Table 4 also shows the results for these same models
when the data are limited to 1992 and 1993. This may
represent the “cleanest” data set because the estimated
PM10 data are more complete than the measured PM10
data, and the mortality data do not include the anomalies
that appeared in 1994 and 1995. The PM10 coefficients
for the three-day lag and the five-day moving average
for all four mortality end points are comparable or
somewhat stronger than those obtained with the full four
years of data using either measured or estimated PM10.
This is strong support for the conclusion that the problems
with the daily mortality data in 1994 and 1995 did
not result in a false finding of a statistically significant
PM10 effect on mortality.
Additional sensitivity analyses indicated that the
PM10 results are robust to functional form, extreme values,
lag structure, model specification, and corrections
for overdispersion. Table 5 provides a context for the
disease-specific results found using the Bangkok data
versus results from other cities. It indicates that the estimated
RR of mortality (the percent increase relative to
baseline rates of mortality) per unit of PM10 found in
Bangkok is similar to that found in other cities.
DISCUSSION
The results of the analysis of four years of data from Bangkok
indicate a statistically significant association between daily
PM10 and daily mortality. The effects are generally similar
to or greater than those found elsewhere: a 10-mg/m3 change
in PM10 is associated with a 1–2% change in total mortality,
a 3–6% change in respiratory-related mortality, and a
1–2% change in cardiovascular-related mortality. In addition,
the PM10 effects appear across all age groups, including
children, although they are statistically strongest for
older adults. For all of the age-specific cohorts, the magnitude
and statistical significance of the estimated effect of
PM10 increases when multiple-day measures (i.e., five-day
moving average) of exposure are used.
Most previous studies of air pollution and mortality
have been conducted in cities in the United States and
Western Europe—areas with relatively cold winters and
Table 3.