University of Southern California Dissertations and Theses

Data analysis on human exposure to concentrated particulate sulfur

(USC Thesis Other)

Data analysis on human exposure to concentrated particulate sulfur

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content DATA ANALYSIS ON HUMAN EXPOSURE
TO CONCENTRATED PARTICULATE SULFUR
Copyright 2004
by
Ling Yao
A Thesis Presented to the
FAC U LTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN C ALIFO R N IA
In Partial Fulfillm ent o f the
Requirements for the Degree
MASTER OF SCIENCE
(BIOSTATISTICS)
August 2004
Ling Yao
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
UMI Number: 1422423
INFORMATION TO USERS
The quality of this reproduction is dependent upon the quality of the copy
submitted. Broken or indistinct print, colored or poor quality illustrations and
photographs, print bleed-through, substandard margins, and improper
alignment can adversely affect reproduction.
In the unlikely event that the author did not send a complete manuscript
and there are missing pages, these will be noted. Also, if unauthorized
copyright material had to be removed, a note will indicate the deletion.
®
UMI
UMI Microform 1422423
Copyright 2004 by ProQuest Information and Learning Company.
All rights reserved. This microform edition is protected against
unauthorized copying under Title 17, United States Code.
ProQuest Information and Learning Company
300 North Zeeb Road
P.O. Box 1346
Ann Arbor, Ml 48106-1346
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
ACKNOWLEDGEMENTS
M y greatest thanks are directed to my committee chair, D r. Daniel O. Stram,
for his b rillia n t guidance, patience, and commitment o f time. I w ould also like to
extend m y gratitude to the committee members, Dr. Rob M cConnell for his support
and guidance in epidemiological studies and w riting, and Dr. K iros Berhane for his
guidance in statistical concepts. The wisdom and generous help o f a ll three
committee members made this project a rewarding experience. And I would also like
to thank my husband, lie Zhang, for his endless love and support.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
TABLE OF CONTENTS
Acknowledgements ii
List o f Tables iv
List o f Figures v
Abstract vii
Introduction 1
Materials and Methods 2
Results 18
Discussion 33
References 37
iii
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
LIST OF TABLES
Table 1. P-values o f treatment effects for all outcomes 20
Table 2. P-values o f sulfur modification 23
Table 3. P-values o f ambient sulfur effects on the 18 outcomes 26
Table 4. The distribution o f CAP and CAP + NO 2 exposure by seasons 33
iv
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 1.
Figure 2-1.
Figure 2-2.
Figure 3.
Figure 4.
Figure 5.
Figure 6.
Figure 7-1.
Figure 7-2.
Figure 8-1.
Figure 8-2.
Figure 9.
LIST OF FIGURES
D istribution o f A2i and A22 using artificial numbers 6
Hypothetical outcome time patterns at different sulfur 9
concentration o f particle showing time*treatment*chemistry
effects
Hypothetical outcome time patterns at different sulfur 10
concentration o f particle showing time*treatment* chemistry
effects
D istribution o f A2 at different sulfur concentration o f particle 1 1
(hypothetical data)
Possible linear relationship between A2 and sulfur concentration 13
o f particle (hypothetical data)
Hypothetical relationship between Aj, o and Aj, s 15
D istribution o f A2i and A22 o f hemoglobin for COPD group 21
(p-values are from Mest)
Linear relationship between differences o f differences from 24
9 *
baseline to 4-hour after exposure (A i) o f eosinophils percentage
and sulfur concentration o f particle in the COPD group
Linear relationship between difference o f differences from 25
9 ♦ a
baseline to the second day after exposure (A 2) o f eosinophils
percentage and sulfur concentration o f particle in the COPD
group
Linear relationship between Ay, o o f absolute lymphocyte 27
count and Ay, s in the non-COPD group (p = 0.0053).
Showing all three time points: Am , o, A y o, and A;2, o
Linear relationship between Ay, o o f absolute lymphocyte count 28
and Ay, s in COPD group (p = 0.6343). Showing a ll three tim e
points: A y o, A y o , and Ai2,o
Linear relationship between A y, q o f blood fibrinogen and Scap 30
in COPD group
v
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 10-1. Sulfur distribution on CAP days
FigurelO-2. Sulfur distribution on CAP + NO 2 days
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
ABSTRACT
We analyzed data from a randomized experiment in which 48 subjects w ith/w ithout
COPD were exposed to controlled ambient particle pollution. Subjects were exposed
to concentrated ambient particles and filtered air w ith/w ithout NO 2 on different days
and outcomes were measured at three time-points. Our prim ary interests are
m odification o f treatment effects by sulfur concentration o f particles and ambient
sulfur effect. We used descriptive graphs based on differences and mixed effects
models to describe all the data. O nly a few significant treatment effects were found.
For five outcomes we found significant sulfur m odification. Significant ambient
sulfur effects were found on many outcomes in non-COPD group, but there also
exists season confounding. Due to lim itation o f the experiment design and poor
power, we cannot draw reliable conclusion on sulfur m odification and ambient sulfur
effects. But our analyses showed a hint o f possible relationships between exposures
and outcomes.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
INTRODUCTION
Many epidemiological studies have showed that ambient particulate air
pollution is associated w ith daily m ortality, especially w ith respiratory and
cardiovascular disease related m ortality (1-4). In recent years, there have been a few
experimental studies exposing human volunteers to controlled or monitored air
particles and other pollutants (5,6). It is s till unclear which components in the
ambient air have the adverse effects and what are the mechanisms.
Gong et al conducted a randomized crossover experiment on human subjects
to evaluate acute effects o f controlled ambient fine particle pollution exposure (7).
Thirteen elderly volunteer w ith chronic obstructive pulmonary disease (COPD) and 6
age-matched healthy adults were exposed to controlled ambient PM pollution and
their short-term responses were evaluated. A Harvard particle concentrator and a
whole-body chamber were used to expose each subject to concentrated ambient
particles (CAP) or filtered air (FA) on different days separated by at least 14 days
and in a random order. Each exposure lasted 2 hours w ith interm ittent m ild exercise
and a standard sequence o f tests was performed 1.5 hour before the exposure, 4
hours after the exposure, and next morning after the exposure. An exposure*time
interaction term was used to access a response to CAP for variables w ith m ultiple
measurements. Group*exposure or group* exposure*time interaction term was used
to access a difference in response to CAP between healthy and COPD subjects.
Significant overall differences o f FEV, symptom scores, sputum cell counts, etc.
were found between COPD and healthy groups. No overall significant acute effect o f
1
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
CAP exposure was found in this study. But the authors did find suggestive acute
effect o f CAP exposure on S a C > 2, circulating erythrocytes, heart rate variability, and
ectopic heartbeats, which is consistent w ith other previous laboratory and
epidemiologic findings.
Using an expanded data set from the above cited study and sim ilar studies
cited below, our analysis focuses on possible exposure effect m odification by
particle chemistries, i.e. to see i f particle chemistries m odify CAP treatment effects.
We also compared the baseline rate levels o f a whole series o f outcomes (described
in the previous study by Gong et al.) to examine whether there is any relationship
between particle chemistries and the baseline measurements. This is a complex
crossover-designed study w ith m ultiple measurements for each outcome. One o f our
goals o f this analysis is to present data from this complex experiment in ways that
both reflect the design and which are relatively easy to understand.
M ATERIALS AN D METHODS
1. M ATERIALS
We used data from the previously cited study (7); an earlier study applying
the same experimental protocol to younger adult subjects w ith and w ithout asthma
(8); and a later study (s till in progress) o f healthy elderly and COPD subjects
exposed to CAP and FA w ith and without added nitrogen dioxide (NO 2). The target
exposure concentrations were 200 micrograms per cubic meter for CAP and 400
parts per b illio n for NO2, simulating worst-case acute outdoor pollution exposures.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Although the concentration o f CAP was controlled, the chemical composition o f
particles was uncontrolled and highly variable. In particular, concentration o f sulfate
varied more or less independently o f total mass concentrations. In all, there were 18
older adult subjects w ith COPD, 6 healthy older adult subjects w ithout COPD, 12
healthy younger adult subjects, and 12 younger adult subjects w ith asthma. A ll
subjects were exposed to CAP and FA in randomized order. The healthy-elderly and
most o f the COPD subjects were also exposed to CAP + NO 2 and to FA + NO 2. their
order o f exposures was randomized completely w ith respect to CAP, but not w ith
respect to NO 2 : for many o f them, exposures w ith NO 2 occurred several months after
exposures w ithout NO 2. Thus, seasonal differences had to be considered in the data
analysis. We combined all the available data for our analysis. Here again our primary
interest is on the effect o f particulate chemistries as a modifier o f any treatment
effects, and not specifically on the main effects o f either NO 2 or CAP.
There were 18 outcomes measured on all exposure days: systolic and
diastolic blood pressure, blood hemoglobin, absolute eosinophils count and
percentage, absolute lymphocytes count and percentage, absolute monocytes count
and percentage, absolute neutrifils and percentage, BAS, forced expired volume in
one second (FEV), forced vital capacity (FVC), hemocrit, heart rate, red blood cell
count, and white blood cell count. Seven outcomes were measured on CAP and FA
days but not on CAP + NO2 days: blood fibrinogen, symptom scores (cardiac,
miscellaneous subtotals, respiratory, and total), RPP, and vonW illebrand factor (VW
factor).
3
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2. STATISTICAL ANALYSES
2.1. TREATMENT EFFECT
O f the study risk factors that m ight affect individual response, COPD was
considered most important, because o f its expected greater effects on reserve
capability o f the respiratory system and on defense mechanisms against inhaled
pollutants. Accordingly, all analyses were stratified by COPD status. In the previous
study, the follow ing model was fitted to evaluate the treatment (CAP) effects:
Yijk = a; + Tj + Tj*Ak + £ jjk
where Y is one o f the outcomes; i, j, k denote subject, outcome measure time (j = 0,
1, 2. i.e. before treatments, 4 hours after treatments, and the next morning after
treatments), and treatment (CAP or FA), respectively; ctj is a random subject specific
intercept; and sp denotes the random error. Note that our interest is not the main
effect for either time T or treatment A. We expect in fact that there w ill be no effect
o f treatment at tim e zero because that is prior to the beginning o f treatment. Sim ilarly
the main effect o f time is not important either. We expect there to be natural diurnal
changes in outcomes and / or temporal responses to the protocol o f confinement and
m ild exercise that is independent o f exposure. We seek to determine i f these
temporal patterns are m odified by exposure for any o f the outcomes. Hence, the
prim ary interest is in the interaction o f time and CAP exposure (Tj*Ak), which
indicates the tim e trend o f response to the CAP exposure.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
In a simpler way, the relationship between time point and outcome can be
explained as a series o f differences o f Y p . Use the outcome difference between time
point zero and tim e point one as an example:
A iik= Y iik - Y;ok
captures the difference between pre-exposure and first post-exposure measurements
for a single subject. And the difference o f differences:
2
A ji" A iic — A iif
captures the difference between pre-exposure and first post-exposure measurements
on the CAP exposure day compared to the difference between pre-exposure and first
post-exposure measurements on the control (FA) day. We are interested in whether
this A2 n has a mean o f zero. A2 1 w ith a mean o f zero indicates that there is no
difference in the tim e pattern between CAP and FA exposures, i.e. there is no
treatment effect at time point one.
For a study w ith one pre-exposure measurement and only one post-exposure
measurement per subject, a paired /-test o f mean (A2 u) = 0 would be a very
reasonable and generally quite efficient analysis. In this study, we have more than
one post-exposure measurement for each individual and each exposure. Because
there is correlation between differences on the same individual, basing the analysis
on separate /-tests fo r the different A2 is not an optimal analysis. However, a graph o f
these differences is a relatively simple way to display the data and help us to
t 2
understand the possible effects o f exposures. Figure 1 shows the distributions o f A i
and A22 using hypothetical numbers, where
5
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 1: distribution o f A2i and A2 2 using hypothetical numbers.
10 D
50
0
50
D O
d e 11 o 2 deltal
d e l t a
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
A2, = Ajic - A jif = (Yao - Yioc) - ( Y iif- Y m )
and
A2 2 = Aj2c - Aj2f = (Yi2c - YiOc) - (Y i2f - YiOf).
This method is straight forward in displaying the data. It helps us to see i f
there is any hint that the difference o f difference has a mean o f zero, i.e. i f there is
any treatment effect at time o f either 4 hours after exposure or the next day. In order
to fu lly analyze the data, we fit a random effects model which has a random effect
for an individual. Throughout this paper we describe both graphical analyses based
on A and A2 as w ell as the random effects models used to describe all the data.
2.2. SULFUR M O D IFIC ATIO N
In this paper, we are prim arily focused on whether there exist effects due to
particle chemistries, which is even more complicated. The follow ing model is fitted
to evaluate i f the treatment effect is m odified by one o f the particle chemistries,
sulfur concentration o f particles.
Yjjjc = dj + S k + Tj* S k + Tj*Ak*Sk + £ij_k
This model can be used to evaluate three different questions:
(1) Whether even before the controlled exposure, the ambient sulfur
concentration o f particles would have an effect on the outcome.
(2) Whether the time effect is m odified by the sulfur concentration o f
particles.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
(3) Whether the treatment*time interaction is m odified by sulfur
exposures.
The interaction o f Tj*Ak*Sk would indicate the sulfur m odification, where Sk
denotes the sulfur concentration o f particles measurement during the exposure. Since
all sulfur measurements on FA days were zero, we cannot estimate Tj*Ak*Sk based
on the available data. In stead we used the follow ing model:
Yjj_k - cii + Tj* Ak + Tj*Ak*Sk + £y_k
to estimate sulfur modification. The analyses were done by two steps:
(1) analyze data from CAP and FA days only;
(2) fo r the 18 outcomes measured on CAP + NO 2 days, analyze data from all
CAP, CAP + NO2, and FA days, adjusted for season.
I f sulfur did m odify the treatment effects, the time trend o f outcome would be
different according to different sulfur concentration o f particles. Figure 2-1 and
figure 2-2 show possible hypothetical time patterns at different sulfur concentration
o f particles.
Sim ilarly, we can also use the distribution o f A2 to illustrate the sulfur
'y
modification. Figure-3 shows that A has different means at different sulfur
concentration o f particles.
Figures 2-1, 2-2, and 3 showed above use sulfur as a categorical variable. In
our data, sulfur concentration o f particles was measured as a continuous variable.
Again, i f we only have one post-exposure measurement, we could fit a simple linear
model to find out i f the differences o f differences are different according to the sulfur
8
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
O u tc o m e
Figure 2-1: Hypothetical outcome time patterns at different sulfur concentration of
particle showing time*treatment*chemistry effects.
S*Time*Exposure
500
400
300
200
100
3 2 0
— Subj 1 FA
-•-S u b ) 1 CAP S-'Low”
-a -S u b j 1 CAP S - ’High”
Tim e
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Outcom e
Figure 2-2: Hypothetical outcome time patterns at different sulfur concentration of
particle showing time*treatment*chemistry effects.
S*Time*Exposure
600
500
400
300
200
100
2.5 2 1.5 0.5 0
— Subj 1 FA
-* -S u b j1 CAPS-'Low" |
Subj 1 CAP S="High"
Time
10
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure-3: Distribution o f A2 at different sulfur concentration of particle (hypothetical
data).
75
5D
25
D
5 D
75
0
sulfur
11
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
concentration o f particles. Due to repeated measurements, we cannot use the simple
linear model to evaluate the relationship. But it w ill certainly help us to display and
understand the data and the true relationships. Figure 4 shows a possible linear
relationship between A2 and sulfur concentration o f particles.
2.3. AM B IE N T SULFUR EFFECT
Because the subjects in the experiments were also exposed to the ambient air
pollution, it is possible that particle chemistry has effects on the outcomes even
before experimental exposure begins. The potential effects o f ambient exposure may
last during and after the experimental exposure. Therefore, we fit the follow ing
model to evaluate the potential ambient sulfur effect, i f we have outcome and
ambient sulfur measurement for all exposure days then at tim e point zero we have:
Yi_0_k tti Sk + £ > i_ 0 _ k
where Y;_o_k indicates the outcome measurement before the beginning o f the
experimental exposure.
During the experiments, sulfur concentration o f particles was measured
inside the chamber. For CAP and CAP + NO 2 days, the sulfur concentration o f
particles inside the chamber can reasonably represent the ambient sulfur
concentration o f particles. However, sulfur concentration o f particles inside the
chamber was zero on FA days, which means we do not have inform ation o f ambient
sulfur concentration on FA days. As long as outcomes were measured on both CAP
12
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 4: Possible linear relationship between A2 and sulfur concentration of particle
(hypothetical data).
5 r
0
4 0
3D
20
1 0
D
1 0
2 0
3D
40
50
6 0 50 40 3D 2 0 I 0
sulfur
13
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
and CAP + NO 2 days, we can sim ply drop the FA day. Considering only
measurements at baseline, to describe the data we can compute Aj, o and Ai, s and fit:
Aij o — Aj, s S i_ 0 _ k
where Ai, 0 — Y ; _ o_c a p - Yi_o_cA P +N 0 2 , and A,i s - Sj_cAP - S i_cA P +N 0 2 - Figure 5
shows a possible hypothetical relationship between Aj, q and Aj, s -
To utilize all the inform ation on CAP and CAP + NO 2 days we actually fit a
mixed effects model:
Yy_k = a* + S k + Sk*Tj + Ak*Tj*Sk + sy_k
where k = CAP, CAP + NO 2.
MISSING D A TA
A t the time o f this paper we do not yet have the data fo r some outcomes on
the CAP + NO 2 days but only data on CAP and FA days. For these outcomes we can
compute:
Ai C A P , O = Y ij C A P - Yj_j_FA.
But we cannot compute
Ai_cAP, s = Si C A P - S i_F A because 8]_ fa was missing.
We now consider a partial analysis o f these data restricting ourselves to the
CAP and FA days where all outcomes were measured. Note that on the CAP days
when the S;_ cap is high we would expect that Sscap - Si F A is also high and when the
S ;_ cap is low that S^cap - Si_ fa is also low. Therefore we consider a model in which
the outcome differences at baseline are explained by the sulfur concentration o f
14
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 5: Hypothetical relationship between A * , q and A j, s.
30
20
0
5D 30 2 0
d e 1 1 a _ s u I f u r
15
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
particles on the CAP day and compare the parameter between the two models. Using
Si C A P as a substitute for Aj_C A p , s = Si_C A p - Sjjfa- i.e. we fit the follow ing model
Ai_cAP,o - P’ Scap + a ’ .............................. (1 )
In place o f model (2):
Aj_cAP, o = P (S cap - S fa ) + £ ..................... (2 ).
We now consider the relationship between the regression parameter P’ in
model (1) and P in model (2). Since all subjects received CAP or FA exposure in a
random order, we have: Var(ScAp) = Var(SFA)- Hence, from model (2):
P = Cov[Ai_cAP, o , (S cap - S fa )] / Var(ScAP - S fa )
- 2Cov(Aj_cAP, 0 ; S cap) / 2Var(ScAp)
- Cov(Ai_cAP, O) S cap) / V ar(S cA p)= P’
So that the regression parameters in model (1) and (2) are the same. Now
consider the variance o f the estimates.
From model (2) we have:
A i cA P , o = P(Scap - S fa ) + £ • By adding and subtracting P E (S fa ) we get
Aj_cAP, o = P (Scap - (S fa - E (S fa )) - E (S fa )) + £
= PScap - P E (S fa ) - P(SF A - E (S f a ) ) + s
Here we can estimate E (S fa ) = E (S c a p ), but we do not measure S fa - This
means that the variance term when fittin g model (2) w ill be equal to p2 Var(ScAp) +
Var(s).
W hile p in models (1) and (2) are the same, a pA 2 estimate from fittin g (2) is
less variable than p i from fittin g (1), in particular s’ has variance
16
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
p2 Var(ScAp) + a2 , where < r2 = Var(e).
This indicates that pA j has a larger variance than P
Var(pA 2 ) V a r(e )/V a r(S cA r-S F A)
Var(pA i) Var(8’)/V a r(S C Ap)
o2 / 2 Var(ScAp)
[p 2 V a r(S CAp) + a 2]/V a r(S c A p )
a2
2[p2 Var(ScAp) + o2 ]
This equals to 1/2 when the true value o f P is zero and is always less than 1/2
when | P | ^ 0. Thus using model (1) rather than model (2) gives us some o f the
inform ation we would like to have but never more than 1/2 o f that is missing.
To utilize outcome data at all time points, we can also consider using mean
Scap as a substitute o f the missing S fa - Starting w ith the follow ing model:
Yjjk — p . + C t j + PSk + E jjk
When S k was not measured, i.e. on the FA days we can fit:
Yjjk = p + o s + pE(Sk) + (PSk - pE(Sk)) + B ijk
Here E(Sk) is the mean o f the values o f sulfur. Again because o f the
randomization, we can assume that E(Sk) = E(S) for all Ij_k and we estimate E(S)
from the CAP days, when Sulfur concentration o f particles was measured. So from
model
Yijk = p + PE(S) + O j + P (S k - E(S)) + S jjk ,
17
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
the term PE(S) modifies the mean p, while P(Sk - E(S)) inflates the variance
o f random intercept a;. Replacing the intercept we have
Yyk = p + PE(S) + a{ + 8ijk,
where V a r(a i*) = Var(a;) + p2 Var(Sk). So using an indicator o f exposure days,
we fit the follow ing model:
Yjjk = p + P (I(k=l)*E (S ) + I(k=2)*Sk)+ otj* + sp.
SEASON EFFECTS AND CONFOUNDING
In this study, CAP and CAP + NO2 exposures are not randomly assigned to
all subjects. Since sulfur concentration o f particles may vary in different seasons,
and weather-related factors may also have effects on human’s cardiopulmonary
system, we adjusted the above model for season (denoted by W ijk):
Y p = c ti + S k + Wijk + Sk*Tj + Ak*Tj*Sk + £ijk
where k = CAP, CAP + NO2. According to the date o f experiments, we
defined season as: Spring = (March, A p ril, May); Summer = (June, July, August);
Fall = (September, October, November); W inter = (December, January, February).
RESULTS
1. TREATMENT EFFECT
We evaluated the treatment effects on 25 outcomes. As reported previously,
few effects o f treatment were detected. Normal force vita l capacity (FVC) was the
only outcome that showed statistically significant response to CAP exposure (p =
18
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
0.0454) among COPD subjects. In the non COPD group, forced expired volume in
one second (FEV, p = 0.0023) and white blood cell count (W BC, p = 0.0414)
showed statistically significant responses to CAP exposure. Some other outcomes
showed m arginally significant responses (0.05 < p < 0.10), including hemoglobin,
FEV, and red blood cell count in the COPD group, and neutrophils and FVC in non-
COPD group. Table 1 shows p-values o f treatment effects for all outcomes.
Since we have 2 post-exposure measurements fo r each subject and each
outcome, the distribution o f
A 2 ] = A j i c - A m = ( Y j i c - Y i o c ) - ( Y i i f - Y m )
and
A 2 2 = A j 2 c - A i 2 f = ( Y i 2 c - Y o c ) - (Y m - Y j o f )
w ill help us to understand the direction o f response changes. Figure 6 shows A2i and
A22 o f hemoglobin for COPD group, where the p-value fo r tim e by treatment
interaction was 0.0754. We can see that compared to FA exposure, hemoglobin after
CAP exposure increased at tim e point one (4 hours after exposure) and then
decreased at tim e point two (the next morning).
2. SULFUR M O D IFIC ATIO N
Using data from CAP and FA days only, we found that among COPD
subjects, the treatment effects on respiratory symptom scores, eosinophils
percentage, and red cell counts were significantly m odified by sulfur concentration
o f particles. In non-COPD group, only diastolic blood pressure showed significant
19
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 1: p-values o f treatment effects for all outcomes.
outcome Treatment Effects (Time*Exposure)
COPD non-COPD
Systolic blood pressure 0.9192 0.8078
Blood fibrinogen 0.4771 0.2549
Hemoglobin 0.0754 0.3477
RPP 0.4008 0.6508
Cardio symptom score 0.9932 0.1767
Misc symptom score 0.2477 0.4237
Respiratory symptom score 0.9157 0.3812
Total symptom score 0.9602 0.397
Absolute Eosinophils 0.8594 0.2007
Absolute Lympocytes 0.6206 0.1888
Absolute Monocytes 0.7153 0.2355
Absolute Neutrifils 0.2196 0.0716
BAS 0.4045 0.9749
Diastolic blood pressure 0.8981 0.3196
Eosinophils percentage 0.9969 0.6457
FEV 0.0539 0.0023
FVC 0.0454 0.074
Hemocrit 0.3213 0.1024
Heart rate 0.5673 0.2384
Lyphocytes percentage 0.9425 0.2595
Monocyptes percentage 0.2616 0.9224
Neutrifils percentage 0.7581 0.4733
Red blood cell count 0.0506 0.3971
VW factor 0.5696 0.4806
White blood cell count 0.1612 0.0414
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 6: D istribution o f A2i and A22 o f hemoglobin for COPD group (p-values are
from t-test).
For A2i (delta=l): mean = 0.08, SD = 0.676, p = 0.1714;
For A22 (delta = 2): mean = -0.225, SD = 0.684, p < 0.0001
2
1
0
1
■ 2
2 1
d
e
1
t
a
s
q
r
-2
time
21
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
m odification by sulfur concentration o f treatment effects. Using a ll available data
and adjusting for season for the 18 outcomes measured on CAP + NO 2 days, we
found two additional outcomes, FEV and hemocrit, had significant modification, by
sulfur concentration o f treatment effects. Table 2 presents p-values o f all sulfur
modification.
Again, using difference o f the outcome changes between CAP and FA
exposures (A2 ) can help us to understand the direction o f the relation o f treatment
effect and sulfur m odification. Figure 7-1 and figure 7-2 show that A2i and A22 o f
eosinophils percentage in the COPD group decreased as sulfur concentration o f
particles level increased, which indicates that A2] and A22 are d iffe r by sulfur level.
3. AM BIEN T SULFUR EFFECT
From the mixed effects model fo r ambient sulfur effect fo r 18 outcomes w ith
both CAP and CAP + NO 2 exposures, sulfur concentration o f particles level showed
significant effects on 10 outcomes in the non-COPD group. Sulfur concentration o f
particles showed no significant effect on any o f the 18 outcomes among COPD
subjects. Table 3 shows the p-values o f sulfur effects on the 18 outcomes.
We plot the differences between outcome measurements on CAP and CAP +
NO2 days at all three tim e points (Ajj, q = Yjj_cAP - Yij_ c a p + n o 2s j — 0, 1,2) against the
differences between sulfur measurements on the two days (Ay, § - Sj cap -
S i_ cA P + N 0 2)- Figure 8-1 and figure 8-2 show the plots o f absolute lymphocyte count
for non-COPD and COPD subjects, respectively. We can see that Ajj, 0 increases w ith
22
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 2: p-values of sulfur modification.
outcome Sulfur Modification
CAP vs. FA CAP, CAP+N02, and FA*
COPD non-COPD COPD non-COPD
Systolic blood pressure 0.1378 0.9685 0.0807 0.2098
Blood fibrinogen 0.108 0.7852
Hemoglobin 0.1996 0.7856 0.0917 5354
RPR 0.0962 0.577
Cardio symptom score 0.8811 0.962
Misc symptom score 0.5194 0.1185
Respiratory symptom score 0.0193 0.8116
Total symptom score 0.1904 0.3239
Absolute Eosinophils 0.1327 0.3345 0.2706 0.4075
Absolute Lympocytes 0.1607 0.559 0.318 0.1077
Absolute Monocytes 0.839 0.7419 0.2606 0.7104
Absolute Neutrifils 0.0503 0.8966 0.2214 0.9514
BAS 0.9103 0.777 0.3939 0.4687
Diastolic blood pressure 0.7181 0.0001 0.3723 <0.0001
Eosinophils percentage 0.0358 0.1715 0.0257 0.0923
FEV 0.6273 0.0878 0.463 0.0219
FVC 0.8892 0.2122 0.9074 0.1206
Hemocrit 0.155 0.2832 0.1402 0.0201
Heart rate 0.8367 0.1287 0.7191 0.2896
Lyphocytes percentage 0.0837 0.8374 0.2565 0.6554
Monocyptes percentage 0.4282 0.4842 0.4061 0.3245
Neutrifils percentage 0.0685 0.8603 0.0997 0.6516
Red blood cell count 0.0326 0.3508 0.0392 0.1031
VW factor 0.1309 0.9668
White blood cell count 0.13 0.9271 0.3892 0.9606
* Adjusted for season
23
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 7-1: Linear relationship between difference o f differences from baseline to 4-
hour after exposure (A2 i)* o f eosinophils percentage and sulfur concentration o f
particle** in the COPD group.
Deltal_sqr
2
1
0
1
2
3
80 60 70 20 30 40 50 0 10
Sulfur
* A2! = (Y i l c - Y i 0 c ) - ( Y i l f - Y i o f ) ;
* * U n it o f S u l f u r : mg/m 3 .
24
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 7-2: Linear relationship between difference o f differences from baseline to the
second day after exposure (A2 2)* o f eosinophils percentage and sulfur concentration
o f particle** in the COPD group.
delta2_sqr
2
1
0
-1
-2
-3
-4
0 10 20 30 40 50 60 70 80
Sulfur
* A2 2 = ( Y i 2 c - Y i 0 c ) - ( Y i 2 f - Y i 0f ) ;
* * U n it o f S u l f u r : mg/m3 .
25
* * * *
r r m rrrn 11 r" |'i 111111 nyi 1111 n 111 m ii 111 i 111
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 3: P-values o f ambient sulfur effects on the 18 outcomes.
outcome COPD
Sulfur Effects* Sulfur Effects** Season Effects
Systolic blood pressure 0.1723 0.4542 0.2571
Hemoglobin 0.8716 0.5466 0.0018
Absolute Eosinophils 0.6599 0.7126 1
Absolute Lympocytes 0.7119 0.658 0.3609
Absolute Monocytes 0.8797 0.3683 0.0015
Absolute Neutrifils 0.8738 0.4143 0.1132
BAS 0.3279 0.1787 0.3551
Diastolic blood pressure 0.4671 0.5674 0.0964
Eosinophils percentage 0.7319 0.4347 0.8253
FEV 0.7859 0.3352 0.4603
FVC 0.2444 0.5887 0.6408
Hemocrit 0.9677 0.4766 0.0011
Heart rate 0.2183 0.3348 0.9499
Lyphocytes percentage 0.6343 0.9308 0.3698
Monocyptes percentage 0.7029 0.0401 0.0002
Neutrifils percentage 0.6123 0.1999 0.2838
Red blood cell count 0.7899 0.2515 0.0002
White blood cell count 0.7683 0.4911 0.1546
outcome non-COPD
Sulfur Effects* Sulfur Effects** Season Effects
Systolic blood pressure 0.061 0.5069 0.0256
Hemoglobin 0.4061 0.2742 0.1154
Absolute Eosinophils <0.0001 0.089 <0.0001
Absolute Lympocytes 0.7169 0.8623 0.9678
Absolute Monocytes 0.6308 0.1227 0.0937
Absolute Neutrifils 0.0057 0.0898 0.0012
BAS 0.4372 0.4848 0.893
Diastolic blood pressure 0.0086 0.2517 0.2262
Eosinophils percentage 0.0161 0.7701 0.0536
FEV <0.0001 0.5428 0.0007
FVC <0.0001 0.4604 <0.0001
Hemocrit 0.7321 0.4517 0.0007
Heart rate 0.9367 0.0517 0.0007
Lyphocytes percentage 0.0053 0.0335 0.0008
Monocyptes percentage 0.0043 0.0014 0.0401
Neutrifils percentage 0.0041 0.01 0.0017
Red blood cell count 0.649 0.0943 0.0012
White blood cell count 0.029 0.2921 0.0146
*sulfur effects not adjusted for season
**sulfur effects adjusted for season
26
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 8-1: Linear relationship between Ay, o o f absolute lymphocyte count and Ay, s
in the non-COPD group (p = 0.0053). Showing all three tim e points: Ajo, o, Aii, o, and
A i 2 , o -
delta Y
1 0
0
-10
-20
-30
-40
-3 0 -20 -10 0 10 20
d elta S
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 8-2: Linear relationship between Ay, 0 o f absolute lymphocyte count and Ay, s
in COPD group (p = 0.6343). Showing all three time points: A;o, o, Aji, o, and Ai2, o-
d elta Y
- 1 0 -
- 11:
- 12 :
30 20 10 0 -3 0 -20 -10 -5 0 -40
d elta S
28
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
the increasing o f Ay, s, which indicates that ambient sulfur concentration o f particles
level increase the absolute lymphocyte count, especially fo r non-COPD subjects.
For the seven outcomes not measured for the day o f CAP + NO 2 exposure,
using model Y p - p + P(I(k=T)*E(8) + I(k=2)*Sk)+ a;* + ep, we found that blood
fibrinogen (p = 0.0059), miscellaneous symptom scores (p = 0.0208), respiratory
symptom scores (p = 0.0428), and total symptom scores (p = 0.0183) in COPD group
significantly responded to sulfur concentration o f particles level. W hile in the non-
COPD group, total symptom scores had a significant response (p = 0.0456) and
respiratory symptom scores had m arginally significant response (p = 0.0944) to
sulfur concentration o f particles.
The plot in figure 9 shows the relationship o f Ay, o and sulfur concentration
o f particles on CAP days among COPD subjects, where Ay, o is the difference o f
blood fibrinogen between CAP and FA days at different tim e points.
4. SEASON EFFECTS AND CONFOUNDING
From figure 10-1 and 10-2 we can see that the distribution o f sulfur
concentration o f particles vary by seasons. Sulfur concentration is relatively high in
summer and low in winter. Due to the experiment design, the assignment o f CAP or
CAP + NO 2 exposure was not random and more subjects had their CAP + NO 2
exposure during winter. Table 4 shows the distribution o f the exposure assignments
by seasons.
29
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 9: Linear relationship between Ajj, o o f blood fibrinogen and Scap in COPD
group.
delta Y
500 -
400
300
200
100
-100
-200
-300
-400
-500
50 40 30 20
su lfu r c
30
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 10-1: Sulfur distribution on CAP days.
80
60
S
C 40
A
P
20
0
31
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Winter
i ------------------------1
Spring Summer
S E A S O N C
Fall
Figure 10-2: Sulfur distribution on CAP + NO2 days.
80 -
0 -
- - - - - - - - - - - - - - - - !- - - - - - - - - - - - - - - - - - - - - - - 1 - - - - - - - - - - - - - - - - - - - - - - - - 1 - - - - - - - - - - - - - - - - - - - - - - - - r -
Winter Spring Summer Fall
S E A S O N B
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 4: The distribution o f CAP, FA, and CAP + NO2 exposure by seasons.
Exposure Season
Winter Spring Summer Fall
CAP 14 15 1 1 8
FA 12 19 6 1 1
CAP + N02 29 5 6 7
After adjusting for season, one outcome (monocytes percentage) among
COPD subjects showed significant response to sulfur concentration o f particles,
w hile non-COPD subjects only have three outcomes (lymphocyte percentage,
monocytes percentage, and neutrifils percentage) significantly respond to sulfur
concentration o f particles. Meanwhile, season have significant effects on most (12
out o f 18) outcomes in non-COPD group, and showed significant effects on 5
outcomes in COPD group. The p-values were showed in table 3.
DISCUSSION
1. TREATMENT EFFECTS AN D SULFUR M O D IFIC ATIO N
The results for treatment effects are sim ilar to the conclusion that Gong et al.
derived from the same study w ith smaller sample size. Although one outcome in the
COPD group and two outcomes in the non-COPD group had statistically significant
responses to CAP exposure, we cannot exclude the possibility that these significant
results are random because tests for m ultiple outcomes w ill certainly increase the
probability o f type I error.
33
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
However, we found many significant results when we allowed the sulfur
chemistries to m odify the treatment effect. Because the order o f FA and CAP
exposures were randomized, season is not an important confounder in the analyses
using data from CAP and FA days only. To increase the power, we used the data
from all three exposures. In this case, we needed to adjust fo r season, because CAP +
NO 2 exposures were not randomized and most o f them were performed during
winter (see table 4). I f sulfur m odification did exist, we may need to reconsider the
results for treatment effect. However, a larger sample size is needed to detect a
treatment effect when the sulfur m odification is involved.
Both treatment effects and sulfur modifications are indicated by interaction
terms. In this case, each interaction term w ill have more than one estimated
parameters due to m ulti-com bination o f different strata. This makes it d iffic u lt to
interpret the directions o f the treatment effects and sulfur m odifications based on the
estimated parameters, especially when both positive and negative estimated
parameter exist. As described in the ‘statistical analysis’ section, differences o f
difference and their plots can help us to understand the directions o f changes better.
A box-plot o f A2! = Aiic - A if = (Y ;ic - Y i0 c ) - (Y ilf - Y i0 f) and A2 2 = Ai2 c - Ai2f = (Yi2 c
- Yioc) - (Y i2f - Y j0 f) can help to show whether the difference o f difference fo r a
certain outcome has a mean equal to, less than, or greater than zero. Sim ilarly, a plot
o f this difference o f difference against sulfur concentration o f particles w ill show us
the direction o f sulfur modification. Because o f repeated measurements fo r each
outcome, we used the mixed effects model to test i f there exist significant treatment
34
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
effects and / or sulfur m odification. But using the plots described above can clearly
show the directions o f changes. Even when there is poor power to detect the
significant effect due to small sample size, the plot may s till show us a hint o f such
an effect.
2. AM B IE N T SULFUR EFFECTS
To evaluate the ambient sulfur effects, we need sulfur concentration o f
particles data on at least two exposure days. We do not have data fo r ambient sulfur
concentration o f particles level on FA days for all outcomes. For the 18 outcomes
that sulfur concentration o f particles was measured on both CAP and CAP + NO 2
days we can sim ply drop FA days and perform the analysis. For 7 outcomes we only
have sulfur concentration o f particles data on CAP days, we used partial analyses
described in the method section to estimate the ambient sulfur effects using the
available CAP-day sulfur data. W hile using expectation o f sulfur concentration o f
particles (estimated by the mean o f sulfur on CAP days) as a substitute, we found 4
in 7 outcomes among COPD subjects had significant responses to the sulfur
concentration o f particles and one significant effect in the non-COPD group. We also
tried to use the expectation sulfur approach to estimate the 18 outcomes w ith sulfur
measurements on both CAP and CAP + NO 2 days (ignoring the measurement on
CAP + NO 2 days) and found only 5 outcomes w ith significant response to ambient
sulfur concentration o f particles. The results indicate that although missing values
w ill certainly decrease the power to detect possible effects, we can s till estimate the
35
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
effects using available data. Under the condition o f randomization, we can use the
expectation value (estimated by the mean o f available values) as a substitute o f the
missing values is a better method compared to directly using the available values a
substitute.
We expected that COPD subjects were more susceptible to pollutants
including sulfur. But sulfur concentration o f particles showed no significant effects
o f 18 outcomes among COPD subjects, w hile in non-COPD group, 10 o f the 18
outcomes had significant responses to ambient sulfur concentration o f particles.
Because the exposure assignments were not totally random, and there was a
correlation between season and sulfur concentration, an alternative explanation could
be an effect o f season. The results show that season is a confounder for sulfur effects.
Especially when the exposures were not evenly distributed among different seasons,
we need to adjust for season.
36
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
REFERENCES
5. Brauer M , Ebelt ST, et al. Exposure o f chronic obstructive pulmonary disease
patients to particles: respiratory and cardiovascular health effects. Journal o f
exposure analysis and environmental epidemiology. 11(6): 490-500, 2001 Nov-
Dee.
2. Bremner SA, Anderson HR, et al. Short term association between outdoor air
pollution and m ortality in London 1992-4. Occupational & Environmental
Medicine 1999; 56(4): 237-44.
6. Ghio AJ, K im C, D evlin RB. Concentrated ambient air particles induce m ild
pulmonary inflam m ation in healthy human volunteers. Am J Respir C rit Care
Med. 162: 981-8, 2000.
7. Gong H, Linn WS, et al. Exposures o f elderly volunteers w ith and without
chronic obstructive pulmonary disease (COPD) to concentrated ambient fine
particle pollution. In press for Inhalation Toxicology 2004
8. Gong H, Linn WS, et al. Controlled exposures of healthy and asthmatic
volunteers to concentrated ambient fine particles in Los Angeles. Inhalation
Toxicology 15: 305-25.
9. Gong H, Linn WS, et al. Separate and combined effects of ambient fine
particulate pollution and nitrogen dioxide in elderly volunteers (abstract).
Am. J. Respir. Crit. Care Med. 167: A499 (2003).
1. Katsouyanni K , Touloumi G, et al. Short term effects o f ambient sulphur
dioxide and particulate matter on m ortality in 12 European cities: results from
time series data from the APHEA project. B ritish M edical Journal 1997; 314:
1658-63.
3. Samet JM, D om inici F, et al. Fine particulate air pollution and m ortality in 20
U.S. cities, 1987-1994. N Engl J Med 2000; 343(24): 1742-9.
4. Van der Zee SC, Hoek G, et al. Acute effects o f air pollution on respiratory
health o f 50-70 yr old adults. European Respiratory Journal 2000; 15: 700-9.
37
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

A descriptive analysis of medication use by asthmatics in the Children's Health Study, 1993

PDF

Family history, hormone replacement therapy and breast cancer risk on Hispanic and non-Hispanic women, The New Mexico Women's Health Study

PDF

Influence of body weight and weight change on carotid IMT and IMT progression: A pooled analysis using subject-level data from four clinical trials

PDF

Evidence for type 2 diabetes related quantitative trait locus on chromosome 6q: Joint analysis of Finnish and Mexican American families

PDF

Identifiability of constrained generalized linear Age-Period-Cohort models: Criteria for choosing equality constraints

PDF

Gene mapping using haplotype data

PDF

Candidate gene association analysis by principal components

PDF

beta3-adrenergic receptor gene Trp64Arg polymorphism and obesity-related characteristics among African American women with breast cancer: An analysis of USC HEAL Study

PDF

Effects of myeloperoxidase (MPO) polymorphism and tobacco smoke on asthma and wheezing in Southern California children

PDF

Dietary fiber intake and atherosclerosis progression: The Los Angeles Atherosclerosis Study

PDF

Comparisons of metabolic factors among gestational diabetes mellitus probands, siblings and cousins

PDF

Determinants of mammographic density in African-American, non-Hispanic white and Hispanic white women before and after the diagnosis with breast cancer

PDF

A pilot survey of medical abortion knowledge and practices among obstetrician/gynecologists and family practitioners in Los Angeles County

PDF

Identifying susceptibility genes for complex diseases by accounting for epistasis in studies of candidate genes

PDF

Extent, prevalence and progression of coronary calcium in four ethnic groups

PDF

BRCA1 mutations and polymorphisms in African American women with a family history of breast cancer identified through high throughput sequencing

PDF

A linear model for measurement errors in oligonucleotide microarray experiment

PDF

Development and evaluation of standardized stroke outcome measures in a population of stroke patients in rural China

PDF

Association between body mass and benign prostatic hyperplasia in Hispanics: Role of steroid 5-alpha reductase type 2 (SRD5A2) gene

PDF

Evaluation of the accuracy and reliability of self-reported breast, cervical, and ovarian cancer incidence in a large population-based cohort of native California twins

Asset Metadata

Creator Yao, Ling (author)

Core Title Data analysis on human exposure to concentrated particulate sulfur

School Graduate School

Degree Master of Science

Degree Program Biostatistics

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag biology, biostatistics,health sciences, public health,OAI-PMH Harvest

Language English

Contributor Digitized by ProQuest (provenance)

Advisor Stram, Daniel O. (committee chair), Berhane, Kiros (committee member), McConnell, Rob (committee member)

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c16-319441

Unique identifier UC11328987

Identifier 1422423.pdf (filename),usctheses-c16-319441 (legacy record id)

Legacy Identifier 1422423.pdf

Dmrecord 319441

Document Type Thesis

Rights Yao, Ling

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the au...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus, Los Angeles, California 90089, USA