Does Temperature Modify the Association between Air Pollution and Mortality? A
Multicity Case-Crossover Analysis in Italy
M. Stafoggia1,2, J. Schwartz3, F. Forastiere1, C. A. Perucci1, and the SISTI Group
1Department of Epidemiology, Rome E Health Authority, Rome, Italy.
2Department of Biostatistics, Harvard School of Public Health, Boston, MA.
3Department of Environmental Health, Harvard School of Public Health, Boston, MA.
Received for publication November 21, 2007; accepted for publication March 6, 2008.
Adverse health effects of particulate matter <10 lm in aerodynamic diameter (PM10) and high temperatures are
well known, but the extent of their interaction on mortality is less clear. This paper describes effect modification of
temperature in the PM10–mortality association and tests the hypothesis that higher PM10effects in summer are due
to enhanced exposure to particles. All deaths of residents of nine Italian cities between 1997 and 2004 were
selected. The case-crossover approach was adopted to estimate the effect of PM10on mortality by season and
temperature level. Three strata of temperature corresponding to low, medium, and high ‘‘ventilation’’ were identi-
fied, and the interaction between PM10and temperature within each stratum was examined. Season and temper-
ature levels strongly modified the PM10–mortality association: for a 10-lg/m3variation in PM10, a 2.54% increase in
risk of death in summer (95% confidence interval: 1.31, 3.78) compared with 0.20% (95% confidence interval:
?0.08, 0.49) in winter. Analysis of the interaction between PM10and temperature within temperature strata resulted
in positive but, in most cases, nonstatistically significant coefficients. The authors found much higher PM10effects
on mortality during warmer days. The hypothesis that such an effect is attributable to enhanced exposure to
particles in summer could not be rejected.
association; climate; effect modifiers (epidemiology); environmental exposure; mortality; particulate matter;
Abbreviation: PM10, particulate matter <10 lm in aerodynamic diameter.
Numerous epidemiologic studies have reported associa-
tions between exposure to outdoor particulate matter and
daily mortality (1–4). The association is consistent in many
countries and concerns the overall population, but effects are
higher in elderly people (5), less-affluent persons (6, 7), and
those who are more vulnerable because of chronic conditions
such as diabetes (5, 7, 8), heart failure (9), myocardial in-
farction (8), and chronic respiratory conditions (10).
A strong association between high summer temperatures
and mortality also has been detected, with a generally
J-shaped exposure response, immediate lags, and similar
patterns in different countries (11–13), although the overall
effect strongly depends on local characteristics (14), cli-
matic conditions, and the availability of air-conditioning
systems (15). The effect is more pronounced in the elderly
(15), women (16), and people with chronic conditions such
as psychoses (11), depression (11), cerebrovascular diseases
(11), diabetes (15, 17), and chronic obstructive pulmonary
It has been common practice in epidemiologic studies to
adjust for temperature in the analyses of health effects of
particulate matter and, to a lesser extent, to adjust for air
pollutants when studying the effects of temperature on
mortality. Surprisingly, the issue of effect modification that
Correspondence to Massimo Stafoggia, Department of Epidemiology, Rome E Health Authority, Via Santa Costanza 53, 00198 Rome, Italy
American Journal of Epidemiology
ª The Author 2008. Published by the Johns Hopkins Bloomberg School of Public Health.
All rights reserved. For permissions, please e-mail: email@example.com.
American Journal of Epidemiology Advance Access published April 11, 2008
by guest on June 13, 2013
temperature may exert on the air pollution–mortality rela-
tion has been largely neglected, so that the extent of the
interaction between the environmental exposures is less
clear. Some papers have been published in the last few years
on season-specific analyses of the association between par-
ticulate matter and mortality (7, 18–20), with some evidence
that the higher effects are found in the warmer months. A
few studies tried to further explore the interaction between
particulate matter and temperature (21–23), identifying
a significant enhanced effect for increasing values of both
particulate matter and temperature. However, the form and
the possible mechanisms ofinteraction are largelyunknown.
A recent scientificdebate (24) indicates an increased levelof
exposure during the warm period due to open windows and
outdoor activities or seasonal changes in the chemical com-
position of particulate matter as two alternative hypotheses
that should be better evaluated.
The objective of this study was to provide further insight
into the subject, using data from nine Italian cities. We used
a staged approach of 1) examining effects by season, 2)
examining effects by strata defined by temperature, and 3)
including interactions with temperature within those strata.
Doing so addresses the hypothesis that the effect modifica-
tion by temperature level on the relation between particulate
matter <10 lm in aerodynamic diameter (PM10) and mor-
tality is predominantly due to a higher exposure to air pol-
lution during warmer months associated with open windows
and more outdoor activities (the ‘‘ventilation’’ hypothesis).
In particular, we postulate that the effect of PM10on mor-
tality would be higher in the high-ventilation days but that
the interaction between PM10and apparent temperature on
those days would not be significant.
MATERIALS AND METHODS
The study population consisted of inhabitants aged 35
years or older in nine Italian cities between 1997 and
2004. For each city, we selected all residents who died of
natural causes (International Classification of Diseases,
Ninth Revision, codes 1–799) in the city. Information on
the underlying cause of death was available (for all but
one city), and we divided them into three groups: cardiovas-
cular diseases (codes 390–459), respiratory diseases (codes
460–519), and other natural diseases.
Information on daily environmental variables was ob-
tained from the Italian Air Force Meteorological Service,
which provided data on air temperature (degrees Centi-
grade), dew point temperature (degrees Centigrade), and
barometric pressure (hectopascals). Apparent temperature
is a climatological index that takes into account the physical
stress during warm days due to the combined effect of air
temperature and humidity (25, 26).
Hourly PM10data from city monitors located in residen-
tial areas were provided by the regional environmental
protection agencies. Data were collected for each city ac-
value was calculated for each monitor by averaging hourly
values, and a daily average for each city was estimated from
the monitor-specific daily means. Missing data on the aggre-
Air Pollution and Health: A European Approach (APHEA)
The current- and the preceding-day PM10means (lag 0–1)
were averaged as the exposure variable, based on previous
We collected information on the following variables as
potential confounders of the association between mortality
and PM10: population decrease during the summer period,
day of the week, holidays, barometric pressure, and influ-
enza epidemics (defined as the annual 3-week period of
maximum incidence of flulike illness based on estimates
of weekly influenza incidence, as reported by the Italian
National Health Service).
The case-crossover design was used to study the associ-
ation between PM10/apparent temperature and mortality
(28). It is a variant of the case-control design, in which each
subject is matched to himself or herself; controls are chosen
as times (i.e., days) in which the event did not occur. It
follows that all time-invariant individual characteristics
are adjusted by design, while other time-dependent covari-
ates can be controlled for by modeling.
We used the time-stratified approach (29) to select control
days, with controls chosen every 3 days in the same month
and year as the event day. Hence, for example, if one subject
died on November, 20, 2000, his or her control days would
have been the 2nd, 5th, 8th, 11th, 14th, 17th, 23rd, 26th, and
29th of November 2000. This approach controls for season
by matching month and year. It also partly controls for other
variables such as weather, since all comparisons are made
within the same month.
We fit conditional logistic regression for each city, where
the outcome variable was the indicator of case/control day
and the exposure variable was PM10(lag 0–1) (alone and in
combination with apparent temperature (lag 0–1); refer to
the information below), and we further controlled for baro-
metric pressure (one linear term), cold apparent temperature
at lag 1–6 (one linear term for values below 9?C), day of the
week, decrease in population during summer, holidays, and
influenza epidemics. All city-specific analyses were per-
formed by using the function coxph( ) in the R statistical
software package (30, 31).
effects meta-analysis, using maximum likelihood as the
estimation method (32).Reportedwere pvaluesofheteroge-
neity between cities. Refer to the online material for further
details (this information is posted on the Journal’s website
We performed the analysis in four stages. First, we evalu-
ated the association of PM10(lag 0–1) with all natural-cause
2 Stafoggia et al.
by guest on June 13, 2013
mortality by season while adjusting for apparent temperature
within each season stratum. In both the full-year and the
season-specific models, apparent temperature (lag 0–1) was
controlled for by using a penalized spline (33) with an effec-
tive number of degrees of freedom chosen by minimizing
Akaike’s Information Criterion (34). In addition, apparent
temperature (lag 1–6) below 9?C was adjusted for with a lin-
ear term to control for potential confounding of low temper-
atures. Refer to the online supplement for further details.
Second, we defined three strata of daily apparent temper-
ature (lag 0–1) according to city-specific percentiles of the
apparent temperature distributions: below the 50th percen-
tile (relatively low temperature, heat on, windows closed,
resulting in low ventilation), between the 50th and 75th
percentiles (transient period with intermediate temperature
and mid-ventilation levels), and above the 75th percentile
(relatively high temperature, windows open, high ventila-
tion). Air-conditioning prevalence was very low in Italy
during the study period, which enabled us to classify days
by temperature instead of season, thus capturing warm days
in the spring and fall into our high-temperature (and hence
high-ventilation) category. We thus analyzed the PM10–
mortality association stratifying by the new categories while
adjusting for apparent temperature within each stratum to
take into account potential residual confounding.
Third, we explored potential nonlinearities in the joint
association of PM10/apparent temperature with natural mor-
tality for the two largest cities in the study: Milan and Rome.
They were chosen because they comprise 65 percent of the
studied population and are representative of the different
climate conditions in Italy, namely, continental in northern
Italy versus Mediterranean in central-southern Italy. Three-
dimensional surfaces were obtained by using thin-plate
splines (35). This step was performed by fitting a Poisson
generalized additive model as an alternative to conditional
logistic regression because previous investigations have
shown that the two methods are equivalent as long as time
trend is suitably controlled for (36). Again, refer to the on-
line supplement for further details.
Finally, we added an interaction term between PM10and
apparent temperature within each stratum from the previous
temperature-stratified analysis to check whether the greater
effect of particulate matter on warmer days was due to higher
exposure to PM10(absence of linear interaction on a log
scale) or the interactive effect of the two exposures was still
present even after having stratified by ‘‘ventilation.’’ We an-
alyzed the ‘‘interaction term’’ again by shifting the second
cutoff point of apparent temperature to higher values, from
the 75th percentile to the 90th, to determine whether a poten-
tial interaction between PM10and apparent temperature on
natural mortality became evident on the warmest days.
All previous analyses were repeated by considering
cause-specific mortality as alternative outcomes.
Several sensitivity analyses were performed to check the
robustness of results to model specifications, dose-response
shapes of confounders, and study periods. Specifically, an
alternative strategy for selecting the control days within the
time-stratified approach was applied, choosing as control
days the same days of the week within the same month
and year as the event day. Second, the confounding effect
of cold temperatures was adjusted for by using penalized
splines of apparent temperature (lag 1–6) or linear splines
with one inner knot chosen differently for each city. Third,
the analyses were rerun excluding 2003–2004, a period
characterized by extremely high temperatures and increased
use of air-conditioning. More details are reported in the
Table 1 displays the study population by city and cause of
death. A total of 321,024 residents aged 35 years or older
died from natural causes: 41.2 percent of the deaths were
caused by cardiovascular diseases and 6.9 percent by respi-
The distribution of death events by season and apparent
temperature is displayed, for each city, in table 2. There
were no meaningful differences across cities.
The main characteristics of the two environmental expo-
sures, PM10 and apparent temperature, are displayed in
table 3, as well as their Pearson correlation coefficients
(and corresponding p values). Daily mean PM10ranged
from 35.1 lg/m3in Pisa to 71.5 lg/m3in Turin. There
was a clear north-south gradient in terms of daily temper-
atures, with the coldest cities located in northern Italy (Turin,
Milan, Mestre, Bologna) and the warmest in central (Pisa,
Florence, Rome) and southern (Taranto and Palermo) Italy.
The cutoffpointschosento defineapparenttemperaturestrata
reflect this trend. The correlation between PM10and apparent
temperature was somewhat heterogeneous across the cities
and was always negative, with the exception of Palermo.
The pooled association between PM10and mortality, by
cause of death, season, and apparent temperature strata, is
reported in table 4, where results are expressed as percent
increases in risk, and 95 percent confidence intervals, cor-
responding to a 10-lg/m3variation in PM10. Reported are
p values for heterogeneity among city-specific estimates.
There was strong effect modification by season, with a gen-
eral pattern of a stronger effect in summer (2.54 percent, 95
percent confidence interval: 1.31, 3.78 for all natural mor-
tality) and a lower effect in winter (0.20 percent, 95 percent
confidence interval: ?0.08, 0.49). This pattern was similar
for all, cardiovascular, and respiratory mortality. Table 4
also reports results stratified by apparent temperature strata.
Again, the effect of PM10was considerably higher as tem-
perature increased. Results were homogeneous between
centers in the low-temperature and winter/spring strata,
but some heterogeneity was present for higher temperatures,
although with different patterns according to cause of death:
greater differences between cities in the mid-temperature
and fall season strata for natural and cardiovascular mortal-
ity, greater heterogeneity in the high-temperature and sum-
mer season strata for respiratory mortality.
Table 5 shows the regression coefficients, standard errors,
and p values for the interaction terms PM103 apparent
temperature, by apparent temperature strata and causes of
Temperature, Air Pollution, and Mortality in Italy3
by guest on June 13, 2013
death. Also reported are p values for heterogeneity among
city-specific estimates. There was a suggestion of a positive
linear interaction (on a log scale) within the highest stratum
of apparent temperature for all groups of causes of death (p
of about 0.15) and in the mid-stratum of apparent tempera-
ture for cardiovascular mortality (p ¼ 0.014). Results were
homogeneous among centers (p for heterogeneity was al-
The above model assumes a linear interaction between
the two exposures. To determine whether we were missing
a more complex interaction between PM10and apparent
temperature, we fit a nonlinear interaction by using thin
plate splines, and the results are shown in figure 1 as
three-dimensional exposure-response surfaces and contour
plots for Milan and Rome. Although the patterns were dif-
ferent in the two cities, the estimated numbers of deaths
were lowest for low values of PM10and apparent tempera-
ture and increased exponentially for increasing temperatures
and PM10, without any sign of a threshold.
Figure 2 reports the coefficients of the interaction terms
between PM10and apparent temperature on all natural mor-
tality in the third stratum of apparent temperature defined
on the basis of a city-specific cutoff point ranging from the
75th to the 90th percentile of apparent temperature. The
nine Italian cities, by season, level of apparent temperature, and city, 1997–2004
Study population: distribution of death events (numbers and percentages) among people aged ?35 years and residing in
Season Apparent temperature*
No.% No.% No.% No.% No.% No.% No.%
Bologna4,58527.64,130 24.93,942 23.73,95523.8 8,49451.13,88723.4 3,99324.0 16,612
Florence4,03228.1 3,56324.8 3,41923.8 3,342 23.37,47952.1 2,781 19.43,50324.4 14,356
Mestre1,47629.2 1,22524.2 1,13722.51,21524.12,757 54.61,182 23.4 1,075 21.3 5,053
Milan 17,95328.715,65725.014,00222.4 14,94723.9 33,67553.8 15,40224.6 13,24921.1 62,559
Palermo4,66328.4 4,200 25.73,91524.0 3,58321.9 7,646 46.8 3,85423.6 3,33820.4 16,331
Pisa 97126.1 93125.193225.187923.71,88650.896325.9861 23.2 3,713
Rome40,987 28.136,63725.134,25223.5 33,96923.379,389 54.4 31,68221.733,48123.0 145,845
Taranto1,20327.91,13626.41,086 25.2 88420.52,277 52.8 96422.31,063 24.74,309
Turin 15,07728.9 12,80624.512,12023.212,24323.426,934 51.6 11,88022.7 11,59222.252,246
Total90,917 28.3 80,28525.074,805 23.375,017 23.4170,53753.172,59522.672,155 22.5 321,024
* The totals of death events over apparent temperature strata do not equal the overall totals (last column) because of days for which data on
apparent temperature were missing.
yFor all cities, percentage ¼ 100.
years and residing in nine Italian cities, by cause of death, 1997–2004
Study population: numbers and percentages of deaths of people aged ?35
Bologna2000–20036,69040.3 1,334 8.08,58851.716,612
Mestre1999–2001 2,08841.32745.4 2,691 53.3 5,053
Milan1999–200424,380 39.05,236 8.432,94352.6 62,559
Pisa 2000–20031,687 45.4273 7.41,75347.23,713
Rome1998–200461,452 42.2 8,361 5.7 76,03252.1145,845
* For all cities, percentage ¼ 100.
yPercentage of the total number of deaths from natural causes in the eight cities (all but
Palermo) with available information on cause of death (N ¼ 304,693).
4Stafoggia et al.
by guest on June 13, 2013
interaction term never reached statistical significance (with
the exception of the 79th percentile), and it decreased to
zero for the warmest cutoff points, indicating a lack of (lin-
ear) interaction between the two environmental exposures
on the warmest days.
Results from the sensitivity analyses (table E1 in the on-
line Appendix) did not alter the main conclusions.
The present study was designed to investigate the inter-
active effect of particulate matter and apparent temperature
on mortality in nine Italian cities. The increasing effect of
PM10by temperature level is suggestive of a synergism be-
tween the two exposures. The three-dimensional exposure-
response surfaces for Milan and Rome (figure 1) showed
that the interactive patterns are quite complex; the expected
numbers of deaths tended to increase steeply in Milan as
temperature and PM10increased, while the joint relation
was more irregular in Rome. We evaluated the hypothesis
that the effect modification was predominantly due to
greater exposure to air pollution during warm months by
examining effects within strata of daily temperature and
by adding linear interaction terms between PM10and appar-
ent temperaturewithin strata of temperature itself. We found
limited evidence to reject the ‘‘ventilation’’ hypothesis.
Numerous studies have investigated the role of season as
an effect modifier in the air pollution–mortality association,
finding consistent results in both North America and
Europe. Katsouyanni et al. (37) identified higher effects of
PM10, black smoke, and sulfur dioxide on total mortality
and 75th percentile of apparent temperature; and Pearson correlation coefficient between
PM10and apparent temperature for nine Italian cities, 1997–2004
Environmental variables: mean (SD*) of PM10*; mean (SD), 50th percentile,
Mean (SD) of
Apparent temperature (?C) Pearson correlation
Bologna2000–2003 50.4 (31.7) 13.7 (10.0) 1322?0.45
Florence 2000–200337.5 (16.6) 15.2 (9.2) 15 22 0.002
Mestre 1999–2001 48.1 (26.8) 14.7 (10.4)15 24 0.000
Milan 1999–200457.9 (38.0) 13.5 (9.3)13 22 0.000
Palermo 2002–2004 36.2 (21.7) 19.0 (7.9)1826 0.000
Pisa2000–2003 35.1 (14.9) 14.8 (8.8)14 22?0.30
Rome1998–2004 47.3 (19.9)15.7 (8.6) 16230.346
Taranto 2001–2003 59.8 (18.9)16.6 (8.2)16 23 0.027
Turin 1997–200371.5 (38.1) 11.9 (9.4)11 200.000
* SD, standard deviation; PM10, particulate matter <10 lm in aerodynamic diameter.
Italian cities, by cause of death, season, and apparent temperature level on the day of death, including percent increases in risk (%)
and 95% confidence intervals corresponding to a 10-mg/m3variation in the pollutant as well as pH*,y 1997–2004
Pooled results: association between PM10* (lag 0–1) and mortality among people aged ?35 years and residing in nine
Other causes of death All natural causes
% 95% CI*
% 95% CIpH
% 95% CIpH
% 95% CIpH
Full year0.630.31, 1.380.224 0.980.27, 1.70 0.5980.370.09, 0.660.681 0.530.25, 0.800.161
Winter 0.15?0.29, 0.59
0.668 0.41?0.67, 1.51
Spring 0.720.4972.99 0.932 0.290.569 0.62 0.645
Summer2.900.0353.89 0.19, 7.730.088 2.150.5362.54 1.31, 3.78 0.011
Fall1.370.43, 2.32 0.0080.45?1.11, 2.03 0.8220.70?0.41, 1.83 0.002 1.210.37, 2.060.000
<50th percentile0.31?0.06, 0.67
0.750 0.21?0.06, 0.47
50th–75th percentile 2.05 0.0003.150.3561.080.1631.60 0.003
>75th percentile 2.681.20, 4.170.3604.120.44, 7.93 0.1032.300.411 2.551.58, 3.520.027
* PM10, particulate matter <10 lm in aerodynamic diameter; pH, p value for heterogeneity of city-specific results; CI, confidence interval.
y The null hypothesis corresponds to perfect homogeneity.
Temperature, Air Pollution, and Mortality in Italy5
by guest on June 13, 2013
in warmer months, studying 12 European cities within the
Air Pollution and Health: A European Approach project;
Michelozzi et al. (18) found a significant effect of nitrogen
dioxide and PM10on mortality in Rome in the warm season
only; more recently, Peng et al. (19), using data from 100 US
cities within the National Morbidity, Mortality, and Air Pol-
lution Study, found a significant association between PM10
and mortality in summer (lag 0) and spring (lag 1) only;
a study conducted in Belgium (20) identified a strong in-
teraction between PM10and season for both overall and
cause-specific mortality. On the other hand, studies con-
ducted in Asia gave contrasting results (38, 39).
The interaction between air pollution and temperature has
been explored to a lesser extent in the epidemiologic liter-
ature. One of the first papers that shed light on the subject
(21) identified a strong interaction between sulfur dioxide
and temperature (<30?C vs. ?30?C) in Athens, Greece,
while the main effect of sulfur dioxide was not significant.
Samet et al. (40) studied the effect of total suspended par-
ticles and sulfur dioxide on mortality in Philadelphia, Penn-
sylvania, while adjusting for weather in four different
models and testing for potential effect modification; they
found little evidence of an interaction, regardless of the
approach used to model weather. In 2001, Katsouyanni
et al. (2) addressed the issue of effect modification within
the Air Pollution and Health: A European Approach
project by performing a meta-analysis of city-specific re-
sults regarding the association between PM10/black smoke
and mortality in 14 European centers and adding mean daily
temperature in the meta-regression stage: temperature con-
tributed to explain residual variation in city-specific results
in a highly significant way, and the estimated increases in
number of deaths for a 10-lg/m3variation in PM10/black
smoke were much higher in the 75th percentile of temper-
ature than in the 25th percentile.
In recent years, a few studies (21–23, 41) have explored
the interactive patterns of PM10and temperature on mortal-
ity in more detail by using more flexible models. In all cases,
a consistent increased risk of death for increasing values of
both environmental variables has been confirmed.
Several explanations have been proposed for the effect
modification of temperature in the air pollution–human
health association. Gordon (42) observed that a synergism
between the two exposures is biologically plausible because
the thermoregulatory system responds to heat stress by
activating three key mechanisms to dissipate excess heat:
cardiovascular, respiratory, and sudomotor (sweating). Ac-
tivation of these thermoeffector systems can have direct or
indirect effects on the entry of toxicants into the body, thus
augmenting total intake of airborne pollutants. On the other
hand, it is possible that the higher PM10effects during sum-
mer are due merely to greater exposure in that period or
simply to a better exposure measurement, since people are
more likely to keep the windows open or spend time out-
doors, and outdoor monitoring-based exposure better re-
flects actual individual exposure. Personal exposure studies
have shown substantially higher personal/outdoor particu-
late matter slopes during periods when windows are open
compared with periods when windows are closed (43). Fi-
nally, particulate matter composition may be different by
Pooled results: linear interaction between PM10* (lag 0–1, 10 mg/m3) and apparent temperature (lag 0–1, 1?C) against mortality among people aged ?35 years and
residing in nine Italian cities, by cause of death and apparent temperature level on the day of death, including regression coefficients (beta), p values, and pH*,y
Other causes of death
All natural causes
0.600 0.000246 (0.000269)
0.525 ?0.001120 (0.003480)
0.411 ?0.001526 (0.001207)
0.567 0.000584 (0.000880)
0.512 0.002396 (0.001629)
* PM10, particulate matter less than 10 lm in aerodynamic diameter; pH, p value for heterogeneity of city-specific results; SE, standard error.
y The null hypothesis corresponds to perfect homogeneity.
6Stafoggia et al.
by guest on June 13, 2013
season, with the most toxic components at higher concen-
trations during warmer periods (19).
Our results seem to be consistent with the ‘‘ventilation’’
hypothesis, even though other explanations cannot be ruled
out. We found limited evidence of linear interaction, within
temperature strata, for total mortality (the interaction term
was statistically significant only when we shifted the cutoff
point of temperature at percentile 79th). No interaction of
the two environmental exposures emerged for respiratory
mortality and other causes of death, whereas the linear in-
teraction term was highly significant for cardiovascular
mortality in the third quartile of apparent temperature and
suggestive of a departure from a multiplicative model in the
Several strengths of the present study deserve consider-
ation. First, it included more than 300,000 deaths from nine
cities evenly spread across Italy, with detailed information
on causes of death: the large quantity of data enabled us to
explore several components of the PM10–temperature inter-
action. Second, to our knowledge, it is the first study to
explore the subject by using case-crossover methodology.
Third, we explored the PM10–temperature interaction two
different ways: by stratifying for apparent temperature and
assuming a linear relation between PM10and mortality and
by allowing both environmental exposures to affect mortal-
ity nonparametrically. Fourth, we explicitly addressed the
issue about possible causes of the effect modification.
Some limitations should be mentioned. First, given the
observational nature of the study, we could not identify any
causal effect of environmental exposures, only suggestive
associations and interactions. In this sense, residual bias due
to unmeasured confounding is still possible, and none of the
the present study. Nonetheless, the approach used controls
diameter (PM10; lag 0–1) and apparent temperature (lag 0–1) with all natural mortality among Italian residents of Milan, 1999–2004 (left), and
Rome, 1998–2004 (right), aged ?35 years. Three-dimensional plots in the upper half, contour plots in the lower half.
City-specific results: exposure-response surfaces relative to the joint association of particulate matter <10 lm in aerodynamic
Temperature, Air Pollution, and Mortality in Italy7
by guest on June 13, 2013
for potential individual confounders by design, and strong
residual confounding due to unmeasured time-dependent
covariates is unlikely. Second, measurement error in the
exposures cannot be excluded: daily PM10and temperature
values were estimated from central monitoring stations, so
that personal exposure may have differed substantially from
estimated exposure. However, it is unlikely to be differen-
tial. On the contrary, misclassification of the outcome is
doubtful in this study because of the broad categorization
of causes of death. Third, all analyses were not adjusted for
ozone, which is known to be highly correlated with temper-
ature and positively associated with mortality in warmer
periods. However, the correlation between PM10and ozone
is generally low; therefore, it is unlikely to be a confounder
of the PM10–mortality association (but it can affect the ex-
tent of the effect modification by temperature). Finally, we
did not know the ventilation status of the homes of the
decedents in our study, and we assumed that there were
more open windows on warm days, an assumption clearly
subject to error.
In conclusion, we identified strong effect modification by
season and temperature in the association between PM10
and natural and cause-specific mortality. The linear interac-
tion terms within temperature strata were in most cases not
statistically significant, and the hypothesis of higher effects
of PM10during warmer months due to higher exposure to air
pollution cannot be rejected. However, there was a sugges-
tion of a more than multiplicative pattern within the third
and fourth quartiles of apparent temperature for cardiovas-
cular mortality that deserves further research. Because air
pollution and high temperatures are strong predictors of
mortality, correct identification of their joint effect on health
outcomes is of primary interest from a public health per-
spective, especially considering the current debate over the
potential effects of climate changes.
This study was funded by the Lazio Region Health
The authors thank Margaret Becker for her help in edit-
ing the manuscript.
SISTI Group (Italian Study on Susceptibility to Temper-
ature and Air Pollution) members—Bologna: D. Agostini,
S. De Lisio, R. Miglio, P. Pandolfi, and C. Scarnato;
Firenze: A. Biggeri, E. Chellini, and S. Mallone; Mestre:
L. Simonato andR.Tessari;Milano:L.Bisanti,M.Rognoni,
and A. Russo; Palermo: A. Cernigliaro and S. Scondotto;
Pisa: M. Vigotti; Roma: V. Belleudi, F. de’Donato,
F. Forastiere, P. Michelozzi, C. A. Perucci, S. Picciotto, and
M. Stafoggia; Taranto: R. Primerano and M. Serinelli;
Torino: G. Berti, E. Cadum, N. Caranci, M. Chiusolo, and
Conflict of interest: none declared.
temperature (lag 0–1) and total natural mortality among residents of nine Italian cities aged ?35 years, by percentile of apparent temperature. Each
diamond represents the regression coefficient (and the vertical line the corresponding 95% confidence interval (CI)) of the linear interaction term
between PM10and apparent temperature on mortality, within the subset of days when the apparent temperature was above the relative percentile.
Pooled results: linear interaction between particulate matter <10 lm in aerodynamic diameter (PM10; lag 0–1) and apparent
8 Stafoggia et al.
by guest on June 13, 2013
1. Samet JM, Zeger SL, Dominici F, et al. The National
Morbidity, Mortality, and Air Pollution Study. Part II:
morbidity and mortality from air pollution in the United
States. Res Rep Health Eff Inst 2000;94(pt 2):5–70; dis-
2. Katsouyanni K, Touloumi G, Samoli E, et al. Confounding
and effect modification in the short-term effects of ambient
particles on total mortality: results from 29 European cities
within the APHEA2 project. Epidemiology 2001;12:
3. Schwartz J. The effects of particulate air pollution on daily
deaths: a multi-city case crossover analysis. Occup Environ
4. Biggeri A, Bellini P, Terracini B. Meta-analysis of the Italian
studies on short-term effects of air pollution-MISA 1996–2002
(Italian). Epidemiol Prev 2004;28(suppl):4–100.
5. Aga E, Samoli E, Touloumi G, et al. Short-term effects of
ambient particles on mortality in the elderly: results from 28
cities in the APHEA2 project. Eur Respir J Suppl 2003;40:
6. Zanobetti A, Schwartz J. Race, gender, and social status as
modifiers of the effects of PM10on mortality. J Occup Environ
7. Zeka A, Zanobetti A, Schwartz J. Individual-level modifiers
of the effects of particulate matter on daily mortality. Am J
8. Bateson TF, Schwartz J. Who is sensitive to the effects of
particulate air pollution on mortality? A case-crossover anal-
ysis of effect modifiers. Epidemiology 2004;15:143–9.
9. Kwon HJ, Cho SH, Nyberg F, et al. Effects of ambient air
pollution on daily mortality in a cohort of patients with con-
gestive heart failure. Epidemiology 2001;12:413–19.
10. Sunyer J, Schwartz J, Tobias A, et al. Patients with chronic
obstructive pulmonary disease are at increased risk of death
associated with urban particle air pollution: a case-crossover
analysis. Am J Epidemiol 2000;151:50–6.
11. Stafoggia M, Forastiere F, Agostini D, et al. Vulnerability to
heat-related mortality: a multicity, population-based, case-
crossover analysis. Epidemiology 2006;17:315–23.
12. Michelozzi P, De Sario M, Accetta G, et al. Temperature and
summer mortality: geographical and temporal variations in
four Italian cities. J Epidemiol Community Health 2006;60:
13. Braga AL, Zanobetti A, Schwartz J. The time course of
weather-related deaths. Epidemiology 2001;12:662–7.
14. O’Neill MS, Zanobetti A, Schwartz J. Modifiers of the tem-
perature and mortality association in seven US cities. Am J
15. Medina-Ramon M, Schwartz J. Temperature, temperature
extremes, and mortality: a study of acclimatization and effect
modification in 50 United States cities. Occup Environ Med.
Advance Access: June 28, 2007. (DOI:10.1136/oem.2007.
16. Rooney C, McMichael AJ, Kovats RS, et al. Excess mortality
in England and Wales, and in Greater London, during the 1995
heatwave. J Epidemiol Community Health 1998;52:482–6.
17. Schwartz J. Who is sensitive to extremes of temperature?
A case-only analysis. Epidemiology 2005;16:67–72.
18. Michelozzi P, Forastiere F, Fusco D, et al. Air pollution and
daily mortality in Rome, Italy. Occup Environ Med 1998;
19. Peng RD, Dominici F, Pastor-Barriuso R, et al. Seasonal
analyses of air pollution and mortality in 100 US cities. Am J
20. Nawrot TS, Torfs R, Fierens F, et al. Stronger associations
between daily mortality and fine particulate air pollution in
summer than in winter: evidence from a heavily polluted re-
gion in western Europe. J Epidemiol Community Health
21. Katsouyanni K, Pantazopoulou A, Touloumi G, et al. Evidence
for interaction between air pollution and high temperature in
the causation of excess mortality. Arch Environ Health
22. Roberts S. Interactions between particulate air pollution and
temperature in air pollution mortality time series studies.
Environ Res 2004;96:328–37.
23. Ren C, Tong S. Temperature modifies the health effects of
particulate matter in Brisbane, Australia. Int J Biometeorol
24. Hanninen O, Jantunen M, Nawrot TS, et al. Response to
findings on associations between temperature and dose re-
sponse coefficient of inhalable particles (PM10). J Epidemiol
Community Health 2007;61:838–9.
humidity index based on human physiology and clothing sci-
ence. JApplied Meteorol 1979;18:861–73.
26. Kalkstein LS, Valimont KM. An evaluation of summer dis-
comfort in the United States using a relative climatological
index. Bull Am Meteorol Soc 1986;67:842–8.
27. Katsouyanni K, Schwartz J, Spix C, et al. Short term effects of
air pollution on health: a European approach using epidemi-
ologic time series data: the APHEA protocol. J Epidemiol
Community Health 1996;50(suppl):s12–18.
28. Maclure M. The case-crossover design: a method for studying
transient effects on the risk of acute events. Am J Epidemiol
29. Levy D, Lumley T, Sheppard L, et al. Referent selection in
casecrossover analyses of acute health effects of air pollution.
30. R Development Core Team. R: a language and environment
for statistical computing. Vienna, Austria: R Foundation for
Statistical Computing, 2004. (http://www.R-project.org).
31. Eisen EA, Agalliu I, Thurston SW, et al. Smoothing in occu-
pational cohort studies: an illustration based on penalised
splines. Occup Environ Med 2004;61:854–60.
32. van Houwelingen HC, Arends LR, Stijnen T. Advanced
methods in meta-analysis: multivariate approach and meta-
regression. Stat Med 2002;21:589–624.
33. Wood SN, Augustin NH. GAMs with integrated model
selection using penalized regression splines and applications
to environmental modelling. Ecol Modell 2002;157:157–77.
34. Hastie TJ, Tibshirani RJ. Generalized additive models. New
York, NY: Chapman and Hall, 1990.
35. Wood SN. Thin plate regression splines. J R Stat Soc (B)
36. Lu Y, Zeger SL. On the equivalence of case-crossover and time
series methods in environmental epidemiology. Biostatistics
37. Katsouyanni K, Touloumi G, Spix C, et al. Short-term effects
of ambient sulphur dioxide and particulate matter on mortality
in 12 European cities: results from time series data from the
APHEA project. Air Pollution and Health: a European
Approach. BMJ 1997;314:1658–63.
38. Wong CM, Ma S, Hedley AJ, et al. Effect of air pollution on
daily mortality in Hong Kong. Environ Health Perspect 2001;
39. Xu Z, Gao J, Dockery DW, et al. Air pollution and daily
mortality in residential areas of Beijing, China. Arch Environ
Temperature, Air Pollution, and Mortality in Italy9
by guest on June 13, 2013
40. Samet J, Zeger S, Kelsall J, et al. Does weather confound or
modify the association of particulate air pollution with mor-
tality? An analysis of the Philadelphia data, 1973–1980.
Environ Res 1998;77:9–19.
41. Muggeo VMR. Bivariate distributed lag models for the anal-
ysis of temperature-by-pollutant interaction effect on mortal-
ity. Environmetrics 2007;18:231–43.
42. Gordon CJ. Role of environmental stress in the physiolog-
ical response to chemical toxicants. Environ Res 2003;92:
43. Sarnat JA, Koutrakis P, Suh HH. Assessing the relationship
between personal particulate and gaseous exposures of senior
citizens living in Baltimore, MD. J Air Waste Manag Assoc
10 Stafoggia et al.
by guest on June 13, 2013