Atmospheric Environment 43 (2009) 142–152
Contents lists available at ScienceDirect
Atmospheric Environment journal homepage: www.elsevier.com/locate/atmosenv
Air pollution and mortality: A history H.R. Anderson* Division of Community Health Sciences, St George’s, University of London, Cranmer Terrace, London SW17 0RE, United Kingdom
a b s t r a c t Keywords: Air pollution Mortality History
Mortality is the most important health effect of ambient air pollution and has been studied the longest. The earliest evidence relates to fog episodes but with the development of more precise methods of investigation it is still possible to discern short-term temporal associations with daily mortality at the historically low levels of air pollution that now exist in most developed countries. Another early observation was that mortality was higher in more polluted areas. This has been confirmed by modern cohort studies that account for other potential explanations for such associations. There does not appear to be a threshold of effect within the ambient range of concentrations. Advances in the understanding of air pollution and mortality have been driven by the combined development of methods and biomedical concepts. The most influential methodological developments have been in time-series techniques and the establishment of large cohort studies, both of which are underpinned by advances in data processing and statistical analysis. On the biomedical side two important developments can be identified. One has been the application of the concept of multifactorial disease causation to explaining how air pollution may affect mortality at low levels and why thresholds are not obvious at the population level. The other has been an increasing understanding of how air pollution may plausibly have pathophysiological effects that are remote from the lung interface with ambient air. Together, these advances have had a profound influence on policies to protect public health. Throughout the history of air pollution epidemiology, mortality studies have been central and this will continue because of the widespread availability of mortality data on a large population scale and the weight that mortality carries in estimating impacts for policy development. Ó 2008 Elsevier Ltd. All rights reserved.
1. Introduction Mortality is probably the earliest health outcome to be investigated by mathematical methods. Using parish records it was observed in the 17th century that the relationship between age and mortality could be described by an exponential function and this gave rise to the idea that human states could be modelled mathematically much as had been done successfully for physical phenomena. Modern environmental epidemiology grew out of a confluence of routine registration of mortality, measurement of environmental factors and appropriate statistical theory (Lilienfeld, 1980). Routine recording of mortality in England dates back to the parish Bills of Mortality, which were first introduced in London in 1532 largely as a response to fear of the plague. Bills of Mortality were the basis of one of the first quantitative analyses of mortality (Graunt, 1939). By the middle of the 1800s, following the nationwide introduction of civil registration and the census, it became
* Tel.: þ44 20 8725 5419; fax: þ44 20 8725 3584. E-mail address:
[email protected] 1352-2310/$ – see front matter Ó 2008 Elsevier Ltd. All rights reserved. doi:10.1016/j.atmosenv.2008.09.026
possible to analyse mortality rates. The pioneer of this method was William Farr, the first Chief Statistician of England’s General Register Office. Drawing on actuarial techniques he analysed mortality with the purpose of understanding environmental causes of mortality (Langmuir, 1976). One example was the finding in 1848–49 of an inverse relationship between elevation above the high water mark of the Thames river and the risk of death from cholera (this fitted the current causal theory of miasma). Until the middle of the 20th century when suitable methods of exposure measurement and analysis became available, the study of air pollution and mortality was limited to observations of increased mortality during fogs. Mortality is now recorded for civil and legal purposes in most countries and remains the cornerstone of public health statistics. It has had an especially profound influence on the growth in our understanding of the health effects of ambient air pollution and remains important not only as an outcome in epidemiological investigations, but for the setting of standards and guidelines (WHO, 2006), health impact assessment (Cohen et al., 2004) and cost–benefit analysis of policy options (DEFRA, 2007a; National Research Council, 2002). Environmental epidemiology aims to understand the relationship between environment and health in human populations, the
H.R. Anderson / Atmospheric Environment 43 (2009) 142–152
Air pollution episodes have no formal definition but usually refer to obviously high concentrations of air pollution lasting over several days or weeks (Anderson, 1999). Winter episodes occur during anticyclonic weather conditions associated with temperature inversions
Greater London
3 18 80 18 82 18 91 18 92 19 48 19 52 19 56 19 57 19 59 19 62 19 75 19 91
2.1. Air pollution episodes
London Administrative County 4500 4000 3500 3000 2500 2000 1500 1000 500 0
18 7
2. Evidence from studies of temporal variability
in which wind speeds are low and emitted pollutants cannot be adequately dispersed. Air pollution episodes have occurred in cities such as London for many centuries and have long been known to be associated with increases in mortality as recorded by the Bills of Mortality which showed increases in mortality during ‘‘stinking fogs’’ in the 17th century in London (Brimblecombe, 1987). In England, increased deaths associated with episodes became easier to identify following the introduction in 1837 of death registration. Some of the episodes known to be associated with an increase in mortality in London since then are summarized in Fig. 1. It is surprising however that until the 1930s, it was generally believed that the increases in mortality observed during episodes were mainly due to the, sometimes, extremely cold weather that accompanies winter temperature inversions, rather than to the accompanying air pollution (Russell, 1926). This position was understandable because at that time, while temperature was clearly associated with mortality, available statistical techniques were insufficient to identify the smaller independent effects of air pollution against the background variability of daily mortality. An air pollution episode which occurred in 1930 in a small industrial town in the Meuse Valley, Belgium is an important landmark in air pollution epidemiology (Firket, 1931, 1936). A large proportion of the population became acutely affected with respiratory symptoms and in those with chronic cardiorespiratory problems there was a worsening of their clinical state. There was 10 times the expected number of deaths (60 vs 6). Autopsies showed acute irritation of the respiratory tract. A similar episode occurred in Donora PA (population 13,000) in 1948 (Schrenk et al., 1949) in which there was a 6-fold increase in deaths. As in the Belgian episode, it occurred during a temperature inversion in a small industrial town situated in a river valley. These two episodes led to widespread acceptance that air pollution, at least in very high concentrations, could increase mortality short-term independently of the low temperature associated with winter stagnation episodes. The mechanism was thought to be a toxic inflammation of the airways. It is notable that in both of these episodes there were reports of deaths in apparently healthy animals. The effects observed in these episodes could not plausibly be attributed to weather conditions alone. The Belgian and US episodes affected small populations and the number of excess deaths was small. Further, both had a strong industrial element suggesting specific toxic chemicals might have played a role. In a remarkable prediction, Firket, in his paper on the Meuse episode commented ‘‘A Londres, ou la mortalite´ journalie`re moyenne est de 151.38, on aurait a deplorer 3.179 mort imme´diates’’ (Firket, 1931). The risk to public health of air pollution episodes resulting from non-industrial sources was finally established without doubt when London experienced an intense episode from 5 to 9 December 1952. A prolonged temperature inversion with little or no wind had enabled the build up of pollutants emitted mainly from the burning of coal in domestic grates. Smoke
Number of deaths
ultimate objective being to use this knowledge to improve public health. Conceptually, the epidemiological approach comprises two stages. The first is to establish using statistical methods whether or not the hazard is associated with health after accounting for chance and known sources of bias or confounding. The second stage is causal inference but this is problematic because most epidemiological evidence is of an observational rather than experimental nature. Causal judgment brings in non-epidemiological evidence, the most important being that from clinical and laboratory experiments. The epidemiology of air pollution and mortality is further constrained by the necessity of opportunistically using data about mortality and air pollution that have been collected for some other purpose. The investigation of associations requires data on exposure and outcome that are heterogeneous. For example, if air pollution data did not vary from day to day or from place to place, it would not be possible to investigate the association between pollution and health outcomes using epidemiological methods. There are two main types of epidemiological approach to the study of air pollution and mortality; studies of short-term temporal variability, also called short-term exposure studies, which include studies of air pollution episodes and of daily mortality; and studies of spatial variability, also called long term exposure studies, which compare mortality across areas varying in air pollution concentrations. The World Health Organization (WHO) has periodically reviewed the evidence on air pollution and health and recommended guidelines that it considered would adequately protect public health. In the late 1970s, WHO concluded that the level of long term exposure to suspended particulate matter above which health effects might be observed was around 150 mg m3 (WHO, 1979). After applying a ‘‘safety factor’’ of 2, a possible health based guideline from 60 to 90 mg m3 was suggested. In 1987 using these principles, WHO published the first global guidelines for ambient particulates, sulphur dioxide, nitrogen dioxide and ozone (WHO, 1987). The guidelines for particulate matter (measured as black smoke with the reflectance method) and sulphur dioxide were linked. The recommended long term (annual) guidelines were 50 mg m3 and 50 mg m3 respectively, while the 24 h guidelines were 125 mg m3 and 125 mg m3 respectively. The 24 h guideline for thoracic particles (the nearest equivalent to PM10) was 70 mg m3. Less than 20 years later, WHO published updated separate guidelines for particulate matter and sulphur dioxide (WHO, 2006). The annual and 24 h guidelines for PM10 were lowered to 20 and 50 mg m3 respectively. The 24 h sulphur dioxide guideline was reduced to 20 mg m3 with no annual guideline. These reductions of more than 50% were based not on a lowest observed effect to which a safety factor had been applied, as in the 1987 guidelines, but on evidence that health effects were observable across the range of ambient exposures without a discernable threshold. The most important body of evidence supporting this radical change in concept came from epidemiological studies of mortality. This paper traces the historical development of our understanding of the relationship between ambient air pollution and mortality and attempts to identify the most important conceptual and methodological factors responsible for the marked reduction in the levels of ambient air pollution believed to be harmful to human populations.
143
Fig. 1. Estimated excess deaths in London fogs 1873–1991. Assembled from Brimblecombe (1987) and other sources.
144
H.R. Anderson / Atmospheric Environment 43 (2009) 142–152
concentrations measured by the reflectance black smoke method exceeded 4000 mg m3, with an average level of about 1600 mg m3 at the County Hall over the four days. These concentrations were about 3–5 times normal for this period. This was the first major episode in a large city to be investigated intensively and the ensuing report estimated that in Greater London there had been an excess of 4000 deaths in the three weeks following the onset of the episode (Ministry of Health, 1954). This was remarkably close to the prediction of 3100 by Firket based on the Meuse episode. Mere inspection of the counts of deaths is sufficient to conclude that an epidemic of deaths was associated with the episode (Fig. 2). Statistically, this was not in itself sufficient to clinch the causality of air pollution because the weather was exceptionally cold at the same time and there was the additional possibility of a coincident event such as an influenza epidemic. Influenza was not however prevalent at the time. The contribution of low temperature was indicated by comparing events within London with those in other ‘‘Great Towns’’ which were also experiencing cold weather, but with much less pollution (Fig. 2). The latter showed a not inconsiderable increase of 15% in mortality during the week of the London episode, but this was dwarfed by the increase of over 400%
in London strongly suggesting that additional factors were at work, of which the most likely candidate was air pollution. Interestingly, as in the Meuse and Donora episodes, deaths of healthy farm animals (in the city at that time for an agricultural show) were also reported. One feature of Fig. 2 is that the epidemic appeared to be biphasic with a second more prolonged increase extending from the end of December to mid-February amounting to a further 8000 deaths above those predicted from earlier years. The possible causes of this second wave were considered to be any or all of: a delayed effect of the air pollution episode; an effect of a further increase in pollution in January and February; and the coincidence of an influenza epidemic (Wilkins, 1954). A recent re-examination of the episode has however cast doubt on whether the second wave could be blamed entirely on influenza (Bell and Davis, 2001). Further winter episodes associated with increases in mortality occurred in London (Fig. 1), as well as in some other large cities such as New York, until the 1990s (Anderson, 1999). Air pollution concentrations in London and most other major western cities have fallen markedly to historically low levels over the last century. Since 1922, when black smoke began to be monitored, concentrations have fallen from around 400 mg m3 to
Fig. 2. London October 1952–March 1953. (a) Weekly number of deaths in Greater London compared with those in 140 Great Towns, and with the corresponding period in 1951– 1952. (b) Mean values of sulphur dioxide compared with 1951–1952.
H.R. Anderson / Atmospheric Environment 43 (2009) 142–152
less than 10 mg m3 (Air Quality Expert Group, 2005). In London, the last winter air pollution episode to cause major public health concern was in 1991. The city experienced a winter temperature inversion with air stagnation lasting several days typical of the conditions previously associated with air pollution episodes. On this occasion, the pollutants that accumulated were not those from domestic fuel burning as in 1952, but from mobile sources with a contribution from space heating using natural gas. The result was a 5-fold increase in nitrogen dioxide (double the prevailing WHO guidelines – with an hourly peak of 423 ppb), 3.5fold increase in black smoke and 2-fold increase in sulphur dioxide. Although the relative increase in air pollution was quite similar to that observed in the 1952 episode the absolute levels were very much lower. Inspection of time-series of daily mortality revealed only the slightest discernible increase amongst the other daily variations (Fig. 3). Nevertheless, analysis involving control years and control areas revealed a 10% increase in mortality attributable to air pollution (Anderson et al., 1995). This compares with a greater than 400% increase in mortality observed during the week of the 1952 fog. This episode occurred during a period when emissions of oxides of nitrogen had reached their highest levels in the UK. Now that nitrogen oxide emissions have fallen due to exhaust treatment and other measures, it is most unlikely that such an episode could be repeated even under similar weather conditions. Furthermore, under climate change scenarios, cold weather with low wind speeds is predicted to become less common (Department of Health, 2008). Conversely, global warming will increase the number of hot sunny summer days and, depending on the presence of precursor pollutants including oxides of nitrogen, may increase the risk of ozone episodes which are also associated with increased mortality. In the recent epidemic of deaths associated with the 2003 European heat wave, it was estimated that ozone contributed from 2.5% to 85.3% of excess mortality across 9 French cities (Filleul et al., 2006). Other ongoing sources of episodes causing a potential increase in mortality are those due to biomass burning and dust storms, both of which may increase as a result of climate change. 2.2. Daily time-series studies Although it was originally believed that the important effects of air pollution occurred mainly during episodes, it was conjectured that air pollution might also have effects on mortality at lower
All cause mortality Max hourly average NO2
300
concentrations. In London, for example, while it was observed in the 1950s that air pollution was correlated with mortality during non-episode periods, limitations of statistical methods for controlling for confounding and serial correlation and of data processing prevented analyses that would convincingly separate the effects of air pollution from those of confounding factors such as temperature (Martin and Bradley, 1960). This question was addressed by the application of time-series techniques to daily time series of mortality and air pollution which rely on sufficient heterogeneity between mortality and other factors to separate the effects of air pollution from those of other factors such as weather, day of the week, influenza epidemics, season and long term trends. One of the earliest time-series studies used London data for 1958– 1972 and while this observed an association between Black Smoke and daily mortality it was concluded that the relative effects were greater at higher than lower concentrations, this being compatible with a no-observable-effects level (Mazumdar et al., 1982). Subsequently, Ostro (1984), using the same London data, was the first to show that there was no evidence of a no-observable-effects level and this was confirmed by Schwartz and Marcus (1990) with the same data but using different statistical methods. Time-series studies of morbidity outcomes such as hospital admissions followed quickly (Samet et al., 1981). The relative ease and low cost with which existing mortality and air pollution data can be obtained and analysed facilitated a rapid increase in single city studies during the 1990s. This trend reversed around the turn of the century partly as a result of editorial resistance (Samet, 2002) (Fig. 4). The consistency of positive results has been remarkable (Pope and Dockery, 2006). As an example, Fig. 5 shows the estimates and 95% confidence intervals extracted from peer reviewed papers up to 2006 reporting associations between PM10 and daily cardiovascular mortality (Department of Health Committee on Medical Effects of Air Pollution, 2006). Nearly all the estimates are in a positive direction and the majority are statistically significant. The summary estimate for all cardiovascular mortality is nearly 1% for a 10 mg m3 increase in PM10. Estimates for subgroups such as ischaemic heart disease mortality also tend to be positive. The evidence for other particle metrics, ozone, nitrogen dioxide and carbon monoxide is similar. As daily time-series studies grew in number, it became apparent that although there was a tendency for estimates to be positive, the effects were small and heterogeneous between studies. The reasons for heterogeneity were postulated to include variations in exposure measurement error, toxicity of the pollution and vulnerability of the population, but the relative contributions of these factors were
450
35
400 30
300 250
150 200 150
100
100
No of publications
350
ppb NO2
Numbers of deaths
250 200
145
Multicity studies Single city studies
25 20 15 10
50 0 2 19 9
2
0
24 /
01 /
19 9
1 10 /
01 /
19 9
1 27 /
12 /
19 9
1 13 /
12 /
19 9
1 11 /
19 9 29 /
11 / 15 /
01 /
11 /
19 9
1
0
5
19 8 19 2 8 19 3 8 19 4 8 19 6 8 19 6 8 19 7 8 19 8 8 19 9 90 19 9 19 1 9 19 2 9 19 3 9 19 4 9 19 5 9 19 6 97 19 9 19 8 9 20 9 0 20 0 0 20 1 0 20 2 03 20 0 20 4 0 20 5 06 20 07
50
Fig. 3. London November 1991–January 1992. Daily mortality and maximum hourly average nitrogen dioxide concentrations.
Fig. 4. Annual numbers of peer-reviewed publications with estimates of effect of air pollution on daily mortality (Courtesy of Air Pollution Epidemiology Database, St George’s, University of London.).
146
H.R. Anderson / Atmospheric Environment 43 (2009) 142–152
Cardiovascular mortality and PM10 heart failure, all, Netherlands, Hoek, 2001 dysrhythmias, all, Netherlands, Hoek, 2001 embolism + thrombosis, all, Netherlands, Hoek, 2001 ihd, all, Hong Kong, Wong, 2002 ihd, all, Montreal1, Goldberg, 2001 ihd, all, Montreal, Goldberg, 2001 ihd, all, Netherlands, Hoek, 2001 ami, all, 10 US Cities, Braga, 2001 circulatory, all, Birmingham, UK, Wordley, 1997 circulatory, all, Cook County, Illinois, Ito , 1996 cerebrovascular, all, Maricopa, Moolgavkar, 2000 cerebrovascular, all, Hong Kong, Wong, 2002 cerebrovascular, all, Seoul, Hong, 2002 cerebrovascular, all, Cook, Moolgavkar, 2000 cerebrovascular, all, Netherlands, Hoek, 2001 cerebrovascular, all, Los Angeles, Moolgavkar, 2000 cardiovascular, all, Ogden, Pope , 1999 cardiovascular, all, Palermo, Biggeri, 2001 cardiovascular, all, Huelva, Daponte , 1999 cardiovascular, all, Le Havre, Zeghnoun, 2001 cardiovascular, all, Strasbourg, Zeghnoun, 2001 cardiovascular, all, Mexico City, Castillejos, 2000 cardiovascular, all, Phoenix, Mar, 2000 cardiovascular, all, Utah Valley, Pope III, 1996 cardiovascular, all, Utah County, Pope , 1992 cardiovascular, all, Rome, Biggeri, 2001 cardiovascular, all, Santa Clara County, Fairley, 1999 cardiovascular, all, Provo/Orem, Pope , 1999 cardiovascular, all, Coachella Valley, Ostro, 1999 cardiovascular, all, Bangkok, Ostro, 1999 cardiovascular, all, Florence, Biggeri, 2001 cardiovascular, all, Inchon1, Hong, 1999 cardiovascular, all, Wayne County, Lippmann, 2000 cardiovascular, all, Bologna, Biggeri, 2001 cardiovascular, all, Coachella Valley, Ostro, 2000 cardiovascular, all, 3 Spanish Cities, Ballester, 2002 cardiovascular, all, Rouen, Zeghnoun, 2001 cardiovascular, all, Madrid, Galan, 1999 cardiovascular, all, Montreal, Goldberg, 2001 cardiovascular, all, Paris, Zeghnoun, 2001 cardiovascular, all, Salt Lake City, Pope, 1999 cardiovascular, all, Erfurt, Wichmann, 2000 cardiovascular, all, Santiago, Ostro, 1996 cardiovascular, all, Inchon, Hong, 1999 cardiovascular, all, Turin, Biggeri, 2001 cardiovascular, 65+, Krakow, Szafraniec, 1999 cardiovascular, 65+, Sao Paulo, Gouveia, 2000 cardiovascular, all, London, Bremner , 1999 cardiovascular, all, West Midlands, Anderson, 2001 cardiovascular, all, Milan, Biggeri, 2001 cardiovascular, all, Hong Kong, Wong, 2001 cardiovascular, all, Hong Kong, Wong, 2002 cardiovascular, all, Santiago, Sanhueza, 1999 cardiovascular, all, Netherlands, Hoek, 2000 cardiovascular, all, Netherlands, Hoek, 2001 cardiovascular, 65+, Helsinki, Ponka , 1998 cardiovascular, all, Melbourne, Simpson, 2000 cardiovascular, all, Seville, Ocana-Riola, 1999 cardiovascular random effects estimate cardiac, all, Maricopa, Moolgavkar, 2000 cardiac, all, Birmingham, Alabama, Schwartz, 1993 cardiac, all, Buffalo, Gwynn, 2000 cardiac, all, Los Angeles, Moolgavkar, 2000 cardiac, all, Lyon, Zmirou, 1996 cardiac, all, 10 US Cities, Braga, 2001 cardiac, all, Cook, Moolgavkar, 2000
-4
-2
0
2
4
6
8
Percentage change 10 unit increase Fig. 5. Relative risks (95% confidence intervals) for associations between PM10 and daily mortality for cardiovascular disease (Department of Health Committee on Medical Effects of Air Pollution, 2006, reprinted with permission).
(and still are) poorly understood. Heterogeneity was also likely to have been affected by differences in the way the exposure and outcome series had been assembled, analysed and published. An important aspect of this was choice of the lag. Mortality on a particular day is likely to be affected by air pollution levels over a number of prior days but because there was insufficient knowledge about the mechanism connecting exposure to death, it was not possible to postulate a lag a priori. It is now recognized that a distributed lag model which captures the combined effect of all lags is the best approach to this problem (Schwartz, 2000). However, most early studies reported the results of single day lags and tended to highlight the ‘‘best’’ one in a positive direction. While this might have been the closest to the underlying reality, it could also have been influenced by random variation with a consequent risk of bias, depending on whether only positive lags were selected. It is likely that this practice, together with publication bias in favour of studies with positive results resulted in estimates that could be inflated by as much as a factor of 2 (Anderson et al., 2005).
The statistical approach to analyzing time series of mortality has undergone important development since the 1950s (Schwartz, 1994; Katsouyanni et al., 1996; Health Effects Institute, 1995, 1997; Dominici et al., 2004; Touloumi et al., 2004, 2006; Bell et al., 2004; Peng et al., 2006). Over the past two decades the most popular statistical approach has been Poisson regression (outcome is a daily count, often small in number) with air pollution variables included as a linear predictors and controlling for time-varying confounders which are factors that are potentially related both to mortality and pollution. These are long term time trends, season and weather. A number of techniques have been to used to model trends and seasonality in mortality counts, all based upon some function of time. These functions have included simple dichotomous variables to indicate month of study, parametric cyclical functions (sine and cosine terms) to model seasonal fluctuations and more sophisticated parametric and non-parametric techniques such as regression and penalized spline smoothing or local smoothing based on ‘‘loess’’. Weather factors are connected with
H.R. Anderson / Atmospheric Environment 43 (2009) 142–152
conditions affecting the dispersion, transport and production of pollutants and some, such as temperature, are relatively stronger risk factors for mortality than is air pollution. Especially with daily data from large cities or multiple cities, regression techniques have the potential to identify statistically significant relative risks that are very small, typically less than 1.01% (1% increase) for 10 mg m3 PM10. As experience with regression techniques increased it quickly became apparent that the results might be sensitive to the analytic model, especially the way in which time-varying confounders were controlled (Moolgavkar et al., 1995). These and other unresolved issues, such as how to measure the independent effects of individual pollutants, challenged the scientific credibility and regulatory relevance of time-series evidence. The question of sensitivity to the statistical approach and the specification of weather variables was formally addressed by a series of studies which compared different modelling approaches applied to the same data sets (e.g. Health Effects Institute, 1995, 1997). The general conclusion of these studies was that the results of different methods were in reasonable agreement and were sufficiently robust to support the case for an association between ambient particles and daily mortality, at least for policy purposes. By the late 1990s, the method of choice became Poisson regression using a generalized additive model (GAM) and nonparametric smoothing of time-varying confounders but this was dramatically called into question when it was discovered that problems with the default convergence criteria used in the Splus GAM function could bias estimates upwards. Around the same time, it was recognized that this method could also underestimate standard errors. These findings invited further attacks on time-series results and to clarify the situation, a formal comparison method with improved methods using the multicity studies from Europe (APHEA 2), the US (NMMAPS) and Canada was conducted (Health Effects Institute, 2003). For NMMAPS, this led to a substantial revision downwards of the PM10 estimate from 0.44% to 0.22% per 10 mg m3. The effect of reanalysis on estimates from other studies was however very small: for the APHEA 2 cities, for example it resulted in a combined estimate of 0.59%, only 4% less than the previous estimate of 0.62%. Other approaches to the analysis of timeseries data, based on case-series methods have also been tried and have obtained results that are similar to those using smooth function of time to control for seasonality (Neas et al., 1999). Although progressive refinement of statistical methods has generally led to smaller estimates than observed in early analyses this has not seriously undermined the consensus that air pollution does have measurable associations with daily mortality that cannot be readily explained by other factors. Recent work using simulated data and multicity mortality data suggests that there is no single model that will both reduce bias and maximize prediction and that in sensitivity analyses, the range of model choice may result in estimates with a 2-fold range. Nevertheless the lower estimates still exclude zero risk (Peng et al., 2006; Touloumi et al., 2006). In order to address issues of bias, small effect size and heterogeneity, and to consolidate the evidence generally, the concept of the planned multicity study was developed. The first of these was APHEA (Air Pollution and Health, a European Approach) which involved 15 and 29 European cities in APHEA 1 (Katsouyanni et al., 1995) and APHEA 2 (Katsouyanni et al., 2001) respectively and this was followed by the US NMMAPS (National Mortality and Morbidity Air Pollution Study) (Samet et al., 2000). Up to mid-2008, there were 66 peer reviewed papers based on about 15 studies of from 2 to 92 cities. The results of multicity studies have largely confirmed those
147
of single city studies and enjoy a greater scientific and policy status on account of their standardized approaches. They also have the advantage of being more amenable to the investigation of heterogeneity (Katsouyanni et al., 2001; Bell and Dominici, 2008). Despite general acceptance that the association between air pollution and daily mortality may be causal, the usefulness of the approach, especially for policy began to be questioned (McMichael et al., 1998). A particular concern was that time series coefficients could not be used convincingly to estimate the reduction in life expectancy, a point that has been elaborated subsequently by others (Kunzli et al., 2001; Rabl, 2003). More recently, theoretical arguments have been put forward which integrate time-series and cohort results and which raise the interesting possibility that contrary to earlier assumptions, time-series studies may have the potential to estimate effects on life expectancy (Burnett et al., 2003). In spite of these unresolved questions, time series studies of mortality continue to be applied for scientific and policy purposes. One reason is that many countries, including some with highly polluted cities, are demanding local evidence rather than relying on that from western cities and daily mortality studies are the only feasible method of obtaining this in a reasonable time. Another is that time-series studies are increasingly being used to study the relative toxicity of components of the air pollution mixture with the ultimate aim of informing targeted control policies (Anderson et al., 2001; Laden et al., 2000). Yet another is the investigation of trends in time-series effect estimates over time as a method for the environmental tracking of toxicity and for evaluating trends in health impact (Burnett et al., 2005; Dominici et al., 2007). 2.3. Intervention studies Empirical evidence that mortality is reduced when concentrations of air pollutants fall would provide important evidence for the causality of observational associations reported by temporal and spatial studies. Such evidence, which is termed ‘‘accountability’’ in the US, also provides evidence that regulatory interventions result in health benefits (Health Effects Institute Accountability Working Group, 2003; National Research Council, 2002). Mostly, annual trends in air quality occur gradually and it is difficult, if not impossible, to separate the effects of changes in air pollution from those of probably more important secular trends in other factors affecting mortality. For this reason, studies in which decrements in mortality can be related temporally to the rapid introduction of abatement measures are of great importance. Two important examples are the evaluation of the effect on mortality of a ban on the burning of coal in Dublin (Clancy et al., 2002) and of the overnight introduction of low sulphur fuel in 1990 in Hong Kong (Hedley et al., 2002). Both of these studies observed reductions in mortality and in the case of Hong Kong there was some evidence of a long term increase in life expectancy. It is notable that in spite of their importance, these reports were published more than a decade after the actual intervention. 3. Studies of spatial variability It had been observed for many centuries that cities tended to be relatively unhealthy environments, with crowding and lack of hygiene promoting mortality from respiratory and enteric diseases and with the availability of vital statistics this was confirmed. Air pollution was another urban hazard of concern but was difficult to link convincingly to increased mortality because associations were likely to be confounded by factors linked both to living in a more polluted area and to increased mortality. Smoking, crowding, poverty, occupational hazards and selective migration are examples. The inferring of associations at an individual level from those observed at a group level is known as the ‘‘ecological fallacy’’
148
H.R. Anderson / Atmospheric Environment 43 (2009) 142–152
(Morgenstern, 1998). This problem is inherent in air pollution studies because exposure assessment is mostly at the group rather than the individual level. In one of the first spatial studies of air pollution, an analysis of large towns of England observed a moderate to strong correlation (above r ¼ 0.5) between smoke emissions, indicated by sales of coal, and deaths from bronchitis, pneumonia, respiratory tuberculosis, lung cancer and other causes (Daly, 1959). However, there were also positive associations between these causes of mortality and lower social class, overcrowding, population density and lower education and Daly recognized that these factors were all likely to correlate with air pollution. It was found however that an association with air pollution could still be identified after adjusting statistically for these other factors. In the US similar studies were carried out with similar results but in all of these studies the potential problem of ecological confounding remained (Evans et al., 1984; Lave and Seskin, 1972). More refined approaches of this type are still reported (Elliott et al., 2007). The need for better data on exposure and confounding factors at an individual level led to the development of the cohort approach which is now regarded as the most influential, both scientifically and for policy relevance. The essence of a mortality cohort study is that it compares the incidence of mortality over time in populations with different levels of long term exposure to air pollution while controlling at the individual level for confounding factors such as smoking or educational status. Nevertheless, quantification of exposure remains a major difficulty because it remains at the group level and it is difficult, if not impossible, to assess exposure adequately over the lifetime of the cohort. Some early cohort studies of smoking and lung cancer did consider the role of the urban environment, and while concluding that it might have a small effect were concerned about the likelihood of residual confounding by smoking (Doll, 1959, 1978). The earliest cohort study to focus on air pollution compared the risk of death after 14– 16 years in a cohort of over 8000 adults living in 6 US cities. After adjustment for smoking and other risk factors, mortality in the most polluted city was 26% higher than in the least polluted city (Dockery et al., 1993). Although there was adjustment for confounding factors at the individual level, the statistical power of this study was low on account of having only 6 cities and a relatively low number of participants with which to estimate mortality risks. On the other hand the exposure assessment was study-directed, thus reducing the potential for exposure misclassification. The need for a greater number of cities and a larger sample of subjects was addressed by a study bolted on to an existing cohort set up by the American Cancer Association (ACS) to investigate the effects of smoking and other risk factors on cancer (Pope et al., 1995). The initial results were based on over half a million subjects living in 151 US Metropolitan Areas. After controlling for a range of individual level confounders, such as smoking, there was a 17% increase in mortality when the area with the lowest concentration of PM2.5 was compared with the highest. Taken together, the policy importance of these cohort results was such that there was pressure to reanalyze the data independently but when this was done the original results were confirmed (Health Effects Institute, 2000). At the time of writing the number of substantial cohort studies of adults has increased to about 10, far fewer than the number of time-series studies. In spite of this, cohort evidence has had a very strong influence on both science and policy. In part this is because cohort studies have the important advantage over time-series studies of being able to provide coefficients, which when combined with life-tables, can directly estimate the number of life-years lost. The results of the ACS study have been used by many agencies to estimate the burden of air pollution mortality (Cohen et al., 2004) and for cost–benefit analyses of policy options (Commission of the European Communities, 2007; DEFRA, 2007a).
The earliest cohort studies compared cities using exposure data from fixed site monitors. More recently, with the development of more sophisticated spatial modelling techniques, cohort studies have studied the effects of variations in chronic exposure on mortality within cities or airsheds. In this design, the analysis is based on variability in exposure on top of a city or regional background exposure that is shared by all cohort members. An early example was from Hamilton in which proximity to major roads was found to shorten life expectancy (Finkelstein et al., 2004). This was followed by an analysis of the Los Angeles sub-group of the ACS cohort (Jerrett et al., 2005). The mortality risks obtained in the latter study were substantially higher than those obtained in the main ACS analysis and the favoured explanation for this was that exposure had been more accurately estimated. The first large European cohort study of mortality and air pollution was conducted in the Netherlands and found evidence of increased risks that were to some extent compatible with those of the US ACS study, thus providing some of the first major European evidence of its type (Beelen et al., 2008). As with the ACS, this was a cohort set up to study lifestyle and cancer. Because the Netherlands as a whole is exposed to a large and fairly homogeneous regional background of fine particles, heterogeneity of exposure relied mainly on using traffic data to model spatial variability. What this and similar city or regional studies cannot show is whether the observed associations between air pollution and mortality apply also to exposure to the background concentrations. 4. Causality and mechanisms The foregoing sections have traced the development of epidemiological evidence for associations between ambient air pollution and mortality. While there remains some uncertainty about unmeasured confounding and various statistical and exposure measurement issues, it is generally accepted that these associations are unlikely to be wholly explained by chance, bias or confounding. However, since the evidence is based on observational studies using imperfect data and analytic techniques, it is unsafe to infer that the residual associations are completely explained by a causal relationship. While the epidemiology of air pollution mortality has been developing, there have been parallel developments in causal thinking, which have had an important bearing on the interpretation of this evidence. Probably the two most important have been the appreciation of the concept of multifactorial causation and the development of a framework for making causal judgments. In the 16th century Paracelsus wrote ‘‘all things are poisons, and nothing is without poison, only the dose permits something not to be poisonous’’. He was referring to the fact that the same substance may be both toxic at high doses and benign or even therapeutic at low doses. This can be generalized to the concept that many harmful environmental substances, while toxic in high doses, are safe in low doses. This concept was and is central to the control of many occupational and environmental hazards where there is dependence on the identification of a no-observable-effect level in recommending a concentration that protects health (WHO, 1987). This thinking probably governed the early view that air pollution largely affected populations at times of air pollution episodes and was not of concern at other times. It explicitly underlay the early WHO Guidelines for outdoor air pollution which relied considerably on evidence from episodes by recommending guideline values below which increases in mortality could not be observed. Similar thinking was current in the US around the same time (Ware et al., 1981). Thus, in the 1987 guidelines, a level was decided below which health effects were thought unlikely to occur (around 100 mg m3 annual mean for both smoke and sulphur dioxide) and a safety factor of 2 was applied. Beginning with the second revision of WHO guidelines for particulate matter in 2000 (WHO, 2000) and
H.R. Anderson / Atmospheric Environment 43 (2009) 142–152
continuing with the most recent (WHO, 2006) the concept of a no observed effects level was abandoned in favour of a model in which no threshold of adverse effect within the usual ambient range was assumed. Under this concept, the guideline, while set at a level that gave reasonable protection for public health, was higher than that at which effects could be observed. This shift in thinking was strongly influenced by the accumulating results of time-series studies of mortality, which tended not to observe a threshold of effect within the ambient range. An example from the APHEA study of 29 European cities is shown in Fig. 6 (Samoli et al., 2005). Equally influential has been that no threshold for cardiopulmonary mortality was observed in the large multicity ACS cohort study of chronic exposure and mortality (Pope et al., 2002) (Fig. 7). The associations between low levels of pollution and mortality and the absence of a threshold within the ambient range could not be explained by the single-cause concept that prevailed in earlier times and which, even now, underlies much lay thinking about cause. The key to understanding why small exposures can cause death and why there is a lack of threshold lies in the multifactorial causal concept (Academy of Medical Sciences Working Group, 2007). A cause of a specific disease event is defined by Rothman and Greenland (2002) as ‘‘an antecedent event, condition or characteristic that was necessary for the occurrence of that disease at the moment it occurred, given that other conditions are fixed’’. Nearly all events are caused by a combination of factors that collectively constitute ‘‘sufficient cause’’ but individually would not be sufficient or even necessary to cause the event. The causal components for a ‘‘sufficient cause’’ may vary from person to person and from population to population. It has been pointed out by Rothman and Greenland (2002) that the strength of a cause may have great public health significance while having little biological significance, depending on the strength of other component causes. As applied to air pollution and mortality, the argument is that air pollution acts as an additional stress in persons who are already in a precarious clinical situation as a result of advanced chronic disease (such as chronic obstructive lung disease) or a spell of life threatening acute disease (such as acute pneumonia) (Anderson et al., 2003). The causes of the conditions that are responsible for this vulnerability are multifactorial and air pollution adds one more factor. This is the proposed mechanism by which small concentrations of air pollution may cause one person to experience failure of a vital system and die while the vast majority of the population survive, mostly without any adverse symptoms. At a population level one reason why we do not generally observe a threshold is because the population comprises individuals with varying thresholds. These are theoretical explanations so far supported by very little direct evidence. Another consequence of multifactorial causation is that it is vulnerability due to the
Fig. 6. 29 European cities. Exposure response for PM10 and daily mortality (adapted from Samoli et al., 2005).
149
collective effect of other causal components rather than the air pollution that largely determines whether or not an individual experiences a clinical event such as death. Thus is not possible to identify, say, among deaths from acute myocardial infarction, which individuals were toppled by air pollution. The policy implications of the lack of threshold are profound since under this model the net health benefit is larger with the ‘‘exposure reduction’’ approach, in which the whole population experiences a reduction in exposure, than with approaches that concentrate on reducing exposure below an arbitrary standard in a minority of the population (DEFRA, 2007a, 2007b). The appreciation of multifactorial causation is not in itself sufficient to attribute causation to an association. In this respect, an important advance in causal thinking, which continues to be influential in the interpretation of associations between air pollution and health, arose from the need to understand the causal nature of smoking related health effects (US Department of Health Education and Welfare, 1964). This was articulated by Hill (1965) who attempted to determine ‘‘.in what circumstances we can pass from this observed association to a verdict of causation.’’ The application of this conceptual framework is preceded by establishing that the observed associations cannot be explained by chance, bias or confounding. Hill described nine ‘‘aspects’’ or ‘‘viewpoints’’ of an association that should be considered in coming to a judgment as to causality. He emphasized that the purpose of this was to come to a decision concerning control of the hazard, not to establish scientific proof. These aspects were to ‘‘help us make up our minds on the fundamental question – is there any other way of explaining the set of facts before us, is there any other answer equally, or more, likely than cause and effect’’. This statement is equivalent to the balance of probabilities concept used in civil law and is appropriate for public health action. The nine aspects were: strength of association, consistency of evidence, specificity of effect, temporality, biological gradient (exposure response), coherence, experiment and analogy. Of these, only temporality is necessary and none are sufficient. In interpreting associations between mortality and air pollution some of these viewpoints are present while others are absent or debatable. Because the process of deciding causality using this framework requires judgment (WHO Working Group, 2000), it is not surprising that a range of interpretations exists, most clearly illustrated by the different positions taken by environmental activists, governments, industries, public health authorities and individual scientists and citizens. The commonest misinterpretation of this approach is to consider that it is a process of concluding scientific proof rather than, as intended by Hill, a method of making a decision relating to public health protection. While biological plausibility based on known mechanisms is not a necessary attribute, if it is present it has a strong influence on causal attribution. Experimental studies of human subjects and of animals have provided important insights into potential mechanisms by which air pollution might increase mortality, but they do not tell us whether these mechanisms are in fact responsible. The history of air pollution epidemiology reveals an unwillingness to accept epidemiological associations that did not make sense in terms of known or presumed mechanisms. An important example of this is the delay in accepting that air pollution may cause effects beyond the lung, most importantly on the cardiovascular system. In early air pollution episodes, it was considered that death occurred as a result of a toxic bronchitis and that cardiovascular deaths were secondary to respiratory stress. This view persisted until the early 1990s (Bates, 1992). However, the report on the 1952 London fog episode clearly shows an increase in cardiovascular deaths, which, while not as great in relative terms as the number of deaths from respiratory causes, was greater in absolute terms because of a higher baseline rate (Ministry of Health, 1954). This applied also to hospital admissions, for which diagnosis is known to be more
150
H.R. Anderson / Atmospheric Environment 43 (2009) 142–152
Fig. 7. American Cancer Society cohort. Exposure response for PM2.5 and mortality by cause (Pope et al., 2002, reproduced with permission).
accurate than death certificates. Furthermore, autopsy evidence showed that most ischaemic heart disease deaths during the episode did not show evidence of associated chronic or acute respiratory disease, thus effectively excluding an effect secondary to respiratory effects. This cardiovascular dimension to mortality was ignored, probably because there was no concept to explain it. However, large numbers of daily time series studies have now observed short-term associations with cardiovascular mortality even at low levels compared with episode conditions. Similarly, the spatial analysis by Daly (1959) did not even consider cardiovascular disease, presumably because it was not considered plausible. The ACS cohort study, however, found that the major association was with cardiovascular rather than pulmonary mortality. Acceptance of a link with cardiovascular effects increased greatly following the publication, by Seaton et al. (1995), of a hypothesis to explain how small particles might cause the release of inflammatory mediators from lung cells into the systemic circulation. Some of these substances might affect haemostatic processes and thus trigger cardiac events or promote atherogenesis. This is a rapidly expanding field of research and various complex networks of mechanisms have been proposed, supported in some cases by evidence (Health Effects Institute, 2002; Brook et al., 2004; Department of Health Committee on Medical Effects of Air Pollution, 2006). Most proposed mechanisms depend on pathophysiological changes that originate in the lung but there is also some evidence that ultrafine particles may even enter the bloodstream directly. The recognition that particles may have an effect beyond the respiratory system has opened up the theoretical possibility of chronic inflammatory effects on virtually any part of the body. Another finding of the report on the 1952 London fog episode was an increase in infant mortality but as was the case for cardiovascular effects, this was ignored, presumably because it lacked plausibility. By the end of the 1970s, however, it was clear from studies of smoking in pregnancy that inhalation of combustion products could affect foetal growth and increase the risk of sudden infant death syndrome. This was thought to be due to an intrauterine mechanism and at that time there was little appreciation that postnatal exposure to second hand smoke could be harmful to the infant. It is now recognized that exposure to second hand cigarette smoke is an important risk factor for sudden infant death
syndrome and this gave plausibility to early studies associating air pollution with infant mortality (Bobak and Leon, 1992; Woodruff et al., 1997). There is now a growing literature on effects on reproductive and perinatal outcomes (WHO European Centre for Environment and Health, 2005). An important causal question that mortality studies have not satisfactorily addressed is that of the culprit pollutants and sources. By their nature epidemiological studies investigate complex mixtures from multiple local and distant sources. Occasionally the source may be clear, such as in biomass burning or dust storms, but the exposure of a population such as a city is to pollution from a range of sources both inside and outside the area and which together comprise a complex and varying mixture of directly emitted and secondary components. Those measures available to epidemiology in the early years (smoke and sulphur dioxide) were not chosen with toxicology in mind but because they indicated the main components of the emissions (coal smoke) and these were until recently the main pollutants measured in many highly polluted cities, such as in Asia (Health Effects Institute International Oversight Committee, 2004). For epidemiology it is an inconvenient fact that the main measured pollutants (particles, sulphur dioxide, nitrogen dioxide, ozone and carbon monoxide) are all associated statistically with daily mortality and that it has proved very difficult, using statistical methods, to disentangle their independent effects. The same is true for understanding the effects of respirable particles differentiated by size, chemistry and source. A recent workshop that addressed this issue concluded that while it was likely that particle toxicology varies, it is not possible to quantify this, and that from a policy perspective, all types need to be considered for regulation (WHO Regional Office for Europe, 2007). It is now appreciated that detailed information on the speciation of particles needs to be linked with feasible epidemiological study designs. 5. From science to policy In the field of air pollution epidemiology, research has mostly been driven by policy rather than pure scientific curiosity. The main research funding has been from responsible government agencies while science funding sources have tended to be more interested in basic research into questions raised by epidemiology carried out for
H.R. Anderson / Atmospheric Environment 43 (2009) 142–152
policy purposes. However, because of the skills required, much of the research has been done not by government bodies but by academic scientists in universities and institutes. Early research into the effects of air pollution on mortality concentrated on identifying whether or not an effect could be identified in the simple hope that if this were the case, then the source would be regulated so that ‘‘safe’’ levels could be achieved. However, regulation has economic and political implications that usually must be based on more than the mere fact that air pollution is a risk factor for death. At one extreme authorities in some very polluted cities are still demanding local evidence of any harm to the population. At the other extreme, more advanced administrations are demanding data on quantification of health effects that go beyond just identifying that air pollution is a hazard. They now require exposure–response relationships and estimates of health impact, such as life years lost, that can be incorporated into cost benefit based appraisals of policy options. 6. Conclusions Thomas Kuhn, in the Theory of Scientific Revolutions, postulated that major changes in scientific knowledge amount to paradigm shifts, a paradigm being a set of internally consistent theories and beliefs (Kuhn, 1962). Paradigm shifts can be attributed to an interaction between developments in methods and theoretical concepts. While this idea is often exemplified by major paradigm shifts such as from Newtonian to Einsteinian Physics, or from Galenic to modern pathophysiology, it is a useful tool for thinking about how we have moved from not regarding even high levels of pollution as hazardous, through being hazardous only in episodes, through to the current position which accepts that air pollution increases mortality at historically low levels and that there is no discernible threshold at the population level within the ambient range. Some of our advances in understanding have occurred due to improvements in the technology for estimating exposure to air pollution and our capacity to process and analyse with powerful statistical techniques the large data sets associated with mortality studies. From the biomedical perspective, the most influential developments have been the application of the concept of multifactorial causation of disease and the appreciation that air pollution may cause adverse effects beyond the lung. These various advances have had major implications for policies to protect public health. Studies of mortality have been central throughout and this will continue because of the availability of mortality data on a large scale and because of the importance of mortality in estimating health impact. Acknowledgements I wish to thank Richard Atkinson and Graziella Favarato in preparing Fig. 4 from the Air Pollution Epidemiology Database, and Mary Field-Smith for helping to prepare the manuscript. References Academy of Medical Sciences Working Group, 2007. Identifying the Environmental Causes of Disease: How Should We Decide What to Believe and When to Take Action. Academy of Medical Sciences, London. Air Quality Expert Group (AQEG), 2005. Particulate Matter in the United Kingdom. DEFRA, London. Anderson, H.R., 1999. Health effects of air pollution episodes. In: Holgate, S.T. (Ed.), Air Pollution and Health. Academic Press, London, pp. 461–482. Anderson, H.R., Atkinson, R.W., Bremner, S.A., Marston, L., 2003. Particulate air pollution and hospital admissions for cardiorespiratory diseases: are the elderly at greater risk? European Respiratory Journal Supplement 40, 39s–46s. Anderson, H.R., Atkinson, R.W., Peacock, J.L., Sweeting, M.J., Marston, L., 2005. Ambient particulate matter and health effects: publication bias in studies of short-term associations. Epidemiology 16 (2), 155–163. Anderson, H.R., Bremner, S.A., Atkinson, R.W., Harrison, R.M., Walters, S., 2001. Particulate matter and daily mortality and hospital admissions in the west midlands conurbation of the United Kingdom: associations with fine and coarse
151
particles, black smoke and sulphate. Occupational and Environmental Medicine 58 (8), 504–510. Anderson, H.R., Limb, E.S., Bland, J.M., Ponce de Leon, A., Strachan, D.P., Bower, J.S., 1995. Health effects of an air pollution episode in London, December 1991. Thorax 50 (11), 1188–1193. Bates, D.V., 1992. Health indices of the adverse effects of air pollution: the question of coherence. Environmental Research 59, 336–349. Beelen, R., Hoek, G., van den Brandt, P.A., Goldbohm, R.A., Fischer, P., Schouten, L.J., Jerrett, M., Hughes, E., Armstrong, B., Brunekreef, B., 2008. Long-term effects of traffic-related air pollution on mortality in a Dutch cohort (NLCS-AIR study). Environmental Health Perspectives 116 (2), 196–202. Bell, M.L., Davis, D.L., 2001. Reassessment of the lethal London fog of 1952: novel indicators of acute and chronic consequences of acute exposure to air pollution. Environmental Health Perspectives 109 (Suppl. 3), 389–394. Bell, M.L., Samet, J.M., Dominici, F., 2004. Time-series studies of particulate matter. Annual Reviews of Public Health 25, 247–280. Bell, M.L., Dominici, F., 2008. Effect modification by community characteristics on the short-term effects of ozone exposure and mortality in 98 US communities. American Journal of Epidemiology 167 (8), 986–997. Bobak, M., Leon, D.A., 1992. Air pollution and infant mortality in the Czech Republic, 1986–88. Lancet 340, 1010–1014. Brimblecombe, P., 1987. The Big Smoke: a History of Air Pollution in London Since Medieval Times. Methuen, London. Brook, R.D., Franklin, B., Cascio, W., Hong, Y., Howard, G., Lipsett, M., Luepker, R., Mittleman, M., Samet, J., Smith Jr., S.C., Tager, I., 2004. Air pollution and cardiovascular disease: a statement for healthcare professionals from the Expert Panel on Population and Prevention Science of the American Heart Association. Circulation 109 (21), 2655–2671. Burnett, R.T., Dewanji, A., Dominici, F., Goldberg, M.S., Cohen, A., Krewski, D., 2003. On the relationship between time-series studies, dynamic population studies, and estimating loss of life due to short-term exposure to environmental risks. Environmental.Health Perspectives 111 (9), 1170–1174. Burnett, R.T., Bartlett, S., Jessiman, B., Blagden, P., Samson, P.R., Cakmak, S., Stieb, D., Raizenne, M., Brook, J.R., Dann, T., 2005. Measuring progress in the management of ambient air quality: the case for population health. Journal of Toxicology and Environmental Health A 68 (13–14), 1289–1300. Clancy, L., Goodman, P., Sinclair, H., Dockery, D.W., 2002. Effect of air-pollution control on death rates in Dublin, Ireland: an intervention study. Lancet 360 (9341), 1210–1214. Cohen, A., Anderson, H.R., Ostro, B., Pandey, K.D., Krzyzanowski, M., Kuenzli, N., Gutschmidt, K., Pope, C.A., Romieu, I., Samet, J., Smith, K.R., 2004. Mortality impacts of urban air pollution. In: Ezzati, M. (Ed.), Comparative Quantification of Health Risks: Global and Regional Burden of Disease Attributable to Selected Major Risk Factors. WHO, Geneva, pp. 1353–1434. Commission of the European Communities, 2007. Impact Assessment: Annex to the Communication on Thematic Strategy on Air Pollution. European Union.
. Daly, C., 1959. Air pollution and causes of death. British Journal of Preventive and Social Medicine 13 (1), 14–27. DEFRA, 2007a. An Economic Analysis to Inform the Air Quality Strategy: Updated Third Report of the Interdepartmental Group on Costs and Benefits. . DEFRA, 2007b. The Air Quality Strategy for England, Scotland, Wales and Northern Ireland, vol. 1. The Stationery Office, London. Department of Health, 2008. Health Effects of Climate Change in the UK 2008. An update of the Department of Health Report 2001/2002. Department of Health, London. Department of Health Committee on Medical Effects of Air Pollution, 2006. Cardiovascular Disease and Air Pollution. Department of Health. Dockery, D.W., Pope, A.C.D., Xu, X., Spengler, J.D., Ware, J.H., Fay, M.E., Ferris Jr., B.G., Speizer, F.E., 1993. An association between air pollution and mortality in six U.S. cities. New England Journal of Medicine 329 (24), 1753–1759. Doll, R., 1978. Atmospheric pollution and lung cancer. Environmental Health Perspectives 22, 23–31. Doll, R., 1959. Smoking and lung cancer. Report to the sub-committee for the study of the Risks of Cancer from Air Pollution and the Consumption of Tobacco. Acta Unio Internationalis Contra Cancrum 15, 1283–1296. Dominici, F., Sheppard, L., Clyde, M., 2004. Health effects of air pollution: a statistical review. International Statistical Review 71, 243–276. Dominici, F., Peng, R.D., Zeger, S.L., White, R.H., Samet, J.M., 2007. Particulate air pollution and mortality in the United States: did the risks change from 1987 to 2000? American Journal of Epidemiology 166 (8), 880–888. Elliott, P., Shaddick, G., Wakefield, J.C., de, H.C., Briggs, D.J., 2007. Long-term associations of outdoor air pollution with mortality in Great Britain. Thorax 62 (12), 1088–1094. Evans, J.S., Toteson, T., Kinney, P.L., 1984. Cross-sectional mortality studies and air pollution risk assessment. Environment International 10, 55–83. Filleul, L., Cassadou, S., Medina, S., Fabres, P., Lefranc, A., Eilstein, D., Le, T.A., Pascal, L., Chardon, B., Blanchard, M., Declercq, C., Jusot, J.F., Prouvost, H., Ledrans, M., 2006. The relation between temperature, ozone, and mortality in nine French cities during the heat wave of 2003. Environmental Health Perspectives 114 (9), 1344–1347. Finkelstein, M.M., Jerrett, M., Sears, M.R., 2004. Traffic air pollution and mortality rate advancement periods. American Journal of Epidemiology 160 (2), 173–177. Firket, J., 1936. Fog along the Meuse Valley. Transactions of the Faraday Society 32, 1192–1197.
152
H.R. Anderson / Atmospheric Environment 43 (2009) 142–152
Firket, J., 1931. Sur les causes des accidents survenus dans la vallee de la Meuse, lors des brouillards de decembre 1930. Bulletin de l’scademie Royale Medicine de Belgique 11, 683. Graunt, J., 1939. Natural and political observations mentioned in a following index, and made upon the Bills of Mortality, 1662. London. Printed by Tho. Roycroft, for John Martin, James Allestry and Tho. Dicas. 1662 edn. Johns Hopkins Press, Baltimore. Health Effects Institute, 1995. Particulate Air Pollution and Daily Mortality. Replication and Validation of Selected Studies. The Phase 1 Report of the Particle Epidemiology Evaluation Project. Health Effects Institute, Cambridge, MA. Health Effects Institute, 1997. Particulate Air Pollution and Daily Mortality. Analyses of the Effects of Weather and Multiple Air Pollutants. The Phase 1B Report of the Particle Epidemiology Evaluation Project. Health Effects Institute, Cambridge, MA. Health Effects Institute, 2002. Understanding the Health Effects of Components of the Particulate Matter Mix: Progress and Next Steps. Health Effects Institute, Cambridge, MA. Health Effects Institute, 2000. Reanalysis of the Harvard Six Cities Study and the American Cancer Society Study of Particulate Air Pollution and Mortality. HEI Special Report. Health Effects Institute, Cambridge. Health Effects Institute, 2003. Revised Analyses of Time-series Studies of Air Pollution and Health. Special Report. Health Effects Institute, Boston, MA. Health Effects Institute International Oversight Committee, 2004. Health Effects of Outdoor Air Pollution in Developing Countries of Asia: a Literature Review. Special Report 15. Health Effects Institute, Boston, MA. Hedley, A.J., Wong, C.M., Thach, T.Q., Ma, S., Lam, T.H., Anderson, H.R., 2002. Cardiorespiratory and all-cause mortality after restrictions on sulphur content of fuel in Hong Kong: an intervention study. Lancet 360 (9346), 1646–1652. HEI Accountability Working Group, 2003. Assessing Health Impact of Air Quality Regulations: Concepts and Methods for Accountability Research. Communication 11. Health Effects Institute, Boston, MA. Hill, A.B., 1965. The environment and disease: association or causation? Proceedings of the Royal Society of Medicine 58, 295–300. Jerrett, M., Burnett, R.T., Ma, R., Pope III, C.A., Krewski, D., Newbold, K.B., Thurston, G., Shi, Y., Finkelstein, N., Calle, E.E., Thun, M.J., 2005. Spatial analysis of air pollution and mortality in Los Angeles. Epidemiology 16 (6), 727–736. Katsouyanni, K., Touloumi, G., Samoli, E., Gryparis, A., Le Tertre, A., Monopolis, Y., Rossi, G., Zmirou, D., Ballester, F., Boumghar, A., Anderson, H.R., Wojtyniak, B., Paldy, A., Braunstein, R., Pekkanen, J., Schindler, C., Schwartz, J., 2001. Confounding and effect modification in the short-term effects of ambient particles on total mortality: results from 29 European cities within the APHEA2 project. Epidemiology 12 (5), 521–531. Katsouyanni, K., Zmirou, D., Spix, C., Sunyer, J., Schouten, J.P., Ponka, A., Anderson, H.R., Le Moullec, Y., Wojtyniak, B., Vigotti, M.A., Bacharova, L., 1995. Short-term effects of air pollution on health: a European approach using epidemiological time-series data. The APHEA project: background, objectives, design. European Respiratory Journal 8, 1030–1038. Katsouyanni, K., Schwartz, J., Spix, C., Touloumi, G., Zmirou, D., Zanobetti, A., Wojtyniak, B., Vonk, J.M., Tobias, A., Ponka, A., Medina, S., Bacharova, L., Anderson, H.R., 1996. Short term effects of air pollution on health: a European approach using epidemiologic time series data: the APHEA protocol. Journal of Epidemiology and Community Health 50 (Suppl. 1), S12–S18. Kunzli, N., Medina, S., Kaiser, R., Quenel, P., Horak Jr., F., Studnicka, M., 2001. Assessment of deaths attributable to air pollution: should we use risk estimates based on time series or on cohort studies? American Journal of Epidemiology 153 (11), 1050–1055. Kuhn, T., 1962. The Structure of Scientific Revolutions. University of Chicago Press, Chicago. Laden, F., Neas, L.M., Dockery, D.W., Schwartz, J., 2000. Association of fine particulate matter from different sources with daily mortality in six U.S. cities. Environmental Health Perspectives 108 (10), 941–947. Langmuir, A.D., 1976. William Farr: founder of modern concepts of surveillance. International Journal of Epidemiology 5 (1), 13–18. Lave, L.B., Seskin, E.P., 1972. Air pollution, climate and home heating: their effects on US mortality rates. American Journal of Public Health 62, 909–916. Lilienfeld, A.M., 1980. Times, Places and Persons. Johns Hopkins Press, Baltimore. Martin, A.E., Bradley, W.H., 1960. Mortality, Fog and Atmospheric Pollution (Section I – General) of Bulletin Issued from the Office of the Ministry of Health. Ministry of Health, London. Mazumdar, S., Schimmel, H., Higgins, I.T., 1982. Relation of daily mortality to air pollution: an analysis of 14 London winters, 1958/59–1971/72. Archives of Environmental Health 37, 213–220. McMichael, A.J., Anderson, H.R., Brunekreef, B., Cohen, A.J., 1998. Inappropriate use of daily mortality analyses to estimate longer-term mortality effects of air pollution. International Journal of Epidemiology 27, 450–453. Ministry of Health, 1954. Mortality and Morbidity During the London Fog of December 1952. Reports on Public Health and Medical Subjects No. 95. HMSO, London. Moolgavkar, S.H., Luebeck, E.G., Hall, T.A., Anderson, E.L., 1995. Particulate air pollution, sulfur dioxide, and daily mortality: a reanalysis of the Steubenville data. Inhalation Toxicology 7, 35–44. Morgenstern, H., 1998. Ecological studies. In: Rothman, K.J., Greenland, S. (Eds.), Modern Epidemiology, second ed. Little Brown, Boston, pp. 459–480. National Research Council, 2002. Estimating the Public Health Benefits of Proposed Air Pollution Regulations. National Academies Press, Washington.
Neas, L.M., Schwartz, J., Dockery, D.,1999. A case-crossover analysis of air pollution and mortality in Philadelphia. Environmental Health Perspectives 107 (8), 629–631. Ostro, B., 1984. A search for a threshold in the relationship of air pollution to mortality: a reanalysis of data on London winters. Environmental Health Perspectives 58, 397–399. Peng, R.D., Dominici, F., Louis, T.A., 2006. Model choice in time series studies of air pollution and mortality. Journal of the Royal Statistical Society A 169 (part 2), 179–203. Pope III, C.A., Burnett, R.T., Thun, M.J., Calle, E.E., Krewski, D., Ito, K., Thurston, G.D., 2002. Lung cancer, cardiopulmonary mortality, and long-term exposure to fine particulate air pollution. Journal of the American Medical Association 287 (9), 1132–1141. Pope III, C.A., Dockery, D.W., 2006. Health effects of fine particulate air pollution: lines that connect. Journal of the Air and Waste Management Association 56 (6), 709–742. Pope III, C.A., Thun, M.J., Namboodiri, M.M., Dockery, D.W., Evans, J.S., Speizer, F.E., Heath Jr., C.W., 1995. Particulate air pollution as a predictor of mortality in a prospective study of U.S. adults. American Journal of Respiratory and Critical Care Medicine 151 (3 Pt 1), 669–674. Rabl, A., 2003. Interpretation of air pollution mortality: number of deaths or years of life lost? Journal of the Air & Waste Management Association 53, 41–50. Rothman, K.J., Greenland, S., 2002. Causation and causal inference. In: Detels, R. (Ed.), Oxford Textbook of Public Health, fourth ed. Oxford University Press, Oxford, pp. 641–653. Russell, W.T., 1926. The relative influence of fog and low temperature on the mortality from respiratory disease. Lancet 2, 1128–1130. Samet, J.M., 2002. Air pollution and Epidemiology: ‘‘deja vu all over again?’’. Epidemiology 13 (2), 118–119. Samet, J.M., Bishop, Y., Speizer, F.E., Spengler, J.D., Ferris Jr., B.G., 1981. The relationship between air pollution and emergency room visits in an industrial community. JAPCA 31, 236–241. Samet, J.M., Zeger, S.L., Dominici, F., Curriero, F., Coursac, I., Dockery, D.W., Schwartz, J., Zanobetti, A., 2000. The national morbidity, mortality and air pollution study part II: morbidity, mortality and air pollution in the United States. Health Effects Institute 94 (Part II). Samoli, E., Analitis, A., Touloumi, G., Schwartz, J., Anderson, H.R., Sunyer, J., Bisanti, L., Zmirou, D., Vonk, J.M., Pekkanen, J., Goodman, P., Paldy, A., Schindler, C., Katsouyanni, K., 2005. Estimating the exposure-response relationships between particulate matter and mortality within the APHEA multicity project. Environmental Health Perspectives 113 (1), 88–95. Schrenk, H.H., Heimann, H., Clayton, G.D., Gafafer, W.M., Wexler, H., 1949. Air pollution in Donora, PA. Public Health Bulletin No. 306. Schwartz, J., 2000. The distributed lag between air pollution and daily deaths. Epidemiology 11 (3), 320–326. Schwartz, J., Marcus, A., 1990. Mortality and air pollution in London: a time series analysis. American Journal of Epidemiology 131 (1), 185–194. Schwartz, J., 1994. Nonparametric smoothing in the analysis of air pollution and respiratory illness. The Canadian Journal of Statistics 22, 471–487. Seaton, A., MacNee, W., Donaldson, K., Godden, D., 1995. Particulate air pollution and acute health effects. Lancet 345, 176–178. Touloumi, G., Atkinson, R., Le Tertre, A., Samoli, E., Schwartz, J., Schindler, C., Vonk, J.M., Rossi, G., Saez, M., Rabczenko, D., Katsouyanni, K., 2004. Analysis of health outcome time series data in epidemiological studies. Environmetrics 15, 101–117. Touloumi, G., Samoli, E., Pipikou, M., Le, T.A., Atkinson, R., Katsouyanni, K., 2006. Seasonal confounding in air pollution and health time-series studies: effect on air pollution effect estimates. Statistics in Medicine 25 (24), 4164–4178. US Department of Health Education and Welfare, 1964. Smoking and Health: Report of the Advisory Committee to the Surgeon General of the Public Health Service. US Government Printing Office, Washington, DC. Ware, J.H., Thibodeau, L.A., Speizer, F.E., Colome, S., Ferris Jr., B.G., 1981. Assessment of the health effects of atmospheric sulfur oxides and particulate matter: evidence from observational studies. Environmental Health Perspectives 41, 255–276. WHO, 1987. Air quality guidelines for Europe. In: European Series No. 23. WHO Regional Publications, Copenhagen. WHO, 2000. Air quality guidelines for Europe, second ed.. In: European Series, No. 91 WHO Regional Office for Europe, Copenhagen. WHO European Centre for Environment and Health, 2005. Effects of Air Pollution on Children’s Health and Development – a Review of the Evidence. WHO Regional Office for Europe, Bonn. WHO Working Group, 2000. Evaluation and use of epidemiological evidence for environmental health risk assessment: WHO Guideline Document. Environmental Health Perspectives 108, 997–1002. WHO, 2006. Air Quality Guidelines: Global Update 2005, Particulate Matter, Ozone, Nitrogen Dioxide and Sulphur Dioxide. WHO Regional Office for Europe, Copenhagen. WHO, 1979. Sulphur oxides and suspended particulate matter. In: Environmental Health Criteria No. 8. WHO. WHO Regional Office for Europe, 2007. Health Relevance of Particulate Matter from Various Sources: Report on a WHO Workshop, Bonn, Germany, March 2007. WHO, Copenhagen. Wilkins, E.T., 1954. Air pollution and the London fog of December, 1952. Journal of the Royal Sanitary Institute 74, 1–21. Woodruff, T.J., Grillo, J., Schoendorf, K.C., 1997. The relationship between selected causes of postneonatal infant mortality and particulate air pollution in the United States. Environmental Health Perspectives 105 (6), 608–612.