|
Executive
Summary
Background
and objectives: Locally derived data are needed for the SAR
Government and regulatory agencies to support the drafting
and implementation of effective legislation for the management
and control of the environment. In 1996, the Environmental
Protection Department commissioned the Chinese University
of Hong Kong to carry out a study on health effects of air
pollution and in 1997 the University of Hong Kong to carry
out a follow-up validation study. The study performed by the
University of Hong Kong identified some irregularities in
the way the data were selected, managed and analysed in the
previous study. It also obtained effect estimates which were
more comparable to the results obtained by the collaborating
centres of Air Pollution on Health: a European Approach (APHEA).
Finally it pointed to new directions in methodology and in
modelling methods adopted by the APHEA Phase Two study which
could be used in Hong Kong.
In this
study, reported here, the problems of the earlier Hong Kong
studies were resolved in that (i) three years of data were
used to obtain better models; (ii) non-linearity effects in
the pollutant and covariates were modelled using a generalized
additive modelling method; and (iii) patterns in the daily
variations were explained as far as possible by using more
appropriate wave lengths, introducing influenza epidemics
in the models and using more recently developed statistical
methods to control for autocorrelation in the data.
Health
outcomes: Daily counts of hospital admissions with a discharge
diagnosis of respiratory disease (ICD9 460-519), chronic obstructive
pulmonary disease (COPD) (490-496 excluding 493), asthma (493),
cardiac disease (390-429) and ischaemic heart disease (IHD)
(410-414), for all ages, as well as hospital deaths due to
respiratory (460-519) and cardiovascular (390-459) diseases,
for all ages, were used as the 7 major health outcomes. Among
them, admissions were sub-categorised as follows: for respiratory
disease, into 0-14, 15-64, 65-74 and 75+ age groups; for asthma,
into 0-14 and 15-64 age groups; for cardiac disease, into
15-64, 65-74 and 75 age groups and for IHD, into 15-64, 65-74
and 75+ age groups. These categories for all ages and specific
age groups formed 19 categories of health outcomes for analysis.
Summary
of findings: The effects for each pollutant without adjustment
for co-pollutant, on the above-mentioned 19 health outcomes
and with adjustment for the most confounding co-pollutant,
on the above 7 major health outcomes are summarized in this
section.
NO2:
NO2 was shown to have strong effects on all the
19 health outcomes under study without adjustment for co-pollutants,
except on the IHD 15-64 and all ages groups. The effects on
the 7 major health outcomes, with adjustment for co-pollutant
and autocorrelation, were significant at the 5% level except
IHD admissions and cardiovascular hospital deaths for all
ages. The effects on asthma admissions and respiratory deaths
for all ages were likely to be non-linear. The effects on
respiratory, COPD, cardiac and IHD admissions for all ages
were higher in the cool than in the warm season.
SO2:
SO2 did not have an effect on most of the 19 health
outcomes under study without adjustment for co-pollutant.
With adjustment for co-pollutant and autocorrelation, it had
an effect on COPD admissions for all ages (p=0.005). The effect
on cardiac admissions for all ages was higher in the cool
than in the warm season.
RSP: RSP
was associated with most of the 19 health outcomes under study
without adjustment for co-pollutant. With adjustment for co-pollutants
and autocorrelation, it had an effect on respiratory (p=0.023)
and asthma (p=0.045) admissions for all ages and the effects
were likely to be non-linear. The effect on cardiac admissions
for all ages was higher in the cool than in the warm season.
O3:
O3 had effects on most of the 19 health outcomes
under study without adjustment for co-pollutant. With adjustment
for co-pollutant and autocorrelation, it had strong effects
on respiratory and asthma admissions for all ages (p=0.000).
The effects on COPD, cardiac and IHD admissions for all ages
were stronger in the cool than in warm season.
FSP: FSP
(PM2.5) had effects on most of the 19 health outcomes
under study without adjustment for co-pollutant. With adjustment
for co-pollutant, the effects were not significant for all
of the 7 major health outcomes for all ages (p>0.192). But
as the pollutant was measured in only one monitoring station
and there were missing data problems, the effects need to
be further investigated.
Public
health implications: The results from this study showed strong
short-term effects of air pollutants on hospital admissions
and hospital deaths.
In the
daily number of hospital admissions for all ages, there was
a 4-8% increase due to respiratory disease; 5-15% increase
due to chronic obstructive pulmonary disease and 11-22% increase
due to asthma. There was also a 3-7% increase in admissions
due to a cardiac disease. There was a 3-5% increase due to
a cardiovascular disease (ICD9 390-459), which includes cardiac
disease as well as cerebrovascular disease and diseases of
arteries, arterioles, capillaries, veins and lymphatics. These
percentage increases are all expressed as rates per 50 ug/m3
increase in the concentration of air pollutants (NO2,
SO2, O3 and RSP).
In the
daily number of hospital deaths for all ages, there was a
10% increase due to respiratory disease and 8% increase due
to cardiovascular disease for every 50 ug/m3 increase
in NO2. For other pollutants the increases were
not significant and they need to be further investigated.
The results
suggest that air pollution may show stronger health effects
in Hong Kong than those obtained in European cities.
Conclusions:
The results from this study show that short-term effects of
air pollutants on health are strong. It is both feasible and
worthwhile to develop further studies on the health effects
of air pollution with a longer series of data, using the methodology
adopted by the APHEA protocol. Territory-wide SAR mortality,
based on daily mortality counts from the Census and Statistics
Department, should be included as one of the main health outcomes
to be studied.
1.0
Background and introduction
In 1997,
the Department was offered a contract by the Environmental
Protection Department (EPD) to study the short-term effects
of ambient air pollution on public health using hospital admission
data in Hong Kong for the period January 1995 to June 19971
and to validate the findings of another similar study on data
for the period January 1994 to June 1996.2 The
results revealed that, similar to other studies overseas,
there were significant effects on morbidity and mortality
of individual pollutants i.e. nitrogen dioxide (NO2),
sulphur dioxide (SO2), respirable suspended particulates
(RSP) and ozone (O3). The study performed by the
Department modified the way the data were selected, managed
and analysed in the previous study and obtained effect estimates
which were more comparable to the results obtained by collaborating
centres of Air Pollution on Health: a European approach (APHEA).
In line with new directions in methodology and in modelling
methods adopted by the APHEA-2 protocol,3 our report
also pointed out some limitations of both of the Hong Kong
studies and indicated what should be done in any future studies.
The limitations
in the Hong Kong studies will be further resolved and discussed
in this new phase of study. First, with three years of data,
a better prediction model will be developed. Second, non-linearity
effects in the pollutant(s) or covariates will be modelled
using a generalized additive model (in statistical package
S-Plus). Third, unexplained patterns in the daily variations
will be explained as far as possible by (i) using more appropriate
wave lengths for seasonality and terms for time trends, to
eliminate as far as possible all unobserved confounding effects;
(ii) introducing influenza epidemics in the model to control
as much of the observed confounding effects as possible in
addition to those due to air temperature and humidity etc;
and (iii) using more recently developed statistical methods
to control for any auto-correlation and/or extra-Poisson variations
in the data.
2.0
Scope and objectives
The following
tasks will be completed by the Department of Community Medicine.
-
Single
pollutant: A model for each health outcome is fitted with
terms to account for all long-term and seasonal patterns
and other possible confounding effects. Short-term daily
variations are then studied and accounted for by daily
variations in SO2, NO2, RSP and
O3 individually, based on data from 1 January
1995 to 31 December 1997.
-
Multiple
pollutants: From each single pollutant model, the joint
effects of each pollutant with other pollutants are studied
by putting them in the model. Interactions and multi-collinearity
among pollutants are also studied and taken account of,
based on the same data as in a) above.
-
Single
pollutant PM2.5: The effect due to PM2.5
is studied in a similar way as in a) above, but using
data from 1 August 1995 to 31 December 1997 when the PM2.5
concentrations are available from the EPD.
-
Multiple
pollutants with PM2.5: Study b) above but with
PM2.5 instead of RSP. Compare the models with
either PM2.5 or RSP in terms of stability (reliability)
in effect estimates.
-
Verification
of model: The best model developed from a) to d) above
is then used to obtain the predicted daily numbers of
admissions for each health outcome. The observed number
of admissions in the data 1 January 1998 to 30 June 1998
are compared with the predicted for verification (i.e.
predictive validity) of model.
3.0
Materials and methods
| 3.1 |
Study
design
The
same design as in the previous report was used. It utilized
routinely collected hospital admission data, air pollutant
concentration data and meteorological data by the Hospital
Authority, Environmental Protection Department and Observatory.
Variations in the daily number of hospital admissions
were studied, and their relationships with each of the
pollutants were modelled to assess the effects of air
pollution on health after adjustment for confounding
factors.
|
| |
|
| 3.2 |
Databases
In
the APHEA-2 protocol there were changes in age group
definitions and missing data handling methods from those
in APHEA. All the changes were incorporated in this
report as far as possible. In order to be more comparable
with results from APHEA studies, there were also some
alterations in the choice of health outcomes, in that
the category of chronic obstructive pulmonary disease
(ICD9 490-496) excluded asthma (ICD9 493), and cardiac
diseases (ICD9 390-429) was used instead of circulatory
disease (ICD9 390-459) in the admission categories.
Hospital
admission data: The hospitals included for compilation
of hospital admissions were the publicly funded hospitals
(accounting for 90% of hospital beds in Hong Kong) with
an accident & emergency department, or those which
were the referral base from the accident & emergency
department of another hospital or had a 24-hour outpatient
department. One other hospital which was the only hospital
in the most polluted district in Hong Kong was also
included. All hospitals should have a computerized system
for inputting and retrieval of patient data (Basic Table
A1).
Information
for patients admitted to the above-mentioned hospitals
between 1.1.1995 and 30.6.1998, on date of admission,
discharge diagnosis (in ICD9 codes) and age (and other
information not required in this study) were retrieved
from the databases. The number of cases by discharge
diagnosis and age, grouped according to APHEA-2 protocol
(hospital deaths not in APHEA-2 categories, were also
included) were summarized in Table 1 below.
| Table
1: |
Number
of patients admitted between 1.1.1995 and 30.6.1998
by discharge diagnoses and age groups |
| Number
of cases |
| Discharge
diagnosis (ICD9 codes) |
Age
groups |
1995-1997 |
1998
(Jan-June) |
| Respiratory
disease (460-519) |
All
ages* |
223,225 |
46,877 |
| 0-14
yrs |
74,164 |
16,450 |
| 15-64
yrs |
48,909 |
9,451 |
| 65-74
yrs |
38,783 |
7,640 |
| 75+ |
61,318 |
13,329 |
Chronic
obstructive pulmonary disease
(490-496 excluding 493) |
All
ages* |
53,374 |
9,795 |
| Asthma
(493) |
All
ages* |
27,080 |
4,286 |
| 0-14
yrs |
14,005 |
2,133 |
| 15-64
yrs |
8,520 |
1,332 |
| Cardiac
disease (390-429) |
All
ages* |
108,150 |
18,835 |
| 15-64
yrs |
37,195 |
5,961 |
| 65-74
yrs |
32,205 |
5,715 |
| 75+ |
37,356 |
6,976 |
| Ischaemic
heart disease (410-414) |
All
ages* |
39,483 |
6,940 |
| 15-64
yrs |
14,259 |
2,402 |
| 65-74
yrs |
13,969 |
2,507 |
| 75+ |
11,201 |
2,029 |
| Deaths:
Respiratory (460-519) |
All
ages* |
11,798 |
2,095 |
| Cardiovascular
(390-459) |
All
ages* |
11,202 |
1,779 |
*
Including patients without information on age.
Pollution
data: Hourly concentrations of air pollutants (Table
2) by monitoring stations (Table 3) were provided by
the Air Services Group of the Environmental Protection
Department.
| Table
2: |
List
of pollutants used in the study |
| Pollutant |
Unit |
| 1.
Nitrogen Dioxide {24-hr} (NO2) |
micrograms/cubic
metre |
| 2.
Sulphur Dioxide {24-hr} (SO2) |
micrograms/cubic
metre |
| 3.
Respirable Suspended Particulates {24-hr} (RSP)* |
micrograms/cubic
metre |
| 4.
Ozone {8-hr} (O3) |
micrograms/cubic
metre |
| 5.
Fine Suspended Particulates {24-hr} (FSP) |
micrograms/cubic
metre |
*
Measured by Tapered Element Oscillating Microbalance
(TEOM), an instrument for the continuous measurement
of particulate matter in air.
| Table
3: |
List of air pollution monitoring stations |
| Station |
Sampling
height |
Above
ground |
Date
start operation |
| 1.
Central/West (C/W) |
78m |
18m
(4th floor) |
11/83 |
| 2.
Kwai Chung (KC) |
82m |
25m
(6th floor) |
7/88 |
| 3.
Kwun Tong (KT) |
34m |
25m
(6th floor) |
7/83 |
| 4.
Sham Shui Po (SSP) |
21m |
17m
(4th floor) |
7/84 |
| 5.
Shatin (ST) |
27m |
21m
(5th floor) |
7/91 |
| 6.
Tai Po (TP) |
31m |
25m
(6th floor) |
2/90 |
| 7.
Tsuen Wan (TW) |
21m |
17m
(4th floor) |
8/88 |
Two
stations were excluded, one in Yuen Long (in suburban
area) due to the extent of missing data and one in Mong
Kok (in urban area) which provided concentrations measured
on ground level.
In
each station, for O3 the 8-hourly (9 am -
5 pm when O3 concentrations were highest)
means were taken as daily data if there were at least
6 valid hourly data each day; and for the other pollutants
the 24-hourly means were used as daily data if there
were at least 18 valid hourly data. Other data were
missing data. The percentages of non-missing daily data
were shown in Table 4 below.
| Table
4: |
Percentage
of valid measures of air pollutants in each monitoring
station |
|
Station# |
| Pollutant |
C/W |
KC |
KT |
SSP |
ST |
TP |
TW |
| NO2
(24-hr average) |
| 1995 |
93 |
92 |
96 |
89 |
97 |
97 |
95 |
| 1996 |
93 |
96 |
95 |
90 |
96 |
95 |
93 |
| 1997 |
91 |
90 |
87 |
98 |
93 |
89 |
85 |
| 1995-97 |
92 |
93 |
93 |
92 |
95 |
93 |
91 |
| 1998* |
97 |
96 |
94 |
85 |
97 |
98 |
96 |
| SO2
(24-hr average) |
| 1995 |
98 |
97 |
97 |
93 |
99 |
- |
95 |
| 1996 |
98 |
95 |
98 |
89 |
99 |
- |
98 |
| 1997 |
92 |
93 |
93 |
96 |
94 |
- |
82 |
| 1995-97 |
96 |
95 |
97 |
93 |
97 |
- |
91 |
| 1998* |
97 |
98 |
94 |
86 |
97 |
- |
94 |
| RSP
(24-h average) |
| 1995 |
85 |
95 |
79 |
- |
87 |
- |
94 |
| 1996 |
99 |
98 |
92 |
- |
87 |
- |
86 |
| 1997 |
90 |
92 |
85 |
- |
91 |
- |
90 |
| 1995-97 |
91 |
95 |
85 |
- |
89 |
- |
90 |
| 1998* |
99 |
95 |
95 |
- |
99 |
- |
92 |
| O3
(8-hr average) |
| 1995 |
94 |
93 |
- |
- |
- |
- |
- |
| 1996 |
92 |
95 |
- |
- |
- |
- |
- |
| 1997 |
87 |
90 |
- |
- |
- |
- |
- |
| 1995-97 |
91 |
93 |
- |
- |
- |
- |
- |
| 1998* |
92 |
96 |
- |
- |
- |
- |
- |
| FSP
(24-hr average) |
| 1995
(Aug-Dec) |
- |
- |
- |
- |
- |
- |
94 |
| 1996 |
- |
- |
- |
- |
- |
- |
82 |
| 1997 |
- |
- |
- |
- |
- |
- |
92 |
| 1995
(Aug)-97 |
- |
- |
- |
- |
- |
- |
88 |
| 1998* |
- |
- |
- |
- |
- |
- |
97 |
*
the data are available only for first half year of 1998
- not available
# Abbreviations referred to Table 3.
Daily
mean concentrations representing the pollutant levels
in all Hong Kong areas were obtained from non-missing
data for each of the pollutants under study by means
of the following steps:
-
Estimate
the annual mean concentration of each pollutant
in each monitoring station and subtract this annual
mean from the daily concentrations of the corresponding
year. The resulting series is regard as "centered".
-
Take the arithmetic mean over all monitoring stations
of these centered series day by day.
-
Finally add the annual mean of all stations to the
series obtained in step 2. The series for each health
outcome is used for the analysis.
Overall,
daily means obtained using APHEA-2 method were highly
correlated (Spearman's correlation coefficient above
0.99) with those obtained using the previous APHEA method
(Basic Table D3).
Meteorological
data: Daily means of temperature and relative humidity
were obtained from the Environmental Protection Department
which derived the data from the Observatory.
|
| |
|
| 3.3 |
Statistical modelling
The
statistical modelling methods followed the guideline
recommended by the APHEA-2 protocol and are outlined
as follows:
| a) |
Poisson
regression with adjustment for over-dispersion using
quasi-likelihood4 approach was used to
model the hospital admission counts of the health
outcomes under study. |
| |
|
| b) |
Using
Poisson regression mentioned in a) above, a core
model was established to remove all the long term
trends and seasonal variations by means of the
following covariates:
| Covariate |
Explanation |
| t
(day) |
Daily
trend |
| t2 |
Daily
curvature |
| cos
(2?kt/365.25) |
Seasonality |
| sin
(2?kt/365.25) |
for
k cycles per year to be determined by spectral
analysis |
| Monday |
Monday
vs Sunday |
| Tuesday |
Tuesday
vs Sunday |
| Wednesday |
Wednesday
vs Sunday |
| Thursday |
Thursday
vs Sunday |
| Friday |
Friday
vs Sunday |
| Saturday |
Saturday
vs Sunday |
| Holiday1 |
Holidays |
| Holiday2 |
Day
after holidays |
| Temperature |
Linear
effect of temperature |
| Humidity |
Linear
effect of relative humidity |
| Influenza |
-
Dummy variable for weeks with number of
influenza admissions (ICD9 487) above
75 percentile in a year
|
The
numbers of seasonal cycles per year were obtained
from spectral analysis5 instead of
1-4 cycles used in the previous study.
|
| |
|
| c)
|
The
residuals from fitting the core model for each health
outcome, were examined for any discernible patterns,
lack of normality and autocorrelation in them. |
| |
|
| d) |
If
there are no discernible patterns and lack of normality
in the residuals, then potential confounding is
not evident in the daily variations. Air pollutant
concentrations were then fitted to obtain the air
pollution effects for pollutants. Autocorrelations
were taken account of for identified models. |
| |
|
| e) |
In
order to obtain an indicator of the amount (proportion)
of variations in the health outcome explained, the
number of admissions was tranformed by taking the
logarithm and multiple regression was then applied.
The value R2 was used to quantify the
proportion of variations explained by the core model. |
| |
|
| f) |
Effects
were obtained per unit change in pollutant levels
measured by current day and average of current day
up to a previous day, called the best lag day, to
be determined by AkaikeÕs Information Criterion
with the minimum value. The best lag day was determined
up to previous 5 days for ozone and up to previous
3 days for the other pollutants. |
| |
|
| g)
|
Interaction
effect between each pollutant (in the original scale)
and each of the other pollutants was checked by
first dichotomizing the other pollutants using the
median and multiplying each of them to obtain the
interaction term, and then introducing the interaction
term into the model. When there was an interaction
effect, the air pollutant effect was estimated for
high and low levels of the other pollutant. |
| |
|
| h) |
Interactions
between pollutant concentrations and season were
checked by defining a season dummy variable with
1 for cool (January-April; November-December) and
0 for warm (May-October) season, multiplying the
season dummy variable with concentrations to obtain
the interaction term and fit each interaction term
to the model. When there was an interaction effect,
relative risks in cool and warm seasons were obtained
respectively. |
| |
|
| i)
|
Effects
due to copollutants were estimated by putting the
other pollutants one by one in the model. Effect
with adjustment for a copollutant which was associated
with the maximum adjusting effect was used. |
| |
|
| j)
|
Autocorrelations
were adjusted for using Newy-West method.6
Non-linear copollutant effects were studied by
means of generalised additive model.7
Non-linear
dose-response relationships of pollutants were
also studied using generalised additive model.
|
| |
|
| k)
|
Validation
of the models for 1995-97 was achieved by comparing
the observed numbers of health outcomes in the first
half year of 1998 with the predicted from the model
developed using 1995-97 data. |
|
4.0
Findings
| 4.1 |
Descriptive
statistics
Summary
statistics of all the health outcomes used in this study
are shown in Tables 5-6 below.
| Table
5: |
Summary
statistics of daily hospital admissions (including
deaths) by disease groups (1995-97) |
| Admission |
No.
(days) |
Mean |
SD |
Min |
P25 |
Median |
P75 |
Max |
| Respiratory |
1096 |
203.7 |
38.49 |
116 |
175 |
197 |
227 |
335 |
| Asthma |
1096 |
24.7 |
8.59 |
7 |
19 |
23 |
29 |
63 |
| COPD |
1096 |
48.7 |
12.27 |
22 |
40 |
48 |
56 |
94 |
| Cardiac |
1096 |
98.7 |
23.28 |
40 |
80 |
101 |
114 |
179 |
| IHD |
1096 |
35.0 |
10.26 |
8 |
28 |
35 |
44 |
76 |
| Death: |
|
Cardiovascular
|
1096 |
10.2 |
3.77 |
1 |
8 |
10 |
13 |
24 |
|
Respiratory
|
1096 |
10.8 |
4.06 |
1 |
8 |
10 |
13 |
27 |
Min:
Minimum; P25: 25% centile; P75: 75% centile; Max: Maximum
Note
: COPD - Chronic obstructive pulmonary disease excluding
asthma
IHD
- Ischaemic heart disease
| Table
6: |
Summary
statistics of daily hospital admissions (including
deaths) by disease groups (1998 Jan-June) |
| Admission |
No.
(days) |
Mean |
SD |
Min |
P25 |
Median |
P75 |
Max |
| Respiratory |
181 |
259.0 |
43.39 |
174 |
227 |
259 |
285 |
389 |
| Asthma |
181 |
23.7 |
7.2 |
10 |
19 |
22 |
28 |
48 |
| COPD |
181 |
54.1 |
11.86 |
32 |
45 |
53 |
62 |
90 |
| Cardiac |
181 |
104.1 |
22.27 |
48 |
88 |
106 |
118 |
166 |
| IHD |
181 |
38.3 |
10.52 |
15 |
31 |
39 |
45 |
65 |
| Death: |
< | |