/m . time-series of daily mortality at ≥ 50 years of and the smoothed residual effects of estimated We estimated the effects of long-term age (to provide an approximate match to the random intercepts and temporal trends at BS exposure on mortality up to 1998 using aging cohort during this period) from 1974 monitoring sites around each address loca- Cox proportional hazards models, with base- to 1998 in the general population (approxi- tion (with estimated individual exposures for line hazard functions stratified by 1-year age mately 1 million persons) of the contiguous 1970–1979 ranging from 6.4 to 55.3 μg/m ) groups and sex while incorporating the same Glasgow plus Renfrew–Paisley conurbation. (Yap et al. 2012). individual baseline covariates as in the short- Briefly, the model incorporated smooth func - We evaluated the multilevel model in a term effects models described above (Yap tions of time (natural cubic splines, with seven cross-validation study (Beverland et al. 2012). et al. 2012). degrees of freedom per year, to capture sea- First we identified BS monitoring sites with Models for short-term pollution effects sonal and other long-term effects), indicator > 80% data (39 sites) and imputed missing using individually weighted exposures. We variables for day of the week, and measures data with a site-specific time-series model with calculated the ratios of individual participants’ of BS (at single central monitoring site) and a flexible trend and month and day effects to geometric mean long-term exposures to BS temperature at lags ≤ 30 days grouped into give 39 sites with “complete” data. Ten-year (multilevel model estimates for 1970–1979 6-day periods (Carder et al. 2008). A simple Poisson model was assumed because there was 3 a Table 1. Mean BS concentrations (μg/m ) over preceding 3 days for cases and controls. no evidence of overdispersion. Cohort/mortality/ Results b c year of case death n Cases Controls Cases – controls Table 1 shows the numbers of cases with Renfrew–Paisley known exposure data available by decade, and All-cause 1971–1979 1,046 37.36 37.86 –0.50 compares mean values of 3-day mean BS over 1980–1989 2,823 19.90 18.63 1.27 cases and controls in each of the analy sis data 1990–1998 4,186 11.38 11.02 0.36 sets. In the Renfrew–Paisley area, the mean All 8,055 17.74 17.17 0.57 exposure for cases exceeded that for controls Cardio vascular in all comparisons except for all-cause deaths 1971–1979 446 38.59 38.08 0.51 in the 1970s. This suggests, prima facie, that 1980–1989 1,347 19.25 18.21 1.04 there was an association between short-term 1990–1998 1,850 11.64 11.23 0.41 All 3,643 17.75 17.10 0.65 BS exposure and mortality. We observed no Respiratory clear trend between decades, nor systematic 1971–1979 68 44.51 43.38 1.13 differences by cause of death. There was an 1980–1989 187 22.74 20.35 2.39 overall positive difference between case and 1990–1998 396 11.05 10.60 0.45 control exposure for each cause of death for All 651 17.90 16.83 1.07 the Collaborative cohort, but the differences Collaborative by decade were quite variable. All-cause Table 2 presents the estimated short-term 1971–1979 242 42.51 38.06 4.45 1980–1989 603 18.98 19.05 –0.07 effects for each cohort and mortality outcome, 1990–1998 571 12.65 12.59 0.06 using three averaging periods of exposure, and All 1,416 20.45 19.70 0.75 either using or not using the individual spa- Cardio vascular tial weights. In each case, the pollution effect 1971–1979 120 43.22 40.63 2.59 (percent increase in odds of mortality for a 1980–1989 278 19.11 18.52 0.59 10-μg/m increment in average BS concentra- 1990–1998 226 11.52 12.77 –1.25 tion) was estimated with and without adjust- All 624 21.00 20.69 0.31 Respiratory ing for the risk factors. The table illustrates 1971–1979 7 41.43 48.07 –6.64 how the estimates changed with increasing 1980–1989 50 22.75 21.43 1.32 degrees of adjustment for risk factors, spatial 1990–1998 54 12.87 13.09 –0.22 weighting, and bilinear tempera ture. All 111 19.12 19.05 0.07 Estimated effects on all-cause mortality. Each case–control set was given equal weight (i.e., the mean control exposure was calculated in each set and then Generally positive associations between short- these means were averaged). Number of deaths (cases); about 7% of deaths were excluded because the 3-day mean c term BS exposure and all-cause mortality were of BS was missing. Controls of same sex born within 1 year of case and reaching control date in the same calendar month as case death. estimated for both cohorts. Significant effects | | 1282 volume 120 number 9 September 2012 • Environmental Health Perspectives Short and long-term exposure–mortality associations were found in the Renfrew–Paisley cohort, par- respectively. Exposure–mortality associations BS in the Renfrew–Paisley cohort, but reduced ticularly when the individually weighted expo- were generally smaller and nonsignificant in those of the 31-day mean. Adjusting for tem- sures were used. Adjusting for risk factors and the Collaborative cohort, with some nonsig- perature made a substantial difference, reduc - temperature, the effect of 3-day mean BS was nificant negative associations for central site ing estimates by about 33% for the 3- and estimated as a 1.0% [95% confidence interval exposure over the 31-day period. Apart from 7-day means and considerably more for the (CI): –0.2, 2.3%] increase in hazard using cen- the smaller number of Collaborative cohort 31-day mean at the central site. trally estimated exposures and a 1.8% (95% participants included in our analyses, possi- Estimated effects on cardio vascular mortal- CI: 0.1, 3.5%) increase using individually ble reasons for the different results in the two ity. The pattern of associations with cardio- weighted exposures (Table 2). Corresponding cohorts are that they differed with regard to vascular mortality was similar to that for estimates of the effect of a 31-day mean BS age, social class, employment, geographical all-cause mortality, with higher estimates for were higher: 1.1% (95% CI: –2.3, 4.7%) and distribution, and election. the longer exposure periods and when indi- 3.4% (95% CI: –0.7, 7.7%) for centrally esti- Adjusting for risk factors did not greatly vidually weighted exposures were used. For the mated and individually weighted exposures change effect estimates of 3- and 7-day mean Renfrew–Paisley cohort the 31-day BS mean a 3 Table 2. Estimated percent change in mortality with a 10-μg/m increase in BS based on nested case–control analyses according to exposure time period (BS averaging period), cohort, and model. Days –1 to –3 Days 0 to –6 Days 0 to –30 Percent change Percent change Percent change (95% CI) p-Value (95% CI) p-Value (95% CI) p-Value All-cause Renfrew–Paisley CSBS 1.4 (0.3, 2.6) 0.011 1.7 (0.2, 3.2) 0.028 4.1 (1.2, 7.0) 0.005 CSBS + risk factors 1.5 (0.4, 2.6) 0.009 1.7 (0.2, 3.3) 0.027 3.7 (0.9, 6.7) 0.011 CSBS + risk factors + temperature 1.0 (–0.2, 2.3) 0.100 1.1 (–0.7, 2.8) 0.231 1.1 (–2.3, 4.7) 0.516 IWBS 2.3 (0.7, 3.8) 0.003 2.9 (0.9, 5.0) 0.005 6.3 (2.7, 10.0) 0.001 IWBS + risk factors 2.3 (0.8, 3.9) 0.003 3.0 (0.9, 5.1) 0.004 5.8 (2.2, 9.6) 0.001 IWBS + risk factors + temperature 1.8 (0.1, 3.5) 0.040 2.3 (0.0, 4.6) 0.051 3.4 (–0.7, 7.7) 0.106 Collaborative CSBS 1.3 (–1.0, 3.7) 0.270 1.7 (–1.6, 5.1) 0.305 –2.0 (–8.0, 4.4) 0.531 CSBS + risk factors 0.6 (–1.9, 3.2) 0.636 0.5 (–2.9, 4.1) 0.771 –3.1 (–9.2, 3.4) 0.345 CSBS + risk factors + temperature –0.1 (–2.9, 2.7) 0.921 –0.7 (–4.5, 3.3) 0.736 –5.0 (–12, 2.6) 0.194 IWBS 2.6 (0.3, 4.8) 0.024 3.4 (0.4, 6.5) 0.027 3.8 (–1.1, 8.9) 0.128 IWBS + risk factors 1.6 (–0.8, 4.1) 0.188 1.8 (–1.4, 5.1) 0.276 2.0 (–3.0, 7.2) 0.446 IWBS + risk factors + temperature 1.1 (–1.4, 3.8) 0.390 1.1 (–2.3, 4.7) 0.528 2.0 (–3.4, 7.6) 0.482 Cardio vascular Renfrew–Paisley CSBS 1.8 (0.1, 3.5) 0.042 2.9 (0.6, 5.3) 0.013 7.4 (3.1, 11.9) 0.001 CSBS + risk factors 1.4 (–0.3, 3.1) 0.109 2.6 (0.2, 5.0) 0.031 6.5 (2.1, 11.1) 0.003 CSBS + risk factors + temperature 0.9 (–1.0, 2.8) 0.351 1.5 (–1.1, 4.2) 0.271 2.9 (–2.3, 8.4) 0.276 IWBS 2.4 (0.1, 4.7) 0.043 3.8 (0.7, 7.0) 0.016 8.7 (3.2, 14.5) 0.002 IWBS + risk factors 2.0 (–0.3, 4.4) 0.089 3.5 (0.4, 6.8) 0.027 8.2 (2.6, 14.1) 0.004 IWBS + risk factors + temperature 1.4 (–1.2, 4.0) 0.287 2.2 (–1.3, 5.8) 0.225 4.1 (–2.2, 10.7) 0.208 Collaborative CSBS 0.3 (–3.2, 3.9) 0.875 1.7 (–3.2, 6.9) 0.507 –3.5 (–12, 6.0) 0.454 CSBS + risk factors –1.3 (–4.8, 2.4) 0.484 –0.3 (–5.1, 4.9) 0.919 –3.9 (–13, 6.0) 0.424 CSBS + risk factors + temperature –2.1 (–6.0, 1.9) 0.303 –1.1 (–6.5, 4.7) 0.712 –5.3 (–16, 6.1) 0.349 IWBS 1.5 (–1.9, 4.9) 0.394 3.0 (–1.6, 7.8) 0.202 1.3 (–5.8, 8.8) 0.735 IWBS + risk factors –0.2 (–3.6, 3.4) 0.930 1.0 (–3.6, 5.8) 0.681 0.3 (–7.0, 8.1) 0.944 IWBS + risk factors + temperature –0.6 (–4.3, 3.2) 0.747 0.6 (–4.4, 5.9) 0.820 0.4 (–7.5, 8.9) 0.930 Respiratory Renfrew–Paisley CSBS 2.8 (–1.3, 7.1) 0.181 3.1 (–2.3, 8.9) 0.270 10.1 (–0.6, 21.9) 0.065 CSBS + risk factors 1.7 (–2.6, 6.1) 0.449 1.9 (–3.6, 7.7) 0.503 5.6 (–4.8, 17.2) 0.304 CSBS + risk factors + temperature –0.2 (–5.0, 4.7) 0.920 –0.6 (–6.7, 6.0) 0.857 –0.6 (–12, 12.9) 0.932 IWBS 3.8 (–1.5, 9.4) 0.167 5.1 (–2.0, 12.7) 0.167 19.4 (5.2, 35.5) 0.006 IWBS + risk factors 2.0 (–3.6, 7.8) 0.495 3.0 (–4.2, 10.6) 0.425 12.1 (–1.4, 27.4) 0.082 IWBS + risk factors + temperature –0.4 (–6.4, 6.1) 0.912 0.2 (–7.6, 8.6) 0.964 7.2 (–7.5, 24.2) 0.358 Collaborative CSBS 0.0 (–9.6, 10.7) 0.998 0.7 (–12, 15.0) 0.922 –24 (–44, 3.2) 0.079 CSBS + risk factors 0.9 (–8.9, 11.6) 0.869 0.7 (–12, 14.9) 0.923 –23 (–43, 4.1) 0.089 CSBS + risk factors + temperature –1.2 (–12, 10.4) 0.826 –2.8 (–16, 12.8) 0.704 –29 (–50, 0.8) 0.055 IWBS 2.6 (–5.8, 11.8) 0.556 4.1 (–6.6, 16.1) 0.463 –16 (–34, 6.5) 0.151 IWBS + risk factors 2.4 (–6.1, 11.7) 0.589 2.5 (–8.2, 14.5) 0.656 –17 (–34, 4.6) 0.114 IWBS + risk factors + temperature 1.1 (–7.8, 10.9) 0.817 0.7 (–11, 13.4) 0.914 –19 (–38, 4.0) 0.097 Abbreviations: CI, confidence interval; CSBS, BS at (single) central monitoring site; IWBS, individually weighted BS from ratio of individual participant's long-term geometric mean exposure to BS, estimated by multilevel model, to geometric mean BS at the central monitoring site (time-series of daily values at central site were multiplied by individual ratios to produce individual-specific time-series for 1974–1998 reflecting both spatial and temporal variation); risk factors, adjustment for potential confounding variables measured at baseline—smoking history, social class, body mass index, marital status, systolic blood pressure, and total cholesterol; temperature, linear effects of temperature > and < 11°C. Estimates in this table are odds ratios (ORs) from conditional logistic regression models comparing the pollution experienced by the case immediately before death and that experienced by controls of the same age in approximately the same time period and at approximately the same time of year. | | Environmental Health Perspectives • volume 120 number 9 September 2012 1283 Beverland et al. was associated with larger mortality increases Table 3 also displays estimates of short- medium-term exposure on mortality within than the 3- and 7-day means [after adjusting term associations derived from time-series cohort participants directly to estimated effects for risk factors, temperature and spatial vari- modeling of the daily variation in mortality for of long-term exposure on mortality within the ation, a 10-μg/m increment in average BS the entire population of the Glasgow conurba- same cohorts (Yap et al. 2012). We also com- exposure over the previous 31 days was asso- tion that was > 50 years of age (Beverland pared the estimated effects of short- and medi - ciated with a 4.1% (95% CI: –2.2, 10.7%) et al. 2007). The time-series estimates were um-term exposure with rate–ratio estimates increase in cardiovascular mortality for this smaller than the corresponding short-term obtained from Poisson regression analysis of cohort]. In the Collaborative cohort there were estimates in the two cohorts, but they had nar- daily mortality within the general population no statistically significant associations and the rower confidence intervals, being based on a from which the majority of cohort participants estimated effect sizes were small and, in some much larger population. were selected (Beverland et al. 2007). instances, negative. For the Renfrew–Paisley cohort, larger Discussion Estimated effects on respiratory mortality. exposure–mortality associations were noted in There were very few statistically significant There have been a limited number of studies the nested case–control analyses than in the associations and estimates generally had very comparing short- and long-term effects of air time-series analyses using a single monitoring wide confidence intervals, making any infer - pollution in related study populations [e.g., site (Table 3); the difference being consistent ences extremely tentative. In the Collaborative Klemm and Mason (2003), Laden et al. for both all-cause and cardio vascular mortality. cohort, the 31-day exposures were associated (2006), and Schwartz et al. (1996) as reviewed This may be a result of more realistic expo- with strongly negative, although nonsignifi - by Pope (2007)]. However, to the best of our sure classification in the cohort-based analyses cant, effects; whereas in the Renfrew–Paisley knowledge, direct estimation of the short-term (Armstrong 1998; Sheppard et al. 2012). In cohort, the 31-day exposures were associated effects of air pollution within a cohort study both nested case–control and time-series analy- with large positive, but generally nonsig- has rarely been attempted. In our case–control ses, the pollution– mortality associations were nificant, effects. For 3- and 7-day mean BS, analysis, pollution on the days preceding the larger for medium-term (31-day) compared to the estimates were generally positive before death of a cohort participant was compared short-term (3-day) averaging/lag periods. adjusting for temperature. with the pollution preceding the days when However, the short- and medium-term Comparison with long-term effect the controls reached the age at which the case effect estimates from the nested case– control estimates. Table 3 compares selected estimates died. We matched for sex, month of death, analysis of the Renfrew–Paisley cohort were of the short and medium-term effects of BS on and within 1 year of birth and adjusted for substantially smaller than the equivalent effects mortality from Table 2 with estimated long- several individual-level risk factors strongly estimated for long-term exposure (Beverland term effects in the same cohorts, obtained associated with mortality. We also adjusted et al. 2007; Yap et al. 2012), which is consis- from proportional hazards modeling (Yap et al. exposure estimates for spatial variation in per- tent with a review of exposure–outcome asso- 2012). In the Renfrew–Paisley cohort long- sonal exposure, and adjusted for temperature. ciations at different time scales (Pope 2007). term pollution exposure–mortality associations This methodology might be expected to pro - For example, Yap et al. (2012) observed that were substantially greater in magnitude than duce estimates of short-term effects that are a 10-μg/m increase in average BS over the short-term exposure–mortality associations more realistic than those available from time- decade 1970–1979 was associated with a 10% for all three categories of mortality. Long-term series modeling and more comparable with (95% CI: 4, 17%) increase in the hazard of effects were much smaller in the Collaborative the estimates available from survival analy- all-cause mortality as compared with our esti- cohort than in the Renfrew–Paisley cohort and sis of long-term follow-up in cohorts. We mate of a 1.8% (95% CI: 0.1, 3.4%) increase were nonsignificant. compared the estimated effects of short- and in hazard for a 10-μg/m increase in average BS over 3 days. Estimated associations for the Table 3. Comparison of estimated magnitudes of associations [percent change (95% CI)] between short- Collaborative cohort were generally positive and long-term exposure to BS and mortality in the Renfrew–Paisley and Collaborative cohorts and in the but much less consistent in magnitude, with no population > 50 years of age of Glasgow, Renfrew, and Paisley conurbation with follow-up to 1998. significant pollution effects observed in separate a,b a,b c Mortality/population group Short-term (3-day) Medium-term (31-day) Long-term (1970–1979) analyses of this cohort. All-cause Our analyses of the effects of long-term Time-series 0.2 (0.0, 0.4) 0.9 (0.3, 1.5) — and individually weighted short-term BS expo- Renfrew–Paisley cohort 1.8 (0.1, 3.5) 3.4 (–0.7, 7.7) 10 (4, 17) sure assumed that the pattern of intraurban b,d Collaborative cohort 1.1 (–1.4, 3.8) 2.0 (–3.4, 7.6) 1 (–4, 6) spatial variations in long-term average BS Combined cohort 1.6 (0.2, 3.0) 2.9 (–0.5, 6.2) 5 (1, 9) exposure in 1970–1979 was largely sustained Cardio vascular Time-series 0.1 (–0.2, 0.4) 0.3 (–0.7, 1.2) — over the subsequent 1980–1998 period, which Renfrew–Paisley cohort 1.4 (–1.2, 4.0) 4.1 (–2.2, 10.7) 11 (1, 22) could not be verified because of the substan - b,d Collaborative cohort –0.6 (–4.3, 3.2) 0.4 (–7.5, 8.9) 3 (–5, 12) tial reduction in the BS monitoring network Combined cohort 0.8 (–1.4, 2.9) 2.7 (–2.4, 7.8) 7 (0, 13) during this later period. Similar assumptions Respiratory about relative invariance of spatial contrasts Time-series 0.3 (–0.2, 0.8) 3.1 (1.4, 4.9) — b in long-term air pollution exposure have been Renfrew–Paisley cohort –0.4 (–6.4, 6.1) 7.2 (–7.5, 24.2) 26 (2, 55) b,d made, by necessity, in almost all epidemiologi- Collaborative cohort 1.1 (–7.8, 10.9) –19.5 (–37.7, 4.0) –3 (–21, 18) cal studies of long-term intra urban air pollu- Combined cohort 0.1 (–5.1, 5.3) –2.6 (–15.2, 10.0) 11 (–3, 28) tion exposure effects. These assumptions have Table details percent increases in mortality associated with 10-μg/m increments in average BS. Rate ratios estimated by Poisson regression modeling (adjusted for temperature) for population > 50 years of age in the been partly supported by observations of “sta- contiguous Glasgow, Renfrew, and Paisley conurbation (Beverland et al. 2007). Odds ratios estimated by conditional bility” in spatial contrasts among measurement logistic regression modeling (adjusted for temperature) on matched case–control sets and adjusted for baseline risk sites in a small number of studies (Eeftens factors (smoking history, social class, body mass index, marital status, systolic blood pressure, and total cholesterol) and spatial variation (Table 2). Hazard ratios estimated by Cox regression modeling using long-term exposures esti- et al. 2011; Hoek et al. 2008). An additional mated from the spatio temporal model and adjusted for baseline risk factors listed in footnote b. Short-term effects were limitation is that exposure misclassification estimated for the Glasgow conurbation subset of the Collaborative cohort (n = 3,818); long-term effects were estimated e may have resulted from a lack of information for the full Collaborative cohort (n = 6,680). Combined cohort risk estimates computed from individual cohort estimates about participant mobility. (as outlined in above footnotes) using weights proportional to the inverse variance of risk estimates in individual cohorts. | | 1284 volume 120 number 9 September 2012 • Environmental Health Perspectives Short and long-term exposure–mortality associations Samoli et al. (2001) used time-series of air pollution, e.g., fine or ultrafine particles General Register Office. 1966. Classification of Occupations 1966. London:HMSO. methods to estimate an average effect on all- or specific transition metals (Heal et al. 2005; Goodman PG, Dockery DW, Clancy L. 2004. Cause-specific cause mortality of 3.1% (95% CI: 2.4, 3.9%) Hochadel et al. 2006; Janssen et al. 2011). mortality and the extended effects of particulate pollu- for a 50-μg/m increase in BS averaged over tion and temperature exposure. Environ Health Perspect Conclusions 112:179–185. the day of death and the previous day in four Hart CL, MacKinnon PL, Watt GC, Upton MN, McConnachie A, western European cities. A study using case– After adjusting for individual-level risk fac- Hole DJ, et al. 2005. The Midspan studies. Int J Epidemiol crossover analysis (Zeka et al. 2005), which tors, temperature, and geographical variation 34(1):28–34. Hawthorne VM, Watt GCM, Hart CL, Hole DJ, Smith GD, Gillis CR. has similarities to our approach, was applied in BS pollution, short and medium-term BS 1995. Cardiorespiratory disease in men and women in urban to PM data for 20 U.S. cities between 1989 exposure–mortality associations estimated Scotland: Baseline characteristics of the Renfrew/Paisley and 2000 and found that a 10-μg/m increase using cohort-based nested case–control analy- (Midspan) study population. Scott Med J 40:102–107. Heal MR, Hibbs LR, Agius RM, Beverland LJ. 2005. Total and in PM averaged over the day of death ses were of greater magnitude than associa- water-soluble trace metal content of urban background and the previous 2 days was associated with tions estimated for similar geographical areas PM , PM and black smoke in Edinburgh, UK. Atmos 10 2.5 0.45% (95% CI: 0.25, 0.65%), 0.50% (95% using single-pollution-site time-series analyses. Environ 39(8):1417–1430. CI: 0.25, 0.75%) and 0.87% (95% CI: However, short and medium-term exposure– Hochadel M, Heinrich J, Gehring U, Morgenstern V, Kuhlbusch T, Link E, et al. 2006. Predicting long-term average concentra- 0.38, 1.36%) increases in all-cause, cardio- mortality associations were of substantially tions of traffic-related air pollutants using GIS-based infor - vascular, and respiratory mortality, respec- lower magnitude than long-term exposure- mation. Atmos Environ 40(3):542–553. tively. Using data from Dublin for 1980–1996, mortality associations observed in the same Hoek G, Beelen R, de Hoogh K, Vienneau D, Gulliver J, Fischer P, et al. 2008. A review of land-use regression models to Goodman et al. (2004) estimated the effects of cohorts using survival analysis. These observa - assess spatial variation of outdoor air pollution. Atmos a 10-μg/m increase in 3-day average BS as tions indicate the importance of intraurban Environ 42(33):7561–7578. 0.4% (95% CI: 0.3, 0.6%), 0.4% (95% CI: variations in long-term pollution climates Janssen NAH, Hoek G, Simic-Lawson M, Fischer P, van Bree L, ten Brink H, et al. 2011. Black carbon as an additional indi- 0.2, 0.7%), and 0.9% (95% CI: 0.5, 1.2%) when estimating associations between expo- cator of the adverse health effects of airborne particles on all-cause, cardiovascular, and respiratory sure and mortality and suggest that public compared with PM and PM . Environ Health Perspect 10 2.5 mortality, respectively. Although none of these health impacts of air pollution may be domi- 119:1691–1699. Klemm RJ, Mason R. 2003. Replication of reanalysis of studies used adjustments for spatial variation nated by long-term exposure determined by Harvard Six-City Mortality study. In: Revised Analyses of that were comparable with our approach, and geographical differences in pollution climates. Time-Series of Air Pollution and Health. Special Report. they involved slightly different lag periods, the Boston:Health Effects Institute, 165–172. Available: http:// pubs.healtheffects.org/getfile.php?u=21 [accessed 8 May order of magnitude of the short-term effects Refe Rences 2012]. is not inconsistent with ours, especially given Künzli N. 2005. Unifying susceptibility, exposure, and time: Armstrong BG. 1998. Effect of measurement error on epide- the relatively wide confidence intervals associ - Discussion of unifying analytic approaches and future miological studies of environmental and occupational ated with our estimates. However, while the directions. J Toxicol Environ Health A 68(13–14):1263–1271. exposures. Occup Environ Med 55(10):651–656. Laden F, Schwartz J, Speizer FE, Dockery DW. 2006. Reduction use of spatially weighted exposures increased Beverland IJ, Robertson C, Yap C, Heal MR, Cohen GR, in fine particulate air pollution and mortality—extended Henderson DEJ, et al. 2012. Comparison of models for our effect estimates, even when using centrally follow-up of the Harvard Six Cities study. Am J Respir Crit estimation of long-term exposure to air pollution in cohort measured exposures our case–control estimates Care 173(6):667–672. studies. Atmos Environ; doi:10.1016/j.atmosenv.2012.08.001. Pope CA III. 2007. Mortality effects of longer term exposures to for 3-day BS were generally higher than those Beverland IJ, Yap C, Robertson C, Agius RM, Hole DJ, Cohen GR, fine particulate air pollution: Review of recent epidemio - et al. 2007. Department of Health research project (Ref: observed in general population time-series and logical evidence. Inhal Toxicol 19:33–38. 0020015). Health Effects of Long-term Exposure to Air case–crossover analyses. Pope CA III, Dockery DW. 2006. Health effects of fine particu- Pollution in Scotland. Available: http://www.dh.gov.uk/prod_ late air pollution: Lines that connect. J Air Waste Manag Another European multicity study using consum_dh/groups/dh_digitalassets/@dh/@en/documents/ Assoc 56(6):709–742. digitalasset/dh_092831.pdf [accessed 8 May 2012]. time-series methods reported city-specific all- Samoli E, Schwartz J, Wojtyniak B, Touloumi G, Spix C, Brook RD, Rajagopalan S, Pope CA III, Brook JR, Bhatnagar A, cause mortality effects of a 10-μg/m increase Balducci F, et al. 2001. Investigating regional differences Diez-Roux AV, et al. 2010. Particulate matter air pollution in short-term effects of air pollution on daily mortality in in PM over a 40-day period ranging from 10 and cardiovascular disease an update to the scientific the APHEA project: a sensitivity analy sis for controlling statement from the American Heart Association. Circulation –0.9% to 4.0% (Zanobetti et al. 2002). In our long-term trends and seasonality. Environ Health Perspect 121(21):2331–2378. case–control analy sis, we found associations of 109:349–353. Carder M, McNamee R, Beverland I, Elton R, Cohen GR, Boyd J, Schwartz J, Dockery DW, Neas LM. 1996. Is daily mortality all-cause mortality with 10-μg/m increases in et al. 2005. The lagged effect of cold temperature and wind associated specifically with fine particles? J Air Waste average BS over the day of death and the pre- chill on cardiorespiratory mortality in Scotland. Occup Manag Assoc 46(10):927–939. Environ Med 62(10):702–710. ceding 30 days of 3.4% (95% CI: –0.7, 7.7%) Sheppard L, Burnett R, Szpiro A, Kim S-Y, Jerrett M, Pope C III, Carder M, McNamee R, Beverland I, Elton R, Van Tongeren M, (Renfrew–Paisley after full adjustments) and et al. 2012. Confounding and exposure measurement Cohen GR, et al. 2008. Interacting effects of particulate error in air pollution epidemiology. Air Qual Atmos Health 2.0% (95% CI: –3.4, 7.6%) (Collaborative), pollution and cold temperature on cardiorespiratory mor- 5(2):203–216. tality in Scotland. Occup Environ Med 65(3):197–204. which are within this range and slightly greater World Health Organization. 1977. Manual of the International Carracedo-Martinez E, Taracido M, Tobias A, Saez M, Figueiras A. than the equivalent effect of 1.1% (95% CI: Statistical Classification of Diseases, Injuries and Causes of 2010. Case–crossover analysis of air pollution health effects: Death. Ninth Revision. Geneva:World Health Organization. 0.8, 1.3%) noted over a 40-day period in time- a systematic review of methodology and application. Environ Yap C, Beverland IJ, Robertson C, Heal MR, Cohen GR, Health Perspect 118:1173–1182. series analyses of BS and all-cause mortality in Henderson DEJ, et al. 2012. Association between long- Collett D. 1994. Modelling Survival Data in Medical Research. Dublin (Goodman et al. 2004). term exposure to air pollution and specific causes of London:Chapman and Hall. mortality in Scotland. Occup Environ Med; doi:101136/ The multicity studies cited above empha- Davey Smith G, Hart CL, Hole DJ, MacKinnon PL, Gillis CR, oemed-2011-100600. Watt GCM, et al. 1998. Education and occupational social size marked unexplained heterogeneity between Zanobetti A, Schwartz J, Samoli E, Gryparis A, Touloumi G, class: Which is the more important indicator of mortality city-specific estimates, so it is conceivable that Atkinson R, et al. 2002. The temporal pattern of mortality risk? J Epidemiol Community Health 52:153–160. responses to air pollution: A multicity assessment of mor- there are specific conditions in the Glasgow Department for Environment Food and Rural Affairs. 2005. tality displacement. Epidemiology 13(1):87–93. National Air Quality Data Archive. Available: http://uk-air. conurbation that might explain differences Zeka A, Zanobetti A, Schwartz J. 2005. Short term effects of defra.gov.uk/data/ [accessed 1 October 2005]. between our results and those of others. It is particulate matter on cause specific mortality: Effects Eeftens M, Beelen R, Fischer P, Brunekreef B, Meliefste K, Hoek G. of lags and modification by city characteristics. Occup also possible that BS gives a better measure 2011. Stability of measured and modelled spatial contrasts in Environ Med 62(10):718–725. than PM for the more damaging elements NO over time. Occup Environ Med 68(10):765–770. | | Environmental Health Perspectives • volume 120 number 9 September 2012 1285

Journal

Environmental Health Perspectives – Unpaywall

Published: Sep 1, 2012

There are no references for this article.