University of Southern California Dissertations and Theses

Spatial analysis of PM₂.₅ air pollution in association with hospital admissions in California

(USC Thesis Other)

Spatial analysis of PM₂.₅ air pollution in association with hospital admissions in California

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content SPATIAL ANALYSIS OF PM
2.5
AND HOSPITAL ADMISSIONS

1
Spatial Analysis of PM
2.5
air pollution
in association with hospital admissions in California

by
Mei Yu Yeh

A Thesis Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
MASTER OF SCIENCE
(APPLIED OF BIOSTATISTICS AND EPIDEMIOLOGY)

May 2018

Copyright 2018 Mei Yu Yeh
SPATIAL ANALYSIS OF PM
2.5
AND HOSPITAL ADMISSIONS

2

Table of Contents
Dedication…………………………………………………………………………………..……..3
Acknowledgements…………………………………………………………………..………....…4
List of Tables/ Figures………………………………………………………………………….…5
Abstract……………………………………………………………………………………….…...7
Introduction………………………………………………………………………………….…….8
Methods……………………………………………………………………………………….…...9
Results…………………………………………………………………………………….……...15
Discussion……………………………………………………………………………….……….30
References……………………………………………………………………..…...………...…..33
SPATIAL ANALYSIS OF PM
2.5
AND HOSPITAL ADMISSIONS

3
Dedication
This thesis is dedicated to my parents and my brother who always support me and
encourage me to do the things I want to do. Thanks for them always be with me when I face
challenges of my whole master life. Because of them, I got the strength to complete the thesis in
time and chase my dreams.

SPATIAL ANALYSIS OF PM
2.5
AND HOSPITAL ADMISSIONS

4
Acknowledgements
I would like to convey my deepest appreciation to my master thesis committee: Dr.
Meredith Franklin, Dr. Steven Yong Cen, and Dr. Rima Habre. At First, I appreciate the chair of
thesis committee Dr. Meredith Franklin, who always endless guiding and helping me with
patience and passion when I face the problem or struggling for the analytic session.
In addition, I would like to thank for other committees. Dr. Steven Yong Cen, who
graciously provided the NIS data and give me his suggestions and support. And I am also
thankful to Dr. Rima Habre, who are willing to share her professional advice and guidance.
It is pleasure to thank once again for those who made this thesis possible.

SPATIAL ANALYSIS OF PM
2.5
AND HOSPITAL ADMISSIONS

5
List of Tables/ Figures
1. Figure 1. Map showing the number of PM
2.5
monitors; PM
2.5
chemical speciation
monitors; and hospital locations………………………………………………………....10
2. Figure 2. Semivariogram interpretation and schematic…………………………………12
3. Figure 3. Map showing mean PM
2.5
concentrations in California from 2010-2016 at each
of the 134 EPA monitoring sites…………………………………………………………16
4. Figure 4. Mean PM
2.5
concentrations 2010-2016 at EPA AQS sites……………………17
5. Figure 5. Empirical semivariograms of PM
2.5
concentrations……………………….18-19
6. Table 1. Spatial parameters of seasonal data in Northern and Southern California from
Weighted Least Squares (WLS) and Maximum Likelihood (ML) models……………...20
7. Figure 6. Plot showing the fitted theoretical semivariance functions for seasonal data in
Northern and Southern California………………………………………………………..21
8. Figure 7. The seasonal predicted PM
2.5
concentrations over California based on universal
kriging...........................................................................................................................22-23
9. Table 2. Estimated LOOCV MSE and R
2
………………………………………………24
10. Figure 8. Cross validation results observed versus predicted values from the test
locations.…………………………………………………………………………………25
11. Table 3. Northern and Southern California specific mean and standard deviation (SD) for
the analyzed causes of hospital admissions during the year 2010-2011, and corresponding
averaged PM
2.5
over the warm and cool season.………………………………………....26
12. Table 4. Poisson regression results for CVD and Respiratory hospital admissions…….27
13. Figure 9. Monthly time series of CVD hospital admissions…………………………….28
14. Figure 10. Monthly time series of respiratory hospital admissions……………………..28
SPATIAL ANALYSIS OF PM
2.5
AND HOSPITAL ADMISSIONS

6
15. Figure 11. Biplots of first two principal components of PM
2.5
chemical speciation
concentrations in Bakersfield and Riverside, CA showing loading on dust source and
secondary formation/regional transport………………………………………………….29
16. Table 5. Regional and seasonal difference in PM
2.5
composition……………….………30

SPATIAL ANALYSIS OF PM
2.5
AND HOSPITAL ADMISSIONS

7
Abstract
Background: Many studies have shown that exposure to particulate matter air pollution with
aerodynamic diameter less than 2.5µm (PM
2.5
) is detrimental to health. We focus on examining
the spatial distribution of PM
2.5
concentrations in California, and its relationship with cause-
specific hospital admissions.
Methods: Spatial statistical methods were used to examine the spatial distribution of PM
2.5
measured at EPA Air Quality System fixed-site monitors in California from 2010 to 2016.
Regionally stratified seasonal kriging models were developed to spatially interpolate PM
2.5
concentrations to the zip code level where hospital admissions from the National Inpatient
Sample were recorded. Using Poisson time series regression, the associations between monthly
PM
2.5
and cause-specific hospital admissions including cardiovascular disease and respiratory
disease were assessed.
Results: We found that a 10µg/m
3
increase in monthly PM
2.5
exposure

was associated with a 1.48%
increase in cardiovascular disease hospital admissions (p<0.0001, 95% CI= (1.33, 1.64))
,
and a

0.24% increase in respiratory disease hospital admissions (p=0.001, 95% CI= (0.098, 0.386))
.

Conclusion: Observed positive associations between PM
2.5
and hospital admissions in California
suggest that exposure to PM
2.5
air pollution remains a serious threat to health.
Key words: particulate matter, PM
2.5
, spatial, kriging, LOOCV, hospital admissions

SPATIAL ANALYSIS OF PM
2.5
AND HOSPITAL ADMISSIONS

8
Introduction
Despite the air getting cleaner under strict regulations set by the U.S. Environmental
Protection Agency (EPA), air pollution is still a serious issue.
1
The American Lung
Association’s annual “state of the air” in 2017 stated that California continues to dominate the 6
of the top 10 most air polluted cities in U.S.
2
One study stated that California has different
ambient particle chemistry, temporal patterns, and size distribution than other places in the U.S.
3

Southern California in particular has some of the worst air quality in the U.S.
4
Wind speed is
lower around Los Angeles than other urban cities but also sits in a basin and tends to have strong
atmospheric inversions, resulting in lower rates of atmospheric mixing particularly at nighttime
and early morning.
5
What is more, weather conditions include long hours of sunshine particularly
in summer that commonly leads to exceedances of the federal ozone standard (due to
photochemistry), yet high levels of fine particles and dust tend to peak in late autumn and
winter.
6

In comparison to larger particles
,
PM
2.5
(particles with aerodynamic diameter less than 2.5
µm), has been shown to lead to greater health problems since they can be inhaled and reach deep
into the alveolar region of the lungs and may also get into the bloodstream.
7
That is, the smaller
the particles, the greater potential toxicity leading to serious health problems.
8,9
Short-term
exposure to PM
2.5
has been associated with morbidity and mortality.
8,10,11,12,13,14,15,16
It is also
suspected that certain chemical components of PM
2.5
are more deleterious to health than others.
8
PM
2.5
is ubiquitous in our environment, and has been associated with cardiovascular, respiratory,
and other serious diseases.
8,7,17
Previous studies have shown specifically that daily fluctuations
in PM
2.5
can lead to increased hospital admission for cardiovascular and respiratory
diseases.
12,18,19,20

SPATIAL ANALYSIS OF PM
2.5
AND HOSPITAL ADMISSIONS

9

In this study, we focused on assessing the association between PM
2.5
concentrations and
cause-specific hospital admissions in California, obtained from the Healthcare Utilization Project
(HCUP) National Inpatient Sample (NIS). Seasonally and regionally stratified geostatistical
models were developed and applied to monthly PM
2.5
concentrations from 2010 to 2014
measured at EPA Air Quality System monitoring stations to derive exposure surfaces over
California. These models were then used to estimate PM
2.5
exposures at the zip code level, which
were matched with monthly hospital admissions data to explore the cause-specific associations
between PM
2.5
and outcomes including cardiovascular diseases (CVD) and respiratory diseases.
Characteristics of PM
2.5
chemical compositions in Northern and Southern California were
examined as effect modifiers of the PM
2.5
and hospital admission associations.
Methods
Exposure Estimation
Air Pollution Data. In this study, daily PM
2.5
mass and chemical speciation data were
acquired for California from the EPA Air Quality System (AQS) (aqs.epa.gov/aqsweb/airdata/).
In the downloaded data, variables including location information (latitude, longitude, state code,
county code), concentration, units (e.g. µg/m
3
for PM
2.5
), Parameter Occurrence Codes (POC),
daily AQI value, daily observation count, percent complete, and AQS parameter code. We
restricted PM
2.5
concentrations to be from filter-based 24-hour integrated samplers following the
Federal Reference Method (FRM). We examined a subset of approximately 60 chemical
components from Chemical Speciation Network (CSN) PM
2.5
filters (out of 101 components in
total) including metals (e.g. Fe, Pb, Ni, Al), carbon (e.g. elemental EC and organic OC) and ions
(e.g. sulfate SO
4
2-
, nitrate NO
3
-
). While PM
2.5
mass was measured on a near daily the CSN
SPATIAL ANALYSIS OF PM
2.5
AND HOSPITAL ADMISSIONS

10
sampling schedule is on a 1-in-3 or 1-in-6 day schedule. In addition to a less frequent sampling
schedule, there are far fewer CSN monitors (N=23) than PM
2.5
monitors (N=134) in California.
(Figure 1)

Figure 1. Map showing the number of PM
2.5
monitors (black points); PM
2.5
chemical speciation
monitors (red stars); and hospital locations (blue triangles)

As such, spatial statistical methods, which enable the interpolation of point referenced
data to generate a surface over a geographic area, are not reliably applied to the CSN data given
their sparse locations. We therefore focused on applying spatial statistical methods to PM
2.5
SPATIAL ANALYSIS OF PM
2.5
AND HOSPITAL ADMISSIONS

11
concentrations, generating exposure estimates at the zip code level for health effects assessments,
and then supplemented these exposure estimates with a post-hoc analysis of chemical
characteristics of PM
2.5
using the prominent CSN sites in Northern (Bakersfield) and Southern
(Riverside) California. We conducted a principal component analysis (PCA) of the chemical
speciation data from these two sites to identify and compare major sources in the regions.
21
We
also examined the relative contributions of the major tracer species from the identified source
components to the seasonal PM
2.5
mass.
Spatial Statistical Modeling. Due to differences in PM
2.5
physical and chemical
characteristics and temporal patterns, the study domain was divided into Northern and Southern
California, with the boundary defined by the latitude of Santa Barbara (34.4 decimal degrees).
For spatial model building, the daily data were aggregated into warm and cool seasonal averages
with warm defined by April through September and cool by October through March. This
stratification was taken to improve the condition of spatial stationary for model building. Figure
3 shows maps of the regions and PM
2.5
monitors with seasonally averaged concentrations.
Latitude and longitude were projected to UTM, with zone 10N for Northern CA and zone 11N
for Southern CA.
We applied geostatistical methods to the regional and seasonally stratified PM
2.5

concentrations including semivariograms and interpolation via kriging. The semivariance γ(h) is
commonly estimated using the binned semivariogram defined by:

where the distance lag (bin) is h, the number of pairs for a particular distance lag is N(h) and
Z(s
i
)-Z(s
j
) is the difference in PM
2.5
measured at locations s
i
and s
j
which are identified by the
projected x,y coordinates. The plot with distance lag (h) on the x-axis and γ(h), the semivariance,
SPATIAL ANALYSIS OF PM
2.5
AND HOSPITAL ADMISSIONS

12
on the y-axis is the semivariogram, which is used to explore the spatial variability between all
pairwise locations and the most appropriate theoretical semivariogram.
22,23
Using Weighted
Least Squares (WLS) and maximum likelihood (ML) methods, various theoretical
semivariograms, including exponential, Gaussian, and Matern functions, were fitted to the
empirical data. The spatial parameters that describe the theoretical semivariograms are the range
(𝜑), sill (𝜎
2
), and nugget (𝜏
2
). The range (𝜑) indicates the distance where the semivariance
reaches an asymptote; the sill (𝜎
2
) refers to the value of the y-axis reached at the range; the
nugget (𝜏
2
) represents the semivariance while the smallest distance (i.e. as hà0).
24
An
illustration of the empirical and theoretical semivariogram is shown in Figure 2 (left). The
corresponding spatial covariance functions of the best fitting theoretical models were used for
spatial interpolation (Exponential example shown in Figure 2, right).

Figure 2. Semivariogram interpretation and schematic (obtained from Journal of Engineering
[image] Available at: https://www.hindawi.com/journals/je/2013/960105/fig2/ [Accessed
3/28/2018].)

Example:
Theoretical Function Exponential

Semivariance:
g(h)=𝜏
2
+ 𝜎
2
(1-exp(-fh))

Corresponding covariance:
C(h)= 𝜏
2
+ 𝜎
2
exp(-fh)
SPATIAL ANALYSIS OF PM
2.5
AND HOSPITAL ADMISSIONS

13
Several spatial interpolation techniques were compared, including Simple, Ordinary, and
Universal Kriging. Kriging is the stage of classical geostatistical analysis that weights the
surrounding measured values to predict the Z(s) at unmeasured locations.
25
The general kriging
equation is:
25,26

where Z(s) is the predicted values at unobserved location s
0
; µ(s) is the nonrandom trend
function based on observed locations s=(x,y); and e(s) are the real-valued errors characterized by
a spatial variance-covariance function previously defined through the theoretical semivariogram
function. The different types of kriging are differentiated through µ(s), where simple kriging
assumes µ(s) is a fixed constant, ordinary kriging estimates µ(s) as the mean of the spatial
process, and universal kriging allows for µ(s) to represent a trend (or drift) in x and y that can be
either linear, quadratic, or some function of spatially varying covariates, as shown by Σβ
k
x
k
(s
i
) in
the above equation.
27
Cross-Validation. To determine the true prediction accuracy of the prediction models
that were used, leave-one-site-out cross validation (LOOCV) was applied. This involved taking
one PM
2.5
monitoring site out at a time, fitting the kriging model to the remaining sites (training
set), and interpolating (predicting) for the left out site (test set). The cross-validation R
2
was
computed from a linear regression of the observed and predicted values of the test set to quantify
how well the model did at predicting the PM
2.5
concentrations at the left out site
28
:

R
cv
2
=1−
+ ,
-
.,
-
/
+ ,
-
.,
/

Numerator: Model error

Denominator: Variance in the dependent variable
SPATIAL ANALYSIS OF PM
2.5
AND HOSPITAL ADMISSIONS

14
Where 𝑦
1
are the observed PM
2.5
values from left out site; 𝑦
1
are the predicted PM
2.5
values from
model at the left out site; and y is the mean of the observed PM
2.5
values from the left out site.
The MSE from the cross validation was also assessed to see if there was any systematic bias
between the observations and predictions. Good kriging models provided small MSE and high
R
cv
2
.
Health Outcome Data
The NIS (National (Nationwide) Inpatient Sample) are databases developed through the
HCUP (Healthcare Cost and Utilization Project).
28,29
The NIS consists of the largest publicly
available inpatient database in the U.S. Annually, HCUP collects inpatient data for more than
seven million hospital stays across the nation.
29,30
For each individual record in the NIS data,
information including hospital zip code, month and year of hospital admission, length of stay,
diagnosis codes, procedure codes, age, and race are provided. We defined diseases by their ICD-
9 codes and focused on disease outcomes that had previously been identified in studies of air
pollution health effects cardiovascular diseases (CVD, ICD-9 390-429) including congestive
heart failure (ICD-9 428), atrial fibrillation (ICD-9 427.31), and myocardial infarction (ICD-9
410), and respiratory diseases including acute respiratory disease (ICD-9 460-461), pneumonia
(ICD-9 480-486), cardiopulmonary disease (COPD, ICD-9 492-496) and other respiratory
diseases (ICD-9 510-519).
Epidemiological Analysis
Using the regionally stratified kriging models described above, monthly PM
2.5
concentrations were spatially interpolated to the centroid of each hospital zip code identified in
the NIS. Monthly counts of cause-specific hospital stays were calculated by hospital zip code to
which the interpolated PM
2.5
concentrations were spatially and temporally linked. Poisson time
SPATIAL ANALYSIS OF PM
2.5
AND HOSPITAL ADMISSIONS

15
series regression models in a generalized additive model (GAM) framework were used to assess
the association between cause-specific hospital stays and PM
2.5
concentrations
.
31, 32
The Poisson
regression models were defined as:
log[E(Y
i
)]=a+bX
i
+f(t)+ε
i

where Y
i
is the monthly count of the cardiovascular or respiratory outcome at hospital i, a is the
constant (intercept) term, X
i
is the monthly PM
2.5
concentration at hospital i with linear
parameter b, f(t) is a smooth function of time, in months, represented by a cubic regression
spline with 4 degrees of freedom per year, and ε
i
is the N(0, σ
2
) error (residual) term. Our
primary interest was regression coefficient (b) for PM
2.5
.
33

All spatial and epidemiological modeling were conducted using the R language (version
R 3.4.1).
Results
Air Pollution
There were 134 PM
2.5
EPA AQS monitoring sites across California that were in operation
between 2010-2016 (Figure 3). We found higher mean PM
2.5
concentrations in the central valley
region near Bakersfield as well as urban centers of Los Angeles county, and lower mean PM
2.5
concentrations in more rural areas of Northern California, particularly nearby Lake Tahoe.
Overall, the mean PM
2.5
concentrations were higher in Southern California, which has been
noted in many previous studies.
6

SPATIAL ANALYSIS OF PM
2.5
AND HOSPITAL ADMISSIONS

16

Figure 3. Map showing mean PM
2.5
concentrations in California from 2010-2016 at each of the
134 EPA monitoring sites

Stratifying by region (Northern and Southern California) and season (warm and cool), we
note the warm season in Northern California had lowest overall mean PM
2.5
concentrations
(Figure 4). There were higher mean PM
2.5
concentrations in the northeastern part of Northern
California along the mountain where population density is low and there are few industrial
sources. For the cool season in Northern California the mean PM
2.5
concentration was generally
lower at each site; however, there were some higher concentrations exceeding the National
Ambient Air Quality Standard near Bakersfield (16.55 µg/m
3
). In the warm season in Southern
California, there were some lower mean PM
2.5
concentration sites in the northeast, but overall,
mean PM
2.5
concentrations were higher. We note higher mean PM
2.5
concentration centered in
the urban areas of Los Angeles and San Diego (around 11.2 µg/m
3
in both warm and cool
seasons) (Figure 4).
SPATIAL ANALYSIS OF PM
2.5
AND HOSPITAL ADMISSIONS

17

Figure 4. Mean PM
2.5
concentrations 2010-2016 at EPA AQS sites: Northern California warm
season (top-left); Northern California cool season (top-right); Southern California warm season
(bottom-left); Southern California cool season (bottom-right)

Empirical semivariograms
Empirical semivariograms for seasonal PM
2.5
in each region of California are shown
using 20 distance bins in Figure 5. All semivariograms showed clear spatial patterns with
SPATIAL ANALYSIS OF PM
2.5
AND HOSPITAL ADMISSIONS

18
smaller semivariance at smaller distances, and larger semivariance at further distances. Northern
California was in general a larger geographic area, so the distances between AQS monitoring
sites was up to approximately 1,000 km. For better interpretation and comparability with
Southern California we limited all semivariograms to have a maximum distance of 350 km
(Figure 5).
In Southern California (Figure 5, bottom panel), we noted greater spatial heterogeneity
with the sill being reached by a range of 100 km. In Northern California, the sill was not attained
until a range of about 200 km. In both Northern and Southern California the semivariances are
much larger in the cool season, indicating greater heterogeneity in concentrations, even at short
distances. This seasonal distinction in spatial variability is likely a result of more secondary
PM
2.5
formation in warm, sunny, summer months leading to less variable concentrations than
seen in cooler months.

SPATIAL ANALYSIS OF PM
2.5
AND HOSPITAL ADMISSIONS

19

Figure 5. Empirical semivariograms of PM
2.5
concentrations: for warm season in Northern
California (top-left); for cool season in Northern California (top-right); for warm season in
Southern California (bottom-left); for cool season in Southern California (bottom-right).

Exponential, Gaussian, and Matern theoretical semivariogram models were fit to the data
by WLS and ML methods. All of fitted theoretical models are illustrated in Figure 6. It is
important to note that while the WLS models may appear to be a better fit, these models are fit to
the binned semivariograms and tend to be visually more appealing. Maximum likelihood models
are fit to all of the data (unbinned) so will not be influenced by the binning scheme used for
illustrative purposes. Spatial parameters including the nugget (𝜏
2
), sill (𝜎
2
), and range (𝜑) were
estimated and the models were assessed for goodness of fit by their AIC and SSE (Table 1).

SPATIAL ANALYSIS OF PM
2.5
AND HOSPITAL ADMISSIONS

20
Table 1. Spatial parameters of seasonal data in Northern and Southern California from Weighted
Least Squares (WLS) and Maximum Likelihood (ML) models.

Warm Northern Cool Northern Warm Southern Cool Southern
California California California California

WLS
Model Matern Gaussian Gaussian Gaussian
𝜏
2
1.90 5.86 1.78 5.33
𝜎
2
6.46 34.12 4.67 11.95
𝜑 385.24 70.46 36.03 44.63
SSE 145.05 99.34 40.43 38.06
MLE
Model Exponential Spherical Spherical Spherical
𝜏
2
0.73 3.90 0.21 0.83
𝜎
2
5.22 33.68 5.36 11.54
𝜑 10.53 111.6 35.35 18.33
AIC 379.6 513.7 221.7 259.2

The spatial parameters presented in Table 1 are the estimates from the models with
smallest SSE (WLS) and AIC (ML). In general, the parameter estimates from the WLS and ML
were similar, however the range parameter estimate from the Matern fit in the warm season in
Northern California was implausibly large (385.24 km), and the resulting SSE was much larger
(145.05) than the other models (all <100). Given the preference for ML estimation to all of the
data and its widespread use in statistical estimation, the best fitting models with smallest AIC
were chosen as: ML-Exponential for Northern California warm season, ML-Spherical for
Northern California cool season, and ML-Spherical for Southern California warm and cool
season.
SPATIAL ANALYSIS OF PM
2.5
AND HOSPITAL ADMISSIONS

21

Figure 6. Plot showing the fitted theoretical semivariance functions for seasonal data in Northern
and Southern California. PM
2.5
empirical binned variogram (circles) with exponential (red),
Gaussian (pink), Spherical (orange) and Matern (yellow) WLS fits, and exponential (green),
Gaussian (blue), Spherical (purple) and Matern (light blue) ML fits.

In both Northern and Southern California, we note larger estimated sills in the cool
season (33.68 and 11.54 µg
2
/m
6
, respectively units of variance), than the warm season (5.22 and
5.36 µg
2
/m
6
, respectively), which follows our observation from the visual inspection of the
empirical binned semivariograms in Figure 4. The spherical semivariogram provided the best
SPATIAL ANALYSIS OF PM
2.5
AND HOSPITAL ADMISSIONS

22
ML fit (smallest AIC) to the cool PM
2.5
data in both regions and the warm data in Southern
California. The exponential semivariogram provided the best fit in Northern California in the
warm season. In general, the models did not fit as well in Northern California as in Southern
California (AIC North 379.6-513.7, AIC South 221.7-259.2), likely due to the larger geographic
area and more variable concentrations (very low in less populated, rural areas, while high in the
Central Valley).
Kriging
Universal kriging was applied to interpolate PM
2.5
using the best fitting seasonal and
regional theoretical semivariogram models described above. We chose universal kriging with a
linear trend in x and y due to its smaller AIC as compared to simple and ordinary kriging. For
illustration, we predicted average seasonal PM
2.5
surfaces in each of the regions in a fine spatial
grid (Figure 7).

SPATIAL ANALYSIS OF PM
2.5
AND HOSPITAL ADMISSIONS

23

Figure 7. The seasonal predicted PM
2.5
concentrations over California based on universal kriging

Overall, we found that interpolated PM
2.5
concentrations over Southern California are
generally higher in urban areas and the spatial patterns were comparable between seasons. In
Northern California, we notice very distinct areas of high concentrations, particularly in the
Central Valley near Bakersfield, where in the cool season PM
2.5
concentrations exceeded the
annual National Ambient Air Quality Standard of 12 – 15 µg/m
3
. This is likely attributed to
nitrate photochemistry, whereby in the cool season in the Central Valley there is increased
secondary formation of ammonium nitrates in the cool months. Overall in the warm season the
concentrations are very low in Northern California.
Leave one Site Out Cross Validation
Iteratively leaving one entire site out at a time provided cross validation test sets for
which we were able to assess the kriging models (Table 2 and Figure 8). The resultant LOOCV
R
2
was highest for the cool season in Northern California (R
2
= 0.50), and although quite a bit
lower in the warm season (R
2
= 0.32), the MSE was very low (0.98) in the warm season. The
SPATIAL ANALYSIS OF PM
2.5
AND HOSPITAL ADMISSIONS

24
results for Southern California were not as promising, with LOOCV R
2
equal to 0.21 in the cool
and only 0.18 in the warm season (Table 2). One reason for these poor results could be that the
chosen regions were too large for kriging, and more localized models should have been fit, or
that there was significant spatio-temporal interaction that was not appropriately captured.
Kriging was applied to data from a six-year period over 2010-2016. Kriging model building on
finer spatial and temporal scales could have resulted in better model stability.
Table 2. Estimated LOOCV MSE and R
2

Warm Northern Cool Northern Warm Southern Cool Southern
MSE 0.98 3.29 2.77 3.05
R
2
0.32 0.50 0.18 0.21
SPATIAL ANALYSIS OF PM
2.5
AND HOSPITAL ADMISSIONS

25

Figure 8. Cross validation results observed versus predicted values from the test locations.

Epidemiological Results
In total, from 2010 to 2011 there were 204, 255 CVD (combined CHF, MI, AF) and
183,450 respiratory admissions (combined COPD, acute respiratory disease, pneumonia and
other respiratory) at 1,200 hospitals in California from the inpatient sample. Unfortunately, NIS
SPATIAL ANALYSIS OF PM
2.5
AND HOSPITAL ADMISSIONS

26
data after 2011 did not contain the hospital zip code, so these data could not be included in the
epidemiological analysis as we could not link PM
2.5
concentrations. A further consequence of
this data limitation is that we used combined categorization of cardiovascular and respiratory
disease because we did not have enough statistical power to examine specific subclasses of
disease. Characteristics of the study population, including means and standard deviations of
CVD and respiratory hospital stays stratified by region and season are shown in Table 3.
Table 3. Northern and Southern California specific mean and standard deviation (SD) for the
analyzed causes of hospital admissions during the year 2010-2011, and corresponding averaged
PM
2.5
over the warm and cool season.

CVD Respiratory* PM
2.5
Warm** PM
2.5
Cool**
Mean(SD) Mean(SD) Mean(SD) Mean(SD)

Northern CA 154.42 (123.75) 155.05 (117.99) 6.83 (4.15) 8.76 (8.55)
Southern CA 184.28 (119.54) 183.07 (117.78) 9.69(3.14) 9.51(4.58)
* Respiratory combined acute, pneumonia, COPD and other respiratory diseases.
** Monthly mean PM
2.5
for the warm season or cool season

There were on average 154 CVD and 155 respiratory admissions per hospital and month
in Northern California, and 184 CVD and 183 respiratory admissions per hospital and month in
Southern California.
Looking across warm and cool seasons on PM
2.5
monthly mean concentration, the warm
season (6.83 µg/m
3
) is lower than the warm season (8.76 µg/m
3
) in Northern California;
however, the warm season (9.69 µg/m
3
) is higher than the cool season (9.61 µg/m
3
) in Southern
California.
The Poisson regression results were expressed as percent increases in cause-specific
hospital admissions per 10 µg/m
3
increase in PM
2.5
concentrations (Table 4). We found a
statistically significant 1.48% increase in CVD hospital admissions (p<0.0001, 95%CI:1.33,
SPATIAL ANALYSIS OF PM
2.5
AND HOSPITAL ADMISSIONS

27
1.64%), and a statistically significant 0.24% increase in respiratory hospital admissions with
10µg/m
3
increase in PM
2.5
(p=0.001, 95%CI: 0.098, 0.386%).
Table 4. Poisson regression results for CVD and Respiratory hospital admissions

PM
2.5
coeff.* Z value Pr( >|z| ) 95% CI N obs in regression
CVD 1.48 18.6 <0.0001 (1.328, 1.640) 1211
North 4.93 12.4 <0.0001 (4.153, 5.713) 588
South 0.93 9.09 <0.0001 (0.719, 1.114) 623
Respiratory 0.24 3.29 0.001 (0.098, 0.386) 1239
North 6.58 16.9 <0.0001 (5.815, 7.344) 597
South 0.14 1.47 0.14 (-0.047, 0.332) 642

* Showing the percent increase in hospital admissions per 10 µg/m
3
increase in PM
2.5

Similar temporal trends were found in the hospital admissions for both CVD and respiratory
diseases (Figures 9 and 10), with peaks in March and lowest numbers of cases in August and
September.
We also examined whether region (North versus South) modified the association between
PM
2.5
and the outcomes. We found statistically significant interactions for both disease outcomes,
so we stratified the model. After stratification we noted much larger and statistically significant
effect estimates for Northern California than Southern California (Table 4). In Northern
California, there was a 6.58% increase in respiratory disease admissions (95% CI: 5.83, 7.34%),
and a 4.93% increase in CVD admissions (95% CI: 4.15, 5.71%) per 10 µg/m
3
increase in PM
2.5
.
SPATIAL ANALYSIS OF PM
2.5
AND HOSPITAL ADMISSIONS

28

Figure 9. Monthly time series of CVD hospital admissions

Figure 10. Monthly time series of respiratory hospital admissions

SPATIAL ANALYSIS OF PM
2.5
AND HOSPITAL ADMISSIONS

29
The larger estimates Northern California led us to believe that differences in particle
composition between regions were potentially driving the observed differences in health effects.
Our post-hoc analysis of the CSN concentrations from the Bakersfield site in Northern California
and Riverside site in Southern California identified some notable differences. The PCA
identified the first three source components common to Bakersfield and Riverside to be 1)
soil/dust (loading on Si, Al) 2) secondary formation/regional (loading on NO
3
-
, SO
4
2-
, NH
4
+
) and
3) industrial (loading on Fe, Ni, Cu), which explained over 50% of the total variability in PM
2.5

(Figure 11). While there are no striking differences in the major identified sources between
Bakersfield and Riverside, we note that there was a stronger signal of nitrate in Bakersfield in the
second source factor than Riverside, which had slightly more loading on sulfate.

Figure 11. Biplots of first two principal components of PM
2.5
chemical speciation concentrations
in Bakersfield (left) and Riverside (right), CA showing loading on dust source (yellow oval) and
secondary formation/regional transport (blue oval).

SPATIAL ANALYSIS OF PM
2.5
AND HOSPITAL ADMISSIONS

30
Furthermore, the analysis of nitrate, sulfate, Si and Nu as tracers of the three sources further
identified differences between the regions and indicated seasonal differences (Table 5). Nitrates
showed strong seasonal differences with relative concentrations (as % of PM
2.5
mass) higher in
cool (34%) than warm (15%) months in Northern California. While much less distinct, the
opposite trend was true in Southern California where the proportion of nitrate in PM
2.5
was
higher in warm (38%) than cool (27%). The most striking regional and seasonal distinction was
in the contribution of Si to the mass, which was nearly three times higher in Bakersfield in the
warm season (4%) versus the cool season (1.9%), and nearly four times higher than Riverside
(1.2%)
Table 5. Regional and seasonal difference in PM
2.5
composition

% PM
2.5
Mass
PM
2.5
species Northern California
(Bakersfield)
Southern California
(Riverside)

Cool Season
NO
3
-
34% 27%
SO
4
2-
8% 7%
Cu 0.002% 0.01%
Si 1.9% 1.4%
Warm Season
NO
3
-
15% 38%
SO
4
2-
16% 16%
Cu 0.01% 0.04%
Si 4% 1.2%

Discussion
Two large publicly available datasets were harnessed in this study: particulate matter air
pollution data from the EPA, and health data from the HCUP National Inpatient Survey. Spatial
statistical methods enabled the interpolation of PM
2.5
concentrations over California, which were
SPATIAL ANALYSIS OF PM
2.5
AND HOSPITAL ADMISSIONS

31
then linked to each of the hospital zip codes. With nearly 200,000 hospital admissions examined,
we were able to detect statistically significant associations between exposure to ambient PM
2.5
and cause-specific diseases. We found that increases in monthly PM
2.5
concentrations were
significantly associated with increases in both CVD and respiratory hospital admissions.
Our results are in line with previous studies showing the risk for hospitalization of
cardiovascular disease is positive association with PM
2.5
among those who aged equal and more
than 65 years old.
34,35
In addition, the relation between the PM
2.5
and the cardiorespiratory
hospital admission is positive among the univariate model, which is consistent to our results for
the respiratory hospital admission.
Although the finding is consistent with other studies, large-scale studies of this type of
the association between PM
2.5
and hospital admission are infrequent due to limitations in data
availability. Especially, NIS is a very large database, which requires sophisticated and careful
interpretation. The strength of NIS which recorded with each discharge for the patients count
and the volume of data; however, lack of information for knowing the different comorbidities
from complications since it only included the pre-discharge information.
36
The limitation of the
NIS is that it did not provide sufficient information to examine confounders such as family
history and alcohol intake.
37
Due to different studies have concerned different covariates in their
model, the result might be different level of degree for the association between the PM
2.5
and
cause-specific hospital admission. In future analyses, we plan to link the NIS data with zip code-
level census information in order to examine potential population-based confounding. Since we
took a Poisson regression time series approach, which has been used in many similar studies of
the short-term effect of air pollution on health,
38
the most important confounding factors to
control for are seasonal and other temporal trends. The distribution of underlying factors such as
SPATIAL ANALYSIS OF PM
2.5
AND HOSPITAL ADMISSIONS

32
age, sex and race will not change from month to month within a hospital, so they will likely not
confound the monthly fluctuations in the environmental variables examined. Nevertheless, since
we pooled many hospitals together in this study, differences across hospitals could lead to
heterogeneity that we did not properly account for in our analyses. Last but not least, we
potentially underestimated the complication rate because this database only provide inpatient
events (inpatient complications).
37
In the process of data analysis, we cannot be guaranteed to
eliminate all biases using the complex spatial statistical models. Despite considering many
factors during the model building for spatial interpolation there is still the possibility of exposure
measurement error in the estimated PM
2.5
exposures used in the epidemiological assessment.
Furthermore, our LOOCV results were not very strong, indicating that we may have introduced
some bias and error in our exposure estimation. In future analyses, we will consider smaller
regions for building the kriging models, which should improve our exposure estimates because
we will better capture location variations in PM
2.5
.
The findings in this study call for action given the observed statistically significant
associations between the cause-specific admission and PM
2.5
concentration. It is better to further
explored the potentially toxic chemical components of PM
2.5
(such as Ni, Cu, V, Fe, etc.) so that
we could clearly know what the specific components will result in the increasing of the cause-
specific hospital admission. Furthermore, with a longer time series of exposure and subsequently
larger number of hospital admissions (pre-dating 2010, for instance) we could examine more
refined cause-specific outcomes such as stroke, diabetes and myocardial infarction (MI). Based
on this, it could provide the suggestion for setting the PM
2.5
standard which establish enough
margin of safety to NAAQS as avoiding the higher risk of hospital admission.
SPATIAL ANALYSIS OF PM
2.5
AND HOSPITAL ADMISSIONS

33
References
1.Health and Environmental Effects of Particulate Matter (PM) | US EPA. (2017, June 26).
Retrieved from http://www.epa.gov/pm-pollution/health-and-environmental-effects-
particulate-matter-pm.
2.How healthy is the air you breathe? (n.d.). Retrieved from http://www.lung.org/our-
initiatives/healthy-air/sota/.
3.Blanchard, C. L., Tanenbaum, S., & Motallebi, N. (2011). Spatial and Temporal
Characterization of PM
2.5
Mass Concentrations in California, 1980–2007. Journal of the
Air & Waste Management Association, 61(3), 339-351. doi:10.3155/1047-3289.61.3.339.
4.Kim, E. Turkiewicz, K. Zulawnick, S.A., Magliano, K.L. (2010). Sources of fine particles
in the South Coast area, California. Atmospheric Environment, 44(26),3095-3100.
5.Hu, S., Fruin, S., Kozawa, K., Mara, S., Paulson, S. E., & Winer, A. M. (2009). A wide
area of air pollutant impact downwind of a freeway during pre-sunrise
hours. Atmospheric Environment, 43(16), 2541-2549.
6.Air Quality | Region 9: Southern California | US EPA. (n.d.). Retrieved from
https://archive.epa.gov/socal/web/html/index-4.html.
7.Ostro, B., Feng, W., Broadwin, R., Green, S., & Lipsett, M. (2006). The Effects of
Components of Fine Particulate Air Pollution on Mortality in California: Results from
CALFINE. Environmental Health Perspectives, 115(1), 13-19. doi:10.1289/ehp.9281.
8.Zanobetti, A., Franklin, M., Koutrakis, P., & Schwartz, J. (2009). Fine particulate air
pollution and its components in association with cause-specific emergency
admissions. Environmental Health, 8(1). doi:10.1186/1476-069x-8-58.
9.NRC (National Research Council). (2004). Research Priorities for Airborne Particulate
SPATIAL ANALYSIS OF PM
2.5
AND HOSPITAL ADMISSIONS

34
Matter. doi:10.17226/10957.
10.Atkinson, R., Ross Anderson, H., Sunyer, J., Ayres, J., Baccini, M.,
Vonk, J.,Katsouyanni, K, et al. (2001). Acute Effects of Particulate Air Pollution on
Respiratory Admissions. American Journal of Respiratory and Critical Care
Medicine, 164(10), 1860-1866. doi:10.1164/ajrccm.164.10.2010138.
11.Le Tertre, A. (2002). Short-term effects of particulate air pollution on cardiovascular
diseases in eight European cities. Journal of Epidemiology & Community Health, 56(10),
773-779. doi:10.1136/jech.56.10.773.
12.Burnett, R. T., Smith-doiron, M., Stieb, D., Cakmak, S., & Brook, J. R. (1999). Effects of
Particulate and Gaseous Air Pollution on Cardiorespiratory Hospitalizations. Archives
of Environmental Health: An International Journal, 54(2), 130-139.
doi:10.1080/00039899909602248.
13.Schwartz, J. (1998). Air Pollution and Hospital Admissions for Heart Disease in Eight US
Counties. Epidemiology, 9(Supplement), S76. doi:10.1097/00001648-199807001-
00222.
14.Schwartz, J., & Morris, R. (1995). Air Pollution and Hospital Admissions for
Cardiovascular Disease in Detroit, Michigan. American Journal of
Epidemiology, 142(1), 23-35. doi:10.1093/oxfordjournals.aje.a117541.
15.Sunyer, J., Antó, J. M., Murillo, C., & Saez, M. (1991). Effects of Urban Air Pollution on
Emergency Room Admissions for Chronic Obstructive Pulmonary Disease. American
Journal of Epidemiology, 134(3), 277-286. doi:10.1093/oxfordjournals.aje.a116081.
16.Kloog, I., Ridgway, B., Koutrakis, P., Coull, B. A., & Schwartz, J. D. (2013). Long- and
SPATIAL ANALYSIS OF PM
2.5
AND HOSPITAL ADMISSIONS

35
Short-Term Exposure to PM
2.5
and Mortality. Epidemiology, 24(4), 555-561.
doi:10.1097/ede.0b013e318294beaa.
17.Laden, F., Neas, L. M., Dockery, D. W., & Schwartz, J. (2000). Association of Fine
Particulate Matter from Different Sources with Daily Mortality in Six U.S.
Cities. Environmental Health Perspectives, 108(10), 941-947. doi:10.1289/ehp.00108941.
18.Fine Particulate Air Pollution and Mortality in 20 U.S. Cities. (2001). New England
Journal of Medicine, 344(16), 1253-1254. doi:10.1056/nejm200104193441614.
19.Zmirou, D., Schwartz, J., Saez, M., Zanobett, A., Wojtyniak, B., Touloumi, G., …
Katsouyanni, K. (1998). Time-Series Analysis of Air Pollution and Cause Specific
Mortality. Epidemiology, 9(5), 495-503. doi:10.1097/00001648-199809000-00005.
20.Short-term Effect of Ambient Particles on Mortality: A Combined Analysis in 30
European Cities Within The APHEA Project. (2000). Epidemiology, 11(4), S106.
doi:10.1097/00001648-200007000-00226.
21.Hasheminassab, S., Daher, N., Ostro, B.D., Sioutas, C., 2014. Long-term source
apportionment of ambient fine particulate matter (PM
2.5
) in the Los Angeles Basin: A
focus on emissions reduction from vehicular sources. Environ. Pollut. 193, 54–64.
doi:10.1016/j.envpol.2014.06.012.
22.Clark, R. G., & Allingham, S. (2010). Robust Resampling Confidence Intervals for
Empirical Variograms. Mathematical Geosciences, 43(2), 243-259. doi:10.1007/s11004-
010-9314-5.
23.Matheron G (1963) Geostatistics. Economic Geology 58 (2), 1246–1266
24.The Importance of the Model Choice for Experimental Semivariogram Modeling and Its
SPATIAL ANALYSIS OF PM
2.5
AND HOSPITAL ADMISSIONS

36
Consequence in Evaluation Process. (2013, October 7). Retrieved from
https://www.hindawi.com/journals/je/2013/960105/.
25.Cressie, N. A. C. (1993). Statistics for Spatial Data. Wiley Series in Probability and
Mathematical Statistics: Applied Probability and Statistics. New York: John Wiley &
Sons, Inc.
26.Zhongquan Lv, & Mingyang Li. (2010). Application of Spatial Balanced Sampling on
forest resources inventory. 2010 International Conference on Computer Application and
System Modeling (ICCASM 2010). doi:10.1109/iccasm.2010.5620396.
27.Brus, D. J., & Heuvelink, G. B. (2007). Optimization of sample patterns for universal
kriging of environmental variables. Geoderma, 138(1-2), 86-95.
doi:10.1016/j.geoderma.2006.10.016.
28.Alexander, D. L., Tropsha, A., & Winkler, D. A. (2015). Beware of R2: Simple,
Unambiguous Assessment of the Prediction Accuracy of QSAR and QSPR
Models. Journal of Chemical Information and Modeling, 55(7), 1316-1322.
doi:10.1021/acs.jcim.5b00206.
29.Kaulfus, A., Alexander, S., Zhao, S., Oster, R. A., OʼKeefe, L. C., & Bartolucci, A.
(2017). The Inherent Challenges of Using Large Data Sets in Healthcare
Research. CIN: Computers, Informatics, Nursing, 35(5), 221-225.
doi:10.1097/cin.0000000000000359.
30.HCUP-US NIS Overview. (n.d.). Retrieved from https://www.hcup-
us.ahrq.gov/nisoverview.jsp.
31.Klakattawi, H., Vinciotti, V., & Yu, K. (2018). A Simple and Adaptive Dispersion
Regression Model for Count Data. Entropy, 20(2), 142. doi:10.3390/e20020142.
SPATIAL ANALYSIS OF PM
2.5
AND HOSPITAL ADMISSIONS

37
32.Sahu, S. K., & Kota, S. H. (2017). Significance of PM
2.5
Air Quality at the Indian Capital.
Aerosol and Air Quality Research, 17(2), 588-597. doi:10.4209/aaqr.2016.06.0262.
33.Dominici, F., Peng, R. D., Bell, M. L., Pham, L., McDermott, A., Zeger, S. L., &
Samet, J. M. (2006). Fine Particulate Air Pollution and Hospital Admission for
Cardiovascular and Respiratory Diseases. JAMA, 295(10),
1127.doi:10.1001/jama.295.10.1127.
34.Moolgavkar, S. H. (2000). Air Pollution and Hospital Admissions for Chronic Obstructive
Pulmonary Disease in Three Metropolitan Areas in the United States. Inhalation
Toxicology, 12(s1), 75-90. doi:10.1080/089583700750019512.
35.Moolgavkar, S. H. (2000). Air Pollution and Daily Mortality in Three U.S.
Counties. Environmental Health Perspectives, 108(8), 777. doi:10.2307/3434732.
36.Burnett, R. T., Cakmak, S., Brook, J. R., & Krewski, D. (1997). The role of particulate
size and chemistry in the association between summertime ambient air pollution and
hospitalization for cardiorespiratory diseases. Environmental health perspectives, 105(6),
614.
37.Alluri, R. K., Leland, H., & Heckmann, N. (2016). Surgical research using national
databases. Annals of Translational Medicine, 4(20), 393-393.
doi:10.21037/atm.2016.10.49.
38.Bhaskaran, K., Gasparrini, A., Hajat, S., Smeeth, L., & Armstrong, B. (2013). Time series
regression studies in environmental epidemiology. International journal of
epidemiology, 42(4), 1187-1195.

Abstract (if available)

Abstract Background: Many studies have shown that exposure to particulate matter air pollution with aerodynamic diameter less than 2.5μm (PM₂.₅) is detrimental to health. We focus on examining the spatial distribution of PM₂.₅ concentrations in California, and its relationship with cause-specific hospital admissions. ❧ Methods: Spatial statistical methods were used to examine the spatial distribution of PM₂.₅ measured at EPA Air Quality System fixed-site monitors in California from 2010 to 2016. Regionally stratified seasonal kriging models were developed to spatially interpolate PM₂.₅ concentrations to the zip code level where hospital admissions from the National Inpatient Sample were recorded. Using Poisson time series regression, the associations between monthly PM₂.₅ and cause-specific hospital admissions including cardiovascular disease and respiratory disease were assessed. ❧ Results: We found that a 10μg/m³ increase in monthly PM₂.₅ exposure was associated with a 1.48% increase in cardiovascular disease hospital admissions (p<0.0001, 95% CI= (1.33, 1.64)), and a 0.24% increase in respiratory disease hospital admissions (p=0.001, 95% CI= (0.098, 0.386)). ❧ Conclusion: Observed positive associations between PM₂.₅ and hospital admissions in California suggest that exposure to PM₂.₅ air pollution remains a serious threat to health.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Assessment of the mortality burden associated with ambient air pollution in rural and urban areas of India

PDF

Associations of ambient air pollution exposures with perceived stress in the MADRES cohort

PDF

Spatial modeling of non-tailpipe emissions and its association with children's lung function

PDF

Examining exposure to extreme heat and air pollution and its effects on all-cause, cardiovascular, and respiratory mortality in California: effect modification by the social deprivation index

PDF

Associations between ambient air pollution and hypertensive disorders of pregnancy

PDF

Assessing the impact of air pollution on adverse birth outcomes in a low resource setting

PDF

Disparities in exposure to traffic-related pollution sources by self-identified and ancestral Hispanic descent in participants of the USC Children’s Health Study

PDF

Native American ancestry among Hispanic Whites is associated with higher risk of childhood obesity: a longitudinal analysis of Children’s Health Study data

PDF

Personal exposure to particulate matter PM2.5 sources during pregnancy and birthweight

PDF

Association of traffic-related air pollution and lens opacities in the Los Angeles Latino Eye Study

PDF

Effectiveness of individual- and household-level protective actions in reducing symptoms associated with hydrogen sulfide chronic low-level exposure

PDF

Air pollution and breast cancer survival in California teachers: using address histories and individual-level data

PDF

Predictive factors of breast cancer survival: a population-based study

PDF

The association between sun exposure and multiple sclerosis

PDF

Air pollution, mitochondrial function, and growth in children

PDF

Contemporary outcomes for adult congenital heart surgery in an adult tertiary care hospital

PDF

Predicting hospital length of stay (LOS) using the National Inpatient Sample

PDF

Dietary and supplementary folate intake and prostate cancer risk

PDF

A cohort study of air-pollution and childhood obesity incidence

PDF

Effect of biomass fuel exposure on infant respiratory health outcomes in Bangladesh

Asset Metadata

Creator Yeh, Mei Yu (author)

Core Title Spatial analysis of PM₂.₅ air pollution in association with hospital admissions in California

School Keck School of Medicine

Degree Master of Science

Degree Program Applied Biostatistics and Epidemiology

Publication Date 04/13/2018

Defense Date 04/13/2018

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

Tag hospital admissions,kriging,LOOCV,OAI-PMH Harvest,particulate matter,PM2.5,spatial

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Franklin, Meredith (committee chair), Cen, Steven (committee member), Habre, Rima (committee member)

Creator Email meiyuyeh@usc.edu,mimiko1109@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c40-492552

Unique identifier UC11268639

Identifier etd-YehMeiYu-6250.pdf (filename),usctheses-c40-492552 (legacy record id)

Legacy Identifier etd-YehMeiYu-6250.pdf

Dmrecord 492552

Document Type Thesis

Format application/pdf (imt)

Rights Yeh, Mei Yu

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 a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA