Study objective: The 1995 UK Environment Act required local authorities to review air quality and, where UK National Air Quality Strategy objectives (except ozone) are likely to be exceeded in 2005, to declare local air quality management areas and prepare action plans. This study modelled the impacts on health of reductions from current levels of PM10 to these objectives.
Design: The framework for conducting quantified health impact assessment assessed causality, then, if appropriate, examined the shape and magnitude of the exposure-response relations. The study modelled declines in pollution to achieve the objectives, then modelled the numbers of deaths and admissions affected if air pollution declined from existing levels to meet the objectives, using routine data.
Setting: Westminster, central London.
Main results: Attaining the 2004 PM10 24 hour objective in Westminster results in 1–21 lives no longer shortened in one year (annual deaths 1363). Reducing exceedences from 35 to seven almost doubles the estimates. The 2009 objective for the annual mean requires a substantial reduction in PM10, which would delay 8–20 deaths. About 20 respiratory and 14–20 circulatory admissions would be affected and around 5% of emergency hospital attendances for asthma by attaining the lower annual mean target. The effects of long term exposure to particulates may be an order of magnitude higher: models predict about 24 deaths are delayed by reaching the 2004 annual target (40 μg/m3[gravimetric]) and a hundred deaths by reducing annual mean PM10 to 20 μg/m3[gravimetric].
Conclusions: Modelling can be used to estimate the potential health impacts of air quality management programmes.
- air pollution
- health impact assessment
- APHEA, short term effects of air pollution on health: European approach
- NMMAPS, national mortality and morbidity air pollution study
- COPD, chronic obstructive pulmonary disease
- TEOM, tapered element oscillating microbalance
- ACS, American Cancer Society study
- GAM, generalised additive models
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
- APHEA, short term effects of air pollution on health: European approach
- NMMAPS, national mortality and morbidity air pollution study
- COPD, chronic obstructive pulmonary disease
- TEOM, tapered element oscillating microbalance
- ACS, American Cancer Society study
- GAM, generalised additive models
The United Kingdom 1995 Environment Act required the development of a National Air Quality Strategy for the UK.12 This set standards for eight outdoor air pollutants: sulphur dioxide, lead, carbon monoxide, ozone, nitrogen dioxide, benzene, 1,3-butadiene, and particulates, measured as the mass of PM10 particles, collected through a size selective inlet that collects 50% of particles of 10 μm aerodynamic diameter, >95% of 5 µm particles and <5% of 20 µm diameter particles (hence PM10: particulate mass with less than 10 μm diameter). The act required each local authority to review air quality and, where target levels other than for ozone are likely to be exceeded in the year 2005, to declare a local Air Quality Management Area, and prepare an action plan.
Many of the highest levels in the UK of PM10 and nitrogen dioxide are found in central London.34 Westminster City Council, a central London borough with an estimated resident population of 216 244 in 2000 and an average annual mortality of 1363, has designated the whole borough an Air Quality Management Area as it will continue to exceed the UK National Air Quality Strategy targets for PM10 and nitrogen dioxide, although it will meet the objectives for the other pollutants.
We modelled the acute impacts on health of meeting the UK National Air Quality Strategy objectives, which require a reduction from the current levels of these two pollutants to the targets shown in table 1. We applied the method to Westminster, using a model that is generalisable to any area with raised levels of pollution.
We developed a four stage framework for quantifying the potential health impacts of air pollution reduction (fig 1).
We conducted a systematic literature review of epidemiological papers examining the effects of particulates and/or nitrogen dioxide, following the principles given by the World Health Organisation guidance5 both for searching for relevant publications and for appraising the studies. A comprehensive bibliographic search was conducted, using Medline (1966–2002); BIDS-Embase (1980–2002); Biological Abstracts (1991–1998) and the Health Management Information Consortium (HMIC) CD-ROM, which included “grey” literature. Other sources of information were expert reports commissioned by the government and publications cited in other references. Search terms are detailed by Mindell.6
The search produced 410 studies of particulate matter, published by December 2001, of which 38 were of poor quality or irrelevant. An additional 52 studies were published in 2002. Assessment of the effects of air pollution is based on studies with good data collection, appropriate analysis, and in circumstances that can be extrapolated to pollution in London.6 Altogether 220 epidemiological studies were found on the effects of nitrogen dioxide, of which nine were of poor quality or irrelevant. An additional 27 studies were published in 2002.
Is there a causal association?
The published studies and reports were examined firstly for evidence of an association. Where one of the pollutants was associated with a specific effect on health, consideration was given to whether or not this was likely to be attributable to confounding. Where unconfounded associations were found, consideration was then given to the possibility that these were likely to be causal relations, by assessing the evidence using Bradford Hill’s viewpoints.7
The systematic review concluded that in the case of acute exposure to ambient particles, the evidence for a causal relation was sufficiently strong for the range of health outcomes given in table 2. Far fewer studies have examined the long term effects of air pollution; particulates have been associated with increased mortality in three American8–10 and one Dutch11 cohort studies, even after re-analysis adjusting for a wide range of socio-demographic and other potential confounders.12 The effect is likely to be causal.
For nitrogen dioxide, the evidence is much less clear. Although recent studies have generally shown associations between the ambient level of the gas and a number of cardiorespiratory outcomes, the evidence for causality is much weaker. The consensus is that the relation is probably not causal but that nitrogen dioxide is almost certainly acting as a proxy measure for another pollutant from traffic, probably fine particulates. We have therefore not presented results of health effects for reducing nitrogen dioxide.
Estimate of the health effect of a unit change in exposure
There are three elements to this: determining the shape of the relation, whether there is a threshold, and the magnitude of the effect.
Several studies have examined different models and have demonstrated a log-linear relation between average particulate concentration and relative risk.13–23 Most studies have assumed that the relation between relative risk and pollution is linear, that is relative risk or percentage change in outcome of a given increment in pollutant does not change across the range of pollutant levels examined. For brevity, the term linear is used here in this sense. Although some studies found the relation to be curvilinear,2425 these are at concentrations well in excess of those now found in the UK; the WHO advised that in the range 20–200 μg/m3 PM10, the relation is linear.26
No threshold (or at least none above about 20 μg/m3) has been found for acute effects on mortality in Madrid,27 Milan,28 Seattle,29 Birmingham (Alabama),20 London,30 or Atlanta,31 or for deaths from COPD or heart failure in Milan,28 for admissions for pneumonia or chronic obstructive airways diseases in older adults in Birmingham (Alabama),32 nor for IHD admissions in Detroit.18 Recent studies confirmed that a linear model performed best for total and cardiorespiratory mortality2333; any threshold is ≈15 μg/m23 or <5 μg/m3 PM10.33 We therefore assumed there is no threshold for the effects of particulates, in line with World Health Organisation conclusions.2634
We present models for those outcomes where the evidence for a causal relation with particulates is at least moderately strong in our view or where reports for government or international bodies have so concluded.
For each acute outcome-pollutant pair for which there was strong or moderate evidence of a causal relation, we chose the effect estimates from time-series studies that used untransformed pollution levels that were most relevant to current pollution in central London:
results of pooled analyses from APHEA studies (short term effects of air pollution on health: European approach) in Western Europe, also a primary study, as the sources and levels of pollution were similar to London and the power of the pooled study was substantially greater than for most of the single studies available. Recently, results from the APHEA-2 studies, covering a population of more than 43 million people in 29 European cities have been published.38–41
estimates based on recent meta-analyses not subject to publication bias (the combined NMMAPS (national mortality and morbidity air pollution study)—APHEA meta-analysis)42 or conducted for the World Health Organisation,2643 including those that were the basis of the 1998 COMEAP report on the burden of disease in the UK attributable to pollution.44
Meta-analyses were not used as the sole effect estimate, nor did we conduct our own, because:
many studies of ostensibly the same end points use different case definitions;
many studies use different age groups;
studies used different metrics of pollutants;
each meta-analysis should contain only results from uni-pollutant models or with the same co-pollutant; and
studies using single and cumulative lags should be analysed separately.
Although some authors have converted other measures of particulate matter into “PM10-equivalents” for their reviews or meta-analyses, we did not do this because of their well documented variability.2444–50
Where they exist, effect estimates that allow for the effects of other relevant pollutants would have been used, to permit addition of effects without double-counting. However, these were seldom available. Most studies have also found greater effects of pollution using cumulative rather than single day lag. Ideally, effect estimates that incorporate both co-pollutants and cumulative exposure would have been used but in practice, these are seldom available. Often, neither are available.
Although the three sources of effect estimates were chosen a priori, in practice, they were not available for all outcomes. Table 3 gives the range of effects estimates considered for acute effects of exposure to particulates, using the criteria above.
We also estimated the effect on all cause mortality of chronic PM10 exposure, although the paucity of good quality cohort studies and the different metrics used limited the effect estimates that could be selected. Most studies used fine particulates (PM2.5) rather than PM10 as their metric: where this was the case, we applied this effect estimate to the PM10 levels although this would be an overestimate because some of the PM10 is PM2.5–10.
the effect observed in those with more than high school education in the reanalysis of the six cities study, to eliminate as far as possible many of the potential socioeconomic confounders12: 0.3% per 1 μg/m3 PM2.5;
the all cause effect estimate from the 16 year follow up of the American Cancer Society cohort51: 0.4% per 1 μg/m3 PM2.5;
the variance weighted average of the 1995 ACS follow up and the six cities study calculated by Künzli et al in their three country study for the World Health Organisation52: 0.43% per 1 μg/m3 PM10; and
the effect estimate deemed “most likely” by COMEAP in their 2001 statement53: 0.1% per 1 μg/m3 PM2.5.
We also estimated the effects on life expectancy of long term exposure to reduced particulate levels. The best estimate is that lifelong exposure to 10 μg/m3 PM10 shortens average life expectancy by 18.6 weeks per person across the whole population (13 days per 1 μg/m3 PM10).54 This assumes that the population is exposed to that pollution from birth until death or 105 years of age and that there is no delay before effects are seen. We applied the effects to the population now alive, with the full range of ages present, and made the assumption that the effects are spread evenly over the years of exposure. We assumed a life expectancy of 75 years.
Calculating the effect of a reduction in pollution
We based our calculations for the effects of short term changes in pollution on the following formula:
ΔE = change in number of outcome events, β = exposure-response coefficient, ΔC = change in ambient concentration, and E = baseline number of outcome events.
(This assumes a log-linear exposure-response function with no threshold, and is arithmetically equivalent to the formula given by Ostro and Chestnut.55)
Change in ambient concentration
Air pollution measurements
We used data from three years (1996–1998) for the baseline for air pollutant levels because meteorological conditions can vary considerably from year to year. In particular, these change the peak levels and the number of days on which higher levels occur, even if emissions remain constant.
Continuous measurements of PM10 have been made in Westminster by a roadside tapered element oscillating microbalance (TEOM) monitor on Marylebone Road since 1995. The particulate objectives are set for gravimetric measurements. TEOM monitors under-record by about 30% in comparison with this,2 so 38.5 μg/m3[TEOM] approximates to 50 μg/m3[gravimetric]. The local authority supplied us with hourly PM10 levels for the three years, from which we calculated 24 hour mean PM10 for each day. Background PM10 TEOM measurements in Bloomsbury, near Westminster in central London, were also obtained from the London Borough of Camden.
Modelling the declines in air pollution
Irrespective of the policy instruments proposed or the feasibility of achieving the targets, there are three general ways in which the hourly or daily pollution profile could change (fig 2).
In Model A, all peaks above the standard (excluding permitted exceedences) have been eliminated. Levels on other days remain unchanged.
Model B assumes that the mean percentage decline in pollutant on the peak days, based on model A, also occurs on the remaining days.
Model C uses the percentage reduction required to eliminate the highest peak (excluding permitted exceedences) to reduce each day’s pollution.
Ambient particulates cause adverse health effects in urban areas even at current UK levels.
Reducing pollution in Westminster to national targets proposed for 2009 will delay 8–20 deaths as a result of acute exposure and about 100 per year as a result of long term exposure.
The achievable health gain from reducing pollution is far less than the total health damage attributable to pollution.
Models A and C provide lower and upper bounds, respectively; B models an intermediate condition. (Model A is the pattern that might be obtained by closing a city to motor vehicles when high pollutant levels are forecast; model C represents what might be seen in the long term, for example because of traffic reduction plus reduced transboundary pollution.) The target is not necessarily achieved by model B, as applying the average decline over all the exceedence days results in some of those days remaining above the standard but this is probably a more realistic estimate of the magnitude, if not the distribution, of the decline in pollutant. The simplest change is the reduction required to reach the target for the annual mean (model D in tables 2–4).
We also estimated the health effects of reducing PM10 levels to zero to compare with the Department of Health’s report on the burden of disease attributable to particulates.44
Baseline number of outcome events
We used health outcome data for the same three calendar years (1996–1998) for the baseline; this reduces errors attributable to random fluctuations of small numbers of events. Mortality data came from the Office for National Statistics and emergency hospital admissions from Hospital Episode Statistics. No data were available on accident and emergency or general practice consultations locally, so results were calculated as a percentage of baseline age and disease specific consultations.
Creating the spreadsheets
Modelling the declines in air pollution
For each applicable combination of air quality target and pollution reduction model for each of the three baseline years, spreadsheets calculate the magnitude of change in ambient air pollution. Technical problems with the measuring devices can cause missing data for hours or days. Ignoring these underestimates the magnitude of the decline in pollution. We therefore imputed mean values to the missing data.
For each air pollution reduction model, estimates of the number of lives no longer shortened or of emergency hospital admissions or consultations not occurring or not brought forward in a year were calculated automatically by multiplying the daily event number by the chosen effect estimate and the decline in pollutant summed over a year. In view of the uncertainties attached to both the effect estimates and the decline in pollution, we used a sensitivity analysis approach with three estimates for each where possible.
Short term effects
Attaining the 2004 24 h PM10 objective at the Westminster roadside monitoring site, using the TEOM equivalent of 38.5 μg/m3, results in one to four fewer deaths brought forward, based on the London studies, three to 21 based on the pooled APHEA-2 results from Western Europe, one to nine, based on Stieb et al’s 2002 meta-analysis, and one to 10 based on the World Health Organisation meta-analysis used by the Department of Health44(table 4). Reducing the permitted number of exceedences from 35 to the seven permitted in 2009 almost doubles the estimates. More respiratory and cardiovascular emergency admissions than deaths are delayed, because of the higher numbers of baseline events but the order of magnitude is similar (table 5). For example, about five respiratory and three to five cardiovascular admissions would be prevented or delayed by achieving the 2004 daily objective with an intermediate decline in pollution (model B). The fewer permitted exceedences for 2009 increases these to eight respiratory and six to eight cardiovascular admissions.
The annual mean PM10 levels changed less than the number and extent of the exceedences of the 24 hour standard in the three baseline years. Achieving the annual mean objective for 2004 would delay between two and four deaths, using most effect estimates (table 4). More admissions are delayed or prevented by achieving these objectives (table 5). The 2009 objective for the annual mean requires a substantial reduction in PM10, which would delay eight to 20 deaths, using the same effect estimates. About 20 respiratory and 14–20 circulatory admissions would be affected as well as around 5% of A&E attendances for asthma by attaining the lower annual mean target.
To check the models, we estimated the effects of reducing PM10 pollution in Westminster to zero, to calculate the burden of disease attributable to particulates (tables 4 and 5). Deaths attributable to PM10 ranged from 1.0% of the total, using the London exposure-effect coefficient, to 5.6%, using the distributed lag 0–40d APHEA-2 effect estimate (2.7% using the effect estimate that included nitrogen dioxide in the analysis). More than one in eleven deaths in Westminster from obstructive lung diseases in the 65–74 age group were attributable to PM10. For emergency hospital admissions, the main groups affected by particulates are also adults with obstructive airways diseases: PM10 caused almost 5% of emergency admissions in adults under 65 with asthma and up to 3.5% of admissions in those aged ⩾65 with chronic obstructive pulmonary disease or asthma.
Local authorities’ legal duty for air quality management was introduced to obtain health benefits; achievable health gain from reducing pollution is much less than the total health damage attributable to pollution.
With no threshold of effect for particulates, more stringent air quality targets yield greater health benefits.
Most air pollution in the UK comes from motor vehicles. Benefits to health and wellbeing of reducing traffic—reductions in injuries, greenhouse gas emissions, noise, and community severance from reduced car use accompanied by large positive health effects from increasing walking and cycling, are likely to be substantially greater than benefits from applying technical fixes to current travel patterns.
Long term effects
The effects on mortality of long term exposure to particulates may be an order of magnitude higher (table 5). Results using the 16 year follow up from the American Cancer Study are similar to those using Künzli and colleagues’ estimate for the long term effect of PM10 on mortality, and slightly higher than using Krewski’s reanalysis of the six cities: about 20 deaths are delayed by reaching the 2004 annual target (40 μg/m3[gravimetric]) and a hundred deaths by reducing annual mean PM10 to 20 μg/m3[gravimetric]. Calculations using the COMEAP “most likely” effect estimate are one quarter the magnitude, being closer to the acute effects of PM10, particularly when a distributed lag is used for those acute effects.
The mean reduction in annual PM10 for the roadside monitor is 4 ug/m3 for the 2004 target and 19.5 μg/m3 for the 2009 target. Ten year exposure to these lower levels would increase life expectancy by 7.3 and 34 days respectively, on average.‡
Choice of monitoring site
Particulate levels measured at the roadside in Westminster were higher than those at the background site in nearby Bloomsbury. The latter already met the 2004 annual mean target and almost met the 2004 24 hour target. To meet the 2009 targets, the decline required in pollution was one third to one half that required to achieve the same levels at the roadside in Westminster. The estimated health effects were lower by the same proportion (data not shown). Ten years exposure to the 2009 target rather than 1996–98 average levels at the Bloomsbury background site would increase average life expectancy by 19 days.§
Attributable risk and achievable health gain
The comparison of our zero pollution findings with the Department of Health report44 demonstrates two main points. Our results using the same effect estimate suggests that 2.6% of deaths in Westminster were associated with baseline pollution levels. This compares well with the estimate for the Department of Health that 1.9% of urban deaths are attributable to PM10 pollution,44 as PM10 levels are higher in central London than almost anywhere else in the UK. Similarly, PM10 in 1996–1998 accounted for 2.8% of all respiratory admissions, a little more than the report’s national estimate of 2.0%.
Secondly, the number of deaths or hospital admissions affected by the proposed reduction in air pollution is small compared with the total that is attributable to particulate air pollution. Attribution of a quantified burden of disease to a particular risk factor is quite different from quantifying the achievable health gain for a specified policy or target. This corresponds to a focus on the change in exposure rather than its level.56
We have assumed that the observed effects of pollution are reversible by reduced exposure; this has been shown in a number of locations.34
Choice of health outcome measures
Most HIA guidance advocates a holistic approach to health, including positive measures of good health.57 However, we were limited to using routine data, which are available only for disease measures, although they are not ideal outcome measures because few people are affected. As prevalence of mild disease is more common, more residents are likely to be affected by pollution exacerbating milder symptoms. While there is evidence that pollution affects lower respiratory symptoms and changes in lung function, we were not able to include these outcomes because of the lack of suitable routine data or of sufficient good quality information about general practitioner consultations.
The specific causes of mortality and morbidity were those identified by our systematic literature review as having at least moderate evidence for a causal relation with particulates. Few published meta-analyses have considered this first step,6 the 1998 COMEAP report being a notable exception.44 Omitting this crucial first step would have led us to report effects on mortality and hospital admissions of declines in levels of nitrogen dioxide. Our decision not to do so is in line with the most recent authoritative review.34
Estimation of exposure-response functions
The choice of exposure-response coefficients depends on what is available. This is subject to change as new studies are published or methods are re-evaluated. The spreadsheets are designed so that these coefficients can be replaced easily.
International collaboration has resulted in studies using identical methods in a number of cities. In addition to examining local effects, pooled analyses are planned at the start; all study sites provide relevant information, regardless of their results. These calculations provide a much more robust estimate of the true effect.
Meta-analyses also combine results from a number of studies to give a more precise estimate. Expert reports have generally put most emphasis on meta-analyses but unlike pooled analyses, they are subject to publication bias so are more likely than pooled analyses to overestimate effects. Publication bias is the most probable explanation for the higher rate found in earlier American studies than the larger, more recent ones in the USA and in Europe.
The largest study is the American national morbidity, mortality, and air pollution study (NMMAPS), based on deaths in the 90 largest cities in the USA.5859 Like many time series studies, it was analysed using generalised additive models (GAM). Investigators have found recently that when estimated regression coefficients are small and the default settings on the gam function of S-Plus software do not assure convergence of its iterative estimation procedure, the results overestimate the regression coefficients by 36%–42%.60 The effect estimate is more sensitive to the choice of method for adjustment for temporal effects than to the convergence criteria used.61 The greater the degree of concurvity (the non-parametric analogue of multicolinearity) the greater the overestimation of the regression coefficient.60 Concurvity also underestimates variance in the model and therefore the standard errors of linear predictors produced by the gam function are biased downwards, inflating type I errors.62 Stieb et al have shown a 23%–41% reduction in effect estimates from meta-analyses of primary PM10 studies not analysed with GAM compared with meta-analyses of studies using GAM analyses.63
Reanalysis of the NMMAPS data using generalised linear models (GLM) and natural cubic splines instead of GAM and smoothing splines to adjust for confounders halves the pooled estimate for non-traumatic mortality from +0.41% to +0.21% per 10 μg/m3 PM10; GAM used with stricter convergence criteria gives an estimate of 0.27%.6064 The highest relative risk, for cardiorespiratory deaths, changed much less.6165 Ramsay et al repeated Stieb et al’s meta-analysis of 31 studies of daily PM10 and mortality.66 They found a slightly lower effect estimate (0.18% instead of 0.20% per 10 μg/m3 PM10) but the variance in the original meta-analysis was underestimated by 23%.62 These effect estimates are all within the same order of magnitude as those shown in table 3(listed there per 1 μg/m3 PM10).
The effect of any reduction in the magnitude of the effect estimate once problems with GAM and with concurvity have been dealt with is to reduce the results by the same proportion. For example, the predicted number of deaths no longer delayed by reaching the annual mean target for 2009 therefore falls from 11 to 5.6 when the original is replaced with the revised estimate from NMMAPS (table 4). Corrected GAM analyses do not, however, affect the results from the pooled APHEA studies.67 This is because multisite time series are analysed using hierarchical models that estimate uncertainty in the pooled estimate using total variance.64 The WHO has reported that the unpublished meta-analysis from St George’s, London, of 26 studies that have not used GAM in analysis resulted in an effect estimate of 0.4% per 10 μg/m3 PM10, similar to the uncorrected NMMAPS result.34 The effect estimate using a 40 day distributed lag suggests that effect estimates from analyses using a few days’ lag are probably underestimates.41
Most studies of long term effects of particulates have used fine particulates not PM10 as the metric; the WHO have recently recommended that a standard for PM2.5 be established because of the increasing evidence that this fraction contains the principal harmful particles.34 Our models used PM10 not only because that is the particulate measure used for the objectives in the current and proposed UK National Air Quality Strategy but also because PM10 is now measured routinely but PM2.5 is not. The results for the long term effects of PM10 using PM2.5 is an overestimate because some of the PM10 is PM2.5–10; these coarser particles may also affect health63 but the evidence for that is less consistent.34
Another contentious issue is whether effect estimates from uni-pollutant or multi-pollutant models should be used. It is advisable to use results from multi-pollutant models when examining the effect of more than one pollutant. As our initial step indicated that the apparent effects of nitrogen dioxide are probably a proxy for effects of other traffic related pollutants and pollutants other than particulates and nitrogen dioxide are not of concern in central London, it is debatable whether it is appropriate to make such adjustments. In practice, few studies have reported such effect estimates. We have given results from the APHEA-2 distributed lag study41 for analyses including and excluding nitrogen dioxide.
Confidence intervals allow for random error but not for systematic error. The three estimates for the decline in pollution for each target (models A, B, and C), excluding annual mean, and the use of a number of exposure-response coefficients constitute built in sensitivity analyses. Analyses using pollution reductions based on each baseline year separately can be used for further sensitivity analyses (data available on request). Fortuitously, meteorological conditions resulted in particularly frequent exceedences of the PM10 daily objective in 1996 and 1997 and unusually few in 1998. We have also calculated the effects based on air pollution measurements from two different sites.
There has been much media attention given to the annual fluctuations in peak levels and number of exceedences of these two pollutants. These are reflected in differing falls in pollution required to reach the hourly or 24 hour objectives using the different years as baseline. Use of mean values does not distinguish between longer exposures to lower levels and shorter exposures to higher levels of pollution. Abbey and Burchette used a number of metrics, namely the frequency of exceedences and the excess concentration for four predetermined cut off values of TSP in the well designed California Seventh Day Adventists cohort study. Excess concentration >200 μg/m3 was the metric that correlated best with incidence of bronchitis, asthma or COPD and with change in severity of asthma.68
Validity of using fixed monitor measurements of air pollution
Ambient pollution is only a proxy for personal exposure and dose, so studies relating levels of air pollution at a fixed site monitor to health outcomes probably underestimate the effect of pollution.69 However, it is important to use the same definition of exposure as that used in the studies from which the exposure-response functions have been derived.70 There is no reason to believe that the relation between residents’ exposure and the fixed site monitor is substantially different in central London from the studies we used. Those most affected by pollution are those who are already ill, who are unlikely to be highly mobile, so their spatial relation to the fixed site monitor is likely to be more consistent than for healthy people.
Comparing PM10 levels at the two sites, hourly values, which vary more than 24 hour mean levels, were higher at the roadside monitor in Marylebone Road than the background site at Bloomsbury in 1995 but for many months, the levels varied almost in parallel.71 Since then, peak levels have fallen. In 1996 and 1998, 24 h-PM10 was >50 μg/m3[TEOM] at the Bloomsbury background site more often than at the Marylebone roadside site.72 Crossroads sites, such as the Marylebone Road roadside monitor, are midway between a severe canyon site and a background site, picking up fresh emissions from the nearest street in certain wind directions and benefiting from the ventilation of the canyon down the side roads in other wind directions.73 They are therefore not as different from background measurements as might be imagined.
Roadside levels generally represent less well most people’s exposure to pollution but PM10 and PM2.5 levels outside individuals’ homes tracked fixed site PM10 levels.74 Westminster City Council’s Air Quality Management Plan aims to bring roadside levels down to meet the standards. Background levels in Westminster would therefore fall to levels below the targets. As there is no observed threshold for effects on health and the exposure-response relation is approximately linear in this range, the effect is determined by the magnitude of the reduction not the absolute level reached. Therefore these calculations are likely to be reasonable estimates of the effects on deaths and admissions brought forward.
Magnitude of results
It is important to note that the figures we have calculated for the acute effects of pollutants are unlikely to represent a decline in annual mortality as most if not all of the affected people would have died within the year. The results shown are the number of people whose death in that year would have been brought forward had the pollution not declined.
Ostro and Chestnut argue that a proportion of these deaths would have been so premature that a change in annual mortality does occur75 but this is disputed.7677 It had been assumed that most of the people affected by air pollution are probably dying a few days, weeks, or months before they would have died without exposure to the air pollution but Zeger et al showed that the theory that deaths due to acute pollution exposure are due only to “harvesting” of the very frail is inconsistent with their analysis of the data,78 which showed that Poisson regression time-series analyses have underestimated the exposure-effect coefficient.7879 For pneumonia,80 myocardial infarction,80 and all cause mortality,7880 exposure to ambient air pollution is probably shortening life to a greater extent than was previously thought. This adds weight to our use of the larger, distributed lag effect estimates from APHEA-2.41 Additionally, the long term effects of chronic exposure to lower levels of pollution are generally held to be considerably larger than the acute effects.78 However, even over a lifetime’s exposure, the effects of these reductions on life expectancy is small compared with the average 12 years of life lost by smokers killed by tobacco.81
Impacts on inequalities in health
Those at greatest risk from air pollution are those with pre-existing cardiorespiratory disease.348283 Both cardiovascular and respiratory disease are more common in less affluent and less educated groups in the UK. In Canada, the highest prevalence of several risk factors for respiratory disease was in the area with the highest ambient pollution levels.84 In London there is a linear relation between levels of PM10 and deprivation; with the greatest predicted falls in the areas of greatest deprivation.85 Reducing air pollution will therefore decrease inequalities both because exposure will be reduced most in deprived areas and because those who will benefit most are those with worse health, the very young and older people.
The spreadsheets we have developed are designed to model the most severe impacts on residents’ health of reductions from current levels of PM10 in a local authority area to the UK National Air Quality Strategy targets. They have been piloted by three local authorities and a health authority in west London and have been used by 35 other local authorities in the UK. The same method could be applied to other pollutants.
Models A, B, and C represent patterns of decline in pollutant levels but do not correspond directly to policy options to achieve such declines. The next step will be to combine the approach used in this study with one based on actual sources and the ways in which emissions can be reduced.
This paper has considered only the direct health gains of air pollution reduction, such as would be obtained by reducing emissions using technical fixes without changing behaviour. The benefits to health and wellbeing of changing travel patterns, with reductions in injuries, greenhouse gas emissions, noise and community severance from reduced car use8687 accompanied by the large positive health effects from increasing walking88 and cycling,89 have the potential to increase much further the health benefits of air quality management.56
We thank Guy Denington, Hugh Donohoe, Amanda Gudgeon, and Kyri Eleftheriou-vaus, Royal Borough of Kensington and Chelsea; Chris Cawley, Steve Neville, and Joy Thompson, City of Westminster; and Mark Amos, Dorothy Gregson, John H James, and Ian Stinson of Kensington and Chelsea and Westminster Health Authority for their help with this work. The spreadsheets are available to local authorities and NHS organisations from the authors.
↵* Relative risk (RR) for a 1 μg/m3 increase in pollutant =eβ where β is the regression coefficient for a 1 μg/m3 change in pollutant. Percentage increase in outcome for a 1 μg/m3 increase in pollutant =100(RR–1).
↵† 95% confidence intervals (CI) for RR for a 1 μg/m3 increase in pollutant =e(β±1.96SE), where β is the regression coefficient for a 1 μg/m3 change in pollutant and SE is the standard error of β.
↵‡ 4.2 μg/m3×13d per 1 μg/m3 PM10×10/75=7.3d; 19.5 μg/m3×13d per 1 μg/m3 PM10×10/75=34d
↵§ 11 μg/m3×13d per 1 μg/m3 PM10×10/75=19d