Short Term Exposure to Air Pollution and Stroke
Short Term Exposure to Air Pollution and Stroke
We assessed the abstracts of 2748 articles and reviewed 238 relevant studies in depth. Of these, we identified 103 that fulfilled the inclusion criteria (Appendix 2 and Appendix 3 ). Sixty nine studies used a time series design, 33 used a case crossover design, and one used both study designs incorporating over 6.2 million events across 28 countries ( Appendix 3 ). Twenty five (24%) studies presented analyses stratified by type of stroke (haemorrhagic and ischaemic), though only a minority of studies reported on haemorrhagic strokes (15 studies, 15%). Most studies presented risk estimates for stroke defined from administrative databases using the ICD-9 and ICD-10 codes ( Appendix 4 ). Most studies adjusted for other meteorological parameters including time trends, seasonality, and temperature ( Appendix 4 ). Not all studies provided risk estimates across all pollutants ( Appendix 3 ).
Of the 103 studies that met the inclusion criteria, we excluded nine studies from meta-analysis. Of these nine studies, five presented estimates as a subset of the parent study and therefore only estimates from the parent study were included. Four studies were excluded as risks were presented per category of pollutant concentration rather than per unit increment, the increment was not defined, or the effect estimates were not presented as either relative risks or odds ratios.
There was a positive association between all gaseous and particulate air pollutants and admission to hospital for stroke or mortality from stroke, with the weakest association seen for ozone (Figure 1). Individual forest plots for each of the pollutants are presented in Appendix 6.
(Enlarge Image)
Figure 1.
Association between gaseous and particulate air pollutants and admission for stroke or mortality from stroke stratified by time lag (days).
Both PM2.5 and PM10 were positively associated with admission to hospital for stroke or mortality from stroke, with a stronger association for PM2.5. The increase in relative risk was 1.011 (95% confidence interval 1.011 to 1.012) per 10 µg/m increase in PM2.5 concentration (Figure 1). The association between PM2.5 and stroke was evident on the day of the event (lag 0) and was present for up to two days (lag 2) before the event.
These associations persisted when we stratified by outcome (admission or death), sex, age (>65), or study design (Figure 2 and Appendix 7) for either PM2.5 or PM10. Few studies reported summary estimates by type of stroke, but when they were reported, the association for PM2.5 was 1.010 (95% confidence interval 1.008 to 1.011; I=69%) for ischaemic stroke and 1.004 (0.978 to 1.029; I=14%) for haemorrhagic stroke ( Appendix 7A and Appendix 7B ). Risks for ischaemic stroke and haemorrhagic stroke associated with PM10 exposure were comparable (1.002 (0.999 to 1.004; I=23%) and 1.002 (0.997 to 1.006; I=0%), respectively).
(Enlarge Image)
Figure 2.
Associations across all gaseous and particulate air pollutants stratified by outcome, age, and study design.
Nitrogen dioxide was the most commonly measured gaseous pollutant, with a 1.014 (95% confidence interval 1.009 to 1.019) relative increase in risk of admission to hospital for stroke or mortality from stroke per 10 ppb increment across two million events (Figure 1). Both sulphur dioxide (1.019 (1.011 to 1.027) per 10 ppb) and carbon monoxide (1.015 (1.004 to 1.026) per 1 ppm) were also positively associated with admission and mortality. Ozone however, showed only a weak association (1.001 (1.000 to 1.002) per 10 ppb; Figure 1).
These associations persisted when we stratified by outcome, age, and study design. All gaseous pollutants except ozone showed a positive and consistent relation with ischaemic stroke ( Appendix 7A and Appendix 7B ). Nitrogen dioxide exposure showed a consistent association with both ischaemic and haemorrhagic stroke (1.024 (95% confidence interval 1.010 to 1.038, I=56%) and 1.024 (1.003 to 1.045; I=42%), respectively; see Appendix 7A and Appendix 7B ).
The association between gaseous pollutants and stroke was related to lag in exposure (days), with the strongest associations evident for pollutant concentrations on the day of the event (lag 0) and diminishing with longer lag periods (Figure 1).
Most studies (80%) originated from high income countries, with only 21 (20%) originating from low or middle income countries. Both nitrogen dioxide and particulate matter (PM10) were commonly measured across high and low to middle income countries. Studies for these pollutants originated from Latin America (including Brazil, Chile, and Mexico), South Africa, China, Thailand, Iran, and South Africa (Figure 3). Most studies from low and middle income countries originated from mainland China (14 studies).
(Enlarge Image)
Figure 3.
Cartogram identifying association between nitrogen dioxide and particulate matter (PM10) and admission to hospital for stroke or mortality from stroke in countries stratified as high and low to middle income.
Pooled estimates from studies originating in low and middle income countries showed a stronger association than high income countries for nitrogen dioxide (1.019 (95% confidence interval 1.011 to 1.027) v 1.012 (1.006 to 1.017)) and PM10 (1.004 (1.002 to 1.006) v 1.002 (1.001 to 1.003)) (Figure 3). The median pollutant concentrations for nitrogen dioxide and PM10 were higher in low and middle income countries (median pollutant concentration 27.6 ppb (interquartile range 23.8–29.6 ppb) for nitrogen dioxide and 50.2 µg/m (32.6–65.7 µg/m) for PM10) than in high income countries (median pollutant concentration 22.6 ppb (19.4–28.3 ppb) for nitrogen dioxide and 25.3µg/m (23.8–29.6 µg/m) for PM10).
There was no difference in our overall effect estimates when we removed studies at increased risk of bias ( Appendix 8 ). Publication bias (Egger's test for asymmetry P<0.05) was observed for all pollutants except sulphur dioxide and PM2.5 ( Table 1 ; Appendix 4 ). Adjustment for asymmetry with the trim and fill method did not alter the effect direction but, as expected, did attenuate the effect size. We observed heterogeneity across all pollutants, and this was most evident with PM2.5 (I=86%) and least evident with PM10 (I=24%).
Results
We assessed the abstracts of 2748 articles and reviewed 238 relevant studies in depth. Of these, we identified 103 that fulfilled the inclusion criteria (Appendix 2 and Appendix 3 ). Sixty nine studies used a time series design, 33 used a case crossover design, and one used both study designs incorporating over 6.2 million events across 28 countries ( Appendix 3 ). Twenty five (24%) studies presented analyses stratified by type of stroke (haemorrhagic and ischaemic), though only a minority of studies reported on haemorrhagic strokes (15 studies, 15%). Most studies presented risk estimates for stroke defined from administrative databases using the ICD-9 and ICD-10 codes ( Appendix 4 ). Most studies adjusted for other meteorological parameters including time trends, seasonality, and temperature ( Appendix 4 ). Not all studies provided risk estimates across all pollutants ( Appendix 3 ).
Of the 103 studies that met the inclusion criteria, we excluded nine studies from meta-analysis. Of these nine studies, five presented estimates as a subset of the parent study and therefore only estimates from the parent study were included. Four studies were excluded as risks were presented per category of pollutant concentration rather than per unit increment, the increment was not defined, or the effect estimates were not presented as either relative risks or odds ratios.
There was a positive association between all gaseous and particulate air pollutants and admission to hospital for stroke or mortality from stroke, with the weakest association seen for ozone (Figure 1). Individual forest plots for each of the pollutants are presented in Appendix 6.
(Enlarge Image)
Figure 1.
Association between gaseous and particulate air pollutants and admission for stroke or mortality from stroke stratified by time lag (days).
Particulate Pollutants
Both PM2.5 and PM10 were positively associated with admission to hospital for stroke or mortality from stroke, with a stronger association for PM2.5. The increase in relative risk was 1.011 (95% confidence interval 1.011 to 1.012) per 10 µg/m increase in PM2.5 concentration (Figure 1). The association between PM2.5 and stroke was evident on the day of the event (lag 0) and was present for up to two days (lag 2) before the event.
These associations persisted when we stratified by outcome (admission or death), sex, age (>65), or study design (Figure 2 and Appendix 7) for either PM2.5 or PM10. Few studies reported summary estimates by type of stroke, but when they were reported, the association for PM2.5 was 1.010 (95% confidence interval 1.008 to 1.011; I=69%) for ischaemic stroke and 1.004 (0.978 to 1.029; I=14%) for haemorrhagic stroke ( Appendix 7A and Appendix 7B ). Risks for ischaemic stroke and haemorrhagic stroke associated with PM10 exposure were comparable (1.002 (0.999 to 1.004; I=23%) and 1.002 (0.997 to 1.006; I=0%), respectively).
(Enlarge Image)
Figure 2.
Associations across all gaseous and particulate air pollutants stratified by outcome, age, and study design.
Gaseous Pollutants
Nitrogen dioxide was the most commonly measured gaseous pollutant, with a 1.014 (95% confidence interval 1.009 to 1.019) relative increase in risk of admission to hospital for stroke or mortality from stroke per 10 ppb increment across two million events (Figure 1). Both sulphur dioxide (1.019 (1.011 to 1.027) per 10 ppb) and carbon monoxide (1.015 (1.004 to 1.026) per 1 ppm) were also positively associated with admission and mortality. Ozone however, showed only a weak association (1.001 (1.000 to 1.002) per 10 ppb; Figure 1).
These associations persisted when we stratified by outcome, age, and study design. All gaseous pollutants except ozone showed a positive and consistent relation with ischaemic stroke ( Appendix 7A and Appendix 7B ). Nitrogen dioxide exposure showed a consistent association with both ischaemic and haemorrhagic stroke (1.024 (95% confidence interval 1.010 to 1.038, I=56%) and 1.024 (1.003 to 1.045; I=42%), respectively; see Appendix 7A and Appendix 7B ).
The association between gaseous pollutants and stroke was related to lag in exposure (days), with the strongest associations evident for pollutant concentrations on the day of the event (lag 0) and diminishing with longer lag periods (Figure 1).
Stratification By Category of National Income
Most studies (80%) originated from high income countries, with only 21 (20%) originating from low or middle income countries. Both nitrogen dioxide and particulate matter (PM10) were commonly measured across high and low to middle income countries. Studies for these pollutants originated from Latin America (including Brazil, Chile, and Mexico), South Africa, China, Thailand, Iran, and South Africa (Figure 3). Most studies from low and middle income countries originated from mainland China (14 studies).
(Enlarge Image)
Figure 3.
Cartogram identifying association between nitrogen dioxide and particulate matter (PM10) and admission to hospital for stroke or mortality from stroke in countries stratified as high and low to middle income.
Pooled estimates from studies originating in low and middle income countries showed a stronger association than high income countries for nitrogen dioxide (1.019 (95% confidence interval 1.011 to 1.027) v 1.012 (1.006 to 1.017)) and PM10 (1.004 (1.002 to 1.006) v 1.002 (1.001 to 1.003)) (Figure 3). The median pollutant concentrations for nitrogen dioxide and PM10 were higher in low and middle income countries (median pollutant concentration 27.6 ppb (interquartile range 23.8–29.6 ppb) for nitrogen dioxide and 50.2 µg/m (32.6–65.7 µg/m) for PM10) than in high income countries (median pollutant concentration 22.6 ppb (19.4–28.3 ppb) for nitrogen dioxide and 25.3µg/m (23.8–29.6 µg/m) for PM10).
Bias and Heterogeneity
There was no difference in our overall effect estimates when we removed studies at increased risk of bias ( Appendix 8 ). Publication bias (Egger's test for asymmetry P<0.05) was observed for all pollutants except sulphur dioxide and PM2.5 ( Table 1 ; Appendix 4 ). Adjustment for asymmetry with the trim and fill method did not alter the effect direction but, as expected, did attenuate the effect size. We observed heterogeneity across all pollutants, and this was most evident with PM2.5 (I=86%) and least evident with PM10 (I=24%).
Source...