Cell Biol Toxicol (2015) 31:131–147 DOI 10.1007/s10565-015-9296-7
REVIEW ARTICLE
The effects on health of ambient particles: time for an agonizing reappraisal? Robert L. Maynard
Received: 18 February 2015 / Accepted: 25 February 2015 / Published online: 29 April 2015 # Springer Science+Business Media Dordrecht 2015
Keywords Air pollution . Ambient particles . Nitrogen dioxide . Particulate matter
John Foster Dulles used the expression ‘agonizing reappraisal’ when reflecting on the foreign policy of the United States in 1953 (Duchin 1992); the phrase might be applied, now, to our knowledge of the effects of ambient particles on health and to policies based thereon. Nobody seriously doubts that ambient concentrations of inhalable particles have effects on health; what is open to doubt is whether these effects can be reliably attributed to specific components of these particles or, even, to inhalable particles within specific size ranges. In addition, the role of nitrogen dioxide (NO2), concentrations of which are often well correlated with those of ambient particulate matter (PM), is due for reconsideration. Modern research on the effects of air pollutants on health began with the London Smog of 1952: more than 4000 unexpected deaths occurred during and after a week of remarkably high concentrations of coal smoke and sulfur dioxide (SO2) (Ministry of Health 1954). That such conditions could affect health was known (Firket 1936; Schrenk et al 1949), but the London Smog focused attention on the problem. Legislation followed remarkably quickly in the UK and in the United States (US), and levels of pollution fell. By the 1970s, levels in London were no longer seen as dangerous to health, and R. L. Maynard (*) Birmingham University, Birmingham, UK e-mail: [email protected]
UK research faded away (Lawther et al 1970; Waller 1971). American and some European workers persisted, and by the late 1980s, it was becoming clear that effects on health could be detected at much lower concentrations of ambient PM than had been regarded as harmless in the UK. A period of reappraisal followed: the editions of the World Health Organization’s Air Quality Guidelines for Europe (AQGs) reflect this and later reappraisals (World Health Organization: 1987, 2000, 2006). Standards for ambient air quality were tightened. The evidence that led to this change in thinking came not from toxicological studies but from epidemiological work. Two types of study dominated the field: timeseries studies and cohort studies.
Time-series studies The majority of time-series studies in this field relate daily concentrations of air pollutants to daily counts of events reflecting effects on health, for example, deaths or hospital admissions. Careful adjustment for confounding factors that themselves vary on a daily basis, for example, ambient temperature, is essential (Hurley 2001). The results of several hundred such studies have been published, far too many to be considered seriatim, but the techniques of meta-analysis allow a synthesis of their findings. One such synthesis has been undertaken in the UK, where a Department of Health (DH)-funded data base of time-series studies has been established. Professor Ross Anderson published a report on this work in 2007: it runs to more than 700 pages; many of
132
which show forest plots and tables of the findings of individual studies (Anderson et al 2007). Use was made of the database in an earlier report by the DH Committee on the Medical Effects of Air Pollutants (COMEAP): Fig. 1 is taken, with permission, from this report (COMEAP 2006). The references to the individual studies, given on the y-axis, are difficult to read; that they are many is obvious. This diagram was prepared in 2005; it would be all but impossible to include all the studies now available in a legible format. It will be seen that the majority of studies show that increases in daily average concentrations of PM are associated with an increase in the risk of death from cardiovascular diseases. In addition, the majority of the lower confidence intervals do not cross the vertical zero-effects line: these are statistically significant associations. Harder to see is the open diamond which shows the results of a meta-analysis of the studies. It will be noted that the confidence intervals around the meta-analysis coefficient are relatively tight: tighter that those of the individual studies. The power of metaanalysis lies in combining the results of individual studies. Fig 1 Cardiovascular mortality and PM10. The y-axis contains the identifying data for each study in order of diagnosis, age group, city, first author and year of publication. Details of the individual studies are provided in the Appendices to the COMEAP report. Figure reproduced with permission of the UK Department of Health
Cell Biol Toxicol (2015) 31:131–147
The metric of PM used in these studies is PM10. This is the mass concentration of particles of generally less than 10 μm aerodynamic diameter. This measure of PM conforms well with the thoracic fraction of inhalable particles, defining those particles liable to deposition within the airways of the thorax: the conducting airways and gas exchange region of the lung. Smaller fractions can be defined: PM2.5, defined in the same way as PM10 but with reference to an aerodynamic diameter of 2.5 μm, is used to reflect those particles with a high probability of deposition in the gas exchange zone and as a convenient way of separating particles such as sulfate and nitrate (secondary particulate) and those produced by combustion sources (2.5 μm). The majority of time-series studies have used PM10 as a metric of ambient particles, but an increasing number of studies have used PM2.5. That PM10 includes PM2.5 will be obvious. The findings of the time-series studies are remarkable. Reference to Fig. 1 suggests that a 10 μg.m−3 increase in daily average PM10 is associated with a 0.9 % increase in daily mortality from cardiovascular diseases, a small coefficient, but one that implies a
Cell Biol Toxicol (2015) 31:131–147
large effect on public health when applied at a population scale. Reference to the original COMEAP report will show many more forest plots, yet more will be found in the meta-analysis by Anderson et al. (2007). To the toxicologist, 10 μg.m−3 is a low concentration: industrial standards for most, but not all, particles are expressed in mg.m−3. Binns et al (1978) reported the total particulate matter delivered by one cigarette as 18.5 mg. If we assume that all ambient particles monitored as PM10 are deposited in the lung (a significant overestimate), and that 20 m3 of air are inspired per day, then the daily dose of these particles would be 200 μg. How can such a low dose cause death and illness leading to hospital admissions? The studies shown in Fig. 1 involve, generally, single pollutant models. However, if two pollutant models including both PM and NO2 are employed then, very surprisingly for some, the apparent effect of PM is not notably resistant or robust to adjustment for concentrations of NO2 though the NO2 coefficients are relatively robust to adjustment for concentrations of PM (Anderson et al., 2007). This is disturbing. A great deal of effort has been put into trying to explain the effects on health of low concentrations of PM: could it be that more attention should have been focused on the oxides of nitrogen? This question remains unanswered, but we shall return to it shortly.
Cohort studies The second very strong strand of evidence of the effects of PM on health comes from cohort studies. These studies have compared the risks of death between populations living in areas with differing long-term average, ambient, concentrations of PM and some other air pollutants. The Harvard Six Cities Study and the American Cancer Society (ACS) Cohort Study, its re-analysis and its followup studies have dominated the field (Dockery et al 1993; Pope et al 1995, 2002; Health Effects Institute 2000). More recent work from Europe and Asia has added to our knowledge (Hoek et al 2013). Once again, adjustment for potential confounding factors is critically important, but here, the confounding factors differ from those of concern in time-series studies. In the latter, personal factors, which vary little from day-to-day (smoking habits, occupational
133
history, educational status, poverty) matter very little; in cohort studies, these personal factors matter a great deal. It will be realized that time-series studies can be done without personal knowledge of any of the ‘participants’; cohort studies require detailed knowledge of participating individuals. That these studies that are sound in methodological terms is accepted. Exhaustive re-analysis of the Harvard Six Cities Study and the ACS study revealed no significant flaws (Health Effects Institute 2000). Table 1 shows data from the ACS Study (Pope et al. 2002) Here, we are dealing with long-term exposure to PM2.5. Note that the size of the effect on cardiopulmonary mortality, per 10 μg.m−3 of this metric of PM, is about 10 times as great as the effect on cardiovascular disease deaths shown by the time-series studies discussed above (9 % and compared with 0.9 %). Details of the study are provided in a COMEAP report and, of course, in the original publications (COMEAP 2009). A 10 μg.m−3 increment in PM2.5 is associated with a 6 % increase in the likelihood of death from all nonaccidental causes. The effect on the risk of death from cardiopulmonary disease is greater (9 %), and reanalysis showed that the majority of this effect fell on deaths from cardiovascular rather than respiratory (pulmonary) causes (Health Effects Institute 2000). The time-series studies and the cohort studies have in common the fact that no obvious threshold of effect could be detected. Of course, no certain estimates of effects could be made for concentrations of PM lower than those recorded, but in the ACS study, effects were present at as low a concentration as 8 μg.m−3 (PM2.5). These finding are as surprising, to the toxicologist, as those of the time-series studies.
Table 1 Adjusted mortality relative risk, RR, (95 % confidence intervals) associated with a 10 μg.m−3 increase in ambient concentrations of particle monitored as PM2.5. The average RR was calculated as the average of the RRs for the two periods during which PM2.5 had been measured. The original studies should be consulted for details Cause of death
Average relative risk
All cause
1.06 (1.02, 1.11)
Cardiopulmonary
1.09 (1.03, 1.16)
Lung cancer
1.14 (1.04, 1.23)
All other causes
1.01 (0.95, 1.06)
134
Intervention studies The studies described above are observational studies; the toxicologist naturally asks whether any experimental studies have been done. It will be realized that population-based experimental studies on the effects of ambient air pollutants on health will be difficult to organize, though accidents, happy and unhappy, may be studied. Into the latter category fall all the studies of unusually severe episodes of air pollution. Sudden and sustained reductions in levels of air pollution do not, in general, occur as a result of chance: deliberate actions are necessary. Some result from sudden changes in policy, examples include the banning of the sale of coal for domestic heating in Dublin in 1990, the reduction in the sulfur content of oil in Hong Kong in 1990, the reduction of traffic in Beijing during the Olympic Games of 2008 and the traffic congestion charging scheme introduced in London in 2003 (Clancey et al 2002; Hedley et al 2002; Li et al 2010; Tonne et al 2008). Strikes leading to a cessation of industrial activity can also lead to sudden reduction in emissions of pollutants: the nationwide copper smelter strike in the United States of 1967/1968 and the closure and reopening of a steel mill in the Utah Valley, USA, in 1986 provide examples (Pope et al 2007; Pope 1989). Twenty-eight such events have been reviewed by Henschel et al. (2012); in the great majority of cases, a reduction, sometimes a marked reduction, in levels of air pollution followed the events. This reduction was followed, in general, by clear improvements of health including reductions in hospital admissions for cardiopulmonary diseases and deaths. Interestingly, as pointed out by the authors, the improvements in health exceeded those predicted by the time-series evidence reviewed above. Much of the evidence relating to the effects of PM on health has been reviewed by Pope and Dockery: no unbiased reader of their review would be left with doubts about the effect of ambient levels of PM on health (Pope and Dockery 2006).
Cell Biol Toxicol (2015) 31:131–147
disease (COMEAP 2010; Cohen et al 2005). That such effects occur in developing countries with high levels of air pollutants is, perhaps, not surprising; that effects continue to occur in the UK where levels of air pollutants in many towns and cities are probably lower than at any time in the past several hundred years, is remarkable. This large impact on public health, valued on a global scale in billions of pounds, calls for action. But action to reduce levels of air pollutants is costly and must be appropriately directed. The obvious approach is to reduce sources of pollutants but this is far from easy. Efforts to reduce traffic have been made in urban areas, and ‘low-emission zones’ are becoming popular, at least with those responsible for taking action to reduce emissions of pollutants. If the individual pollutants responsible for the effects on health could be identified, then more specific actions to reduce the emissions of these substances could be taken. But this has proved curiously difficult. Many of the major air pollutants are emitted by the same sources, and distinguishing their effects has proved very difficult indeed. Much emphasis has been placed on PM, but this has led to the possible effects of gases such as NO2 being, at least comparatively, ignored. Concentrations of particulate matter monitored as PM2.5 and nitrogen dioxide (better described as nitrogen oxides, NOx, for both nitric oxide (NO) and, NO2, should be included) are closely correlated, and distinguishing their effects has, until recently, been all but impossible. Recent developments in epidemiological methods and in the techniques of meta-analysis have eased the difficulty, and it is now clear that ambient concentrations of NO 2 do have effects on health (WHO 2013). These concentrations tend to be low,
REVIEW ARTICLE
The effects on health of ambient particles: time for an agonizing reappraisal? Robert L. Maynard
Received: 18 February 2015 / Accepted: 25 February 2015 / Published online: 29 April 2015 # Springer Science+Business Media Dordrecht 2015
Keywords Air pollution . Ambient particles . Nitrogen dioxide . Particulate matter
John Foster Dulles used the expression ‘agonizing reappraisal’ when reflecting on the foreign policy of the United States in 1953 (Duchin 1992); the phrase might be applied, now, to our knowledge of the effects of ambient particles on health and to policies based thereon. Nobody seriously doubts that ambient concentrations of inhalable particles have effects on health; what is open to doubt is whether these effects can be reliably attributed to specific components of these particles or, even, to inhalable particles within specific size ranges. In addition, the role of nitrogen dioxide (NO2), concentrations of which are often well correlated with those of ambient particulate matter (PM), is due for reconsideration. Modern research on the effects of air pollutants on health began with the London Smog of 1952: more than 4000 unexpected deaths occurred during and after a week of remarkably high concentrations of coal smoke and sulfur dioxide (SO2) (Ministry of Health 1954). That such conditions could affect health was known (Firket 1936; Schrenk et al 1949), but the London Smog focused attention on the problem. Legislation followed remarkably quickly in the UK and in the United States (US), and levels of pollution fell. By the 1970s, levels in London were no longer seen as dangerous to health, and R. L. Maynard (*) Birmingham University, Birmingham, UK e-mail: [email protected]
UK research faded away (Lawther et al 1970; Waller 1971). American and some European workers persisted, and by the late 1980s, it was becoming clear that effects on health could be detected at much lower concentrations of ambient PM than had been regarded as harmless in the UK. A period of reappraisal followed: the editions of the World Health Organization’s Air Quality Guidelines for Europe (AQGs) reflect this and later reappraisals (World Health Organization: 1987, 2000, 2006). Standards for ambient air quality were tightened. The evidence that led to this change in thinking came not from toxicological studies but from epidemiological work. Two types of study dominated the field: timeseries studies and cohort studies.
Time-series studies The majority of time-series studies in this field relate daily concentrations of air pollutants to daily counts of events reflecting effects on health, for example, deaths or hospital admissions. Careful adjustment for confounding factors that themselves vary on a daily basis, for example, ambient temperature, is essential (Hurley 2001). The results of several hundred such studies have been published, far too many to be considered seriatim, but the techniques of meta-analysis allow a synthesis of their findings. One such synthesis has been undertaken in the UK, where a Department of Health (DH)-funded data base of time-series studies has been established. Professor Ross Anderson published a report on this work in 2007: it runs to more than 700 pages; many of
132
which show forest plots and tables of the findings of individual studies (Anderson et al 2007). Use was made of the database in an earlier report by the DH Committee on the Medical Effects of Air Pollutants (COMEAP): Fig. 1 is taken, with permission, from this report (COMEAP 2006). The references to the individual studies, given on the y-axis, are difficult to read; that they are many is obvious. This diagram was prepared in 2005; it would be all but impossible to include all the studies now available in a legible format. It will be seen that the majority of studies show that increases in daily average concentrations of PM are associated with an increase in the risk of death from cardiovascular diseases. In addition, the majority of the lower confidence intervals do not cross the vertical zero-effects line: these are statistically significant associations. Harder to see is the open diamond which shows the results of a meta-analysis of the studies. It will be noted that the confidence intervals around the meta-analysis coefficient are relatively tight: tighter that those of the individual studies. The power of metaanalysis lies in combining the results of individual studies. Fig 1 Cardiovascular mortality and PM10. The y-axis contains the identifying data for each study in order of diagnosis, age group, city, first author and year of publication. Details of the individual studies are provided in the Appendices to the COMEAP report. Figure reproduced with permission of the UK Department of Health
Cell Biol Toxicol (2015) 31:131–147
The metric of PM used in these studies is PM10. This is the mass concentration of particles of generally less than 10 μm aerodynamic diameter. This measure of PM conforms well with the thoracic fraction of inhalable particles, defining those particles liable to deposition within the airways of the thorax: the conducting airways and gas exchange region of the lung. Smaller fractions can be defined: PM2.5, defined in the same way as PM10 but with reference to an aerodynamic diameter of 2.5 μm, is used to reflect those particles with a high probability of deposition in the gas exchange zone and as a convenient way of separating particles such as sulfate and nitrate (secondary particulate) and those produced by combustion sources (2.5 μm). The majority of time-series studies have used PM10 as a metric of ambient particles, but an increasing number of studies have used PM2.5. That PM10 includes PM2.5 will be obvious. The findings of the time-series studies are remarkable. Reference to Fig. 1 suggests that a 10 μg.m−3 increase in daily average PM10 is associated with a 0.9 % increase in daily mortality from cardiovascular diseases, a small coefficient, but one that implies a
Cell Biol Toxicol (2015) 31:131–147
large effect on public health when applied at a population scale. Reference to the original COMEAP report will show many more forest plots, yet more will be found in the meta-analysis by Anderson et al. (2007). To the toxicologist, 10 μg.m−3 is a low concentration: industrial standards for most, but not all, particles are expressed in mg.m−3. Binns et al (1978) reported the total particulate matter delivered by one cigarette as 18.5 mg. If we assume that all ambient particles monitored as PM10 are deposited in the lung (a significant overestimate), and that 20 m3 of air are inspired per day, then the daily dose of these particles would be 200 μg. How can such a low dose cause death and illness leading to hospital admissions? The studies shown in Fig. 1 involve, generally, single pollutant models. However, if two pollutant models including both PM and NO2 are employed then, very surprisingly for some, the apparent effect of PM is not notably resistant or robust to adjustment for concentrations of NO2 though the NO2 coefficients are relatively robust to adjustment for concentrations of PM (Anderson et al., 2007). This is disturbing. A great deal of effort has been put into trying to explain the effects on health of low concentrations of PM: could it be that more attention should have been focused on the oxides of nitrogen? This question remains unanswered, but we shall return to it shortly.
Cohort studies The second very strong strand of evidence of the effects of PM on health comes from cohort studies. These studies have compared the risks of death between populations living in areas with differing long-term average, ambient, concentrations of PM and some other air pollutants. The Harvard Six Cities Study and the American Cancer Society (ACS) Cohort Study, its re-analysis and its followup studies have dominated the field (Dockery et al 1993; Pope et al 1995, 2002; Health Effects Institute 2000). More recent work from Europe and Asia has added to our knowledge (Hoek et al 2013). Once again, adjustment for potential confounding factors is critically important, but here, the confounding factors differ from those of concern in time-series studies. In the latter, personal factors, which vary little from day-to-day (smoking habits, occupational
133
history, educational status, poverty) matter very little; in cohort studies, these personal factors matter a great deal. It will be realized that time-series studies can be done without personal knowledge of any of the ‘participants’; cohort studies require detailed knowledge of participating individuals. That these studies that are sound in methodological terms is accepted. Exhaustive re-analysis of the Harvard Six Cities Study and the ACS study revealed no significant flaws (Health Effects Institute 2000). Table 1 shows data from the ACS Study (Pope et al. 2002) Here, we are dealing with long-term exposure to PM2.5. Note that the size of the effect on cardiopulmonary mortality, per 10 μg.m−3 of this metric of PM, is about 10 times as great as the effect on cardiovascular disease deaths shown by the time-series studies discussed above (9 % and compared with 0.9 %). Details of the study are provided in a COMEAP report and, of course, in the original publications (COMEAP 2009). A 10 μg.m−3 increment in PM2.5 is associated with a 6 % increase in the likelihood of death from all nonaccidental causes. The effect on the risk of death from cardiopulmonary disease is greater (9 %), and reanalysis showed that the majority of this effect fell on deaths from cardiovascular rather than respiratory (pulmonary) causes (Health Effects Institute 2000). The time-series studies and the cohort studies have in common the fact that no obvious threshold of effect could be detected. Of course, no certain estimates of effects could be made for concentrations of PM lower than those recorded, but in the ACS study, effects were present at as low a concentration as 8 μg.m−3 (PM2.5). These finding are as surprising, to the toxicologist, as those of the time-series studies.
Table 1 Adjusted mortality relative risk, RR, (95 % confidence intervals) associated with a 10 μg.m−3 increase in ambient concentrations of particle monitored as PM2.5. The average RR was calculated as the average of the RRs for the two periods during which PM2.5 had been measured. The original studies should be consulted for details Cause of death
Average relative risk
All cause
1.06 (1.02, 1.11)
Cardiopulmonary
1.09 (1.03, 1.16)
Lung cancer
1.14 (1.04, 1.23)
All other causes
1.01 (0.95, 1.06)
134
Intervention studies The studies described above are observational studies; the toxicologist naturally asks whether any experimental studies have been done. It will be realized that population-based experimental studies on the effects of ambient air pollutants on health will be difficult to organize, though accidents, happy and unhappy, may be studied. Into the latter category fall all the studies of unusually severe episodes of air pollution. Sudden and sustained reductions in levels of air pollution do not, in general, occur as a result of chance: deliberate actions are necessary. Some result from sudden changes in policy, examples include the banning of the sale of coal for domestic heating in Dublin in 1990, the reduction in the sulfur content of oil in Hong Kong in 1990, the reduction of traffic in Beijing during the Olympic Games of 2008 and the traffic congestion charging scheme introduced in London in 2003 (Clancey et al 2002; Hedley et al 2002; Li et al 2010; Tonne et al 2008). Strikes leading to a cessation of industrial activity can also lead to sudden reduction in emissions of pollutants: the nationwide copper smelter strike in the United States of 1967/1968 and the closure and reopening of a steel mill in the Utah Valley, USA, in 1986 provide examples (Pope et al 2007; Pope 1989). Twenty-eight such events have been reviewed by Henschel et al. (2012); in the great majority of cases, a reduction, sometimes a marked reduction, in levels of air pollution followed the events. This reduction was followed, in general, by clear improvements of health including reductions in hospital admissions for cardiopulmonary diseases and deaths. Interestingly, as pointed out by the authors, the improvements in health exceeded those predicted by the time-series evidence reviewed above. Much of the evidence relating to the effects of PM on health has been reviewed by Pope and Dockery: no unbiased reader of their review would be left with doubts about the effect of ambient levels of PM on health (Pope and Dockery 2006).
Cell Biol Toxicol (2015) 31:131–147
disease (COMEAP 2010; Cohen et al 2005). That such effects occur in developing countries with high levels of air pollutants is, perhaps, not surprising; that effects continue to occur in the UK where levels of air pollutants in many towns and cities are probably lower than at any time in the past several hundred years, is remarkable. This large impact on public health, valued on a global scale in billions of pounds, calls for action. But action to reduce levels of air pollutants is costly and must be appropriately directed. The obvious approach is to reduce sources of pollutants but this is far from easy. Efforts to reduce traffic have been made in urban areas, and ‘low-emission zones’ are becoming popular, at least with those responsible for taking action to reduce emissions of pollutants. If the individual pollutants responsible for the effects on health could be identified, then more specific actions to reduce the emissions of these substances could be taken. But this has proved curiously difficult. Many of the major air pollutants are emitted by the same sources, and distinguishing their effects has proved very difficult indeed. Much emphasis has been placed on PM, but this has led to the possible effects of gases such as NO2 being, at least comparatively, ignored. Concentrations of particulate matter monitored as PM2.5 and nitrogen dioxide (better described as nitrogen oxides, NOx, for both nitric oxide (NO) and, NO2, should be included) are closely correlated, and distinguishing their effects has, until recently, been all but impossible. Recent developments in epidemiological methods and in the techniques of meta-analysis have eased the difficulty, and it is now clear that ambient concentrations of NO 2 do have effects on health (WHO 2013). These concentrations tend to be low,
