American Journal of Epidemiology
Copyright ª 2006 by the Johns Hopkins Bloomberg School of Public Health
All rights reserved; printed in U.S.A.
Vol. 164, No. 1
DOI: 10.1093/aje/kwj147
Advance Access publication April 19, 2006
Original Contribution
Declining Vulnerability to Temperature-related Mortality in London over the
20th Century
Claire Carson, Shakoor Hajat, Ben Armstrong, and Paul Wilkinson
From the London School of Hygiene and Tropical Medicine, London, United Kingdom.
Received for publication March 21, 2005; accepted for publication January 6, 2006.
The degree to which population vulnerability to outdoor temperature is reduced by improvements in infrastructure, technology, and general health has an important bearing on what realistically can be expected with future
changes in climate. Using autoregressive Poisson models with adjustment for season, the authors analyzed
weekly mortality in London, United Kingdom, during four periods (1900–1910, 1927–1937, 1954–1964, and
1986–1996) to quantify changing vulnerability to seasonal and temperature-related mortality throughout the 20th
century. Mortality patterns showed an epidemiologic transition over the century from high childhood mortality to low
childhood mortality and towards a predominance of chronic disease mortality in later periods. The ratio of winter
deaths to nonwinter deaths was 1.24 (95% confidence interval (CI): 1.16, 1.34) in 1900–1910, 1.54 (95% CI: 1.42,
1.68) in 1927–1937, 1.48 (95% CI: 1.35, 1.64) in 1954–1964, and 1.22 (95% CI: 1.13, 1.31) in 1986–1996. The
temperature-mortality gradient for cold deaths diminished progressively: The increase in mortality per 1C drop
below 15C was 2.52% (95% CI: 2.00, 3.03), 2.34% (95% CI: 1.72, 2.96), 1.64% (1.10, 2.19), and 1.17% (95% CI:
0.88, 1.45), respectively, in the four periods. Corresponding population attributable fractions were 12.5%, 11.2%,
8.7%, and 5.4%. Heat deaths also diminished over the century. There was a progressive reduction in temperaturerelated deaths over the 20th century, despite an aging population. This trend is likely to reflect improvements in
social, environmental, behavioral, and health-care factors and has implications for the assessment of future
burdens of heat and cold mortality.
climate; greenhouse effect; mortality; seasons; temperature; weather
Abbreviation: CI, confidence interval.
Most temperature-related deaths, whether caused by heat
or by cold, are theoretically preventable; yet there is abundant evidence of vulnerability to both high and low outdoor
temperatures, even in high-income countries. The events
occurring in Europe in August 2003 demonstrated the potential threat of unaccustomed heat (1, 2), while Britain is
among those countries that year after year experience a substantial burden of excess winter deaths (3, 4).
The magnitude of temperature-related mortality may be
thought to reflect shortcomings of public health, but the
factors that contribute to it may be related as much to broad
social, environmental, and behavioral influences as to defi-
ciencies of the health-care system. As living standards improve, it would be reasonable to expect reduced population
vulnerability to both cold and heat because of the accompanying improvements in infrastructure, technology, support
services, and general health. The degree to which this is true
has an important bearing on the question of what realistically can be expected in the future with regard to both winter
deaths and the likely direct health impacts of climate
change. To gain insight into these wider, wealth-related
influences, we examined the past trends of seasonal and
temperature-related mortality in London, United Kingdom,
using data spanning the 20th century that captured several
Correspondence to Dr. Shakoor Hajat, Public and Environmental Health Research Unit, London School of Hygiene and Tropical Medicine,
Keppel Street, London WC1E 7HT, United Kingdom (e-mail: [email protected]).
77
Am J Epidemiol 2006;164:77–84
78
Carson et al.
phases of progressive socioeconomic and epidemiologic
development.
MATERIALS AND METHODS
We analyzed British data for four periods in the 20th
century selected to avoid times of war and influenza pandemics: 1900–1910, 1927–1937, 1954–1964, and 1986–
1996. For the first three of these periods, weekly counts of
deaths occurring in London, by age and cause-of-death
group, were obtained from the Weekly Returns of Birth
and Deaths, Infectious Disease, Weather published by Her
Majesty’s Stationery Office (later the Government Statistical Office). For the last period, we obtained an electronic file
of deaths occurring in London from the Office for National
Statistics and derived from it weekly counts of death by age
and cause of death to match data from earlier periods.
Meteorologic data (daily maximum and minimum temperatures) for each period were obtained from the monitoring station at Kew (unpublished data, British Atmospheric
Data Centre, Chilton, United Kingdom (http://badc.nerc.ac.
uk/home/index.html)). Daily mean temperatures were calculated from the daily minimum and maximum and then
averaged by week to correspond with the weeks of mortality
data.
For the last period, 1986–1996, we obtained from the
same weather station data on daily relative humidity, which
were then converted into weekly mean values. Additionally
for the last period, we obtained weekly counts of clinical
specimens of influenza A reported to the Health Protection
Agency and daily air pollution data (24-hour mean particulate matter 10 lm in diameter (lg/m3)) for the Bloomsbury
automatic monitoring station (unpublished data, National
Air Quality Information Archive, Culham, United Kingdom
(http://www.aeat.co.uk/netcen/airqual/welcome.html)). For
1954–1964, the Weekly Returns (5) provided tabulations
of the mean, minimum, and maximum air pollution levels
as measured by the Owen’s Smoke Filter, which collects
particulate matter by filter (6).
Statistical methods
To analyze excess mortality occurring during the winter
season, we defined winter as December–March, according
to the method of Curwen (3). These are the coldest months
in Britain, and they contain the large majority of days with
a maximum temperature below 5C. We examined weekly
mortality in relation to ambient temperature using Poisson
generalized linear models allowing for overdispersion. Our
modeling followed approaches that are well-established in
air pollution studies (7). Cubic splines of time with equally
spaced knots were used to control for secular trends (e.g.,
demographic shifts) and confounding by seasonally varying factors other than temperature. We used 7 df per year
for these splines (approximately equivalent to a 2-month
moving average). This number of degrees of freedom was
chosen as a compromise between providing adequate control for unmeasured confounders and leaving sufficient information from which to estimate temperature effects. Plots
of partial autocorrelation functions and of fitted values over
time (see Web figure 1, which is posted on the Journal’s
website (http://www.aje.oxfordjournals.org)) suggested that
this was an adequate amount of adjustment for seasonal
trends. The amount of variation explained by the regression model in Gaussian form was at least 85 percent in
each case.
For graphical presentation, natural cubic splines (cubic
splines constrained to be linear beyond the data range) were
used to construct graphs of mortality as smoothed functions
of temperature, with 1 df for every 5C of temperature.
To quantify any effect of cold on death, we used simple
linear ‘‘hockey-stick’’ models—models which assume a
log-linear increase in risk below a threshold of 15C for the
average temperatures from week 0 and week 1 (i.e., the
mean of lag zero and 1 week). This was chosen as the most
appropriate common threshold based on the temperaturemortality graphs and the likelihood profiles for the data
set of combined periods obtained by maximum likelihood
estimation. In addition, when estimating the cold thresholds
separately for each time period, we found a value of 15C to
be compatible for each decade except the final one, where
a threshold of 19C (95 percent confidence interval (CI): 18,
20) was estimated. We assumed a common threshold in the
analysis in order to focus on one parameter of interest—
the slopes—in all of our models. Likelihood profiles of
the combined data set suggested a heat threshold at 15C
as well; thus, for heat-related deaths, we assumed a loglinear increase in risk above this same threshold.
We undertook sensitivity analyses to examine the influence on results of varying the degree of seasonal control
(df/year) and of including air pollution data (later two periods
of study) and influenza data and 4-df natural cubic splines
of weekly mean relative humidity (latest period only).
All analyses were conducted in Stata 8.2 (8).
RESULTS
Epidemiologic trends
The data for the four 20th-century time periods show the
epidemiologic transition over the century from high childhood mortality to low childhood mortality and towards a predominance of chronic disease mortality at older ages in the
later periods (table 1). The proportion of deaths occurring
among children under age 15 years fell from 38.5 percent in
1900–1910 to 1.5 percent in 1986–1996, while deaths in the
age group 65 years increased from 29.4 percent to 79.7
percent. At the same time, the proportion of deaths due to
cardiovascular disease rose from 12.1 percent to 42.3 percent, while deaths from other causes declined. Note that the
change in weekly counts (numbers) of deaths reflects, on the
one hand, an increasing population and, on the other, falling
age-specific mortality. Some change can also be attributed
to change in the geographic/population coverage of the data
sources.
As might be expected, there were comparatively small
differences between periods in weekly mean, minimum,
and maximum temperatures. The warmest of the four periods was 1986–1996, and the coolest was 1954–1965.
Am J Epidemiol 2006;164:77–84
Vulnerability to Temperature-related Mortality
79
TABLE 1. Mortality patterns and temperatures in London, United Kingdom, over the course of the 20th century, by period
Period
1900–1910
1927–1937
1954–1964
1986–1996
No. of observations (weeks)
564
572
559
572
Mean no. of deaths per week
1,431.1 (1,036–1,866)*
1,007.5 (713–1,579)
810.0 (614–1,176)
1,318.9 (1,108–1,644)
Percentage of deaths by age groupy
Children (<15 years)
38.5
13.3
4.9
1.5
Adults (15–64 years)
32.0
40.5
31.4
18.8
Elderly (65 years)
29.4
46.1
63.7
79.7
Cardiovascular disease
12.1
27.9
33.3
42.3
Respiratory
18.9
20.0
14.1
14.0
Noncardiorespiratory
69.0
52.1
52.6
43.7
Percentage of deaths by cause
Weekly mean of daily temperatures (C)
Daily mean temperatures
10.0 (2.1–1.8)
10.4 (2.4–19)
9.6 (0.9–16.8)
10.8 (2.6–19.4)
Daily maximum temperatures
16.8 (7.5–27.1)
17.2 (7.9–27.7)
14.5 (5.1–23.3)
18.2 (8.4–29.2)
Daily minimum temperatures
3.2 (ÿ4.3–11.2)
3.7 (ÿ3.7–11.6)
4.0 (ÿ3.9–12.1)
2.8 (ÿ4.5–10.8)
* Numbers in parentheses, 5th–95th percentile range.
y For the first time period, the age groupings were <20, 20–59, and 60 years.
Seasonal mortality
Figure 1 shows the time series of weekly all-cause mortality in London over the four time periods. The seasonal
fluctuation in deaths had a more complex pattern in 1900–
1910 than in the other periods, with indications of summer
as well as winter peaks. Patterns by cause of death are
shown in Web figure 2 (http://www.aje.oxfordjournals.org).
Table 2 shows that the ratio of winter deaths to nonwinter
deaths was 1.24 (95 percent CI: 1.16, 1.34) in 1900–1910,
1900–1910
200
0
100
100
0
1900
1902
1904
1906
1908
1910
1927
1929
1933
1935
1937
1994
1996
300
200
300
100
200
0
0
1954
1931
1986–1996
1954–1964
100
Ratio of observed/expected deaths
200
300
300
1927–1937
1956
1958
1960
1962
1964
1986
1988
1990
1992
FIGURE 1. Pattern of weekly all-cause mortality in London, United Kingdom, over the course of the 20th century, by period.
Am J Epidemiol 2006;164:77–84
80
Carson et al.
TABLE 2. Ratio of winter:nonwinter mortality and percentage of deaths attributable to the winter season in London, United
Kingdom, over the course of the 20th century, by period and cause of death
Period
1900–1910
Estimate
95% CI*
1927–1937
Estimate
95% CI
1954–1964
Estimate
95% CI
1986–1996
Estimate
95% CI
Ratio of winter death rates to
nonwinter death ratesy
All causes
1.24
1.16, 1.34
1.54
1.42, 1.68
1.48
1.35, 1.64
1.22
1.13, 1.31
Cardiovascular disease
1.30
1.05, 1.60
1.53
1.31, 1.81
1.45
1.23, 1.71
1.23
1.10, 1.38
Respiratory
2.05
1.74, 2.45
2.32
1.92, 2.86
2.70
2.09, 3.61
1.59
1.31, 1.96
Noncardiorespiratory
1.07
0.98, 1.17
1.31
1.16, 1.48
1.23
1.07, 1.41
1.10
0.98, 1.23
Percentage of deaths attributable to
winter
All causes
7.31
4.82, 9.86
15.0
11.9, 18.1
13.7
10.2, 17.3
6.55
4.00, 9.17
Cardiovascular disease
8.73
1.68, 16.3
14.7
8.99, 20.8
12.8
6.93, 18.9
6.95
3.05, 11.0
19.3, 32.0
30.0
22.9, 37.6
35.8
26.3, 46.0
Respiratory
Noncardiorespiratory
25.4
2.18
ÿ0.70, 5.15
9.06
4.98, 13.3
6.93
2.34, 11.7
16.1
3.13
9.03, 23.6
ÿ0.60, 7.02
* CI, confidence interval.
y Winter was defined as weeks wholly within the months of December, January, February, and March; the nonwinter period was defined as
weeks wholly outside of these months.
1.54 (95 percent CI: 1.42, 1.68) in 1927–1937, 1.48 (95
percent CI: 1.35, 1.64) in 1954–1964, and 1.22 (95 percent
CI: 1.13, 1.31) in 1986–1996. The amplitude of the winter
rise, largest in 1927–1937, diminished in subsequent
periods. The largest contrast was between 1986–1996 and
earlier years. The change in the seasonal pattern for cardiovascular and respiratory disease broadly paralleled that of
all-cause mortality (figure 2), although, from table 2, the
greatest winter:nonwinter ratio for respiratory death was
in 1954–1964.
The percentage of deaths attributable to the winter season
reflects the change in the winter:nonwinter ratio of deaths.
In proportionate terms, the largest attributable fraction in
each of the four periods was for deaths from respiratory
disease, maximal at just under 36 percent in 1954–1965.
The percentage of all cardiovascular deaths attributable to
cold was greatest in 1927–1937, at 14.7 percent; it declined
to less than 7 percent in the last period of analysis.
Temperature-mortality relation
Plots of the temperature-mortality relation for all-cause
mortality (figure 3) showed an increase in risk at low temperatures in each period, but the strength of association
gradually declined over the century. Relations by cause of
death and age group are shown in Web figures 3 and 4
(http://www.aje.oxfordjournals.org). Table 3 quantifies both
the heat effect and the cold effect under models assumed to
follow a log-linear relation. This confirms that the gradient
of the (low) temperature-mortality relation declined progressively over the century, from a 2.5 percent increase in
mortality for each degree-Celsius fall in temperature below
15C in 1900–1910 to approximately a 1.2 percent increase
in mortality per degree-Celsius fall in temperature in 1986–
1996. The steepest temperature-related gradients were seen
for respiratory and cardiovascular death. The gradient for
cardiovascular disease declined substantially across the four
periods (from an approximately 3.4 percent increase per
degree Celsius to approximately 1 percent), while that for
respiratory disease showed a more moderate decline (from
3.9 percent to 2.8 percent). Corresponding to these declines
in the temperature-mortality gradient, there was also a decline in the proportion of deaths attributable to cold, from
more than 12.5 percent in 1900–1910 to 5.4 percent in
1986–1996.
In 1986–1996, the proportion of respiratory deaths attributable to cold (12.6 percent) was higher than the proportions
for cardiovascular disease death (4.6 percent) and noncardiorespiratory death (3.6 percent), yet respiratory deaths
made up a smaller proportion of all cold-related deaths than
cardiovascular deaths because of the lower overall frequency of respiratory death.
The temperature-mortality plots (figure 3) and hockeystick models (table 3) also provided some indication of
heat-related mortality in the earlier periods of analysis, but
not in 1954–1964 or 1986–1996. In the earliest periods, the
evidence was strongest for noncardiovascular disease. By
1986–1996, mortality appeared to continue to decline at
mean temperatures above 15C.
To check that the results of the temperature-mortality
gradient were not substantially biased by omission of air
pollution as an explanatory factor, we constructed models
that included pollution measures for selected periods for
which air pollution data were obtainable. For 1992–1996,
the percent increase in mortality per degree Celsius below
the cold threshold was 1.27 (95 percent CI: 0.86, 1.68)
Am J Epidemiol 2006;164:77–84
Vulnerability to Temperature-related Mortality
39
10 15 20 25
5
0
26
1954–1964
1986–1996
26
39
52
Week
39
52
39
52
10 15 20 25
5
0
-5
10 15 20 25
5
0
13
13
Week
-5
1
-5
1
52
Mean temperature (°C)
26
Week
100 200 300 400 500
13
1927–1937
100 200 300 400 500
-5
0
5
10 15 20 25
100 200 300 400 500
1
100 200 300 400 500
Mortality count (as % of minimum)
1900–1910
81
1
13
26
Week
FIGURE 2. Average seasonal fluctuation in mortality and mean weekly temperature in London, United Kingdom, over the course of the 20th
century, by period. Mortality curves are expressed as percentages of the minimum mortality rate for the year. Dotted-and-dashed line (– d –),
cardiovascular mortality; solid line (—), respiratory mortality; dotted line (d d d), noncardiorespiratory mortality; dashed line (- - -), temperature.
without adjustment for air pollution or influenza, 1.27 (95
percent CI: 0.86, 1.69) with inclusion of weekly mean particulate matter 10 lm in diameter, and 1.32 (95 percent CI:
0.90, 1.75) with additional adjustment for influenza A. In
1954–1964, the figures were 1.64 (95 percent CI: 1.10, 2.19)
without adjustment for pollution, 1.85 (95 percent CI: 1.30,
2.40) with adjustment for particulate pollution measured by
the Owen’s Smoke Filter, and 1.91 (95 percent CI: 1.30,
2.52) with pollution adjustment and omission of years
(1957 and 1958) with high influenza counts. Sensitivity
analyses also indicated that, for the final period, control
for relative humidity left both heat and cold estimates
largely unchanged; and, in all decades, when either fewer
degrees of freedom were used in the smoothing splines for
the seasonal control (4 df/year) or more were used (10 df/
year and 12 df/year), the relative differences in slopes between the decades remained comparable.
DISCUSSION
Temperature-related mortality is of current scientific and
public health interest in the United Kingdom because of the
persistently high number of excess winter deaths (9) (which
is paradoxically higher in Britain than in many other colderAm J Epidemiol 2006;164:77–84
winter countries of continental Europe (10)) and also, more
generally, because of debates about the vulnerability of
European and other populations to the projected increased
frequency of heat waves under global climate change (11).
The study on which we report here provides a novel perspective on these issues by yielding evidence about the
change over time in population susceptibility to the health
impacts of heat and cold and the extent to which such susceptibility appears to be determined by broadly wealthrelated factors. To our knowledge, it is the only such study
that has examined the change in vulnerability in a single
location, and it is the more informative because this period
coincides with rapid demographic and epidemiologic transitions that many other countries are also likely to undergo
over the next century.
Its results provide persuasive evidence that vulnerability
to cold in particular, but perhaps to heat as well, has decreased over the course of the 20th century, despite the
aging of the population and a progressive increase in the
prevalence of cardiorespiratory disease. This finding, which
is consistent with previous reports of seasonal (12) and coldrelated (13) mortality across more limited time spans, suggests that protective influences have outweighed the trends
of aging and increasing prevalence of cardiorespiratory disease which would otherwise tend to increase susceptibility.
82
Carson et al.
1927–1937
40
20
40
0
20
-20
0
-20
-10
0
10
20
-10
30
0
10
20
30
20
30
1986–1996
-20
0
0
20
20
40
40
60
60
1954–1964
-20
% change in mortality relative to annual mean
60
60
1900–1910
-10
0
10
20
30
-10
0
10
Mean temperature (°C)
FIGURE 3. Percent change in all-cause mortality by mean weekly temperature in London, United Kingdom, over the course of the 20th century,
by period. Graphs are based on cubic smoothing splines (1 df per 5C of temperature). Results were adjusted for season.
The decline in vulnerability is apparent both from the
simpler seasonal analyses and, more conclusively, from
the specific temperature-based analyses and plots. The
changes in seasonal patterns are arguably the more complicated to interpret, because they reflect a range of factors that
vary from season to season. It seems probable, for example,
that the apparent increase in the ratio of winter deaths to
nonwinter deaths between 1900–1910 and 1927–1937 predominantly reflects changes in summer deaths rather than an
increase in risk of winter death. There is irregularity in the
weekly counts of deaths in 1900–1910 compared with subsequent periods, with indications of summer peaks as well
as winter peaks (figure 1). This may well reflect the relatively high mortality from diarrheal disease, the prevalence
of most bacterial forms of which increases during warmer
weather. During the early part of the 20th century, the death
rate from diarrheal disease (dysentery, cholera, enteritis) fell
substantially, from 1,100 per million persons to 140 per
million persons between 1900 and 1930 (12), and we may
presume a corresponding fall in summer death rates, particularly in children.
The evidence of the temperature-mortality (time-series)
analyses was very clear for cold deaths, which essentially
showed a monotonic decline over the course of the century.
The pattern for heat-related deaths was not as clear. There
was evidence of heat-related deaths in the first half of the
century, particularly for non-cardiovascular disease deaths,
but not in the second half for any cause of death. Patterns of
heat-related death are likely to have been influenced by the
weekly aggregation of data in these analyses. Cardiovascular heat deaths appear to occur at very short lags (0–1 day)
and may in part be due to short-term displacement (14, 15),
so it is likely that an effect of heat on daily mortality would
have been attenuated in our weekly-aggregated analysis.
The weekly aggregation may not have attenuated noncardiorespiratory deaths, however. Previous analyses in populations where noncardiorespiratory causes are principal
contributors to heat deaths showed that the time lag may
be longer and less influenced by short-term mortality displacement (15). Thus, it is possible that the heat deaths
apparent in the first two time periods largely reflect diarrhea
and infectious disease deaths, which were much reduced by
the second half of the century.
The historical nature of our data gave rise to some
limitations of analysis, specifically from the absence of air
pollution and influenza A data for most periods, but our
sensitivity analyses suggest that this is unlikely to have
materially altered the pattern of results. Similarly, changes
may have occurred over time in the recording and classification of specific causes of death, but this should not affect
analyses of all-cause mortality, and any bias in the causespecific results should have been minimized by our use of
broad categories from the International Classification of
Diseases.
Given the increase in population age over the 20th century, the decline in vulnerability to cold and heat is most
readily explained by beneficial changes relating to increasing wealth. This conclusion is supported by the age-specific
Am J Epidemiol 2006;164:77–84
Vulnerability to Temperature-related Mortality
83
TABLE 3. Gradients of the temperature-mortality relation for cold and heat and the percentage of deaths attributable to each factor
over the course of the 20th century, by period and cause of death, London, United Kingdom
Period
1900–1910
Estimate
95% CI*
1927–1937
Estimate
95% CI
1954–1964
Estimate
95% CI
1986–1996
Estimate
95% CI
Cold-related mortality
Gradient of cold slopey
All causes
2.52
2.00, 3.03
2.34
1.72, 2.96
1.64
1.10, 2.19
1.17
0.88, 1.45
Cardiovascular disease
3.40
2.76, 4.04
2.54
1.88, 3.22
2.27
1.63, 2.91
0.99
0.62, 1.36
Respiratory
3.93
3.11, 4.75
2.90
1.85, 3.95
2.39
1.17, 3.62
2.84
2.19, 3.50
Noncardiorespiratory
1.86
1.29, 2.43
2.02
1.31, 2.73
0.86
0.31, 1.40
0.76
0.45, 1.07
8.74
5.93, 11.5
5.42
4.13, 6.69
4.60
2.92, 6.25
Percentage of deaths
attributable to cold
All causes
12.5
10.1, 14.9
11.2
8.40, 14.0
Cardiovascular disease
16.5
13.7, 19.2
12.1
9.13, 15.1
11.8
8.67, 14.9
Respiratory
18.8
15.2, 22.2
13.7
9.03, 18.1
12.4
6.34, 18.1
Noncardiorespiratory
9.44
6.66, 12.1
9.77
6.48, 13.0
4.67
1.72, 7.53
12.6
3.55
9.86, 15.2
2.13, 4.95
Heat-related mortality
Gradient of heat slopez
1.02
ÿ0.16, 2.21
1.53
0.15, 2.93
0.29
ÿ1.95, 2.59
ÿ1.34
ÿ1.94, –0.75
ÿ0.41
ÿ1.96, 1.16
1.29
ÿ0.18, 2.78
0.11
ÿ2.51, 2.79
ÿ1.79
ÿ2.57, –1.00
Respiratory
1.84
ÿ0.82, 4.57
2.45
ÿ0.45, 5.43
2.35
ÿ4.40, 9.57
ÿ0.84
ÿ2.31, 0.65
Noncardiorespiratory
1.25
0.05, 2.47
1.37
ÿ0.08, 2.83
0.30
ÿ1.75, 2.39
ÿ1.03
ÿ1.65, –0.41
All causes
Cardiovascular disease
Percentage of deaths
attributable to heat
0.40
ÿ0.06, 0.86
0.89
0.09, 1.69
0.06
ÿ0.39, 0.50
ÿ0.90
ÿ1.31, –0.50
ÿ0.16
ÿ0.79, 0.45
0.76
ÿ0.10, 1.61
0.02
ÿ0.49, 0.53
ÿ1.21
ÿ1.74, –0.67
Respiratory
0.72
ÿ0.33, 1.75
1.42
ÿ0.27, 3.08
0.45
ÿ0.88, 1.76
ÿ0.56
ÿ1.56, 0.43
Noncardiorespiratory
0.49
0.02, 0.96
0.80
ÿ0.04, 1.64
0.06
ÿ0.34, 0.46
ÿ0.69
ÿ1.11, ÿ0.27
All causes
Cardiovascular disease
* CI, confidence interval.
y Percent increase in deaths per degree Celsius below 15C.
z Percent increase in deaths per degree Celsius above 15C.
results, which suggest more impressive downward trends
(at least in the age groups 15–64 and 65 years) than do
analyses based on all-ages mortality as a whole (see Web
figure 5 (http://www.aje.oxfordjournals.org)). We cannot
quantify or even identify all of the modifying factors that
have contributed to this reduced susceptibility, but probable
candidates include developments in health care (including
influenza vaccination (16)), improved nutrition (especially
during winter months), better support services, and improved housing. Over the course of the century, improvement in life expectancy was gradual, but the rate of increase
was significantly greater in the first half of the century than
in the later half. However, the increase in gross domestic
product per person was faster in the second half of the century: 0.7 percent per annum between 1900 and 1948 as compared with 2.2 percent per annum in the period 1948–1998
(17). Stepwise changes in access to health care occurred in
the United Kingdom with the introduction of national health
insurance (1911) and the establishment of the National
Health Service (1948), while important advances occurred
Am J Epidemiol 2006;164:77–84
in medical and surgical treatment of myocardial ischemia in
the last quarter of the century. Indoor temperatures rose as
a result of improvements in building materials and standards, such that even today, the age of a property remains an
important determinant of winter temperatures (9, 18), which
in turn correlate with the risk of winter death (9). Increased
car ownership, climate-controlled transportation and shopping facilities, and improved clothing fabrics are also likely
to have reduced potentially important outdoor exposure to
cold (19–21) and, to a lesser degree, heat. Although the
contribution of each of these specific factors is unclear, it
is evident that public health threats from high and low temperatures in London have been substantially modified over
time. It is reasonable to conclude that a similar modification of risk will occur among populations in other settings,
particularly in low- and middle-income countries, as they
grow richer.
In conclusion, the observed patterns of seasonal and
temperature-related mortality in London suggest that there
has been a substantial reduction in population vulnerability
84
Carson et al.
to temperatures over the 20th century which is likely to
reflect changes related to increasing wealth. This observation has bearing on the likely future burdens of heat and cold
mortality in the context of social and environmental change,
both in the United Kingdom and elsewhere.
10.
11.
ACKNOWLEDGMENTS
Dr. Paul Wilkinson is supported by a Public Health Career
Scientist Award (NHS Executive grant CCB/BS/PHCS031).
Conflict of interest: none declared.
12.
13.
14.
REFERENCES
15.
1. Kovats S, Wolf T, Menne B. Heatwave of August 2003 in
Europe: provisional estimates of the impact on mortality.
Eurosurveill Wkly 2004;8(11). (Electronic article). (http://
www.eurosurveillance.org/ew/2004/040311.asp#7).
2. World Health Organization and European Environment
Agency. Extreme weather and climate events and public health
responses. Report on a WHO meeting, Bratislava, Slovakia,
9–10 February 2004. Copenhagen, Denmark: World Health
Organization and European Environment Agency, 2004.
(http://www.euro.who.int/document/E83004.pdf).
3. Curwen M. Excess winter mortality: a British phenomenon?
Health Trends 1990/91;22:169–75.
4. Aylin P, Morris S, Wakefield J, et al. Temperature, housing,
deprivation and their relationship to excess winter mortality in
Great Britain, 1986–1996. Int J Epidemiol 2001;30:1100–8.
5. Registrar General. The Registrar General’s weekly returns
for England and Wales. London, United Kingdom: HMSO,
1954–1964.
6. Magill PL. Analytical methods—particulate matter. In:
Magill PL, Holden FR, Ackley C. Air pollution handbook
1956. New York, NY: McGraw-Hill, 1956:23–24.
7. Dominici F, Sheppard L, Clyde M. Effects of air pollution:
a statistical review. Int Stat Rev 2003;71:243–76.
8. Stata Corporation. Stata statistical software, release 8.2. College Station, TX: Stata Corporation, 2005.
9. Wilkinson P, Landon M, Armstrong B, et al. Cold comfort: the
social and environmental determinants of excess winter death
16.
17.
18.
19.
20.
21.
in England, 1986–1996. York, United Kingdom: Joseph
Rowntree Foundation, 2001.
Healy JD. Excess winter mortality in Europe: a cross country
analysis identifying key risk factors. J Epidemiol Community
Health 2003;57:784–9.
Houghton JT, Ding Y, Griggs DJ, et al, eds. Climate change
2001: the scientific basis. The contribution of Working Group I
to the Third Assessment Report of the Intergovernmental
Panel on Climate Change. New York, NY: Cambridge University Press, 2001.
McDowall M. Long term trends in seasonal mortality. Popul
Trends 1981;26:16–19.
Donaldson GC, Keatinge WR. Mortality related to cold
weather in elderly people in southeast England, 1979–94. BMJ
1997;315:1055–6.
Pattenden S, Nikiforov B, Armstrong B. Mortality and temperature in Sofia and London. J Epidemiol Community Health
2003;57:628–33.
Hajat S, Armstrong BG, Gouveia N, et al. Mortality displacement of heat-related deaths: a comparison of Delhi,
Sao Paulo, and London. Epidemiology 2005;16:613–20.
Armstrong BG, Mangtani P, Fletcher A, et al. Effect of influenza vaccination on excess deaths occurring during periods
of high circulation of influenza: cohort study in elderly people.
BMJ 2004;329:660. (Electronic article). (http://bmj.com/cgi/
content/abstract/329/7467/660).
Hicks J, Allen G. A century of change: trends in UK statistics
since 1900. (House of Commons research paper 99/111).
London, United Kingdom: House of Commons Library, 1999.
(http://www.parliament.uk/commons/lib/research/rp99/rp99111.pdf).
Oreszczyn T, Hong SH, Ridley I, et al. Determinants of winter
indoor temperatures in low income households in England.
Energy and Buildings 2005;38:245–52.
Keatinge WR, Coleshaw SR, Holmes J. Changes in seasonal
mortalities with improvement in home heating in England
and Wales from 1964 to 1984. Int J Biometeorol 1989;33:
71–6.
The Eurowinter Group. Cold exposure and winter mortality
from ischaemic heart disease, cerebrovascular disease, respiratory disease, and all causes in warm and cold regions of
Europe. Lancet 1997;349:1341–6.
Donaldson GC, Keatinge WR. Cold related mortality in
England and Wales; influence of social class in working and
retired age groups. J Epidemiol Community Health 2003;
57:790–1.
Am J Epidemiol 2006;164:77–84