Particles size and composition in Mediterranean countries:
geographical variability and short-term health effects
MED-PARTICLES Project 2011-2013
Under the Grant Agreement EU LIFE+ ENV/IT/327
Particles size and composition in Mediterranean countries:
geographical variability and short-term health effects
MED-PARTICLES
ACTION 11.
Health effects of PM10, coarse particles (PM2.5-10) and fine particles (PM2.5) on daily
total and cause-specific hospitalizations: city-specific results and meta-analysis
Summary: Report on the methodology and the results obtained analyzing the short term effect of air
pollution on hospital admissions in the MED-PARTICLES project.
-----------------------------------------------------------
Methods
Daily counts of emergency hospital admissions were collected from National or regional health
information systems for 10 European cities: Milan, Turin, Bologna, Parma, Reggio Emilia,
Modena, Rome, Marseille, Madrid and Barcelona. Since Parma, Reggio Emilia and Modena are
very close and share common environmental and socio-demographic characteristics, they have been
analyzed altogether as a single conurbation called “Emilia Romagna”. Only hospitalizations relative
to residents aged 15+ years were considered. Two study outcomes were defined on the basis of the
primary discharge diagnosis: cardiovascular hospitalizations (International Classification of
Diseases, 9th revision – ICD9: 390-459; 10th revision – ICD10: I00-I99), and respiratory
hospitalizations (ICD9: 460-519; ICD10: J00-J99). All data were extracted and collected according
to a common protocol.
Daily concentrations of PM2.5, PM10, nitrogen dioxide (NO2) and ozone (O3) were collected in each
city from multiple monitors belonging to the air quality monitoring networks. Daily mean
concentrations were computed for PMx and NO2, whereas daily maximum 8-hour running mean
was calculated for O3. Daily PM2.5-10 was calculated for each station as the difference between PM10
and PM2.5, provided that they were measured on the same stations and with the same sampling
methodology. Daily mean air temperature was collected from each center, using airport
meteorological stations where available.
Time-varying confounders were constructed according to a common protocol, under the assumption
of similar behavior of the study populations during holidays and vacation periods. They include:
holidays (a four-level variable assuming value “3” on Christmas and Easter; “2” in the surrounding
periods of Christmas and Easter; “1” on isolated holidays; “0” on other days); summer population
decrease (a three-level variable assuming value “2” in the 2-week period around the 15th of August;
value “1” from the 16th of July to 31st of August, with the exception of the aforementioned period;
value “0” all other days); influenza epidemics (a two-level variable assuming value “0” on normal
days, value “1” on days with particularly high influenza episodes. These were identified using
national influenza surveillance systems, where available, or on the basis of the daily counts of
hospitalizations for influenza (ICD9: 487; ICD10: J09-J11), as an alternative).
The analyses were carried out using a two-stage approach. In the first stage, city-specific overdispersed Poisson regression models were fitted, in which the dependent variable was the daily
count of hospitalizations, the exposure was the (lagged) daily concentration of PMx, and an
extensive list of variables was added for confounding adjustment. The adjustment model was
defined a priori and was common to all cities, in order to reduce the potential for heterogeneity in
the city-specific results. The confounders were: long-term and seasonal time trends, air temperature,
holidays, summer population decrease, and influenza epidemics.
Time trend was adjusted for by introducing a three-way interaction term between year, month and
day of the week.
High temperatures were adjusted for by calculating the average of current and previous day
temperature (lag 0-1) and by fitting a natural spline with two inner knots of the lagged variable only
for days with lag 0-1 temperature above the city-specific median value. Similarly, low temperatures
were adjusted for by fitting a natural spline with one inner knot of lag 1-6 air temperature only for
days below the city-specific median value. The other confounders, including holidays, summer
population decrease and influenza epidemics, were adjusted for with indicator variables.
Once the adjustment model was defined at the city level, the pollutant was added to the regression
model. The lag structure of the association between PMx concentrations and daily counts of causespecific hospitalizations was inspected by applying cubic polynomial distributed lag models, from
lag 0 to 5, where all lags were entered simultaneously in the model and were constrained to follow a
polynomial shape. In addition, three cumulative lag structures were a priori defined to represent
immediate, delayed or prolonged effects: lags 0-1, lags 2-5 and lags 0-5, respectively. Finally, for
each PM/outcome combination, one of these three alternatives was chosen as the “reference lag” for
the following analyses, on the basis of the meta-analytical results from the distributed and
cumulative lags models.
The robustness of results to other pollutants co-exposure was evaluated by fitting bi-pollutant
models, in which the co-pollutant was modeled with the same cumulative lag as that chosen as
reference for each specific exposure/outcome combination. Models including ozone as co-pollutant
were also fitted by restricting the study period to the warm season only, defined as April to
September.
In the second stage of the analysis, the city-specific results were pooled by using random-effects
meta-analytical procedures. The presence of heterogeneity in the city-specific results was evaluated
by applying the X2 test from Cochran’s Q statistic and its amount was quantified by computing the
I2 statistic (Higgins and Thompson 2002), which represents the proportion of total variation in
effect estimates that is due to between-cities heterogeneity.
All the results are expressed as percent increases of hospitalizations, and 95% CI, relative to fixed
increments in each PM fraction: 10 g/m3 for PM2.5, 6.3 g/m3 for PM2.5-10 and 14.4 g/m3 for
PM10. These increments have been chosen in order to represent the same amount of daily variability
across pollutants, so to allow a better inter-pollutant comparison of results.
Results
The population base consisted of subjects aged 15+ years resident in 8 cities of Southern Europe,
totaling more than 11 million inhabitants (Table 1). The mean daily counts of hospitalizations
ranged from 18 in the Emilia Romagna conurbation to 116 in Madrid for cardiovascular diseases,
and from 8 in Bologna to 97 in Madrid for respiratory conditions. The study periods were recent
and comparable across cities, the only exception being Marseille with 2001-2003 data.
The pooled results from polynomial distributed lag models show an immediate effect of the three
PM fractions on cardiovascular hospitalizations, up to lag 1, whereas the effect was prolonged until
day five with regard to respiratory admissions.
Table 2 reports the effect estimates of PM on hospitalizations for the three cumulative lags 0-1, 2-5
and 0-5. A strong and similar effect on cardiovascular hospitalizations was apparent for all PM
fractions: increments of 10 g/m3 in PM2.5, 6.3 g/m3 in PM2.5-10 and 14.4 g/m3 in PM10 (lag 0-1)
were associated with % increases of 0.43 (95% confidence interval (CI): 0.04, 0.83), 0.50 (95% CI:
0.14, 0.85) and 0.47 (95% CI: 0.00, 0.95), respectively. Effect estimates on cardiovascular
morbidity were null for the delayed cumulative lags 2-5, and small and non-significant on the
cumulative lags 0-5. Hence, coupled with the evidence coming from the distributed-lag models
result, lag 0-1 was adopted for this outcome on the following analyses.
The effect estimates of all three PM fractions on respiratory hospitalizations were higher at the
cumulative lag 0-5, though statistically significant only for PM10 and highly heterogeneous across
cities. In particular, effect estimates were: 1.25% (95% CI: -0.02, 2.54) for 10 g/m3 increase in lag
0-5 PM2.5, 1.31% (95% CI: -0.10, 2.74) for 6.3 g/m3 increase in lag 0-5 PM2.5-10, and 1.13% (95%
CI: 0.14, 2.12) for 14.4 g/m3 increase in lag 0-5 PM10. Noteworthy, estimates were more
homogeneous at the immediate lag 0-1, and statistically significant for the coarse fraction (0.62%,
95% CI: 0.11, 1.13). The cumulative lag 0-5 was considered to be the reference lag for the
association between any PM fraction and respiratory hospitalizations and was adopted in further
analyses.
Table 1. Study population. Resident population and emergency hospital admissions for cardiovascular and respiratory causes among subjects age
15+ years in the 8 cities of MED-PARTICLES
Population
Study
period
Date
Milan
Turin
Emilia Romagna
Bologna
Marseille
Rome
Barcelona
2006 - 2010
2006 - 2010
2008 - 2010
2006 - 2010
2001 - 2003
2006 - 2010
2003 - 2010
Madrid
City
Total
a
Cardiovascular admissions
Respiratory admissions
n
n
x 1,000
p/years
Daily
mean
n
x 1,000
p/years
Daily
mean
Jan 1st, 2008
Jan 1st, 2008
Jan 1st, 2009
Jan 1st, 2008
Census 1999
Jan 1st, 2008
Jan 1st, 2007
1,299,633
908,263
529,699
372,256
796,525
2,718,768
1,595,110
71,779
48,967
19,717
34,568
46,905
153,176
147,890
11.0
10.8
12.4
18.6
19.6
11.3
11.6
39.3
26.8
18.0
18.9
42.8
83.9
50.6
34,427
21,761
10,164
14,103
18,069
53,825
135,356
5.3
4.8
6.4
7.6
7.6
4.0
10.6
18.9
11.9
9.3
7.7
16.5
29.5
46.3
2004 - 2009
Jan 1st, 2007
3,132,503
201,041
13.5
115.9
167,850
11.3
96.7
-
-
11,352,757
724,043
12.5
51.2
455,555
7.8
32.2
Cities are ordered by latitude, North to South
Table 2. Association between PM and hospitalizations: % increase of hospital admissions, and 95%
CI, associated with increases of 10, 6.3 and 14.4 g/m3 for PM2.5, PM2.5-10 and PM10, respectively
Pollutant
Cardiovascular admissions
Respiratory admissions
% increase (95% CI)
% increase (95% CI)
0-1
2-5
0-5
0.43 (0.04, 0.83)
0.02 (-0.45, 0.49)
0.36 (-0.21, 0.93)
0.40 (-0.20, 1.00)
1.01 (-0.17, 2.20)a
1.25 (-0.02, 2.54)a
0-1
2-5
0.50 (0.14, 0.85)
-0.22 (-0.63, 0.20)
0.62 (0.11, 1.13)
0.84 (-0.48, 2.18)a
0-5
0.21 (-0.38, 0.79)
1.31 (-0.10, 2.74)a
0-1
2-5
0-5
0.47 (0.00, 0.95)
-0.08 (-0.51, 0.35)
0.26 (-0.24, 0.77)
0.58 (0.11, 1.05)
0.95 (-0.05, 1.95)a
1.13 (0.14, 2.12)a
Lag
PM2.5
PM2.5-10
PM10
a
Statistically significant heterogeneity, as indicated by p<0.10 from Cochran’s Q and I2 >50%
The pooled results from single and bi-pollutant models are reported in Table 3. When evaluated
simultaneously on cardiovascular hospitalizations, fine and coarse particles displayed the same
effect estimate (not statistically significant 0.33% increase of admissions). In contrast, the coarse
particles displayed an effect twice that of fine particles on respiratory admissions (0.84% increase
for a 6.3 g/m3 increment of lags 0-5 PM2.5-10 versus 0.43% increase for a 10 g/m3 increment of
lags 0-5 PM2.5). The adjustment for NO2 did not substantially affect the association between PM2.510
and either study outcome, whereas it had opposite effects on the association between fine
particles and hospitalizations for cardiovascular or respiratory causes. No confounding from ozone
was apparent in the all-year analysis, nor in the analysis restricted to the warm period. Noteworthy,
the results from both single and bi-pollutant models displayed much higher effect estimates, always
statistically significant, when the analyses were restricted to the warm period April to September.
Table 3. Association between PM and hospitalizations from single and bi-pollutant models: %
increase of hospital admissions, and 95% CI, associated with increases of 10, 6.3 and 14.4 g/m3
for PM2.5, PM2.5-10 and PM10, respectively
Cardiovascular admissions
(lag 0-1)
Respiratory admissions
(lag 0-5)
% increase (95% CI)
% increase (95% CI)
all-year
0.43 (0.04, 0.83)
1.25 (-0.02, 2.54)a
+ PM2.5-10
all-year
0.33 (-0.12, 0.78)
0.43 (-0.83, 1.70)
+ NO2
all-year
0.14 (-0.34, 0.62)
1.72 (0.43, 3.03)
+ O3
all-year
0.35 (-0.02, 0.73)
0.93 (-0.49, 2.36)a
april-september
1.75 (0.68, 2.84)
4.49 (1.72, 7.35)a
april-september
2.11 (1.13, 3.10)
4.51 (1.35, 7.77)a
all-year
0.50 (0.14, 0.85)
1.31 (-0.10, 2.74)a
+ PM2.5
all-year
0.33 (-0.07, 0.73)
0.84 (-0.42, 2.12)
+ NO2
all-year
0.30 (-0.09, 0.70)
1.15 (-0.44, 2.77)a
+ O3
all-year
0.53 (0.18, 0.89)
1.20 (-0.33, 2.76)a
april-september
0.90 (0.28, 1.52)
3.01 (0.68, 5.39)a
april-september
0.93 (0.37, 1.49)
3.03 (0.44, 5.68)a
all-year
0.47 (0.00, 0.95)
1.13 (0.14, 2.12)a
+ NO2
all-year
0.17 (-0.44, 0.79)
1.50 (0.20, 2.83)
+ O3
all-year
0.43 (-0.07, 0.94)
0.92 (-0.14, 1.99)a
april-september
2.07 (1.30, 2.85)
4.33 (2.58, 6.11)
april-september
2.25 (1.45, 3.06)
4.98 (1.98, 8.06)
Pollutant
PM2.5
PM2.5
+ O3
PM2.5-10
PM2.5-10
+ O3
PM10
PM10
+ O3
a
Period
Statistically significant heterogeneity, as indicated by p<0.10 from Cochran’s Q and I2 >50%
Conclusions: We investigated the association between daily concentrations of fine and coarse particles
and hospitalizations for cardiorespiratory conditions in eight Mediterranean cities, finding evidence of
harmful effects of the PM metrics on both study outcomes
We identified positive associations of coarse particles with both cardiovascular (0.46% increase; 95%
CI: 0.10, 0.82 with a 6.3-μg/m3 increase in lag 0–1 PM2.5–10) and respiratory admissions (1.24%
increase; 95% CI: –0.32, 2.82) with a 6.3-μg/m3 increase in lag 0–5 PM2.5–10). Corresponding
associations with a 10-μg/m3 increase in PM2.5–10 were higher than previously reported (0.73%; 95%
CI: 0.16, 1.30%, and 1.95%; 95% CI: –0.51, 4.48%, respectively).
Associations between fine particles and cardiovascular hospitalizations in our study were not affected
by coarse PM co-exposure, but the associations of both fine and coarse PM with respiratory admissions
decreased to nonsignificance when evaluated together in two-pollutant models.