C International Epidemiological Association 1998 Printed in Great Britain International Journal of Epidemiology 1998.

-27-45O-453

Inappropriate use of daily mortality analyses

to estimate longer-term mortality effects
of air pollution
Anthony J McMichael,a H Ross Anderson,*3 Bert Brunekreefc and Aaron J Cohend

Background To avoid the usual problems of multi-population correlation studies of air pol-
lution and mortality, and for reasons of convenience, daily time-series mortality

Downloaded from https://academic.oup.com/ije/article/27/3/450/625389 by guest on 28 May 2022

studies within single populations have recently become popular in air pollution
epidemiology. Such studies describe how the short-term distribution of deaths
relates to short-term fluctuations in air pollution levels. The regression-based risk
coefficients from these acute-effects studies have been widely used to estimate
the excess annual mortality within a population with a specified average level of
air pollution. Such calculations are inappropriate. Since daily time-series data pro-
vide no simple direct information about the degree of life-shortening associated
with the excess daily deaths (many of which are thought to be due to exacerba-
tion of well-advanced disease, especially cardiovascular disease), such data cannot
contribute to the estimation of the effects of air pollution upon chronic disease
incidence and long-term death rates. Yet it is that category of effect that is of most
public health importance.
Conclusion Such effects are best estimated from long-term cohort studies that incorporate
good knowledge of local (or personal) exposure to air pollutants and of potential
confounders. Time-series studies, properly evaluated, can identify the existence
of acute toxic effects of transient peak levels of air pollution; they are thus useful
for monitoring acute toxicity and for identifying the most noxious pollutants.
However, to quantify the long-term health impacts of air pollution we cannot use
acute-effects data.
Keywords Air pollution, mortality, study design, time-series data, acute effects, chronic effects
Accepted 16 September 1997

Urban air pollution is a major focus of public health concern premature mortality, based on changes in annual age-specific
and regulatory activity. Yet there is an unresolved, indeed poorly death rates.
recognized, tension between the prevailing type of epidemio- Time-series analyses, conducted in many Western cities, have
logical data and the information most needed to make policy. consistently shown that upwards excursions in daily pollutant
The setting of appropriate air quality guidelines requires know- levels are accompanied by extra daily deaths. 1 This provides
ledge of the population burden of illness and premature death clear evidence of a measurable 'acute' effect of air pollution
avoidable by control of specified air pollutants. However, most episodes upon daily mortality. At the least, it indicates that
epidemiological studies of the mortality impacts of specific air air pollution causes alterations in the day-to-day distribution of
pollutants have been based on daily time-series data. These deaths within that population, presumably by precipitating
do not allow estimation of the usual population indices of death in susceptible persons (usually with advanced disease).
However, the more important public health question is whether
sustained exposure to elevated levels of air pollution affects
* Department of Epidemiology and Population Health, London School ol annual death rates, by inducing or contributing to the develop-
Hygiene and Tropical Medicine. Keppel Street. London WC1E 7HT, UK.
b ment of life-shortening chronic disease. That latter question
Department of Public Health Sdence, St Georges Hospital Medical School,
Cranmer Terrace, London SVV17 ORE. UX. cannot be addressed by daily time-series mortality data since,
c
Department of Epidemiology and Public Health, University of Wageningen. by their nature (including the lack of an external comparison
PO Box 238, 6700 AE Wageningen, The Netherlands. population), they provide no direct information on the loss of
d
Health Effects Institute, 955 Massachusetts Avenue, Boston, MA 02139. USA. person-time associated with the 'extra' deaths. Nor can they tell

450
 INAPPROPRIATE USE OF DAILY MORTALITY ANALYSES 451

us about the contributions of transient peaks in exposure to particle and sulphur dioxide pollution.8 Using risk coefficients
longer-term disease processes. from local daily mortality time-series studies, the authors
Much recent public discussion and economic impact assess- estimated the proportional change in daily deaths achievable by
ment has confused this issue. A recent briefing report prepared reducing current annual mean pollution levels to zero, and then
internally for the British Parliament illustrates the general multiplying the estimated daily mortality saving by 365. This
problem.2 The report explicitly recognizes that there are 'acute' calculation assumes that none of those deaths would otherwise
and 'chronic' effects of air pollution on mortality. But it never- have occurred during that year, whereas the underlying time-
theless states, on the basis of time-series findings, that 'if meas- series data neither allow nor suggest that assumption.
ures were taken to reduce average annual levels of PM 10 Relatedly, a recent estimate was made of the annual health
(i.e. fine respirable particulates), each 1 |ig/m 3 reduction might costs attributable to air pollution from road transport in the UK
reduce the number of deaths each year by 400 or so.' This (£19.7 billion in 1993). 10 Pollutant-specific risk coefficients
statement, like various others published recently, inappro- from daily time-series studies were, again, used to estimate the
priately scales up the transient excess daily mortality from time- annual extra number of lives lost. For each death a standard
series studies in order to estimate the impact of sustained economic cost was assumed. Yet the real costs to society vary as

Downloaded from https://academic.oup.com/ije/article/27/3/450/625389 by guest on 28 May 2022

pollutant exposure on annual death rates. a function of the amount of life actually lost. A similar con-
What, then, do daily time-series studies actually measure? By fusion was evident in a World Bank working paper, published
regressing the daily death counts on daily air pollutant levels, in 1994, that used a time-series-based approach to estimate the
we can estimate the relative risk of immediate death associated mortality impact of air pollution in order 'to (help) analyse the
with a given short-term increase in air pollution. For example, economics of pollution control in developing countries'.
it has been estimated that there is an approximately I % increase This confusion has been compounded by a recent technical
in daily deaths for a day with a level of particulates (PM ]0 ) report from WHO entitled 'A Methodology for Estimating Air
10 |ig/m 3 higher than another day.3 (Capital cities in Europe Pollution Health Effects'.11 The report acknowledges that, since
typically have mean PM 10 levels around 20-40 ug/m 3 .) Indeed, 'cumulative long-term exposure to particles may also contribute
since no 'threshold' (no-effect) level of exposure is evident, a to premature mortality, it is likely that the acute effect (time-
general linear relationship is presumed such that even the many series) exposure studies ... underestimate the effects' (p.10).
days with relatively low PM 10 levels are presumed to entail an Surprisingly, and without satisfactory explanation, the report
increment in daily deaths. proposes averaging the dose-response coefficients from daily
Summing these small daily increments over a whole year time-series and long-term follow-up studies to estimate mortal-
yields a large number of 'extra' deaths associated with daily ity impacts from particulates.
upward fluctuations in levels of PM 10 (or other air pollutants).
The annual sum can run into several thousands, even for small
countries such as the Netherlands. 4 But how should such sum-
Short-term mortality displacement
mation be interpreted? We simply do not know whether, for a or 'harvesting'
population with a given long-term mean exposure, one sequence There is no doubt that brief elevations of various air pollutants
of daily PM 10 fluctuations causes more or less total deaths cause extra deaths, and that some reduction in life-span (i.e.
during a whole year than does some other sequence of PM 10 loss of person-time) thus occurs. There is, however, need for a
exposures—or, indeed, than does a non-fluctuating exposure clearer understanding of the 'losses' involved. Deaths in serious
level. Presumably, many of these extra daily deaths (which air pollution episodes and during heat-waves are known to
must come from some unidentified, constantly refilling, pool of occur disproportionately among the very old and very sick. I2/ '
susceptibles) would have occurred during that particular year Hence, it is likely that many, perhaps most, of the excess daily
anyway, even without fluctuations in air pollution. However, deaths evident in time-series data result from short-term dis-
since we cannot identify those actual individuals who die placement of the time of death (e.g. from well-advanced cardio-
because of a transient air pollution episode (nor, therefore, can vascular disease), sometimes called 'harvesting'. Every such
we estimate their pre-episode life expectancy), it is not possible extra death involves some degree of prematurity, from as little
to estimate directly the loss of person-time associated with the as a day to many years. For the important public health task of
estimated number of extra deaths. Hence, summing the series estimating the impact of air pollution on population mortality,
of acute mortality excesses, without knowledge of the loss of
measured as changes in the annual death rate, we need to be
person-time, cannot yield an estimate of the change in annual
able to estimate the person-years of life lost. This issue is well-
death rate. 5
understood by epidemiologists in other contexts. For example,
avoidable deaths at young age (e.g. traffic accidents killing
young adults) entail a greater loss of person-time than do
Examples of inappropriate use of avoidable deaths in late adulthood. 14 ' 15
time-series data There has been some recent recognition of the existence of
This confusion has not been confined to government docu- short-term displacement of mortality by transient elevations of
ments. Various scientiOc papers and working documents have air pollution. The UK Government's Expert Panel on Air Quality
blurred the distinction between study types, 6 or have inap- Standards, reviewing time-series studies, recently questioned
propriately used the published results of daily time-series studies 'whether such excess death rates (sic) represented either more
to estimate impacts on annual death rates, 3 ' 7 " 9 For example, people dying overall, i.e. an increase in absolute mortality, or
inappropriate estimates were made for Paris, in 1995, of rather the deaths of already ill people being brought forward,
the annual number of lives that could be saved by eliminating perhaps by only a few days, and therefore fewer dying over the
452 INTERNATIONAL JOURNAL OF EPIDEMIOLOGY

subsequent period.' 16 The Committee on Medical Effects of Air Despite the new enthusiasm for daily mortality time-series
Pollution also noted the problem. 17 Meanwhile, the US Gov- studies, the basic conceptual problem persists. By design, daily
ernment's Environmental Protection Agency, in its Paniculate mortality time-series studies show only how the short-term
Matter Criteria Document 18 and Staff Paper,19 distinguished distribution of deaths relates to short-term fluctuations in air
between the results of daily time-series studies and those of pollution levels. Nevertheless, time-series studies have drawn
cross-sectional and cohort studies (but then attempted awk- attention to the existence of acute toxic effects of low levels of
wardly to estimate the proportion of the annual death rate air pollution; 3435 they are useful for monitoring acute airborne
attributable to fluctuations in paniculate exposure). toxicity and for identifying the most noxious components or
The challenge, then, is how to elucidate the relative size and mixtures; and they can be used to explore the phenomenon of
nature of the acute and chronic effects of air pollution upon mortality displacement.
mortality. Two studies have sought to quantify aspects of short- In order, now, to differentiate between the acute and longer-
term mortality displacement within daily time-series data sets. term mortality impacts of air pollution two types of research are
Cifuentes and Lave describe short-term mortality displacement needed:
associated with increases in paniculates-related daily mortality (1) Assessment of the extent to which the observed increases

Downloaded from https://academic.oup.com/ije/article/27/3/450/625389 by guest on 28 May 2022

in Philadelphia between 1980 and 1989, but observe that, empir- in daily mortality represent short-term mortality displacement.
ically, the displacement does not affect the estimated regression This will require either the development of statistical ap-
coefficient.20 In a study in Erfurt, Germany, Spix et al estimated proaches to the estimation, from daily time-series data, of the
that the apparent effect of air pollution on daily mortality was amount of person-time lost, or studies that provide more de-
reduced by about one-third on days that had been preceded by tailed descriptions of the clinical characteristics of people who
a 15-day period of high daily average mortality—which was die on days with increased air pollution—including studies that
presumed to have reduced the number of susceptible people analyse jointly morbidity and mortality rates in the same locales
who would otherwise have been at increased risk of death from over the same time interval. Such studies are now under-
elevated air pollution. 21 way. 36 ' 37
Meanwhile, there have been few epidemiological studies of (2) Additional studies of the impact of long-term exposure
the impact of long-term air pollution exposure on life expect- to air pollution on mortality from chronic diseases. Cohort
ancy. To date, three cohort studies of long-term exposure to air (or, sometimes, case-control studies) can provide the best such
pollution have been reported, all from the US. 22 " 24 Two of these information, provided that there is satisfactory control of con-
studies suggest that prolonged exposure to elevated, though founding effects.
moderate, levels of airborne particulates increases the long-term Finally, it should be noted that the abovementioned limita-
death rate. • A US environmental advocacy group, the Natural tion upon the type of health impact assessment derivable from
Resources Defense Council, recently estimated the mortality daily time-series data applies particularly to mortality studies.
impact of paniculate matter using risk estimates from the Whereas death is a once-only event, many non-fatal health
American Cancer Society cohort study, 23 and pointing out that, events can occur repeatedly in an individual. Time-series ana-
although most of the recent risk assessments had used daily lyses may therefore be especially informative for non-fatal
time-series results to estimate effects, 'it seems logical that pro- outcomes.
longed exposure presents the greatest threat to public health.' 25

Conclusions
Research: past developments, current needs The many recent time-series analyses of daily air pollutant level
It is instructive for epidemiologists to consider how this con- in relation to daily deaths indicate that several categories of air
fusion over acute and chronic mortality impacts of air pollution pollution pose an acute health hazard. However, we must be
has arisen. cautious in interpreting these findings. Daily time-series ana-
Several well-recognized difficulties in quantifying the mort- lyses, well able to describe variations in the daily distribution of
ality effects of air pollution 26 ' 27 explain much of the evolution deaths (particularly in the vulnerable subgroups who are the
of epidemiological research methods. The earliest studies were 'canaries' of our modern cities), do not provide a satisfactory
based on the dramatically obvious acute effects on mortality, basis for estimating the longer-term change in mortality attri-
within a single population, of major air pollution episodes (e.g. butable to sustained exposure to air pollution.
the London smog of 1952). 28 Early studies of the long-term
mortality impacts of air pollution compared populations in
different locations that had different pollutant exposures. 29 " 3 ' Acknowledgements
However, since those latter studies were prone to inter- We have received helpful comments from Jonathan Samet
population confounding (e.g. by differences in prevalence of (Johns Hopkins University), Claudia Spix (University of Mainz),
cigarette smoking) 32 a preference has emerged over the past John C. Bailar III (University of Chicago) and Tord Kjellstrom
decade for analysing time-trends within a single population. This (WHO Geneva).
avoids inter-population confounding. Although, contemporary
air pollution episodes (e.g. London in 199133) entail lower ex-
posure levels and lesser health impacts than did their pre- Disclaimer
decessors, sophisticated and sensitive statistical techniques The views expressed in this paper are those of Dr Cohen and do
are now available for the formal analysis of within-population not necessarily reflect the opinions of the Health Effects
time-series data. Institute or its sponsors.
 INAPPROPRIATE USE OF DAILY MORTALITY ANALYSES 453

Spix C. Heinrich J, Dockery D et al. Air pollution and daily mortality

References in Erfurt, East Germany, 1980-1989. Environmental Health Perspectives
1 1993,101:518-26.
The APHEA project. Short term effects of air pollution on health'
a European approach using epidemiological time series data. J Epi- 1
Dockery DW, Pope CA, Xu X et al An association between air
demiol Community Health. 1 9 9 6 ; 3 0 ( S u p p l . l ) . (Seven of the thirteen pollution and mortality in six US cities. New Engl J Med 1993:329:
papers present daily time-senes studies of air pollution and mortality.) 1753-59
2 23
Parliamentary Office of Science and Technology. Fine Particles and Pope CA, Thun MJ, Mohan M et al. Particulate air pollution as a
Health London: POST, 1996. predictor of mortality in a prospecuve study of US adults. Am J Respir
3 Crit Care Med 1995,151:669-74.
Ostro B. Estimating the Health Effects of Air Pollution. A Method with an
Application to Jakarta (Policy Research Working Paper No. 1301 ) 24
Abbey DE, Mills PK, Petersen FF, Beeson WL. Long-term ambient
Washington, DC: World Bank, 1994. concentrations of total suspended particulates and oxidants as related
4 to incidence of chronic disease in California Seventh-Day Adventists.
Health Council of the Netherlands. Particulate Air Pollution. Health
Council Report 1995/14 (ISBN 90-5549-089-X), The Hague, 1995. Environ Health Pcrspcct 1991,94:43-50
5 25
Elandt-Johnson RC, Johnson NL. Survival Models and Data Analysis Shprentz DS, Bryner GC, Schprentz JS, Hawkins DG. Breath-taking
New York: Wiley, 1980. Premature Mortality due to Particulate Air Pollution in 239 American Cities

Downloaded from https://academic.oup.com/ije/article/27/3/450/625389 by guest on 28 May 2022

6 New York: Natural Resources Defense Council, 1996.
Moolgavkar SH, Luebeck EG. A critical review of the evidence on
26
particulate air pollution and mortality. Epidemiology 1996:7:420-28. Katsouyanni K. Research methods in air pollution epidemiology.
7 In- Fletcher T, McMichael AJ (eds) Health at the Crossroads- Transport
B o b a k M, Feachem RGA. Air pollution and mortality in central and
Policy and Urban Health. Chichester J o h n Wiley and Sons, 1997,
eastern Europe: an estimate of the impact. Eur J Public Health 1995;
pp 51-60.
5:82-86.
27
8
B o w n W. Dying from too much dust. New Scientist 12 March 1994: Dockery DW Epidemiologic study design for investigating respiratory
health effects of complex air pollution mixtures. Environ Health
12-13.
Perspect 1993;101(Suppl.4):187-91.
9
Qu£nel P, Zmirou D, Le Tertre A et al. Impact de la pollution atmos- 28
p h i r i q u e urbaine de type acid-particulaire sur la mortality quoti- Ministry of Health. Mortality and Morbidity during the London Fog of
dienne a Lyon et dans l'agglomeration parisienne. Santi Pubhque December 1952. Reports on Public Health and Medical Subjects No 95.
1995;6(4):1-14. London: HMSO, 1954.
29
10 Gardiner MJ, Crawford MD, Morris JN. Patterns of mortality in
Maddison D, Pearce D, Johansson O et al. Blueprint 5: The True Costs of
Road Transport. London: Earthscan, 1996. middle and early old age in the county boroughs of England and
Wales. Br J Prev Soc Med 1969:23:133-40
Ostro B. A Methodology for Estimating Air Pollution Health Effects 30
Chinn S, Florey C duV, Baldwin IG, Gorgol M. The relation of
WHO/EHG/96.5. Geneva: WHO, 1996
12
mortality in England and Wales 1969-73 to measurements of air
Schwartz J. What are people dying of on high air pollution days? pollution. J Epidemiol Community Health 1981:35:174-79
Environ Res 1996:64:26-35. 31
13
Evans JS, Josteson T, Kinney PL. Cross sectional mortality studies and
Kalkstein L. Health and climate change: direct impacts in cities. Lancet air pollution risk assessment. Environ Inl 1984:10:55-83.
1993;342:1397-99. 32
Commission of the European Communities. Air Pollution Epidemiology
Petiti DB. Meta-analysis, Decision Analysis and Cost-effeaiveness Analysis— Series Report No 4. Study Designs Brussels: Commission of The Euro-
Methods for Quantitative Synthesis in Medicine. Oxford: Oxford University pean Communities, 1993.
Press, 1994. 33
Anderson HR, Limb ES, Bland JM et al. Health effects of an air
World Bank World Development Report 1993. Investing in Health Oxford. pollution episode in London, December 1991. Thorax 1995,50:
Oxford University Press, 1993. 1188-93.
16
Department of the Environment (UK). Expert Panel on Air Quality 34
Pope CA. Combustion Source Paniculate Air Pollution and Human Health
Standards. Particles London: HMSO, 1995. Causal Associations or Confounding? Paniculate Matter: Health and
Committee on Medical Effects of Air Pollution (COMEAP). Non- Regulatory Issues (VIP.49) Proceedings, I n t e r n a t i o n a l Specialty
Biological Particles and Health. London: HMSO, 1995. Conference hosted by Air and Waste Management Association, 4—6
US Environmental Protection Agency. Air Quality Criteria for Particulate April, 1995, Pittsburgh, PA, 1995, pp.60-77.
35
Matter. Report No. EPA/600/P-95/001. Washington, DC: EPA, 1996. Utell MJ, Samet JM. Particulate air pollution and health: New
US Environmental Protection Agency (OAQPS Staff Paper). Review of evidence on an old problem (Editorial) Am Rev Resp Dis 1993,147:
the National Ambient Air Quality Standards for Particulate Matter. Policy 1334-35
36
Assessment of Scientific and Technical Information. Research Triangle Park, Spix C. Daily time series of mortality counts: Estimating the harv-
NC: US Environmental Protection Agency, 1996. esting effect. Statistics in Medicine (In press), 1998.
20 Cifuentes L, Lave LB. Association of daily mortality and air pollution in 37
Health Effects Institute. Program Summary Research on the Effects of
Philadelphia, 1983-1988 Carnegie Mellon University Working Paper, Paniculate Air Pollution on Mortality and Morbidity Cambridge, MA:
Chicago, 1997. Health Effects Institute, April 1996.