Using linked mortality and census data, we report a significant positive association between estimated long-term exposure to air pollution (PM10) and mortality in New Zealand urban areas. This setting includes approximately 75% of the New Zealand population, who are exposed to relatively low levels of PM10 compared with other countries.
Selection bias and confounding seem unlikely to explain our results. There is no evidence of selection bias (). The results persist after controlling for plausible confounders, including multiple measures of socioeconomic position and smoking. It seems unlikely that mismeasured or unknown confounders might explain the remaining association.
There is likely to be substantial misclassification of the air pollution exposure. Most of this misclassification was probably non-differential by mortality risk (meaning we have probably significantly underestimated the true strength of association). There is potential differential misclassification of PM10 by mortality risk in our study, because our assessment was based on modelling in one city using proxies, including domestic heating, estimates of industrial emissions and vehicle kilometres travelled within small areas. If these proxies are not such reliable predictors of PM10 in other cities, and are (say) correlated with socioeconomic position, then it may be that our PM10 estimates are also capturing aspects of socioeconomic exposure. However, the fact that an association remained after extensive control of socioeconomic factors, including individual level income and education, makes this an unlikely explanation of the results. The use of modelled estimates of PM10 exposure will tend to smooth the data and reduce the resulting CI. However, this should not affect the central effect estimates.
It is possible that less healthy people might migrate towards health services (or other service amenities) that happen to be in more polluted areas—a form of reverse causation or endogeneity. However, we think this is unlikely to be an important factor as New Zealand cities are relatively small (maximum 1.4 million), and most suburbs in New Zealand's main cities have relatively good access to hospitals.
The odds of all-cause mortality in adults (aged between 30 and 74 years at census) increased by 7% (95% CI 3% to 10%) per 10 μg/m3
increase in average PM10
exposure. Our observations are consistent with an increasing number of studies of long-term exposure to particulate matter and mortality.3–10
The original US Six Cities Study reported an adjusted mortality rate ratio of 1.27 (95% CI 1.08 to 1.48) for the most polluted compared with the least polluted city, corresponding to 18.2 and 46.5 μg/m3
—equivalent to an increase in mortality of approximately 10% per 10 μg/m3
. In the US Nurses Health Study, there was a 16% (5% to 28%) increase in all-cause mortality per 10 μg/m3
It is not yet clear to what extent the heterogeneity in reported dose response in those studies is related to differences in the accuracy of exposure measurement, to differences in the toxicity of complex mixtures of pollutants at differing levels of exposure, and/or differences in the sensitivity of exposed populations.
The exposure measures in other studies are not directly comparable. Recent studies use the more specific measure PM2.5
(particulate matter with an aerodynamic diameter less than 2.5 μm) rather than PM10
, whereas several European studies use black smoke or total suspended particulates as the exposure measure. The association between particulate air pollution exposure and mortality is usually found to be strongest for finer fractions, such that the dose response for PM2.5
is greater than for PM10
, which in turn is greater than for total suspended particulates. An extended follow-up of the US Six Cities Study reported a 14% (6% to 22%) increase in mortality per 10 μg/m3
whereas the American Cancer Society Study reported 6% (2% to 11%) and the Nurses Health Study 26% (2% to 54%), while coarse particulate matter exposure (PM10–2.5
) was not associated with an increase in mortality in that study.27
An analysis based on electoral wards in the UK found a 1.3% (1% to 1.6%) increase in all-cause mortality per 10 μg/m3
increase in black smoke.10
In 18 regions in France, the PAARC study reported a 7% (3% to 10%) increase per 10 μg/m3
increase in black smoke and a 5% (2% to 8%) increase for total suspended particulates.7
In The Netherlands, there was a 5% (0% to 10%) increase in mortality per 10 μg/m3
increase in black smoke.
Our study assessed the association of PM10
and mortality over 3 years. In the US Nurses Health Study, mortality was most strongly associated with average PM10
exposures in the 24 months before death. In the UK, the association, although weak, was stronger for exposure in the 4 years before death. However, a re-analysis of the American Cancer Society Study found no clear effect of exposure period.17
A priori, we hypothesised that the association of PM10 with mortality in our study would be stronger for a cohort restricted to those census respondents who lived in the same census area at the time of the 1991 census. We also hypothesised that the association of PM10 with mortality would be stronger for cardiorespiratory deaths. Our results were consistent with these a priori hypotheses, strengthening the ability to make causal inference.
There is some evidence, most consistently in studies with individual measurements of social factors, that more deprived populations are particularly sensitive to air pollution effects.5
Our ability to detect a true difference by ethnicity in sensitivity to air pollution was limited by the relatively small Maori population. Although not significant, the difference in our central effect estimates for European and Maori was substantial: 7% versus 20%. The reasons for this difference, if real, are not clear. This might reflect a higher prevalence of pre-existing cardiorespiratory disease among Maori, or a difference in the toxicity of air pollution to which different ethnic groups are typically exposed. Alternatively, this finding may reflect biases in exposure estimates. We found no other suggestion of interaction of social factors with PM10
in the association with mortality.