American Journal of Epidemiology
Copyright © 1999 by The Johns Hopkins University School of Hygiene and Public Health
All rights reserved
Vol. 149, No. 12
Printed in U.S.A.
Effects of Temperature and Snowfall on Mortality in Pennsylvania
Michael L. Gorjanc,1 W. Dana Flanders,2 James VanDerslice,3 Joel Hersh,4 and Josephine Malilay1
The relation between exposure to severe cold weather and mortality is examined in a retrospective study of
deaths occurring during the month of January from 1991 to 1996 in Pennsylvania. Using division-days as units
of observation (n = 1,560) aggregated from death certificates and geographic divisions, the authors estimated
mortality rates for total deaths and deaths due to ischemic heart disease, cerebrovascular diseases, and
respiratory diseases by analyses based on generalized estimating equations. Total mortality increased on days
of "extreme" climatic conditions, that is, when snowfall was greater than 3 cm and when temperatures were
below -7°C (rate ratio (RR) = 1.27, 95 percent confidence interval (Cl) 1.12-1.44). On days of extreme
conditions, mortality due to ischemic heart diseases tripled among males aged 35-49 years (RR = 3.54, 95
percent Cl 2.35-5.35), increased for men aged 50-64 years (RR = 1.77, 95 percent Cl 1.32-2.38), and rose for
males aged 65 years and older (RR = 1.58, 95 percent Cl 1.37-1.82), when compared with milder conditions.
Among females, mortality for those aged 65 years and older increased for respiratory causes (RR = 1.68, 95
percent Cl 1.28-2.21) and cerebrovascular causes (RR = 1.47, 95 percent Cl 1.13-1.91). Cold and snow
exposure may be hazardous among men as young as 35 years. Am J Epidemiol 1999; 149:1152-60.
cerebrovascular disorders; cold climate; heart diseases; mortality; respiration disorders; weather
On January 6-8, 1996, severe winter storms struck
the eastern United States. From Arkansas east through
the Carolinas and north to Maine, the death toll attributed to the blizzard of 1996 rose to 154 people. More
than half of these deaths occurred in the state of
Pennsylvania alone, where daily snowfalls at times
exceeded 30 inches (76.2 cm). Some weather stations
reported more than 50 inches (127 cm) of snow on the
ground in the days after the blizzard (1, 2). January
temperature readings in Pennsylvania fluctuated from
a high of 67°F (19.4°C) to a low of -27°F (-32.8'C).
Previous studies have suggested that weather events
can affect mortality. In Massachusetts and Rhode
Island, Glass and Zack (3) observed that deaths from
ischemic heart disease rose significantly (8 percent)
during and in the week immediately following blizzards. Deaths from ischemic heart disease accounted
for 90 percent of these excess deaths. Faich and Rose
(4) also reported that mortality increased from 27
deaths per day (the average number of deaths from the
same time period of the previous 5 years) to 48 deaths
on the day of the storm. Larsen (5, 6) found that unusually cold weather from September to June was associated with a proportional increase in the crude death
rate in the United States and that cold temperatures
were followed by higher mortality for all cause-ofdeath categories except malignant neoplasms. A similar study in Britain (7) determined that deaths from circulatory and respiratory diseases rose markedly during
the winter months. Gyllerup et al. (8) found significant
associations between cardiovascular mortality and
temperature exposure among both sexes and all age
groups. The strongest association was observed for
men 40-64 years of age. Cold temperatures explained
coronary mortality better than the four strongest
socioeconomic factors combined and, for men aged
40-64 years, temperature exposure was more strongly
associated with coronary mortality than was a combination of tobacco use, serum cholesterol, and the use
of antihypertensive drugs (9, 10).
Both Bull and Morton (11) and Rogot and Padgett
(12) have described the lagged effects of temperature
on mortality for different causes of death.
Temperatures 1-2 days before death were most highly
Received for publication May 11, 1998, and accepted for publication October 13, 1998.
Abbreviations: Cl, confidence interval; ICD-9, International
Classification of Diseases, Ninth Revision; RR, rate ratio.
1
Environmental Hazards Epidemiology Section, Health Studies
Branch, Division of Environmental Hazards and Health Effects,
National Center for Environmental Health, Centers for Disease
Control and Prevention, Atlanta, GA.
2
Department of Epidemiology, The Rollins School of Public
Health of Emory University, Atlanta, GA.
3
University of Texas-Houston School of Public Health, El Paso,
TX.
4
Bureau of Epidemiology, Pennsylvania Department of Health,
Harrisburg, PA.
Reprint requests to Dr. Josephine Malilay, Environmental Hazards
Epidemiology Section, Health Studies Branch, Division of
Environmental Hazards and Health Effects, National Center for
Environmental Health, Centers for Disease Control and Prevention,
4770 Buford Highway, NE, Mailstop F-46, Atlanta, GA 30341-3724.
1152
Effects of Temperatures and Snowfall on Mortality
associated with deaths from ischemic heart disease,
whereas deaths from cerebrovascular diseases were
associated with temperatures 3 days before the day of
death. In the case of respiratory diseases such as pneumonia, Bull and Morton noted the longest lag of 7
days.
Physiologic changes during exposure to cold have
been shown to contribute to cardiovascular causes of
death. After 6 hours of mild surface cooling of subjects, Keatinge et al. (13) found that subjects' blood
viscosity rose 21 percent with the fraction of plasma
occupied by platelets increasing by 15 percent. The
increases in platelets started after 1 hour in the cold
and may not have reached a peak even after 6 hours of
experimental cooling. Keatinge et al. suggest this lag
may account for the approximate 24-hour delay seen
between the fall in temperatures and the peak in the
number of deaths from coronary thrombosis reported
elsewhere (12). The effects of these biologic mechanisms can be exacerbated by the additional stress of
physical activity such as snow shoveling (14, 15).
Although few studies have focused on the association of mortality and snowfall, previous studies have
demonstrated an association between mortality and
extremely cold temperatures. However, the point at
which climatic conditions reach a critical mass at
which excess deaths begin to occur is unknown. This
study presents the results of the effects of temperature
and snowfall on daily mortality stratified by age and
sex. By addressing the severity of exposure and the climatic conditions at which deaths begin to occur, we
hope that public health recommendations can be targeted to appropriate risk groups to prevent mortality
from cold-weather events.
MATERIALS AND METHODS
We used mortality data derived from death certificates supplied by the Pennsylvania Department of
Health and weather data obtained from the National
Climatic Data Center. The study period covered 6
months, each January from 1991 through 1996 (2, 16).
However, to eliminate the effects on mortality that
occur in association with the New Year holiday, only
data from January 6 to January 31 were used. The
weather data included daily temperature and precipitation observations, recorded at 146 weather stations
throughout Pennsylvania. Mortality and weather data
were aggregated temporally by day and geographically
by "weather division," regions of the state defined by
the National Oceanic and Atmospheric Administration
as having "similar climatological characteristics" (2).
With one exception, weather divisions were defined by
the external boundaries formed by groups of contiguous counties. The use of weather division as the unit of
Am J Epidemiol Vol. 149, No. 12, 1999
1153
geographic aggregation ensured sufficient numbers of
deaths per day for analysis.
Daily averages of high temperatures and snowfalls
were computed for each division by using data from all
stations located in a given weather division. We chose
high temperatures as the measure of temperature exposure, since they generally occur during daytime hours,
when people are most likely to be outdoors.
The death certificate data included all recorded deaths
during the study period. For each death, we obtained
age, sex, underlying cause of death, and day, year, and
county of death. Deaths were grouped by sex and the
following age groups: 0-34 years, 3 5 ^ 9 years, 50-64
years, and 65 years and older. Further, we classified
deaths according to the International Classification of
Diseases, Ninth Revision (ICD-9), into the following
categories: ischemic heart diseases (ICD-9 codes
410-414), cerebrovascular diseases (ICD-9 codes
430-438), and respiratory diseases (ICD-9 codes
460-519) (17). Deaths were assigned to a weather division based on the county where the death occurred. In
the single case where a county was bisected by a
weather-division boundary, the population distribution
of that county was spatially analyzed using a geographic information system and 1990 census data, and
the proportion of deaths assigned to each weather division was based on the proportion of population residing
in that weather division (18). Thus, the data set used in
this analysis consisted of daily mortality and weather
observations for each of the 10 weather divisions, resulting in 1,560 "division-days" of observation.
To explore the relations between mortality and
weather, we aggregated cause-specific deaths into
groups based on the weather conditions during which
they occurred. Mean division-day snowfall and high
temperatures were rounded to the nearest centimeter or
degree, respectively. This resulted in mutually exclusive, ordered categories of snowfall and temperature to
which each death was assigned. A death rate for each
snowfall and temperature category was then computed
in which the numerator consisted of the total number of
deaths that occurred on a division-day with that specific
snowfall or temperature. The denominator of the rate
was computed by summing the corresponding exposed
populations from the weather division in which each
death occurred. The resulting cause-specific crude death
rates were then plotted against the rounded snowfall and
temperature values.
We also computed mortality rates based on personyears of exposure to selected climatic conditions for
both total and cause-specific deaths. Using breaks in
their distributions, we divided division-day weather
averages into the following categories: snowfall, <3 and
>3 cm; temperature, <-7°C, >-7°C to <0°C, and >0°C.
1154
Gorjanc et al.
A finer stratification of weather variables using five
snow categories and six temperature categories yielded
qualitatively similar results, but it resulted in too few
division-days in some categories for stable analysis.
Thus, we assigned total and cause-specific deaths to the
appropriate snow-temperature categories and computed
mortality rates for each weather category.
We used generalized estimating equations to model
the effects of temperature and snowfall on mortality
(19, 20). In these analyses, we allowed for temporal
correlation and treated outcomes in different weather
divisions as independent. To assess temporal correlation, the Durbin-Watson statistic was calculated for
total deaths by weather-division. Independent clusters were defined as all observations in each of 10
weather divisions for each of six Januarys (n = 60).
For all models, first order, autoregressive, workingcorrelation matrices were used. To assess goodness of
fit, we used deviance and scaled Pearson chi-square,
assessed the importance of higher-order and interactive terms, and plotted residuals. We fit models for
men and women in each age group and for each
cause-of-death category. Because the number of
deaths in an area is affected by population, the division number was used as an independent variable to
account for differences in population and to control
for possible confounding by unobserved factors associated with different areas of the state. Division 6,
having the smallest population, was specified as the
reference division. To assess spatial correlations of
death rates, we produced a variogram and calculated
the correlation of distance between each pair of
weather-divisions and the corresponding pair's difference in death rates (21). We also studied the predictive ability of using different lag periods for the
weather variables. On the basis of the deviance
divided by the degrees of freedom for the Poisson
models, we subjectively selected 0-day lags for all
but respiratory outcomes, for which we used 3-day
lags (figure 1). Because the deviance suggested possible overdispersion, we estimated the scale parameter by dividing the deviance by the degrees of freedom (d-scale option).
We also conducted similar supplemental analyses
using an alternative generalized estimating equations
model. In the alternative generalized estimating equations model, we addressed spatial and temporal correlation by defining independent clusters (n = 18) as all
observations within successive 6- or 7-day periods
from all weather divisions. We divided January into
three periods (two 7-day periods and one 6-day
period) and allowed for the possibility that daily
observations within these periods were correlated.
Observations or subjects, consisting of the different
periods, were assumed to be independent. To ensure
independence of clusters, we omitted observations for
3 days between each successive period. For those outcomes with substantial numbers of deaths, results
1.3
1.28
1.26
total deaths
• respiratory deaths
1.24
I 1.22
1.2
1.18
3
4
lags in days
FIGURE 1. Lagged deviances from generalized estimating equations analysis of total deaths and deaths from respiratory causes,
Pennsylvania, 1991-1996. Total deaths include ischemic heart disease, cerebrovascular disease, and respiratory causes.
Am J Epidemiol
Vol. 149, No. 12, 1999
Effects of Temperatures and Snowfall on Mortality
TABLE 1.
Descriptive statistics for temperature and snowfall, Pennsylvania, 1991-1996
High temperature
Mean
temperature
CO
Mean
snowfall
(cm)
<-5°C
(division-days)
<0°C
(division-days)
>5°C
(division-days)
1991
1992
1993
1994
1995
1996
2.4
2.4
2.7
-3.2
4
1.2
0.79
0.59
0.31
2.24
0.26
2.87
14
25
2
85
4
45
70
83
58
177
64
109
75
85
54
19
88
64
2
0
0
16
0
20
Mean
1.6
1.17
22
93
64
6
from these supplemental analyses were qualitatively
similar to results from the models presented here.
Thus, we did not include results from these supplemental analyses.
RESULTS
Descriptive analysis
We observed significant variability in weather
across the six Januarys of the study period. Whereas
weather in 1991 through 1993 varied relatively little,
weather patterns in 1994 through 1996 varied substantially (table 1). January 1994 was clearly the coldest of
the 6 months, recording the lowest daily high temperature. Conditions were far more temperate in 1995,
when the warmest high and low temperatures were
recorded, as well as the lowest snowfall of the 6
months. Contrasting months based on arbitrary fiveunit categories, January 1996 was a month of extremes
with more cold days than any other January except in
1994, despite the fact that nearly 25 percent of the high
temperatures were over 5°C. The most division-days
of snowfall greater than 10 cm were also reported in
January 1996, largely because of the massive blizzard
of January 6-8. Far more days with snowfall greater
than 3 cm were recorded during 1994 and 1996 than
during any of the other years.
Of the 71,823 deaths that occurred throughout the
six study periods, the highest number of deaths was
recorded in 1994 with 13,485 (table 2), including the
highest number of deaths in each disease category. In
1994, deaths from respiratory diseases, ischemic heart
diseases, and cerebrovascular diseases were 27 percent, 17 percent, and 14 percent higher, respectively,
than for the six-January average. Similarly, the moderate overall death count recorded in 1996 was mirrored
by moderate numbers of cause-specific deaths, with
the exception of cerebrovascular deaths, which were 7
percent higher than the six-January average of 781
deaths. The Durbin-Watson statistic ranged from a low
of 1.14 to a high of 1.91 (mean = 1.63) across the 10
Am J Epidemiol
1155
Vol. 149, No. 12, 1999
Snowfall
>10cm
(division-days)
weather-divisions. Some of these values were significantly less than 2, particularly in the weather-divisions
with the largest populations, suggesting temporal autocorrelation for total deaths in at least some divisions.
Little correlation between the distance separating
weather divisions and the difference in their respective
death rates was noted (r = -0.086), and the variogram
did not suggest any pattern in the covariance.
Death rates for total deaths, ischemic heart diseases,
and cerebrovascular diseases showed slight to moderate
inverse relations with temperature. Deaths due to
ischemic heart diseases exhibited the most pronounced
association with temperature, increasing by as much as
40 percent over a 40°F drop in temperature from the
highest to lowest points. The relation described between
death rates for respiratory diseases and temperature was
less clear. Rates for total deaths and deaths due to
ischemic heart diseases were both positively associated
with the amount of snowfall, although there were actually fewer observations for days with the heaviest snowfall. The death rates for cerebrovascular and respiratory
causes did not demonstrate clear relations with snowfall.
The rate of total deaths and deaths by specific causes
generally increased as the weather became more severe
(table 3). Total deaths, deaths due to ischemic heart diseases, and deaths due to cerebrovascular diseases all
TABLE 2. Total deaths and deaths from selected causes,
Pennsylvania, 1991-1996
Total
no. of
deaths
No. of
ischemic
heart
disease
deaths
No. of
cerebrovascular
disease
deaths
2,537
2,766
2,479
1,006
1,399
1,016
1,568
1,275
1,165
6,856
7,238
6,916
7,869
7,376
7,338
1,071
7,265
1991
1992
1993
1994
1995
1996
11,135
12,172
11,106
13,485
11,927
11,996
3,158
2,517
2,656
736
769
695
890
759
837
Mean
11,971
2,685
781
deaths
No. of
deaths
from
all other
causes
1156
Gorjancetal.
TABLE 3. Total number of deaths, number of deaths from selected causes, and mortality rates for
selected snowfall and temperature categories, Pennsylvania, 1991-1996
Total no. of deaths
No. of ischemic
heart disease
deaths
No. of
cerebrovascular
disease deaths
No. of
respiratory
disease deaths
<-7
>-7 to <0
>0
4,491 (3.5)*
21,872 (3.2)
39,421 (3.2)
1,036(0.80)
4,802(0.71)
8,778(0.71)
299 (0.23)
1,446(0.21)
2,548 (0.21)
499 (0.38)
2,222 (0.33)
4,056 (0.33)
<-7
>-7 to <0
>0
820 (3.7)
3,946 (3.5)
1,311 (3.4)
209 (0.94)
935 (0.84)
333 (0.86)
56 (0.25)
259 (0.23)
76 (0.20)
76 (0.34)
441 (0.40)
127(0.33)
Snow
(cm)
Temperature
<3
>3
•Numbers in parentheses, mortality per 105person-years (rate).
increased with increasing snowfall and decreasing temperature. In addition, within each category of snowfall,
these death rates increased with decreasing temperature. Only deaths due to respiratory diseases did not follow this general dose-response pattern, although the
highest rate did occur when snowfall was over 3 cm.
Analyses based on generalized estimating equations
Initial analyses suggested that we would generally
obtain the smallest deviances and largest likelihoods
for a lag of zero for total deaths, ischemic heart disease
deaths, and cerebrovascular deaths. For respiratory
deaths, this occurred with a lag of 4 days. Therefore,
we used these lags in all subsequent models (figure 1).
Analyses also indicated that the association between
snowfall and mortality depended on temperatures
(effect modification). In general, analyses based on
generalized estimating equations tended to show that
decreasing temperature within a snowfall category was
associated with an increasing mortality rate. Similarly,
snowfall over 3 cm was generally associated with
higher rates of death.
Total deaths. Overall, results of the generalized
estimating equations analyses suggested that total mortality was higher on days with lower temperatures and
more snowfall than on other days. Rate ratios for total
deaths were highest in the most extreme weather categories. Total mortality was higher when snowfall was
greater than 3 cm and temperatures were above -7°C
(rate ratio (RR) = 1.13, 95 percent confidence interval
(CI) 1.07-1.20) than under conditions for the referent
exposure (>0°C and <3 cm of snowfall). A higher rate
was observed when snowfall exceeded 3 cm and temperatures were below -7°C (RR - 1.27, 95 percent CI
1.12-1.44). Within specific age categories, significant
increases in mortality were also noted, although many
of the confidence intervals only narrowly excluded one.
The oldest men and women experienced increases in
mortality rates, showing a dose-response pattern of
increased mortality as temperatures dropped within
each snow category (table 4). Significant increases in
mortality were also noted for men aged 50-64 years in
the high snowfall category, with rates increasing as the
temperatures dropped (table 5).
Deaths due to ischemic heart diseases. The risk
of death due to ischemic heart diseases was higher on
days when snowfall was higher and temperatures were
lower, a pattern similar to, although stronger than, that
for total deaths (tables 4-6). This was especially true
for men, being inversely associated with age. When
temperatures were below -7°C and snowfall was
greater than 3 cm, the mortality rate among men aged
35-49 years was 3.54 times higher than when temperatures were above freezing and snowfall was less than
3 cm (95 percent CI 2.35-5.35); among men aged
50-64 years, the rate was 1.77 times higher (95 percent
CI 1.32-2.38), while men aged 65 and older experienced rates 1.58 times higher (95 percent CI
1.37-1.82) (tables 4-6). A distinct dose-response relation existed between the severity of the weather conditions and mortality rates among the youngest and oldest men. The same relation was not observed for men
aged 50-64 years, although mortality was highest during the most severe weather conditions.
Among women, weather-related increases in death
rates from ischemic heart diseases were significant
only for those aged 65 years or older, and the rate was
highest when snowfall was greater than 3 cm and temperatures were below -7°C (RR = 1.29, 95 percent CI
1.11-1.49) (table 4). Within each snowfall category,
deaths increased with decreasing temperatures.
Deaths due to cerebrovascular diseases. Women
aged 65 years and older experienced a significantly
higher death rate for cerebrovascular diseases during
the periods when temperatures were below -7°C and
when snowfall was high (RR = 1.47, 95 percent CI
1.13-1.91) (table 4). The oldest men also experienced
increased mortality when snowfall was high; the rate
Am J Epidemiol Vol. 149, No. 12, 1999
Effects of Temperatures and Snowfall on Mortality
1157
TABLE 4. Rate ratios (RRs) and 95% confidence intervals (CIs) for mortality among people aged 65 years and older derived
from general estimating equations analysis for selected categories of temperature and snowfall, Pennsylvania, 1991-1996
Snow
(cm)
Temperature
(°C)
Both sexes
RR
Males
95% Cl
Females
95% Cl
RR
RR
95% Cl
Total deaths
Low (<3)
High (>3)
High (>0)
Mid (>-7 to 0)
Low (<-7)
High (>0)
Mid (>-7 to 0)
Low (<-7)
1.00*
1.03*
1.15*,*
1.10*,*
1.16*,*
1.28*,*
High (>0)
Mid (>-7 to 0)
Low (<-7)
High (>0)
Mid (>-7 to 0)
Low (<-7)
1.00
1.06*
1.20*
1.18*
1.25*
1.41*
High (>0)
Mid (>-7 to 0)
Low (<-7)
High (>0)
Mid (>-7 to 0)
Low (<-7)
1.00
1.05
1.19*
1.17*
1.23*
1.39*
Reference
0.96-1.15
1.01-1.39
1.05-1.31
1.08-1.40
1.15-1.69
High (>0)
Mid (>-7 to 0)
Low (<-7)
High (>0)
Mid (>-7 to 0)
Low (<-7)
1.00
1.04
1.18
1.21*
1.26*
1.43*
Reference
0.98-1.11
0.98-1.41
1.06-1.39
1.09-1.46
1.14-1.79
Reference*
0.99-1.06
1.10-1.21
1.01-1.19
1.08-1.25
1.14-1.43
1.00*
1.03*
1-16*,*
1-16*,*
1-16*,*
1.36*,*
Reference
0.98-1.07
1.07-1.25
1.07-1.24
1.07-1.26
1.10-1.67
1.00*
1.03*
1.16*,*
1.06*
1.17*,*
1.20*,*
Reference
0.99-1.08
1.08-1.25
0.95-1.19
1.08-1.28
1.01-1.43
Reference
0.97-1.13
1.14-1.37
1.13-1.41
1.18-1.49
1.37-1.82
1.00
1.08*
1.16*
1.11
1.20*
1.29*
Reference
1.04-1.14
1.06-1.26
0.98-1.26
1.06-1.36
1.11-1.49
Reference
0.91-1.14
0.89-1.40
0.96-1.45
0.97-1.48
0.97-1.78
1.00
1.08
1.24
1.18*
1.27*
1.47*
Reference
0.95-1.22
0.99-1.56
1.03-1.35
1.08-1.50
1.13-1.91
Reference
0.92-1.08
0.90-1.28
0.97-1.32
0.96-1.32
0.96-1.54
1.00
1.08
1.28*
1.31*
1.42*
1.68*
Reference
0.99-1.18
1.02-1.61
1.13-1.52
1.21-1.66
1.28-2.21
Deaths from ischemic heart diseases
Low (<3)
High (>3)
Reference
1.01-1.12
1.13-1.27
1.07-1.30
1.13-1.39
1.26-1.58
1.00
1.05
1.25*
1.26*
1.32*
1.58*
Deaths from cerebrovascular diseases
Low (<3)
High (>3)
1.00
1.02
1.12
1.18
1.20
1.32
Deaths from respiratory diseases
Low (<3)
High (>3)
1.00
0.99
1.07
1.13
1.12
1.21
* Interaction of snowfall and temperature was significant,
t Reference group.
* p < 0.05.
increased with falling temperatures and was highest
during the most extreme weather conditions, but confidence intervals included one (table 4).
Deaths due to respiratory diseases. Increases in
mortality rates from respiratory causes were significant only for women aged 65 years and older, with all
increases following the dose-response pattern seen in
other cause-of-death categories (table 4). During the
most severe weather conditions, mortality due to respiratory causes was over two thirds higher than when
temperatures were above 0°C and snowfall was below
3 cm (RR = 1.68, 95 percent Cl 1.28-2.21).
DISCUSSION
Results of this study support previous findings that
colder temperatures and increased snowfall tend to be
followed by an increased number of deaths (3-12, 22).
An increased risk of death from ischemic heart diseases
among middle-aged men was particularly noteworthy.
Am J Epidemiol Vol. 149, No. 12, 1999
Evidence suggests that the associations between
higher mortality and extreme weather conditions are
rooted in physiology. Keatinge et al. (13) describe
platelet-count and blood-viscosity increases that may
occur when populations are exposed to cold temperatures. That platelets are "concerned in coagulation of
the blood and in contraction of the clot, and hence in
hemostasis and thrombosis" (23, p. 1304) makes
increases in mortality more comprehensible, and perhaps most importantly, more preventable. When people engage in typical winter activities such as snow
shoveling that account for both cold exposure and
increased physical stress, these biologic mechanisms
likely increase the risk of death due to ischemic heart
diseases and cerebrovascular diseases (14, 15).
Deaths due to ischemic heart diseases were found to
have strong significant associations with temperatures.
This was especially true for men, who experienced the
highest risk of death in the youngest age group. In a
short-term study, a significant 16 percent rise in mor-
1158
Gorjanc et al.
TABLE 5. Rate ratios (RRs) and 95% confidence intervals (CIs) for mortality among people aged 50-64 years derived from
general estimating equations analysis for selected categories of temperature and snowfall, Pennsylvania, 1991-1996
Snow
(cm)
Temperature
(°C)
Both sexes
RR
Males
95% Cl
RR
Females
95% Cl
RR
95% Cl
Reference
0.94-1.08
0.92-1.19
1.04-1.29
1.04-1.32
1.03-1.44
1.00
0.98
0.89
1.01
1.00
0.90
Reference
0.90-1.07
0.68-1.16
0.90-1.15
0.88-1.13
0.67-1.21
Reference
0.85-1.14
0.97-1.50
1.20-1.80
1.15-1.82
1.32-2.38
1.00
1.02
0.59
0.84
0.85
0.49
Reference
0.81-1.29
0.28-1.24
0.48-1.47
0.47-1.53
0.19-1.25
1.00
1.23
1.19
1.39
1.71
1.66
Reference
0.86-1.75
0.62-2.30
0.88-2.19
0.99-2.95
0.74-3.69
1.00
0.83
0.67
0.98
0.82
0.66
Reference
0.65-1.06
0.36-1.25
0.62-1.54
0.50-1.33
0.30-1.41
Total deaths
Low (<3)
High (>3)
High (>0)
Mid (>-7 to 0)
Low (<-7)
High (>0)
Mid (>-7 to 0)
Low (<-7)
1.00*
0.99*
0.97*
0.99*
1.15*,*
1.36*,*
High (>0)
Mid (>-7 to 0)
Low (<-7)
High (>0)
Mid (>-7 to 0)
Low (<-7)
1.00
0.99
1.04
1.29*
1.27*
1.33*
High (>0)
Mid (>-7 to 0)
Low (<-7)
High (>0)
Mid (>-7 to 0)
Low (<-7)
1.00
1.37*
1.30
0.98
1.35
1.28
Reference
1.13-1.67
0.77-2.20
0.66-1.46
0.88-2.05
0.66-2.47
High (>0)
Mid (>-7 to 0)
Low (<-7)
High (>0)
Mid (>-7 to 0)
Low (<-7)
1.00
0.87
0.88
1.25
1.09
1.09
Reference
0.71-1.07
0.60-1.28
0.91-1.71
0.77-1.55
0.67-1.79
Reference*
0.94-1.04
0.87-1.07
0.86-1.15
1.03-1.29
1.07-1.72
1.00
1.01
1.05
1.16*
1.17*
1.22*
Deaths from ischemic heart diseases
Low (<3)
High (>3)
Reference
0.88-1.11
0.85-1.26
1.07-1.55
1.05-1.55
1.03-1.73
1.00
0.98
1.20
1.47*
1.44*
1.77*
Deaths from cerebrovascular diseases
Low (<3)
High (>3)
1.00
1.49*
1.39
0.67
1.00
0.94
Reference
1.14-1.95
0.70-2.79
0.35-1.31
0.50-2.02
0.36-2.45
Deaths from respiratory diseases
Low (<3)
High (>3)
§
§
§
§
§
§
* Interaction of snowfall and temperature was significant.
t Reference group.
*p<0.05.
§ Not included because too few deaths in these categories.
tality from ischemic heart diseases was observed after
"cold snaps" (22). In a longer-term study in Sweden, a
similar trend showed that the strongest association
between coronary mortality and temperature was
among men aged 40-64 years (8). A follow-up study
found long-term associations between climate and
coronary mortality that were not explained by most
socioeconomic factors nor by accounting for established risk factors, such as serum cholesterol levels or
use of antihypertensive drugs (8-10). In our study,
increased death due to ischemic heart diseases was
also strongly associated with snowfall, with highly significant increases in mortality rates from ischemic
heart diseases associated with snowfall greater than 3
cm. Mortality from ischemic heart diseases among
men was higher than among women, with significant
findings for men in all age groups, as well as the oldest women. Other studies (3, 22) have noted that
deaths from ischemic heart diseases rose by 22 percent
in a "blizzard week" and by as much as 88 percent
after heavy snowfalls among men aged 65 years or
less. Physiologic studies of cold exposure and the
physical stresses of outdoor activities suggest that traditional male activities such as snow shoveling may
account for many of these deaths (13-15).
In this study, deaths due to cerebrovascular diseases
tended to be higher during severe weather among those
aged 50 years and older. For men aged 50-64 years, the
pattern was not clear-cut, while for women in this age
group and for those of both sexes aged 65 years and
older, a pattern of higher mortality during periods of
higher snowfall and lower temperatures was observed.
Similarly, other studies have shown inverse and approximately linear associations between stroke mortality and
temperature over the greater part of the temperature
range (12), findings which are consistent with our own.
Am J Epidemiol
Vol. 149, No. 12, 1999
Effects of Temperatures and Snowfall on Mortality
1159
TABLE 6. Rate ratios (RRs) and 95% confidence intervals (CIs) for mortality among people aged 35-49 years derived from
general estimating equations analysis for selected categories of temperature and snowfall, Pennsylvania, 1991-1996
Snow
(cm)
Temperature
(°C)
Both sexes
RR
Males
RR
95% Cl
Females
95% Cl
RR
95% Cl
Reference
0.86-1.06
0.98-1.41
0.79-1.11
0.75-1.07
0.86-1.41
1.00
0.89
0.89
0.99
0.88
0.88
Reference
0.79-1.00
0.67-1.17
0.81-1.21
0.71-1.08
0.62-1.23
Reference
0.83-1.36
1.50-2.89
1.30-2.22
1.30-2.50
2.35-5.35
t
t
t
t
t
t
Total deaths
Low (<3)
High (>3)
Reference*
0.87-1.00
0.89-1.30
0.85-1.08
0.79-1.00
0.82-1.28
High (>0)
Mid (>-7 to 0)
Low (<-7)
High (>0)
Mid (>-7 to 0)
Low (<-7)
1.00
0.93
1.07
0.96
0.89
1.03
High (>0)
Mid (>-7 to 0)
Low (<-7)
High (>0)
Mid (>-7 to 0)
Low (<-7)
1.00
1.01
1.804:
1.594:
1.604:
2.864:
High (>0)
Mid (>-7 to 0)
Low (<-7)
High (>0)
Mid (>-7 to 0)
Low (<-7)
1.00
0.85
0.83
0.58
0.49
0.48
Reference
0.56-1.29
0.29-2.35
0.25-1.36
0.20-1.24
0.13-1.85
High (>0)
Mid (>-7 to 0)
Low (<-7)
High (>0)
Mid (>-7 to 0)
Low (<-7)
1.00
1.06
0.23
1.52
1.61
0.35
Reference
0.71-1.58
0.04-1.42
0.92-2.51
0.87-2.97
0.05-2.32
1.00
0.95
1.17
0.94
0.89
1.10
Deaths from ischemic heart diseases
Low (<3)
High (>3)
Reference
0.81-1.26
1.31-2.47
1.17-2.14
1.14-2.25
1.86-^.38
1.00
1.06
2.084:
1.704:
1.804:
3.544:
Deaths from cerebrovascular diseases
Low (<3)
High (>3)
1.00
0.81
1.33
0.32
0.26
0.42
Reference
0.46-1.44
0.48-3.69
0.07-1.53
0.05-1.35
0.06-2.75
t
t
t
t
t
t
Deaths from respiratory diseases
Low (<3)
High (>3)
t
t
t
t
t
t
t
t
t
t
t
t
* Reference group.
t Not included because too few deaths in these categories.
4: p S 0.05.
We found both temperature and snowfall to be significantly associated with deaths from respiratory diseases for women aged 65 years and older. Because the
effects of weather on deaths from respiratory diseases
were lagged 3 days, it is possible that those who died
from respiratory diseases may have died sometime
before or after the third day. If this did indeed occur,
the 3-day lag period that we used may have underestimated the effects of weather on deaths from respiratory diseases. Other studies have shown that deaths
due to respiratory diseases were associated with temperatures recorded 5 or more days before the day of
death (11, 24). For example, Bull and Morton (24)
found deaths due to pneumonia to be associated with
temperatures 7 or more days before.
In this study, we addressed day-to-day changes in
mortality as they were affected by day-to-day changes
in temperature and snowfall. We did not address the
time and place of death nor individual factors, such as
Am J Epidemiol
Vol. 149, No. 12, 1999
the use of medications, preexisting medical conditions,
and lifestyles, including socioeconomic status and
alcohol consumption. Although many of these issues
have long been considered risk factors for death by the
medical and public health communities, some research
has shown that exposure to cold temperatures is actually a much stronger risk factor (10). Further, we noted
that wind speed, humidity, and radiant heat energy are
important in assessing stress from cold exposure.
These data, however, are not easily available in the
same comprehensive statewide coverage as are temperature and snowfall data (25).
A limitation inherent in our study was the high
degree of autocorrelation in temperature data. Because
of this pattern in the temperature data, it was difficult
to determine whether the models were detecting an
association of temperature with mortality for the specified period or whether they were actually detecting an
association with temperatures the day before or after
1160
Gorjancetal.
the lag period. From a public health standpoint, however, precautionary health advisories based on forecasts for severe winter-weather events would still be
possible. Conversely, slight changes in periods correlating snowfall with mortality could produce widely
varying results. This is due to the more random nature
of snowfall occurrence. In this study, nearly 50 percent
of all division-days recorded no snowfall at all.
To conclude, cold temperatures and significant
snowfall appear to contribute to significantly higher
rates for deaths from ischemic heart diseases among
men. Given what we know about the body's normal
thermoregulatory responses to cold temperatures and
snowfall, including increased platelet counts and vasoconstriction of the peripheral venous system, and
about the effects of the physical stress of outdoor winter activities such as snow shoveling, it appears biologically plausible that a series of events triggered by
adverse weather may lead to death for those at greatest
risk. Although cold-weather conditions are especially
dangerous for those aged 65 years and older, this study
suggests that men as young as 35 years, who traditionally have not been cautioned in winter-weather health
advisories, may also be at risk of death.
We suspect that many middle-aged men are unaware
of the risk posed to them by severe winter weather.
Results of this study, however, suggest that public
health advisories concerning the hazards of cold and
snow exposure may need to be extended to include
middle-aged men as young as 35 years old. We also
recommend further study of the effects of coldweather conditions on mortality among men aged
35-49 years, including the time and place of death, as
well as the individual risk factors for death among
those in this age group.
ACKNOWLEDGMENTS
The authors acknowledge the technical assistance of Dr.
Pamela Anderson, National Institute of Occupational Safety
and Health, and Dr. Carol Crawford, Dr. David Olson, Larry
Killen, Dr. Michael McGeehin, Dr. Rossanne Philen, and
Larry Posey, National Center for Environmental Health,
Centers for Disease Control and Prevention. They also thank
Dr. Tom Ross from the National Climatic Data Center for
providing the climatic data used for this study.
REFERENCES
1. Lott N, Ross T, Sittel M. The winter of '95-'96...a season of
extremes. Asheville, NC: National Climatic Data Center, 1996.
(National Climatic Data Center technical report 96-02).
2. National Climatic Data Center cooperative data. (CD-ROM).
Asheville, NC: National Climatic Data Center, National
Environmental Satellite, Data and Information Service,
National Oceanic and Atmospheric Administration, 1997.
3. Glass R, Zack M. Increase in deaths from ischemic heart disease after blizzards. Lancet 1979;3:485-7.
4. Faich G, Rose R. Blizzard morbidity and mortality: Rhode
Island, 1978. Am J Public Health 1979;69:1050-2.
5. Larsen U. The effects of monthly temperature fluctuations on
mortality in the United States from 1921 to 1985. Int J
Biometeorol 1990;34:136-45.
6. Larsen U. Short-term fluctuations in death by cause, temperature, and income in the United States, 1930-1985. Soc Biol
1990;37:172-87.
7. Langford IH, Bentham G. The potential effects of climate
change on winter mortality in England and Wales. Int J
Biometeorol 1995;38:141-7.
8. Gyllerup S, Lanke J, Lindholm LH, et al. High coronary mortality in cold regions of Sweden. J Intern Med 1991;230:
479-85.
9. Gyllerup S, Lanke J, Lindholm LH, et al. Socioeconomic factors in the community fail to explain the high coronary mortality in cold parts of Sweden. Eur Heart J 1992;13:878-81.
10. Gyllerup S, Lanke J, Lindholm LH, et al. Cold climate is an
important factor in explaining regional differences in coronary
mortality even if serum cholesterol and other established risk
factors are taken into account. Scott Med J 1993;38:169-72.
11. Bull GM, Morton J. Environment, temperature, and death
rates. Age Ageing 1978;7:210-24.
12. Rogot E, Padgett SJ. Associations of coronary and stroke mortality with temperature and snowfall in selected areas of the
United States, 1962-1966. Am J Epidemiol 1976; 103:565-75.
13. Keatinge WR, Coleshaw SR, Cotter F, et al. Increases in
platelet and red cell counts, blood viscosity, and arterial pressure during mild surface cooling: factors in mortality from
coronary and cerebral thrombosis in winter. Br Med J (Clin
Res Ed) 1984;289:1405-8.
14. Franklin BA, Hogan P, Bonzheim K, et al. Cardiac demands of
heavy snow shoveling. JAMA 1995;273:880-2.
15. Franklin BA, Bonzheim K, Gordon S, et al. Snow shoveling: a
trigger for acute myocardial infarction and sudden death. Am J
Cardiol 1996;77:855-8.
16. Mortality, Pennsylvania, 1991-1996. (Computerized data
files). Harrisburg, PA: Pennsylvania Department of Health,
1996.
17. International classification of diseases. Ninth Revision, clinical modification. Vol 1. 4th ed. Washington, DC: Department
of Health and Human Services, 1991. (DHHS publication no.
(PHS) 91-1260).
18. US Bureau of the Census. Statistical abstract of the United
States: 1995. 115th ed. Washington, DC: US Department of
Commerce, 1995.
19. Liang KY, Zeger SL. Data analysis using generalized linear
models. Biometrika 1986;73:13-22.
20. SAS Institute, Inc. SAS/STAT user's guide. Cary, NC: SAS
Institute, Inc, 1997.
21. Devine O, Parrish RG. Monitoring the health of populations.
In: Stroup DF, Teutsch SM, eds. Statistics in public health:
quantitative approaches to public health problems. New York,
NY: Oxford University Press, 1997.
22. Anderson TW, Rochard C. Cold snaps, snowfall and sudden
death from ischemic heart disease. Can Med Assoc J
1979;121:1580-3.
23. Platelet. Dorland's illustrated medical dictionary. 28th ed.
Philadelphia: WB Saunders Company, 1994:1304.
24. Bull GM, Morton J. Relationships of temperature with death
rates from all causes and from certain respiratory and arteriosclerotic diseases in different age groups. Age Ageing
1975;4:232^6.
25. Kilbourne EM. Cold environments. In: Noji EK, ed. The public health consequences of disasters. New York, NY: Oxford
University Press, 1997.
Am J Epidemiol Vol. 149, No. 12, 1999