Am. J. Respir. Crit. Care Med.,
Volume 158, Number 1, July 1998, 341-342
A REANALYSIS OF THE OZONE/
ASTHMA RELATIONSHIP
To the Editor:
Thurston and colleagues (1) conclude that there are monotonic relationships between ozone (O3) and (1) reduced lung
function, (2) increased number of asthma symptoms, and (3)
increased use of asthma medication in asthmatic children at
concentrations well below the 1-h standard of 120 ppb. The
primary basis for these conclusions is the data plotted in Figure 1 of their paper. Since it appeared to me that the linear
lines derived from their regression analyses were heavily influenced by a small number of data points at the highest concentrations, I decided to test the robustness of the regressions by
removing some of the higher values.
I reconstructed their data set by interpolating the data contained in Figure 1 of their paper. To see how close the reconstructed data was to their data, I tried to reproduce their regression lines and correlation coefficients. Although, I could
not reproduce them exactly, I believe they are close enough
for the purposes of this exercise. The relationships I found
were: peak flow difference = 
0.071 O3 + 27.7 (r = 0.44);
medication use = 0.0017 O3 + 0.23 (r = 0.58); symptoms = 0.0029 O3 + 0.33 (r = 0.68). These compare with the relationships that they report: peak flow difference = 0.073 O3 + 27.5 (r = 0.43); medication use = 0.0018 O3 + 0.22 (r = 0.62); symptoms = 0.0029 O3 + 0.31 (r = 0.71).
When I removed the days with the two highest O3 concentrations (160 and 147 ppb), the relationship between peak flow
difference and O3 was not significant as the r decreased to
0.11 and the slope of the regression line decreased to 0.02
with a standard error of ± 0.05. For the other two health endpoints, it was necessary to delete the four highest O3 days before the relationships became nonsignificant. Incidentally, removal of the four highest days removes only the days which
experienced O3 greater than or equal to the present 1-h standard of 120 ppb. For the medication/O3 relationship, r decreased to 0.09 and the slope to 0.0003, with a standard error
of ± 0.0011. For the symptoms/O3 relationship, the r decreased
to 0.07 and the slope to 0.0002, with a standard error of ± 0.001.
Thus, when the days on which O3 was equal to or greater
than the current 1-h standard are eliminated from the analysis,
the remaining data do not support the conclusion of monotonic relationships. This is an extremely important issue because EPA extrapolates reported dose-response functions to
the entire U.S. population when preparing their risk assessments used to set National Ambient Air Quality Standards
(NAAQs). Whether monotonic relationships exist below the
current 1-h standard or whether there exists a threshold concentration has enormous policy implications in setting NAAQS.
The recent decision by the EPA Administrator to create a
more stringent O3 standard is predicated upon the existence of
such a monotonic relationship. Although the authors' conclusions are consistent with the Administrator's action, these
data do not support it.
George T.
Wolff
General Motors Public Policy Center
Detroit, Michigan
6.
Thurston, G. D.,
M. Lippman,
M. B. Scott, and
J. M. Fine.
1997.
Summertime
haze air pollution and children with asthma.
Am. J. Respir. Crit. Care
Med.
155:
654-660
[Abstract].
 |
From the Author : |
In our children's asthma camp study, we found significant associations between increased daily maximum ozone concentrations in ambient air in rural Connecticut and (1) increased
numbers of asthma attacks (as indicated in this study by increased doctor-diagnosed beta-agonist medication use; (2) decreased lung function; and (3) increased respiratory symptoms. The fact that Dr. Wolff is able to drive these associations
to statistical nonsignificance by removing up to 4 of the 15 study days is not surprising to those who have studied statistics. Reducing the number of observations in a data set increases the standard error of the estimate (in this case, of the
slope), which in turn reduces the statistical significance of the
relationship. Thus, Dr. Wolff's reanalysis "results" are more
likely due to a lack of power in the reduced data set analysis to
detect an effect, rather than to any purported lack of an effect
by ozone at levels below a 120 ppb 1-h maximum.
As Dr. Wolff is so interested in this question, I have taken
the summary data from this study (i.e., as daily averages of over 50 children on each study day) and applied a more robust data smoothing approach to search for any indication of a
threshold of effects (using the statistical package S-Plus [Seattle, WA]). I show the result of this effort for the case of the
mean number of beta-agonist medications, as prescribed for
the children each day by our on-site physicians, plotted versus
the daily ozone concentration (Figure 1). This plot suggests a
threshold at approximately 80 ppb daily 1-h maximum, which
would convert to a much lower 8-h average threshold value (at
roughly 60-70 ppb). Thus, if there is a threshold in these data,
it would appear from this smoothed plot to be well below the
old 120 ppb 1-h ozone standard. The smoothed plots of lung
function change and chest symptoms (not presented here)
similarly suggest thresholds, if any, occurring below a 120 ppb
1-h maximum value. Therefore, a more suitable examination
for thresholds in these data suggests that the U.S. EPA indeed
acted appropriately in tightening the ozone standard in 1997.
View larger version (21K):
[in this window]
[in a new window]
|
Figure 1.
A smoothed plot of physician-prescribed daily beta-
agonist medications required by children at a summer "Asthma
Camp."
|
|
An analysis of a much larger data set, having a greater statistical power, is really needed to more definitively answer Dr. Wolff's threshold question, especially if his approach of selective data deletion is to be employed.
George D.
Thurston
NYU School of Medicine
Nelson Institute of Environmental MedicineTuxedo, New York
