American Journal of Epidemiology
Copyright C 1997 by Trie Johns Hopkins University School of Hygiene and Public Hearth
All rights reserved
Vol 145, No 3
Printed in U S A
Ozone, Suspended Particulates, and Daily Mortality in Mexico City
Victor H. Borja-Aburto,1 Dana P. Loomis,2 Shnkant I. Bangdiwala,3 Carl M. Shy,2 and
Ramon A. Rascon-Pacheco1
To investigate acute, irreversible effects of exposure to ozone and other air pollutants, the authors examined
daily death counts in relation to air pollution levels in Mexico City during 1990-1992. When considered singly
in Poisson regression models accounting for penodic effects, the rate ratio for total mortality associated with
a 100-ppb increment in 1-hour maximum ozone concentration was 1 024 (95% confidence interval (Cl)
1.011-1.039). Measures of average ozone concentration were somewhat more strongly related to mortality.
The rate ratio was 1.024 (95% Cl 0.984-1.062) per 100 ppb for sulfur dioxide and 1 050 (95% Cl 1.030-1.067)
per 100 /ig/m 3 for total suspended particulates. However, when all three pollutants were considered simultaneously, only total suspended particulates remained associated with mortality, indicating excess mortality of
6% per 100 jxg/m 3 (rate ratio = 1.058, 95% Cl 1.033-1.083), consistent with observations in other cities in the
United States and Europe. The authors found no independent effect of ozone, but it is difficult to attribute
observed effects to a single pollutant in light of the complexity and variability of the mixture to which people
are exposed. Nevertheless, particulate matter may be a useful indicator of the nsk associated with ambient air
pollution. Am J Epidemiol 1997;145:258-68
air pollution; mortality; environment
During the past decade, an extensive literature has
developed concerning the acute effects of particulate
air pollutants on mortality in cities in the Americas and
Europe (1-4). Epidemiologic studies have found a
positive association between air pollution and total,
respiratory, and cardiovascular mortality at pollutant
exposure levels below the current United States National Ambient Air Quality Standard, and these observed effects have been similar across studies (2, 3).
However, the association of daily mortality with other
pollutants has not been so thoroughly examined.
Ozone exposure is associated with an array of transient adverse respiratory effects in healthy people, as
well as in people with preexisting respiratory disease,
such as asthma, and with increased frequency of hospital and emergency room visits (5). An association
between ozone and mortality can also be hypothesized
on the basis of knowledge of ozone's acute and intermittent pulmonary effects, combined with observations of the relation of particulate air pollutants to
mortality. Epidemiologic findings (6) offer some support for this possibility.
This epidemiologic study was undertaken to evaluate the hypothesis that air pollution, with emphasis on
ozone exposure, is related to excess daily mortality in
Mexico City, where ozone levels range widely and
exceed the United States and international air quality
standards almost every day.
MATERIALS AND METHODS
This study examined daily variations in mortality in
relation to air pollution in a defined area of Mexico
City for the 3-year period from 1990 through 1992.
Details of the research methods are provided elsewhere (7).
Received for publication March 25, 1996, and in final form October 28, 1996
Abbreviations: RR, rate ratio; Cl, confidence interval, TSP, total
suspended particulates; PM-10, particulate matter with an aerodynamic diameter of less than 10 fi/n; ICD-9, International Classification of Diseases, Ninth Revision.
1
Institute Nacional de Salud Publica, Cuemavaca, Morelos,
Mexico.
2
Department of Epidemiology, School of Pubic Health, University of North Carolina, Chapel Hill, NC.
3
Department of Btostatistics, School of Public Health, University
of North Carolina, Chapel Hill, NC.
Repnnt requests to Dr. Dana P. Loomis, Department of Epidemiology, CB#7400, McGavran-Greenberg Hall, University of North
Carolina, Chapel Hill, NC 27599-7400.
Study area
The research area encompassed the Distrito Federal
(Federal District) of Mexico, which includes more
than eight million inhabitants, approximately 50 percent of the population of the Mexico City metropolitan
area (figure 1). A small area of the Distrito Federal,
with 31,944 inhabitants (Milpa Alta), was excluded
258
Ozone, Suspended Particulates, and Daily Mortality 259
Area
• Southwest
a Southeast
•a Downtown
Northeast
a Northwest
a Excluded
Explanation
[~"}
Monitor sites
Mexico City
Metropolitan Area
D.F. border
FIGURE 1. Political boundaries of Mexico City Metropolitan Area, Distrito Federal and subareas delimited for this study
because it does not have an air-monitoring station. The
Distnto Federal has a young population, 30 percent of
whom are under age 15 years and only 5 percent of
whom are 65 or older and, consequently, a relatively
low crude death rate of 5.4 X 1,000"'.
Air pollution differences within Mexico City can be
large due to meteorologic factors, particularly wind
direction. To take this heterogeneity into account, we
performed subanalyses of five geopolitically defined
subareas (figure 1), each containing an air qualitymonitoring station, in addition to considering the Distrito Federal as a whole. The five areas vary in population size and death rate (table 1).
Air pollution measurements
The Departamento del Distrito Federal supplied data
on air quality and weather (tables 1 and 2). Exposure
measurements were based on the nine monitoring stations with the most complete information for the study
period; these stations record concentrations of sulfur
dioxide, carbon monoxide, ozone, nitrogen oxides, and
an array of meteorologic parameters, measured hourly,
and total suspended particulates (TSP), recorded every
sixth day as a 24-hour integrated measure. Air pollutant concentrations are measured using US Environmental Protection Agency standard methods: ozone by
TABLE 1. Descriptive statistics for air quality and mortality data for five subareas of Mexico City,
1990-1992*
PdramAtAr
roitUlroltJI
Population (thousands)
Mean daily deaths
Mean daily noninjury deaths
Mean ozone, 1-hour daily
maximum (ppb)
Mean TSP (ng/n>»)
Mean PM-101 (ng/rn3)
Mean sulfur dioxide (ppb)
Mean carbon monoxide (ppb)
Mean relative humidity (%)
Mean minimum temperature (°C)
Area
Southwest
Southeast
Central
Northeast
2,082
25
1,968
22
22
19
1,971
34
32
1,268
18
17
881
15
14
177
116
137
335
109
213
140
220
49
5,479
54
10
39
6,101
54
11
148
191
88
56
6,011
49
12
69
6,002
44
11
51
5,232
47
11
* Restricted to 211 days when total suspended particulates (in ng/m2) (TSP) were available,
t PM-10, particulate matter with aerodynamic diameter <10 urn
Am J Epidemiol
Vol. 145, No. 3, 1997
Northwest
260
Bona-Aburto et al.
TABLE 2.
Distributions of dally data on mortality, meteorology, and air pollution, Mexico City, 1990-1992
Parameter
No
Minimum
5%
25%
50%
75%
95%
Maximum
t
Air quality and weather*
Ozone, 1-hour daily maximum
Ozone, daily mean
Ozone, moving average
Ozone, 8AM-6PM average
Sulfur dioxide daily, mean
Carbon monoxide, daily mean
Temperaturet
TSP*
Relative humidity
1,072
1,072
1,066
1,067
1,066
1,066
1,057
216
1,065
25
13
16
17
15
2,032
14
66
19
68
29
45
43
26
3,536
6.3
103
28
Mortal ity§
All ages
Ages <5 years
Ages >65 years
1,084
1,084
1,084
69
3
25
83
7
39
122
42
74
70
44
4,980
9.0
151
39
94
11
46
155
54
94
87
53
5,798
11.4
204
49
181
69
114
107
61
6,580
12.8
271
59
225
83
139
130
780
7,703
14.0
368
73
285
126
179
170
110
11,972
16.8
456
85
103
14
52
115
16
59
132
22
71
154
28
85
* Ozone, sulfur dioxide, and carbon monoxide measured in parts per billion.
t Temperature, mean dally minimum temperature (In °C)
t TSP, total suspended particulates (in u.g/m3)
§ Total mortality, excluding external causes of death.
ultraviolet photometry, sulfur dioxide by pulsed fluorescence, carbon monoxide by nondispersive infrared
photometry, nitrogen oxides by chemoluminescence,
and TSP by the gravimetric method. Gravimetric measurements of paniculate matter with aerodynamic diameter under 10 microns (PM-10) were available only
for 1991 and 1992 from one monitor in the central city
at La Merced, operated by the Centro de Ciencias de la
Atm6sfera of Universidad Nacional Aut6noma de
M6xico (Dr. Irma Rosas, personal communication,
1993).
Some values of gaseous pollutant concentrations for
particular stations and days were missing. Prior to
estimating average levels, these missing values were
imputed using the mean of all available stations with a
station-specific correction factor (6). No more than 10
percent of the values for ozone levels had to be imputed for each station for the complete series from
1990 to 1992. Missing values were treated as missing,
rather than imputed, if more than 30 consecutive or
more than 90 separated observations were absent in a
year.
The primary metric of ozone exposure was the daily
1-hour maximum concentration, the basis of United
States and international standards. The daily maximum 1-hour concentration was defined as the highest
value among the valid hourly measurements for each
day. Exposures to particulates, sulfur dioxide, and
carbon monoxide were measured by the 24-hour mean
concentration. Exposures to these agents were examined with lags of 1-4 days, as well as for the current
day.
In an exploratory analysis, we also examined the
association of mortality with four other indices of
ozone exposure: the mean concentration for 24 hours,
the mean between 8 a.m. and 6 p.m. each day, an
8-hour moving average around the daily maximum,
and 3-day cumulative exposure. The 8 a.m. to 6 p.m.
mean concentration was calculated as the sum of the
available values during the time interval, divided by
the number of valid hourly measurements; the 24-hour
mean was calculated as the sum of the available measurements divided by the number of valid number of
hours during each calendar day; the 8-hour moving
average was computed by selecting the maximum
value of the day and then determining the average of
the previous 4 hours, the maximum value, and the 3
hours after the maximum value; and the 3-day cumulative index was the sum of the means for the current
day and the 2 preceding days.
To estimate these exposure indicators, at least six
observations per day were required. If less than six
observations were available, that day was treated as
missing.
Mortality data
The Instituto Nacional de Estadistica, Geografia e
Inform£tica of Mexico supplied detailed, individual
mortality data in the form of individual records. Each
record includes age, sex, date of death, date of death
registration, delegaci6n (county) of death, delegacidn
of residence, and cause of death. The data were reduced to total and cause-specific numbers of deaths
per day. Only deaths of residents of the Distrito Federal that occurred within the Distrito Federal were
considered.
Am J Epidemiol
Vol. 145, No. 3, 1997
Ozone, Suspended Particulates, and Daily Mortality
The number of daily deaths, rather than the death
rate, is used as the outcome measure, since the size of
the population had not substantially changed during
the period. The deaths were divided into the following
three groups according to the International Classification of Diseases, Ninth Revision (ICD-9): 1) respiratory diseases, including acute respiratory infection
(ICD-9 codes 460-466), pneumonia and influenza
(ICD-9 codes 480-487), chronic obstructive pulmonary disease and allied conditions (ICD-9 codes 490496), pneumoconiosis and other lung diseases due to
external agents (ICD-9 codes 500-508), and symptoms involving the respiratory system and other chest
symptoms (ICD-9 codes 768 and 786); 2) cardiovascular diseases, defined as hypertensive disease (ICD-9
codes 401-405), ischemic heart disease and diseases
of pulmonary circulation (ICD-9 codes 410-417),
stroke (ICD-9 codes 430-438), and symptoms involving the cardiovascular system (ICD-9 code 785); and
3) external causes (codes E800-E999).
Data analysis
Poisson regression was used to model the daily
death counts as a function of air pollution parameters
while controlling simultaneously for potentially confounding covariates. To control for periodic phenomena and seasonal trends, we conducted a sensitivity
analysis to test different smoothed functions of time as
explanatory variables in the regression model.
An array of knotted cubic spline functions were fit
to the data to directly remove the longer wavelength
fluctuations (8). A sine-cosine function was also considered as an alternative function to remove long-cycle
variations. To control for possible residual cyclic variation, indicator variables for month and day of the
week were also tested for inclusion in the model.
Results from regressions using these functions were
compared with those using only temperature (a known
predictor of seasonal variations in mortality) to control
for long-term cycles to select the simplest model that
best explained variations in mortality.
The model that fit the mortality data best according
to goodness-of-fit statistics and visual examination of
residuals was developed before adding terms for air
pollutants. Among different indices of temperature,
the daily minimum with a 0-day lag was the best
predictor of daily mortality; mean and maximum temperatures have a limited range in Mexico City. Relative humidity was also considered for inclusion but
was not associated with mortality.
Other air pollutants known or suspected to be related to mortality were also evaluated. TSP, sulfur
dioxide, and carbon monoxide were considered. Nitrogen oxides were hypothesized to be an ozone preAm J Epidemiol
Vol. 145, No. 3, 1997
261
cursor and were not evaluated. In addition, TSP is a
weaker predictor of mortality than is PM-10 (2, 3), so
its validity as a measure of potential confounding by
particulate pollution was evaluated by comparing the
estimated association of ozone and mortality in the
central area on days when both were available adjusted
alternately for TSP and PM-10.
The ability to evaluate the effect of particulate matter was limited by the every-sixth-day sampling regime used for TSP. To enhance the comparability of
single- and multiple-pollutant models, all analyses reported here are based only on the subset of days when
TSP was available. Models for gaseous pollutants
based on all available days (approximately 1,070)
gave similar results for single pollutants (7).
Regressions were stratified by age group (all ages,
less than age 5 years, and age 65 years and over) to
identify differential effects by age. To account for
serial correlation common in longitudinal data (9), the
final models were reestimated using the iteratively
weighted and filtered least-squares method (1, 10), an
extension of Poisson regression designed to account
for possible overdispersion and autocorrelation. To
evaluate the evidence of dose-response relation between mortality and ozone and to allow for possible
nonlinearity, additional analyses were performed with
ozone and TSP divided into quintiles, indicated by
nominal variables.
RESULTS
The 1-hour daily maximum value of ozone was
above the US standard of 120 ppb approximately 75
percent of the days (table 2), with no apparent seasonal
trend (figure 2). Other characteristics of mortality and
air quality are shown in table 2. Minimum temperature, relative humidity, sulfur dioxide, and particulate
matter showed strong seasonal variation (data not
shown). Temperature varied from almost 0°C in
December-January to 17°C in May, while the highest
levels of particulates and sulfur dioxide were present
during January and February, when thermal inversions
typically occur in Mexico City.
TSP measurements for the central area of Mexico
City were highly correlated with same-day measurements of PM-10 from the independent monitor (correlation coefficient = 0.82), with a PM-10:TSP ratio
of approximately 0.50.
The four indices of ozone exposure were highly
correlated. The maximum correlation was observed
between the 8 a.m.-6 p.m. average and the moving
average with a correlation coefficient over 0.91, since
the maximum value usually occurred at 2 p.m.
Pollutant levels varied across the city (table 1), with
the highest ozone levels in the southwest, elevated
262
Borja-Aburto et al.
27O
22O -
n
o.
a.
x"
a
17O "
12O -
o
8
70 "
20 "
182
365
557
Time in days
730
92O
1095
FIGURE 2. Time series of 1-hour maximum ozone levels (in parts per billion), Mexico City, 1990-1992 Each O represents 1 day of
observation
TSP concentration in the southeast, and the highest
sulfur dioxide concentration in the northeast. Temperature and humidity were similar across areas, however.
Spatial variations were examined with pairwise correlations; gaseous pollutants were weakly correlated
among the five areas, with correlation coefficients
around 0.50. In contrast, there were strong spatial
correlations among areas for paniculate matter, with
correlation coefficients around 0.85.
Total mortality peaked during the winter months.
However, exploratory analyses showed that the number of registered deaths for the last day of the year
dropped unexpectedly, probably due to administrative
problems with death registration. Therefore, the interval December 29 to January 1 was excluded from all
analyses.
Poisson regression models with minimum temperature, a spline function with six knots, and a sine-cosine
function all fit the mortality time series well. Adding
indicator variables for season, month, and day-ofweek effects did not significantly reduce the variance.
Results using all of these methods to control for longterm trends in mortality were similar, but the model
using only temperature was preferred because it has
stronger biologic justification and removed long-cycle
variation with the fewest variables. Only results based
on this basic model are shown here. Similar results
from models based on the spline function are reported
elsewhere (7). Figure 3 shows the residuals of the
model with minimum temperature as the only predictor of mortality. The overdispersion parameter (10) for
this basic model indicated 12 percent excess dispersion of mortality.
Total mortality, cardiovascular mortality, and mortality for those over age 65 years were associated with
1-hour maximum ozone concentration after adjustment for temperature to control for seasonal effects
(table 3). However, these rate ratios diminished after
adjustment for TSP (table 3). Similar results were
obtained when maximum ozone concentration was
categorized in quintiles rather than modeled as a continuous variable (figures 4 and 5). No other meaningful changes in the estimates were observed when sulfur dioxide or carbon monoxide was added with TSP
already in the model.
The simple association of peak ozone concentration
with total mortality was strongest in the northwest
(rate ratio (RR) = 1.083 per 100 ppb) and the northeast (RR = 1.030) areas and weaker or nonexistent in
the central area (RR = 1.010), the southeast (RR =
0.970), and the southwest (RR = 0.961). However,
adjustment for TSP also eliminated the positive associations between ozone and mortality in all subareas
except the northwest (RR = 1.030). The smaller number of deaths within subareas made confidence intervals wider at this level of analysis; the confidence
interval for the northwest region before adjustment for
TSP was the only one that did not include unity.
Further analyses were performed by season. Mexico
City's tropical climate has only two seasons: rainy
(May-October) and dry (November-April). When considered without other air pollutants, adjusting only for
temperature, peak ozone concentration and total mortality were related in the rainy season (RR = 1.019, 95
percent confidence interval (CI) 1.006-1.039) and
only weakly associated in the dry season (RR = 1.012,
Am J Epidemiol
Vol. 145, No. 3, 1997
Ozone, Suspended Particulates, and Daily Mortality
263
2O
15 1O 5"
V
e
e
a.
O -5 -1O
-15 "
-2O "
182
365
557
730
92O
1095
Time in days
FIGURE 3. Residuals from a Poisson regression model using minimum temperature as a predictor of daily mortality, Mexico City,
1990-1992. Each O represents 1 day of observation.
TABLE 3. Mortality rate ratios (or a 100-ppb Increase In daily maximum 1-hour ozone exposure
(restricted to 211 days when ozone, TSP,* and temperature were available), Mexico City, 1990-1992
Adjusted for
temperature
Total mortality
Respiratory mortality
Cardiovascular disease mortality
Ages <5 years
Ages >65 years
Adjusted for
temperature and TSP
RR»
95% Cl*
RR
95% Cl
1.024
1 023
1 036
0 967
1 039
1 011-1.039
0.981-1.067
1.006-1.066
0.931-1.003
1.019-1 060
0.982
0.981
0 900-1 014
1024
0 962
0 991
0 890-1 082
0 956-1.096
0.882-1 049
0.946-1.036
* TSP, total suspended particulates; RR, relative risk; Cl, confidence Interval, estimated by Poisson
regression.
95 percent Cl 0.993-1.032). However, with TSP also
in the model, mortality and ozone concentration were
not related in either season (RR = 0.994 and 0.990 for
dry and rainy seasons, respectively).
Indicators of average ozone exposure were generally more strongly associated with mortality than the
1-hour maximum, particularly for cardiovascular disease mortality and among persons over age 65 years
(table 4). Cumulative exposure over 3 days was not as
strongly associated, with rate ratios generally in the
same range as those for the 1-hour maximum (table 4).
In general, associations involving average and cumulative ozone levels were diminished or eliminated by
adjustment for TSP, like those for the peak concentration. However, an exception to this pattern was observed for cardiovascular disease mortality, for which
the rate ratios remained notably elevated with TSP in
the model (table 4).
The effects of several pollutants besides ozone were
also noteworthy when they were considered in sepaAm J Epidemiol
Vol. 145, No. 3, 1997
rate models. The rate ratio for total mortality was
1.024 (95 percent Cl 0.984-1.062) per 100 ppb of
sulfur dioxide and 1.050 (95 percent Cl 1.030-1.067)
per 100 /xg/m3 of TSP. In contrast, carbon monoxide
was only weakly associated with mortality (RR =
1.013, 95 percent Cl 1.003-1.023) and was not considered further. When all pollutants were simultaneously included in the model (table 5), the rate ratio
for TSP remained elevated, indicating excess mortality
of 6 percent per 100 jig/m3 (RR = 1.058, 95 percent
Cl 1.033-1.083), while the rate ratio for ozone was
reduced to 1.0. The rate ratio for sulfur dioxide was
not notably affected by adding TSP to the model. The
independent effect of TSP was stronger for respiratory
mortality (RR = 1.095 per 100 /xg/m3, 95 percent Cl
1.013-1.184) (table 5). TSP behaved similarly when
analyzed as a categorical variable (figure 6).
The effect of TSP was consistent across areas of the
city (7). In contrast, TSP was more strongly related to
total mortality during the dry season (RR = 1.037, 95
264
Borja-Aburto et al.
1 04I
I
S
I
II
1-
•
0 96-
n
Q? —
62 5
142 8
1173
1632
217 8
Ozone 1-hour maximum
FIGURE 4. Rate ratios by quintile of 1-hour maximum ozone concentration for association with total mortalrty adjusted for temperature,
Mexico City, 1990-1992.
1 04-
o
CO
s
ir
1-
•
i
i
ii
ii
0 96-
62 5
1173
1428
1632
217 8
Ozone 1-hour maximum
FIGURE 5. Rate ratios by quintile of 1-hour maximum ozone concentration for association with total mortality adjusted for minimum
temperature plus total suspended particulates, Mexico City, 1990-1992.
percent CI 1.011-1.063) when TSP levels are highest,
than during the rainy season (RR = 1.011, 95 percent
CI 0.963-1.061), although both point estimates were
included in each others' confidence intervals. Sulfur
dioxide was also associated with mortality in the dry
season (RR = 1.048, 95 percent CI 0.930-1.180), but
not in the wet months (RR = 0.994, 95 percent CI
0.838-1.179). The effect of temperature was similar
regardless of season, however.
Examination of various lag intervals indicated the
strongest association for same-day exposures. The association was diminished with 1-day lags for all polAm J Epidemiol
Vol. 145, No. 3, 1997
Ozone, Suspended Particulates, and Daily Mortality
265
TABLE 4. Mortality rate ratios for 100-ppb increase In alternative measures of ozone exposure
(restricted to 211 days when ozone andTSP* were available), Mexico City, 1990-1992
Adjusted for
temperature
Adjusted (or
temperature and TSP
RR*
95%CI»
RR
95% Cl
Total mortality
24-hour mean
8-hour moving average
8 AM-6 PM mean
3-day cumulative
1.058
1.043
1.040
1.016
1.022-1.094
1.021-1 064
1.018-1 062
1.006-1 054
0.975
0.999
0.992
1005
0 901-1.055
0 953-1.047
0.946-1.041
0.986-1.026
Respiratory mortality
24-hour mean
8-hour moving average
8 AM-6 PM mean
3-day cumulative
1.015
1 034
1 028
1.026
0.912-1 131
0.969-1.101
0.963-1 093
1.000-1.054
0.924
0.969
0 957
0 989
0 724-1.179
0 839-1 116
0 828-1 107
0 932-1 051
Cardiovascular disease mortality
24-hour mean
8-hour moving average
8 AM-6 PM mean
3-day cumulative
1.121
1.077
1.072
1.038
1.042-1 207
1.031-1 125
1.025-1 122
1 020-1.057
1.080
1.066
1.072
1.041
0 913-1.280
0 964-1 178
0 969-1 189
0.998-1.086
Ages <5 years
24-hour mean
8-hour moving average
8 AM-6 PM mean
3-day cumulative
0.820
0 927
0933
0.962
0.747-0.902
0.876-0.980
0.881-0 988
0.940-0 985
0.786
0 928
0.927
0.963
0 634-0.973
0.817-1.053
0 814-1.054
0 912-1.016
Ages >65 years
24-hour mean
8-hour moving average
8 AM-6 PM mean
3-day cumulative
1 122
1.074
1.069
1.069
1 069-1 178
1 043-1.010
1.038-1.010
1.024-1 049
1025
1.022
1.015
1.026
0 918-1.147
0.957-1 092
0.948-1.085
0 997-1 055
* TSP, total suspended particulates; RR, relative nsk; Cl, confidence Interval, estimated by Poisson
regression.
lutants except sulfur dioxide and disappeared with lags
of 2 or more days (table 6).
No interaction terms were significant, and no serious problems of collinearity were evident in the models, since all condition indices (an index of collinearity) were low (11). As a control on the analytic
methods, we examined the effect of ozone and TSP on
deaths from external causes (data not shown). As
expected, no association was found. Relative to standard Poisson regression, the use of iteratively
weighted and filtered least-squares models did not
materially change either the point estimates of the
association of mortality with air pollutants or their
precision.
DISCUSSION
Analysis of mortality among the population of Mexico City from 1990 to 1992 shows increases in deaths
with air pollution levels on the same day and on the
previous day. An association was observed with the
Am J Epidemiol
Vol. 145, No. 3, 1997
daily maximum concentration of ozone, the basis of
current standards (RR = 1.024, 95 percent Cl 1.0111.039 per 100 ppb) and also with the mean concentrations of sulfur dioxide and particulate matter, when
they were considered in separate regression models.
However, when the effect of all pollutants was modeled simultaneously, only TSP had a noteworthy independent effect, with excess mortality increasing 6
percent per 100 /xg/m3. These relations were consistent across the city.
This relative increase is potentially important given
the large population exposed in the Federal District.
Assuming that the entire population of eight million is
exposed, approximately 4,500 excess deaths per year
would be associated with a change of 100 /xg/m3 in
TSP.
An exploratory analysis of other indices of ozone
exposure suggested that average and cumulative exposures may be independently associated with mortality from cardiovascular diseases. Rate ratios were on
the order of 1.07-1.08 for a 100-ppb increase in con-
266
Borja-Aburto et al.
TABLE 5. Adjusted rate ratios (per 100-unlt change In
pollutant concentration and 10°C change In temperature) from
IWFLS* log-linear regressions containing minimum
temperature, ozone,TSP,* and sulfur dioxide, whh 0 day-lag
(restricted to 211 days when ozone and TSP were available),
Mexico City, 1990-1992
RR»
95% Cl*
Total mortality
Minimum temperature
Sulfur dioxide
TSP
Ozone maximum
0.881
1.022
1.058
0.999
0.825-0 942
0 898-1 163
1.033-1083
0.942-1.019
Ages >65 years
Minimum temperature
Sulfur dioxide
TSP
Ozone maximum
0 801
1.022
1 059
0.989
0 738-0 976
0 871-1.201
1 027-1 090
0.941-1 041
Respiratory mortality
Minimum temperature
Sulfur dioxide
TSP
Ozone maximum
0.638
0.907
1.095
1.002
0.518-0.786
0.600-1.372
1.013-1.184
0.885-1.135
Cardiovascular mortality
Minimum temperature
Sulfur dioxide
TSP
Ozone maximum
0 803
1 071
1.052
1.021
0.715-0.901
0.855-1.341
1.009-1 099
0.948-1.100
• IWFLS, Iteratively weighted and filtered least-squares method
for Polsson regression, developed by Samet and Zeger (1), TSP,
total suspended particulatss; RR, rate ratio; Cl, confidence
interval.
centration after adjustment for TSP, although statistical precision was limited.
Previous epidemiologic studies have yielded varied
results with respect to ozone and mortality. Kinney
and Ozkaynak (6) reported positive associations of
mortality with ozone levels in long-term studies of Los
Angeles, California, and New York City. However,
subsequent analyses of data from Los Angeles indicated a relative risk of 1.02 for a 143-ppb increment in
ozone concentration, with the relative risk diminishing
to 1.0 when a term for particulate pollutant concentration was added to the model (12). Ozone was also
related to mortality in an analysis of data from Philadelphia, Pennsylvania, but only during the summer
(13). A study in Santiago, Chile, found no relation
between same-day ozone concentration and total mortality, but ozone concentration lagged by 1 day had a
relative risk of 1.04 for a 100-ppb change (14). No
association was found in Detroit, Michigan; St. Louis,
Missouri; or two Tennessee cities (3, 15). A study of
infant respiratory deaths in Sao Paulo, Brazil, also
indicated no association with ozone (16).
In contrast, excess mortality is consistently associated with particulate pollutant levels in cities with
differences in air pollution and population characteristics. In a meta-analysis, Schwartz (2) reported an
estimated risk ratio of 1.06 (95 percent CI 1.05-1.07)
for a 100-jLtg/m3 increase in TSP mass, essentially
identical to the rate ratio we observed. The overall
congruence of our findings with this pattern, which
has been reported in many recent studies in the Americas and western Europe (3, 4, 13, 14, 17-20), is
particularly noteworthy considering that in Mexico
City TSP measurements were available for only every
sixth day. Mexico City also has vastly different demographic characteristics from most cities in North
America and Europe and a lower crude death rate
because of its young age distribution.
Our findings concerning particulates do differ in
detail from those of several studies in Philadelphia,
where repeated reanalyses have been carried out. In
contrast to results reported by Li and Roth (21), we
found the results were insensitive to the modeling
approach. Unlike Moolgavkar et al. (13), we found no
evidence of mutual confounding between sulfur dioxide and particulates and no evidence that the relations
between pollutants changed with season. Our analysis
did suggest differences in the effect of particulates by
season, but interpretation is complicated by statistical
imprecision and seasonal differences in the range of
exposures.
Misspecification of exposure is a significant concern in environmental epidemiologic studies. We assigned air pollutant levels from fixed, outdoor monitors to individuals who died to estimate their exposure.
In Mexico City, neither hospitals nor homes are tightly
sealed. However, ozone is highly reactive, so levels
indoors are lower than those outdoors. Reports from
the southern area of Mexico City (22, 23) found ratios
of outdoor/indoor concentrations of 2:1 to 3.5:1. Mexico City residents spend, on average, 2.76 hours outdoors each day (24, 25) in contrast to 1 hour or less
among residents of US cities (26-28). Thus, it appears
that ozone concentrations measured by fixed monitors
may overestimate actual exposure.
In this study, however, exposure measurement error
is of the Berkson type because all deaths on any given
day are grouped and assigned a common value for
exposure and covariates. Nondifferential Berkson-type
errors do not bias observed exposure-response relations (29). Exposure measurement error in epidemiologic studies can generally be regarded as nondifferential with regard to disease status unless there is
evidence to the contrary, and in this study, there is no
reason to suspect any other pattern, despite undoubted
imperfections in the air pollution data.
The influence of annual, seasonal, and weekly cycles on mortality was examined using various methods
Am J Epidemiol
Vol. 145, No. 3, 1997
Ozone, Suspended Particulates, and Daily Mortality
0 95
103 1
161 1
205 76
259 94
267
373 34
TSP
FIGURE 6. Rate ratios by quintlle of 24-hour integrated concentration of total suspended particulates (TSP) for association wrth total
mortality, adjusted for minimum temperature, sulfur dioxide, and ozone, Mexico City, 1990-1992.
TABLE 6. Ad|usted rate ratios (per 100-unH change In pollutant level or 10°C change In temperature)
from IWFLS* log-linear regressions containing ozone, TSP,* and sulfur dioxide, with 1- and 2-day lags
and unlagged minimum temperature (restricted to 211 days when ozone and TSP were available), Mexico
City, 1990-1992
1-day lag
2-day lag
RR*
95% Cl*
RR
95% Cl
Total mortality
Sulfur dioxide
TSP
Ozone maximum
1.126
1 025
0.990
0 967-1 313
0 997-1 055
0.949-1.031
1.045
1.024
0.974
0 931-1.173
1 000-1.048
0 937-1.012
Ages >65 years
Sulfur dioxide
TSP
Ozone maximum
1.093
1.045
1.006
0.913-1.308
1.011-1 079
0.955-1.059
0.922
1.014
1.028
0.796-1 069
0.985-1.045
0 976-1.082
• IWFLS, iteratlvely weighted and filtered least-squares method for Poisson regression, developed by Samet
and Zeger (1); TSP, total suspended particulates; RR, rate ratio, Cl, confidence interval.
to control for periodic mortality cycles. However, the
results were insensitive to the method used to control
for these variations (7).
In other locations, excess mortality is associated
with both high and low air temperatures (19, 30).
However, the temperature in Mexico City is moderate
(extremes of >36°C or <0°C are almost never observed), so neither mean nor maximum temperature
was strongly predictive of mortality.
While we observed associations between several air
pollutants and mortality, few of these associations
were statistically independent. In light of the complexity and variability of the atmospheric mixtures to
Am J Epidemiol
Vol. 145, No. 3, 1997
which urban populations are exposed, attempts to examine the effect of ozone while controlling for other
pollutants raise both philosophic and technical challenges (31-34). Statistical techniques allow one agent
to be assessed while artificially holding the level of
others constant, even though this situation may not
exist in nature. We also found no evidence of statistical interaction between air pollutants, but causal interactions through which the occurrence of health effects
is governed by specific combinations of pollutants
may nevertheless be present. Fortunately, isolating the
specific, biologically active pollutant responsible for
empiric associations with mortality may not be neces-
268
Borja-Aburto et al.
sary to protect public health. Pollutants produced by
burning fossil fuels are interrelated, so to control their
adverse effects, it may be sufficient to identify an
index pollutant that consistently predicts those effects.
Our observation of an independent association between cardiovascular disease mortality and indicators
of average exposures to ozone over a day or several
days suggests that these indices may be worth further
investigation as supplements or alternatives to the
1-hour maximum on which standards are now based.
In addition, the consistency of our findings about
particulate pollutants with those of numerous other
studies indicates that it may be reasonable to consider
particulate concentrations as an indicator for the risk
associated with air pollution, a use that does not require certainty about the agent's causal role.
14.
15.
16.
17.
18
19
20.
21
ACKNOWLEDGMENTS
22
Supported by a grant from the Health Effects Institute.
Dr. Victor Borja-Aburto was supported in part by
CONACYT-Mexico while at the University of North Carolina, Chapel Hill.
23.
REFERENCES
24.
1 Samet JM, Zeger SL, Berhane K Particulate air pollution and
daily mortality Replication and validation of selected studies
Cambridge, MA. Health Effects Institute, 1995
2 Schwartz J. Air pollution and daily mortality, a review and
meta-analysis. Environ Res 1994;64:36-52.
3 Dockery DW, Schwartz J, Spengler JD Air pollution and
daily mortality associations with particulates and acid aerosols. Environ Res 1992,59:362-73.
4 Spix C, Heinnch J, Dockery D, et al. Air pollution and daily
mortality in Erfurt, East Germany, 1980-1989 Environ
Health Perspect 1993;101:518-26.
5. Lippmann M Health effects of troposphenc ozone: review of
recent research findings and their implications to ambient air
quality standards J Exp Anal Environ Epidemiol 1993;3p
103-29.
6 Kinney P, Ozkaynak H Associations between ozone and daily
mortality in Los Angeles and New York City. (Abstract) Am
Rev Respir Dis 1992;145.A95
7. Loomis DP, Borja-Aburto VH, Bangdiwala SI, et al. Ozone
exposure and daily mortality in Mexico City a Ume-senes
analysis. HEI Research Report no. 75 Cambridge, MA:
Health Effects Institute, 1996
8. Durrleman S, Simon R. Flexible regression models with cubic
splines. StatMed 1989;8.551-61.
9. Liang KY, Zeger SL. Regression analysis for correlated data.
Annu Rev Public Health 1993; 14:43-68.
10. McCullagh P, Nelder JA. Generalized linear models 2nd ed.
Cambridge, MA Chapman & Hall, 1989200
11. Kleinbaum DG, Kupper LL, Muller KE. Applied regression
analysis and other multivariable methods. Boston, MA: PWSKent Publishing, 1988:213.
12. Kinney PL, Ito K, Thurston GD. A sensitivity analysis of
mortality/PM-10 associations in Los Angeles. Inhalation
Toxicol 1995;7:59-69
13 Moolgavkar SH, Luebeck EG, Hall TA, et al. Air pollution
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
and daily mortality in Philadelphia. Epidemiology 1995;6:
476-84.
Ostro B, Sanchez JM, Aranda C, et al. Air pollution and
mortality results from a study of Santiago, Chile J Expo Anal
Environ Epidemiol 1996,6:97-114.
Schwartz J. Particulate air polluUon and daily mortality in
Detroit. Environ Res 1991,56:204-13.
Saldiva PHN, Lichtenfels AJFC, Paiva PSO, et al. Association
between air polluuon and mortality due to respiratory diseases
in children in Sao Paulo Brazil A preliminary report. Environ
Res 1994;65:218-25
Schwartz J, Dockery DW Particulate air pollution and daily
mortality in Steubenville, Ohio. Am J Epidemiol 1992,135.
12-19
Schwartz J. Air polluUon and daily mortality in Birmingham,
Alabama. Am J Epidemiol 1993;137:1136-47.
Touloumi G, Pocock SJ, Katsouyanni K, et al Short-term
effects of air pollution on daily mortality in Athens, a time
series analysis. Int J Epidemiol 1994;23957-67.
Verhoeff AP, Hoek G, Schwartz J, et al Air pollution and
daily mortality in Amsterdam, the Netherlands Epidemiology
1996;7.225-30.
Li Y, Roth HD. Daily mortality analysis by using different
regression models in Philadelphia County, 1973—1990. Inhalation Toxicol 1995;7:45-58.
Cortez-Lugo M, Garcia-Franco M, Ramirez-Solis Y, et al
Evaluation of interior and exterior air quality in a kindergarten
in southern Mexico City. Proceedings of the Sixth International Conference on Indoor Air Quality and Climate, Helsinki, Finland, July 4 - 8 , 1993. 19933:223-6.
Perez-Neria J, Villegas DH, Rojas-Ramos M, et al Contaminacion del aire en intenores. Comparacion simultanea de la
concentracion de ozono intra y extramuros (In Spanish). Rev
Inst Nal Enf Resp Mex 1994;7:14-20.
Rojas-Bracho L. Evaluacion del grado de exposicion a aeroparticulas en los habitantes de la zona metropolitana de la
ciudad de Mexico (In Spanish). Master's Thesis. Mexico
City, Mexico. Universidad Nacional Autonoma de Mexico,
1994.
Fernandez-Bremauntz A, Quentin MJ. A survey of commuter
travel habits in the metropolitan area of Mexico City J Exp
Anal Environ Enid 1992;Suppl 2:1-17.
Schwab M, Colome SD, Spengler JD, et al. Activity patterns
applied to pollutant exposure assessment: data from a personal
monitoring study in Los Angeles Toxicol Indust Health 1990;
6:517-32
Clayton CA, Pemtt RL, Pellizzan ED, et al. Particle total
exposure assessment methodology (PTEAM) study distribution of aerosol and elemental concentrations in personal, indoor, and outdoor air samples in a southern California community. J Exp Anal Environ Epidemiol 1993;3 227-50.
Freeman NCG, Waldman JM, Lioy PJ. Design and evaluation
of a location and acUvity log used for assessing personal
exposure to air pollutants. J Exp Anal Environ Epidemiol
1991;! 327-38.
Armstrong BG. The effects of measurement errors on relaUve
risk regressions Am J Epidemiol 1990;132:l 176-84
Kunst A, Looman CWN, Mackenbach JP. Outdoor air temperature and mortality in the Netherlands: a time-senes analysis Am J Epidemiol 1993;137:331-41.
Samet JM. Particulate air polluUon and mortality: the Philadelphia story. Epidemiology 1995;6.471-3.
Dockery DW, Schwartz J Particulate air polluUon and
mortality: more than the Philadelphia story. Epidemiology
1995,6 629-32.
Moolgavkar SH, Luebeck EG, Hall TA, et al. Particulate air
pollution and mortality. (Letter). Epidemiology 1996;7:
212-13.
Dockery DW, Schwartz J. The authors reply. (Letter). Epidemiology 1996;7:213-14.
Am J Epidemiol Vol. 145, No. 3, 1997