Impact of Outdoor Air Pollution on Survival After Stroke
Population-Based Cohort Study
Background and Purpose— The impact of air pollution on survival after stroke is unknown. We examined the impact of outdoor air pollution on stroke survival by studying a population-based cohort.
Methods— All patients who experienced their first-ever stroke between 1995 and 2005 in a geographically defined part of London, where road traffic contributes to spatial variation in air pollution, were followed up to mid-2006. Outdoor concentrations of nitrogen dioxide and particulate matter <10 μm in diameter modeled at a 20-m grid point resolution for 2002 were linked to residential postal codes. Hazard ratios were adjusted for age, sex, social class, ethnicity, smoking, alcohol consumption, prestroke functional ability, pre-existing medical conditions, stroke subtype and severity, hospital admission, and neighborhood socioeconomic deprivation.
Results— There were 1856 deaths among 3320 patients. Median survival was 3.7 years (interquartile range, 0.1 to 10.8). Mean exposure levels were 41 μg/m3 (SD, 3.3; range, 32.2 to 103.2) for nitrogen dioxide and 25 μg/m3 (SD, 1.3; range, 22.7 to 52) for particulate matter <10 μm in diameter. A 10-μg/m3 increase in nitrogen dioxide was associated with a 28% (95% CI, 11% to 48%) increase in risk of death. A 10-μg/m3 increase in particulate matter <10 μm in diameter was associated with a 52% (6% to 118%) increase in risk of death. Reduced survival was apparent throughout the follow-up period, ruling out short-term mortality displacement.
Conclusions— Survival after stroke was lower among patients living in areas with higher levels of outdoor air pollution. If causal, a 10-μg/m3 reduction in nitrogen dioxide exposure might be associated with a reduction in mortality comparable to that for stroke units. Improvements in outdoor air quality might contribute to better survival after stroke.
Stroke is a major cause of mortality worldwide, accounting for 9% of all deaths around the world and 10% to 12% of deaths in Western countries.1 There is increasing evidence linking outdoor air pollution and the incidence of stroke, including evidence from time-series, case-crossover, and cohort studies.2–4
Survival after stroke has been poor, with 50% of patients dying within 1 year of sustaining a stroke.1 Increasing efforts are being made to improve survival, and these include early therapeutic interventions and organized stroke care.1 However, the impact of outdoor air pollution on survival after stroke has not been examined. Exposure to outdoor air pollution is potentially modifiable, and even a small, adverse effect on survival can have important implications in public health terms because exposure to this potential prognostic factor is common.
Traffic-related pollution is a major and increasing contributor to outdoor air pollution in both developed and developing economies. Traffic-related pollution levels vary substantially at a fine spatial scale, as pollutant concentrations may decrease rapidly within short distances of roads.5 Studies examining the health effects of outdoor air pollution therefore would benefit from use of exposure estimates that are able to capture and incorporate exposure variation at a fine spatial scale.
We examined the effect of exposure to outdoor air pollution on survival after stroke by using the South London Stroke Register, a population-based, prospective, cohort study with multiple active surveillance methods designed to capture and follow up all incident cases of first-ever stroke at all ages occurring within the population living in a defined geographic area.6 We linked very-high-resolution modeled air pollution concentrations to patients in the cohort to carry out this study.7
Patients and Methods
Study Area and Patients
The register commenced in 1995, with a resident population of 272 000 in the register area at the 2001 census. The area boundary expanded by 2-fold in November 2004. Hospital and community notification sources included accident and emergency records, hospital staff, brain imaging requests, death certificates, coroners’ records, general practitioners, community nurses and therapists, bereavement officers, social services, hospital-based stroke registries, general practice computer records, and notification by patients or relatives. Estimated completeness of data capture was 80% to 88%.8,9
Patients were interviewed according to a standardized questionnaire and examination schedule. Information was obtained from hospital and general practice records, or next of kin when it could not be obtained from patients. We set out a priori to include all variables that had previously been included in a survival analysis of the data.10 The information used was related to sociodemographic factors (age, sex, social class, ethnicity), smoking and alcohol consumption, pre-existing medical conditions (hypertension, coronary heart disease, diabetes, atrial fibrillation, transient ischemic attack), prestroke functional ability (Barthel Index,11 living alone), severity of stroke (Glasgow Coma Scale,12 ability to swallow, urinary continence), and hospital admission. We used the income domain of the Index of Multiple Deprivation, the standard index used by government agencies in England, to adjust for socioeconomic confounding at the neighborhood level.13 We assigned to cases the deprivation score of the lower superoutput area that contained their residential postal code centroid. There were, on average, 28 residential postal codes per lower superoutput area and 53 people per postal code.
We defined the cohort as all patients who experienced their first-ever stroke between January 1, 1995, and December 31, 2005. Patients were followed up at 3 months by a register team field worker. Subsequently, field worker visits were made at 1 year after stroke and annually thereafter, with a postal questionnaire being used when a visit was impractical.14 Nonresponse was followed up by contact with the patients’ general practitioners, the health services authority, and next of kin. Patients were flagged at the National Health Service Central Register, and the Stroke Register was notified of deaths through this system. Date of death was confirmed by the Office for National Statistics. Survival time was from date of stroke to date of death. For patients with no record of death, survival time was censored on June 30, 2006. The study had approval from the ethics committee of Guy’s and St. Thomas’ Hospital Trust, King’s College Hospital, London.
Exposure to Outdoor Air Pollution
We used modeled outdoor air pollution data produced for the Greater London Authority. Modeled nitrogen dioxide (NO2) and particulate matter <10 μm in diameter (PM10) concentrations were available for 2002 at a 20×20-m grid point resolution, produced by using a system capable of modeling >1 million individual sources with different source characteristics.15–17 The model took into account a range of pollution sources and emissions, including major and minor road networks modeled with detailed information on vehicle stock, traffic flows, and speed on a link-by-link basis; and pollution sources in the London Atmospheric Emissions Inventory, including large and small regulated industrial processes, boiler plants, domestic and commercial combustion sources, agriculture, rail, ships, airports, and pollution carried into the area by prevailing winds. The modeling system has been used extensively in London for assessing the impacts of traffic and air quality management schemes, including the London Air Quality Strategy and congestion charging.18,19
Evaluation of the model had previously been undertaken by comparing modeled values with measured values from 56 NO2 and 38 PM10 monitoring sites throughout London (2 of which were in the study area). As part of the evaluation, detailed examination was performed to accurately pinpoint the measurement site locations down to 1-m accuracy levels and in particular to ensure that the separation distance from local roads was as accurate as possible because of the large concentration gradients that exist close to roads. The model predictions and measurements were then compared at these precise locations by using sites with data-capture rates of >75% throughout the year.
Air pollution values were linked to patients by using their residential postal code at the time of their stroke. Postal codes were assigned the value of the pollution grid point nearest to the postal code centroid, with average values used from equidistant grid points. Procedures were carried out with the use of ESRI ArcGIS version 9.0. For patients who moved during the study period, their postal code at death or at the end of the follow-up period was also assigned a pollution value when it was within the Greater London area. For these patients, their exposure was taken as the average of pollution values at the start and end of their follow-up period. For patients who had moved out of Greater London, pollution values could not be assigned to their postal code location at the end of their follow-up time. We therefore used half their follow-up time, censored their contribution to the study at that point, and used the pollution value at the time of stroke.
We used STATA version 9 for statistical analyses. We examined unadjusted associations between pollutants and survival from Kaplan-Meier curves. We used Cox regression to adjust for potential confounding factors. NO2 and PM10 were examined in separate analyses. Age and area-level deprivation were fitted as continuous variables. All other confounders were entered as categorical variables. The analyses were stratified by stroke subtype, Glasgow Coma Score, and social class, according to the modeling strategy described previously, wherein assumptions for proportional-hazards modeling were examined by using Schoenfeld residuals.10 We examined assumptions for NO2 and PM10 graphically (complementary log-log plots) and by statistical testing (ESTAT PHTEST option in STATA).
Elements of data were missing for various reasons, including death soon after stroke and incomplete information from next of kin and medical records. We therefore first analyzed data from cases with complete information only on all variables. We then imputed values for missing variables by using the ICE multiple-imputation procedure within STATA and reran the analyses on 10 imputed datasets.20 In the imputation, we included all explanatory variables and outcome information, as previous work has shown that outcome information should be included to avoid bias in estimation of effects.21 Pollutant values, deprivation, age, sex, outcome, and length of follow-up were not imputed because there were no missing values for these variables.
Sensitivity analyses included comparing the effects of pollutants on survival within 28 days of stroke with their effects on survival from 28 days onward, restricting the analyses to patients experiencing a stroke before November 2004 when the stroke area expanded by 2-fold, restricting the analyses to patients experiencing cerebral infarction, and excluding patients with high pollution exposure values.
We examined whether pollutant effects were modified by selected variables. We hypothesized that effects might be greater in older patients; smokers; patients with pre-existing hypertension, coronary heart disease, or diabetes; and patients who were incontinent of urine or unable to swallow. We obtained hazard ratios (HRs) and 95% CIs by running models separately for subgroups and assessed significance by using single models incorporating interaction terms.
The cohort comprised 3323 new cases of stroke, 3 of whom were excluded (1 withdrew and 2 had missing postal codes). There were 1856 deaths among 3320 patients in 9278 years of follow-up time. Median survival was 3.7 years (interquartile range, 0.1 to 10.8). Six hundred seventy-two of the 3320 patients moved during the study period, with 118 moving out of Greater London. Mean exposure concentrations were 41 μg/m3 (SD, 3.3) μg/m3 for NO2 and 25 (SD, 1.3) μg/m3 for PM10. Exposures ranged from 32.2 to 103.2 μg/m3 for NO2 and 22.7 to 52 μg/m3 for PM10.
Figure 1 shows maps of modeled outdoor NO2 and PM10 concentrations in 2002 zoomed in on part of the study area with a raster map for reference. The Stroke Register area was located within a busy, built environment in inner London, where the main contributor to spatial variation in pollution levels was road traffic–related pollution, as demonstrated by the maps. Figure 2 shows scatterplots comparing modeled and monitored values. The scatterplots show generally good agreement between monitored and modeled values (r=0.91 for NO2; r=0.90 for PM10).
Table 1 shows characteristics of patients in the cohort categorized by median levels of exposure to pollutants. In the higher-NO2 exposure category, there was a higher proportion of patients in the white group and a lower proportion in the black group. There was also a lower proportion of patients in the nonmanual social class, a higher proportion living alone before stroke, a marginally lower proportion with diabetes, and a lower proportion admitted to stroke units after stroke. There were broadly similar variations in relation to PM10. In addition, there were minor differences with regard to Barthel Index and atrial fibrillation.
The Kaplan-Meier plots (Figure 3) with pollutant values dichotomized by median values show that patients exposed to higher pollution concentrations had lower unadjusted survival than did those exposed to lower pollution concentrations. Reduced survival was apparent throughout the 11.5-year follow-up period. The unadjusted effect of the dichotomized categories was significant for NO2 (P=0.005) and borderline for PM10 (P=0.067) according to the log-rank test.
Table 2 shows unadjusted and adjusted HRs with 95% CIs for the effects of pollutants on survival after stroke. The unadjusted and adjusted HRs restricted to the 1787 cases with complete information on all potential confounders were higher than the results for all 3320 cases with imputation of missing data for potential confounding variables. The results presented here therefore focus on analyses of all 3320 cases. The unadjusted HR per 10-μg/m3 increase in NO2 was 1.23 (95% CI, 1.08 to 1.41), and the adjusted HR was 1.28 (95% CI, 1.11 to 1.48). For PM10, the unadjusted HR per 10-μg/m3 increase was 1.50 (95% CI, 1.07 to 2.09), whereas the adjusted HR was 1.52 (95% CI, 1.06 to 2.18).
Results of the sensitivity analyses showed that for NO2, the effects before and after a 28-day poststroke cutoff were similar, that the expansion of the area made little difference in the results, and that the results restricted to cerebral infarction were similar to those for all stroke. Exclusion of patients with high NO2 values marginally increased the effect size. The pattern for PM10 was more variable, and the estimates from the sensitivity analyses had wide CIs. The estimates were, however, still compatible with the result from the main analysis.
Table 3 compares HRs for selected subgroups. We found no evidence that older age, smoking, pre-existing medical conditions (hypertension, coronary heart disease, diabetes), or severity of stroke (urinary incontinence, inability to swallow) significantly exacerbated the adverse effects of NO2 or PM10 on survival after stroke.
We found that patients who experienced stroke had reduced survival if they lived in areas with higher levels of outdoor air pollution, after a wide range of potential confounding factors was taken into account. The effect was seen with both PM10 and NO2 but was more significant for NO2, a marker for traffic-related pollution. Reduced survival was apparent for the entire follow-up period, clearly demonstrating that the adverse effect of air pollution was not a mortality-displacement phenomenon, whereby death after stroke was hastened by a few days or weeks among patients already likely to die within a short period.
Ours is the first study to examine the effect of outdoor air pollution on survival after stroke, and the magnitude of association is consequential. If causal, a 10-μg/m3 reduction in NO2 exposure would be associated with a 22% decrease in mortality after stroke, an effect size comparable to that for stroke units1 (1/1.28=0.78, a 22% decrease). Our results add to the growing body of evidence linking air pollution and stroke. Daily time-series and case-crossover studies have found short-term effects of outdoor air pollution on stroke incidence when hospital admissions are used as a proxy.2,3,22 The adverse effect of long-term exposure to outdoor air pollution on stroke incidence has also been reported.4 We have previously observed higher stroke mortality rates in areas close to main roads.23 We have also previously found that rate ratios were greater for mortality than for admissions in areas with higher pollution levels,24 which would be consistent with mortality rate ratios reflecting both increased incidence and subsequent reduced survival after stroke in more polluted areas.
Although there was a wide range in exposures in our study, most of the cohort was exposed to a narrow exposure band. The study area comprised a small part of Greater London and was relatively homogeneous in terms of population exposure to air pollution. Nevertheless, we were able to detect an association between air pollution and stroke survival that we attribute to the model’s ability to detect variation in pollution exposure at a fine spatial scale. Previous studies have found that within-city estimates of effect were larger than between-city effect estimates.4,25 We have expressed effect estimates in terms of a 10-μg/m3 change in NO2 (1 part per billion=1.91 μg/m3) and PM10 concentrations, as this is becoming a standard quantity for expressing effect estimates for these pollutants and this magnitude of change lies within the overall range of values observed in our study.
The association we have observed could be explained by various potential mechanisms. Stroke may compromise respiratory function through paralysis and immobility, and stroke survivors may be more susceptible to respiratory infection. These consequences may increase stroke survivors’ susceptibility to the adverse respiratory effects of air pollution.26 There is also evidence linking impaired respiratory function to fatal stroke.27 Cardiovascular events are a significant cause of mortality after stroke, and there is increasing evidence linking air pollution and cardiovascular outcomes through a variety of mechanisms, including increased coagulability, cardiac arrhythmias, and atherogenesis.28 We were unable to investigate cause-specific mortality, as this information was incomplete.
There are a number of potential limitations to our study. We used modeled exposure estimates from a single year, and pollution levels could have varied across the study period. The use of estimates from a single year assumes that the rank order of levels of these pollutants within the study area did not change during the study period. Although this assumption may not necessarily hold, inspection of measured pollution levels during the study period in Greater London suggested that in general, the rank order remained relatively stable. The level of agreement between monitored and modeled values in terms of absolute concentrations was generally good for both pollutants. We took location of residence at the start and end of the study into account when assessing exposure to outdoor air pollution. Nevertheless, our exposure assessment is necessarily limited for various reasons, including a lack of information on daily activity levels and time spent indoors and in traffic. We did not have exact address locations and used postal code centroids as a proxy, which could have resulted in exposure assignment error. If the limitations in exposure assessment resulted in classic measurement error, the true underlying effect would be underestimated, but a Berkson-type error would not be expected to bias health effect estimates. Exposures derived from prediction models are associated with nontrivial prediction error, but properly incorporating that error into health effects models is difficult, and the optimal procedure for doing so is still uncertain.
We adjusted for a number of potential confounders, including individual social class and neighborhood-level socioeconomic deprivation, and the association was independent of these covariates. However, the possibility of residual or unmeasured confounding due to socioeconomic and other factors cannot be ruled out. We addressed the missing data issue by using multiple imputation. Imputation needs to be used with care, and the potential for bias from inadequate use of imputed data has been highlighted.29 Although the imputation procedure may not have a sound theoretical basis, it appears to perform well in practice.30 Expansion of the Stroke Register area in the last year of the recruitment period could have affected completeness of case capture but had little effect on the results of our study.
In conclusion, our study provides evidence that patients who have sustained a stroke may have lower survival if they live in areas with higher levels of outdoor, including traffic-related, air pollution. The results suggest that improvements in outdoor air quality, including a reduction in traffic-related pollution, might contribute to better survival after stroke.
Sources of Funding
We would like to thank the Colt Foundation, which supported this study through a research grant. C.D.W. acknowledges financial support from the Department of Health via the NIHR Comprehensive Biomedical Research Centre Award to Guy’s and St. Thomas’ NHS Foundation Trust in partnership with King’s College London.
This work used Crown copyright data supplied by Ordnance Survey/EDINA. The views expressed in this article do not necessarily reflect the views of the funding bodies.
- Received September 11, 2009.
- Revision received December 18, 2009.
- Accepted December 30, 2009.
Tsai SS, Goggins WB, Chiu HF, Yang CY. Evidence for an association between air pollution and daily stroke admissions in Kaohsiung, Taiwan. Stroke. 2003; 34: 2612–2616.
Wellenius GA, Schwartz J, Mittleman MA. Air pollution and hospital admissions for ischemic and hemorrhagic stroke among Medicare beneficiaries. Stroke. 2005; 36: 2549–2553.
Department of transport: air quality. In: Design Manual for Roads and Bridges, Vol 11: Environmental Assessment. London, UK: HSMO; 1994: section 3, pt 1.
Stewart JA, Dundas R, Howard RS, Rudd AG, Wolfe CD. Ethnic differences in incidence of stroke: prospective study with stroke register. BMJ. 1999; 318: 967–971.
Tilling K, Sterne JA, Wolfe CD. Estimation of the incidence of stroke using a capture-recapture model including covariates. Int J Epidemiol. 2001; 30: 1351–1359.
Heuschmann PU, Grieve AP, Toschke AM, Rudd AG, Wolfe CD. Ethnic group disparities in 10-year trends in stroke incidence and vascular risk factors: the South London Stroke Register (SLSR). Stroke. 2008; 39: 2204–2210.
Wolfe CD, Smeeton NC, Coshall C, Tilling K, Rudd AG. Survival differences after stroke in a multiethnic population: follow-up study with the South London Stroke Register. BMJ. 2005; 331: 431–433.
Communities and local government: indices of deprivation. Available at: http://www.communities.gov.uk/communities/neighbourhoodrenewal/deprivation/ Last accessed September 1, 2009.
Sarker SJ, Heuschmann PU, Burger I, Wolfe CDA, Rudd AG, Smeeton NC, Toschke AM. Predictors of survival after haemorrhagic stroke in a multi-ethnic population: the South London Stroke Register (SLSR). J Neurol Neurosurg Psychiatry. 2008; 79: 260–265.
Environmental Research Group. Available at: http://www.erg.kcl.ac.uk/ Last accessed September 1, 2009.
Carslaw DC, Beevers SD, Fuller G. An empirical approach for the prediction of annual mean nitrogen dioxide concentrations in London. Atmos Environ. 2001; 35: 1505–1515.
Fuller GW, Carslaw DC, Lodge HW. An empirical approach for the prediction of daily mean PM10 concentrations. Atmos Environ. 2002; 36: 1431–1441.
Greater London Authority. Cleaning London’s Air: the Mayor’s Air Quality Strategy. Available at: http://www.london.gov.uk/mayor/strategies/air_quality/air_quality_strategy.jsp Last accessed September 1, 2009.
Royston P. Multiple imputation of missing values: update of ICE. STATA J. 2005; 5: 527–536.
Maheswaran R, Elliott P. Stroke mortality associated with living near main roads in England and Wales: a geographical study. Stroke. 2003; 34: 2776–2780.
Maheswaran R, Haining RP, Brindley P, Law J, Pearson T, Fryers PR, Wise S, Campbell MJ. Outdoor air pollution and stroke in Sheffield, United Kingdom: a small-area level geographical study. Stroke. 2005; 36: 239–243.
Carlin JB. Multiple imputation needs to be used with care and reported in detail. Available at: http://www.bmj.com/cgi/eletters/bmj.39261.471806.55v1#174954. Last accessed September 1, 2009.