Mortality due to chronic exposure to PM

There are far fewer published studies on the mortality effects of long-term exposures to ambient PM than on the mortality effects of acute exposure. The current number of published cohort studies is five. The total number of publications in scientific journals about these five cohort studies is, of course, much larger, but essentially all conclusions are based on the information contained in just five cohort studies.

1. The first one is the Harvard Six Cities (HSC) study (Dockery et al., 1993). In the HSC study, the associations between mortality and long-term exposure to sulphate and several PM size fractions were investigated in six American cities. The HSC comprised 8111 white subjects, who were followed for 111 076 person years. This study suggests that the fine mass component (PM2.5) is more strongly correlated with

mortality than is the coarse component (PM15 or 10 minus 2.5). Overall, the results showed

statistically significant relationships between long-term exposures to PM and/or sulphates and excess mortality. A further follow-up of this study has recently been completed, but not yet published.

2. The American Cancer Society (ACS) study (Pope et al., 1995, with a follow-up in 2002) is the second study of long-term exposure. In the ACS, the risk factor data of 500 000 adults were linked with air pollution data and combined with vital status and cause of death between 1982 and 1998. This is the largest cohort available to date. The results of the ACS indicate that fine particle and sulphur oxide-related pollution were associated with all-cause, cardiopulmonary and lung cancer mortality. Each 10 µg/m3 elevation in PM2.5 was associated with an approximately 4%, 6% and 8% increased

risk in the respective causes of death. Measures of coarse PM, TSP and gaseous pollutants such as NO2, CO and O3 were not consistently associated with mortality.

3. The Adventist Health Study on Smog (AHSMOG) (Abbey et al., 1999) represents the third major U.S. prospective cohort study of chronic PM exposure-mortality effects. In 1977 the study enrolled 6338 non-smoking non-Hispanic white Seventh-Day

Adventist residents of California between the ages of 27 and 95. The authors found long-term ambient concentrations of PM10 to be associated with increased risks of all

natural cause mortality in males, mortality with any mention of non-malignant respiratory causes in models that included both sexes, and lung cancer mortality in males. Lung cancer analyses were based on small numbers (18 deaths for females and 12 for males) requiring further exploration in follow-up studies.

4. In a fourth cohort study, Lipfert et al. (2000a) reported results from a large-scale mortality analysis using a prospective cohort of up to 70 000 men assembled by the U.S. Veterans Administration (VA) who were diagnosed as hypertensive in the mid- 1970s. The study cohort was male, middle-aged (51 + 12 years) and included a larger proportion of African-Americans (35%) than the U.S. population as a whole and a large percentage of current or former smokers (81%). Contrary to three of the previous cohort studies, no associations with particulate matter were found in relation to

mortality.

5. In the Netherlands, the association between mortality and distances to major roads of the home addresses of respondents in combination with a regional and urban

background is being studied in an ongoing cohort study – The Netherlands Cohort Study on Diet and Cancer (NLCS). A cohort of 120 000 55- to 69-year-old subjects

investigated a random sample of 4492 people from the full cohort. Long-term exposure to traffic-related air pollutants (BS and NO2) was estimated for the 1986

home address. During the follow-up period, 489 people died. Cardiopulmonary mortality was significantly associated with living near a major road (RR 1.95). For total deaths the RR for living near a major road was 1.41, though statistically not significant. Non-cardiopulmonary and non-lung cancer deaths were unrelated to air pollution.

A grant has been obtained from HEI to include all 15 000 deaths that have occurred so far in the cohort for the analysis of a relationship with traffic-related air pollution. Results are expected in 2005. With support from NAP, a small extension of the analysis will be conducted in the course of 2002.

In qualitative terms it can be argued that the US results (ACS and HSC) pointing to health effects associated with chronic exposure to PM have been corroborated by this Dutch study, which focuses on the effects of traffic (Hoek et al., 2002c).

Weighing all the evidence of these cohort studies and assigning specific significance to the fact that the first two studies comprised a sample of the general population and have been thoroughly re-analysed, some of the authors are convinced that chronic exposure to ambient levels of PM in the US have led to the reported health effects. Both studies (HSC and ACS) have been supplemented by a follow-up, which generally points in the same direction as well, though only one of the follow-ups (ACS) has as yet been published. The AHSMOG

population, however, consisting of Seventh-Day Adventists, is very different in a number of lifestyle factors from the general population, so that findings in this particular population cannot easily be extrapolated. The HSC and NLCS studies, and to a somewhat lesser extent the ACS study, are truly population-based. This is not so for the veterans study (VA), for which only a preliminary report is available. Clearly, the weight of the evidence is with the HSC, ACS and NLCS studies at the moment.

However, some of the authors conclude that the overall picture from these five studies is mixed in regard to the evidence concerning chronic exposure to PM and associated mortality, as some of the studies point to PM effects and some do not. These authors think it is too early to come to a conclusion. Not finding a significant PM effect in two of the studies does not, of course, prove that there are no health effects from chronic exposure to PM. The information from the other studies indicates that the associations can be called qualitative at the least and warrants the serious attention of those involved with environmental health effects. The question of whether these studies can also be used for the quantitative risk estimation of chronic PM effects in the Netherlands is of a different order and will be explored later on. A legitimate question to ask when interpreting the results of ‘ecologic’ studies is whether the individual concentrations are represented well enough by the ambient concentrations of the central site monitors. Exposure in the US cohort studies of chronic exposure to PM is described at a group level instead of at an individual level. This may lead to the so-called ecological fallacy. The group-level information on all the people living within a radius of a number of kilometres is represented by the concentration at a central site monitor or a number of monitors considered to be representative for the sought-after personal exposure to ambient PM. Reviewing the epidemiological literature on the relation between ambient air pollution exposure and cancer, which included the first two cohort studies (HSC and ACS),

Katsouyanni and Pershagen (1996) concluded: ‘A major drawback in the studies has been the inadequate characterisation of air pollution exposure. First, measurements of air pollution in

the study areas should span the time period relevant for the disease etiology and preferably should include concentrations of suspected carcinogens. Second, an estimation of individual exposure should be based on exposure studies, studies of the time activity patterns, and the geographic distribution of pollutants in a micro-scale. Exposure studies should provide data on how individual exposure is related to the levels measured at fixed monitors, considering different activities and transportation used. Further methods for retrospective exposure assessment covering periods of several decades should be developed. The results of such studies could be used as input in large analytic epidemiological investigations to address the problems of measurement error and reduce uncertainties in the RR estimates.’

The currently used exposure measures (in HSC, ACS, AHSMOG) are those during the final number of years immediately preceding death. Unfortunately, we do not really know the time window for relevant exposure. Referring to cessation of smoking, a 50% reduction for lung cancer risk takes approximately 10 years, and after 20 years it has decreased to a background level. For cardiopulmonary diseases this period appears to be much shorter and may be in the order of years. If different individual measures of exposure comprising such a window of exposure, which are of course much harder to come by, had been used, the magnitude of the assigned exposures would probably have been larger than the currently assigned PM

exposures in the studies. Quantitatively, this might mean that the currently reported relative risks in the chronic studies, assuming that they would remain statistically significant with the new exposure measures, would have been lowered by a similar amount to that by which the exposures have been augmented. These arguments about exposure validity indicate that the time domain of extrapolation needs to be explored very carefully.

The effect of the reduction in PM over the last few decades should show up in the follow-up period in lower numbers of health effects associated with PM compared with the previous period in the cohort studies for chronic exposure. Unfortunately, the currently reported follow- up of the ACS study does not order its information in a way to present such a picture. The associated extra risk of cardiopulmonary mortality in the US per 10 µg/m3

decreases from 13% to 4% when the first period is compared with the total period, which includes the first period and the follow-up. A more or less similar RR would seem the logical result to expect when the decreases in PM concentrations over the last few decades are taken into account and chronic exposure to PM is causal for the health effects. A decrease in RR could be due to a change in the ambient PM mix. However, for lung cancer in the ACS the picture is reversed, and lung cancer mortality per 10 µg/m3

increases from 1.4% to 8% when the first period is compared with the total period. Finding both results in the same study seems contradictory. Krewski et al. (2000) indicated that in the ACS study the relationship between PM exposure and mortality disappeared for high SES. A finding like this mostly points to some sort of effect modification, though it could also point to an insufficient control for occupational exposures of those with less than high school education. In the ACS follow-up (Pope et al., 2002) the health effects of PM, which are significant for those with an education level less than high school, disappear in Figure 4 for those with more than high school education. This may indicate that something in the lifestyle or diet of the group in question can mitigate or prevent the PM-associated health effects. Further research in this direction is warranted, as it may lead us to lifestyle or other personal factors that could be influenced to reduce PM risks to the population.

Lipfert et al. (2000a) concluded from the VA as a new insight: ‘... the general decline of mortality responses to air pollution with increasing follow-up time. This trend could suggest depletion of the cohort of its most susceptible subjects, a concentration-response threshold, increasing uncertainty about the exposures and the characteristics of the cohort, or all of these. A bona fide chronic effect would be expected to be manifested throughout the period of follow-up, especially as the cohort ages, not just at the beginning. It thus follows that other such cohort studies should also examine the ramifications of the timing of air pollution exposure.’ A substantiation against an extrapolation in time can be found in a very recent cross-sectional study for five specific periods from 1960 to 1997, which examined temporal and spatial relationships between air pollution and age-specific mortality rates for US counties (Lipfert and Morris, 2002). On the basis of attributable risks computed for overall average concentrations, the strongest associations were found in the earlier periods, with attributable risks usually less than 5%. Stronger relationships were seen when mortality and air quality were measured in the same period and when locations were limited to those of the HSC. Responses to PM, CO and SO2 declined over time, suggesting that the results of previous

studies may no longer be applicable.

Although most of the mortality studies focused on the total population or the elderly, new studies on mortality in the very young suggest PM effects in a susceptible sub-population of children (Bobak and Leon, 1999; Loomis et al., 1999; Lipfert et al., 2000b). Post-neonatal mortality has been associated with PM levels. In the current NAP project on PM-related health effects we have tried to collect data on post-natal and post-neonatal deaths in the Netherlands to study this phenomenon and to find out if it occurs in the Netherlands too. Unfortunately, we were unable to obtain these data because they were not available for privacy reasons.

a. Transferability to the Netherlands of health effects of chronic exposure

Two of the previously presented long-term studies indicate (relatively) high mortality risks for rather low levels of PM10, PM2.5 or sulphates in the US. A quite recent re-analysis of the data

of the HSC and the ACS study (Krewski et al., 2000) sponsored by the HEI indicated that the original calculations and data handling satisfied all the necessary scientific and technological standards.

An excess of 20 µg/m3 PM2.5 in the HSC is associated with 28% excess mortality, leading to a

relative risk of RR = 1.28 per 20 µg/m3 PM2.5 (US-EPA, 2001). If the results of these US

studies are also applicable to the Netherlands (with an annual average concentration of approximately 20–25 µg/m3 PM2.5), the magnitude of the RR signifies that the health effects

of chronic exposure to PM constitute a huge problem with a large health impact. If true for the Netherlands, the American HSC figures imply on average an excess mortality of 30%,

resulting in a serious shortening of our life span. However, the most recent RR from the follow-up of the ACS is somewhat lower than that from the HSC. Pope et al. (2002) indicate a 4% rise in all-cause mortality and a rise of 8% in lung cancer mortality for a 10 µg/m3

elevation in PM2.5. Translated into the current situation in the Netherlands, with

approximately 20–25 µg/m3 PM2.5, this would mean an additional all-cause mortality of

approximately 10% and an additional lung cancer mortality of 20%. These health implications remain considerable; careful exploration is necessary to see whether such an extrapolation from the US to the Netherlands is permitted. At the average concentrations, the health effects estimated by Pope et al. (2002) are even higher.

Some of the authors argue that there is no fundamental difference between sources and air pollution concentrations in the US and the Netherlands. Differences within the US are

probably larger than between relevant areas in the US and the Netherlands. Long-term effects have been found both in the west and in the east in the US, which is why some of the authors conclude that the long-term effects found in the US can be extrapolated to the Netherlands. The health impact of PM in the Netherlands can be quantified on the basis of the HSC and ACS results.

However, some of the authors point more to the differences than to the existing similarities in PM air pollution in the US and the Netherlands. A formal argument could be used to declare such an extrapolation non-permissible. For instance, strictly scientifically speaking,

extrapolating a regression (and these epidemiological results are essentially regressions) outside of its geographical and time domain is never permitted. The air pollution mix in the Netherlands and that in the US are quite different in a number of aspects, and so are the health effects probably, too. Concentrating on one aspect only, namely sulphate, the latest exposures in the ACS presented in Pope et al. (2002) show US sulphate levels to be 6.2 µg/m3 with an SE of + 2.0. This leads to the conclusion that the current annual average levels of sulphate in the Netherlands, which are some 2 µg/m3, are presently outside the 95% confidence interval of the US sulphate data in the ACS. Of course, sulphate levels in the Netherlands were also higher in the past, but for an up-to-date risk estimate we need to use our current levels. Using the HSC data at face value for the Netherlands may lead to inconsistencies in the risk estimates when different indicators for the air pollution mix are used, e.g. when the current increase in all-cause mortality in the Netherlands is estimated using two different indicators from the HSC. Using the Dutch PM2.5 data and HSC RR leads to an increase in all-cause

mortality of approximately 30% in the Netherlands, as has been indicated above. When, on the other hand, the HSC RR data are used and sulphate is taken as an indicator instead of PM2.5, then the previously found increase in all-cause mortality in the Netherlands completely

disappears, because the ambient levels of sulphate in the Netherlands are considerably lower than those in even the least polluted of the six American cities in the HSC, and the RR in the HSC were calculated in comparison with the location with the lowest pollution levels. Ayres (2002) cautioned in an editorial how unwise it was to extrapolate effect size coefficients from one area to another with the questions currently still open.

In reality, it is necessary to determine very carefully, conclude some of the authors, which constraints need to be met concerning the air pollution mix and geographical location (spatial) or time period (temporal) in order to validate an extrapolation and use the results obtained in the US for a quantitative risk assessment elsewhere.

These conclusions or differing views concerning the quantitative extrapolation of foreign data are no reason for complacency, however. All these arguments and the qualitative information from the four US studies plus the preliminary results of the Dutch cohort study constitute a strong plea for a well-designed prospective Europe-wide cohort study, to provide us with some of the much desired answers to the question of the health effects of chronic exposure to PM.

position on the research agenda is justified. Once the causal factors of these PM-associated health effects have been elucidated, it may become possible in the future to develop cost- effective mitigation or abatement strategies, or both.