University of Southern California Dissertations and Theses

Understanding the role of population and place in the dynamics of seasonal influenza outbreaks in the United States

(USC Thesis Other)

Understanding the role of population and place in the dynamics of seasonal influenza outbreaks in the United States

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

Understanding the role of population and place in the dynamics of seasonal influenza outbreaks in the
United States

by

Emily Ann Serman

A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(POPULATION, HEALTH AND PLACE)

August 2022

Copyright © 2022 Emily Ann Serman
ii

Dedication

To my family, friends, and colleagues whose support made this work possible

iii

Acknowledgements
I am so grateful to have had the love and support of so many people over the course of, not just
this Ph.D. but, my entire academic career. From California to Rhode Island, Maryland, Alabama,
New Hampshire, Germany, and Slovenia, I’m beyond lucky to have you all in my corner. A
special thank you to my parents who encouraged all of my interests as a kid and for allowing
your children to bloom in their own ways.
My dear USC friends Avery, Johanna, Kate, Lois, and Xiaozhe: I truly do not think I
would have been able to get this finished without you! The long nights at school doing
homework, the in-person and virtual writing sessions, spring break trips, delicious meals, and
more. My friendships with you are the best thing to come from my time in California.
My co-advisors Jennifer and Meredith: Thank you for everything! I appreciated all the
pep talks, encouragement, and support. I feel extremely lucky to have your guidance through this
process!
My committee members Prabhu, Eileen, and Darren: Thank you for your thoughtful
feedback and help in improving these papers.
My JPL colleagues Heidar, Joao, and Sharon: I learned so much during my time at JPL.
Thank you so much for creating such a supportive environment to learn and grow as a
researcher.

iv

Table of Contents
Dedication ....................................................................................................................................... ii
Acknowledgements ........................................................................................................................ iii
List of Tables ................................................................................................................................. vi
List of Figures ............................................................................................................................... vii
Abbreviations ............................................................................................................................... viii
Abstract ........................................................................................................................................... x
Chapter 1 Introduction .................................................................................................................... 1
1.1. Background .........................................................................................................................3
Chapter 2 Spatial variation in humidity and the onset of seasonal influenza across the
contiguous United States ......................................................................................................... 9
2.1. Introduction .........................................................................................................................9
2.2. Data ...................................................................................................................................12
2.3. Methods.............................................................................................................................13
2.3.1. Data Processing ........................................................................................................13
2.3.2. Segmented Regression .............................................................................................14
2.3.3. ANOVA Analysis ....................................................................................................16
2.4. Results ...............................................................................................................................17
2.5. Discussion .........................................................................................................................24
Chapter 3 Identifying spatial and temporal trends in age-stratified influenza-specific
hospital admission rates by race 2008-2017 .......................................................................... 29
3.1. Introduction .......................................................................................................................29
3.2. Data ...................................................................................................................................34
3.3. Methods.............................................................................................................................36
3.4. Results ...............................................................................................................................38
v

3.4.1. Age group: under 5 ..................................................................................................38
3.4.2. Age group: 5-17 years ..............................................................................................40
3.4.3. Age group: 18-49 years ............................................................................................42
3.4.4. Age group: 50-64 years ............................................................................................44
3.4.5. Age group: 65 years and older .................................................................................45
3.5. Discussion .........................................................................................................................47
Chapter 4 The role of demographics and socioeconomics in the timing of
influenza-related hospital utilization ..................................................................................... 54
4.1. Introduction .......................................................................................................................55
4.1.1. Humidity and influenza............................................................................................56
4.1.2. Demographics and socioeconomic status and influenza ..........................................57
4.1.3. Study aims ................................................................................................................60
4.2. Data ...................................................................................................................................61
4.3. Methods.............................................................................................................................63
4.4. Results ...............................................................................................................................65
4.5. Discussion .........................................................................................................................71
Chapter 5 Conclusion .................................................................................................................... 75
References ..................................................................................................................................... 81
Appendix A Chapter 2 Supplemental Materials ........................................................................... 88
Appendix B Chapter 3 Supplemental Materials ......................................................................... 126
Appendix C Chapter 4 Supplemental Materials ......................................................................... 127

vi

List of Tables
Table 2.1 The resulting breakpoint values estimated from the segmented linear regression
analysis, as well as the adjusted R
2
values for the regressions. The states are grouped into
the NOAA regions used for the ANOVA analysis ....................................................................... 21

Table 2.2 Results from the Tukey Honest Significant Difference test on all of the possible
regional pairwise combinations. The differences in mean breakpoints between pairwise
combinations, lower and upper limits of the 95% confidence interval, as well as the
statistical significance of the comparisons are listed. ................................................................... 23

Table 3.1 Average annual populations for each age and race/ethnicity category in each
state between 2008-2017............................................................................................................... 36

Table 4.1: Distribution of timing of hospital utilizations and demographic profiles of
Mesa and Tucson. Race and age category data is an average annual number calculated
from American Community Survey 5-year estimates that span the analysis period. ................... 61

Table 4.2: Results of the city-stratified logistic regression models and chi-squared tests
for difference between city-stratified coefficients. P-values are marked with statistical
significance indicated with asterisks. ............................................................................................ 70

vii

List of Figures
Figure 2.1. Average weekly humidity versus average weekly incidence for Arkansas
(2003-2015). The red and green lines represent the results of the segmented regression,
their intersection is the breakpoint. ............................................................................................... 16

Figure 2.2 Observed average weekly humidity (kg/kg) versus average ILI count for
each state. The scatter plots for each state are colored by the corresponding NOAA
region they were assigned for the ANOVA analysis and are arranged in their
approximate geographic location. ................................................................................................. 18

Figure 2.3 Each box represents a state, denoted by its state abbreviation (state
abbreviations are listed in Table 2.1 for reference), and is colored according to the
humidity value at the breakpoint which was determined with segmented regression. ................. 20

Figure 2.4 Average annual humidity values for each state (x) versus the extracted
breakpoint for humidity (y) with a simple linear association (blue line). ..................................... 21

Figure 3.1 This plot displays average admission rate ratios in children under 5 years
of age for each race/ethnicity group as compared to White non-Hispanics. ................................ 40

Figure 3.2 This plot displays average admission rate ratios for children 5-17 years
old for each race/ethnicity group as compared to White non-Hispanics. ..................................... 42

Figure 3.3 This plot displays average admission rate ratios for adults 18-49 years
old for each race/ethnicity group as compared to White non-Hispanics. ..................................... 43

Figure 3.4 This plot displays average admission rate ratios for adults 50-64 years
old for each race/ethnicity group as compared to White non-Hispanics. ..................................... 45

Figure 3.5 This plot displays average admission rate ratios for adults 65 years and
older for each race/ethnicity group as compared to White non-Hispanics. .................................. 47

Figure 4.1 Comparison of the odds ratio of early hospital utilization for each
race/ethnicity category compared to White non-Hispanics. ......................................................... 65

Figure 4.2: Comparison of the odds ratio of early hospital utilization for each
age category compared to children aged 5-17. ............................................................................. 67

Figure 4.3 Comparison of the odds ratio of early hospital utilization for each insurance type
compared to individuals with Private insurance. .......................................................................... 68

viii

Abbreviations
ACS American Community Survey
AHRQ Agency for Healthcare Research and Quality
AIRS Atmospheric Infrared Sounder
ANOVA Analysis of Variance
CDC Centers for Disease Control
CHAMPVA Civilian Health and Medical Program of the Department of Veteran’s Affairs
ED Emergency Department
EOS Earth Observing System
FluSurv-NET Influenza Hospitalization Surveillance Network
GFT Google Flu Trends
HCUP Healthcare Cost and Utilization Project
HHS Health and Human Services
HIV/AIDS Human Immunodeficiency Virus/Acquired Immunodeficiency Syndrome
HSD Honest Significant Differences
HVAC Heating, Ventilation, Air Conditioning
ICD-9/10 International Disease Classification, Ninth/Tenth Revision
ICU Intensive Care Unit
ILI Influenza-like Illness
MMR Mass Mixing Ratio
NASA National Aeronautics and Space Administration
NCHS National Center for Health Statistics
NOAA National Oceanic and Atmospheric Administration
ix

SEDD State Emergency Department Database
SID State Inpatient Database
US United States
UV Ultraviolet
ZCTA Zip Code Tabulation Area
x

Abstract
Influenza, or flu, is a contagious respiratory illness with symptoms such as fever, body aches,
fatigue, and cough. Despite access to a yearly vaccine, the United States (US) Centers for
Disease Control (CDC) estimates that between 3-11% of the US population, tens of millions of
people, have symptomatic influenza infections each year. Despite the high morbidity of influenza
as well as the mortality risks it poses for vulnerable individuals, our understanding of the
fundamental transmission dynamics of the disease are still not well understood. Research across
disciplines has shown increasing evidence for the impacts of local climate, network and mobility
functions, demographic and socioeconomic status, as well as the virus’s antigenic characteristics
on our physical vulnerability, our prevention behaviors, and our perception of risk. An increased
understanding of the factors that impact both the timing and magnitude of seasonal outbreaks can
aid in the development of more accurate influenza forecasting models and inform public health
campaigns aimed at preventing and reducing the burden of influenza. This dissertation explores
the role of population and place in these dynamics by: 1. quantifying state-level humidity
thresholds that signal the onset of seasonal outbreaks, 2. Exploring age-stratified racial and
ethnic disparities in influenza-related hospital admissions across space and time, and 3. assessing
the role of underlying population characteristics in the timing of seasonal outbreaks when
humidity is controlled for.
These three studies contribute to the body of influenza work by demonstrating spatial
variation in humidity-influenza relationships, as well as, racial and ethnic disparities.
Furthermore, it provides evidence from a case study for the role of population characteristics in
the timing of influenza outbreaks. The humidity thresholds identified in study one were strongly
associated with annual average humidity for each state (R
2
= 0.90). This finding provides
xi

additional support to the theory of individual level adaptations potentially being the driving
mechanism of the influenza-humidity relationship. In study two, there were notable spatial and
temporal differences in racial and ethnic disparities that highlight the need for sub-national and
age-specific exploration of disparate trends in influenza hospitalizations. Black non-Hispanics,
particularly children, were generally admitted more frequently than White non-Hispanics in all
three states. Hispanics had the most spatial variation in their hospital admissions compared to
White non-Hispanics. Finally, study three provides evidence for the role of population
characteristics, specifically age structure and insurance type, in the timing of influenza outbreaks
in Mesa and Tucson, Arizona. There were statistically significant differences between the cities
in the likelihood of adults aged 18 and older requiring an early hospital utilization compared to
5-17 year olds (18-49 years: Chi-squared=10.50, p-value=0.001; 50-64 years: Chi-squared=5.89,
p-value=0.015; over 65: Chi-squared=7.68, p-value=0.006), as well as for individuals on
Medicaid compared to those with private insurance (Chi-squared=6.33, p-value=0.012). This
result has important implications for the use of population data in models that aim to predict
seasonal influenza outbreaks and for our understanding of factors that contribute to their timing.
The findings of these studies demonstrate the need for local conditions to be considered
in influenza forecasting efforts as well as public health interventions. Both environmental
conditions and population characteristics influenced the timing of influenza outbreaks while age
structure and race/ethnicity played a role in the distribution of cases once outbreaks were
underway. Future work should consider broader spatial and temporal contexts to identify the
scope of variation in these relationships across the United States.

1

Chapter 1 Introduction
One hundred years ago, the 1918 influenza virus pandemic killed over 50 million people
worldwide, including 675,000 people in the United States (CDC 2018). A century and four more
pandemics later, scientists continue to search for more effective prevention and prediction
techniques and endeavor to better understand individual risk factors associated with severe
disease. Despite access to a yearly vaccine, the United States Centers for Disease Control (CDC)
estimates that nearly 10% of the US population, tens of millions of people, are infected with
influenza virus each year (CDC 2021a). Due, in part, to the complex nature of influenza
susceptibility, our understanding of the fundamental transmission dynamics of the disease are
still not well understood (Killingley and Nguyen-Van-Tam 2013). For example, interdisciplinary
research has shown evidence for a variety of influencing factors including, but not limited to:
local climate (Dalziel et al. 2018, Shaman et al. 2010, Towers et al. 2013, Lowen et al. 2007,
Barreca and Shimshack 2012, Soebiyanto and Kiang 2014, Grais and Ellis 2004, Lowen & Steel
2014, Serman et al. 2022), network and mobility functions (Dalziel et al. 2018, Pei et al. 2018,
Maliszewski and Wei 2011, Grais and Ellis 2004, Chao et al. 2010, Gao et al. 2015),
demographic characteristics and socioeconomic status (CDC 2021b, Thompson et al. 2011,
Placzek and Madoff 2014, Sloan et al. 2015, Gounder et al. 2014, Maliszewski and Wei 2011,
Suryaprasad et al. 2013, Truelove et al. 2010, Kwan-Gett et al. 2009, Kumar et al. 2015, Wegner
and Naumova 2011, Maliszewski and Wei 2011, Ponnambalam et al. 2011, Sebastiani et al.
2006, O’Halloran et al. 2021), as well as the virus’s antigenic characteristics (Du et al. 2017,
Towers et al. 2013) on our physical vulnerability, our prevention behaviors, and our perception
of risk. Since 2013, the CDC has hosted an annual seasonal influenza prediction challenge
encouraging both academic and private research groups to create models which can predict the
2

timing, magnitude, and intensity of each years’ outbreak (CDC 2019). While this effort has led to
advancements in influenza forecasting, nearly a decade later, groups are still working to create
models that are consistently accurate, once again highlighting the complex nature of influenza
transmission dynamics (CDC 2019). Influenza forecasting is a national priority due to its high
morbidity. Research that increases our understanding of the contributing factors of an influenza
infection can aid public health entities in anticipating the timing and magnitude of seasonal
outbreaks and therefore provide opportunities to create more targeted vaccine campaigns and to
reduce the overall burden of influenza in the community.
The ultimate goal of the following body of work is to contribute to the understanding of
influenza transmission dynamics by investigating how climate and population characteristics
impact the timing and magnitude of influenza outbreaks in the United States. This narrative will
be divided into three sections. The first focuses on the relationship between humidity and the
timing of seasonal influenza outbreaks at the state-level. Specifically, we quantified humidity
thresholds that signal the start of influenza outbreaks from decadal week-of-year humidity and
influenza incidence averages. Then we considered how this relationship varied over space. In the
second section we describe and compare age and race stratified influenza hospitalizations for
multi-year periods in a collection of states. The purpose of this was to determine if and how
influenza hospitalizations varied by race/ethnicity and whether those relationships also varied
over space and time. The final installment considers a case study of two Arizona cities to
determine the role of population characteristics in the timing of seasonal influenza outbreaks
when humidity is controlled for.

3

1.1. Background
The seasonality of influenza has been linked to seasonal climate changes. There are several
hypotheses about what is influencing the transmission cycle including: changes in host immune
capability, melatonin and vitamin D levels in winter, behavior (eg: staying indoors and the effect
of HVAC systems in winter), as well as environmental factors like temperature, humidity, and
UV irradiation (Lowen & Steel 2014). Evidence supporting the relationship between humidity
and influenza has been strengthened by laboratory studies (Kudo et al. 2019, Lowen et al. 2007,
Lowen & Steel 2014), resulting in three leading theories as to why the two may be closely
linked. They are as follows: 1) virus survival increases as humidity levels decrease; 2) droplet
size decreases with decreasing humidity, allowing particles to travel farther; and 3) low humidity
dries out the mucous membranes of the nose leaving the subject more susceptible to infection
(Lowen et al 2007). More recent work by Kudo et al. (2019) showed that mice kept in low
relative humidity conditions had difficulties in mucociliary clearance, impaired tissue repair in
their airways, and even produced lower amounts of interferon-stimulating genes that blocked
virus spread compared to mice kept in high relative humidity conditions. The first section of the
following text contributes to the understanding of how local environmental conditions influence
influenza transmission. While other studies have considered the impact of humidity on the
magnitude and seasonality of influenza outbreaks (Dalziel et al., 2018; Soebiyanto & Kiang,
2014; Tamerius et al., 2019), this work specifically investigates the onset of outbreaks. In turn,
this can improve the use of variables such as humidity in models that attempt to predict the
timing of seasonal epidemics (Shaman et al., 2010; Shaman et al., 2017). Given that the best
method for influenza prevention is an annual vaccine, an early warning system would allow local
health workers and community members to be better prepared for the start of seasonal
4

transmission. This study analyzed weekly averages from a decade of data to estimate the
relationship between humidity and influenza at the state level for the contiguous United States
and related the patterns found to location specific average annual humidity conditions. This study
used a novel combination of datasets including humidity from NASA’s Atmospheric Infrared
Sounder (AIRS) and incident influenza estimates from Google Flu Trends (GFT).
Though climatic factors are thought to drive the onset of seasonal outbreaks, there is
evidence that, once the outbreak begins, morbidity is not equally distributed among racial and
socioeconomic groups. Recent studies have prompted the inclusion of minority race groups,
particularly minority children, as high risk for severe influenza on CDC websites (CDC 2021b,
O’Halloran et al. 2021). In this context, severe influenza is defined as cases which lead to
hospitalization, Intensive Care Unit (ICU) admission, and in-hospital death (CDC 2021b,
O’Halloran et al. 2021). However, both recent and past investigations of racial and ethnic
disparities have had either a narrow spatiotemporal focus or failed to address spatial and
temporal variation in which groups were more frequently hospitalized with an influenza
diagnosis. For example, while we can compare multiple state-level analyses of the 2009 H1N1
“swine flu” pandemic, they only provide a cross-sectional view of influenza hospitalizations for
a single season. And, while from a pandemic preparedness perspective there is great value in
understanding the dynamics of a major global outbreak on sub-groups of the population, these
results may not be generalizable to an “average” influenza season. Studies that used multiple
seasons of influenza data generally focused on social determinants of influenza hospitalizations
(Tam et al. 2014, Sloan et al. 2015, Chandrasekhar et al. 2017, Hadler et al. 2016). Of these
studies, only two included data from sites across the US but reported their results as a national
estimate (Chandrasekhar et al. 2017, Hadler et al. 2016). Only one, Chandrasekhar et al. (2017),
5

reported state level results but they did not elaborate on any spatial variation nor did they
consider multi-category stratifications of the data (ie: race and age, race and income). There is
considerable heterogeneity in the age, racial/ethnic, and socioeconomic composition of the US
population over space. As such, reporting national level or county specific results may not be
generalizable to other spatial contexts. Therefore, just as we should consider several seasons to
understand how rates of influenza hospitalizations may change over time or in pandemic and
non-pandemic seasons, we should also compare a variety of locations. Furthermore, none of the
aforementioned analyses addressed trends in disparities, likely because the average time period
considered was 5.5 years. Long term trends are important because they can indicate success of
public health campaigns and highlight persistent or increasing disparities that require further
attention. Furthermore, they can provide insight to changes in the overall burden of influenza on
the nation’s healthcare system. Studies designed with a limited spatial or temporal scope are
limited in their ability to contribute to our understanding of racial and ethnic disparities in
influenza hospitalizations within the US, particularly trends in disparate rates over time. The
work presented in the second section of this dissertation aims to address the spatial and temporal
limitations of previous studies by considering both a longer time period of analysis and by
comparing sub-national influenza hospitalization data for multiple locations. This study utilized
data created and maintained by the Agency for Healthcare Research and Quality (AHRQ),
Healthcare Cost and Utilization Project (HCUP) to quantify rates of influenza-specific
hospitalizations for several age and race/ethnicity groups across nine influenza seasons (2008-
2009 through the 2016-2017) in three states (Arizona, Florida, and Maryland). Specifically, we
were interested in identifying if and how rates changed over time, and whether there were
geographical differences. Comparisons of influenza-specific hospital admissions between White
6

non-Hispanic, Black non-Hispanic, Hispanic, Asian/Pacific Islander, and Native
American/Alaskan Native populations for ages younger than 5 years old, 5-17 years old, 18-49
years old, 50-64 years old, and those 65 and older are presented. The results will increase our
understanding of documented racial and ethnic disparities in influenza admissions by providing a
longer temporal comparison of admission rates, identifying state-level differences in admission
rates by race and age, as well as quantifying trends in influenza-specific hospitalizations for this
time period. This can be used to inform public health initiatives such as information and
vaccination campaigns by state and local agencies aimed at reducing influenza hospitalizations,
particularly amongst those populations who are at an increased risk of severe disease.
As outlined above, there is substantial evidence for the role of humidity in the timing of
influenza outbreaks. However, given the number of factors that contribute to an influenza
infection, and the impact of social determinants on the distribution of influenza cases, we should
consider the role of population characteristics as well as environmental conditions on the timing
of seasonal outbreaks. Kumar et al. (2015) is an example of how this research might be
accomplished. Kumar et al. (2015) created an agent-based model to test whether population
structure could account for the differences in attack rates seen during the H1N1 pandemic. In this
context, population structure meant accounting for population density, age structure, average
household size, and contact rates (Kumar et al. 2015). Interestingly, Kumar et al. (2015) found
that census tracts with high levels of poverty saw earlier and steeper increases in hospitalizations
than areas that were less impoverished and that this matched the hospital data from the pandemic
event. They also found that when individual susceptibility and household income were modeled
to be inversely proportional, the attack rates in their simulations were comparable to the attack
rates seen during the pandemic, suggesting that individual poverty could account for area-level
7

inequalities (Kumar et al. 2015). The final study presented here presents a case study on the
impact of demographic and socioeconomic characteristics on early hospital utilizations during
seasonal influenza outbreaks in two Arizona cities. This analysis focused on the cities Mesa and
Tucson, located in south-central and southern Arizona respectively, for the 2008-09 through the
2016-17 influenza seasons. Mesa and Tucson were selected for this case study because they have
distinct differences in the timing of their influenza-related hospital utilization despite having very
similar climate conditions. Over the course of the nine seasons, Tucson had a 5% higher hospital
utilization early in the season compared to Mesa. Since humidity is thought to trigger the onset
of seasonal outbreaks and the conditions in these two cities are comparable, we hypothesized that
differences in the underlying population characteristics explained part of the variation seen in the
timing of their hospital utilizations. Mesa and Tucson have similar age structures, though Mesa
has a slightly larger population of individuals over 65 and Tucson has more individuals in the 18-
49 year category. Additionally, while they have comparable Asian/Pacific Islander, Black non-
Hispanic, and Native American/Alaskan Native populations, Tucson has a substantially larger
Hispanic population than Mesa, with Hispanics making up almost 40% of the population in
Tucson compared to 26% in Mesa. This study used city-stratified logistic regressions to
determine the probability of subgroups of the population (by race/ethnicity, age, and insurance
type) of presenting at the hospital during the earliest months of the influenza outbreaks compared
to later in the season. Given the evidence for links between humidity and the timing of seasonal
influenza outbreaks, this study controlled for humidity using a state- and seasonality-specific
humidity measure derived from thresholds published in the first section. Results from this study
will increase our understanding of populations at higher risk of early hospital utilization in these
two cities. Worby et al. (2015) showed that increasing vaccination rates amongst groups of the
8

population which “lead” influenza outbreaks reduced the overall burden of the seasonal
outbreak. As such, the results of this study provide an example of how one might use information
on past seasons to develop more targeted intervention strategies, such as vaccination campaigns,
that are tailored to meet the specific needs of the community.

9

Chapter 2 Spatial variation in humidity and the onset of seasonal influenza
across the contiguous United States

In recent years, environmental factors, particularly humidity, have been used to inform influenza
prediction models. This study aims to quantify the relationship between humidity and influenza
incidence at the state-level in the contiguous United States. Piecewise segmented regressions
were performed on specific humidity data from NASA’s Atmospheric Infrared Sounder (AIRS)
and incident influenza estimates from Google Flu Trends (GFT) to identify threshold values of
humidity that signal the onset of an influenza outbreak. Our results suggest that influenza
incidence increases after reaching a humidity threshold that is state-specific. A linear regression
showed that the state-specific thresholds were associated with annual average humidity
conditions (R
2
=0.9). Threshold values statistically significantly varied by region (F-
statistic=8.274, p<0.001) and of their 36 pairwise combinations, 13 pairs had at least marginally
statistically significant differences in their means. All of the significant comparisons included
either the South or Southeast region, which had higher humidity threshold values. Results from
this study improve our understanding of the significance of humidity in the transmission of
influenza and reinforce the need for local and regional conditions to be considered in this
relationship. Ultimately this could help researchers to produce more accurate forecasts of
seasonal influenza onset and provide health officials with better information prior to outbreaks.
2.1. Introduction
Every year, millions of people in the United States are infected with the contagious respiratory
illness known as the influenza, or flu, virus. Symptoms of influenza range from mild to severe,
and in some cases require hospitalization that can even lead to death. In the United States,
10

seasonal influenza outbreaks generally begin in fall and peak over the course of the winter or into
the spring months (CDC, 2016). However, the driving factors behind this seasonality are not well
understood. There are several hypotheses about possible influences on the transmission cycle
including: changes in host immune capability, melatonin and vitamin D levels in winter,
behavior (e.g. staying indoors and the effect of HVAC systems in winter), as well as
environmental conditions like temperature, humidity, and UV irradiation (Lowen & Steel, 2014).
Studies focused on the relationship between seasonal influenza and environmental conditions
have found the strongest associations with temperature and humidity in temperate regions of the
world (Dalziel et al., 2018; Lowen et al., 2007; Lowen & Steel, 2014; Shaman et al., 2017;
Soebiyanto & Kiang, 2014; Tamerius et al., 2019). This study focuses on the role of humidity as
a driving factor of state-level seasonal influenza outbreaks in the contiguous United States.
It is theorized that the relationship between humidity and influenza might be explained by
all or some the following three mechanisms: 1) virus survival increases as humidity levels
decrease; 2) droplet size decreases with decreasing humidity, allowing particles to travel farther
and remain suspended in the air for longer; and 3) low humidity dries out the mucous membranes
of the nose leaving the subject more susceptible to infection when they encounter the virus
(Lowen et al, 2007; Lowen & Steel, 2014). However, a more recent study by Kudo et al. (2019)
presented further evidence for impaired host responses to influenza virus when in low humidity
conditions. They found that mice exposed to low relative humidity conditions had more severe
influenza infections compared to those kept in high relative humidity environments and were not
able to clear the virus from their bodies’ due to impaired physical mechanisms and immune
systems responses. Not only did the mice in low humidity have difficulties in mucociliary
11

clearance, they also showed signs of impaired tissue repair in their airways, and had lower
production of interferon-stimulated genes that block virus spread.
While laboratory experiments on animals have utilized relative humidity to better
understand the physical and immunological responses to influenza exposure (Kudo et al., 2019;
Lowen et al., 2007; Lowen & Steel, 2014), from a modeling perspective, absolute humidity
seems a better proxy for human conditions (Shaman et al., 2010; Shaman et al., 2017). Absolute
humidity is the mass of water vapor divided by the total volume of air, while relative humidity is
the ratio of the vapor pressure to the saturation vapor pressure with respect to water and, perhaps
most importantly, is influenced by temperature. Shaman et al. (2017) describe several reasons
why absolute humidity is the more appropriate measure for modeling in human populations. For
example, outdoors, relative humidity is highest in the winter time, while absolute humidity is at
its lowest (Lowen & Steel, 2014; Shaman et al., 2010; Shaman et al., 2017). However, indoors,
where people in high income countries such as the United States, spend the majority of their time
in the winter months, both absolute and relative humidity are low (Lowen & Steel, 2014;
Shaman et al., 2010; Shaman et al., 2017). Therefore, outdoor absolute humidity is easily used to
approximate indoor conditions, where the virus is more likely transmitted from person to person
(Shaman et al., 2017). It is common for specific humidity to be used, in the context of
meteorology and climate, as a replacement for absolute humidity, and this analysis, like
Tamerius et al. (2019), follows that precedent. Specific humidity is the mass of water vapor
divided by the total mass of air.
The analysis presented here contributes to the understanding of how local environmental
conditions influence influenza transmission. While other studies have considered the impact of
humidity on the magnitude and seasonality of influenza outbreaks (Dalziel et al., 2018;
12

Soebiyanto & Kiang, 2014; Tamerius et al., 2019), this work specifically investigates the onset
of outbreaks. In turn, this can improve the use of variables such as humidity in models that
attempt to predict the timing of seasonal epidemics (Shaman et al., 2010; Shaman et al., 2017).
Given that the best method for influenza prevention is an annual vaccine, an early warning
system would allow local health workers and community members to be better prepared for the
start of seasonal transmission. This study analyzes weekly averages from a decade of data to
estimate the relationship between humidity and influenza at the state level for the contiguous
United States and relates the patterns found to location specific average annual humidity
conditions. This study uses a novel combination of datasets including humidity from NASA’s
Atmospheric Infrared Sounder (AIRS) and incident influenza estimates from Google Flu Trends
(GFT).
2.2. Data
The AIRS instrument on-board NASA’s Earth Observing System (EOS) Aqua satellite, launched
in 2002, provides profiles of atmospheric conditions including temperature and humidity (e.g.,
Tian et al., 2013, 2017). Version 6 (V6) Level 3 AIRS observations are available twice daily
from an ascending (daytime, ~1:30pm local time) and descending (nighttime, ~1:30am local
time) path in a 1-degree by 1-degree grid. This study uses water vapor Mass Mixing Ratio
(MMR) at the near-surface level as a substitute for specific humidity values. Water vapor MMR
is the ratio of the mass of water vapor in an air parcel to the mass of dry air for the same parcel
(i.e. g/kg dry air) and near surface MMR is a close proxy to specific humidity (Camuffo, 2014).
In 2008, Google launched the GFT product, which aggregated Google search queries on
influenza activity for more than 25 countries (Olsen et al., 2013). The data provide an estimated
number of Influenza-Like-Illness (ILI) related physicians visits per 100,000 people for each
13

week of the year (Ginsberg et al., 2009). In this context, a “case” is defined as a query for
phrases and keywords such as “flu-like symptoms” or “influenza remedies” in a Google search
engine (Ginsberg et al., 2009). These data are available as weekly estimates from the start of the
2003 influenza season through the end of the 2015 season for select cities, the lower 48
contiguous states, and for all 10 Health and Human Services regions. This study focuses on the
state-level data.
There has been some discussion about the utility and accuracy of GFT data as an
estimation for influenza incidence. In particular, there are arguments that GFT overestimates the
burden of influenza infections, with the 2012-13 season being noted as a particularly extreme
occurrence of this (Olsen et al., 2013; Kandula & Shaman, 2019). This analysis acknowledges
the limitations of GFT as an estimation of influenza in the US and excludes data from the 2009-
10 pandemic and the documented overestimation of the 2012-13 to reduce the influence of these
extreme conditions. Additionally, it should be noted that this is an analysis of weekly averages
over several seasons which may also reduce the impact of inconsistencies in the data.
2.3. Methods
The following sub-sections describe the methods for data processing, segmented linear
regression, and ANOVA analysis.
2.3.1. Data Processing
AIRS 1-degree by 1-degree grid cells were averaged for the 48 contiguous states to generate
state-specific MMR. For a given cell, the location of its center was used to pair the cell (as a
whole) with a state. As mentioned previously, the daily AIRS data contains both an ascending
and descending humidity reading. These readings were averaged together to produce a single
14

measurement for each day and then aggregated to the weekly time scale over the period 2003-
2015 to complement the GFT data.
The GFT data is provided as weekly incidence estimates for each state. Not all states
have data for the entire time period of interest (2003-2015), but this study processed what was
available. In order to avoid skewing from extreme conditions and estimations, the H1N1
pandemic of 2009-2010 and the documented overestimation of GFT data for the 2012-2013
influenza season (Olsen et al., 2013) were excluded from this analysis.
Both the AIRS and GFT datasets for each state were aggregated once more to find the
average humidity and influenza conditions by week of the year over the 2003-2015 time period.
Since the aim of this analysis is to better understand the humidity conditions leading up to and at
the start of the influenza season, the weekly averages were restricted to the period from week 36,
the beginning of September, to the peak average week for each state. “Peak average week” refers
to the week of the year with the highest average influenza over the course of the time series.
These averages were plotted versus influenza incidence and a segmented linear regression was fit
to the data in order to estimate the breakpoint humidity value.
2.3.2. Segmented Regression
This analysis used segmented regression (Muggeo, 2008) to assess the association between
humidity and ILI. Segmented regression is essentially a piecewise linear regression (Eq 1)
whereby breakpoints (or knots, 𝜓 ) are determined iteratively given a starting value. We estimate
a one-breakpoint model with the following form:
𝑦 = 𝛽 0
+ 𝛽 1
𝑥 + 𝛽 2
( 𝑥 − 𝜓 ̃
)
+
+ 𝛾𝐼 ( 𝑥 > 𝜓 )
̃
−

15

where y is the dependent variable, ILI, x is the independent variable, humidity, 𝛽 0
is the
intercept, 𝛽 1
is the slope describing the association between humidity and ILI before the
breakpoint 𝜓 , 𝛽 2
is the slope describing the association between humidity and ILI after
breakpoint 𝜓 , and I is the indicator function (equal to 1 when true). To obtain the optimal
estimate of the breakpoint, 𝜓 ̂
, the linear model is fit iteratively. Through 𝛾𝐼 ( 𝑥 > 𝜓 )
̃
−
the
breakpoint is updated by 𝜓 ̂
= 𝜓 ̃
+ 𝛾 ̂ /𝛽 ̂
2
. Additional details are available in Muggeo (2003) and
Muggeo (2008).
Breakpoints indicate the intersection of parts of the line segment that have different
slopes. In the context of this analysis, the “breakpoint humidity value” is the humidity level
which precedes a sharp increase in influenza cases. An example of this plot, for the state of
Arkansas, can be seen in Figure 2.1 where the breakpoint approximation, located where the red
and green colored segments intersect, is 0.0052 (note that the units are kg/kg because of the use
of MMR as a humidity proxy, and because it is a ratio, units are not included from this point
forward). This study suggests that the breakpoint signals the onset of an influenza outbreak and,
for the example of Arkansas, once the local humidity conditions reach approximately 0.0052, the
number of influenza cases could be expected to rise.
16

Figure 2.1. Average weekly humidity versus average weekly incidence for Arkansas (2003-
2015). The red and green lines represent the results of the segmented regression, their
intersection is the breakpoint.

2.3.3. ANOVA Analysis
We tested for regional differences in the extracted breakpoint values using an ANOVA
regression. States were grouped according to the nine climatological regions previously defined
by the National Centers for Environmental Information (Karl and Koss, 1984). Since the
ANOVA analysis provides only an overall association, a Tukey Honest Significant Differences
(HSD) test was used to determine which of the pairwise regional combinations had statistically
significant differences.
17

2.4. Results
Figure 2.1 shows the 2003-2015 average weekly humidity versus the average weekly influenza
incidence for the state of Arkansas. The one-breakpoint segmented regression is displayed by its
piecewise linear shape (red and green lines) separated by the estimated breakpoint ψ ̂. For
Arkansas, ψ ̂=0.0052.
Scatter plots of weekly humidity climatology versus the average weekly influenza
incidence, as seen in Figure 2.1, were created for each individual state and can be found in
Appendix A (Figures A.1-A.48). The scatter plots, without the overlaying regressions, are also
shown colored by their climatological region and arranged in their approximate geographic
position in the United States in Figure 2.2. Viewing the state-level plots in this manner reveals
how this relationship varies over space by making clear the similarities in the “shape” of the
onset of seasonal outbreaks, as well as highlighting potential limitations of the NOAA climatic
groupings in this context. States in the West and Northwest regions (California and Nevada;
Oregon, Washington, and Idaho, respectively) have similarly steep increases in influenza cases at
both higher and lower humidity values. In the Northeast region, the scatterplots show a smaller
increase at higher humidity values and a generally lower increase in cases than the rest of the
country at humidity values lower than the breakpoint. Finally, the “rounded” less distinct breaks,
as compared to the West and Northwest, in the incidence/humidity relationship found in the
South and Southeast regions are noted. The inclusion of Kansas in the South region is the most
notable potential misfit of the NOAA regional classifications. Kansas appears to have the steeper
more distinct slopes more similar to its northern neighbors Missouri, Nebraska, and Iowa than
the more rounded shape and higher incidence rates seen in states to its south such as Oklahoma,
Arkansas, and Texas. The adjusted R
2
values, indicating how well the segmented regression
18

Figure 2.2 Observed average weekly humidity (kg/kg) versus average ILI count for each state.
The scatter plots for each state are colored by the corresponding NOAA region they were
assigned for the ANOVA analysis and are arranged in their approximate geographic location.

19

model fit, range from 0.51 for the state of Michigan to 0.99 for California with a standard
deviation of 0.13. The complete list of adjusted R
2
values can be found in Table 2.1 along with
each states’ estimated breakpoint values, all of which are grouped by climate region. These
breakpoint values, ψ ̂, are displayed by state in Figure 2.3. In this figure, each state is represented
by a box, once again displayed in its approximate geographic position in the US, and the color
corresponds with the humidity value extracted at the breakpoint. States in the Southeast and
Southern regions had generally higher breakpoint values and Florida and Wyoming have the
highest and lowest breakpoints, respectively. The extracted humidity breakpoint values for each
state were also plotted versus each state’s average annual specific humidity and fitted with a
linear regression (Figure 2.4). The overall relationship between these two humidity values has an
R
2
of 0.90 and the trend is highly linear.

20

Figure 2.3 Each box represents a state, denoted by its state abbreviation (state abbreviations are
listed in Table 2.1 for reference), and is colored according to the humidity value at the breakpoint
which was determined with segmented regression. Wyoming has the lowest breakpoint humidity
value and Florida has the highest.

21

Figure 2.4 Average annual humidity values for each state (x) versus the extracted breakpoint for
humidity (y) with a simple linear association (blue line).

Table 2.1 The resulting breakpoint values estimated from the segmented linear regression
analysis, as well as the adjusted R
2
values for the regressions. The states are grouped into the
NOAA regions used for the ANOVA analysis.
Region State Abbreviation
Estimated Breakpoint
(kg/kg)
Adjusted
R-Squared
Value
Central
Illinois IL 0.00389 0.799
Indiana IN 0.00350 0.973
Kentucky KY 0.00463 0.735
Missouri MO 0.00431 0.744
Ohio OH 0.00406 0.718
Tennessee TN 0.00474 0.960
West Virginia WV 0.00420 0.847

22

East North
Central
Iowa IA 0.00345 0.630
Michigan MI 0.00410 0.506
Minnesota MN 0.00325 0.545
Wisconsin WI 0.00366 0.613
Northeast
Connecticut CT 0.00462 0.589
Delaware DE 0.00614 0.875
Massachusetts MA 0.00384 0.717
Maryland MD 0.00452 0.877
Maine ME 0.00346 0.908
New Hampshire NH 0.00296 0.716
New Jersey NJ 0.00470 0.865
New York NY 0.00407 0.660
Pennsylvania PA 0.00360 0.827
Rhode Island RI 0.00485 0.862
Vermont VT 0.00310 0.893
Northwest
Idaho ID 0.00342 0.654
Oregon OR 0.00339 0.881
Washington WA 0.00336 0.947
South
Arkansas AR 0.00532 0.793
Kansas KS 0.00387 0.697
Louisiana LA 0.00769 0.915
Mississippi MS 0.00724 0.772
Oklahoma OK 0.00545 0.916
Texas TX 0.00587 0.914
Southeast
Alabama AL 0.00601 0.885
Florida FL 0.00937 0.911
Georgia GA 0.00644 0.955
North Carolina NC 0.00567 0.933
South Carolina SC 0.00604 0.938
Virginia VA 0.00455 0.969
Southwest
Arizona AZ 0.00404 0.797
Colorado CO 0.00295 0.941
New Mexico NM 0.00326 0.913
Utah UT 0.00291 0.918
West
California CA 0.00446 0.994
Nevada NV 0.00293 0.899
West North
Central
Montana MT 0.00294 0.631
North Dakota ND 0.00274 0.609
Nebraska NE 0.00339 0.792
South Dakota SD 0.00306 0.588
Wyoming WY 0.00233 0.893

23

In addition to spatial variation in the breakpoint, we also considered variation in the
slopes of the piecewise regression as well as the week of the year that the breakpoint occurred.
However, despite regional similarities in the “shapes” of the epidemic curves, there was little
variation in these values between states and no distinct relationships with annual average
humidity conditions as were seen with the breakpoint. Results from these tests can be seen in
supplemental figures A.50-A.53.
Finally, we have the results of the ANOVA regression and Tukey HSD test. State
breakpoint values statistically significantly varied by region (F-statistic=8.274, p<0.001). All
possible regional pairwise combinations from the Tukey HSD test are listed in Table 2.2 along
with their differences in mean, upper and lower limits and the associated adjusted p-value. Of the
36 possible combinations, 13 of them were at least marginally statistically significant. All of the
significant pairs included one of either the South or Southeast regions. These results can be seen
graphically in Appendix A (Figure A.49), which depicts the significant pairings in red.
Table 2.2 Results from the Tukey Honest Significant Difference test on all of the possible
regional pairwise combinations. The differences in mean breakpoints between pairwise
combinations, lower and upper limits of the 95% confidence interval, as well as the statistical
significance of the comparisons are listed.

Difference 95% Confidence Interval Adjusted

Region Comparison in Means Lower Upper p-value
East North Central-Central -0.00057 -0.00253 0.00139 0.988

Northeast-Central -0.00002 -0.00153 0.00149 1.000
Northwest-Central -0.00080 -0.00295 0.00136 0.949

South-Central 0.00172 -0.00002 0.00346 0.054 .
Southeast-Central 0.00216 0.00042 0.00390 0.006 **
Southwest-Central -0.00090 -0.00285 0.00106 0.847
West-Central -0.00050 -0.00300 0.00201 0.999

West North Central-Central -0.00130 -0.00312 0.00053 0.352
Northeast-East North Central 0.00055 -0.00127 0.00238 0.984

Northwest-East North Central -0.00023 -0.00261 0.00216 1.000
South-East North Central 0.00229 0.00027 0.00430 0.016 *
Southeast-East North Central 0.00273 0.00072 0.00475 0.002 **

24

Southwest-East North Central -0.00033 -0.00253 0.00188 1.000

West-East North Central 0.00008 -0.00263 0.00278 1.000
West North Central-East North
Central -0.00073 -0.00282 0.00137 0.964

Northwest-Northeast -0.00078 -0.00281 0.00125 0.937
South-Northeast 0.00174 0.00015 0.00332 0.023 *
Southeast-Northeast 0.00218 0.00059 0.00376 0.002 **
Southwest-Northeast -0.00088 -0.00270 0.00094 0.808

West-Northeast -0.00048 -0.00288 0.00192 0.999
West North Central-Northeast -0.00128 -0.00296 0.00041 0.267

South-Northwest 0.00252 0.00031 0.00472 0.015 *
Southeast-Northwest 0.00296 0.00075 0.00516 0.002 **
Southwest-Northwest -0.00010 -0.00248 0.00228 1.000
West-Northwest 0.00030 -0.00255 0.00315 1.000

West North Central-Northwest -0.00050 -0.00278 0.00178 0.998
Southeast-South 0.00044 -0.00136 0.00224 0.996

Southwest-South -0.00262 -0.00463 -0.00060 0.004 **
West-South -0.00221 -0.00476 0.00034 0.134

West North Central-South -0.00301 -0.00491 -0.00112 <0.001 ***
Southwest-Southeast -0.00306 -0.00507 -0.00104 <0.001 ***
West-Southeast -0.00266 -0.00520 -0.00011 0.036 *
West North Central-Southeast -0.00346 -0.00535 -0.00157 0.000 ***
West-Southwest 0.00040 -0.00230 0.00311 1.000
West North Central-Southwest -0.00040 -0.00249 0.00170 0.999

West North Central-West -0.00080 -0.00341 0.00181 0.983

Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

2.5. Discussion
The purpose of this analysis is to build upon previous works relating humidity to seasonal
influenza outbreaks (Barreca & Shimshack, 2012; Chattopadhyay et al., 2018; Dalziel et al.,
2018; Shaman et al., 2010; Shaman & Kohn, 2009; Soebiyanto & Kiang, 2014; Tamerius et al.,
2019) by considering how the climate-influenza interaction at the onset of seasonal outbreaks
varies over space. Our findings are generally aligned with previous studies, in that they support
humidity and influenza interaction theories (Barreca & Shimshack, 2012; Dalziel et al, 2018;
Shaman et al., 2010; Shaman & Kohn, 2009; Tamerius et al., 2019). In contrast to the other

25

findings, but similar to Tamerius et al. (2019), the analysis described here provides evidence that
this interaction is relative to local annual humidity (Figure 2.4). This finding may have important
implications for identifying the mechanism that drives the humidity and influenza interaction, as
well as the use of humidity as a driver in influenza prediction models.
As stated in the introduction, there were three early explanations for the importance of
low humidity in influenza transmission. Of these three, it would be expected that the virus’
survival threshold and the impact of humidity on particle size would be related to a humidity
range that would apply in all locations, regardless of average local conditions. However, this
study and the work presented by Kudo et al. (2019), show evidence that supports the third
mechanism involving the role of the mucous membranes of the nasal cavity in protecting
individuals from viruses, which are impacted by average local environmental conditions. Figures
2.2 and 2.3 both provide evidence of spatial variability in both incidence trends during an
average influenza season and the breakpoint humidity values. The ANOVA and Tukey HSD test
provide additional support for distinct regional trends, particularly in the South and Southeast
regions as compared to the rest of the country (Table 2.2).
We believe that our results are consistent with that of Kudo et al. (2019) and that the
impact of humidity might be better considered at the individual or more localized level. For
example, populations adapted to more humid environments with large ranges of seasonal
humidity may not be as susceptible to slight changes in humidity levels as those who live in drier
climates with little seasonal variability. What is considered “low” humidity for a state such as
Florida would still feel quite “high” for inhabitants of a state that is very dry, such as Wyoming,
especially given that even the lowest seasonal ranges of humidity in Florida exceed the highest
ranges of seasonal humidity in Wyoming. This is particularly notable given the significance of

26

the impact of low humidity not only on the mucous membranes of the nasal passage but also
immunological function outlined by Kudo et al. (2019), both of which have substantial impact on
transmission and susceptibility. The linear regression of the estimated breakpoints versus annual
average humidity shown in Figure 4 further supports this argument. In locations where the
annual average humidity is higher, the humidity values associated with the breakpoints, that this
analysis argues signal the onset of the highest weeks of influenza incidence, is also higher. This
linear relationship and insight into potential local thresholds could be used to enhance current
prediction models. Additionally, it could be used to identify a more precise “environmental flu
season” that is dictated by real-time humidity conditions as opposed to a general time period of
October to March.
This finding is partially in agreement with the work of Barreca and Shimshack (2012).
Barreca and Shimshack (2012) performed a regression analysis for 30 years of humidity and
influenza mortality data for a collection of urban counties in the US. They found that the strength
of the association with humidity and temperature depended on the counties’ average humidity
conditions over the course of the year. Specifically, they found that for counties that were
considered to have “high” average humidity during the year, associations were stronger between
mortality and low winter humidity. In “low” average humidity counties and those that were
generally colder, mortality was more strongly correlated with temperature. While we are not
considering mortality data, the differences between “high” average humidity versus “low”
average humidity counties is notable.
Furthermore, Tamerius et al. (2019) also suggest that seasonal characteristics of influenza
vary by regional climatological characteristics across the US. Their investigation of “cross-
seasonal”, or summer transmission, found that the country’s more tropical locales, such as

27

Florida and Hawaii, had higher cross-seasonal transmission and lower seasonal transmission.
This may explain the low incidence values seen for peak weeks in our analysis for Florida and
the more gradual increases in slope as the influenza seasons progress in the South and Southeast
when compared with the West coast of the country where there are sharp increases in slope at
humidity levels lower than the breakpoint (Figure 2.2). Generally, our results are in agreement
with the suggestion that there is a regional or localized trend to the influenza-climate interaction.
This finding is important as it could ultimately lead to an opportunity for more generalized
regional predictions in addition to more localized state predictions. Identifying these regional
transmission patterns could better inform public health campaigns and the distribution bulletins
about increased influenza activity.
Further work is needed to determine whether the trends seen in this analysis are
influenced by population characteristics or additional underlying climatic conditions. Although
there is increasing evidence for the impacts of local climate (Barreca & Shimshack, 2012;
Dalziel et al., 2018; Shaman et al., 2010; Soebiyanto & Kiang, 2014; Tamerius et al., 2019;
Towers et al, 2013), interdisciplinary research has also shown network and mobility functions
(Brownstein et al., 2006; Chao et al., 2010; Gao et al., 2015; Grais & Ellis, 2004; Maliszewski &
Wei, 2011; Pei et al., 2017), demographics (Chitnis et al., 2010; Gounder et al., 2014; Kwan
Ghett et al., 2009; Placzek & Madoff, 2014; Suryaprasadet et al., 2013; Thompson et al., 2011;
Wegner & Naumova, 2011), antigenic characteristics (Du et al., 2017; Towers et al., 2013), as
well as socioeconomic status (Chandrasekhar et al., 2017; Hadler et al., 2016; Kumar et al.,
2015; Ponnambalam et al., 2011; Sloan et al., 2015; Tam et al., 2014) impact our vulnerability to
contracting influenza, our prevention behaviors, and our perception of risk. Additional
interdisciplinary work, similar to that of Chattopadhyay et al. (2018), is needed to parse out the

28

complex, non-linear transmission dynamics of the timing, magnitude, and spatial distribution of
influenza outbreaks, which are not addressed by the regression framework used in this analysis.
Additionally, more flexible, polynomial regressions may improve upon the results presented
here. Future studies should also consider the spatial scale at which this analysis may be
applicable, as another limitation of this study is its use of data aggregated at the state-level. State-
level data, particularly for humidity, can not address intrastate variability in environmental
conditions.
Results from this study improve our understanding of the significance of humidity in the
transmission of influenza. They also reinforce the need for local and regional conditions to be
considered in this relationship. Understanding the variations in this interaction at multiple levels
of spatial and temporal resolution, as well as for varying locales, will help researchers to produce
more accurate forecasts of seasonal influenza onset and provide health officials with better
information prior to outbreaks.

29

Chapter 3 Identifying spatial and temporal trends in age-stratified influenza-
specific hospital admission rates by race 2008-2017

Studies considering racial and ethnic disparities in influenza-specific hospitalizations have
generally failed to address variation in these disparities over space and time. This study
investigates complete hospitalization records in Arizona, Florida, and Maryland for the 2008-
2009 through the 2016-2017 influenza seasons. It compares rates between White non-Hispanic,
Black non-Hispanic, Hispanic, Asian/Pacific Islander, and Native American/Alaskan Native
populations for ages younger than 5 years old, 5-17 years old, 18-49 years old, 50-64 years old,
and those 65 and older. The overarching trends seen in this study showed that Asians
consistently had lower average rates of hospitalization compared to White non-Hispanics in all
three states and over time. Black non-Hispanics and Native American/Alaskan Natives generally
had the highest rates of hospital admission compared to White non-Hispanics, with the largest
disparities occurring in children. Hispanic and Black non-Hispanic populations tended to have
rates that varied by location, with these populations having the highest average rates of
admission in Florida. Future research should consider additional states and longer time series to
determine the extent of spatial and temporal variation of racial and ethnic disparities in influenza
hospitalizations.
3.1. Introduction
Influenza, or flu, is a contagious respiratory illness with symptoms such as fever, body aches,
fatigue, and cough (CDC 2021a). Despite access to a yearly vaccine, the United States (US)
Centers for Disease Control (CDC) estimates that between 3-11% of the US population, tens of
millions of people, have symptomatic influenza infections each year (2021a). It is difficult to

30

quantify the true annual burden of influenza because the majority of cases can be managed at
home and are therefore not captured by healthcare system counts (Reed et al. 2015).
Hospitalization data, from sources such as the CDC’s Influenza Hospitalization Surveillance
Network (FluSurv-NET), are frequently used to estimate influenza hospitalization and by proxy
estimate the overall number of influenza cases in a given year. These approximations can help
public health officials understand the burden on health care infrastructure, as well as inform
vaccination campaigns aimed at reducing the overall incidence of influenza (Chaves et al. 2015).
Adults over the age of 65, young children, and people with conditions such as asthma, heart
disease, HIV/AIDS, and cancer are considered to be at a higher risk for severe influenza
infections (CDC 2021a).
Recent studies have prompted the inclusion of minority race groups, particularly minority
children, as high risk for severe influenza on CDC websites (CDC 2021b, O’Halloran et al.
2021). In this context, severe influenza is defined as cases which lead to hospitalization,
Intensive Care Unit (ICU) admission, and in-hospital death (CDC 2021b, O’Halloran et al.
2021). However, both recent and past investigations of racial and ethnic disparities have had
either a narrow spatiotemporal focus or failed to address spatial and temporal variation in which
groups were more frequently hospitalized with an influenza diagnosis. For example, while we
can compare multiple state-level analyses of the 2009 H1N1 “swine flu” pandemic, they only
provide a cross-sectional view of influenza hospitalizations for a single season. And, while from
a pandemic preparedness perspective there is great value in understanding the dynamics of a
major global outbreak on sub-groups of the population, these results may not be generalizable to
an “average” influenza season. Studies that used multiple seasons of influenza data generally
focused on social determinants of influenza hospitalizations (Tam et al. 2014, Sloan et al. 2015,

31

Chandrasekhar et al. 2017, Hadler et al. 2016). Of these studies, only two included data from
sites across the US but reported their results as a national estimate (Chandrasekhar et al. 2017,
Hadler et al. 2016). Only one, Chandrasekhar et al. (2017), reported state level results but they
did not elaborate on any spatial variation nor did they consider multi-category stratifications of
the data (ie: race and age, race and income). There is considerable heterogeneity in the age,
racial/ethnic, and socioeconomic composition of the US population over space. As such,
reporting national level or county specific results may not be generalizable to other spatial
contexts. Therefore, just as we should consider several seasons to understand how rates of
influenza hospitalizations may change over time or in pandemic and non-pandemic seasons, we
should also compare a variety of locations. Furthermore, none of the aforementioned analyses
addressed trends in disparities, likely because the average time period considered was 5.5 years.
Long term trends are important because they can indicate success of public health campaigns and
highlight persistent or increasing disparities that require further attention. Furthermore, they can
provide insight to changes in the overall burden of influenza on the nation’s healthcare system.
Studies designed with a limited spatial or temporal scope are limited in their ability to contribute
to our understanding of racial and ethnic disparities in influenza hospitalizations within the US,
particularly trends in disparate rates over time.
Regardless of spatial and temporal scale, studies have suggested that minority
populations are at a disproportionate disadvantage for severe influenza relative to non-Hispanic
Whites (Thompson et al. 2011, Truelove et al. 2010, Kwan-Gett et al. 2009, Suryaprasad et al.
2013, Placzek and Madoff 2014, Sloan et al. 2015, Gounder et al. 2014, Maliszewski and Wei
2011, Chandrasekhar et al. 2017, Hadler et al. 2016, Tam et al. 2014, O’Halloran et al. 2021).
However, there is notable variation by place and time in which minority populations have

32

disparate rates of influenza-specific hospital admissions. Of the studies that focused on the H1N1
pandemic, non-Hispanic Blacks, Asians, and Hispanics all had greater rates than non-Hispanic
Whites in Wisconsin (Truelove et al. 2010); in King County, Washington only Hispanics and
non-Hispanic Blacks (Kwan-Gett et al. 2009); in New Mexico only Native Americans and
Hispanics (Thompson et al. 2011); and in Massachusetts Placzek and Madoff (2014) found
disparities for Hispanics only.
Many of the analyses that considered multiple seasons used partial or complete data from
the aforementioned FluSurv-NET. FluSurv-NET is hospitalization data aggregated from a
network of 14 sites across the US (Chaves et al. 2015). These sites include select counties from:
California, Colorado, Connecticut, Georgia, Maryland, Michigan, Minnesota, New Mexico, New
York, Ohio, Oregon, Rhode Island, Tennessee and Utah. Together, these sites have a service area
that includes almost 10% of the US population (Chaves et al. 2015). All but one of the studies
that utilized FluSurv-NET used it to consider the role of socioeconomic status, as well as race
and ethnicity, in influenza hospitalizations. They found that, at the census tract level and across
all locations and various time periods, influenza hospitalizations increased with percent of the
population living below the poverty line (Tam et al. 2014, Sloan et al. 2015, Chandrasekhar et al.
2017, Hadler et al. 2016).
While the impact of socioeconomic status on influenza hospitalization was consistent
over space and time, as with H1N1 focused studies, there was some spatial variation in which
minority populations experienced disparate rates. This seemed to be at least partially due to
which race/ethnicity groups were included in the analysis. For example, in a collection of Middle
Tennessee counties, Sloan et al. (2015) only reported racial disparities when neighborhood
socioeconomic conditions were further stratified by the race groups White, Black, and Other,

33

with Blacks having higher incidence rates across all seven seasons considered. However, Tam et
al. (2014) and Chandrasekhar et al. (2017) both found that non-Hispanic Blacks and Hispanics
were at a higher risk of hospitalizations compared to non-Hispanic Whites. Tam et al. (2014)
reported these findings for New Haven County, Connecticut and Chandrasekhar et al. (2017) for
all 14 FluSurv-NET sites. O’Halloran et al. (2021) was the only study to focus specifically on
racial and ethnic disparities in influenza hospitalization rates within FluSurv-NET and were the
only group to include Native Americans and Alaskan Natives in their analysis. In addition to
disparities for non-Hispanic Blacks and Hispanics noted by Tam et al. (2014) and Chandrasekhar
et al. (2017), O’Halloran et al. (2021) found Native Americans and Alaskan Natives were also at
increased risk of hospitalization compared to non-Hispanic Whites. Hadler et al. (2016), who
used the FluSurv-NET data from all 14 sites, reported that hospitalizations increased with
poverty for all race/ethnicity groups but did not discuss the between group differences or spatial
variation present in their results.
Two studies considered severe influenza infections specifically among Native Americans
and Alaskan Native populations compared to all other race groups. Suryaprasad et al. (2013)
focused on Native American populations in the Southwest and found higher hospitalization rates
for this population during the pandemic compared to all other groups in the US. Similarly,
Gounder et al. (2014) found persistently higher rates of influenza hospitalization for Native
Americans and Alaskan Natives compared to the general US population from 2001-2011.
This study aims to address the spatial and temporal limitations of previous studies by
considering both a longer time period of analysis and by comparing sub-national influenza
hospitalization data for multiple locations. Specifically, we compare average rates of influenza-
specific hospitalizations for several age and race/ethnicity groups across two time periods,

34

comprised of nine influenza seasons in total (2008-2009 through the 2016-2017), in three states
(Arizona, Florida, and Maryland), to determine if and how rates changed over time, and whether
there were geographical differences. Comparisons of influenza-specific hospital admissions
between White non-Hispanic, Black non-Hispanic, Hispanic, Asian/Pacific Islander, and Native
American/Alaskan Native populations for ages younger than 5 years old, 5-17 years old, 18-49
years old, 50-64 years old, and those 65 and older are presented. The results will increase our
understanding of documented racial and ethnic disparities in influenza admissions by providing a
longer temporal comparison of admission rates, identifying state-level differences in admission
rates by race and age, as well as quantifying trends in influenza-specific hospitalizations for this
time period. This can be used to inform public health initiatives such as information and
vaccination campaigns by state and local agencies aimed at reducing influenza hospitalizations.
3.2. Data
Hospitalization data in this analysis are from the State Inpatient Database (SID) which is created
and maintained by the Agency for Healthcare Research and Quality (AHRQ), Healthcare Cost
and Utilization Project (HCUP). The SID includes a wide variety of deidentified patient
information including demographics such as race, age, and home zip code, as well as pertinent
medical information such as diagnosis codes, treatment details, and patient outcomes. The SID is
a census of all patients who were admitted to the facilities that are reported by the state of
interest. Since 1994, the SID has included both community hospitals, including academic
medical centers and specialty facilities, as well as non-community hospitals, which includes
Federal institutions run by Veteran’s Affairs, the Department of Defense, and Indian Health
Services (ARHQ 2019).

35

The analysis presented here considers data from Arizona, Maryland, and Florida for the
years of 2008-2017. This time range covers nine influenza seasons beginning with the 2008-2009
influenza season through the 2016-2017 season; due to the nature of the northern hemisphere
seasons spanning portions of two calendar years. These three sites were chosen based on the
availability of data for the time period of interest, as well as their representation of minority
populations. According to US Department of Health and Human Services (HHS), Arizona ranks
in the top ten for significant populations of Hispanics, Native American/Alaskan Natives, and
Native Hawaiian/Pacific Islanders, while Maryland ranks for Black non-Hispanic populations,
and Florida ranks for Black non-Hispanic, Hispanic, Asian, and Native Hawaiian/Pacific Islander
populations (HHS 2021). It should also be noted that, in the SID, the Asian race category
includes individuals that identify as Native Hawaiian and/or Pacific Islander. Due to the low
numbers of Native American/Alaskan Native populations seen in Maryland and Florida (Table
1), they were not considered in this analysis. These states also provide an interesting case study
of differences in the Hispanic population based on country of origin. For example, Arizona has
the third largest Mexican population in the US, while Florida has the highest Cuban and Puerto
Rican populations in the country (Pew Research Center 2017). These differences are important to
note when interpreting hospitalization trends as they may relate to variation in comorbidity
prevalence, language barriers, vaccine uptake rates, occupation type as it relates to exposure risk,
and trust in US health care services among these subgroups.
Population data were compiled from the bridged-race population estimates by age and
race/ethnicity, made available by HHS, CDC, and National Center for Health Statistics (NCHS)
and accessed through the CDC Wonder data portal. These are mid-year estimates which were
matched with influenza season based on the season start date. For example, the mid-year 2007

36

population estimates are used for the 2007-2008 influenza season that begins in September of
2007. Additionally, the population estimates are given for single year ages so they were
aggregated to the age groups of interest for this study.

Table 3.1 Average annual populations for each age and race/ethnicity category in each state
between 2008-2017

Arizona Maryland Florida

Total
Population*
Percent
Population
Total
Population*
Percent
Population
Total
Population*
Percent
Population
Age category

under 5 years
89,139 6.8% 73,320 6.2% 217,185 5.6%

5-17 years
235,078 17.9% 196,561 16.8% 590,732 15.2%

18-49 years
556,896 42.4% 514,287 43.8% 1,596,315 41.1%

50-64 years
236,032 18.0% 235,418 20.1% 767,377 19.8%

over 65 years
195,602 14.9% 153,740 13.1% 708,182 18.3%
Race/Ethnicity

White non-Hispanic
760,191 57.9% 641,122 54.6% 2,234,242 57.6%

Black non-Hispanic
57,581 4.4% 353,137 30.1% 616,399 15.9%

Hispanic
396,073 30.2% 102,131 8.7% 908,512 23.4%

Asian/Pacific Islander
43,320 3.3% 73,434 6.3% 109,403 2.8%

Native
American/Alaskan
Native
55,583 4.2% 3,504 0.3% 11,235 0.3%

*Average number of persons in each category between 2008-2017

3.3. Methods
Influenza has a distinct seasonal pattern in temperate regions of the world, like the US, with
increases in cases beginning in the fall months and typically peaking over the course of the
winter and early spring (CDC 2021). As such, the first step in this analysis was defining the time
range for influenza seasons. In order to match time periods across states, an influenza season in
this study is defined as October 1 through March 31. This also addresses differences in the
reporting of time between the three states. Arizona provides the month of admission for patients

37

while Maryland and Florida provide the discharge quarter. Since 94% of influenza-specific
hospitalizations in Maryland and Florida last two weeks or less, we make the assumption that
patients were likely to have been discharged during the same quarter they were admitted.
Next, inpatients with influenza-specific diagnosis codes were identified within the SID.
The time period of 2008-2017 includes the transition in medical records from the International
Disease Classification, Ninth Revision (ICD-9) to the International Disease Classification, Tenth
Revision (ICD-10). As such, this analysis includes both the ICD-9 and ICD-10 codes for
influenza, 487-488 and J09-J11 respectively, to identify patients with an influenza diagnosis
anywhere within their listed diagnosis codes (ie: influenza did not need to be their primary
diagnosis). Furthermore, cases were restricted to patients who explicitly identified their home
state as Arizona, Maryland, or Florida, depending on the state being analyzed at that time.
Patients with missing home state information or home states other than the state of interest were
excluded. Patients with missing information on sex or race/ethnicity or who identified as “other”
were also excluded from this study.
Following the precedent of studies by Tam et al. (2014) and Chandrasekhar et al. (2017)
cases were sorted into the following five age categories: under 5, 5-17 years old, 18-49, 50-64,
and 65 and over. The race and ethnicity groups included in this analysis are: White Non-
Hispanic, Black Non-Hispanic, Asian/Pacific Islander, Native American/Alaskan Native, and
Hispanic. The number of hospitalizations per age and race/ethnicity group were divided by the
respective state population of each age and race/ethnicity group to calculate hospital admission
rates per population for each season and state.
Due to the multiple stratifications of the data, and the relatively rare nature of influenza-
specific hospitalization compared to the number of influenza cases in a year, there are some

38

instances of small numbers. Following the NCHS protocol for generating population rates when
there are fewer than 20 cases, we created multi-year averages of influenza hospital admission
rates for each of the five race/ethnicity and five age groups described above. The data were
divided into two time periods. The earlier time period consists of the 2008-2009 season through
the 2012-2013 season, and the later period the 2013-2014 season through the 2016-2017 season.
The use of two time periods also allowed for this study to address whether trends in influenza
hospitalizations changed over the course of the nine seasons considered. Within each age
category, the rates of racial and ethnic minorities were compared to those of White non-
Hispanics. These rate ratios are displayed by age category in Figures 3.1-3.5 and the associated
values are listed in Table B.1 in Appendix B.
3.4. Results
The results for this analysis will be presented by age group.
3.4.1. Age group: under 5
The CDC has noted that racial and ethnic disparities in hospital admissions are particularly high
for children (2021a) and these trends are reflected in this analysis. Figure 1 displays average
admission rate ratios as compared to White non-Hispanics for two time periods for each state.
Each dot is colored and shaped to represent one of the four race/ethnicity groups. This allows us
to visualize the differences in average hospital admission rates as they compare to White non-
Hispanics for all the race/ethnicity groups within each state and between states, as well as see the
changes in the average rates over the span of nine seasons.
On average, Asian children had lower hospital admission rates than White non-Hispanics,
a trend that was consistent across all three states. Hispanic children in Maryland and Arizona had
admission rates that were comparable to White non-Hispanics over the two time periods.

39

However, in the 2008-09 through the 2012-13 influenza seasons Hispanic children in Florida
were admitted approximately 1.75 times as frequently as White non-Hispanic children, a rate
ratio which dropped only slightly in the following time period. Average hospital admission rates
for Black non-Hispanic children in Arizona were just under 1.5 times the average rate of White
non-Hispanics during both time periods. Black non-Hispanics in Maryland also had admittance
rates that were approximately 1.5 times those of White non-Hispanics in the earlier time period
but experienced a substantial increase in disparities over the analysis period. In comparison,
Black non-Hispanics in Florida were admitted at more than twice the rate of White non-
Hispanics during the earlier time period of this analysis, and were admitted at 2.25 times the rate
of White non-Hispanics during the latter period. Native American and Alaskan Native children
were only included in this study for the state of Arizona and they had by far the highest rates of
admission in the state, being more than twice as likely as White non-Hispanic children to be
hospitalized with influenza across both time periods.

40

Figure 3.1 This plot displays average admission rate ratios in children under 5 years of age for
each race/ethnicity group as compared to White non-Hispanics. Averages are displayed for two
time periods: the 2008-09 season through the 2012-13 season and the 2013-14 season through
2016-17 season. Each average value is colored and shaped according to the race/ethnicity group
it represents.

3.4.2. Age group: 5-17 years
As seen in the under 5 age category, Asian children are, on average, admitted less frequently
than White non-Hispanic children in Arizona and Florida (Figure 3.2). Asian and Hispanic
children in Maryland were not included in this age category due to small numbers causing
unstable admission rate estimates. Hispanic children have rates of admission that are slightly
lower than White non-Hispanics in Arizona, but slightly higher rates in Florida across both time

41

periods of the analysis. Native American and Alaskan Native children in Arizona were admitted
at approximately the same rate as White non-Hispanics in the 2008-13 seasons but saw an
increase in admittance during the 2013-17 time period. This pattern is much different from the
trends for younger children and the only instance where Native American and Alaskan Native
individuals was hospitalized less than White non-Hispanics. Once again, Black non-Hispanics
had the highest rates of admission compared to White non-Hispanics, this time in all three states.
Furthermore, Black non-Hispanics in Maryland and Florida saw an increase in the average
admission disparity between Black non-Hispanics and White non-Hispanics across time. Florida
saw the largest increase in disparity, with Black non-Hispanics being admitted, on average, at
nearly twice the rate of White non-Hispanics during the 2013-17 time period, compared to only
1.5 times the rate in the prior period.

42

Figure 3.2 This plot displays average admission rate ratios for children 5-17 years old for each
race/ethnicity group as compared to White non-Hispanics. Averages are displayed for two time
periods: the 2008-09 season through the 2012-13 season and the 2013-14 season through 2016-
17 season. Each average value is colored and shaped according to the race/ethnicity group it
represents.

3.4.3. Age group: 18-49 years
As in the younger age groups, Asians were hospitalized at about half the rate of White non-
Hispanics in all three states (Figure 3.3). Hispanics in this age group were generally admitted at
the same rate as White non-Hispanics, and in Arizona and Florida these rates were stable across
both time periods. Hispanics in Maryland saw a slight increase in average admittance rate
compared to White non-Hispanics over the two time periods considered. Black non-Hispanics

43

once again had disproportionally high rates of hospitalization compared to White non-Hispanics
and were admitted, on average, between 1.5 and 2 times as frequently, with slight increases in
disparity across the time periods, in all three states. Native Americans and Alaskan Natives in
Arizona saw a decrease in their average admission rate ratio over time. In the earlier time period,
they were hospitalized more than twice as often while in the more recent period they were
hospitalized about 1.5 times as frequently than White non-Hispanics.

Figure 3.3 This plot displays average admission rate ratios for adults 18-49 years old for each
race/ethnicity group as compared to White non-Hispanics. Averages are displayed for two time
periods: the 2008-09 season through the 2012-13 season and the 2013-14 season through 2016-
17 season. Each average value is colored and shaped according to the race/ethnicity group it
represents.

44

3.4.4. Age group: 50-64 years
The differences in admission rates between race/ethnicity group in the 50-64 year age category
are similar to those seen in the 18-49 age group. Once again, Asians and Hispanics have lower
and comparable rates of admittance compared to White non-Hispanics in all three states, though
average rates of admission for Hispanics in the earlier time period were slightly higher in this age
group (Figure 3.4). Black non-Hispanics were admitted, on average, between 1.5 and 2 times as
frequently as White non-Hispanics in all three states. While rates were relatively stable in
Maryland and Florida, Arizona saw a reduction in the difference in admissions for Black non-
Hispanics compared to White non-Hispanics across the time periods. Native Americans and
Alaskan Natives in Arizona saw no real change in disparity across the nine seasons considered
here. In both time periods, Native Americans and Alaskan Natives were, on average, admitted
nearly twice as often than White non-Hispanics in the state.

45

Figure 3.4 This plot displays average admission rate ratios for adults 50-64 years old for each
race/ethnicity group as compared to White non-Hispanics. Averages are displayed for two time
periods: the 2008-09 season through the 2012-13 season and the 2013-14 season through 2016-
17 season. Each average value is colored and shaped according to the race/ethnicity group it
represents.

3.4.5. Age group: 65 years and older
Like children under 5, adults over the age of 65 are considered to be at an increased risk of
severe illness from influenza. However, generally, where children under 5 had the largest
disparities between race/ethnicity groups, the 65 and older category had the smallest disparities
(Figure 5). In addition, when comparing hospital admission rates between race/ethnicity groups
in the 65 and older population, there were notable state-level trends that varied from the other

46

four age categories. In Arizona, there is a distinct convergence of the average admittance rates,
resulting in all five groups being admitted at roughly the same rate in the later time period, albeit
slightly more frequently for minorities than White non-Hispanics. Asians in Arizona had a
notable increase in hospital admissions while Native Americans and Alaskan Natives had a
notable decrease. In Maryland, all three minority groups were hospitalized at comparable or
lower rates compared to White non-Hispanics where as in all four younger age groups, Black
non-Hispanics were admitted at least 1.5 times as frequently. In Florida, though Hispanic and
Black non-Hispanic older adults continue to be admitted at higher rates than White non-
Hispanics, this is the only instance where Black non-Hispanics are admitted less than 1.5 times
as frequently and where they have smaller disparities than Hispanics.

47

Figure 3.5 This plot displays average admission rate ratios for adults 65 years and older for each
race/ethnicity group as compared to White non-Hispanics. Averages are displayed for two time
periods: the 2008-09 season through the 2012-13 season and the 2013-14 season through 2016-
17 season. Each average value is colored and shaped according to the race/ethnicity group it
represents.

3.5. Discussion
The 2009 H1N1 pandemic drew attention to the disproportionate burden of influenza
hospitalizations on racial and ethnic minorities in the US. It is important to note, however, that
these disparities are not isolated to pandemic events, but are present in seasonal, non-pandemic
influenza outbreaks as well. Understanding how these differences present over space and time
can aid public health entities in creating more targeted vaccine and information campaigns that

48

aim to reduce disparate outcomes like hospitalization. This study builds on previous works in a
few ways. First, we used a census of influenza-specific hospitalization records for three states to
determine age-stratified, multi-year average hospital admission rates by race/ethnicity, as
opposed to calculating rates from representative samples like FluSurv-NET. Second, we
considered how average hospitalization rates for these groups varied by state. And finally, we
investigated whether there are changes in these average rates over time.
There were overarching trends seen in the analysis presented here. Across all age groups,
time periods, and locations, Asians consistently had lower rates of hospitalization compared to
White non-Hispanics. For children (age groups under 5 and 5-17), Black non-Hispanics
consistently had the highest rates of hospitalization compared to White non-Hispanics. There
was a large increase in average hospital admission rates in Maryland for children under 5 and a
similarly large increase for children 5-17 in Florida. Furthermore, Hispanic children in Florida
had higher rates of admittance than in Arizona or Maryland. This spatial variation for Hispanics
was not seen in older age groups. Though there is no state level comparison for Native American
and Alaskan Native populations, adults 65 and older and adults 18-49 both saw notable decreases
in average disparity compared to White non-Hispanics in Arizona. Adults 50-64 and children
under 5 saw no real change in their average rates of hospitalization over time, while children 5-
17 saw a substantial increase. Generally, the largest increases in disparate rates of hospital
admission occurred in the youngest age groups. There were distinct similarities in the trends and
magnitude of disparities between the under 5 and 5-17 age categories and, separately, the 18-49
and 50-64 age groups, with the 65 and older group having differences that were unique from the
younger ages.

49

The results presented here are generally in agreement with other multi-year and multi-
state studies in that Black non-Hispanics and Native Americans and Alaskan Natives are
hospitalized at disproportionate rates compared to White non-Hispanics across several age
categories (Gounder et al. 2014, Tam et al. 2014, Chandrasekhar et al. 2017, O’Halloran et al.
2021). Because of the state level differences, our results for Hispanic populations were less
generalizable, and are therefore more difficult to compare to rates published in previous studies.
O’Halloran et al. (2021), Tam et al. (2014), and Chandrasekhar et al. (2017) reported that
Hispanics had risk ratios that ranged between 1.2 and 1.87 across their samples. In this study,
which calculates rates from a census of influenza hospital admissions, the median hospital
admission rate ratios for Hispanic populations across all age groups ranges from 0.50-0.95 in
Maryland, from 0.79-1.16 in Arizona, and 0.98-1.74 in Florida (see Table S1). As previously
mentioned, the country of origin varies for Hispanics in the three states considered in this
analysis. The spatially varying differences in admittance rates may reflect cultural or
socioeconomic differences among subgroups of Hispanic populations, highlighting not only the
importance of identifying and understanding spatial variation in disparities but also the need for
more detailed demographic information in health records. Though rates for the Asian and Pacific
Islander group were relatively consistent over space and time in this study, the same might be
said for the need for more complete demographic information for this group as well.
The generally smaller disparities seen in the 65 and older age group of this study was also
an interesting result, as were the again notable state-level differences for Black non-Hispanics
and Hispanics. In Maryland, all minority groups had lower hospital admission rates compared to
White non-Hispanics. This was the only instance in this study where all average hospitalization
rate ratios for minority populations were less than one and the only time this occurred for Black

50

non-Hispanics. Hospitalization rate ratios among adults 65 and older in Arizona saw a striking
convergence over the period of analysis with all minority groups having slightly higher
admittance rates than White non-Hispanics in the most recent time period. This was also the only
instance of Hispanics having higher average admission rates than Black non-Hispanics, noted
here in Florida. O’Halloran et al. (2021), the only other study that considered age-stratified racial
and ethnic differences in hospitalization rates, found similarly smaller racial and ethnic
differences in the oldest age groups of their study. However, O’Halloran et al. (2021) applied
two categories for adults over 65 (65-74 years and 75 and older) and did not account for spatial
variation so this is not a direct comparison. Other possible explanations for this observation may
include vaccination campaigns targeted specifically towards older adults due to their increased
risk of hospitalization, as well as Medicare coverage which covers any fees associated with the
influenza vaccine. There may be additional biological factors associated with aging, such as
increased likelihood of comorbidities and decreased immune function that, to some degree, effect
all older adults regardless of their race or ethnicity. The combined effect of insurance coverage
and increased comorbidities likely increases the number of doctor’s visits for older adults, and
presents more opportunities for them to receive the vaccination while they are already in a
medical setting.
Given the nature of selecting admissions in the SID by influenza-specific diagnosis
codes, it is likely that the rates presented here are an under-estimation of the burden of influenza
hospitalizations. This is in part due to the accuracy of influenza diagnoses not always being
reliable and the presence of other seasonal respiratory illnesses that could lead to confounding.
Furthermore, it is important to acknowledge that influenza hospitalizations are, compared to
influenza cases, relatively rare. Individuals who are hospitalized are likely to have other

51

comorbid conditions that are disproportionally experienced by racial and ethnic minorities. As
such, disparities found by this analysis may not be unique to influenza, but they may also reflect
broader implications of health inequity. In addition to factors of socioeconomic status (Placzek
and Madoff 2014, Sloan et al. 2010, Chandrasekhar et al. 2017, Hadler et al. 2016, Tam et al.
2014, Thompson et al. 2011), prevalence of comorbidities among Black non-Hispanic
(O’Halloran et al. 2021) and Native American/Alaskan Native (Gounder et al. 2014) populations,
have been noted as possible explanations for disparate hospitalization rates. For example,
O’Halloran et al. (2021) found in their sample that Black non-Hispanic children were more likely
than children of other races to have asthma and blood disorders, while Gounder et al. (2014)
reported Native American/Alaskan Native populations of all ages were more likely than other
races to have comorbid conditions that increase risk of severe influenza.
Additionally, though vaccination remains the most effective way to prevent influenza, the
CDC notes lower rates of vaccine uptake by minority groups in the US (CDC 2021b). For the
2019-20 influenza season, vaccination coverage for Hispanics and Black non-Hispanics were
38% and 41% respectively, compared to 52% of Asian/Pacific Islanders and 53% of White non-
Hispanics. Low rates of vaccination translate into hospitalization data, as O’Halloran et al.
(2021) notes in their study sample that hospitalized Black non-Hispanics and Hispanics had
lower vaccination rates across all age groups compared to White non-Hispanics. A 2010 survey
about seasonal influenza vaccinations at low-income clinics in Los Angeles, California showed
that Black non-Hispanics and Hispanics were less likely to have received a vaccine against the
H1N1 pandemic strain than their White non-Hispanic and Asian or Pacific Islander peers
(Redelings et al. 2012). Furthermore, Black non-Hispanics respondents were more concerned
about the safety and efficacy of influenza vaccines than other racial and ethnic groups and were

52

less likely to agree that they could trust the doctors or clinicians who recommended vaccination
(Redelings et al. 2012). Increases in average hospital admission rate ratios between the time
periods considered were seen in 8 out of the 12 age/state combinations for Black non-Hispanics
under 65. These increased disparities coupled with noted vaccine hesitancy among this
population, and higher rates of comorbid conditions noted in literature highlight a critical public
health concern that needs further attention.
The use of state-level data is another limitation of this study. Given that there are between
state differences in disparities of hospital admission rates, and that previous studies have found
strong associations with socioeconomic status and influenza hospitalization, it seems likely that
there will be within state differences in disparate rates as well. Truelove et al. (2010) considered
Wisconsin hospitalization rates during the H1N1 pandemic and found that there were statistically
significant differences in hospitalization rates for the state’s largest city, Milwaukee, compared
to the rest of the state. Future studies should use the work presented here and from previous
analyses to determine the optimal spatial scale at which to discuss racial and ethnic disparities in
influenza hospitalizations.
Thus far, research pertaining to racial and ethnic disparities in influenza hospitalizations
have failed to address spatial and temporal variation in influenza hospitalization disparities. This
study provides evidence for substantial state-level differences for Black non-Hispanic and
Hispanic populations in particular. Understanding how racial and ethnic disparities vary spatially
will allow for more targeted community level information and vaccination campaigns while
analysis of long-term trends will help to determine the effectiveness of these efforts. Future
research on racial and ethnic disparities in influenza hospitalization should consider a wider
array of states to determine the magnitude of spatial variation in influenza hospitalization and a

53

longer time series to better understand the temporal trends. Furthermore, more detailed racial and
ethnic demographic information for admitted individuals may increase our comprehension of
state-level differences that may be explained by racial or ethnic subgroups.

54

Chapter 4 The role of demographics and socioeconomics in the timing of
influenza-related hospital utilization

Despite evidence for the role of demographics and socioeconomic status on the magnitude of
influenza outbreaks, the timing of outbreaks is generally attributed to environmental conditions
such as humidity. The purpose of this study is to explore the impact of race/ethnicity, age, and
insurance type on the odds of early hospital utilization in two Arizona cities. This case study was
performed for the 2008-09 through the 2016-17 influenza seasons in Mesa and Tucson. City-
stratified logistic regressions showed that there were distinct patterns within and between the two
cities. While Mesa had statistically significant relationships between early hospital utilization
and race, there was no statistically significant difference in the coefficients between cities. There
were, however, significant differences between the cities in the age and insurance type
categories. Compared to 5-17 year olds, individuals aged 18-49 years, 50-64 years, and 65 and
older in Tucson had statistically significantly higher odds ratios for early hospitalization
compared to the same groups in Mesa (18-49 years: Chi-squared=10.50, p-value=0.001; 50-64
years: Chi-squared=5.89, p-value=0.015; over 65: Chi-squared=7.68, p-value=0.006).
Additionally, the odds of early hospital utilization for Medicare recipients in Tucson were
statistically significantly higher than for Medicare recipients in Mesa (chi-squared=6.33, p-
value=0.012). Understanding the role of underlying population characteristics on the timing of
influenza outbreaks can support public health officials in creating more targeted vaccination
campaigns and in turn reduce the overall burden of influenza epidemics on the population.

55

4.1. Introduction
The Centers for Disease Control (CDC) estimates that every year, millions of people in the
United States are infected with the contagious respiratory illness known as the influenza or flu
virus. Despite the high morbidity of influenza as well as the mortality risks it poses for
vulnerable individuals, our understanding of the fundamental transmission dynamics of the
disease are still not well understood (Killingley and Nguyen-Van-Tam 2013). In temperate
regions of the world, like the United States, influenza typically begins in the fall and can peak as
late as the spring months (CDC 2021a), but the timing of both the onset and peak of seasonal
epidemics can vary by weeks or months each year. Anticipating the timing of these events can
aid public health workers in planning vaccine campaigns ahead of yearly outbreaks as well as
helping hospitals and doctors’ offices allocate personnel.
A variety of factors increase our susceptibility of contracting influenza. Adults over 65,
young children, and people with asthma, heart disease, diabetes, HIV/AIDS, and cancer are
considered high risk individuals (CDC 2021a). Additionally, research across disciplines has
shown increasing evidence for the impacts of local climate (Dalziel et al. 2018, Shaman et al.
2010, Towers et al. 2013, Lowen et al. 2007, Barreca and Shimshack 2012, Soebiyanto and
Kiang 2014, Grais and Ellis 2004, Lowen & Steel 2014, Serman et al. 2022), network and
mobility functions (Dalziel et al. 2018, Pei et al. 2018, Maliszewski and Wei 2011, Grais and
Ellis 2004, Chao et al. 2010, Gao et al. 2015), demographic and socioeconomic status (CDC
2021b, Thompson et al. 2011, Placzek and Madoff 2014, Sloan et al. 2015, Gounder et al. 2014,
Maliszewski and Wei 2011, Suryaprasad et al. 2013, Truelove et al. 2010, Kwan-Gett et al. 2009,
Kumar et al. 2015, Wegner and Naumova 2011, Maliszewski and Wei 2011, Ponnambalam et al.
2011, Sebastiani et al. 2006, O’Halloran et al. 2021), as well as the virus’s antigenic

56

characteristics (Du et al. 2017, Towers et al. 2013) on our physical vulnerability, our prevention
behaviors, and our perception of risk. Despite the range of contributing factors to an influenza
infection, research on the timing of seasonal outbreaks has generally been focused on the role of
climatic conditions, while demographic and socioeconomic have been used to better understand
disparities in hospitalizations and outbreak magnitude. While both provide information that can
be useful for the health care sector, this analysis will attempt to determine the role of
demographic and socioeconomic measures in the timing of seasonal outbreaks.
4.1.1. Humidity and influenza
The seasonality of influenza has been linked to seasonal climate changes. There are several
hypotheses about what is influencing the transmission cycle including: changes in host immune
capability, melatonin and vitamin D levels in winter, behavior (eg: staying indoors and the effect
of HVAC systems in winter), as well as environmental factors like temperature, humidity, and
UV irradiation (Lowen & Steel 2014). Evidence supporting the relationship between humidity
and influenza has been strengthened by laboratory studies (Kudo et al. 2019, Lowen et al. 2007,
Lowen & Steel 2014), resulting in three leading theories as to why the two may be closely
linked. They are as follows: 1) virus survival increases as humidity levels decrease; 2) droplet
size decreases with decreasing humidity, allowing particles to travel farther; and 3) low humidity
dries out the mucous membranes of the nose leaving the subject more susceptible to infection
(Lowen et al 2007). More recent work by Kudo et al. (2019) showed that mice kept in low
relative humidity conditions had difficulties in mucociliary clearance, impaired tissue repair in
their airways, and even produced lower amounts of interferon-stimulating genes that blocked
virus spread compared to mice kept in high relative humidity conditions.

57

Several studies have found that humidity improves models that predict influenza
outbreaks (Barreca and Shimshack 2012, Dalziel et al. 2018, Shaman et al. 2010, Shaman et al.
2017, Soebiyanto and Kiang 2014, Tamerius et al. 2019) and that relationships between humidity
and influenza may be relative to local climatic conditions (Barreca and Shimshack 2012, Dalziel
et al. 2018, Serman et al. 2022, Shaman et al. 2010, Soebiyanto and Kiang 2014, Tamerius et al.
2019). For example, Barreca and Shimshack (2012) considered 30 years of historic mortality
data from approximately 1972-2002 for 359 “urban” counties; counties with more than 100,000
residents. They found that the strength of the association with humidity and temperature
depended on the counties’ average humidity conditions over the course of the year. In counties
that were considered to have “high” average humidity during the year, associations were stronger
between mortality and low winter humidity. In “low” average humidity counties and those that
were generally colder, mortality was more strongly correlated with temperature. Furthermore,
Serman et al. (2022) identified state specific humidity thresholds that preceded increases in
influenza activity that were closely related to the average annual humidity of each state. Though
these and other studies find strong correlations between environmental conditions and influenza
activity, most list socioeconomic and demographic factors, as well as the impact of the virus
strain, as likely to be important in the spatial variation of seasonal outbreaks.
4.1.2. Demographics and socioeconomic status and influenza
During the 2009-10 H1N1 “swine flu” pandemic a variety of studies were published
demonstrating that minority populations across the country were at an increased risk of
influenza-related hospitalization compared to White non-Hispanics (Maliszewski and Wei 2011,
Truelove et al 2010, Kwan-Gett et al 2009, Suryaprasad et al 2013, Placzek and Madoff 2014,
Thompson et al 2011). Studies outside the context of the pandemic have offered further evidence

58

of increased risk of hospitalization for minority populations compared to White non-Hispanics
(Chandrasekhar et al. 2017, Gounder et al. 2014, Hadler et al. 2016, O’Halloran et al. 2021,
Sloan et al. 2015, Tam et al. 2014). When compared to each other these studies show variation in
which minority populations are over-represented in influenza hospitalizations, suggesting that
additional factors, such as socioeconomic status, may also play a role.
Studies that have considered the effects of socioeconomic status on influenza
hospitalizations have included measures such as percent of the population living below poverty
(Placzek and Madoff 2014, Sloan et al. 2015, Chandrasekhar et al. 2017, Hadler et al. 2016, Tam
et al. 2014, Kumar et al. 2015), median house hold income (Chandrasekhar et al. 2017, Tam et
al. 2014, Thompson et al. 2011, Sloan et al. 2015), percent of households with a female head
(Chandrasekhar et al. 2017, Sloan et al. 2015), household crowding (Sloan et al. 2015, Tam et al.
2014), education levels (Sloan et al. 2015, Tam et al. 2014), and insurance and employment rates
(Sloan et al. 2015) have all been used to assess the impact of socioeconomic status on influenza
hospitalization. Regardless of the measure used, the studies considered here all generally agreed
that as socioeconomic status decreases, hospitalization rates increase (Thompson et al. 2011,
Sloan et al. 2015, Placzek and Madoff 2014, Chandrasekhar et al. 2017, Hadler et al. 2016, Tam
et al. 2014). All studies found that census tracts with the highest rates of poverty had increased
rates of hospitalization (Thompson et al. 2011, Sloan et al. 2015, Chandrasekhar et al. 2017,
Hadler et al. 2016, Tam et al. 2014) and Hadler et al. (2016) emphasized that this disparity was
present within all race and age groups but that minorities still fared worse than non-Hispanic
Whites. Census tracts with low education rates (Sloan et al. 2015, Tam et al. 2014), and high
percentages of female headed households (Chandrasekhar et al. 2017, Sloan et al. 2015) were
also found to have higher rates of influenza hospitalization.

59

It is well understood that living in low socioeconomic areas can increase one’s risks for
poor health outcomes (Krieger et al. 2015). In addition to having generally poor health outcomes,
access to health care, health care affordability, individual preventative behaviors (e.g.:
handwashing), and beliefs about the efficacy and safety of vaccinations are all influenced by
socioeconomic status (Lu et al. 2017, Athamneh and Sansgiry 2014, Herbert et al 2005, Krieger
et al. 2015, Redelings et al. 2012). Though annual vaccination is considered to be one of the best
methods of prevention for influenza, disparities in vaccination rates have been documented in
various demographic and socioeconomic groups (Lu et al 2017, Athamneh and Sansgiry 2014,
Herbert et al 2005, Tam et al. 2014). The CDC reported that for the 2019-20 influenza season,
only 38% of Hispanics and 41% of Black non-Hispanics received influenza vaccines compared
to 52% and 53% of Asian/Pacific Islanders and White non-Hispanics, respectively. Tam et al.
(2014) found that vaccination was strongly associated with age and poverty level and that as
poverty increased vaccination rates decreased.
While the studies summarized here describe racial, ethnic, and socioeconomic disparities
in influenza hospitalizations and deaths, few have used these covariates in predictive models. Of
the studies that did attempt to make predictions, all of them focused on the H1N1 pandemic
(Kumar et al. 2015, Ponnambalam et al. 2011, Maliszewski and Wei 2011) and two of the three
(Ponnambalam et al. 2011, Maliszewski and Wei 2011) focused solely on magnitude and not
timing. Kumar et al. (2015) is the exception, creating an agent-based model to test whether
population structure could account for the differences in attack rates seen during the H1N1
pandemic. In this context, population structure meant accounting for population density, age
structure, average household size, and contact rates (Kumar et al. 2015). Interestingly, Kumar et
al. (2015) found that census tracts with high levels of poverty saw earlier and steeper increases in

60

hospitalizations than areas that were less impoverished and that this matched the hospital data
from the pandemic event. They also found that when individual susceptibility and household
income were modeled to be inversely proportional, the attack rates in their simulations were
comparable to the attack rates seen during the pandemic, suggesting that individual poverty
could account for area-level inequalities (Kumar et al. 2015).
4.1.3. Study aims
Given the number of contributing factors on influenza transmission, additional research on the
impact of underlying population characteristics on the timing of seasonal outbreaks is needed.
This study aims to contribute to the body of knowledge by performing a case study on the impact
of demographic and socioeconomic characteristics on early hospital utilizations during seasonal
influenza outbreaks in two Arizona cities. This study will focus on the cities Mesa and Tucson,
located in south-central and southern Arizona respectively, for the 2008-09 through the 2016-17
influenza seasons. Mesa and Tucson were selected for this case study because they have distinct
differences in the timing of their influenza-related hospital utilization despite having very similar
climate conditions. Over the course of the nine seasons, Tucson had a 5% higher hospital
utilization early in the season compared to Mesa (Table 4.1). This study aims to evaluate the role
of underlying population characteristics that may explain the differences in the timing of hospital
visits between the two cities. Mesa and Tucson have similar age structures, though Mesa has a
slightly larger population of individuals over 65 and Tucson has more individuals in the 18-49
year category (Table 4.1). Additionally, while they have comparable Asian/Pacific Islander,
Black non-Hispanic, and Native American/Alaskan Native populations, Tucson has a
substantially larger Hispanic population than Mesa (Table 4.1). This study used logistic
regression to determine the probability of subgroups of the population stratified by

61

race/ethnicity, age, and insurance type of presenting at the hospital during the earliest months of
the influenza outbreaks compared to later in the season. Given the evidence for links between
humidity and the timing of seasonal influenza outbreaks, we controlled for humidity using a
state- and seasonality-specific humidity measure derived from thresholds published in Serman et
al. (2022). Results from this study will increase our understanding of populations at higher risk
of early hospital utilization which can in turn be used to develop more targeted intervention
strategies such as vaccination campaigns.

Table 4.1: Distribution of timing of hospital utilizations and demographic profiles of Mesa and
Tucson. Race and age category data is an average annual number calculated from American
Community Survey 5-year estimates that span the analysis period.

Mesa Tucson
Timing n % n %

Early 1,263 15.1% 1,768 20.3%

Remainder of season 7,077 84.9% 6,955 79.7%
Race n* % n* %

White non-Hispanic 325,746 66.2% 407,364 51.3%

Black non-Hispanic 15,960 3.2% 29,274 3.7%

Hispanic 129,523 26.3% 316,440 39.9%

Asian/Pacific Islander 11,029 2.2% 23,760 3.0%

Native American/Alaskan Native 10,031 2.0% 16,796 2.1%
Age Category n* % n* %

<5 34,582 6.9% 50,428 6.2%

5-17 87,126 17.4% 131,041 16.2%

18-49 209,563 41.8% 357,831 44.1%

50-64 86,450 17.2% 151,044 18.6%

+65 83,570 16.7% 120,400 14.9%

*Average annual number of persons in each category from ACS estimates

4.2. Data
In this analysis, influenza cases are estimated from the State Emergency Department Database
(SEDD) and the State Inpatient Database (SID) which are created and maintained by the Agency

62

for Healthcare Research and Quality (AHRQ), Healthcare Cost and Utilization Project (HCUP).
The analysis presented here considers data from the Arizona cities of Mesa and Tucson from the
time period of 2008-2017. This time range covers nine influenza seasons beginning with the
2008-2009 season through the 2016-2017 season, due to the nature of the northern hemisphere
season spanning portions of two calendar years. The SID and SEDD include a wide variety of
deidentified patient information including demographics such as race, age, and home zip code, as
well as pertinent medical information such as diagnosis codes, treatment details, and outcomes.
The SEDD includes all patients who visited the Emergency Department (ED) but were not
admitted into the hospital. All admitted patients, including those who originated from the ED are
included in the SID. Therefore, using the datasets in combination provides a complete picture of
hospital utilization during the time period.
Humidity data were downloaded from gridMET a gridded meteorological data set
produced and maintained by the Climatology Lab at the University of California Merced. The
gridded data are available daily, dating from 1979 to present, at a spatial resolution of ~4km or
1/24th degree. It combines meteorological data from several remote sensing sources and is
validated against multiple ground station networks (Abatzoglou 2013). While several climate
variables are available from gridMET, this study utilizes their specific humidity data. This
analysis follows the precedent of other studies (Dalziel et al. 2018, Tamerius et al. 2019, Serman
et al. 2022) in using specific humidity as a proxy for absolute humidity to model the relationship
between local humidity conditions and the timing of influenza outbreaks.
Race/ethnicity and age category summary statistics for Mesa and Tucson were compiled
using data from the American Community Survey 5-year estimates from 2011, 2015, and 2019 at
the ZCTA level.

63

4.3. Methods
First, we identified zip codes that designated Mesa and Tucson as their primary city (Appendix C
Table C.1) and then converted them to Zip Code Tabulation Areas (ZCTAs) using a 2019
crosswalk. All summary statistics and models were performed on ZCTAs aggregated by the city
they represented.
Daily specific humidity obtained from gridMET was first averaged by month. Then the
monthly averages for each ZCTA were calculated by finding the spatially weighted mean. In this
analysis, spatial weights are based on the proportion of the grid cells within the ZCTA. The
calculated average monthly humidity values for each ZCTA were matched to patients according
to their admission month and home ZCTA. As previously mentioned, this analysis controls for
humidity using a state and seasonality specific measure. In an earlier study, Serman et al. (2022)
published state specific humidity breakpoints, or thresholds, that preceded sharp increases in
influenza cases. The specific humidity threshold for the state of Arizona was 0.00404 kg/kg.
Assuming this threshold does precede an increase in influenza cases, one would expect that
individuals presenting at the hospital early in the season would have experienced humidity
conditions in their home zip code that were similar to the threshold. As such, the difference
between the threshold and the monthly average humidity during the month they visited the
hospital is used to control for humidity conditions in the model.
Patients were selected from the SID and SEDD for this analysis based on influenza-
specific diagnosis codes. The time period of this analysis encompasses the transition from the
International Disease Classification, Ninth Revision (ICD-9) to the International Disease
Classification, Tenth Revision (ICD-10). Therefore, this analysis includes both the ICD-9 and
ICD-10 codes for influenza, 487-488 and J09-J11 respectively. Patients with an influenza

64

diagnosis code anywhere within their records were included in this analysis, influenza did not
need to be their primary diagnosis. Patients who did not identify Arizona as their home state or
had missing state information were excluded, as were individuals with missing or “other” race
designations.
The race and ethnicity groups included in this analysis are: White Non-Hispanic, Black
Non-Hispanic, Asian/Pacific Islander, Native American/Alaskan Native, and Hispanic.
Individuals were grouped into five age categories: under 5, 5-17 years old, 18-49, 50-64, and 65
and over. The third independent variable, insurance type, is broken into the following five
categories: Medicare, Medicaid, Private, Self-Pay, and Other. The SID and SEDD insurance type
category “Other” identifies individuals who paid for medical care through government programs
such as Worker’s Compensation, Indian Health Services, and the Civilian Health and Medical
Program of the Department of Veteran’s Affairs (CHAMPVA). For the purpose of this analysis,
individuals who were listed as “Charity” in the SID and SEDD were included in the Other
category.
City-stratified logistic regressions were used to determine the probability of subgroups of
the population having an “early” hospital utilization compared to any other time in the season.
For the purpose of this analysis, an influenza season was defined by identifying the peak month
of influenza-related hospital visits, then including the two months prior to peak and the two
months following peak. This captured over 90% of influenza-related cases within the data set.
Next, an early utilization was defined as the two months prior to the peak month of influenza-
related ED visits and hospitalizations. Cases in the peak month and the two months following
peak make up the comparison group of the remainder of the season. The independent variables in

65

this model were difference from the state specific humidity breakpoint, race/ethnicity, age
category, and insurance type.
The coefficients from the resulting city-stratified models were compared using a chi-
squared test. The final results of the regressions and comparison tests can be found in Table 4.2.
A sensitivity test was conducted to assess the impact of individual seasons on the results, the
results of which can be found in Tables C.3 and C.4 in Appendix C.
4.4. Results

Figure 4.1 Comparison of the odds ratio of early hospital utilization for each race/ethnicity
category compared to White non-Hispanics. Odds ratios are displayed with 95% Confidence
Intervals and cities are differentiated with different color and dot shapes.

66

The results for the city-stratified logistic regression models revealed distinct differences in the
demographic and socioeconomic features that indicated higher odds of an early hospital
utilization. In Tucson, there were no statistically significant relationships between timing and
race when compared to White non-Hispanics (Figure 4.1). When compared to 5-17 year olds,
individuals aged 18-49 (OR=1.33, 95% CI[1.15, 1.53], p<0.000) and 65 and older (OR=1.64,
95% CI[1.22, 2.19], p=0.001) had statistically significantly higher odds of early hospital
utilization (Figure 4.2). Individuals on Medicaid also had higher odds of an early hospital
utilization compared to those with Private insurance coverage (OR=1.18, 95% CI[1.02, 1.36],
p=0.025) (Figure 4.3).

67

Figure 4.2: Comparison of the odds ratio of early hospital utilization for each age category
compared to children aged 5-17. Odds ratios are displayed with 95% Confidence Intervals and
cities are differentiated with different color and dot shapes.

In Mesa, Black non-Hispanics (OR=0.75, 95% CI[0.59, 0.96], p=0.023), Hispanic
(OR=0.84, 95% CI[0.72, 0.96], p-value=0.014), and Asian/Pacific Islander (OR=0.77, 95%
CI[0.54, 1.10], p=0.011) groups all had statistically significantly lower odds of having an early
hospital utilization compared to White non-Hispanics. In addition, the odds of children under 5
(OR=0.76, 95% CI[0.63,0.91], p=0.002) and individuals aged 50-64 (OR=0.74, 95% CI[0.57,
0.96], p-value=0.024) were statistically significantly lower for early hospital utilization
compared to children 5-17 years. Finally, Medicare patients had statistically significantly lower

68

odds of having an early hospital utilization compared to those on private insurance (OR=0.71,
95% CI[0.51, 1.00], p-value=0.049).

Figure 4.3 Comparison of the odds ratio of early hospital utilization for each insurance type
compared to individuals with Private insurance. Odds ratios are displayed with 95% Confidence
Intervals and cities are differentiated with different color and dot shapes.

As seen in Table 4.2, despite Mesa having statistically significant relationships between
early hospital utilization and the Black non-Hispanic, Hispanic, and Asian/Pacific Islander
race/ethnicity groups, there is no statistically significant difference in the coefficients between
cities (Black non-Hispanic: Chi-squared=2.38, p-value=0.123; Hispanic: Chi-squared=1.91, p-

69

value=0.167; Asian/Pacific Islander: Chi-squared=0.73, p-value=0.391). There were, however,
significant differences between the cities in the age and insurance type categories. Compared to
5-17 year olds, individuals aged 18-49 years, 50-64 years, and 65 and older in Tucson had
statistically significantly higher odds ratios for early hospitalization compared to the same groups
in Mesa (18-49 years: Chi-squared=10.50, p-value=0.001; 50-64 years: Chi-squared=5.89, p-
value=0.015; over 65: Chi-squared=7.68, p-value=0.006). Additionally, the odds of early
hospital utilization for Medicare recipients in Tucson were statistically significantly higher than
for Medicare recipients in Mesa (Chi-squared=6.33, p-value=0.012).

70

Table 4.2 Results of the city-stratified logistic regression models and chi-squared tests for
difference between city-stratified coefficients. P-values are marked with statistical significance
indicated with asterisks.
Variables OR Lower Upper p-value OR Lower Upper p-value
Chi
2
p-value
Difference from breakpoint 1.45 1.37 1.54 <0.000 *** 1.39 1.32 1.47 <0.000 *** 0.84 0.360
Race
White non-Hispanic reference - - - reference - - - reference -
Black non-Hispanic 0.75 0.59 0.96 0.023 * 0.98 0.78 1.24 0.857 2.38 0.123
Hispanic 0.84 0.72 0.96 0.014 * 0.95 0.85 1.07 0.408 1.91 0.167
Asian/Pacific Islander 0.52 0.32 0.86 0.011 * 0.70 0.46 1.08 0.105 0.73 0.391
Native American/Alaskan Native 0.77 0.54 1.10 0.144 0.84 0.63 1.12 0.235 0.14 0.706
Age category
under 5 0.76 0.63 0.91 0.002 ** 0.95 0.79 1.13 0.544 3.00 0.083
5-17 reference - - - reference - - - reference -
18-49 1.01 0.86 1.19 0.897 1.33 1.15 1.53 <0.000 *** 5.89 0.015 *
50-64 0.74 0.57 0.96 0.024 * 1.19 0.96 1.47 0.114 7.68 0.006 **
over 65 0.76 0.52 1.11 0.158 1.64 1.22 2.19 0.001 ** 10.50 0.001 **
Insurance type
Private reference - - - reference - - - reference -
Medicare 0.71 0.51 1.00 0.049 * 0.91 0.70 1.18 0.493 1.38 0.239
Medicaid 0.90 0.77 1.05 0.178 1.18 1.02 1.36 0.025 * 6.33 0.012 *
Self-pay 1.15 0.93 1.42 0.197 1.18 0.96 1.45 0.112 0.03 0.858
Other 0.93 0.62 1.40 0.733 0.88 0.67 1.16 0.36 0.05 0.819
Mesa Tucson Difference in
Coefficients
95% CIs 95% CIs

71

4.5. Discussion
Past work in influenza modeling has placed a heavy emphasis on the role of humidity in the
timing of seasonal influenza outbreaks. The purpose of this study was to determine whether
underlying population characteristics such as race/ethnicity, age, and socioeconomic status,
measured by insurance type, could account for differences in outbreak timing when controlling
for humidity. City-stratified models showed that there were distinct differences in the factors that
influenced outbreak timing in each city. However, when compared, the likelihood of an early
hospital utilization among adults 18 and older and individuals on Medicaid were statistically
significantly different in the two cities. The work presented here provides compelling evidence
for the role of underlying population characteristics, particularly age structure, in the onset of
seasonal influenza outbreaks while controlling for humidity. Differences in the city-specific
effect estimates suggests that there may be additional location specific factors that this paper has
not identified. Understanding demographic and socioeconomic drivers of early hospital
utilization can aid public health professionals in planning more targeted pre-season interventions
such as vaccination campaigns.
While there is substantial evidence for an overall disproportionate burden of influenza
hospitalizations on minority populations, results from this study did not suggest that this
disparity translates into the timing of influenza outbreaks. In contrast, minority populations in
Mesa were less likely than White non-Hispanics to have an early hospital utilization. Minority
populations in Tucson saw no statistically significant difference in risk compared to White non-
Hispanics. Age, on the other hand, did seem to play a role in early hospital utilizations. While we
know that very young children and adults aged 65 and older have a higher risk for severe cases

72

of influenza, there is no consensus on whether any one age group drives influenza outbreaks
(Worby et al. 2015).
Worby et al. (2015) considered hospitalization data from the Influenza Surveillance
Hospitalization Network (FluSurv-NET) for the 2009 H1N1 pandemic as well as the 2010-11
through 2013-14 seasons. FluSurv-NET collects data from over 80 counties in almost 20 states
that comprise of almost 10% of the nation’s population. Worby et al. (2015) found that age-
specific risk was impacted by the season’s dominant influenza (sub)type and that children aged
5-17 years had the highest relative risk for early hospitalization more frequently than the other
age groups during the study period and generally “lead” influenza outbreaks.
While this study and that of Worby et al. (2015) considered different outcomes
(hospitalization vs hospital utilization), we further considered general age-related trends in risk
as a comparison. Odds of early hospital utilization by age in Mesa were more similar to the
relative risks found by Worby et al. (2015). In Mesa, children under 5 and adults 50-64 had
statistically significantly lower odds of early hospital utilization compared to children aged 5-17
years. The age-stratified odds of early hospitalization in Tucson were statistically significantly
different from Mesa for the oldest three age groups and were statistically significantly higher
than for children 5-17 years. A potential theory for why Mesa has more similar trends to the
Worby et al. (2015) study could be that the population characteristics of the FluSurv-NET
sample is more similar to Mesa than to Tucson.
Kumar et al. (2015) found that census tracts with high levels of poverty had earlier and
steeper increases in influenza cases. While this study did not consider a direct measurement of
poverty, to qualify for Medicaid in the state of Arizona one must be considered to be low or very
low income and one of the following: be pregnant, responsible for a child under 18, or you or a

73

member of your household must have a disability (Arizona Medical Assistance Program 2022).
Tucson saw a statistically significantly higher probability of early hospital utilization by
Medicaid recipients suggesting a similar link to low income and early influenza to that found by
Kumar et al. (2015). The same was not seen for individuals on Medicaid in Mesa. However,
American Community Survey estimates Tucson has an overall lower median household income
and a higher percentage of people living in poverty compared to Mesa, which may suggest
additional neighborhood level influence that we have not accounted for.
The use of only two cities in this study for comparison is a limitation of this work. While
the models for each individual city are robust, the impacts of population on timing seen in these
two cities may not be generalizable to other locations. In addition to a larger comparison group,
future studies may also consider grouping cities by similar demographics and socioeconomic
conditions in addition to climate. For example, previous work suggests that additional features
such as population density or measures of crowding (Dalziel et al. 2018, Kumar et al. 2015) as
well as epidemic characteristics like influenza type and strain (Worby et al. 2105) may help to
further explain differences in outbreak timing by location. The focus on urban locations is also a
limitation of this analysis as there may be considerable rural/urban differences in the timing of
influenza outbreaks as well as the implementation of public health campaigns. This raises
important questions about public health equity across geographic locations and the focus on
urban locales for interventions and research.
Understanding the role of subgroups of the population in the timing of influenza
outbreaks can provide public health officials with information on how to target vaccine
campaigns more effectively. Worby et al. (2015) simulated the impact of vaccinating the leading
age group of seasonal outbreaks and found that it reduced the overall burden. As such, when

74

individual public health entities understand more completely who in their local community is at
risk for early hospital utilization for influenza, they can target vaccine campaigns to those
subgroups. Using the results from this study as an example, vaccine campaigns in Mesa might
focus on school-aged children (ages 5-17) more specifically while Tucson should target
communities with higher numbers of individuals aged 18-49 and 65 and older, such as college
campuses. Reducing the burden of seasonal influenza through targeted vaccination campaigns is
important because it not only reduces morbidity, but also influenza-related mortality. Population
characteristics provide important insights for influenza prediction models and should be
considered alongside environmental conditions, but their inclusion is even more important for
the implementation of public health interventions. As such, future research should strive to
thoughtfully include population characteristics in models that aim to both explain and predict
influenza outbreaks.

75

Chapter 5 Conclusion
Understanding influenza transmission dynamics and predicting influenza outbreaks is of great
interest to national and international health organizations. And, while advancements have been
made in modeling efforts, there is a lot we still do not understand about the role that local climate
(Dalziel et al. 2018, Shaman et al. 2010, Towers et al. 2013, Lowen et al. 2007, Barreca and
Shimshack 2012, Soebiyanto and Kiang 2014, Grais and Ellis 2004, Lowen & Steel 2014,
Serman et al. 2022), network and mobility functions (Dalziel et al. 2018, Pei et al. 2018,
Maliszewski and Wei 2011, Grais and Ellis 2004, Chao et al. 2010, Gao et al. 2015),
demographic and socioeconomic status (CDC 2021b, Thompson et al. 2011, Placzek and Madoff
2014, Sloan et al. 2015, Gounder et al. 2014, Maliszewski and Wei 2011, Suryaprasad et al.
2013, Truelove et al. 2010, Kwan-Gett et al. 2009, Kumar et al. 2015, Wegner and Naumova
2011, Maliszewski and Wei 2011, Ponnambalam et al. 2011, Sebastiani et al. 2006, O’Halloran
et al. 2021), as well as the virus’s antigenic characteristics (Du et al. 2017, Towers et al. 2013)
play in the timing and intensity of seasonal outbreaks.
Because of its complex nature, research on influenza outbreaks spans several disciplines
but the work is rarely interdisciplinary in nature. For example, research that addresses the
relationship between humidity and influenza does not typically include demographic or
socioeconomic information and generally only addresses the timing of seasonal outbreaks. On
the other hand, work focused on the demographic and socioeconomic characteristics of influenza
cases or hospitalizations typically does not consider the role of environmental conditions, in part
because they aim to explain the magnitude of outbreaks and not timing. Furthermore, both types
of papers generally have a limited spatial and temporal scope that does not provide an
opportunity to address the influence of variation in environmental and population characteristics

76

over space and time. As such, the work presented here attempts to address: 1) spatial variation in
the documented humidity-influenza relationship, 2) spatial and temporal differences in
racial/ethnic disparities influenza hospital admission rates, 3) the role of a population’s
demographic and socioeconomic characteristics in the timing of seasonal influenza outbreaks.
Understanding how relationships between influenza and environmental and population
characteristics vary over space and time has multiple benefits. First, it can aid public health
officials in creating more targeted intervention programs and show how effective current efforts
have been in reducing disparities. Second, it can provide additional context for models that aim
to predict seasonal outbreaks and make them more accurate. The multitude of factors that
influence influenza outbreaks increase the likelihood that the relationships we document may not
be generalizable to broad geographic contexts. Examples of this were seen in the differences in
disparate hospitalization rates of minority populations by state and rural/urban differences, both
documented in the H1N1 pandemic (Placzek and Madoff 2014, Kwan-Gett et al. 2009,
Thompson et al. 2011, Truelove et al. 2010), as well as in the results of the three studies
presented here.
The studies presented here contributed to the field first by quantifying explicit state level
humidity thresholds that seem to signal the onset of seasonal outbreaks. Interestingly, these
thresholds were closely related to the average annual humidity of each state. This presents
further evidence for the role of individual level adaptations to our environments and the role that
may play in influenza infections. Given these thresholds were determined using decadal trends,
future studies should build upon this work by testing the applicability of these thresholds on
single seasons. Additionally, since these thresholds were determined at the state level, and since
some states have considerable within state variation in climate, the spatial scale at which these

77

thresholds are applicable should also be tested. Another way to test the spatial scale of humidity
thresholds may be to determine similar thresholds at the city level. For example, in a large and
diverse state like California, one might identify and compare thresholds for a collection of cities
such as Los Angeles, San Francisco, San Diego, and Sacramento. Another consideration for
future work is the role of climate change. Humidity and temperature are closely related. As such,
it stands to reason that we will see changes in state level humidity profiles. As we continue to
adapt to our surrounds, the thresholds identified in this chapter may also change. Though current
data from Google Flu Trends are no longer publicly available, influenza estimates from other
sources might strive to establish updated thresholds for more recent time periods. For example,
while this work was being completed, JPL began a collaboration with the Los Angeles County
Department of Public Health. The AIRS team first became interested in influenza modeling as a
way to use data from satellite sensors in health applications. Now, they are working with
LACDPH to share data and modeling results. This collaboration not only provides JPL with data
to train and validate their models, it also provides a community health perspective on how the
models might be used. For example, how much lead time is needed from a prediction model to
be most useful for the public health entity. Interdisciplinary cooperation like this provides an
example of how teams across disciplines might come together to advance influenza modeling
capabilities.
Though the first paper of this dissertation deals only with the influenza-humidity
relationship, it provided inspiration on ways to bridge the gap between disciplines. But first, I
wanted to take a closer look at spatial variation between average hospital admission rates by
race/ethnicity and age in three states. The differences by state highlight the problems with
discussing racial and ethnic disparities in influenza hospitalizations at a national scale. The

78

results also suggested that ethnic subgroups may have different rates of influenza hospitalization,
identifying the need for more detailed demographic information to be a part of intake forms and
healthcare data. Furthermore, the study considered change in average admission rates over time,
something that has not been done by past work. This analysis noted that increases in disparate
rates also varied by race/ethnicity and by state with Black non-Hispanics generally seeing the
largest increases in disparities in the most age groups. Increases in disparities were also notably
higher in Florida than in Arizona and Maryland. These findings are important because they
highlight the need for considering local contexts when designing and implementing targeted
vaccine campaigns. Future studies should consider more states and longer time periods to better
understand the extent of the spatial and temporal variation of racial and ethnic disparities.
Furthermore, where it is available, similar studies should be carried out with more detailed racial
and ethnic data. A few states within the State Inpatient Databases (SID) used in this study have
this more detailed information and it should be utilized in an effort to better understand
disparities within racial and ethnic subgroups. Just as with interstate variation in climate, there is
also within state variation of population characteristics. As noted by Truelove et al. (2010), who
found that disparities in H1N1 hospitalizations in the Wisconsin city of Milwaukee were
different than in the rest of the state, the same may be true in other locations in non-pandemic
years. Again, understanding these disparities can inform the design of vaccine campaigns.
In the final paper I attempt to merge what we have learned about both environmental and
population characteristics to identify the role of each in the timing of influenza outbreaks. While
there is a distinct focus on humidity-driven influenza prediction models, there is some evidence
that age structure, in particular, may also play a role (Kumar et al. 2015, Worby et al. 2015). This
analysis compared the timing of seasonal outbreaks in two Arizona cities with similar humidity

79

profiles and found that the likelihood of an early hospital utilization among adults 18 and older
and individuals on Medicaid were statistically significantly different in the two cities. This
information may be helpful for local public health entities while planning influenza vaccination
campaigns. For example, Tucson may want to target adults 18-49 years and 65 and older while
Mesa may want to focus their efforts on school aged children (5-17 years). Having more targeted
vaccination campaigns creates an opportunity to reduce transmission among groups that “lead”
influenza outbreaks, therefore reducing the overall burden of the outbreak on the community
(Worby et al. 2015). A considerable limitation of this study is the number of population
characteristics that were included in the model. While age, race/ethnicity, and insurance type are
important metrics, there are likely additional city characteristics that were not captured by this
study. For example, this study does not include an explicit measure of socioeconomic conditions
within the cities. We know that Tucson has a considerably higher percentage of individuals
living in poverty than Mesa and a lower median household income and, given the considerable
evidence for higher poverty rates leading to increased risk of influenza hospitalization
(Thompson et al. 2011, Sloan et al. 2015, Chandrasekhar et al. 2017, Hadler et al. 2016, Tam et
al. 2014) and decreased likelihood of influenza vaccine uptake (Lu et al. 2017, Athamneh and
Sansgiry 2014, Herbert et al 2005, Krieger et al. 2015, Redelings et al. 2012), it seems to reason
that this may also play a role in the timing of one’s exposure to influenza. In addition, because
this study considers only two locations, it has limited generalizability. Future studies may want
to consider larger cities that will have more instances of influenza related hospital utilization and
potentially more robust patterns of hospital utilization. In addition, this study has an urban bias
and future work should also consider the impact of rural/urban conditions on the timing of
seasonal influenza outbreaks. Influenza research across disciplines has provided strong evidence

80

for a combination of factors which seem to contribute to both the timing and magnitude of
influenza outbreaks. As such, research on the subject should follow the example of JPL and
LACDPH and prioritize a more interdisciplinary and mutually beneficial approach. Furthermore,
researchers should consider the work that has already been done in broader spatial and temporal
contexts to address the generalizability of the national level trends that tend to dominate the
influenza narrative.

81

References
Abatzoglou, J.T. 2013. “Development of gridded surface meteorological data for ecological
applications and modelling.” International Journal of Climatology, 33(1). 121-131.
https://doi.org/10.1002/joc.3413
Agency for Healthcare Research and Quality. 2019. “Overview of the State Inpatient Databases.”
Healthcare Cost and Utilization Project (HCUP). Accessed January 28, 2020.
https://www.hcup-us.ahrq.gov/db/state/siddist/SID_Introduction.jsp
AIRS Science Team/Joao Teixeira. 2013. “AIRS/Aqua L3 Daily Standard Physical Retrieval
(AIRS-only) 1 degree x 1 degree V006”. Goddard Earth Sciences Data and Information
Services Center (GES DISC), Accessed June 2018, 10.5067/Aqua/AIRS/DATA303
Arizona Medical Assistance Program. 2022. “Who is eligible for Arizona Medical Assistance
Program.” Accessed March 16, 2022.
https://www.benefits.gov/benefit/1000#:~:text=To%20be%20eligible%20for%20Arizona,in
come%20or%20very%20low%20income.
Athamneh, L. N., & Sansgiry, S. S. 2014. “Influenza vaccination in patients with diabetes:
Disparities in prevalence between African Americans and Whites | Vacunación de la gripe
en pacientes con diabetes: Diferencia de prevalencia entre Afro-Americanos y blancos.”
Pharmacy Practice, 12(2). DOI: 10.4321/s1886-36552014000200008.
Barreca, A.I., & Shimshack, J.P. 2012. “Absolute Humidity, Temperature, and Influenza
Mortality: 30 Years of County-Level Evidence from the United States.” Am J Epidemiol,
176(Suppl), S114-S122. DOI: 10.1093/aje/kws259.
Brownstein, J.S., Wolfe C.J., & Mandl, K.D. 2006. “Empirical evidence for the effect of airline
travel on inter-regional influenza spread in the United States.” PLoS Med, 3(10): e401.
DOI: 10.1371/journal.pmed.0030401.
CDC. 2016. “Seasonal Influenza, More Information.” Accessed August 2018.
http://www.cdc.gov/flu/about/qa/disease.htm#people
Camuffo, D. 2014. “Theoretical Grounds for Humidity.” In Microclimate for Cultural Heritage:
Conservation, Restoration, and Maintenance of Indoor and Outdoor Monuments, 49-76.
Elsevier Science.
CDC. 2018. “How CDC Classifies Flu Severity.” Accessed August 2018.
https://www.cdc.gov/flu/about/classifies-flu-severity.htm
CDC. 2019. “About CDC’s Flu Forecasting Efforts.” Accessed May 2022.
https://www.cdc.gov/flu/weekly/flusight/about-flu-
forecasting.htm#:~:text=CDC's%20efforts%20with%20forecasting%20began,and%20intens
ity%20of%20flu%20season.

82

CDC. 2021a. “Key Facts About Influenza (Flu).” Accessed August 2021.
https://www.cdc.gov/flu/about/keyfacts.htm
CDC. 2021b. “Flu Disparities Among Racial and Ethnic Minority Groups.” Accessed August
2021. https://www.cdc.gov/flu/highrisk/disparities-racial-ethnic-minority-groups.html
CDC. 2021c. “Past Influenza Seasons.” Accessed August 2021.
https://www.cdc.gov/flu/season/past-flu-seasons.htm
CDC. 2022. “FluView Interactive.” Accessed August 2021.
https://www.cdc.gov/flu/about/classifies-flu-severity.htm
Chandrasekhar, R., Sloan, C., Mitchel, E., Ndi, D., Alden, N., Thomas, A. … Lindegren, M.L.
2017. “Social determinants of influenza hospitalization in the United States.” Influenza and
Other Respiratory Viruses, 11, 479–488. DOI: 10.1111/irv.12483.
Chao, D.L., Halloran, E., & Longini, Jr., I.M. 2010. “School opening dates predict influenza
A(H1N1) outbreaks in the United States.” Journal of Infectious Diseases, 202(6), 877-880.
DOI: 10.1086/655810.
Chattopadhyay, I., Kiciman, E., Elliot, J.W., Shaman, J.L., Rzhetsky, A. 2018. “Conjunction of
factors triggering waves of seasonal influenza.” eLife, 7, e30756. DOI:
10.7554/eLife.30756.
Dalziel, B.D., Kissler, S., Gog, J.R., Viboud, C., Bjornstad, O.N., Metcalf, C.J.E., Grenfell, B.T.
2018. “Urbanization and humidity shape the intensity of influenza epidemics in U.S. cities.”
Science, 362, 75–79. DOI: 10.1126/science.aat6030.
Du, X., King, A.A., Woods, R.J., & Pascual, M. 2017. “Evolution-informed forecasting of
seasonal influenza A(H3N2).” Sci Transl Med, 9(413), 1–28. DOI:
10.1126/scitranslmed.aan5325.
Gao, H., Wong, K.K., Zheteyeva, Y., Shi, J., Uzicanin, A., & Rainey, J.J. 2015. “Comparing
observed with predicted weekly influenza-like illness rates during the winter holiday break,
United States, 2004-2013.” PLoS ONE, 10(12): e0143791.
doi:10.1371/journal.pone.0143791.
Ginsberg, J., Mohebbi, M. H., Patel, R. S., Brammer, L., Smolinski, M. S., & Brilliant, L. 2009.
“Detecting influenza epidemics using search engine query data.” Nature, 457, 1012-1014.
DOI: https://doi.org/10.1038/nature07634.
Google. (n.d.). “Flu search activity (standard deviation from baseline).” Google Flu Trends
Estimates - Google Public Data Explorer.
https://www.google.com/publicdata/explore?ds=z3bsqef7ki44ac_&ctype=l&met_y=flu_ind
ex.

83

Gounder, P. P., Callinan, L. S., Holman, R. C., Cheng, P.-Y., Bruce, M. G., Redd, J. T., …
Hennessy, T. W. 2014. “Influenza hospitalizations among American Indian/Alaska Native
people and in the United States general population.” Open Forum Infectious Diseases, 1(1).
DOI: 10.1093/ofid/ofu031.
Grais, R.F., & Ellis, J.H. 2004. “Modeling the spread of annual influenza epidemics in the U.S.:
The potential role of air travel.” Health Care Management Science, 7(2), 127–134. DOI:
10.1023/b:hcms.0000020652.38181.da.
Hadler, J.L., Yousey-Hindes, K., Pérez, A., Anderson, E.J., Bargsten, M., Bohm, S.R., …
Chaves, S.S. (2016). Influenza-Related Hospitalizations and Poverty Levels — United
States, 2010–2012. MMWR Morb Mortal Wkly Rep, 65(5), 101–105.
DOI: http://dx.doi.org/10.15585/mmwr.mm6505a1
HCUP. 2021. State Inpatient Databases (SID). Healthcare Cost and Utilization Project (HCUP).
2008-2017. Agency for Healthcare Research and Quality, Rockville, MD. www.hcup-
us.ahrq.gov/sidoverview.jsp
HCUP. 2021. State Emergency Department Databases (SEDD). Healthcare Cost and Utilization
Project (HCUP). 2008-17. Agency for Healthcare Research and Quality, Rockville, MD.
www.hcup-us.ahrq.gov/seddoverview.jsp
Hebert, P. L., Frick, K. D., Kane, R. L., & Mcbean, A. M. 2005. “Racial Disparities The Causes
of Racial and Ethnic Differences in Influenza Vaccination Rates among Elderly Medicare
Beneficiaries.” Health Services Research, 40(2), 517–537. DOI: 10.1111/j.1475-
6773.2005.00370.x.
Kandula, S. & Shaman, J. 2019. “Reappraising the utility of Google Flu Trends.” PLoS Comput
Biol 15(8), e1007258. DOI: 10.1371/journal.pcbi.1007258.
Karl, T.R. & Koss, W.J. 1984. “Regional and National Monthly, Seasonal, and Annual
Temperature Weighted by Area, 1895-1983.” Historical Climatology, eSeries 4-3.
Killingley, B. & Nguyen-Van-Tam, J. 2013. “Routes of influenza transmission.” Influenza and
Other Respiratory Viruses, 7(Suppl. 2), 42-51. DOI:10.1111/irv.12080.
Krieger, N., Chen, J.T., Waterman, P.D., Rehkopf, D.H., & Subramanian, S.V. 2005. “Painting a
truer picture of US socioeconomic and racial/ethnic health inequalities: The public health
disparities geocoding project.” American Journal of Public Health, 95(2), 312–323. DOI:
10.2105/AJPH.2003.032482.
Kudo, E., Song, E., Yockey, L.J., Rakib, T., Wong, P.W., Homer, R.J., Iwasaki, A. 2019. “Low
ambient humidity impairs barrier function and innate resistance against influenza infection.”
PNAS, 116(22). e1007258. DOI: 10.1371/journal.pcbi.1007258.

84

Kumar, S., Piper, K., Galloway, D.D., Hadler, J.L., & Grefenstette, J.J. 2015. “Is population
structure sufficient to generate area-level inequalities in influenza rates? An examination
using agent-based models.” BMC Public Health, 15(947), 1–12. DOI: 10.1186/s12889-015-
2284-2.
Kwan-Gett, T. S., Baer, A., & Duchin, J. S. 2009. “Spring 2009 H1N1 influenza outbreak in
King Country Washington.” Disaster Medicine and Public Health Preparedness,
3(SUPPL.2). DOI: 10.1097/DMP.0b013e3181c6b818.
Lowen, A. C., Mubareka, S., Steel, J., Palese, P. 2007. “Influenza virus transmission is
dependent on relative humidity and temperature.” PLoS Pathog, 3(10), e151. DOI:
10.1371/journal.ppat.0030151.
Lowen, A. C., & Steel, J. 2014. “Roles of Humidity and Temperature in Shaping Influenza
Seasonality.” J. Virol. Journal of Virology, 88(14), 7692-7695. DOI:10.1128/jvi.03544-13.
Lu, D., Qiao, Y., Brown, N. E., & Wang, J. 2017. “Racial and ethnic disparities in influenza
vaccination among adults with chronic medical conditions vary by age in the United
States.” PLoS ONE. DOI: 10.1371/journal.pone.0169679.
Maliszewski, P. J., & Wei, R. 2011. “Ecological factors associated with pandemic influenza A
(H1N1) hospitalization rates in California, USA: A geospatial analysis.” Geospatial Health,
6(1), 95–105. DOI: 10.4081/gh.2011.161.
Muggeo, V.M.R. 2003. “Estimating regression models with unknown break-points.” Statistics in
Medicine, 22, 3055–3071. DOI: 10.1002/sim.1545
Muggeo, V.M.R. 2008. “segmented: An R package to fit regression models with broken-line
relationships.” RNews, 8(1), 20—25.
O’Halloran, A.C., Holstein, R., Daily Kirley, P., Alden, N.B., Yousey-Hindes, K., Anderson,
E.J., … Garg, S. 2021. “Rates of influenza-associated hospitalization, Intensive Care Unit
admission, and in-hospital death by race and ethnicity in the United States from 2009 to
2019.” JAMA Network Open, 4(8): e2121880. DOI:10.1001/jamanetworkopen.2021.21880.
Olson, D.R., Konty, K.J., Paladini, M., Viboud, C., Simonsen, L. 2013. “Reassessing Google Flu
Trends Data for Detection of Seasonal and Pandemic Influenza: A Comparative
Epidemiological Study at Three Geographic Scales.” PLoS Comput Biol, 9(10), e1003256.
DOI:10.1371/journal.pcbi.1003256.
Pei, S., Kandula, S., Yang, W., & Shaman, J. 2018. “Forecasting the spatial transmission of
influenza in the United States.” PNAS, 115(11). DOI: 10.1073/pnas.1708856115.
Pew Research Center. 2019. “Facts on Hispanics of Mexican origin in the United States, 2017.”
Accessed August 2021. https://www.pewresearch.org/hispanic/fact-sheet/u-s-hispanics-
facts-on-mexican-origin-latinos/

85

Pew Research Center. 2019. “Facts on Hispanics of Puerto Rican origin in the United States,
2017.” Accessed August 2021. https://www.pewresearch.org/hispanic/fact-sheet/u-s-
hispanics-facts-on-mexican-origin-latinos/
Placzek, H., & Madoff, L. 2014. “Effect of race/ethnicity and socioeconomic status on pandemic
H1N1-related outcomes in Massachusetts.” American Journal of Public Health, 104(1).
DOI: 10.2105/AJPH.2013.301626.
Ponnambalam, L., Samavedham, L., Lee, H.R., & Ho, C.S. 2011. “Understanding the
socioeconomic heterogeneity in healthcare in US counties: the effects of population density,
education and poverty on H1N1 pandemic mortality.” Epidemiol. Infect, 140, 803–813.
DOI:10.1017/S0950268811001464.
Redelings M.D., Piron, J., Smith, L.V., Chan, A., Heinzerling, J., Sanchez, K.M. … Kuo, T.
2012. “Knowledge, attitudes, and beliefs about seasonal influenza and H1N1 vaccinations
in a low-income, public health clinic population.” Vaccine, 30, 454–458. DOI:
10.1016/j.vaccine.2011.10.050.
Reed, C., Chaves, S.S., Kirley, P.D., Emerson, R., Aragon, D., Hancock, E.B., … Finelli, L.
2015. “Estimating influenza disease burden from population-based surveillance data in the
United States.” PLoS ONE, 10(3): e0118369. DOI: 10.1371/journal.pone.0118369.
Sebastiani, P., Mandl, K.D., Szolovits, P., Kohane, I.S., Ramoni, M.F. 2006. “A Bayesian
dynamic model for influenza surveillance.” Statistics in Medicine, 25, 1803-1816. DOI:
10.1002/sim.2566.
Serman, E., Thrastarson, H.Th., Franklin, M., Teixeira, J. 2022. “Spatial Variation in Humidity
and the Onset of Seasonal Influenza Across the Contiguous United States.” GeoHealth,
6(2). DOI: 10.1029/2021GH000469.
Shaman, J., Pitzer, V., Viboud, C., Grenfell, B. & Lipsitch, M. 2010. “Absolute Humidity and
the Seasonal Onset of Influenza in the Continental US.” PLoS Biology, 8(2).
DOI:10.1371/currents.rrn1138.
Shaman, J., Kandula, S., Yang, W., Karspeck, A. 2017. “The use of ambient humidity conditions
to improve influenza forecast.” PLoS Comput Biol, 13(11), e1005844. DOI:
10.1371/journal.pcbi.1005844.
Shaman, J., & Kohn, M. 2009. “Absolute humidity modulates influenza survival, transmission,
and seasonality.” Proceedings of the National Academy of Sciences, 106(9), 3243-3248.
DOI:10.1073/pnas.0806852106.
Sloan, C., Chandrasekhar, R., Mitchel, E., Schaffner, W., & Lindegren, M. L. 2015.
Socioeconomic disparities and influenza hospitalizations, Tennessee, USA. Emerging
Infectious Diseases, 21(9), 1602–1610. DOI: 10.3201/eid2109.141861.

86

Soebiyanto, R.P. & Kiang, R. 2014. “Meterorological parameters as perdictors for seasonal
influenza.” Geocarto International, 29(1), 39–47. DOI:
http://dx.doi.org/10.1080/10106049.2013.799717
Suryaprasad, A., Redd, J. T., Hancock, K., Branch, A., Steward-Clark, E., Katz, J. M., … Wohr,
M. 2013. “Severe acute respiratory infections caused by 2009 pandemic influenza A
(H1N1) among American Indians-southwestern United States, May 1-July 21, 2009.”
Influenza and Other Respiratory Viruses, 7(6), 1361–1369. DOI: 10.1111/irv.12123.
Tam, K., Yousey-Hindes, K., Hadler, J.L. 2014. “Influenza-related hospitalization of adults
associated with low census tract socioeconomic status and female sex in New Haven
County, Connecticut, 2007-2011.” Influenza and Other Respiratory Viruses, 8(3), 274–281.
DOI: 10.1111/irv.12231.
Tamerius, J., Uejio, C., Koss, J. 2019. “Seasonal characteristics of influenza vary regionally
across US.” PLoS ONE, 14(3): e0212511. DOI: 10.1371/journal.pone.0212511.
Thompson, D. L., Jungk, J., Hancock, E., Smelser, C., Landen, M., Nichols, M., … Sewell, M.
2011. “Risk factors for 2009 pandemic influenza A (H1N1)-related hospitalization and
death among racial/ethnic groups in New Mexico.” American Journal of Public Health,
101(9), 1776–1784. DOI: 10.2105/AJPH.2011.300223.
Tian, B., Fetzer, E. J., Kahn, B. H., Teixeira, J., Manning, E., Hearty, T. 2013. “Evaluating
CMIP5 Models using AIRS Tropospheric Air Temperature and Specific Humidity
Climatology.” J. Geophys. Res. Atmos., 118, 114– 134, DOI:10.1029/2012JD018607.
Tian, B., Manning, E., Fetzer, E., Olsen, E., Wong, S. 2017. AIRS/AMSU /HSB Version 6 Level
3 Product User Guide.
Towers, S., Chowell, G., Hameed, R., Jastrebski, M., Kahn, M., Meeks, J. … Harris, G. 2013.
“Climate change and influenza: the likelihood of early and severe influenza seasons
following warmer than average winters.” PLoS Curr., 5(1), DOI:
10.1371/currents.flu.3679b56a3a5313dc7c043fb944c6f138.
Truelove, A. S., Truelove, A.S., Druckenmiller, J. K., Heffernan, R. T., & Davis, J. P. 2010.
“Epidemiologic and clinical features among patients hospitalized in Wisconsin with 2009
H1N1 influenza A virus infections, april to august 2009.” Wisconsin Medical Journal,
109(4), 201–208. DOI: 10.1093/infdis/jiq117.

87

United States Department of Health and Human Services (US DHHS), Centers for Disease
Control and Prevention (CDC), National Center for Health Statistics (NCHS), Bridged-Race
Population Estimates, United States July 1st resident population by state, county, age, sex,
bridged-race, and Hispanic origin. Compiled from 1990-1999 bridged-race intercensal
population estimates (released by NCHS on 7/26/2004); revised bridged-race 2000-2009
intercensal population estimates (released by NCHS on 10/26/2012); and bridged-race
Vintage 2017 (2010-2017) postcensal population estimates (released by NCHS on
6/27/2018). Available on CDC WONDER Online Database. Accessed at
http://wonder.cdc.gov/bridged-race-v2017.html on Apr 6, 2020
United States Department of Health and Human Servies (US DHHS). 2021. “Minority
Population Profiles”. Office of Minority Health. Accessed August 2021.
https://minorityhealth.hhs.gov/omh/browse.aspx?lvl=2&lvlID=26
Wegner, J.B., & Naumova, E.N. 2011. “Seasonal synchronization of influenza in the United
States older adult population.” PloS ONE, 5(4): e10187. DOI:
10.1371/journal.pone.0010187.
Worby, C.J., Chaves, S.S., Wallinga, J., Lipstitch, M. … Goldstein, E. 2015. “On the relative
role of different age groups in influenza epidemics.” Epidemics, 13, 10-16. DOI:
10.1016/j.epidem.2015.04.003.

88

Appendix A Chapter 2 Supplemental Materials

Figure A.1 Average weekly humidity versus average weekly incidence for Alabama (2003-2015). The red
and green lines represent the results of the segmented regression, their intersection is the breakpoint.

89

Figure A.2 Average weekly humidity versus average weekly incidence for Arkansas (2003-2015). The
red and green lines represent the results of the segmented regression, their intersection is the breakpoint.

Figure A.3 Average weekly humidity versus average weekly incidence for Arizona (2003-2015). The red
and green lines represent the results of the segmented regression, their intersection is the breakpoint.

90

Figure A.4 Average weekly humidity versus average weekly incidence for California (2003-2015). The
red and green lines represent the results of the segmented regression, their intersection is the breakpoint.

Figure A.5 Average weekly humidity versus average weekly incidence for Colorado (2003-2015). The
red and green lines represent the results of the segmented regression, their intersection is the breakpoint.

91

Figure A.6 Average weekly humidity versus average weekly incidence for Connecticut (2003-2015). The
red and green lines represent the results of the segmented regression, their intersection is the breakpoint.

Figure A.7 Average weekly humidity versus average weekly incidence for Delaware (2003-2015). The
red and green lines represent the results of the segmented regression, their intersection is the breakpoint.

92

Figure A.8 Average weekly humidity versus average weekly incidence for Florida (2003-2015). The red
and green lines represent the results of the segmented regression, their intersection is the breakpoint.

Figure A.9 Average weekly humidity versus average weekly incidence for Georgia (2003-2015). The red
and green lines represent the results of the segmented regression, their intersection is the breakpoint.

93

Figure A.10 Average weekly humidity versus average weekly incidence for Iowa (2003-2015). The red
and green lines represent the results of the segmented regression, their intersection is the breakpoint.

Figure A.11 Average weekly humidity versus average weekly incidence for Idaho (2003-2015). The red
and green lines represent the results of the segmented regression, their intersection is the breakpoint.

94

Figure A.12 Average weekly humidity versus average weekly incidence for Illinois (2003-2015). The red
and green lines represent the results of the segmented regression, their intersection is the breakpoint.

Figure A.13 Average weekly humidity versus average weekly incidence for Indiana (2003-2015). The red
and green lines represent the results of the segmented regression, their intersection is the breakpoint.

95

Figure A.14 Average weekly humidity versus average weekly incidence for Kansas (2003-2015). The red
and green lines represent the results of the segmented regression, their intersection is the breakpoint.

Figure A.15 Average weekly humidity versus average weekly incidence for Kentucky (2003-2015). The
red and green lines represent the results of the segmented regression, their intersection is the breakpoint.

96

Figure A.16 Average weekly humidity versus average weekly incidence for Louisiana (2003-2015). The
red and green lines represent the results of the segmented regression, their intersection is the breakpoint.

97

Figure A.17 Average weekly humidity versus average weekly incidence for Massachusetts (2003-2015).
The red and green lines represent the results of the segmented regression, their intersection is the
breakpoint.

98

Figure A.18 Average weekly humidity versus average weekly incidence for Maryland (2003-2015). The
red and green lines represent the results of the segmented regression, their intersection is the breakpoint.

Figure A.19 Average weekly humidity versus average weekly incidence for Maine (2003-2015). The red
and green lines represent the results of the segmented regression, their intersection is the breakpoint.

99

Figure A.20 Average weekly humidity versus average weekly incidence for Michigan (2003-2015). The
red and green lines represent the results of the segmented regression, their intersection is the breakpoint.

Figure A.21 Average weekly humidity versus average weekly incidence for Minnesota (2003-2015). The
red and green lines represent the results of the segmented regression, their intersection is the breakpoint.

100

Figure A.22 Average weekly humidity versus average weekly incidence for Missouri (2003-2015). The
red and green lines represent the results of the segmented regression, their intersection is the breakpoint.

Figure A.2. Average weekly humidity versus average weekly incidence for Mississippi (2003-2015). The
red and green lines represent the results of the segmented regression, their intersection is the breakpoint.

101

Figure A.24 Average weekly humidity versus average weekly incidence for Montana (2003-2015). The
red and green lines represent the results of the segmented regression, their intersection is the breakpoint.

102

Figure A.25 Average weekly humidity versus average weekly incidence for North Carolina (2003-2015).
The red and green lines represent the results of the segmented regression, their intersection is the
breakpoint.

103

Figure A.26 Average weekly humidity versus average weekly incidence for North Dakota (2003-2015).
The red and green lines represent the results of the segmented regression, their intersection is the
breakpoint.

104

Figure A.27 Average weekly humidity versus average weekly incidence for Nebraska (2003-2015). The
red and green lines represent the results of the segmented regression, their intersection is the breakpoint.

105

Figure A.28 Average weekly humidity versus average weekly incidence for New Hampshire (2003-2015).
The red and green lines represent the results of the segmented regression, their intersection is the
breakpoint.

106

Figure A.29 Average weekly humidity versus average weekly incidence for New Jersey (2003-2015). The
red and green lines represent the results of the segmented regression, their intersection is the breakpoint.

107

Figure A.30 Average weekly humidity versus average weekly incidence for New Mexico (2003-2015).
The red and green lines represent the results of the segmented regression, their intersection is the
breakpoint.

108

Figure A.31 Average weekly humidity versus average weekly incidence for Nevada (2003-2015). The red
and green lines represent the results of the segmented regression, their intersection is the breakpoint.

Figure A.32 Average weekly humidity versus average weekly incidence for New York (2003-2015). The
red and green lines represent the results of the segmented regression, their intersection is the breakpoint.

109

Figure A.33 Average weekly humidity versus average weekly incidence for Ohio (2003-2015). The red
and green lines represent the results of the segmented regression, their intersection is the breakpoint.

Figure A.34 Average weekly humidity versus average weekly incidence for Oklahoma (2003-2015). The
red and green lines represent the results of the segmented regression, their intersection is the breakpoint.

110

Figure A.35 Average weekly humidity versus average weekly incidence for Oregon (2003-2015). The red
and green lines represent the results of the segmented regression, their intersection is the breakpoint.

111

Figure A.36 Average weekly humidity versus average weekly incidence for Pennsylvania (2003-2015).
The red and green lines represent the results of the segmented regression, their intersection is the
breakpoint.

112

Figure A.37 Average weekly humidity versus average weekly incidence for Rhode Island (2003-2015).
The red and green lines represent the results of the segmented regression, their intersection is the
breakpoint.

113

Figure A.38 Average weekly humidity versus average weekly incidence for South Carolina (2003-2015).
The red and green lines represent the results of the segmented regression, their intersection is the
breakpoint.

114

Figure A.39 Average weekly humidity versus average weekly incidence for South Dakota (2003-2015).
The red and green lines represent the results of the segmented regression, their intersection is the
breakpoint.

115

Figure A.40 Average weekly humidity versus average weekly incidence for Tennessee (2003-2015). The
red and green lines represent the results of the segmented regression, their intersection is the breakpoint.

Figure A.41 Average weekly humidity versus average weekly incidence for Texas (2003-2015). The red
and green lines represent the results of the segmented regression, their intersection is the breakpoint.

116

Figure A.42 Average weekly humidity versus average weekly incidence for Utah (2003-2015). The red
and green lines represent the results of the segmented regression, their intersection is the breakpoint.

Figure A.43 Average weekly humidity versus average weekly incidence for Virginia (2003-2015). The
red and green lines represent the results of the segmented regression, their intersection is the breakpoint.

117

Figure A.44 Average weekly humidity versus average weekly incidence for Vermont (2003-2015). The
red and green lines represent the results of the segmented regression, their intersection is the breakpoint.

118

Figure A.45 Average weekly humidity versus average weekly incidence for Washington (2003-2015).
The red and green lines represent the results of the segmented regression, their intersection is the
breakpoint.

Figure A.46 Average weekly humidity versus average weekly incidence for Wisconsin (2003-2015). The
red and green lines represent the results of the segmented regression, their intersection is the breakpoint.

119

Figure A.47 Average weekly humidity versus average weekly incidence for West Virginia (2003-2015).
The red and green lines represent the results of the segmented regression, their intersection is the
breakpoint.

120

Figure A.48 Average weekly humidity versus average weekly incidence for Wyoming (2003-2015). The
red and green lines represent the results of the segmented regression, their intersection is the breakpoint.

121

Figure A.49 The resulting 95% confidence intervals of the Tukey HSD test on the differences in mean
breakpoints by region are displayed here with significant pairwise combinations displayed in red.

122

Figure A.50 This plot displays the beta 1 slopes from the segmented regression. In the A.1-A.48 plots this
is the red line. Though Indiana is a clear outlier in this plot, Indiana’s regression fit is very high
(R2=0.973), and there is little variation between the other states. We theorize two possible explanations
for this result. First, from a transmission perspective, once there is enough community spread, and the
epidemic is set in motion, environmental conditions play a smaller role. Second, given the nature of the
data being weekly, there are a limited number of data points in the epidemic portion of the curve for a
large portion of the states. With this in mind, we argue that it may not be appropriate to make conclusions
from the individual slope values, though they don’t vary significantly from one another.

123

Figure A.51 This scatter plot displays the values used in Figure A.50. The y-axis is the beta 1 slope values
from the segmented regression (depicted with a red line in the A.1-A.48 figures) and the x-axis is
humidity. This plot shows that, although Indiana is an outlier, there is little variation in the other data
points.

124

Figure A.52 This plot displays the week of the year that corresponds to the breakpoint for each of the
segmented regressions shown in A.1-A.48. In other words, it depicts the average timing of influenza
outbreaks. The approximate week of year that we see the breakpoint has a range of 6 weeks, with 25
states having a breakpoint in week 48 or 49. The standard deviation for the week of year that the
breakpoint fell on was 1.4 and the coefficient of variation 2.9% indicating a very narrow differentiation
between states. There do seem to be some potential regional similarities, or at least similarities with states
that border or may have a lot of movement between them (NY and CT, PA/MD/VA/NJ). There don’t
seem to be any patterns related to average annual humidity conditions specifically, however there may be
other associations related to movement/population dynamics that are outside the scope of this paper.

125

Figure A.53 This plot displays the values presented in figure A.52. The week of the year that corresponds
to the breakpoint for each of the segmented regressions shown in A.1-A.48 are on the y-axis and the
breakpoint humidity value for that state are on the x-axis.

126

Appendix B Chapter 3 Supplemental Materials

Table B.1 Average hospital admission rate ratios for each state, time period, age and
race/ethnicity category

Arizona

Maryland

Florida

2008-2013 2013-2017

2008-2013 2013-2017

2008-2013 2013-2017

Race/
Ethnicity*

Average Average

Average Average

Average Average
under 5
BNH

1.25 1.26

1.49 1.87

2.04 2.26
Hispanic

1.00 1.06

1.06 0.95

1.77 1.70
Asian

0.81 0.78

0.76 0.82

0.49 0.53
NA/AN
2.17 2.28 - - - -
5-17 years
BNH
1.58 1.55 1.52 1.71 1.49 1.92
Hispanic

0.72 0.89

- -

1.23 1.30
Asian

0.93 0.73

- -

0.45 0.55
NA/AN
0.87 1.47 - - - -
18-49 years
BNH
1.41 1.52 1.90 2.08 1.75 1.92
Hispanic

0.91 0.88

0.71 0.95

0.99 0.94
Asian

0.65 0.53

0.21 0.44

0.25 0.30
NA/AN
2.05 1.57 - - - -
50-64 years
BNH
2.38 1.59 1.81 1.96 1.88 1.87
Hispanic

1.27 1.03

0.75 0.81

1.14 0.99
Asian

0.45 0.62

0.30 0.28

0.36 0.31
NA/AN
1.88 1.85 - - - -
65 and older
BNH

1.89 1.41

0.90 0.95

1.22 1.29
Hispanic

1.26 1.14

0.73 0.63

1.45 1.29
Asian

1.08 1.39

0.51 0.55

0.57 0.45
NA/AN

1.99 1.25

- -

- -

*BNH = Black non-Hispanic, Asian = Asian/Pacific Islander, NA/AN = Native American/Alaskan Native

127

Appendix C Chapter 4 Supplemental Materials
Table C.1 Full list of zip codes used in this analysis and their associated Zip Code Tabulation
Area (ZCTA) for both Mesa and Tucson.

Mesa Tucson
zip code ZCTA zip code ZCTA

zip code ZCTA

zip code ZCTA
85201 85201 85701 85701

85721 85719

85746 85746
85202 85202 85702 85701

85722 85719

85747 85747
85203 85203 85703 85705

85723 85723

85748 85748
85204 85204 85705 85705

85724 85724

85749 85749
85205 85205 85706 85706

85725 85713

85750 85750
85206 85206 85707 85707

85726 85726

85751 85715
85207 85207 85708 85708

85728 85718

85752 85741
85208 85208 85709 85745

85730 85730

85754 85745
85209 85209 85710 85710

85731 85710

85756 85756
85210 85210 85711 85711

85732 85711

85757 85757
85211 85201 85712 85712

85733 85719

85775 85701
85212 85212 85713 85713

85734 85756

85213 85213 85714 85714

85735 85735

85214 85213 85715 85715

85736 85736

85215 85215 85716 85716

85740 85704

85216 85208 85717 85719

85741 85741

85274 85202 85718 85718

85743 85743

85275 85213 85719 85719

85744 85747

85277 85215 85720 85701

85745 85745

128

Figure C.1 The map on the left depicts the state of Arizona with several cities and major
roadways for orientation. Detailed maps on the right display the ZCTAs used for the Mesa and
Tucson study areas.

129

Table C.2 Full summary statistics of race, age category, and insurance type for each city.
Individuals were identified from both the SID and the SEDD and the proportion of individuals
from each data set as well as their combined contributions to the overall numbers are listed here.

Race
n
% of
combined
n
% of
combined
n
% of
overall
n
% of
combined
n
% of
combined
n
% of
overall
White 2694 71.2% 1091 28.8% 3785 43.4% 3078 70.8% 1268 29.2% 4346 52.1%
Black 444 85.9% 73 14.1% 517 5.9% 553 86.3% 88 13.7% 641 7.7%
Hispanic 3367 86.2% 537 13.8% 3904 44.8% 2515 87.4% 363 12.6% 2878 34.5%
Asian 134 79.8% 34 20.2% 168 1.9% 148 81.3% 34 18.7% 182 2.2%
NA/AN 286 81.9% 63 18.1% 349 4.0% 242 82.6% 51 17.4% 293 3.5%
Age Category
<5 1308 84.3% 243 15.7% 1551 17.8% 1770 87.2% 259 12.8% 2029 24.3%
5-17 1754 92.5% 142 7.5% 1896 21.7% 1766 94.6% 101 5.4% 1867 22.4%
18-49 2952 87.8% 412 12.2% 3364 38.6% 2213 84.6% 402 15.4% 2615 31.4%
50-64 566 62.3% 342 37.7% 908 10.4% 439 57.5% 324 42.5% 763 9.1%
+65 345 34.4% 659 65.6% 1004 11.5% 348 32.6% 718 67.4% 1066 12.8%
Insurance Type
Medicare 444 36.5% 771 63.5% 1215 13.9% 434 34.3% 831 65.7% 1265 15.2%
Medicaid 3775 86.7% 579 13.3% 4354 49.9% 3541 87.8% 490 12.2% 4031 48.3%
Private 1554 82.9% 321 17.1% 1875 21.5% 1567 80.4% 382 19.6% 1949 23.4%
Self Pay 767 93.2% 56 6.8% 823 9.4% 839 94.2% 52 5.8% 891 10.7%
Other 385 84.4% 71 15.6% 456 5.2% 155 76.4% 48 23.6% 203 2.4%
Tucson Mesa
SEDD SID Combined SEDD SID Combined

130

Table C.3 Results of the sensitivity analysis for Mesa. Each column shows the model results
when the season at the top of the column is removed from the model. Table is divided between
two pages for readability.

Season Removed
OR p-value OR p-value OR p-value OR p-value OR p-value
Difference from breakpoint 2.05 <0.000 1.22 <0.000 1.68 <0.000 1.48 <0.000 1.33 <0.000
Race
White non-Hispanic ref - ref - ref - ref - ref -
Black non-Hispanic 0.68 0.007 0.73 0.018 0.70 0.010 0.80 0.075 0.78 0.060
Hispanic 0.77 0.002 0.81 0.006 0.85 0.038 0.86 0.038 0.85 0.038
Asian/Pacific Islander 0.39 0.005 0.57 0.027 0.51 0.018 0.55 0.019 0.54 0.019
Native American/Alaskan
Native 0.81 0.294 0.76 0.131 0.58 0.013 0.78 0.190 0.82 0.282
Age category
under 5 0.78 0.024 0.78 0.009 0.79 0.025 0.77 0.005 0.76 0.004
5-17 ref - ref - ref - ref - ref -
18-49 1.24 0.024 1.02 0.831 1.06 0.552 0.98 0.789 1.02 0.824
50-64 0.88 0.371 0.68 0.005 0.85 0.252 0.71 0.010 0.77 0.060
over 65 0.87 0.490 0.67 0.043 0.81 0.315 0.75 0.149 0.84 0.398
Insurance type
Private ref - ref - ref - ref - ref -
Medicare 0.73 0.091 0.77 0.129 0.73 0.092 0.71 0.052 0.70 0.057
Medicaid 0.86 0.084 0.91 0.271 0.92 0.356 0.89 0.156 0.96 0.602
Self-pay 1.18 0.184 1.15 0.217 1.10 0.426 1.12 0.320 1.27 0.040
Other 0.90 0.677 1.03 0.896 0.82 0.418 1.01 0.967 1.03 0.882
08-09 09-10 10-11 11-12 12-13
Mesa

131

Season Removed
OR p-value OR p-value OR p-value OR p-value OR p-value
Difference from breakpoint 1.34 <0.000 1.74 <0.000 1.20 <0.000 1.40 <0.000 1.45 <0.000
Race
White non-Hispanic ref - ref - ref - ref - ref -
Black non-Hispanic 0.75 0.025 0.80 0.088 0.80 0.100 0.74 0.024 0.75 0.023
Hispanic 0.84 0.021 0.83 0.018 0.89 0.159 0.81 0.007 0.84 0.014
Asian/Pacific Islander 0.48 0.007 0.53 0.017 0.56 0.037 0.58 0.039 0.52 0.011
Native American/Alaskan
Native 0.77 0.156 0.85 0.401 0.73 0.131 0.81 0.266 0.77 0.144
Age category
under 5 0.75 0.006 0.77 0.007 0.65 <0.000 0.74 0.001 0.76 0.002
5-17 ref - ref - ref - ref - ref -
18-49 1.01 0.871 0.93 0.439 0.95 0.594 0.98 0.808 1.01 0.897
50-64 0.71 0.014 0.69 0.008 0.73 0.032 0.74 0.029 0.74 0.024
over 65 0.73 0.109 0.79 0.252 0.85 0.462 0.62 0.024 0.76 0.158
Insurance type
Private ref - ref - ref - ref - ref -
Medicare 0.69 0.038 0.70 0.055 0.64 0.027 0.77 0.147 0.71 0.049
Medicaid 0.89 0.137 0.92 0.331 0.87 0.102 0.89 0.175 0.90 0.178
Self-pay 1.11 0.336 1.13 0.298 1.13 0.304 1.16 0.172 1.15 0.197
Other 0.92 0.701 0.87 0.540 0.85 0.489 0.92 0.708 0.93 0.733
Mesa
13-14 14-15 15-16 16-17 Full Model

132

Table C.4 Results of the sensitivity analysis for Tucson. Each column shows the model results
when the season at the top of the column is removed from the model. Table is divided between
two pages for readability.

Season Removed
OR p-value OR p-value OR p-value OR p-value OR p-value
Difference from breakpoint 1.76 <0.000 1.44 <0.000 1.83 <0.000 1.38 <0.000 1.28 <0.000
Race
White non-Hispanic ref - ref - ref - ref - ref -
Black non-Hispanic 0.97 0.834 0.91 0.449 1.22 0.138 0.99 0.912 0.94 0.605
Hispanic 0.90 0.365 0.81 0.001 1.21 0.007 0.94 0.332 0.95 0.382
Asian/Pacific Islander 0.65 0.073 0.80 0.372 0.69 0.162 0.70 0.106 0.76 0.207
Native American/Alaskan
Native 0.83 0.226 0.70 0.023 0.87 0.426 0.84 0.259 0.87 0.348
Age category
under 5 0.96 0.655 0.97 0.773 0.97 0.789 0.95 0.547 0.89 0.209
5-17 ref - ref - ref - ref - ref -
18-49 1.36 <0.000 1.21 0.020 1.62 <0.000 1.29 0.001 1.29 0.001
50-64 1.30 0.021 0.92 0.496 1.63 <0.000 1.11 0.365 1.18 0.151
over 65 1.70 0.001 1.09 0.558 2.15 <0.000 1.50 0.009 1.76 <0.000
Insurance type
Private ref - ref - ref - ref - ref -
Medicare 0.96 0.749 0.89 0.414 0.85 0.274 0.95 0.699 0.91 0.482
Medicaid 1.20 0.017 1.21 0.017 1.03 0.740 1.12 0.120 1.23 0.007
Self-pay 1.20 0.101 1.11 0.349 1.07 0.599 1.13 0.268 1.20 0.105
Other 0.79 0.127 1.18 0.291 0.95 0.736 0.83 0.211 0.89 0.428
Tucson
08-09 09-10 10-11 11-12 12-13

133

Season Removed
OR p-value OR p-value OR p-value OR p-value OR p-value
Difference from breakpoint 1.25 <0.000 1.46 <0.000 1.25 <0.000 1.31 <0.000 1.39 <0.000
Race
White non-Hispanic ref - ref - ref - ref - ref -
Black non-Hispanic 1.00 0.992 1.00 0.989 0.90 0.418 0.94 0.608 0.98 0.857
Hispanic 0.98 0.791 0.92 0.177 0.89 0.076 0.99 0.832 0.95 0.408
Asian/Pacific Islander 0.74 0.170 0.68 0.086 0.64 0.063 0.72 0.146 0.70 0.105
Native American/Alaskan
Native 0.90 0.513 0.85 0.302 0.86 0.346 0.83 0.211 0.84 0.235
Age category
under 5 0.89 0.227 0.98 0.809 0.98 0.812 0.96 0.625 0.95 0.544
5-17 ref - ref - ref - ref - ref -
18-49 1.28 0.001 1.39 <0.000 1.27 0.003 1.31 <0.000 1.33 <0.000
50-64 1.19 0.138 1.35 0.010 1.10 0.455 1.09 0.449 1.19 0.114
over 65 1.49 0.011 2.12 <0.000 1.59 0.005 1.66 0.001 1.64 0.001
Insurance type
Private ref - ref - ref - ref - ref -
Medicare 0.94 0.639 0.86 0.284 1.01 0.967 0.87 0.308 0.91 0.493
Medicaid 1.19 0.021 1.24 0.006 1.18 0.042 1.19 0.018 1.18 0.025
Self-pay 1.18 0.132 1.31 0.015 1.24 0.055 1.20 0.088 1.18 0.112
Other 0.83 0.219 0.84 0.237 0.90 0.467 0.85 0.246 0.88 0.36
Tucson
13-14 14-15 15-16 16-17 Full Model

134

Table C.5 A comparison of the city level ratios for both the data used in this study (SID+SEDD
combined) and city level estimates from Google Flu Trends (GFT). The notes in this table were
compiled from historic CDC information to help explain potential differences seen in the ratios
for each data set.

SID+SEDD Combined GFT

Season Mesa Tucson Ratio Season Mesa Tucson Ratio Notes
08-09 487 385 1.26 08-09 59520 51926 1.15
Similar ratio to GFT, CDC
severity ranking low across the
board, no season summary to
refer to. Mesa had more cases
in children
09-10 976 2042 0.48 09-10 102048 108821 0.94
Seems low for Mesa BUT
H1N1 pandemic was *much*
worse for children than adults
and older adults. Cases for
children in Tucson were much
higher
10-11 764 783 0.98 10-11 75249 68306 1.10
CDC severity moderate across
the board, residual H1N1,
vaccine very effective.
Comparable cases for age
categories in these cities
except children <5 higher in
Mesa
11-12 234 549 0.43 11-12 64011 53662 1.19
Low CDC severity across the
board but high % of H1N1 in
Arizona (source: CDC
FluView) could have made this
worse for Tucson which has
more young people, more
cases in young and middle
aged adults in Tucson
12-13 940 886 1.06 12-13 148321 125989 1.18
CDC rated high severity for
older adults (why we might see
more Mesa cases), CDC
seasonal report noted
particularly higher rates of
hospitalization for older adults
and lower vaccine efficacy for
this group, Mesa had slightly
more cases across age
categories
13-14 791 1055 0.75 13-14 67745 74516 0.91
CDC severity moderate across
the board, however CDC noted
that H1N1 was elevated and
had more reports of flu in
young adults and middle-aged
adults - true in age distribution

135

of cases, vaccine was pretty
effective and older people are
more likely to get vaccinated
so that may also contribute to
Mesa's lower cases
14-15 1065 839 1.27 14-15 71492 72575 0.99
Maybe high for Mesa but CDC
severity high for older adults,
cases higher for Mesa across
age categories
15-16 2349 1502 1.56 15-16 - - -
CDC rated moderate for adults,
low for others. % of positive
tests for Arizona (source: CDC
FluView) is much higher than
other seasons so I think the
magnitude is ok but I don't
know why Mesa might be
higher. No real additional
insight from season summary,
cases in Mesa higher across
age categories
16-17 734 682 1.08 16-17 - - -
CDC severity moderate across
the board, numbers are
comparable to '10-11 season
which was also moderate,
CDC season summary did not
provide further insight, Mesa
had slightly higher cases in
kids and +65

Abstract (if available)

Abstract Influenza, or flu, is a contagious respiratory illness with symptoms such as fever, body aches, fatigue, and cough. Despite access to a yearly vaccine, the United States (US) Centers for Disease Control (CDC) estimates that between 3-11% of the US population, tens of millions of people, have symptomatic influenza infections each year. Despite the high morbidity of influenza as well as the mortality risks it poses for vulnerable individuals, our understanding of the fundamental transmission dynamics of the disease are still not well understood. Research across disciplines has shown increasing evidence for the impacts of local climate, network and mobility functions, demographic and socioeconomic status, as well as the virus’s antigenic characteristics on our physical vulnerability, our prevention behaviors, and our perception of risk. An increased understanding of the factors that impact both the timing and magnitude of seasonal outbreaks can aid in the development of more accurate influenza forecasting models and inform public health campaigns aimed at preventing and reducing the burden of influenza. This dissertation explores the role of population and place in these dynamics by: 1. quantifying state-level humidity thresholds that signal the onset of seasonal outbreaks, 2. Exploring age-stratified racial and ethnic disparities in influenza-related hospital admissions across space and time, and 3. assessing the role of underlying population characteristics in the timing of seasonal outbreaks when humidity is controlled for.

These three studies contribute to the body of influenza work by demonstrating spatial variation in humidity-influenza relationships, as well as, racial and ethnic disparities. Furthermore, it provides evidence from a case study for the role of population characteristics in the timing of influenza outbreaks. The humidity thresholds identified in study one were strongly associated with annual average humidity for each state (R2 = 0.90). This finding provides additional support to the theory of individual level adaptations potentially being the driving mechanism of the influenza-humidity relationship. In study two, there were notable spatial and temporal differences in racial and ethnic disparities that highlight the need for sub-national and age-specific exploration of disparate trends in influenza hospitalizations. Black non-Hispanics, particularly children, were generally admitted more frequently than White non-Hispanics in all three states. Hispanics had the most spatial variation in their hospital admissions compared to White non-Hispanics. Finally, study three provides evidence for the role of population characteristics, specifically age structure and insurance type, in the timing of influenza outbreaks in Mesa and Tucson, Arizona. There were statistically significant differences between the cities in the likelihood of adults aged 18 and older requiring an early hospital utilization compared to 5-17 year olds (18-49 years: Chi-squared=10.50, p-value=0.001; 50-64 years: Chi-squared=5.89, p-value=0.015; over 65: Chi-squared=7.68, p-value=0.006), as well as for individuals on Medicaid compared to those with private insurance (Chi-squared=6.33, p-value=0.012). This result has important implications for the use of population data in models that aim to predict seasonal influenza outbreaks and for our understanding of factors that contribute to their timing.

The findings of these studies demonstrate the need for local conditions to be considered in influenza forecasting efforts as well as public health interventions. Both environmental conditions and population characteristics influenced the timing of influenza outbreaks while age structure and race/ethnicity played a role in the distribution of cases once outbreaks were underway. Future work should consider broader spatial and temporal contexts to identify the scope of variation in these relationships across the United States.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Disparities in gallbladder, intra-hepatic bile duct, and other biliary cancers among multi-ethnic populations: a California Cancer Registry study

PDF

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

PDF

Pharmacogenetic association studies and the impact of population substructure in the women's interagency HIV study

PDF

Methodological approaches to assessing diurnal cortisol rythms in epidemiological studies: how many salivary samples are necessary?

PDF

Hepatitis C test result awareness among reported cases in Los Angeles County

PDF

Instability of heart rate and rating of perceived exertion during high-intensity interval training in breast cancer patients undergoing anthracycline chemotherapy

PDF

Disparities in colorectal cancer survival among Latinos in California

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

Diet quality and pancreatic cancer incidence in the multiethnic cohort

PDF

Carcinogenic exposures in racial/ethnic groups

PDF

Early childhood health experience & adult phenotype in twins

PDF

Prostate cancer disparities among Californian Latinos by country of origin: clinical characteristics, incidence, treatment received and survival

PDF

A biodemographic approach to understanding sociodemographic disparities in kidney functioning on three dimensions: individual, population, and cross-national

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

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

Using genetic ancestry to improve between-population transferability of a prostate cancer polygenic risk score

PDF

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

PDF

Racial/ethnic differences in colorectal cancer patient experiences, health care utilization and their association with mortality: findings from the SEER-CAHPS data

PDF

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

PDF

Genetic studies of cancer in populations of African ancestry and Latinos

Asset Metadata

Creator Serman, Emily Ann (author)

Core Title Understanding the role of population and place in the dynamics of seasonal influenza outbreaks in the United States

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Health and Place,Population

Degree Conferral Date 2022-08

Publication Date 07/25/2022

Defense Date 04/29/2022

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

Tag Epidemiology,humidity,influenza,OAI-PMH Harvest,outbreak timing,transmission dynamics

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Ailshire, Jennifer (committee chair), Franklin, Meredith (committee chair), Crimmins, Eileen (committee member), Gounder, Prabhu (committee member), Ruddell, Darren (committee member)

Creator Email emily.a.serman@gmail.com,eserman@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC111375399

Unique identifier UC111375399

Legacy Identifier etd-SermanEmil-10962

Document Type Dissertation

Format application/pdf (imt)

Rights Serman, Emily Ann

Type texts

Source 20220728-usctheses-batch-962 (batch), 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 author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright. The original signature page accompanying the original submission of the work to the USC Libraries is retained by the USC Libraries and a copy of it may be obtained by authorized requesters contacting the repository e-mail address given.

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

Repository Email cisadmin@lib.usc.edu