University of Southern California Dissertations and Theses

Biomarkers of age-related health changes: associations with health outcomes and disparities

(USC Thesis Other)

Biomarkers of age-related health changes: associations with health outcomes and disparities

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content BIOMARKERS OF AGE-RELATED HEALTH CHANGES:
ASSOCIATIONS WITH HEALTH OUTCOMES AND DISPARITIES
By
Qiao Wu
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
(GERONTOLOGY)
May 2024
Copyright 2024 Qiao Wu

ii
Dedication
This dissertation is dedicated
To my best friend and lovely wife, Fengxue, who brought KuKu, the warm sunshine on my lap, and Becky, my endless source of joy, into my life during the creation of this work.

iii
Acknowledgments
Expressing my gratitude to those individuals whose influence and support have turned
this dissertation from a daunting task into an exciting journey might be as challenging as writing
the dissertation per se. While I am indebted to so many more people than I have room to share
here, this dissertation would not have been possible without the unwavering support of a few. At the forefront of my acknowledgments is my mentor and committee chair, Dr. Eileen
Crimmins. Eileen, you have been the most profound inspiration, the cornerstone of guidance and
support, and the Rick Sanchez (and a way better version of him) in this Morty’s academic
adventure. Working alongside you will always be the most gratifying experience when
recounting my time at USC. Thank you for teaching me the way to digest, produce, and
communicate science, and for always being a generous resource of invaluable learning
opportunities. I am aware that anything I write here would end up as an oversimplification of the
tremendous help I received from you, and that realization is precisely why I am truly grateful. I would also like to extend a heartfelt gratitude to my committee members, Dr. Jennifer
Ailshire and Dr. Jung Ki Kim. Jennifer, thank you for being the methodology guru who
constantly illuminates the possibilities of fascinating data and tools. You are the reason I am a
more confident researcher, presenter, writer, and more importantly, a better cook. Jung Ki you
are family to me. I was so spoiled by your thoughtful help and your data magic! Beyond
academia, thank you for being mad at me when I got my first (and second) tattoo and when I
complained as a first-time dad. It did not change my stupid ideas, but it meant a lot to me. You
made me feel at home at times when the idea of home was quite literally thousands of miles
away.

iv
Most importantly, to my best friend and lovely wife, Fengxue. 美美, you have been the
constant support that I could not have even dreamt of throughout these past years, and I could not
have accomplished anything without your encouragement. Your unconditional love and faith
have been the pillars that kept me standing tall and helped me navigate the incessant challenges
and unbearable stress. You are the most wonderful listener, knowledgeable audience, and
constructive critic of not only my research but also all the other crazy and silly thoughts I have. I
am incredibly grateful for your companionship and thrilled to see where the journey ahead takes
us. I also owe immense thanks to my parents. They have been my champions and supporters
for the past 30 years. My love and gratitude for them are, as always, beyond words. It is
impossible to list all the sacrifices they made to support me. It is amusing now to recall that
becoming a scientist used to be my casual response when asked about my dream when I was
little. Now that I approach another career milestone, I feel surprisingly closer to that dream, and I
hope my parents feel proud. Also, I would like to thank my 86-year-old grandmother, Yajun, not
only for her unwavering support but for also being my true inspiration. She exemplifies using her
own experience that aging can be joyful, beautiful, and elegant. She is a true educator and role
model. Through her influence, a tradition of wise, strong, accomplished, and liberal women has
been passed on in my family. I aspire for my daughter, Becky, to embody the same qualities as
she grows up. I would like to give a big shout-out to Dr. Erfei Zhao and Margarita Osuna with a million
glasses of beer. I could never have expected better friends and colleagues than you. And as you
already knew, I could not have survived without you.

v
I thank and blame KuKu. I do not know how to type without something furry on my lap
now. It remains debatable if it makes my research any better, but you significantly improved my
experience of life (p<.001), and for that, I am thankful. I also thank Ginger Root. Your amazing works have been the much-needed boost of
caffeine during the coding and writing sessions.

vi
Table of Contents
Dedication....................................................................................................................................... ii
Acknowledgments..........................................................................................................................iii
List of Tables............................................................................................................................... viii
List of Figures................................................................................................................................ xi
Abstract........................................................................................................................................ xiii
Chapter 1: Introduction ................................................................................................................... 1
The Biodemographic Approach to Understanding Health and Aging.................................... 1
Gaps in the Literature .............................................................................................................. 4
The Current Study ................................................................................................................... 6
Chapter 2: Cardiometabolic Risk Trajectory among Older Americans: Findings from the Health
and Retirement Study .................................................................................................................... 10
Introduction ........................................................................................................................... 10
Data and Methods..................................................................................................................12
Results................................................................................................................................... 18
Discussion............................................................................................................................. 22
Supplemental Information .....................................................................................................31
Chapter 3: The Association between Cardiometabolic Risk and Cognitive Function among Older
Americans and Chinese ................................................................................................................. 39
Introduction ........................................................................................................................... 39
Methods................................................................................................................................. 43
Results................................................................................................................................... 49

vii
Discussion............................................................................................................................. 54
Supplemental Information.....................................................................................................62
Chapter 4: RNA-based Indicators of Cellular Senescence Predict Aging Health Outcomes in the
Health and Retirement Study ........................................................................................................ 73
Introduction ........................................................................................................................... 73
Methods................................................................................................................................. 76
Results................................................................................................................................... 81
Discussion............................................................................................................................. 84
Supplemental Information .....................................................................................................90
Chapter 5: Conclusion and Discussion..........................................................................................98
References................................................................................................................................... 104

viii
List of Tables
Table 2.1. Results of the Growth Curve Model Predicting Total Cardiometabolic Risk ..............27
Table 2.2. Results of the Growth Curve Model Predicting Biomarker Z-Scores......................... 28
Table 2.3. The Growth Curve Model Predicting Biomarker Z-Scores Controlling for
Medication .....................................................................................................................................28
Supplemental Table 2.1. Sample Characteristics at Baseline ....................................................... 31
Supplemental Table 2.2. The High-Risk Thresholds for CMR Risk Factors................................32
Supplemental Table 2.3. Means of Biomarker Z-Score at 3 Waves and T-Tests for the .............32
Supplemental Table 2.4. SBP, DBP, RHR, and TC Mean Values by Wave and Medication
Group.............................................................................................................................................33
Supplemental Table 2.5. CMR Model Results Using All-Wave Participants and Models
Including Those Who Died and Those Who Missed Only One Wave ......................................... 34
Supplemental Table 2.6. Comparison between Biomarker Z-Score Models Using All-Wave
Participants and Models Including Those Who Died ....................................................................35
Supplemental Table 2.7. Comparison between Biomarker Z-Score Models Using All-Wave
Participants and Models Including Those Who Died and Those Who Missed Only One Wave ..36
Supplemental Table 2.8. Comparison between Total CMR Models Using Two Measures: Our
Proposed Measure vs. Mitchell et al. ............................................................................................ 37
Table 3.1. Results from the Baseline Models Predicting Cognitive Function in the Two
Countries....................................................................................................................................... 58
Table 3.2. Results from the Longitudinal Models Predicting Cognitive Function 4 Years after
Initial CMR in Two Countries with Both Initial Levels of CMR and CMR Increase .................. 59

ix
eTable 3.1. Comparison between Initial Samples and Analytic Samples..................................... 62
eTable 3.2. Word Lists and Total Syllable Counts for the CHARLS Word Recall Tests............ 63
eTable 3.3. Word Lists and Total Syllable Counts for the HRS Word Recall Tests.................... 63
eTable 3.4. High-Risk Thresholds for CMR Indicators................................................................ 63
eTable 3.5. Sample Characteristics for the US (HRS) and the Chinese (CHARLS) Samples......64
eTable 3.6. Results from the Cross-Sectional Models with Interaction Terms.............................65
eTable 3.7. Results from the Longitudinal Models with Interaction Terms................................. 65
eTable 3.8. Change Models in the US: CMR Biomarkers with Baseline Adjustment VS. CMR
Biomarkers without Baseline Adjustment.....................................................................................66
Table 4.1. Sample Characteristics................................................................................................. 87
Table 4.2. Results of the OLS Model Predicting the Senescence Scores (N=3,580)....................88
Supplemental Table 4.1. The Gene Lists for Gene Expression Composite Scores.......................90
Supplemental Table 4.2. Results from OLS Regression Models Assessing the Associations
between Senescence Scores and PC GrimAge AA (N=3,580)..................................................... 91
Supplemental Table 4.3. Results from OLS Regression Models Assessing the Associations
between Senescence Scores and PC PhenoAge AA (N=3,580)....................................................91
Supplemental Table 4.4. Results from OLS Regression Models Assessing the Associations
between Senescence Scores and DunedinPACE (N=3,580)......................................................... 92
Supplemental Table 4.5. Results from Logistic Regression Models Assessing the Associations
between Senescence Scores and 4-Year Mortality (N=3,575)......................................................93
Supplemental Table 4.6. Results from Logistic Regression Models Assessing the Associations
between Senescence Scores and 4-Year Mortality Adjusted for PC GrimAge AA (N=3,575)....93

x
Supplemental Table 4.7. Results from OLS Regression Models Assessing the Associations
between Senescence Scores and Multimorbidity (N=3,580)........................................................ 94
Supplemental Table 4.8. Results from OLS Regression Models Assessing the Associations
between Senescence Scores and Multimorbidity Adjusted for PC GrimAge AA (N=3,580)...... 95
Supplemental Table 4.9. Results from OLS Regression Models Assessing the Associations
between Senescence Scores and Biological Age Acceleration (N=2,660)................................... 95
Supplemental Table 4.10. Results from OLS Regression Models Assessing the Associations
between Senescence Scores and Biological Age Acceleration Adjusted for PC GrimAge AA
(N=2,660)...................................................................................................................................... 96

xi
List of Figures
Figure 1.1. The Biodemographic Model of Health and Aging ....................................................... 8
Figure 1.2. Example of Social Factors Link to Health Outcomes through Biology....................... 9
Figure 2.1. Biomarker Z-Score Trajectories................................................................................. 29
Figure 2.2. Trajectories of Individual Biomarkers by Medication Group.....................................30
Supplemental Figure 2.1. Comparison Across Poisson Regression Model Predicted Outcomes
Using Different Samples: CMR as Measured in Mitchell et al. and Samples Present in 4-year and
8-year Analyses............................................................................................................................. 37
Supplemental Figure 2.2. Comparison Across Poisson Regression Model Predicted Outcomes
Using Different CMR Measures (Mitchell et al. is 7 items and this paper is 9 items) and Samples
(4-year and 8-year Sample). .......................................................................................................... 38
Figure 3.1. Baseline Cognitive Function Score by CMR Level in the US (HRS, N=7,413) and
China (CHARLS, N=6,108)..........................................................................................................59
Figure 3.2. Cognitive Function Score for People with Low-Risk versus High-Risk Biomarker
Values of Individual Biomarkers in the US (HRS, N=7,413) and China (CHARLS, N=6,108)..60
Figure 3.3. Estimated Probability of Having High-Risky Biomarker Level by Education Groups
in the US (HRS, N=7,413) and China (CHARLS, N=6,108)....................................................... 61
eFigure 3.1. Distributions of CHARLS Blood-Based Biomarkers at Baseline and Follow-up
Waves before and after Adjustments (left panels: unadjusted; right panels: adjusted).................66
eFigure 3.2. Distributions of HRS Dry Blood Spot Biomarkers at Baseline and Follow-up Waves
before and after Baseline Adjustments (left panels: unadjusted; right panels: adjusted)..............68

xii
eFigure 3.3. Distributions of Cognitive Function for both Countries at the Baseline and the
Follow-Up Waves..........................................................................................................................70
Figure 4.1. The Distribution and Correlation Matrix of the Cellular Senescence Scores.............88
Figure 4.2. The Associations between Cellular Senescence Scores and Multiple Dimensions of
Aging Measures/Outcomes........................................................................................................... 89

xiii
Abstract
Incorporating biomarkers into social science aging research helps understand the risk for
age-related health changes and explains how social and environmental factors get translated to
health. However, the existing literature can benefit from longitudinal and cross-country evidence
and new biomarkers to measure more fundamental mechanisms of aging. Based on the
comparable nationally representative longitudinal surveys with biomarkers on older adults in the
US and China, the current dissertation contributes to the literature by depicting the with-in- person trajectory of cardiometabolic risk (CMR) among older Americans, comparing the cross- sectional and longitudinal associations between CMR and cognitive function between the US and
China, and testing a group of RNA-based cellular senescence measures at the population level, showing their associations with social-behavioral factors and age-related health outcomes. The
results show that the overall CMR can be fairly stable during 8 years of aging among older
Americans. Medication use contributes to the decrease in blood pressure, heart rate, and total
cholesterol. However, glycosylated hemoglobin increases, especially faster among non-Hispanic
Blacks, indicating a need for better blood sugar management; Waist circumference also increases, especially faster among people with higher education levels, indicating a need for a healthier
lifestyle. Both a higher CMR level and a CMR increase are associated with worse cognitive
function and faster cognitive decline among older Americans but not Older Chinese. In China, those with the highest socioeconomic status (SES) seem to have both higher CMR and better
cognitive health. Among them, having a higher CMR is not as harmful as it is for their lower- SES counterparts in terms of cognitive health, but a rapid rise in CMR is additionally harmful. The results also show that among older Americans, older ages, female sex, being non-Hispanic

xiv
Black, and being obese are linked to a higher level of cellular senescence. Cellular senescence
correlates highly with epigenetic aging. It is also associated with health outcomes including
physiological dysregulation and multimorbidity, with and without adjustment for epigenetic
aging, and these associations are mainly driven by macromolecular damage and senescence- associated secretory phenotype. So, the RNA-based cellular senescence measure used in this
dissertation measures aging and adds to our understanding of physiological dysregulation and
age-related diseases in a useful way.

1
Chapter 1: Introduction
Aging is a gradual and highly complex process jointly influenced by a wide range of
intertwined biological and social-behavioral factors. This process is both universal and distinct. It
is a shared experience of all human beings across their life span, while the aging experience can
be highly contextual and variable across individuals. Hence, understanding and modeling aging
requires a comprehensive grasp of both social and biological information, as well as an
awareness of both common underlying mechanisms and potential disparities. This dissertation
attempts to add to our understanding of aging-related health changes and outcomes using high- quality population-level data that recently became available. The Biodemographic Approach to Understanding Health and Aging
In recent decades, there has been a growing interest among social scientists in explaining
how social, economic, psychological, and environmental factors “get under the skin” and affect
health outcomes. To achieve that, it is important to include biology in social science studies of
health and aging, which used to be limited to reported but not measured variables (e.g., self-rated
health, morbidity, disability, etc.). Biodemography is such a multidisciplinary framework that
represents a series of analytical approaches based on biomarkers – the objective measures of
biological mechanisms – at different dimensions of health changes (Crimmins et al., 2010). The complex interaction between biological and social factors and their gradual impact
on the mortality process can be heuristically described in the biodemographic model of health
and aging (Figure 1.1). At the population level, the common age-related health changes, i.e., the
mortality process, can be organized into several sequential dimensions (the grey boxes of Figure

2
1.1): While individuals may skip dimensions or experience reversals and repeats, at the
population level, age-related health change starts with “upstream health changes” – molecular
and cellular changes such as epigenetic changes and cellular senescence. After accumulation, these changes can lead to “downstream health changes and outcomes” including physiological
dysregulation, usually detectable based on clinical indicators, and then followed by health
outcomes including multimorbidity, frailty, disability, functioning loss, and mortality. These
sequential dimensions of health changes are particularly important for aging research. On one
hand, understanding the aging process, to a great extent, is to figure out when, where, and how
fast these health changes take place, as well as what contributes to the accumulation of upstream
damages and the progression of downstream outcomes. On the other hand, interventions to affect
aging increasingly attempt to reduce the level of upstream damage and delay the onset of
downstream outcomes. Figure 1.1 also provides an overview of how the previously discussed age-related health
changes can be influenced by social, environmental, genetic, and even broader contextual factors. Extensive evidence has identified a wide range of social factors that are associated with aging, and they can be characterized into the 5 highly interrelated “social hallmarks of aging”, including
(i) low socioeconomic status, (ii) minority status (based on demographic characteristics), (iii)
adverse life events, (iv) adverse psychological states, and (v) adverse behaviors (Crimmins, 2020). Figure 1.1 shows that demographic factors and socioeconomic status can influence
individuals’ exposure to adverse behaviors, psychological states, and life events, and then affect
biology and downstream health changes. It is important to point out that the model indicated by
Figure 1.1 is heuristic since there are many unrepresented one-way and bidirectional arrows as
well as hidden concepts or mechanisms. The mechanisms currently shown in the model can also

3
be influenced by overarching factors such as epidemiological, physical, socioeconomic, and
policy contexts. Since many important dimensions in the biodemographic model of health and aging
heavily involve underlying biology, the model highlights the importance of incorporating
biological mechanisms, informed by biomarkers, into social science and population health
studies of aging. Biomarkers are extremely valuable in health and aging research for multiple
reasons. First, biomarkers can usually indicate the risk before the onset of diseases, and
sometimes are even modifiable, so they play an important role in disease prevention and
intervention. Second, biomarkers are usually more objective in terms of depicting the prevalence
of diseases and conditions compared to self-reports. For example, according to the China Health
and Retirement Longitudinal Study (CHARLS) data, among middle-aged and older Chinese who
actually have diabetes based on biomarkers (fasting plasma glucose and glycosylated
hemoglobin – HbA1c), most of them (about 60%) are not aware of their diabetic status, remaining undiagnosed (Zhao et al., 2016; Li & Lumey, 2019), and among the ones with
hypertension based on biomarkers (systolic and diastolic blood pressures), around 46% are
unaware (Li & Lumey, 2019). In this case, the collection of biomarkers greatly improves the
knowledge of disease prevalence and also highlights the need for the betterment of chronic
disease screening and management. Third, summary measures of biomarkers can be used to
capture multisystem biology as well as complicated mechanisms/phenotypes such as aging that
are otherwise hard to measure. For example, summary measures like allostatic load (Seeman et
al., 2001), phenotypic age (Liu et al., 2018), and biological age (Levine, 2013; Crimmins et al., 2021) based on clinical-level biomarkers that indicate the functioning of multiple bodily systems
have been developed to measure the overall level of physiological deterioration. To measure

4
aging at the molecular/cellular level, biomarkers are developed based on the biological hallmarks
of aging (López-Otín et al., 2013, 2023). Some most frequently used include DNA-methylation- based epigenetic aging measures such as GrimAge (Lu et al., 2019), PhenoAge (Levine et al., 2018), and the pace of aging (Belsky et al., 2015). Gaps in the Literature
Despite the collective efforts made by researchers of social, medical, and biological
science for decades, most early and some current evidence is still limited by the availability of
data in terms of both how the data were collected and where the data originate. The biological
measures that are currently available seem to be insufficient to capture all facets of aging and to
fully explain the impact of the social environment on health. The Importance of Longitudinal Evidence – Since the collection of biological data can be
costly, especially in large population-representative surveys, early population studies with
biomarkers mainly focus on a single time point, and find that older people are more likely to
have multisystem physiological dysregulation (Crimmins et al., 2003; Levine et al., 2013; Yang
et al., 2011; Geronimus et al., 2006; Seeman et al., 2008). However, cross-sectional studies can
only depict differences by age across individuals instead of changes with age within individuals. The observed differences might be due to the cohort effect, which is very common and could be
particularly strong in societies that undergo rapid social and economic transitions. Hence, more
longitudinal evidence based on biomarkers is required. The Importance of Cross-Country Comparisons – Comparable population-level data
collected across countries provides a unique opportunity to understand aging under various social
contexts. However, cross-country aging studies and data combining social and biological

5
information are only starting to accumulate in recent years. The examinations of human behavior, psychology, and many other social factors are disproportionately based on Western, Educated, Industrialized, Rich, and Democratic (WEIRD) societies (Henrich et al., 2010; Kohrt et al., 2016). Similarly, for a lot of biological data, including genetic and epigenetic data, not all
populations are equally represented. As an example, according to EWAS Atlas, most epigenetic
data (with known ancestry information) predominately come from European ancestry, followed
by Black/African American (EWAS Atlas 2023 https://ngdc.cncb.ac.cn/ewas/statistics). Fortunately, innovative data collection and harmonization are making the examination of bio- social interplay become more and more feasible. For example, the Gateway to Global Aging
(g2aging.org) is a platform harmonizing population surveys on aging, specifically the
International Family of Health and Retirement Studies (HRS Family Studies) which covers many
high-income countries (e.g., US, UK, European countries, etc.) as well as low- and middleincome countries (LMICs) (e.g., China, India, Mexico, etc.) (Lee et al., 2021). Bringing a global
perspective and cross-society comparative evidence to the field of aging research is important. They help elucidate both the universal/fundamental aspects of aging and the aging experience
specific to certain macro-socioeconomic contexts. They also help identify and assess the relative
importance of key socioeconomic determinants of aging, providing implications on how to
improve the health and well-being of the older population (Ailshire & Carr, 2021). The Importance of Measuring Underlying Aging Mechanisms – In the meantime, more
efforts are still needed to better incorporate fundamental aging biology into social studies. A
recent study (Crimmins, 2020) uses both social hallmarks of aging and a wide range of aging
biomarkers to explain age-related health outcomes. The examined biomarkers are frequently
used in population-level studies and cover multiple dimensions and correspond to multiple

6
biological hallmarks of aging, including biological age (calculated based on 10 clinical-level
biomarkers), telomere length, mitochondrial copy number, epigenetic age acceleration (the
Horvath DNA methylation measures, Horvath 2013), and a set of polygenic risk scores relevant
to the outcomes of interest (cognition, coronary artery disease, myocardial infarction, type 2
diabetes, and longevity). Interestingly, the inclusion of biomarkers only partially explained the
strong effects of social hallmarks on aging multimorbidity, disability, cognitive deficiency, and
mortality. These findings, on one hand, highlight the importance of measuring social
determinants when modeling aging, while on the other hand, indicate the potential research
opportunities to improve current biological measures and develop new markers to cover the basic
mechanisms of aging that are yet to be accounted for. The Current Study
To bridge the gaps in the literature, the current study first aims to examine within-person
physiological change with age among older Americans longitudinally in Chapter 2. Specifically, we focus on the 8-year change in the cardiometabolic risk (CMR), which is calculated by a set of
blood-based and anthropometric biomarkers indicating cardiovascular, metabolic, and
inflammatory health. In addition to the general population-level trajectory, we also aim to
examine how the trajectory differs across demographic and socioeconomic population subgroups. Since CMR factors are modifiable risk factors for many health outcomes, including
cognitive problems, the aim of Chapter 3 is to understand the association between CMR and
cognitive functioning, both cross-sectionally and longitudinally, i.e., we examine both the levels
and the changes of CMR and cognitive functioning. At the same time, since such associations
could be largely influenced by social factors, both at the individual level and the level of broader

7
epidemiological and developmental contexts, we compare the associations between the US and
China using comparable population-representative survey data. In this particular case, China
serves as a very good comparison relative to the US, as China has a very different overall
developmental level and epidemiological context, including suboptimal chronic disease
management (Zhao et al., 2016; Li & Lumey, 2019), low education level of the current older
population (Gateway to Global Aging Harmonized Data, 2022), huge urban advantage in terms
of socioeconomic and healthcare resources (Gong et al., 2012), and paradoxically worse
cardiometabolic-inflammatory health among urban compared to rural residents (Zhang &
Crimmins, 2019; Zhao et al., 2016; Hou, 2008). Another aim of the current study is to test a group of innovative biomarkers of cellular
senescence, which is one of the fundamental mechanisms of aging that has not been well
measured at the population level, and assess whether and how they explain health and aging
among older Americans. In recent years, researchers have started to use RNA summary scores in
human samples to understand how the social environment regulates health through gene
expressions. As shown in the social-environmental signal transduction pathway (Figure 1.2), social, environmental, and behavioral factors can potentially influence people’s transcriptional
profile, either through physiochemical and microbial processes involving toxins, pollutants, microbes, and nutrients, or through psychological processes where neural and endocrine
responses are triggered by stresses such as traumatic experience, threats, and uncertainty. In both
scenarios, these exposure-related processes can activate (or repress) the transcription factors that
regulate the expression of certain genes. Gene expression can then influence subsequent
physiological processes through, for example, the production of proteins (Cole, 2014, 2019).

8
In Chapter 4, similar to many existing RNA-based human studies (Cole et al., 2020;
Mann et al., 2020; Levine et al., 2017), we use gene expression composite scores based on RNA
sequencing data, to measure the level of cellular senescence. The strength of our data is that they
come from the HRS and thus are representative of older Americans. We construct the composite
scores based on the hallmarks of the senescence phenotype and examine their associations with
downstream aging outcomes like mortality, multimorbidity, and physiological dysregulation, as
well as the upstream aging mechanism indicated by a set of DNA-methylation-based epigenetic
aging measures. We also aim to examine how demographic, socioeconomic, and behavioral
characteristics link to cellular senescence. Finally, Chapter 5 concludes the significant findings and discusses future research
directions. To summarize, this dissertation contributes to the literature on biodemography by
presenting longitudinal and international evidence, and by measuring cellular senescence using a
group of novel RNA-based gene expression scores at the population level. Tables/Figures
Figure 1.1. The Biodemographic Model of Health and Aging

9
Note. Demographic, socioeconomic, behavioral, and biological influences on health outcomes. For illustration purposes, many arrows and concepts are not presented, and many bidirectional ef ects are not shown. An updated version of Crimmins and
Vasunilashorn 2016. Figure 1.2. Example of Social Factors Link to Health Outcomes through Biology
Note. Derived from Cole 2019 and Cole 2014.

10
Chapter 2: Cardiometabolic Risk Trajectory among Older
Americans: Findings from the Health and Retirement Study
* This chapter was published as Wu Q, Ailshire JA, Kim JK, Crimmins EM. Cardiometabolic risk trajectory among
older Americans: Findings from the Health and Retirement Study. The Journals of Gerontology: Series A. 2021 Dec
1;76(12):2265-74. https://doi.org/10.1093/gerona/glab205
Introduction
Age is a major risk factor for poor health outcomes. Understanding how physiological
status changes with age is an important addition to our understanding of aging health. Cross- sectional data have shown that biological dysregulation is higher at older ages, and that older
people are more likely to have multisystem physiological dysregulation (Crimmins et al., 2003;
Geronimus et al., 2006; Levine, 2013; Seeman et al., 2008; Yang & Kozloski, 2011). Prior
population-based studies have largely relied on cross-sectional data to examine differences in
biological dysregulation across age groups, with relatively less attention to changes with age. Age differences observed in cross-sectional data can conflate cohort differences and mortality
effects with age-related changes in health. There are also limitations to prior studies with longitudinal analysis of changing
physiological status. For instance, several studies used samples with limited representativeness
or were based on a short time frame for studying change. Karlamangla et al. (Karlamangla et al., 2006) found an increase over 2.5-years in allostatic load, a multisystem indicator of
physiological functioning, in a study of “successful agers” from three communities in the U.S. Merkin et al. (Merkin et al., 2014) also found an increase in allostatic load over 7 years of aging

11
in a sample recruited from 6 U.S. counties, from 2000 to 2007, with the increase slower among
those with highest education. Other examinations of change in physiological indicators have focused on younger age
groups or specific demographic groups. Belsky et al. (Belsky et al., 2015) quantified the pace of
physiological deterioration in 18 biomarkers reflecting multiple systems among adults aged 28 to
35 using longitudinal data from the Dunedin Study in New Zealand. O’Keeffe et al. (O’Keede et
al., 2019) found different trajectories in individual cardiovascular and metabolic measures by age
at menopause among middle-aged and older women in the U.K.: systolic blood pressure, waist
circumference, and high-density lipoprotein cholesterol increased with age, while low-density
lipoprotein, triglycerides, and diastolic blood pressure decreased with age. Mitchell et al. (Mitchell et al., 2019) examined race differences in 4-year CMR change in the nationally
representative U.S. Health and Retirement Study (HRS), finding that Blacks experienced an
increase in risk over 4 years of aging while Whites and Hispanics experienced a decrease. Changes in physiological risk that characterize older Americans over a longer period are yet to
be studied. Both individual risk factors (Crimmins et al., 2010; Kim et al., 2019; Seeman et al., 2008)
and summary biological risk measures (Crimmins et al., 2009; Kim et al., 2019; Levine &
Crimmins, 2014; Mitchell et al., 2019; Seeman et al., 2004) vary across age, but also by
race/ethnicity, gender, socioeconomic status (SES), and health behaviors. These factors may
modify the trajectory of CMR with age. In recent decades, the increased usage and effectiveness
of prescription drugs have driven improvement in some risk factors, particularly in blood
pressure and cholesterol (Kim et al., 2019; Crimmins et al., 2010). So, medication use may affect

12
changes over time in overall CMR by changing the prevalence of high risk of some biomarkers. All these factors need to be considered to understand the observed age/time trajectory. In the current study, we examine 8-year change in an index of cardiometabolic risk
(CMR) indicating dysregulation in the cardiovascular and metabolic systems (Mitchell et al., 2019). We use data from a nationally representative sample of Americans over the age of 50. To
better understand what drives the change in overall CMR, we also examine 8-year trajectories in
each biomarker included in total CMR. We hypothesize that total CMR will increase over time
with aging; however, individual biomarker components of the CMR may have different trends, and some of the trends may reflect the availability of medications. We also examine differences
in trajectories of CMR for race/ethnic and SES groups and current smoking status. Our
hypothesis is that racial/ethnic minorities, those with lower education level, and those who
currently smoke will have higher CMR and experience faster elevation in risk with age. Because
the trajectories of some biomarkers may be affected by medication use, we also examine the
effect of adopting or discontinuing medication use on specific markers. Data and Methods
Data
The Health and Retirement Study (HRS) is a nationally representative longitudinal study
that surveys U.S. adults over age 50 every two years. In 2006, the HRS initiated an Enhanced
Face-to-Face Interview (EFTF) where anthropometric measurements were taken, and dried blood
spots (DBS) were collected and subsequently assayed. In 2006, this was done for a random half
of the sample; the other half of the sample had the data collected in 2008. The first half sample

13
had the measures collected again in 2010 and 2014, and the second half in 2012 and 2016
(Crimmins et al., 2013, 2017, 2020). Each respondent could have measures collected at 3 times. Hence, in the current study 2006/2008 are considered the baseline wave (wave 1); wave 2
includes the 2010/2012 interviews; and wave 3 includes the 2014/2016 interviews. The three- waves track individual trajectories for 8 years. Among 12,000 people who participated in both the physical measurement and DBS
biomarker collection at the baseline interviews, 9,173 people had complete baseline data. We
excluded 2,142 people who died before the third wave and an additional 3,503 people who had
missing data for analysis variables. Our final analytical sample consisted of 3,528 individuals
who survived and had complete data for three waves. Most people with missing data were
missing on glycosylated hemoglobin (HbA1c) and high-density lipoprotein cholesterol (HDL-C)
(missing 1,139 and 1,127, respectively) because of limited blood spot samples. Overall, those
who had missing data (n = 3,503) and those who did not have complete data due to death (n =
2,142) had significantly higher baseline CMR, were older, and had lower education than our
final analytical sample. We show differences between these subsamples in Supplemental Table
2.1, and we report results from sensitivity analysis including those who had some missing data
after the main results. We also conducted analysis in the effect of medication on those in the
sample with complete information on blood pressure medication (n=3,497) and cholesterol
medication (n=2,694). Measures
Cardiometabolic risk (CMR) is based on multiple indicators of cardiovascular and
metabolic functioning including systolic blood pressure (SBP), diastolic blood pressure (DBP),

14
resting heart rate (RHR), waist circumference (WC), glycosylated hemoglobin (HbA1c), high- density lipoprotein cholesterol (HDL-C), total cholesterol (TC), and C-reactive protein (CRP). Blood pressure and RHR were measured using an Omron HEM-780N Monitor. SBP, DBP, and
RHR were measured three times (Crimmins et al., 2008); we used the average of the non-missing
measures for this analysis. WC was measured at the level of the navel. The HRS biomarker
values for CRP, TC, HDL-C, and HbA1c were assayed from dried blood spots (DBS). DBS
assay results can vary over time because of the assays used and across laboratories because of
instrumentation differences. In order to make the DBS data comparable over time and to other
population studies based on whole blood assays, HRS DBS biomarker values have been
converted into what are called NHANES (National Health and Nutrition Examination Survey)
equivalent values. Since both NHANES and HRS are intended to represent the U.S. population, the distributions of the biomarkers in HRS should be similar to those in NHANES within the
appropriate age group. The conversion makes the distribution of the values for HRS DBS assays
similar to that among NHANES respondents (Crimmins et al., 2013). For TC, HDL-C, and
HbA1c, the equivalent values in the first wave of the HRS biomarker sample (2006/2008) were
based on NHANES 2005-2006 and 2007-2008; values for HRS wave 2 (2010/2012) were based
on NHANES 2009-2010 and 2011-2012; and values for HRS wave 3 (2014/2016) were based on
NHANES 2011-2012 and 2013-2014 for HRS 2014 and 2013-2014 and 2015-2016 for HRS
2016 (Crimmins et al., 2020). This means that any time trend is similar to that observed in
NHANES. Because NHANES did not provide CRP data from 2011-2014 and because there
appeared to be a reduction in values of CRP in NHANES 2009-2010 with a high concentration at
the lower end of detection which we believe resulted from assay changes, for the current study
the NHANES equivalents for CRP for all waves were normed to NHANES 2005-2008. This

15
means there is no time trend in the adjusted CRP measure in the HRS DBS sample, but values
continue to reflect relative differences across the sample. The summary CMR measure is a count of the number of biomarkers that exceeded the
clinical high-risk thresholds for each biomarker (Supplemental Table 2.2). The high-risk
thresholds are 140 mmHg for SBP and 90 mmHg for DBP, thresholds for hypertension (Lakka et
al. 2002; Mancia et al., 2007). An RHR greater than 90 is normally defined as high-risk (Bird et
al., 2010; Hartaigh et al., 2015; Mitchell et al., 2019; Seccareccia et al., 2001; Seeman et al., 2008). WC is an indicator of abdominal obesity and chronic metabolic dysregulation (O’Keede
et al., 2019); and the thresholds defining high-risk WC are 88 cm (35 inches) for women and 102
cm (40 inches) for men (Mitchell et al., 2019; Mancia et al., 2007). For HbA1c, an indicator of
glucose metabolism and glycemic control over the past two to three months, a level higher than
6.5% is considered high risk (Mitchell et al., 2019). For TC, 240mg/dL is the high-risk cutoff
(Bird et al., 2010; Mitchell et al., 2019). HDL-C lower than 40 mg/dL is used to indicate highrisk (Bird et al., 2010; Crimmins et al., 2003; Mitchell et al., 2019). CRP is an indicator of
systemic inflammation associated with cardiovascular and metabolic diseases (Ridker, 2003). A
CRP level higher than 3 mg/L is considered high-risk (Bird et al., 2010; Mitchell et al., 2019). In addition to the eight high-risk indicators based on individual markers, we included one
more risk factor to account for additional cardiovascular risk, high pulse pressure (PP). Pulse
pressure, the difference between SBP and DBP, is an indicator of arterial stiffness, that gradually
widens after middle age posing additional risk (Ji et al., 2020; Franklin & Wong, 2016; Mancia
et al., 2007). If the SBP was greater than 140 mmHg (systolic hypertension), and the PP was
greater than 70 mmHg, one additional risk was added to the total CMR. So, our CMR measure
ranged from 0 to 9, with higher values indicating higher risk. When focusing on trends in

16
individual markers, to facilitate comparison across biomarkers, each of the markers was
standardized based on its mean and standard deviation at wave 1, and values were expressed in
z-scores.At each wave of the survey, participants were asked whether they use medication to
control blood pressure and cholesterol. For both blood pressure medication and cholesterol
medication variables, we categorized people into 5 mutually exclusive and exhaustive groups:
(Crimmins et al., 2003) Those who never used medication during the study period, (Levine, 2013)
those who started medication at wave 3, (Yang & Kozloski, 2011) those who started medication
at wave 2, (Geronimus et al., 2006) those who used medication for all three waves, and (Seeman
et al., 2008) those who used medication at some point but stopped. While medication groups
were based on three waves, we included them with the baseline characteristics. Because
questions on medications were not asked at all waves, the N was lower when medication was
included. The covariates included age, race/ethnicity, gender, current smoking, and education. All
of which were self-reported and assessed at baseline. Age was categorized into four groups: 50 to
59, 60 to 69, 70 to 79, and 80 and above. Racial/ethnic groups included non-Hispanic White, non-Hispanic Black, and Hispanic, and non-Hispanic others. Education was classified as less
than high school, high school, some college, and college degree or higher. Statistical Analysis
We used growth curve models to examine CMR changes over time and differences in
trajectories by sociodemographic characteristics. The linear multilevel equations are shown
below.

17
(1) Level one:
�푖 = �0푖 + �1푖�푖 + �푖
(2) Level two:
�0푖 = �00 + �01�푖 + �0푖
�1푖 = �10 + �11�푖 + �1푖
The subscript t indicates time and the subscript i indicates an individual respondent. Specifically, the models provided estimates of the average trajectory of CMR for the entire
sample (level one) as well as how baseline characteristics (noted by Z) were linked to variability
in baseline CMR differences and CMR changes over 8 years over time (level two). The
coefficients on “the rate of change over time”, noted as T in the equation, can be interpreted as
the change with one year of aging. In the first model, we estimated the unadjusted intercept and
slope in CMR. The second model was adjusted for age, gender, and race/ethnicity. Model 3
further controlled for education, and model 4 further controlled for current smoking status. The trajectories for each individual biomarker were graphed and compared descriptively
using z-scores. Then the growth curve model was applied to each of the biomarker z-scores to
further understand the direction and pace of its change over time. For some biomarkers (SBP, DBP, RHR, and TC), trajectories were also graphed by medication use during the study period to
examine how medication use affected the changes, and then medication groups were controlled
in individual marker models. Survey weights for the DBS sample at baseline were used to adjust for initial sample
selection and missing data. All analyses were performed using Stata version 16.

18
Results
Sample Characteristics
Supplemental Table 2.1 shows sample characteristics at baseline. The proportion with
high-risk values varied across individual biomarkers in the CMR score. For some markers like
SBP (26.4%), WC (35.5%) and CRP (34.8%), about one-third of the participants were measured
as high-risk; the high-risk percentages of other markers were lower. The largest portion of the
sample was aged 50-59 (41.9%); 35.1% were aged 60-69; 18.8% were aged 70-79; and 4.2%
aged 80 and above. More than four-fifths of the sample was non-Hispanic Whites (83.7%). About 7.1% were non-Hispanic Blacks, 6.4% were Hispanics, and 2.8% were non-Hispanic
others. Females were somewhat more than half the sample (54.0%). Only 13.2% of the
participants were current smokers. In terms of education, 13.3% did not complete high school
while 32.2% completed high school, 24.3% had some college experience, and 30.2% had a
college degree or more. Change in Total Cardiometabolic Risk over Time While Aging
Results from growth curve models of total CMR are presented in Table 2.1. Model 1 is an
unadjusted growth model and shows an average baseline CMR score of 1.97 and no evidence
that CMR increased over time. Model 2 added age, gender, and race/ethnicity. The oldest age
group, ages 80 and above, had significantly higher baseline CMR score (β = 0.46, p < 0.001)
compared to those ages 50 to 69 (Model 2). There was no difference in the rate of change across
age groups. Model 3 and Model 4 added education and current smoking status, respectively. The
inclusions did not change the results on time trends. In the full model (Model 4), compared to

19
those ages 50 to 69, those ages 80 and above had significantly higher baseline CMR score (β =
0.35, p = 0.004), and still, the rate of change did not differ across age groups. The full model also suggests that both non-Hispanic Blacks (β = 0.56, p < 0.001) and
Hispanics (β = 0.29, p = 0.011) had significantly higher CMR scores than non-Hispanic Whites. Current smokers had higher baseline CMR compared to their non-smoking counterparts (β =
0.22, p = 0.019) but did not differ in time trends. Compared to those who did not complete high
school, high school graduates (β = -0.29, p = 0.001), those who had some college experience (β =
-0.42, p < 0.001), and those who had a college degree or more (β = -0.72, p < 0.001) had
significantly lower baseline CMR. Despite the observed baseline differences within the
previously mentioned variables, the rates of change over time were not significantly different
across categories. Change in Individual Cardiometabolic Biomarkers
We next examine change in the individual CMR biomarkers. Figure 2.1 shows the
descriptive trajectory for the mean of each biomarker unadjusted for covariates; all biomarkers
were z-scored for ease of comparison. Most of the observed changes in biomarkers were
statistically significant (p-values are reported in Supplemental Table 2.3). Though the total CMR
score did not change with age, individual biomarkers did change, but in different directions and
with different magnitudes. For instance, PP, HbA1c and WC increased, while RHR, DBP, and
TC decreased. There were no statistically significant changes in SBP, HDL-C, and CRP. To better understand the trajectories of individual markers with age, growth curve models
adjusted for covariates were applied to each of the markers, and the results can be seen in Table
2.2. Over the 8 years, the increases in HbA1c (β = 0.04, p < 0.001), HDL-C (β = 0.02, p = 0.035), and PP (β = 0.04, p < 0.001, adjusted for high SBP) were statistically significant; this indicates

20
worse levels of HbA1c and PP, and improved levels of HDL. In contrast, DBP (β = -0.05, p <
0.001), RHR (β = -0.03, p = 0.002), and TC (β = -0.04, p < 0.001) decreased significantly over
time indicating improvement in risk levels with increasing age. Rates of changes in individual markers across subgroups differed. Older persons (ages
70-79 and ages 80 and above) had faster decrease in blood pressures and slower increase in WC
and PP than the youngest age group (all p-values < 0.05). For non-Hispanic Blacks, blood
pressures (SBP: β = -0.03, p = 0.011; DBP: β = -0.03, p = 0.003) dropped and HbA1c (β = 0.02, p < 0.021) increased faster, compared to non-Hispanic Whites. Women experienced slower
decrease in both indicators of blood pressure (SBP: β = 0.03, p < 0.001; DBP: β = 0.02, p <
0.001), slower increase in HbA1c (β = -0.01, p = 0.029) and CRP (β = -0.02, p < 0.001), slower
improvement in HDL-C (β = -0.03, p < 0.001), and faster increase in PP (β = 0.01, p < 0.001, adjusted for high SBP). Current smokers had a more rapid increase in WC compared to non- smokers (β = 0.01, p = 0.006). Compared to those who did not complete high school, all the
other education groups had less decline in RHR and faster WC increase (all p-values < 0.05). For
those whose SBP was higher than 140 mmHg, PP increased significantly faster (β = 0.02, p <
0.001). Interestingly, the significant decrease in WC shown in Supplemental Table 2.3 was no
longer significant after controlling for covariates in the growth curve model, indicating that the
significant change in WC was the result of differential patterns across age groups, smoking
behavior groups, and education groups. Since DBP, RHR, and TC significantly decreased over
time, and the values of these three markers can be regulated by medication, we further examined
their trajectories by medication use. The Impact of Medication Use

21
Figure 2.2 depicts the descriptive trajectories of SBP, DBP, RHR, and TC by medication
use, unadjusted for covariates. The trajectories of SBP, DBP, and RHR were categorized by
blood pressure medication groups (with a total sample size of 3,497), and the trajectory of TC
was categorized by cholesterol medication group (with a total sample size of 2,694). The size of
each medication group can be found in Supplemental Table 2.4. Most people used medication
during the study period; one-third of the respondents never used blood pressure medication, and
45% used it at all three waves. Only around one fourth of the respondents never used cholesterol
medication, while more than two fifths used it at all three waves. Figure 2.2 shows descriptively
that, in general, biomarker values dropped with medication and were elevated without it. However, people who never used blood pressure medication had relatively low SBP, DBP, and
RHR compared to the other medication groups; while people who never used cholesterol
medication had a relatively high TC level. Overall, medication use seems to have explained the
fluctuations in biomarker trajectories quite well. Since Figure 2.2 shows that medication use may have affected some biomarkers’
trajectories, we further controlled medication use in the growth curve models where the z-scores
of SBP, DBP, RHR, and TC were the dependent variables. Table 2.3 shows the comparison
between the unadjusted and adjusted models. The magnitudes of declines with age in DBP, RHR, and TC were reduced after controlling for medication. Specifically, the coefficient of DBP
(unadjusted: β = -0.049, p < 0.001; adjusted: β = -0.024, p = 0.022) and TC (unadjusted: β = - 0.043, p < 0.001; adjusted: β = -0.026, p = 0.047) were cut in half, and the decrease in RHR
(unadjusted: β = -0.032, p = 0.002; adjusted: β = -0.021, p = 0.038) was reduced by one third. Compared to those who never used medication, almost all other medication groups for SBP, DBP, and TC experienced faster decreases over time with aging (almost all p-values < 0.05). The

22
rates of change did not differ a lot across medication groups for RHR, but clearly those who took
medication all the time had a faster decrease relative to those who never took medication (β = - 0.017, p = 0.004). Sensitivity Analysis
We noted above that people who died during the study period and those who survived but
had missing data were not included in our analytic sample. We conducted additional analyses to
determine if limiting the sample to survivors observed in all three periods produces a different
pattern of findings than using a sample that also includes those who were missing some data or
died in the follow-up period. The comparison of total CMR models (Supplemental Table 2.5) as
well as the biomarker z-score models (Supplemental Tables 2.6 and 2.7) across different samples
shows that when those who died and those who missed one wave were included, the total CMR
significantly decreased over time when including those who died (β = -0.013, p < 0.001) and
when both the decedents and those who missed one wave were included (β = -0.012, p < 0.001). However, the decreases were no longer significant after controlling for covariates. The
biomarker trajectories do not change much after using new samples (Supplemental Tables 2.6
and 2.7) except for that the overtime increase in HDL was no longer significant. Discussion
In this representative national sample with a mean age of 63 at baseline, who survived
and aged over 8 years from 2006/2008 to 2014/2016, we found no evidence of age-related
change in total CMR, and no difference in CMR rate of change across population subgroups. Among the biomarkers included in CMR, risk from glycosylated hemoglobin (HbA1c), waist

23
circumference (WC), and pulse pressure (PP) increased significantly over time with aging. In
contrast, diastolic blood pressure (DBP), resting heart rate (RHR), and total cholesterol (TC)
decreased significantly over the eight years. The mixed models on each of the biomarkers
indicated that the decreases in DBP, RHR, and TC were still significant after controlling for age, race/ethnicity, gender, smoking, and education. The decreases were at least partially explained
by medication use. Although the aging rate of CMR appears relatively constant, it conceals differential aging
in individual biomarkers. Examining the trajectory of each cardiometabolic biomarker in
addition to the CMR summary measure provided us with more nuanced understanding of how
CMR changes with aging over time. Risks of individual biomarkers did not all change in the
same direction or by the same magnitude. This suggests complex associations among
physiological systems included in CMR. Our study design allowed us to contribute to the existing literature. In the past, studies
using cross-sectional data (Crimmins et al, 2003; Yang & Kozloski, 2011) were not able to
differentiate observed age differences from cohort effects or mortality effects, and studies using
longitudinal data were limited either by representativeness (Belsky, 2015; Karlamangla et al., 2006; Merkin et al., 2014) or by a short time frame (Mitchell et al., 2019). However, we showed
how CMR changed with age longitudinally over 8 years of aging in a sample representative of
older Americans who survived. The estimated total CMR was relatively constant over 8 years among people with an
initial average age of 63 in this period; a conclusion which differs from the increase by age
observed in a nationally representative cross-sectional sample in an earlier period (Crimmins et
al, 2003). Due to the longitudinal nature of our data, we were able to assess the within person

24
changes while aging, and by only including the participants who responded in all three waves, our results were not influenced by mortality selection out of the sample. Our results also differ
from the increase in risk with aging found by previous longitudinal studies using
nonrepresentative samples (Belsky, 2015; Karlamangla et al., 2006; Merkin et al., 2014). This
may reflect differences between the national sample we used and less representative community
samples. Moreover, our results reflect a more recent time period when the use of efficacious
medications was spreading rapidly through the population. Our results also showed that despite the decrease in blood pressures, heart rate, and
cholesterol with aging, the constant total CMR resulted from increases in metabolic markers like
HbA1c and WC. Medication use has reduced the CMR and balanced out the changes linked to
metabolism and obesity in a relatively short period of time. Medication can be recognized as
slowing the aging process in terms of cardiometabolic health during this period. We should note that by limiting our sample to those who had complete data for all three
waves, we analyzed aTable slightly healthier and younger sample compared to those who were
in the sample at baseline, those who died, and those who were missing data at later waves
(Supplemental 2.1). Including those who died or were missing for a wave resulted in baseline
CMR difference across age groups becoming more significant, likely due to the fact that those
who died or who failed to complete the three waves of the survey were older. Also, when these
persons were included, the total CMR decreased overtime in unadjusted models as, in general, it
was people with higher CMR who died and were missing (Supplemental Table 2.5). But the
decreases were no longer significant after controlling for age, gender, race/ethnicity, education, and smoking status, and the pattern of the results of the full models did not change much. So, we
believe that including factors related to mortality and data missing in the models has partially

25
accounted for the sample selection, and our results should be seen as representative of the U.S. community-dwelling older population who survived for 8 years. Our growth curve models reveal how baseline characteristics were associated with initial
levels and rates of change. The results indicated relatively higher baseline risk among older age
groups, racial/ethnic minorities, current smokers, and people with lower education level, which
were consistent with previous studies (Crimmins et al., 2009; Levine & Crimmins, 2014;
Mitchell et al., 2019; Seeman et al., 2007). Our findings confirmed the effectiveness of
medication in regulating blood pressure, heart rate, and cholesterol in reducing CMR (Crimmins
et al., 2010; Kim et al., 2019). While we found that non-Hispanic Blacks and Hispanics had higher baseline CMR, our
results indicate that the rates of change across racial/ethnic groups did not differ. Our findings
differed from those reported in a study by Mitchell et al. (2019), which also used the HRS, but
found an increase in CMR over a four-year period (2006-2010/2008-2012) among non-Hispanic
Blacks. There are two key study design differences that account for our discrepant findings. First, Mitchell et al. used a different CMR summary score from ours. Our CMR score included pulse
pressure as well as SBP and DBP, while the CMR score in Mitchel et al. only included pulse
pressure. While pulse pressure is a good indicator of arterial stiffness, we believe that using PP
alone to reflect blood pressure in a summary CMR is not sufficient (Franklin & Wong, 2016;
Kim & Crimmins, 2020; Mosley, 2007; Prospective studies collaboration, 2002). Recent
literature has suggested treating high PP as an additional risk under the condition of high systolic
blood pressure (Butler et al., 2011; Haider et al., 2003; Mancia et al., 2007; Pastor-Barriuso et al., 2003). So, in the current paper, we included both the traditional hypertension indicators (systolic
blood pressure higher than 140 and diastolic blood pressure higher than 90) and added one point

26
to the CMR index if the SBP is higher than 140 mmHg and the PP is higher than 70 mmHg. We
believe the new CMR measure better reflects cardiometabolic risk. Second, our sample is limited
to those who had complete data for all three waves rather than just two waves, as in Mitchell et
al., resulting in a somewhat healthier sample in our analysis that is likely to have experienced
slower cardiometabolic deterioration. When we replicated our analysis using the CMR measure employed in the Mitchell et al. paper, we found an increased CMR over the 8-year study period but this increase did not differ
by race/ethnicity and the final model with all controls did not indicate an increase over time
similar to our results (Supplemental Table 2.8). We attribute the difference in the overall time
change results to the difference in the trends in the blood pressure related indicators: increase in
pulse pressure, lack of change in SBP, and decrease in DBP. In supplemental analysis using
methods similar to Mitchell et al. which allowed us to compare both the measures and the
samples in the two studies (Supplemental Figures 2.1, 2.2), we found that the slopes of
racial/ethnic trajectories did not differ much based on CMR measures, but the increasing trend
for Blacks was attenuated when we only included those who had complete data for all three
waves. This leads us to believe that the differential results on the racial change between two
papers are primarily due to sample selection and the requirement for longer survival. Our study has limitations. First, though we included a longer period of observation than
had been available previously, we were not able to study trajectories over more than 8 years of
aging. It will be useful to extend this analysis to a longer time period in the future. Second, the
number of biomarkers collected from the DBS assays was limited. However, in 2016, HRS
started collecting venous blood from respondents, allowing a wider range of biomarkers in the

27
future. It means that the future study informed by new data may provide a fuller picture of the
trajectory of physiological deterioration. Tables/Figures
Table 2.1. Results of the Growth Curve Model Predicting Total Cardiometabolic Risk
N = 3,528 Model 1 Model 2 Model 3 Model 4
Baseline CMR 1.972*** 1.815*** 2.332*** 2.272*** Rate of Change over Time -0.003 -0.004 -0.010 -0.009
Age Groups - Reference: Ages 50-60
Baseline
Ages 60-69 0.093 0.056 0.066
Ages 70-79 0.108 0.024 0.050
Ages 80 and above 0.458*** 0.325** 0.351** Rate of Change over Time
Ages 60-69 0.001 0.001 0.001
Ages 70-79 -0.006 -0.005 -0.005
Ages 80 and above -0.031 -0.029 -0.029
Gender - Reference: Males
Baseline
Females -0.001 -0.048 -0.042
Rate of Change over Time
Females 0.008 0.009 0.009
Racial/Ethnic Groups - Reference: Non-Hispanic White
Baseline
Non-Hispanic Black 0.709*** 0.572*** 0.563*** Hispanic 0.519*** 0.270* 0.286* Others 0.079 0.053 0.034
Rate of Change over Time
Non-Hispanic Black -0.011 -0.009 -0.009
Hispanic -0.018 -0.015 -0.015
Others 0.020 0.020 0.020
Education - Reference: Less than High School
Baseline
High School -0.304*** -0.290*** Some College -0.435*** -0.416*** College and Higher -0.754*** -0.722*** Rate of Change over Time
High School 0.005 0.004
Some College -0.000 -0.000
College and Higher 0.011 0.011
Currently Smoke - Baseline 0.215* Currently Smoke - Rate of Change -0.003
Log Likelihood -20113.876 -20035.495 -19948.422 -19940.118
LR Test P 0.000 0.000 0.000
Note. 1 The likelihood ratio test compares the current model with the previous model. * p<0.05 ** p<0.01 *** p<0.001

28
Table 2.2. Results of the Growth Curve Model Predicting Biomarker Z-Scores
N = 3,528 SBP DBP CRP TC A1c HDL-C RHR WC
PP
Adjusted
for SBP Baseline CMR 0.126 0.216** -0.054 -0.076 0.089 -0.536*** 0.092 0.631*** -0.365*** Rate of Change over Time -0.004 -0.049*** 0.008 -0.043*** 0.040*** 0.021* -0.032** 0.008 0.035*** Age Groups - Reference: Ages 50-59 Baseline Ages 60-69 0.207*** -0.075 0.020 -0.079 0.089 0.014 -0.131** 0.024 0.322*** Ages 70-79 0.423*** -0.241*** -0.027 -0.294*** 0.080 -0.039 -0.294*** -0.039 0.702*** Ages 80 and above 0.869*** -0.231** 0.212 -0.398*** 0.066 -0.084 -0.344*** -0.183* 1.111*** Rate of Change over Time Ages 60-69 -0.002 -0.006 -0.003 -0.008 -0.012* 0.000 0.004 -0.009* -0.002 Ages 70-79 -0.018* -0.016* 0.007 -0.002 -0.009 0.004 0.009 -0.019*** -0.012* Ages 80 and above -0.049** -0.027* -0.031 0.001 -0.013 0.013 0.034* -0.031*** -0.032** Gender - Reference: Males Baseline Females -0.353*** -0.142*** 0.170*** 0.288*** -0.058 0.634*** 0.129** -0.601*** -0.290*** Rate of Change over Time Females 0.026*** 0.019*** -0.021*** 0.000 -0.011* -0.028*** -0.007 0.001 0.013*** Racial/Ethnic Groups - Reference: Non-Hispanic White Baseline Non-Hispanic Black 0.342*** 0.273*** 0.145** -0.017 0.438*** 0.042 0.197* 0.213*** 0.162*** Hispanic 0.143 0.042 -0.011 0.065 0.371** -0.049 -0.040 -0.075 0.113* Others 0.118 0.168 -0.047 -0.107 0.500** -0.062 0.057 -0.226* 0.030 Rate of Change over Time Non-Hispanic Black -0.028* -0.033** -0.002 -0.010 0.022* -0.017 0.000 -0.004 -0.005 Hispanic -0.002 -0.003 0.002 -0.013 0.000 -0.012 0.004 -0.002 0.000
Others 0.005 -0.003 0.010 0.014 0.003 0.021 0.014 -0.009 -0.002 Education - Reference: Less than High School Baseline High School -0.123* -0.071 -0.044 -0.046 -0.127 0.102 -0.103 -0.176** -0.150*** Some College -0.228*** -0.105 -0.039 0.032 -0.264*** 0.197*** -0.157* -0.264*** -0.169*** College and Higher -0.292*** -0.180** -0.178*** -0.031 -0.258*** 0.373*** -0.179** -0.491*** 0.000 Rate of Change over Time High School 0.000 0.009 -0.008 0.013 0.010 -0.005 0.029*** 0.020*** -0.006
Some College 0.004 0.015 -0.009 0.000 0.002 -0.002 0.017* 0.012* -0.008
College and Higher 0.011 0.017 -0.004 0.024* -0.005 0.000 0.019* 0.020*** 0.000
Currently Smoke - Baseline 0.168** 0.169** 0.057 -0.052 -0.027 -0.115 0.331*** -0.214*** 0.017
Currently Smoke - Rate of Change -0.016 -0.010 0.007 0.012 0.009 0.000 -0.015 0.013** -0.007 High SBP - Baseline - - - - - - - - 1.166*** High SBP - Rate of Change - - - - - - - - 0.019*** Log Likelihood -16053.984 -16174.462 -15544.130 -16861.539 -15316.554 -16225.407 -16443.457 -11139.088 -12599.278 Note. SBP: Systolic blood pressure; DBP: Diastolic blood pressure; PP: Pulse Pressure; RHR: Resting heart rate; HbA1c: Glycosylated hemoglobin; HDL-C: High-density lipoprotein cholesterol; TC: Total cholesterol; WC: Waist circumference; CRP: C-reactive protein. * p<0.05 ** p<0.01 *** p<0.001
Table 2.3. The Growth Curve Model Predicting Biomarker Z-Scores Controlling for Medication
No Medications With Medications SBP
1 DBP
1 RHR 1 TC 1 SBP
2 DBP
2 RHR 2 TC 3 Baseline CMR 0.126 0.216** 0.092 -0.076 -0.233*** -0.103 0.027 -0.023
Rate of Change over Time -0.004 -0.049*** -0.032** -0.043*** 0.017 -0.024* -0.021* -0.026* Age Groups - Reference: Ages 50-60
Baseline Ages 60-70 0.207*** -0.075 -0.131** -0.079 0.142*** -0.126** -0.136** -0.058
Ages 70-80 0.423*** -0.241*** -0.294*** -0.294*** 0.344*** -0.305*** -0.302*** -0.226*** Ages 80 and above 0.869*** -0.231** -0.344*** -0.398*** 0.769*** -0.316*** -0.349*** -0.365*** Rate of Change over Time Ages 60-70 -0.002 -0.006 0.004 -0.008 0.001 -0.002 0.007 -0.003
Ages 70-80 -0.018* -0.016* 0.009 -0.002 -0.016* -0.012 0.013 0.001
Ages 80 and above -0.049** -0.027* 0.034* 0.001 -0.048** -0.024 0.038** 0.005
Gender - Reference: Males Baseline Females -0.353*** -0.142*** 0.129** 0.288*** -0.343*** -0.133*** 0.136*** 0.273***

29
Rate of Change over Time Females 0.026*** 0.019*** -0.007 0.000 0.024*** 0.017** -0.008 -0.001
Racial/Ethnic Groups - Reference: Non-Hispanic White Baseline Non-Hispanic Black 0.342*** 0.273*** 0.197* -0.017 0.222** 0.173* 0.189* 0.012
Hispanic 0.143 0.042 -0.040 0.065 0.113 0.006 -0.046 0.101
Others 0.118 0.168 0.057 -0.107 0.106 0.164 0.063 -0.117
Rate of Change over Time Non-Hispanic Black -0.028* -0.033** 0.000 -0.010 -0.022* -0.025* 0.003 -0.014
Hispanic -0.002 -0.003 0.004 -0.013 -0.001 0.000 0.003 -0.014
Others 0.005 -0.003 0.014 0.014 0.002 -0.006 0.014 0.015
Education - Reference: Less Than High School
Baseline High School -0.123* -0.071 -0.103 -0.046 -0.081 -0.032 -0.085 -0.021
Some College -0.228*** -0.105 -0.157* 0.032 -0.160** -0.042 -0.147* 0.019
College and Higher -0.292*** -0.180** -0.179** -0.031 -0.197*** -0.095 -0.153* -0.066
Rate of Change over Time High School 0.000 0.009 0.029*** 0.013 -0.003 0.006 0.026** 0.011
Some College 0.004 0.015 0.017* 0.000 0.002 0.011 0.014 -0.005
College and Higher 0.011 0.017 0.019* 0.024* 0.005 0.011 0.015 0.016
Currently Smoke - Baseline 0.168** 0.169** 0.331*** -0.052 0.176** 0.169** 0.327*** -0.072
Currently Smoke - Rate of Change -0.016 -0.010 -0.015 0.012 -0.016 -0.010 -0.016 0.021
Medication - Reference: Never Used medication Baseline Started Medication at Wave 3 0.555*** 0.584*** 0.094 0.406*** Started Medication at Wave 2 0.811*** 0.771*** 0.184* 0.241** Medication All Waves 0.549*** 0.450*** 0.044 -0.445*** Stopped Medication at Some Point 0.357*** 0.342*** 0.148 -0.178* Rate of Change While Aging
Started Medication at Wave 3 -0.026* -0.041*** 0.005 -0.094*** Started Medication at Wave 2 -0.083*** -0.083*** -0.017 -0.114*** Medication All Waves -0.027*** -0.035*** -0.017** -0.005
Stopped Medication at Some Point 0.006 -0.004 -0.022 0.026* Log Likelihood -16053.984 -16174.462 -16443.457 -16861.539 -15681.778 -15860.050 -16309.364 -11755.672 Note. 1 The models do not control for medication. N = 3,528
2 The models control for blood pressure medication. N = 3,497
3 The models control for cholesterol medication. N = 2,694
SBP: Systolic blood pressure; DBP: Diastolic blood pressure; RHR: Resting heart rate; TC: Total cholesterol. * p<0.05 ** p<0.01 *** p<0.001 (two-sided t test)
Figure 2.1. Biomarker Z-Score Trajectories

30
Figure 2.2. Trajectories of Individual Biomarkers by Medication Group

31
Supplemental Information
Supplemental Table 2.1. Sample Characteristics at Baseline
People with complete data for all 3 waves
(Reference Group)
People with complete data at baseline People who did not die but had missing data People who died
n = 3,528 n = 9,173 n = 3,503 n = 2,142
Percentage, Mean Percentage, Mean Percentage, Mean Percentage, Mean High-Risk CMR Factors Total CMR Mean 1.75 1.90*** 2.15*** 2.20*** Mean Age 63.28 65.80*** 64.20*** 74.56***
50-60 41.93 34.40*** 38.07** 10.64***
60-70 35.13 33.49* 37.37 21.82***
70-80 18.76 20.12** 16.88* 29.81***
80 and above 4.18 11.99*** 7.68*** 37.73*** Race/Ethnicity
Non-Hispanic White 83.66 81.89** 79.14*** 83.9
Non-Hispanic Black 7.1 8.63*** 9.94*** 9.15* Hispanic 6.4 6.98 8.31** 5.44
Others 2.84 2.49 2.61 1.50** Gender Female 53.95 53.91 55.69 50.08** Male 46.05 46.09 44.31 49.92** Currently Smoke 13.15 14.93** 14.78 19.04*** Education Less than High School 13.32 17.80*** 17.77*** 27.43***

32
High School 32.15 32.75 31.56 36.53** Some College 24.28 23.37 23.65 20.86* College and above 30.25 26.08*** 27.03* 15.18*** Note. The * indicates significant dif erences (t-test) between the reference group (people had complete data for all 3 waves) and the
specified group. Supplemental Table 2.2. The High-Risk Thresholds for CMR Risk Factors
CMR Variables High Risk Group
Cardiovascular Markers
Systolic Blood Pressure (mmHg) ≥ 140
Diastolic Blood Pressure (mmHg) ≥ 90
Pulse Pressure Adjustment If SBP ≥ 140 & PP ≥ 70, CMR+1
Resting Heart Rate (bpm) ≥ 90
Metabolic Markers
Glycated Hemoglobin (%) ≥ 6.5
High-Density Lipoprotein (mg/dL) ≤ 40
Total Cholesterol (mg/dL) ≥ 240
Waist Circumference (inch) ≥ 35 (women); ≥ 40 (men)
Inflammatory Marker C-Reactive Protein (mg/L) ≥ 3
Note. CMR: Cardiometabolic Risk; SBP: Systolic blood pressure; PP: Pulse pressure. Supplemental Table 2.3. Means of Biomarker Z-Score at 3 Waves and T-Tests for the
Differences in Mean Scores across Waves
Wave 1 p Wave 2 p Wave 3
n = 3,528 n = 3,528 n = 3,528
SBP 0 0.481 0.014 0.494 0.028
DBP 0 0.000 *** -0.138 0.000 *** -0.305
RHR 0 0.000 *** -0.121 0.501 -0.107
A1c 0 0.000 *** 0.097 0.000 *** 0.261
HDL 0 0.011 * -0.053 0.000 *** 0.029
TC 0 0.000 *** -0.267 0.759 -0.275
Waist 0 0.000 *** 0.086 0.000 *** 0.133
CRP 0 0.001 ** -0.074 0.842 -0.070
PP 0 0.000 *** 0.143 0.000 *** 0.313
PP among Those Who Have Risky SBP 0 0.004 ** 0.129 0.000 *** 0.398
Note. SBP: Systolic blood pressure; DBP: Diastolic blood pressure; PP: Pulse pressure; RHR: Resting heart rate; HbA1c: Glycosylated hemoglobin; HDL-C: High-density lipoprotein cholesterol; TC: Total cholesterol; WC: Waist circumference; CRP: C-reactive protein. The sample size for the last row (those who have risky SBP) is 987 for wave 1, 1,038 for wave 2, and 1,004
for wave 3.

33
* p<0.05 ** p<0.01 *** p<0.001
Supplemental Table 2.4. SBP, DBP, RHR, and TC Mean Values by Wave and Medication Group
Biomarkers Medication groups Never Used
Medication Started
Medication at
Wave 3
Started
Medication at
Wave 2
Medication All
Waves Stopped
Medication Some Point
DBP (n=3,497) n = 1193 n = 216 n = 302 n = 1573 n = 213
Wave 1 Mean 77.17 82.89 86.70 81.86 80.98
Std. Dev. 9.74 10.47 12.51 10.97 11.06
Wave 2 Mean 77.09 83.15 80.37 79.64 79.30
Std. Dev. 10.64 10.81 10.47 11.31 12.63
Wave 3 Mean 76.13 78.10 78.01 77.39 79.08
Std. Dev. 10.86 10.74 11.63 10.69 13.30
SBP (n=3,497) n = 1193 n = 216 n = 302 n = 1573 n = 213
Wave 1 Mean 121.75 130.86 139.68 133.93 131.92
Std. Dev. 15.70 17.21 20.06 18.54 20.46
Wave 2 Mean 123.98 136.42 131.07 133.18 131.91
Std. Dev. 17.52 19.36 17.56 19.02 21.73
Wave 3 Mean 125.55 130.61 130.49 133.06 135.51
Std. Dev. 18.01 16.90 19.37 19.09 21.39
RHR (n=3,497) n = 1193 n = 216 n = 302 n = 1573 n = 213
Wave 1 Mean 69.91 71.48 72.43 70.09 71.01
Std. Dev. 10.45 11.38 10.43 11.35 12.27
Wave 2 Mean 69.03 69.85 69.85 68.67 69.97
Std. Dev. 10.39 11.80 10.59 11.52 12.83
Wave 3 Mean 69.46 71.42 70.53 68.36 68.83
Std. Dev. 10.70 21.71 13.45 11.51 10.04
TC (n=2,694) n = 752 n = 168 n = 286 n = 1162 n = 326
Wave 1 Mean 212.06 222.11 224.82 190.52 201.62
Std. Dev. 38.10 40.27 43.88 35.69 42.00
Wave 2 Mean 204.52 215.32 185.74 181.76 197.32
Std. Dev. 46.86 38.43 39.31 38.36 45.31
Wave 3 Mean 206.00 185.19 181.00 182.77 203.62
Std. Dev. 40.02 36.68 36.60 35.56 42.41
Note. Data at wave 1 were collected in 2006/2008. Data at wave 2 were collected in 2010/2012. Data at wave 3 were collected in
2014/2016. For SBP, DBP, and RHR, respondents were categorized by blood pressure medication groups. For TC, respondents were
categorized by cholesterol medication groups. The sample size was determined by how many people have complete biomarker (SBP/DBP/RHR/TC-specific) and complete medication (blood pressure/cholesterol-specific) use data for all three waves. SBP: Systolic blood pressure; DBP: Diastolic blood pressure; RHR: Resting heart rate; HbA1c: Glycosylated hemoglobin;
HDL-C: High-density lipoprotein cholesterol; TC: Total cholesterol; WC: Waist circumference; CRP: C-reactive protein.

34 Supplemental Table 2.5. CMR Model Results Using All-Wave Participants and Models Including Those Who Died and Those Who Missed Only One Wave

35 Supplemental Table 2.6. Comparison between Biomarker Z-Score Models Using All-Wave Participants and Models Including Those Who Died

36 Supplemental Table 2.7. Comparison between Biomarker Z-Score Models Using All-Wave Participants and Models Including Those Who Died and Those Who Missed Only One Wave

37
Supplemental Table 2.8. Comparison between Total CMR Models Using Two Measures: Our
Proposed Measure vs. Mitchell et al. N = 3,528 Original CMR Measure Mitchell et al. Measure Model 1 Model 2 Model 3 Model 4 Model 1 Model 2 Model 3 Model 4
Baseline CMR 1.972*** 1.815*** 2.332*** 2.272*** 1.635*** 1.443*** 1.852*** 1.830*** Rate of Change over Time
-0.003 -0.004 -0.010 -0.009 0.012*** 0.015* 0.012 0.012
Age Groups - Reference: Ages 50-60
Baseline Ages 60-69 0.093 0.056 0.066 0.120* 0.089 0.092
Ages 70-79 0.108 0.024 0.050 0.171** 0.103 0.112* Ages 80 and above 0.458*** 0.325** 0.351** 0.342*** 0.233* 0.242** Rate of Change over Time Ages 60-69 0.001 0.001 0.001 0.001 0.001 0.001
Ages 70-79 -0.006 -0.005 -0.005 -0.004 -0.004 -0.004
Ages 80 and above -0.031 -0.029 -0.029 -0.016 -0.016 -0.015
Gender - Reference: Males Baseline Females -0.001 -0.048 -0.042 0.077 0.036 0.039
Rate of Change over Time Females 0.008 0.009 0.009 -0.005 -0.004 -0.004
Racial/Ethnic Groups - Reference: Non-Hispanic White Baseline Non-Hispanic Black 0.709*** 0.572*** 0.563*** 0.524*** 0.413*** 0.410*** Hispanic 0.519*** 0.270* 0.286* 0.402*** 0.203* 0.209* Others 0.079 0.053 0.034 -0.045 -0.065 -0.072
Rate of Change over Time Non-Hispanic Black -0.011 -0.009 -0.009 0.010 0.010 0.010
Hispanic -0.018 -0.015 -0.015 -0.007 -0.005 -0.005
Others 0.020 0.020 0.020 0.024 0.024 0.023
Education - Reference: Less than High School
Baseline High School -0.304*** -0.290*** -0.221** -0.216** Some College -0.435*** -0.416*** -0.334*** -0.327*** College and Higher -0.754*** -0.722*** -0.613*** -0.602*** Rate of Change over Time High School 0.005 0.004 0.003 0.003
Some College -0.000 -0.000 -0.003 -0.003
College and Higher 0.011 0.011 0.005 0.006
Currently Smoke - Baseline 0.215* 0.078
Currently Smoke - Rate of Change -0.003 0.003
Log Likelihood
- 20113.87
6
- 20035.49
5
- 19948.42
2
- 19940.11
8
- 17420.01
4
- 17335.30
9
- 17243.79
7
- 17241.52
2
LR Test P 0.000 0.000 0.000 0.000 0.000 0.000
Supplemental Figure 2.1. Comparison Across Poisson Regression Model Predicted Outcomes
Using Different Samples: CMR as Measured in Mitchell et al. and Samples Present in 4-year and
8-year Analyses

38
Supplemental Figure 2.2. Comparison Across Poisson Regression Model Predicted Outcomes
Using Different CMR Measures (Mitchell et al. is 7 items and this paper is 9 items) and Samples
(4-year and 8-year Sample).

39
Chapter 3: The Association between Cardiometabolic Risk and
Cognitive Function among Older Americans and Chinese
Introduction
Population aging and rising life expectancy are contributing to the increasing prevalence
of dementia around the globe. Worldwide dementia cases are likely to double in 20 years, and
the associated costs are estimated to double in 10 years (Prince et al., 2015). In both the US and
China, the elevating care needs and costs have made dementia an increasing public health
challenge (Wang et al., 2019; Zissimopoulos et al., 2019), requiring attention from policymakers, healthcare professionals, and researchers. Cardiometabolic risk (CMR) factors have been negatively associated with cognitive
health and the neurodegenerative process (Santos et al., 2017), and they are considered among
the most important risk factors for cognitive problems. At the same time, CMR factors have been
the focus of interventions to maintain cognitive health because they can be modified through
medical treatment and behavioral change. As implied by most previous work, elevated levels of
CMR indicators, reflecting vascular aging, metabolic syndrome, and chronic inflammation, may
increase the risk of cognitive decline through pathways including amyloid-beta deposition, cerebral hypoperfusion, and neurovascular unit dysfunction (Santos et al., 2017). Specifically, previous studies based on data from multiple countries suggest that higher levels of systolic
blood pressure (SBP) (Beck et al., 2022; Chanti-Ketterl et al., 2022; Hu et al., 2020; Wei et al., 2018), diastolic blood pressure (DBP) (Hu et al., 2020; Hughes et al., 2020), pulse pressure (PP)
(Sha et al., 2018; Wei et al., 2018), resting heart rate (RHR) (Beck et al., 2022; Hu et al., 2020),

40
glycosylated hemoglobin (HbA1c) (Hu et al., 2020; Ortiz et al., 2022), and c-reactive protein
(CRP) (Alley et al., 2008; Beck et al., 2022), are linked to worse cognitive performance, while
higher levels of high-density lipoprotein cholesterol (HDL) (Chanti-Ketterl et al., 2022;
Svensson et al., 2019) link to better performance. Research has also shown that higher levels of
body mass index (BMI) (Anstey et al., 2011; Pegueroles et al., 2018), waist circumference (WC)
(Liu et al., 2019; Waki et al., 2020), and waist-to-hip ratio (Beck et al., 2022; Liu et al., 2019) are
negatively associated with cognitive function. The relationship between late-life total cholesterol
(TC) and cognitive function is less clear. Recent meta-analyses do not produce consistent
significant associations between late-life TC and cognitive outcomes (Peters et al., 2021) but
suggest that mid-life TC might be a risk factor for late-life dementia (Anstey et al., 2017). Since cardiometabolic markers are often highly correlated, and since they might have
additive effects on health outcomes, summary indicators of cardiovascular, metabolic, and
inflammatory biomarkers, have been used in existing literature to measure aggregate
physiological health and to understand its association with cognitive performance. In many
places around the globe but mainly in high-income countries, a worse CMR summary score has
been cross-sectionally linked to worse cognitive function (Chanti-Ketterl et al., 2022; D’Amico
et al., 2020; Narbutas et al., 2019; Zhou et al., 2021), and longitudinally linked to an overall
cognitive decline (Crook et al., 2018; Karlamangla et al., 2002; Oi & Haas, 2019; Schmitz et al., 2018). However, a longitudinal perspective on the CMR measure is also needed: Since CMR
indicators are the target of interventions and can be modified through medication use, it is
particularly interesting to see whether a change in CMR is accompanied by a simultaneous
change in cognitive function, or whether controlling CMR and preventing its increase can
preserve cognitive health, which is particularly important for countries going through rapid

41
changes in chronic disease management and treatment. While a longitudinal increase in some
CMR factors like HbA1c (Ravona-Springer et al., 2012), WC (Rodríguez-Fernández et al., 2017), and PP (Waldstein et al., 2008) have been associated with worse cognitive outcomes, to our
knowledge, how a change in CMR summary measure is associated with cognitive performance
and cognitive change is yet to be studied due to the lack of longitudinal data. Our hypothesis is
that an increase in CMR will be associated with not only worse cognitive ability but also greater
cognitive decline. However, the relative importance of CMR in the cognitive function of older adults may
depend on other co-occurring risks, largely influenced by broader social, economic, and medical
contexts. The US and China provide a useful comparison for understanding how the relative
importance of CMR for cognitive health varies across contexts. For instance, while a higher
CMR is typically associated with worse cognitive outcomes in the US (Chanti-Ketterl et al., 2022; D’Amico et al., 2020), some previous studies have found summary measures of
cardiometabolic-neuroendocrine biomarkers (Zhou et al., 2021) and metabolic syndrome (Xue &
Niu, 2016) to be unrelated to cognitive function scores among the general Chinese population. These contrasting findings might be attributed to differences in samples and variations in the
measurement of CMR and cognitive function, but the differential CMR-cognition association
between the two countries might also reflect fundamental differences in developmental contexts. To better understand whether and why the link between CMR and cognition might differ
between the two countries, comparable population-representative data based on similar study
designs are needed. In addition to CMR, education is another important predictor of cognitive health: A
higher education level is associated with better cognitive outcomes in both the U.S. (Crimmins et

42
al., 2018) and China (Lei et al., 2014). This protective effect is explained by two main theories
(Juul Rasmussen & Frikke-Schmidt, 2023): One theory considers higher education as a proxy for
a high socioeconomic status (SES). In this case, the role that education plays in the association
between CMR and cognitive health may differ in the US and China since high-SES populations
may have different characteristics. Another theory considers higher education as a source of
cognitive reserve, providing a direct protective effect by influencing brain structure. Around
2014, the percentage of older adults with less than secondary education was less than 15% in the
US but about 85% in China (Program on Global Aging, Health & Policy, University of Southern
California, 2023). The different levels of educational attainment could result in different levels of
cognitive reserve and may lead to differential late-life cognitive outcomes. Hence, it will be
interesting to see whether and how education explains the potential discrepancies in cognitive
function between the US and China. In the current study, we aim to answer the following questions: (i) How does CMR
associate with cognitive function cross-sectionally? (ii) How does CMR change link to cognitive
function and cognitive decline? (iii) Do the previously examined associations differ between the
US and China and why? We hypothesize that cross-sectionally, higher CMR will link to worse
cognitive function, and longitudinally, CMR increase will be associated with both worse
cognitive function, and greater cognitive decline. Similarly, for individual biomarkers, we
hypothesize that those with high-risk values will have worse cognitive function. We also
hypothesize that the associations between CMR and cognitive function might differ between the
two countries, and the differences could be potentially explained by socioeconomic factors such
as education level. The data we use from the two countries are highly comparable and nationally
representative and thus provide unique sources for international comparisons. The datasets from

43
both countries include longitudinal measurements for both CMR and cognitive function, which
have not previously been available. Methods
Data & Samples
Data on older Americans used in the current study come from the Health and Retirement
Study (HRS), a nationally representative longitudinal survey every two years of US adults over
age 50. Beginning in 2006, at each wave half of the sample has anthropometric measurements
taken, and dried blood spots (DBS) collected and subsequently assayed. Data used here for
baseline were collected from half of the sample in 2010, and in 2012 for the other half; follow-up
measures were collected in 2014, and 2016 (Crimmins et al., 2017). Data for the Chinese come from the China Health and Retirement Longitudinal Study
(CHARLS), which is a nationally representative longitudinal study surveying Chinese residents
ages 45 and older. Its survey design is intentionally similar to that of HRS to facilitate cross- country comparison. Venous blood samples and anthropometric measures biomarkers were first
collected in 2011 which is our baseline and again in 2015 for the follow-up (Zhao et al., 2013). We limit the CHARLS respondents to ages 50 and older to match the age range in the two
samples. The timeline of the US sample and the Chinese sample align fairly well as the baseline
and follow-up for the two samples overlap (2010/2012 vs 2011; 2014/2016 vs 2015), and the
interval is 4 years for both. To use the maximum sample size for each analysis, a smaller sample
is used for longitudinal analyses and a larger sample for cross-sectional analyses. Among the 8,393 respondents who participated in the HRS 2010/2012 biomarker
collection, 7,430 had complete biomarker data needed for the analyses. Since the HRS cognitive

44
variables had already been imputed, no one is missing on the selected cognitive function
measures. The US cross-sectional sample also excludes those who had missing values for age, gender, and education resulting in a final size of 7,413. Among those who had complete baseline
biomarker and cognitive function data, only 4,481 had those measures at the follow-up wave. Among those who did not have complete data at the follow-up, 1,149 missed some of the
biomarkers or cognitive test results; 878 died before the follow-up wave; 910 were not
interviewed; and 12 did not have any biomarker or cognitive measures, but their vital status was
not documented. After excluding those missing age, gender, and education, the final size of the
US longitudinal sample is 4,474. Both the longitudinal and cross-sectional samples are
significantly younger and have higher education levels than the original sample, but the
differences are small (eTable 3.1). Among the initial Chinese sample over age 50 of 9,297 respondents in the 2011 blood
collection, 7,700 had complete biomarker data and of these, 6,109 had complete data for the
cognitive function measures. Among them, 6,108 had complete data for age, gender, and
education, and comprise the Chinese cross-sectional sample. Among the cross-sectional sample, 4,106 had complete 2015 biomarker data, and 3,665 had complete 2015 cognitive function data. Among those who had missing values for these measures, 1,688 missed some of the biomarkers
or cognitive test results; 263 died before the follow-up wave; and 493 were not interviewed. So
the sample size of the Chinese longitudinal sample is 3,665. Both the cross-sectional and
longitudinal samples have significantly younger respondents, male respondents, and those with
higher education levels compared to the initial samples. Specifically, the proportions of those
who did not complete primary school education were significantly smaller, while the proportions

45
of those who completed primary school but not high school was higher in both analytical
samples relative to the initial sample (eTable 3.1). Measures
Cognitive Function
Cognitive function is measured using a 25-point summary score based on the overlapping
cognitive tests used in both HRS and CHARLS and is collected at both the baseline and follow- up. The score includes (i) immediate and delayed 10-word recall tests to measure episodic
memory (0 to 20 points), and (ii) a serial sevens subtraction test to measure working memory (0- 5 points). In the word recall tests, each correct response is worth 1 point. During both HRS and
CHARLS interviews, one of the four word recall lists was randomly assigned to each respondent. All words in the lists are common objects/concepts that are used universally across each country, and the numbers of syllables in the Chinese and English words are similar (eTables 3.2 & 3.3). In
the serial sevens test, respondents were asked to subtract by 7 starting from 100 a total of 5 times. Each correct subtraction was scored as 1 point. A higher summary score indicates better
cognitive function. Individual Cardiometabolic Biomarkers and the CMR Summary Measure
CMR is measured with a 9-point summary score (Wu et al., 2021) based on biomarkers
measuring cardiovascular and metabolic functioning available in both HRS and CHARLS. Those
include systolic blood pressure (SBP), diastolic blood pressure (DBP), resting heart rate (RHR), waist circumference (WC), glycosylated hemoglobin (HbA1c), high-density lipoprotein
cholesterol (HDL-C), total cholesterol (TC), and c-reactive protein (CRP). In both studies, SBP,

46
DBP, and RHR were measured 3 times at each wave; we use the average of the non-missing
measures for this analysis. To account for potential assay differences over time, the 2015 CHARLS blood-based
biomarker values (values for CRP, TC, HDL-C, and HbA1c) were adjusted to match the
distribution of the 2011 baseline values. This makes the distributions of the blood-based
biomarkers of the CHARLS respondents the same in 2011 and 2015, with no change over time at
the national level but there is change at the individual level. The distributions of blood-based
biomarkers among all blood sample respondents at two waves are shown in eFigure 3.1. A
detailed description of the adjustment is in the supplementary materials. The HRS biomarker collection was based on DBS, and the biomarker values may vary
over time because of changes in assays and laboratories (Kim et al., 2023). To make DBS data
comparable over time and to other population studies based on venous blood assays, HRS
converted DBS biomarker values (values for CRP, TC, HDL-C, and HbA1c) into NHANES
(National Health and Nutrition Examination Survey) equivalent values (Crimmins et al., 2017). For the main analyses, we use the same approach as used for CHARLS blood-based biomarkers
and eliminate change over time at the population level (eFigure 3.2). However, importantly, even
though there are no over-time changes in HRS and CHARLS blood-based biomarker values at
the population level, individual-level changes can still occur in our analysis. The CMR summary score is a count of the number of 8 biomarkers that exceed clinical
high-risk thresholds, together with a pulse pressure (PP) adjustment (high-risk thresholds can be
found in eTable 3.4): If SBP is greater than 140 mmHg and PP is greater than 70 mmHg, one
additional point is added to the total CMR. The CMR summary score ranges from 0 to 9, with
higher values indicating higher risk. In addition, each of the 9 individual CMR components is

47
coded as a dichotomous variable indicating whether the value is at a risk increasing level. For
both countries, the CMR summary score and the 9 CMR components risk indicators are
calculated both at baseline and follow-up. CMR increase is calculated by subtracting the baseline
CMR score from the follow-up CMR score. A positive value indicates an increase and a negative
value indicates a decrease. A 3-category CMR change variable is generated indicating CMR
increase, no change, and decrease. Covariates
Old age is one of the primary risk factors for cognitive impairment (Levine et al., 2018). In general, women are more likely to appear to have dementia, but largely due to women’s
longer life expectancy and less access to education in earlier cohorts and in less developed
societies (Zhao & Crimmins, 2022). Specifically in the US, older women generally have slightly
higher episodic memory and overall cognitive function, especially at early older ages (Oksuzyan
et al., 2010). On the other hand, in China, women are consistently found to have worse cognitive
outcomes (Lei et al., 2014; Zhang et al., 2019; Zhao et al., 2014). As previously mentioned, education is considered one of the most important predictors of cognitive health. Hence, our
covariates include age, gender, and education level. Age is categorized into 4 groups: 50-59, 60-69, 70-79, and 80 and older. In eTable 3.5, for comparison purposes, education level is initially classified as (i) less than primary school
level (less than 6 years), (ii) above primary school level but less than high school, (iii) high
school, (iv) some college, and (v) college degree or higher for both countries. Because older
Americans and older Chinese have very different overall education levels, in the regression
models, education is coded differently for the two countries to ensure sufficient respondents in

48
each category. For the US sample, the education groups include (i) less than high school, (ii)
completed high school but without a college degree, and (iii) college and above; for the Chinese
sample, the education groups include (i) less than primary school, (ii) above primary school but
less than high school, and (iii) high school and above. Statistical Analysis
We first compare the sample characteristics between the two countries. To examine the
association between summary CMR and cognitive function, two sets of ordinary least squares
(OLS) regression models controlling for covariates are used: (i) The baseline models assess the
associations between baseline CMR and cognitive function based on the cross-sectional samples, and (ii) the change models use baseline CMR and CMR increase to predict follow-up cognitive
function and cognitive change based on the longitudinal samples. When cognitive change is the
outcome of interest, follow-up cognitive function is used as the dependent variable and baseline
cognitive function is added as an independent variable. The models are fitted for the two
countries separately. Then, to further explain the differential results across countries, baseline cognitive
function scores are compared across baseline CMR groups, adjusted for age and gender
separately by country. Similarly, to understand whether and how the observed difference is
influenced by individual biomarkers, baseline cognitive function scores are compared between
the baseline low-risk and high-risk groups for each biomarker. Since SES might play a role in the
differential associations, baseline cognitive function and baseline estimated probabilities of
having a high-risk level for each biomarker are compared across education groups, adjusted for
age and gender, using the cross-sectional samples of the two countries separately. Finally, for

49
both the US and Chinese samples, interaction terms between baseline CMR and education are
added to the cross-sectional models; Interaction terms between CMR increase and education and
between baseline CMR and education are added to the longitudinal models. Results
Selected sample characteristics are compared in eTable 3.5. Overall, both the cross- sectional and the longitudinal samples in the US are older, have more women, and are better
educated. Specifically, for the cross-sectional samples, the older Americans have a mean age of
69.7 (SD=9.1) while the older Chinese have a mean age of 62.2 (SD=8.1). Only 12% of the US
sample is 50-59, while half of the Chinese are this age. The proportion of Americans above age
80 is almost 6 times more than Chinese (3% vs 17%). The proportions of female respondents are
56% and 48% for the US and Chinese sample respectively. Most (88%) of the Chinese sample
have an education level lower than high school, whereas 84% of the US sample have at least
completed high school education. CMR is higher among the US sample compared to the Chinese
sample (2.02 vs 1.54). The difference comes from the markedly higher risk levels of HbA1c, TC, WC, and CRP in the US sample. Despite being older with higher CMR, the US sample has better
cognitive function at both waves (eFigure 3.3). The average baseline cognitive function is 13.6
(SD=4.1) in the US and 10.8 (SD=4.1) in China. In the longitudinal samples, the differences mentioned above are also observed. In
addition, in both countries, respondents are almost evenly distributed across the three CMR
change categories, although the Chinese group whose CMR did not change from baseline to
follow-up was slightly larger.

50
Results from the baseline models examining the association between baseline CMR and
cognitive function based on cross-sectional samples are in Table 3.1. In the US, Model U1 shows
that a higher baseline CMR is associated with a significantly lower cognitive function (b=-0.28, p<0.001), adjusted for age and gender. After controlling for education in Model U2, the negative
association is still significant but with a reduced magnitude (b=-0.08, p<0.016). In addition, older persons have worse cognitive function compared to those aged 50-59 (aged 60-69: b=-0.52, p=0.003; aged 70-79: b=-1.65, p<0.001; aged 80+: b=-3.91, p<0.001). Women have better
cognitive function (b=0.76, p<0.001); and having higher education levels is related to better
cognitive function (Reference: Less than high school; Less than college: b=3.05, p<0.001;
College and above: b=4.97, p<0.001). In China, a significant relationship between CMR and cognition is not found. Nevertheless, older ages still link to worse cognitive function (Aged 60-69: b=-0.32, p=0.037;
Aged 70-79: b=-1.86, p<0.001; Aged 80+: b=-3.55, p<0.001), and higher education levels still
link to better cognition (Reference: Less than primary school; Less than high school: b=2.36, p<0.001; high school and above: b=4.52, p<0.001) in a similar way as in the US models. Unlike
the US models, in Model C1, female gender is significantly associated with worse cognitive
function (b=-0.88, p<0.001), but the association becomes insignificant after controlling for
education in Model C2. Change models use both baseline CMR and CMR change to predict cognitive outcomes
based on the longitudinal samples, and the results are in Table 3.2. In both countries, Models 1 &
2 use follow-up cognitive function as the outcome, and Models 3 & 4 further add baseline
cognitive function as an independent variable, so the outcome becomes cognitive change. For the
US sample, both CMR increase (b=-0.18, p=0.001) and a higher baseline CMR (b=-0.45,

51
p<0.001) are negatively associated with follow-up cognitive function. Model U2 includes
education and the negative coefficient of baseline CMR remains significant with a reduced
magnitude (b=-0.16, p=0.003), but the negative coefficient of CMR increase is only marginally
significant (b=-0.10, p=0.055), indicating that education explains part of the variance that was
previously explained by CMR level and change. The coeffcients of age, gender, and education
are similar to the baseline models. Similarly, Model U3 shows that CMR increase (b=-0.09, p=0.033), worse baseline CMR (b=-0.15, p<0.001), and older ages (Aged 70-79: b=-1.09, p<0.001; aged 80+: b=-2.26, p<0.001) are linked to cognitive decline. After controlling for
education in Model U4, the coefficient of baseline CMR becomes marginally significant with
reduced size (b=-0.08, p=0.064). The coefficient of CMR increase is no longer significant (b=- 0.06, p=0.108). Higher education levels are still protective factors (Reference: Less than high
school; Less than college: b=1.01, p<0.001; College and above: b=1.85, p<0.001). In the Chinese longitudinal sample, neither CMR change nor baseline CMR is
significantly associated with follow-up cognitive function (Table 3.3, Models C1 & C2) or
cognitive decline (Models C3 & C4). But still, higher education levels (e.g., Model C4:
Reference: Less than primary school; Less than high school: b=1.75, p<0.001; high school and
above: b=3.23, p<0.001) and female gender (e.g., Model C4: b=0.40, p=0.021) are protective for
cognitive function while older age (e.g., Model C4: reference: aged 50-59; Aged 60-69: b=-0.80, p<0.001; Aged 70-79: b=-2.35, p<0.001; Aged 80+: b=-3.08, p<0.001) are linked to higher risk. Compared to the US sample, the negative association between older ages and cognitive function
is stronger (e.g., Model C4 vs U4: Aged 60-69: b=-0.80 vs -0.15; Aged 70-79: b=-2.35 vs -1.04;
Aged 80+: b=-3.08 vs -2.20), indicating that the onset of decline in cognitive health may be
earlier in China.

52
Since the link between CMR and cognitive function is only found in the US but not
China, especially in the baseline models, additional analyses based on the cross-sectional
samples further investigate the differential association observed in the two countries. Figure 3.1
shows the estimated cognitive function score by CMR levels, adjusted for age and gender. Consistent with the previous models, in the US, people with higher CMR levels generally have
worse cognitive function. While in China, the difference is not significant, and people with a
CMR level higher than 5 may even have better cognitive function. To understand whether the
difference is driven by individual biomarkers, Figure 3.2 shows the estimated cognitive function
score for low-risk versus high-risk biomarker values, adjusted for age and gender. In the US, people with high-risk markers always have lower cognitive function scores, and the difference
between high- and low-risk groups for SBP, pulse pressure, HbA1c, and CRP is significant. However, in China, for several markers, those with high-risk values seem to have better
cognitive function, and those with a risky HDL level even have significantly better cognitive
function. So the null relationship between CMR and cognitive function is consistent based on
both the CMR summary measure and individual biomarkers. To see whether education can help explain why the association between CMR and
cognitive function is not found in China, the relationship between education level and CMR as
well as the relationship between education level and cognitive function are both visualized in
Figure 3.3. In both countries, higher education levels are associated with significantly better
cognitive function. But unlike the US, where higher education groups are less likely to have
high-risk CMR biomarker values, in China, groups with higher education are not significantly
better off; and for some markers, they seem to have even riskier CMR biomarker values.

53
The role that education potentially plays in the relationship between CMR and cognitive
function is further investigated in the cross-sectional and longitudinal models with interaction
terms. In eTable 3.6, neither the interaction terms in the US or Chinese samples are statistically
significant. However, in the Chinese sample, based on an overall positive estimated coefficient
between baseline CMR and cognitive function (b=0.009, p=0.883), the interaction term between
the highest education level and baseline CMR has an additionally positive association (b=0.610, p=0.113), indicating that a higher CMR among the highest education group could be potentially
good for cognitive health cross-sectionally. On the contrary, in the US sample, although not
statistically significant as well, the estimated associations between CMR and cognitive function
for all education groups are negative. In eTable 3.7, for each country, similar to previous
longitudinal models, the outcome variable for Model 1 is follow-up cognitive function and for
Model 2 is cognitive change. In the US, all the significant coefficients of the interaction terms
are negative, indicating additional negative effects on the associations between CMR and
cognition. In China, the coefficients of the interaction terms between baseline CMR and the
highest education level are positive for both Model1 (b=0.626, p=0.013) and Model 2 (b=0.350, p=0.062), suggesting that for those with the highest education level, compared to those with the
lowest education level, having a higher baseline CMR is associated with additionally better
follow-up cognitive function and slower cognitive decline. However, the coefficients of the
interaction terms between CMR increase and the highest education level are negative for Models
1 (b=-0.536, p=0.050) and 2 (b=-0.350, p=0.079), indicating that for those with the highest
education level, relative to their counterpart with the lowest education level, having a rapid CMR
increase within 4 years of aging is associated with additionally worse follow-up cognitive
function and faster cognitive decline.

54
Discussion
In this study, we examined the association between CMR and cognitive function both
cross-sectionally and longitudinally among older Americans and Chinese using data from
comparable nationally representative surveys. We found distinct patterns of association between
CMR and cognitive function: In the US, a higher CMR is associated with worse cognitive
function cross-sectionally, and increasing CMR over time is associated with worse cognitive
function as well. Although we also found that CMR increase is associated with greater cognitive
decline, after accounting for education this association was no longer statistically significant. However, in our Chinese sample, neither CMR level nor CMR change is significantly associated
with cognitive function or cognitive decline. Further examination of individual biomarkers in an
attempt to understand these complex relationships showed that in the US, people with risk levels
at baseline of SBP, PP, HbA1c, and CRP values have significantly worse cognitive function
(adjusted for age and gender). However, differences between high- and low-risk groups cannot
be found in China. In both countries, a higher education level links to better cognitive health. While higher education is linked to lower CMR in the US, for many biomarkers, the most
educated older Chinese seem to have even higher risk. Since a high education can be linked to
both better cognition and worse CMR for some older Chinese, it could potentially affect the
relative importance of CMR in predicting cognitive outcomes. The models with interaction terms
show that for Chinese older adults with the highest education level, having a higher CMR is not
as harmful as it is for their lower-education counterparts in terms of cognitive health. Nevertheless, a rapid rise in CMR is additionally harmful, possibly because they already have a
higher baseline CMR to start with.

55
The significant associations between CMR and cognitive function in the US are
consistent with much existing literature (Chanti-Ketterl et al., 2022; D’Amico et al., 2020; Hu et
al., 2020). The current study contributes to the literature by showing a significant negative
association between CMR increase over 4 years and cognitive function, suggesting the
importance of monitoring these indicators and controlling them by changing health behaviors as
well as taking medication. The insignificant results in China should not be interpreted as indicating that
cardiometabolic health will not matter in terms of preserving and improving late-life cognitive
health. Instead, it reveals the complicated role that SES plays in the relationship between CMR
and cognition in the Chinese context. Because of the unprecedented rapid economic development
and urbanization that occurred throughout their lifetime, the current older population in China, mostly born before the 1960s, may have a unique social patterning of CMR, very different from
that in high-income societies like the US: On one hand, the fast economic development has led to
rapid increases in levels of education, income, and access to resources like healthcare and social
services, especially among urban dwellers (Gong et al., 2012). On the other hand, rapid
urbanization has led to urban residents with higher socioeconomic status (SES) being more likely
than rural residents to be overweight/obese, hypertensive, diabetic, and to have higher biological
risks indicating cardiovascular, metabolic, and inflammatory health due to the earlier adoption of
an unhealthy diet, reduced level of physical activities, and exposure to certain pollutants (Zhang
& Crimmins, 2019; Zhao et al., 2016). In addition, the healthcare system in China is not focused
on chronic diseases, and their management is far from complete (Li & Lumey, 2019; Zhao et al., 2016), since the epidemiological transition started more recently in China than in the U.S. and is
happening in a shorter period (Zou et al., 2022). As a result, while higher SES is usually

56
associated with lower level and better management of CMR in high-income countries like the US
(Toms et al., 2019; Johnson et al., 2017; Wu et al., 2021), higher SES in China may mean even
higher risk without sufficient management of the risk. The current analyses based on the Chinese
sample do not show an overall association between CMR and cognitive health but still indicate
that having a rapid CMR increase can be harmful even among high SES individuals, and
therefore highlight the importance of improving cardiometabolic health through education, especially in terms of promoting healthy lifestyles. In addition, the insignificant associations may
also be influenced by the stage of life when people develop high CMR: It is possible that older
Chinese experienced a later rise of CMR and hence have not lived with high risks as long as
older Americans. In addition to our main findings, our data indicate that the cognitive function score is
lower among older Chinese compared to older Americans, which is consistent with other recent
findings (Gross et al. 2023). Our sample of older Chinese was born approximately between the
1930s to 1960s. They likely had very few educational, occupational, and healthcare resources
throughout most of their lifespan and many of them experienced adverse and traumatic events
such as wars. Most notably, they have a rudimentary level of education (85% of them did not
complete secondary education). All these adversities could potentially reduce one’s cognitive
reserve and harm one’s cognitive function at older ages. In addition, our baseline models suggest
that the female gender is associated with worse cognitive function in China, which is consistent
with what has been found in previously (Lei et al., 2014; Zhang et al., 2019; Zhao et al., 2014). In addition, after controlling for education, this negative association is eliminated, which is also
consistent with previous studies: Both Zhao et al. (2014) and Lei et al. (2014) suggested that

57
disparity in education is the main cause of women’s worse cognitive health. Lei et al. also
showed that the gender difference was largely mitigated after controlling for education. To make biomarker measures comparable across countries, we applied an adjustment to
both HRS and CHARLS blood-based biomarkers so that assay changes would not have an effect. For HRS, we did a sensitivity analysis without the baseline adjustment. eFigure 3.2 shows the
minor differences in biomarker distributions in the HRS blood-based biomarkers at baseline and
follow-up waves before and after the adjustment. eTable 3.8 shows results from the same
regression model from Table 3.2 but compares the coefficients between the models using CMR
with and without baseline adjustment. Since the results are almost identical, this adjustment had
little effect on the results. The current study has some important strengths. The data come from two nationally
representative surveys designed to facilitate international comparison of high-quality and
comparable data on cognitive function and biomarkers, as well as longitudinal data to examine
temporal change in both CMR and cognitive function. In addition, we adjusted the blood-based
biomarker values to make them comparable over time in both countries using a closely aligned
approach. Our study has limitations. By limiting our analyses to those who have complete data in
both waves, the sample did not include those with missing data or those who died before the
follow-up interview. Our final samples are statistically different from the initial sample as shown
in eTable 3.1. So, our samples are representative of slightly younger and healthier (in terms of
both CMR and cognitive function) sub-populations who survived 4 years in both countries. Our
samples also have more highly educated individuals, especially in China, and women in China
are less represented, presumably for the same reason. Hence, the tradeoff for utilizing the

58
longitudinal nature of the data is that conclusions may be less generalizable to the most
socioeconomically disadvantaged population in China. Even though the two studies were
designed in similar ways to facilitate international comparison, there are inevitable differences in
our samples. Our Chinese sample has more respondents in the youngest age group compared to
the US sample, and thus the estimates in our Chinese models could be more weighted by these
younger respondents. In terms of biomarkers, discrepancies in collection and assays may still
exist in spite of efforts to make the biomarkers comparable over time and across studies. The
cognitive tests were also designed and administered for comparability but still differences may
exist due to language differences. In addition, since the rate of CMR and cognitive change may
differ across individuals, and could be slower among healthier persons, the analyses focusing on
change might benefit from a longer follow-up period, which will be possible in the future. Based on comparable nationally representative longitudinal data in the US and China, the
results of the current paper confirm that high CMR is a risk factor for worse cognitive health in
the U.S., but the cross-sectional and longitudinal relationship between CMR and cognitive
function might differ across societies, depending on how SES factors like education interact with
cardiometabolic health. Our findings are important for identifying high-risk population
subgroups in terms of both cardiometabolic and cognitive problems in these two countries. Tables/Figures
Table 3.1. Results from the Baseline Models Predicting Cognitive Function in the Two Countries
US-HRS (N=7,413) China-CHARLS (N=6,108)
Model U1 Model U2 Model C1 Model C2
Baseline CMR -0.276*** -0.084* 0.152 0.096
Age - Ref: Aged 50-59
Aged 60-69 -0.752*** -0.520** -0.608** -0.323* Aged 70-79 -2.183*** -1.650*** -2.503*** -1.862*** Aged 80+ -4.622*** -3.911*** -4.838*** -3.554*** Female 0.539*** 0.762*** -0.878*** -0.196
Education - Ref: Less than High School (US) / Less than

59
Primary School (CN)
Less than College (US) / Less than High School (CN) 3.045*** 2.364*** College and above (US) / High School and above
(CN) 4.966*** 4.515*** Constant 15.521*** 11.525*** 11.683*** 9.612*** R2 0.144 0.282 0.079 0.21
Note. † p<0.1 * p<0.05, ** p<0.01, *** p<0.001
Table 3.2. Results from the Longitudinal Models Predicting Cognitive Function 4 Years after
Initial CMR in Two Countries with Both Initial Levels of CMR and CMR Increase
US-HRS (N=4,474) China-CHARLS (N=3,655)
Model U1 Model U2 Model U3 Model U4 Model C1 Model C2 Model C3 Model C4
CMR Increase -0.178** -0.098† -0.086* -0.064 -0.23 -0.137 -0.116 -0.076
Baseline CMR -0.404*** -0.160** -0.151*** -0.081† 0.009 0.005 -0.044 -0.036
Baseline Cognitive Function Score 0.708*** 0.648*** 0.518*** 0.422*** Age - Ref: Aged 50-59
Aged 60-69 -0.732*** -0.492* -0.199 -0.154 -1.243*** -0.969*** -0.914*** -0.797*** Aged 70-79 -2.549*** -2.086*** -1.091*** -1.043*** -3.641*** -3.086*** -2.509*** -2.354*** Aged 80+ -5.020*** -4.227*** -2.261*** -2.202*** -5.112*** -4.095*** -3.435*** -3.076*** Female 0.446*** 0.671*** 0.004 0.131 -0.353† 0.388† -0.03 0.401* Education - Ref: Less than High School (US) / Less
than Primary School (CN)
Less than College (US) /
Less than High School (CN) 2.853*** 1.005*** 2.634*** 1.752*** College and above (US) /
High School and above
(CN)
4.971*** 1.848*** 4.923*** 3.228*** Constant 15.774*** 11.703*** 4.472*** 3.951*** 11.654*** 9.260*** 5.454*** 5.022*** R2 0.152 0.281 0.537 0.553 0.094 0.245 0.319 0.377
Note. † p<0.1 * p<0.05, ** p<0.01, *** p<0.001
Figure 3.1. Baseline Cognitive Function Score by CMR Level in the US (HRS, N=7,413) and
China (CHARLS, N=6,108)

60
Figure 3.2. Cognitive Function Score for People with Low-Risk versus High-Risk Biomarker
Values of Individual Biomarkers in the US (HRS, N=7,413) and China (CHARLS, N=6,108)

61
Figure 3.3. Estimated Probability of Having High-Risky Biomarker Level by Education Groups
in the US (HRS, N=7,413) and China (CHARLS, N=6,108)

62 Supplemental Information eTable 3.1. Comparison between Initial Samples and Analytic Samples

63
eTable 3.2. Word Lists and Total Syllable Counts for the CHARLS Word Recall Tests
List A List B List C List D
Words Pinyin Words Pinyin Words Pinyin Words Pinyin Rice 米饭 Mi Fan Stool 凳子 Deng Zi Mountain 山 Shan Water 水 Shui
River 河流 He Liu Foot 脚 Jiao Stone 石头 Shi Tou Hospital 医院 Yi Yuan Doctor 医生 Yi Sheng Sky 天空 Tian Kong Blood 血液 Xue Ye Tree 树木 Shu Mu Clothes 衣服 Yi Fu Money 金钱 Jin Qian Mother 妈妈 Ma Ma Father 爸爸 Ba Ba Egg 鸡蛋 Ji Dan Pillow 枕头 Zhen Tou Shoes 鞋子 Xie Zi Fire 火 Huo Cat 小猫 Xiao Mao Dog 小狗 Xiao Gou Eye 眼睛 Yan Jing Tooth 牙齿 Ya Chi
Bowl 饭碗 Fan Wan House 房子 Fang zi Girl 女孩 Nv Hai Moon 月亮 Yue Liang
Child 小孩 Xiao Hai Wood 木头 Mu Tou House 房子 Fang Zi Village 村子 Cun Zi
Hand 手 Shou School 小学 Xiao Xue Road 马路 Ma Lu Boy 男孩 Nan Hai
Book 书 Shu Tea 茶 Cha Sun 太阳 Tai Yang Table 桌子 Zhuo Zi
Total Syl 18 Total Syl 18 Total Syl 19 Total Syl 18
Note. The English translation of the Chinese words is provided by the CHARLS team in the survey questionnaire. Pinyin is the spellings of Chinese characters in the Latin alphabet based on their pronunciation. The total count of syllables is shown in the last row of the table for each word list. In CHARLS interview, one of the four lists was randomly assigned to each respondent for the word recall tests. All words in the
lists are very common objects/concepts which are used universally across the country. These word lists were repeatedly used in both waves. eTable 3.3. Word Lists and Total Syllable Counts for the HRS Word Recall Tests
List A List B List C List D
Hotel Sky Woman Water
River Ocean Rock Church
Tree Flag Blood Doctor
Skin Dollar Corner Palace
Gold Wife Shoes Fire
Market Machine Letter Garden
Paper Home Girl Sea
Child Earth House Village
King College Valley Baby
Book Butter Engine Table
Total Syllables: 14 Total Syllables: 15 Total Syllables: 15 Total Syllables: 17
Note. The total count of syllables is shown in the last row of the table for each word list. In HRS interview, one of the four lists was randomly assigned to each respondent for the word recall tests. These word lists were repeatedly used in both waves. eTable 3.4. High-Risk Thresholds for CMR Indicators
CMR Variables High Risk Group Reference
Cardiovascular Markers
Systolic Blood Pressure (mmHg) ≥ 140
Lakka et al. 2002; Mancia et al. 2007
Diastolic Blood Pressure (mmHg) ≥ 90
Pulse Pressure Adjustment If SBP ≥ 140 & PP ≥ 70, CMR+1
Mancia et al. 2007; Franklin and
Wong 2016; Ji et al. 2020

64
Resting Heart Rate (bpm) ≥ 90
Seeman et al. 2008; Mitchell, Ailshire, and Crimmins 2019;
Seccareccia et al. 2001; Hartaigh et
al. 2015; Bird et al. 2010
Metabolic Markers
Glycated Hemoglobin (%) ≥ 6.5 Mitchell, Ailshire, and Crimmins
2019
High-Density Lipoprotein (mg/dL) ≤ 40
Crimmins et al. 2003; Mitchell, Ailshire, and Crimmins 2019; Bird et
al. 2010
Total Cholesterol (mg/dL) ≥ 240 Mitchell, Ailshire, and Crimmins
2019; Bird et al. 2010
Waist Circumference (inch) ≥ 35 (women); ≥ 40 (men) Mitchell, Ailshire, and Crimmins
2019; Mancia et al. 2007
Inflammatory Marker
C-Reactive Protein (mg/L) ≥ 3 Mitchell, Ailshire, and Crimmins
2019; Bird et al. 2010
Note. CMR: Cardiometabolic Risk; SBP: Systolic blood pressure; PP: Pulse pressure. eTable 3.5. Sample Characteristics for the US (HRS) and the Chinese (CHARLS) Samples
HRS CHARLS
Cross-Sectional, N=7,413
Longitudinal, N=4,474
Cross-Sectional, N=6,108
Longitudinal, N=3,655
Mean (SD) /
Proportion Mean (SD) /
Proportion Mean (SD) /
Proportion Mean (SD) /
Proportion Age 69.67 (9.06) 68.33 (8.25) 62.17 (8.07) 61.39 (7.13)
Age Group
Aged 50-59 0.12 0.13 0.48 0.48
Aged 60-69 0.44 0.47 0.34 0.38
Aged 70-79 0.28 0.28 0.16 0.13
Aged 80+ 0.17 0.12 0.03 0.01
Female 0.56 0.56 0.48 0.46
Education Level
Less than Primary School (Less than 6
Years) 0.03 0.02 0.43 0.39
Less than High School 0.13 0.11 0.45 0.49
High School 0.33 0.33 0.10 0.10
Some College 0.23 0.23 0.02 0.02
College 0.28 0.31 0.01 0.00
CMR Baseline 2.02 (1.41) 1.91 (1.37) 1.54 (1.42) 1.49 (1.42)
CMR Follow-Up - 2.00 (1.37) - 1.50 (1.34)
CMR Change - 0.09 (1.41) - 0.01 (1.38)
CMR Increase - 0.37 - 0.32
CMR Constant - 0.32 - 0.38
CMR Decrease - 0.31 - 0.30
Biomarker at High Risk Level
SBP Baseline 0.29 0.27 0.32 0.30
SBP Follow-Up - 0.27 - 0.30
DBP Baseline 0.15 0.15 0.14 0.12
DBP Follow-Up - 0.11 - 0.10
PP Additional Risk Baseline 0.10 0.08 0.17 0.15
PP Additional Risk Follow-Up - 0.10 - 0.14
RHR Baseline 0.05 0.04 0.05 0.05
RHR Follow-Up - 0.04 - 0.06
HbA1c Baseline 0.15 0.13 0.05 0.05
HbA1c Follow-Up - 0.16 - 0.06

65
HDL-C Baseline 0.18 0.17 0.28 0.30
HDL-C Follow-Up - 0.18 - 0.30
TC Baseline 0.14 0.14 0.11 0.10
TC Follow-Up - 0.13 - 0.09
Waist Baseline 0.66 0.65 0.22 0.22
Waist Follow-Up - 0.66 - 0.27
CRP Baseline 0.31 0.28 0.20 0.19
CRP Follow-Up - 0.33 - 0.17
Cognitive Function Score Baseline 13.56 (4.13) 14.23 (3.83) 10.77 (4.10) 11.28 (3.88)
Episodic Memory Score Baseline 9.90 (3.29) 10.41 (3.10) 7.20 (3.34) 7.45 (3.25)
Executive Function Score Baseline 3.66 (1.58) 3.82 (1.49) 3.58 (1.59) 3.81 (1.46)
Cognitive Function Score Follow-Up - 13.60 (4.13) - 10.53 (4.13)
Episodic Memory Score Follow-Up - 9.84 (3.32) - 6.82 (3.46)
Executive Function Score Follow-Up - 3.76 (1.54) - 3.72 (1.46)
Note. Abbreviations: SBP: Systolic blood pressure; DBP: Diastolic blood pressure; RHR: Resting heart rate; HbA1c: Glycosylated
hemoglobin; HDL-C: High-density lipoprotein cholesterol; TC: Total cholesterol; WC: Waist circumference; CRP: C-reactive protein. eTable 3.6. Results from the Cross-Sectional Models with Interaction Terms
US - HRS China - CHARLS
N=7,413 N=6,108
Baseline CMR -0.114 0.009
Age - Ref: Aged 50-59
Aged 60-69 -0.527** -0.299* Aged 70-79 -1.659*** -1.838*** Aged 80+ -3.923*** -3.477*** Female 0.762*** -0.210
Education - Ref: Education Level 1
Education Level 2 3.017*** 2.372*** Education Level 3 4.770*** 3.541*** Education * Baseline CMR - Ref: Education Level 1 * Baseline CMR
Education Level 2 * Baseline CMR 0.008 -0.010
Education Level 3 * Baseline CMR 0.103 0.610
Constant 11.608*** 9.741*** R2 0.282 0.215
Note. † p<0.1 * p<0.05, ** p<0.01, *** p<0.001. US Education Levels: Lv1 – Less than High School; Lv2 – Less than College; Lv3 – College and above China Education Levels: Lv1 – Less than Primary School; Lv2 – Less than High School; Lv3 – High School and Above eTable 3.7. Results from the Longitudinal Models with Interaction Terms
US - HRS China - CHARLS
N=4,474 N=3,655
Model 1 Model 2 Model 1 Model 2
CMR Increase 0.235* 0.121 0.008 -0.025
Baseline CMR 0.072 0.076 -0.071 -0.098
Baseline Cognitive Function Score 0.646*** 0.408*** Age - Ref: Aged 50-59
Aged 60-69 -0.515* -0.170 -0.941*** -0.784*** Aged 70-79 -2.106*** -1.059*** -3.005*** -2.322*** Aged 80+ -4.261*** -2.229*** -3.999*** -3.050*** Female 0.654*** 0.122 0.312† 0.354* Education - Ref: Education Lv1
Education Lv2 3.647*** 1.548*** 2.678*** 1.748*** Education Lv3 5.293*** 2.069*** 3.834*** 2.654*** Education*CMR Increase - Ref: Education Lv1*CMR Increase

66
Education Lv2*CMR Increase -0.418** -0.244* -0.027 0.062
Education Lv3*CMR Increase -0.341* -0.161 -0.536† -0.350†
Education*Baseline CMR - Ref: Education Lv1*Baseline CMR
Education Lv2*Baseline CMR -0.354* -0.240* -0.041 0.014
Education Lv3*Baseline CMR -0.092 -0.062 0.626* 0.350†
Constant 3.951*** 3.618*** 9.388*** 5.269*** R2 0.553 0.554 0.264 0.384
Note. † p<0.1 * p<0.05, ** p<0.01, *** p<0.001. US Education Levels: Lv1 – Less than High School; Lv2 – Less than College; Lv3 – College and above China Education Levels: Lv1 – Less than Primary School; Lv2 – Less than High School; Lv3 – High School and Above eTable 3.8. Change Models in the US: CMR Biomarkers with Baseline Adjustment VS. CMR
Biomarkers without Baseline Adjustment
With Baseline Adjustment (Table 3) Without Baseline Adjustment
Model 1 Model 2 Model 3 Model 4 Model 1 Model 2 Model 3 Model 4
CMR Increase -0.178** -0.098† -0.086* -0.064 -0.180** -0.092† -0.086* -0.061
Baseline CMR -0.404*** -0.160** -0.151*** -0.081† -0.406*** -0.157** -0.152*** -0.079†
Baseline Cognitive Function Score 0.708*** 0.648*** 0.708*** 0.648*** Age - Ref: Aged 50-59
Aged 60-69 -0.732*** -0.492* -0.199 -0.154 -0.734*** -0.494* -0.2 -0.155
Aged 70-79 -2.549*** -2.086*** -1.091*** -1.043*** -2.550*** -2.087*** -1.091*** -1.043*** Aged 80+ -5.020*** -4.227*** -2.261*** -2.202*** -5.018*** -4.226*** -2.260*** -2.201*** Female 0.446*** 0.671*** 0.004 0.131 0.446*** 0.671*** 0.004 0.131
Education - Ref: Less
than High School (US)
/ Less than Primary
School (CN)
Less than College
(US) / Less than High
School (CN)
2.853*** 1.005*** 2.854*** 1.005*** College and above
(US) / High School and above (CN)
4.971*** 1.848*** 4.970*** 1.847*** Constant 15.774*** 11.703*** 4.472*** 3.951*** 15.771*** 11.693*** 4.469*** 3.945*** R2 0.152 0.281 0.537 0.553 0.152 0.281 0.537 0.553
Note. † p<0.1 * p<0.05, ** p<0.01, *** p<0.001
For the models without baseline adjustment, the NHANES (National Health and Nutrition Examination Survey) equivalent version of HRS dry blood spot markers are used: the HRS 2010 and 2012 values were adjusted based on a pooled sample of
NHANES 2009-2010 and 2011-2012 values; HRS 2014 values were adjusted based on pooled NHANES 2011-2012 and 2013- 2014; and HRS 2016 values were adjusted based on pooled NHANES 2013-2014 and 2015-2016. So the difference in HRS
adjusted biomarker values between waves reflects change over time. The only exception is CRP. Since the change in NHANES
CRP seemed to reflect assay changes rather than real changes in values, we have now readjusted CRP, so NHANES equivalents are based on NHANES 2005-2008 for all HRS waves. eFigure 3.1. Distributions of CHARLS Blood-Based Biomarkers at Baseline and Follow-up
Waves before and after Adjustments (left panels: unadjusted; right panels: adjusted)
a. Glycosylated Hemoglobin (HbA1c)

67
b. High-Density Lipoprotein Cholesterol (HDL-C)
c. Total Cholesterol (TC)

68
d. C-Reactive Protein (CRP)
eFigure 3.2. Distributions of HRS Dry Blood Spot Biomarkers at Baseline and Follow-up Waves
before and after Baseline Adjustments (left panels: unadjusted; right panels: adjusted)
a. Glycosylated Hemoglobin (HbA1c)

69
b. High-Density Lipoprotein Cholesterol (HDL-C)
c. Total Cholesterol (TC)

70
d. C-Reactive Protein (CRP)
eFigure 3.3. Distributions of Cognitive Function for both Countries at the Baseline and the
Follow-Up Waves
The Cognitive Function Score at the Baseline Wave (%)

71
The Cognitive Function Score at Follow-Up Wave (%)
Adjusting CHARLS 2015 Blood-Based Biomarker Values Based on 2011 Values
The distributions are based on the whole CHARLS sample. Since biomarkers have
different missing patterns, the sample sizes also differ by biomarker by year. The adjusted 2015 values use 2011 values as a reference. The blood-based biomarker
values are first weighted using blood survey weights and transformed into 100 percentiles, separately for the 2011 sample and 2015 sample. 2011 blood weights are used for 2011
biomarkers; 2015 blood weights are used for 2015 biomarkers.

72
To facilitate construction of percentiles when values are discrete and have many
individuals scored at the same value, a very small random number is added to each observed
value. The weighted percentiles are created based on the altered values, and at each percentile, the mean of the actual biomarker values is taken. Due to the skewed distribution of CRP, the
percentile calculation is based on log values. Finally, the 2015 values are regressed on the 2011 values to create an equation that can
be used to convert 2015 values into 2011 values.

73
Chapter 4: RNA-based Indicators of Cellular Senescence Predict
Aging Health Outcomes in the Health and Retirement Study
Introduction
Aging is a gradual process, and at the population level, age-related health changes can be
categorized into several sequential dimensions, captured by the morbidity process model at the
population level, aging begins with molecular/cellular-level changes, followed by physiological
dysregulation indicated by clinical-level biomarkers, and leading to the subsequent diagnosis of
diseases and conditions (Crimmins, 2015; Zhao & Crimmins 2022). Cellular senescence is one of the molecular/cellular-level aging mechanisms that
accumulates with advancing age and plays an important pathogenic role in various adverse
health outcomes (Campisi, 2013; Campisi & Robert 2014; López-Otín et al. 2023, Serrano &
Munoz-Espin, 2021). Various cellular/molecular-level damage, including telomere attrition, DNA damage, oxidative stress, mitochondrial dysfunction, and oncogenic signaling, can
potentially induce cellular senescence, changing normal cells to senescent cells, and in response
to the damage, senescent cells stop proliferation – entering a generally irreversible state of
growth arrest (Gorgoulis et al., 2019; Muñoz-Espín & Serrano, 2014). In this way, cellular
senescence lowers the risk of malignant transformation by preventing the damage from spreading
to the next cell generation, and thus is considered a tumor-suppressive mechanism (Coppé et al., 2010). However, senescent cells are still metabolically active; they release a wide range of proinflammatory cytokines, chemokines, proteases, growth factors, and other bioactive molecules to
the local microenvironment. Despite that the secretion can help with the clearance of damaged

74
cells and the regeneration of broken tissues, it can have deleterious effects on neighboring cells, contributing to persistent chronic inflammation and progressive fibrosis (Coppé et al., 2010;
Franceschi and Campisi, 2014; López-Otín et al., 2023). Such proinflammatory secretion is
termed Senescence-Associated Secretory Phenotype (SASP), and has been suggested to mediate
downstream aging outcomes (Coppé et al., 2010; Franceschi & Campisi, 2014; López-Otín et al., 2023; Yue et al., 2022). Previous studies have linked cellular senescence to pulmonary fibrosis, diabetes, vascular
aging and atherosclerosis, kidney diseases, liver diseases, as well as Alzheimer’s and Parkinson’s
diseases, but most studies rely on animal models or human cells in vitro (Campisi et al., 2011;
Muñoz-Espín & Serrano, 2014; Serrano & Muñoz-Espín 2022). Since SASP can help explain the
association between cellular senescence and many age-related pathologies, it has been the focus
in human samples. Recent work finds a high concentration of SASP molecules to be associated
with lung disease (Aversa et al., 2023), a summary frailty measure combining comorbidity and
limitations in activities of daily living (Schafer et al., 2020), a mobility disability measured by a
short physical performance battery (Fielding et al., 2022), and a higher mortality risk (St. Sauver
et al., 2023). However, the samples in these studies are either of small size/trials or electronic
health records and thus have limited representativeness. In addition, recent works have suggested a more comprehensive approach to capturing
the entire effect of cellular senescence rather than solely relying on SASP – Alongside SASP, other key hallmarks of cellular senescence include cell cycle arrest (CCA) and macromolecular
damage (MD) (Gorgoulis et al., 2019). To profile the distinct aspects of cellular senescence, Dehkordi et al. (2021) developed, tested, and validated three RNA-based scores, corresponding
to CCA, MD, and SASP, in two independent single-nucleus RNA sequencing datasets. These

75
selected genes are reportedly associated with senescence in existing studies based on various cell
and tissue types in human and mouse brains (Bussian et al., 2018; Chinta et al., 2018; Chow et al. 2019; Musi et al., 2018; Ogrodnik et al., 2019, Riessland et al., 2019, Zhang et al., 2019). Nevertheless, how this more comprehensive measure of cellular senescence links to age-related
health outcomes at the population level needs further investigation. The various types of macromolecular damage that trigger cellular senescence could be
induced by social and behavioral factors. For example, sleep disorders can induce cell injury and
oxidative damage (Brown & Naidoo, 2010; Everson et al., 2014). Based on RNA data, Carroll et
al. (2016) find that partial sleep deprivation resulted in the activation of DNA damage repair
(DDR) and SASP, which correspond to two hallmarks of cellular senescence (Carroll et al., 2016). Other studies have found that obesity and a high-fat diet can increase oxidative stress and
entail the accumulation of senescent cells (Minamino et al., 2009; Serrano & Muñoz-Espín 2021;
Tchkonia et al., 2010). Though not directly focusing on cellular senescence, studies have
suggested that chronic cigarette smoking promotes oxidative stress, reduces antioxidative
response, and damages DNA (Calir et al., 2021; Nyunoya et al. 2006, Pole et al., 2016). Alcohol
abuse has also been linked to telomere erosion (Carvalho et al., 2019) and DNA damage
(Kruman et al., 2012; Sun et al., 2023). Hence, all these factors could be potentially associated
with cellular senescence. In addition, these behavioral factors may differ across socioeconomic
groups and racial/ethnic groups. Evaluating these associations at the population level will
contribute to our further understanding of the aging process and the disparity in the aging
experience. The current study creates senescence gene expression composite scores in a nationally
representative of older Americans based on Dehkordi et al.’s gene lists and relates them to social

76
and behavioral factors hypothesized to be associated with cellular senescence. To understand
how these composite scores measure aging, they are also related to multiple dimensions of aging
measures – from upstream biological changes to downstream health outcomes. In this study, we
hypothesize that risky health behaviors or conditions such as smoking, drinking, obesity, and
sleep disorders will all be associated with higher levels of cellular senescence. We also
hypothesize that the expression of senescence genes will be positively associated with measures
of a number of dimensions reflecting aging health change including mortality, multimorbidity, physiological dysregulation indicated by clinical-biomarker-based biological age acceleration, and epigenetic age acceleration measures based on DNA methylation. Since cellular senescence
is considered an underlying mechanism of aging related to but separate from epigenetic changes, another hypothesis of the current study is that the expression of senescence genes will add to our
ability to explain the downstream outcomes in addition to epigenetic age acceleration measures
based on DNA methylation. Methods
Data & Sample
Our study sample comes from the Health and Retirement Study (HRS), a longitudinal
study of US adults older than age 50. In 2016, a random subsample of HRS Venous Blood
Sample (VBS) participants was selected for innovative assays reflecting cellular-/molecular-level
mechanisms of aging, including RNA sequencing. This VBS innovative sub-sample fully
represents the entire HRS sample, and thus is nationally representative. After the quality control
process, the total number of people who have gene expression data is 3,738. Among them, 3,642
have complete demographic and education information. Our main analytical sample includes

77
3,580 who also have complete data for health behaviors, morbidity, and DNA methylation data. A smaller subsample of 3,575 was used for the analysis of mortality, and a subsample of 2,660
was used for the analysis of biological age due to the reduced availability of some biomarkers. Gene Expression Composite Scores
Four gene expression composite scores are used in the current study – The CCA score
(based on 22 genes), the MD score (48 genes), the SASP score (44 genes), and the senescence
summary score (112 genes) which is a combination of the previous three. Specifically, CCA
genes code the proteins that are important for establishing cell cycle withdrawal; MD genes code
the proteins mainly involving DNA repair, oxidative stress, and telomere shortening; SASP
genes code pro-inflammatory cytokines and chemokines, growth modulators, angiogenic factors, and matrix metalloproteinases (Supplemental Table 4.1). IGFBP7 and AKT1 genes are in both
MD and SASP scores, and they are only used once in the senescence summary score. Each score
is calculated by the mean of the z-score standardized log2 transformation of the normalized
transcript abundance value of the corresponding genes. A higher value of the score indicates a
higher level of cellular senescence. Specifically, for respondent i, the Scorei is calculated using the equation below. 푆푐�푖 =
�
�
퐿�2퐶�(�)
푖 − 푚� 퐿�2퐶�(� )
푠 푙�2퐶�(�

�
The subscript x indicates individual genes, and n indicates the total number of genes. CPM stands for count per million, which is the transcript counts of gene x per million total
human transcriptome-mapped RNA sequencing reads, indicating the normalized expression level
of gene x. Since the average expression level can vary substantially across genes, a log2
transformation is applied to the CPM values for each gene to prevent the results from being

78
dominated by a few highly expressed genes. Since expression level can be heteroscedastic, a z- score standardization is applied to the log2 CPM values for each gene to prevent the results from
being dominated by a few highly variable genes. Finally, the scores average the standardized
log2 transformed expression level of all genes. Outcome Measures
Mortality – The mortality information is derived from the vital status of the respondent in
the 2020 HRS interview, and thus measures 4-year mortality. If no contact is made in 2020, the
respondents are presumed alive. Respondents reported deceased as of the 2020 wave and prior
waves after 2016 are coded deceased. Multimorbidity – The multimorbidity measure in the current study is the number of selfreports of the selected physician-diagnosed health conditions (Faul et al., 2023). The 5 selected
conditions include (1) diabetes or high blood sugar; (2) cancer or a malignant tumor of any kind
except skin cancer; (3) chronic lung disease except asthma such as chronic bronchitis or
emphysema; (4) heart attack, coronary heart disease, angina, congestive heart failure, or other
heart problems; and (5) stroke or transient ischemic attack. So, the multimorbidity measure
ranges from 0-5. Biological Age – Biological age is measured using the expanded biological age
(Crimmins et al., 2021) based on 22 clinical-level biomarkers. It is a measure that indicates the
general level of physiological dysregulation and explains multimorbidity well at the population
level. To measure biological age eliminating the effect of chronological age, biological age
acceleration (AA) is used in the current study. This is calculated using the residual that results
from regressing biological age on chronological age. Since biological age (acceleration) is

79
measured in years, it can be interpreted as age acceleration or deceleration compared to
chronological age. Epigenetic Age – Epigenetic age is estimated by 3 DNA methylation clocks – GrimAge
(Lu et al., 2019) and PhenoAge (Levine et al., 2018), trained on aging-related health outcomes, and DunedinPACE (Belsky et al., 2022), trained on within-individual change across 19
biological indicators. To bolster the reliability of and reduce the influence of technical noise on
the epigenetic clock algorithms, the principal component (PC) versions (Higgins-Chen et al., 2022) of GrimAge and PhenoAge are used in the current study. Similarly to biological age, age
acceleration (AA) reflects faster or slower aging according to the PC clocks (Faul et al., 2023). As the outcomes of our models, PC GrimAge AA and PC PhenoAge AA are represented in years. DunedinPACE captures the pace of epigenetic aging. It cannot be interpreted in years, instead, it
indicates the age acceleration per year. Social and Behavioral Measures
Social and behavioral measures include age, sex, race/ethnicity, education, alcohol
consumption, smoking, body mass index (BMI) categories, and sleep disorder. Age is
categorized into 4 groups: 55-64, 65-74, 75-84, and 85 and older. Racial/ethnic groups include
non-Hispanic White, non-Hispanic Black, Hispanic, and non-Hispanic others. Education is
classified as less than high school, high school, some college, and college degree or higher. Alcohol consumption is measured by self-reported total weekly drinks, which is the product of
the average daily number of drinks and the total days of drinking per week. Smoking is measured
by pack-years, which is the product of the average daily number of cigarette packs smoked and
the lifetime years of smoking (Haghani et al., 2020). BMI is calculated using the measured
height and weight, but when physical measures are not available, self-reported BMI values are

80
used. Then, respondents are categorized into normal (BMI<25), overweight (25<=BMI<30), obesity I (30<=BMI<35), and obesity II (BMI>=35). Sleep disorder is indicated by the presence
of insomnia symptoms, which is a combined measure of sleep disturbance with non-restorative
sleep (Kusters et al., 2023). Technical Controls
Since the gene expression profile can vary across cell types, the RNA transcripts of 8
genes (CD3E, CD3D, CD4, CD8A, CD14, CD19, FCGR3A, and NCAM1) indicating the
relative prevalence of major leukocytes are included in the regression models as covariates
(Mann et al., 2020). Dummy-codes indicating 46 batch plates are also included as controls. Statistical Analysis
To understand the social-behavioral pattern of senescence scores, social and behavioral
variables as well as the technical controls were used to predict each score in OLS regression
models.To test the relationships of the senescence scores to aging measures, the associations
between each score and other aging measures at different dimensions were assessed separately in
OLS models. Epigenetic age acceleration measures were first assessed (separately) since they are
based on upstream molecular/cellular biomarkers at the same level as the senescence scores. Then, mortality, multimorbidity, and biological age were separately assessed since they are
considered downstream outcomes of the aging process. Finally, each score and PC GrimAge AA, an epigenetic age acceleration measure, were used together to predict downstream outcomes to
see whether these new scores explain additional variation in the outcome. Survey weights for the HRS VBS innovative sub-sample are used to adjust for initial
sample selection and missing data. All analyses are performed using Stata version 18.

81
Results
Sample Characteristics
Table 4.1 presents the sample characteristics. The analytical sample has a mean age of 69
(SD=9). Slightly more than half of the sample are women (54%). More than three-quarters of the
sample are non-Hispanic Whites (78%). About 10% are non-Hispanic Blacks, 9% are Hispanics, and 3% are non-Hispanic others. The average length of education is 13 (SD=3) years – Specifically, 13% did not complete high school, 30% completed high school, 26% had some
college experience, and 30% had a college degree or more. People in our sample vary in terms of
health behaviors and outcomes. On average, people have 3 (SD=6) drinks of alcohol per week
and have smoked 13 (SD=21) pack-years cumulatively. The mean BMI of the sample is 29
(SD=6), which is considered overweight. Specifically, 37% of the sample are overweight, 22%
have class I obesity, and 16% have obesity above class II. The most prevalent health conditions
among the selected are diabetes and heart problems (both around 25%). The average biological
age of the sample is 68 (SD=12), similar to the average chronological age by design. In terms of
epigenetic age, the sample has a mean GrimAge of 77 (SD=8) and a mean PhenoAge of 66
(SD=11). The means of senescence scores are all close to zero because mathematically they are the
average of a set of z-scores. As Figure 4.1 shows, all scores are normally distributed. The MD, SASP, and senescence summary scores are highly correlated (correlation coefficients range from
0.76 to 0.93), while their correlations with the CCA score are moderate (correlation coefficients
range from 0.28 to 0.52) – The summary score is mainly driven by the MD and SASP scores.

82
The correlations between the senescence scores and other continuous aging measures are overall
weak. Social and Behavioral Factors Predicting the Senescence Scores
Results from the OLS model using social and behavioral factors to predict the senescence
scores can be found in Table 4.2. Older age groups (aged 65-74: β=0.03, p=0.025; aged 75-84: β =0.07, p<0.001; aged 85+: β =0.04, p=0.006) have significantly higher SASP scores compared to
those aged 55-64. For MD and summary scores, only those aged 75-84 (MD: β =0.04, p=0.001;
summary score: β =0.04, p<0.001) are higher than the youngest group. The age pattern of CCA
is very different from the other scores. Those aged 85+ (β =-0.06, p=0.001) and those aged 75-84
(β =-0.07, p<0.001) have lower CCA compared to the youngest. Overall, women have a lower
level of cellular senescence indicated by all scores (β ranges from 0.05 to 0.19 depending on the
scores, all p<0.001). All senescence scores are higher among Non-Hispanic Blacks compared to
non-Hispanic Whites (β ranges from 0.03 to 0.12 depending on the scores, all p<0.05). Those
who received college-level or higher education have lower MD (β =-0.06, p=0.002) and
summary (β =-0.04, p=0.046) scores. In terms of the behavioral factors including BMI category, smoking, drinking, and
insomnia symptoms, only the first is significantly associated with some of the scores, and the rest
all show null results. Among all scores, the MD score is most significantly associated with BMI
categories – A higher MD score is found among those overweight (β =0.03, p=0.024), those with
class I obesity (β =0.06, p<0.001), and those with class II obesity (β =0.08, p<0.001) compared
to the normal BMI group. The summary score is higher among obese individuals (obesity I: β =0.04, p=0.005; obesity II: β =0.08, p<0.001). Only those with class II obesity have a higher
SASP score (β =0.06, p=0.001).

83
The Senescence Scores and Epigenetic Aging Measures
The association between the senescence scores and epigenetic aging measures can be
found in Figure 4.2 (Panel A), and the detailed results of each model can be found in
Supplemental Table 4.2-4.4. The MD, SASP, and senescence summary scores are significantly
associated with faster PC GrimAge acceleration (MD: β=0.22, p<0.001; SASP: β=0.18, p<0.001;
Summary score: β=0.21, p<0.001), PC PhenoAge acceleration (MD: β=0.35, p<0.001; SASP:
β=0.28, p<0.001; Summary score: β=0.32, p<0.001), and DunedinPACE (MD: β=0.26, p<0.001;
SASP: β=0.18, p<0.001; Summary score: β=0.24, p<0.001). The CCA score is only significantly
associated with DunedinPACE (β=0.05, p=0.021). The variance explained across the models
based on the MD, SASP, and summary scores are fairly consistent, while the CCA models
always explain less of the variance in the outcome. The Senescence Scores and Age-Related Health Outcomes
The association between the senescence scores and downstream age-related health
outcomes can be found in Figure 4.2 (Panels B & C), and the detailed results of each model can
be found in Supplemental Tables 4.5-4.10. Only the MD score is associated with 4-year
mortality (OR=4.67, p=0.003). MD, SASP, and summary scores are significantly associated with
multimorbidity (β=0.21, 0.12, & 0.17 respectively, all p<0.001) and faster biological age
acceleration measures (β=0.29, 0.16, & 0.21 respectively, all p<0.001). Overall, for MD, SASP, and the summary scores, the size of the beta coefficients predicting biological age is greater than
those predicting multimorbidity. The size of those predicting epigenetic age is greater than those
predicting biological age. After including the PC GrimAge AA in the previous models, PC
GrimAge AA is always significantly associated with mortality (OR ranges from 1.16 to 1.18, depending on the inclusion of different senescence scores; all p<0.001), multimorbidity (β ranges

84
from 0.20 to 0.22, depending on the inclusion of different senescence scores; all p<0.001), and
biological age (β ranges from 0.29 to 0.31, depending on the inclusion of different senescence
scores; all p<0.001). The association between the MD score and mortality becomes marginally
significant (β=2.71, p=0.056) after the inclusion of PC GrimAge AA. The coefficients of the MD, SASP, and summary scores predicting multimorbidity (MD: β=0.17, p<0.001; SASP: β=0.08, p=0.001; Summary score: β=0.13, p<0.001) and biological age (MD: β=0.23, p<0.001; SASP:
β=0.10, p=0.001; Summary score: β=0.15, p<0.001) are still significant with a reduced size after
the inclusion. Discussion
The current study examines how four RNA-based cellular senescence scores link to
social-behavioral characteristics and measures of other dimensions of aging health in a nationally
representative sample of older Americans. The CCA, MD, and SASP scores correspond to three
hallmarks of cellular senescence, and the senescence summary score is a combination of the
three. We find that cellular senescence is sensitive to certain demographic, socioeconomic, and
behavioral factors: Overall, older individuals tend to have higher senescence scores, especially
for the SASP score. However, the CCA score decreases with age. Women and non-Hispanic
Blacks have higher levels of cellular senescence indicated by all scores. People with a higher
education level seem to have lower MD and summary scores. Among all the behavioral factors
examined in the current study, only BMI has a significant association with the senescence scores. Obese individuals have a higher senescence summary score, and it is mainly driven by MD and
SASP.

85
We also find that cellular senescence links to other dimensions of aging. The senescence
summary score significantly links to all assessed aging measures except for 4-year mortality, and
the associations are driven by the MD and SASP scores, but not the CCA score. At the
downstream dimension, aging health outcomes including 4-year mortality, multimorbidity, and
overall physiological dysregulation indicated by the clinical-level-biomarker-based expanded
biological age acceleration are assessed. At the upstream dimension, DNA-methylation-based
epigenetic age acceleration is assessed. According to the model of mortality process (Crimmins, 2015; Zhao & Crimmins, 2022), same as epigenetic changes (e.g., DNA methylation), cellular
senescence belongs to the dimension of molecular-/cellular-level changes, which occur prior to
the other outcomes. Our results align well with this model since the beta coefficients of MD, SASP, and summary scores are generally greater in the models measuring upstream aging
measures and attenuate gradually as the outcome of the model moves downstream. When putting
each of the senescence scores together with epigenetic age acceleration to predict downstream
age-related outcomes, MD, SASP, and summary scores are still significantly associated with
multimorbidity and biological age. It indicates that by further capturing the underlying
mechanism of aging, the senescence scores may have the potential to explain additional variation
in health outcomes compared to the existing biomarkers. Female sex is associated with a higher level of all senescence scores after controlling for
other social and behavioral factors in our model (Table 4.2). This finding is consistent with the
existing literature. There is evidence that women overall have a lower capacity for DNA damage
repair, greater DNA damage repair decline with age, and female cells have a greater tendency to
undergo cellular senescence in response to genotoxic stress (Ng & Hazrati, 2022; Olivieri et al., 2023). Our study provides preliminary human evidence on the sex difference in cellular

86
senescence at the population level. The effect of minority status (specifically being non-Hispanic
Black versus non-Hispanic White) and high education on senescence scores, especially the MD
score, indicates that the overall level of cellular senescence can be sensitive to socioeconomic
status (SES). People who belong to racial minority groups and receive low-level education may
have more adverse exposures (e.g., environmental, chemical, and psychological) and fewer
protective resources (e.g., occupational, social, and healthcare) and hence accumulate a higher
level of MD throughout their life span. Carroll et al. (2016) found that partial sleep deprivation was associated with elevated
expression levels of both DDR and SASP genes. However, in our model, the presence of
insomnia symptoms is not significantly associated with any of the senescence scores, including
MD and SASP. The discrepancy could be due to several reasons. Carroll et al. constructed their
SASP score using 9 genes including IL6, CSF2, CCL8, IL8, CCL13, ICAM1, CXCL1, CXCL2, and CXCL3. Compared to the 44-gene score used in the current study, the 9-gene SASP score is
less comprehensive and has a strong emphasis on inflammatory response. Similarly, compared to
our 48-gene MD score, their DDR score only focuses on a specific type of macromolecular
damage, and thus may depict a less holistic picture. In the Carroll et al. study, partial sleep
deprivation was mandatorily executed – participants were not allowed to sleep from 11 PM to 3
AM the night of the intervention. However, in the current study, the criteria for insomnia
symptoms is not as strict. In addition, Carroll et al. measured the gene expression profile of their
respondents in the morning right after the night when partial sleep intervention took place. In
summary, the elevation of the DDR and SASP scores in the Carroll et al. study should be
interpreted as an acute inflammatory response to extreme lack of sleep, while the model in the
current study examines a less extreme and long-term association between insomnia symptoms

87
and cellular senescence. In fact, the SASP score among the respondents of the Carrol et al. study
returned to the baseline level after uninterrupted sleep, indicating that the impact was acute and
reversible. The current study has limitations. The genes included in the composite scores mainly
come from studies based on various brain cells, but the RNA data of the current study are based
on venous blood. The expression level of the genes might differ across tissue/cell types and thus
the composite scores need to be further compared between brain and blood cells samples. One
potential source is the Religious Orders Study/Memory and Aging Project (ROSMAP:
doi:10.1038/s41593-018-0154-9), where RNA sequencing data are available from both brain
cells and blood cells. Our analyses reveal the potential of a set of cellular senescence scores in
measuring aging, but the validity of this measure needs to be further tested using different
samples. On one hand, whether this measure applies to a younger population remains unknown
but intriguing. On the other hand, its performance under different social contexts needs to be
assessed. Tables/Figures
Table 4.1. Sample Characteristics
Mean (SD) /
Proportion
Mean (SD) /
Proportion
N=3,580 N=3,580
Senescence Scores Total Drinks Weekly 2.9 (6.3)
Summary Score 0.0 (0.2) Cumulative Packs Smoked 13.1 (20.7)
CCA Score 0.0 (0.2) BMI 29.4 (6.6)
MD Score 0.0 (0.3) Normal 23.8
SASP Score 0.0 (0.3) Overweight 37.4
Age 68.6 (9.2) Obesity I 22.7
Aged 55-64 40.5 Obesity II 16.2
Aged 65-74 35.2 Insomnia Symptoms 20.4
Aged 75-84 17.2 4-Year Mortality (N=3,575) 9.8
Aged 85+ 7.1 Multimorbidity 0.8 (1.0)
Female 54.4 Diabetes 25.5

88
Race Ethnicity Cancer 14.6
Non-Hispanic White 77.5 Lung Disease 11.0
Non-Hispanic Black 10.2 Heart Problems 25.0
Hispanic 8.9 Stroke 8.1
Non-Hispanic Other 3.3 Expanded BioAge (N=2,660) 68.2 (11.7)
Education Expanded BioAge AA (N=2,660) 0.0 (8.1)
Years of Education 13.3 (3.0) PC GrimAge 77.3 (8.0)
Less than High School 14.0 PC PhenoAge 66.0 (10.6)
High School 29.9 DunedinPACE 1.0 (0.1)
Some College 25.9 PC GrimAge AA 0.0 (4.0)
College and Higher 30.2 PC PhenoAge AA 0.0 (6.5)
Note CCA – Cell Cycle Arrest; MD – Macromolecular Damage; SASP – Senescence-Associated Secretory Phenotype; BioAge – Expanded Biological Age; AA – Age Acceleration; PC – Principal Component. Table 4.2. Results of the OLS Model Predicting the Senescence Scores (N=3,580)
N=3,580 CCA MD SASP Sum Score
Age: Ref - Aged 55-64
Aged 65-74 -0.04† 0.01 0.03* 0.01
Aged 75-84 -0.06*** 0.04** 0.07*** 0.04*** Aged 85+ -0.07*** 0.01 0.04** 0.01
Female 0.19*** 0.05*** 0.12*** 0.12*** RE: Ref - Non-Hispanic White
Non-Hispanic Black 0.12*** 0.06*** 0.03* 0.07*** Hispanic -0.02 0.01 0.01 0.00
Non-Hispanic Other 0.00 0.01 0.00 0.01
Education: Ref - Less than High School
High School -0.03 -0.01 0.02 0.00
Some College -0.01 -0.02 0.00 -0.01
College and Higher -0.02 -0.06** -0.01 -0.04* BMI: Ref - Normal
Overweight 0.01 0.03* 0.01 0.02
Obese I 0.01 0.06*** 0.01 0.04** Obese II 0.03 0.08*** 0.06*** 0.08*** Cumulative Packs Smoked 0.03 0.02† 0.00 0.02
Total Drinks Weekly 0.03 -0.01 0.00 0.00
Insomnia Symptoms 0.01 0.02 0.00 0.01
Adjusted R2 0.22 0.66 0.55 0.60
Note † p<0.1 * p<0.05, ** p<0.01, *** p<0.001
The model is adjusted for patch/plate and cell types. Beta coefficients are reported. CCA – Cell Cycle Arrest; MD – Macromolecular Damage; SASP – Senescence-Associated Secretory Phenotype; Sum Score – Senescence Summary Score. Figure 4.1. The Distribution and Correlation Matrix of the Cellular Senescence Scores

89
Note CCA – Cell Cycle Arrest; MD – Macromolecular Damage; SASP – Senescence-Associated Secretory Phenotype; Sum Score – Senescence Summary Score; BioAge – Biological Age; AA – Age Acceleration; PC – Principal Component. Figure 4.2. The Associations between Cellular Senescence Scores and Multiple Dimensions of
Aging Measures/Outcomes
Note

90
† p<0.1 * p<0.05, ** p<0.01, *** p<0.001
CCA – Cell Cycle Arrest; MD – Macromolecular Damage; SASP – Senescence-Associated Secretory Phenotype; Sum Score – Senescence Summary Score; BioAge – Biological Age; AA – Age Acceleration; PC – Principal Component. All models are adjusted for all covariates (age, sex, race/ethnicity, education, BMI categories, smoking, drinking, insomnia
symptoms), patch/plate, and cell types. Odds Ratios are reported for mortality. Beta coefficients are reported for the other outcomes. For the model predicting mortality, N=3,575; for the model predicting BioAgeAA, N=2,660; for all other models, N=3,580. Supplemental Information
Supplemental Table 4.1. The Gene Lists for Gene Expression Composite Scores
CCA Genes MD Genes SASP Genes
CDKN2D SOD1 IGFBP7
ETS2 MAP2K1 VIM
RB1 GSK3B FN1
E2F3 PIK3CA SPARC
CDK6 SOD2 IGFBP4
RBL2 MAPK14 TIMP1
ATM IGF1R TBX2
BMI1 TP53BP1 TBX3
MDM2 NBN COL1A1
CDK4 HRAS COL3A1
CCNE1 CITED2 IGFBP2
E2F1 CREG1 TGFB1I1
CHEK2 ABL1 PTEN
CHEK1 MORC3 CD44
CDKN1A NFKB1 NFIA
TWIST1 AKT1 CALR
CCND1 CDKN1B TIMP2
ETS1 EGR1 CXCL8
TP53 RBL1 IL6
CDKN2A MAP2K6 FGF2
CDK2 IGF1 FGF7
SATB1 IRF3 AKT1
- PCNA CXCL2
- GADD45A VEGFA
- MAP2K3 CXCL1
- IGFBP5 PLAUR
- SIRT1 SERPINE1
- ING1 LMNB1
- TGFB1 GLB1
- TERF2 VEGFB
- CCNB1 CCL2
- PRKCD IL1B
- CDC25C CXCL5
- IGFBP3 SERPINB2
- ALDH1A3 IL11
- MYC IL1A
- NOX4 CCL5
- CCNA2 TNF
- CDKN2C CCL20
- TERT MMP1

91
- ID1 CCL8
- IGFBP7 MMP3
- CDKN1C MMP12
- IRF7 MMP10
- IFNG -
- CDKN2B -
- PLAU -
- IRF5 - Note CCA – Cell Cycle Arrest Score; MD – Macromolecular Damage Score; SASP – Senescence-Associated Secretory Phenotype Score Source: Dehkordi SK, Walker J, Sah E, Bennett E, Atrian F, Frost B, Woost B, Bennett RE, Orr TC, Zhou Y, Andhey PS. Profiling senescent cells in human brains reveals neurons with CDKN2D/p19 and tau neuropathology. Nature aging. 2021
Dec;1(12):1107-16. https://doi.org/10.1038/s43587-021-00142-3
Supplemental Table 4.2. Results from OLS Regression Models Assessing the Associations
between Senescence Scores and PC GrimAge AA (N=3,580)
N=3,580 CCA MD SASP Sum Score
Gene Expression Score 0.02 0.22*** 0.18*** 0.21*** Age: Ref - Aged 55-64
Aged 65-74 -0.03 -0.03† -0.04* -0.03†
Aged 75-84 -0.04* -0.05** -0.05*** -0.05** Aged 85+ -0.04* -0.04** -0.05** -0.04* Female -0.30*** -0.31*** -0.32*** -0.33*** RE: Ref - Non-Hispanic White
Non-Hispanic Black 0.08*** 0.06*** 0.07*** 0.07*** Hispanic -0.03† -0.03† -0.03† -0.03†
Non-Hispanic Other 0.00 0.00 0.00 0.00
Education: Ref - Less than High School
High School -0.06* -0.06* -0.06** -0.06* Some College -0.10*** -0.09*** -0.10*** -0.10*** College and Higher -0.21*** -0.20*** -0.21*** -0.20*** BMI: Ref - Normal
Overweight -0.03 -0.04† -0.03 -0.04†
Obese I -0.05** -0.07*** -0.06** -0.06** Obese II 0.03 0.01 0.02 0.02
Cumulative Packs Smoked 0.35*** 0.35*** 0.35*** 0.35*** Total Drinks Weekly 0.04* 0.05** 0.04* 0.04*
Insomnia Symptoms 0.04** 0.04* 0.04** 0.04** Adjusted R2 0.42 0.44 0.44 0.44
Note † p<0.1 * p<0.05, ** p<0.01, *** p<0.001
All models are adjusted for patch/plate and cell types. Beta coefficients are reported. CCA – Cell Cycle Arrest; MD – Macromolecular Damage; SASP – Senescence-Associated Secretory Phenotype; Sum Score – Senescence Summary Score. Supplemental Table 4.3. Results from OLS Regression Models Assessing the Associations
between Senescence Scores and PC PhenoAge AA (N=3,580)

92
N=3,580 CCA MD SASP Sum Score
Gene Expression Score 0.03 0.35*** 0.28*** 0.32*** Age: Ref - Aged 55-64
Aged 65-74 -0.02 -0.02 -0.03 -0.02
Aged 75-84 -0.05** -0.07*** -0.07*** -0.07*** Aged 85+ -0.04† -0.05* -0.05* -0.05†
Female -0.10*** -0.11*** -0.12*** -0.13*** RE: Ref - Non-Hispanic White 0.00 0.00 0.00 0.00
Non-Hispanic Black 0.20*** 0.18*** 0.19*** 0.18*** Hispanic 0.05* 0.04* 0.04* 0.04* Non-Hispanic Other 0.03* 0.03† 0.03* 0.03* Education: Ref - Less than High School 0.00 0.00 0.00 0.00
High School 0.00 0.00 -0.01 0.00
Some College -0.04 -0.03 -0.04 -0.03
College and Higher -0.11*** -0.09** -0.11*** -0.10*** BMI: Ref - Normal 0.00 0.00 0.00 0.00
Overweight 0.00 -0.01 0.00 0.00
Obese I 0.08*** 0.06** 0.08*** 0.07** Obese II 0.17*** 0.14*** 0.16*** 0.15*** Cumulative Packs Smoked 0.08*** 0.08*** 0.09*** 0.08*** Total Drinks Weekly -0.06*** -0.05** -0.06** -0.06**
Insomnia Symptoms 0.03† 0.03 0.03† 0.03
Adjusted R2 0.25 0.29 0.29 0.29
Note † p<0.1 * p<0.05, ** p<0.01, *** p<0.001
All models are adjusted for patch/plate and cell types. Beta coefficients are reported. CCA – Cell Cycle Arrest; MD – Macromolecular Damage; SASP – Senescence-Associated Secretory Phenotype; Sum Score – Senescence Summary Score. Supplemental Table 4.4. Results from OLS Regression Models Assessing the Associations
between Senescence Scores and DunedinPACE (N=3,580)
N=3,580 CCA MD SASP Sum Score
Gene Expression Score 0.05* 0.26*** 0.18*** 0.24*** Age: Ref - Aged 55-64
Aged 65-74 0.07** 0.06** 0.06** 0.06** Aged 75-84 0.13*** 0.12*** 0.11*** 0.12*** Aged 85+ 0.12*** 0.12*** 0.11*** 0.12*** Female -0.08*** -0.09*** -0.10*** -0.10*** RE: Ref - Non-Hispanic White
Non-Hispanic Black 0.19*** 0.18*** 0.19*** 0.18*** Hispanic 0.09*** 0.09*** 0.09*** 0.09*** Non-Hispanic Other 0.04* 0.04* 0.04* 0.04* Education: Ref - Less than High School
High School -0.07** -0.07** -0.08** -0.07** Some College -0.12*** -0.11*** -0.12*** -0.12*** College and Higher -0.23*** -0.22*** -0.23*** -0.23*** BMI: Ref - Normal
Overweight 0.10*** 0.09*** 0.10*** 0.10*** Obese I 0.14*** 0.13*** 0.14*** 0.13*** Obese II 0.26*** 0.24*** 0.25*** 0.24*** Cumulative Packs Smoked 0.19*** 0.18*** 0.19*** 0.19***

93
Total Drinks Weekly -0.02 -0.02 -0.02 -0.02
Insomnia Symptoms 0.04* 0.04* 0.04* 0.04* Adjusted R2 0.26 0.28 0.27 0.28
Note † p<0.1 * p<0.05, ** p<0.01, *** p<0.001
All models are adjusted for patch/plate and cell types. Beta coefficients are reported. CCA – Cell Cycle Arrest; MD – Macromolecular Damage; SASP – Senescence-Associated Secretory Phenotype; Sum Score – Senescence Summary Score. Supplemental Table 4.5. Results from Logistic Regression Models Assessing the Associations
between Senescence Scores and 4-Year Mortality (N=3,575)
N=3,575 CCA MD SASP Sum Score
Gene Expression Score 0.83 4.67** 2.12 3.48
Age: Ref - Aged 55-64
Aged 65-74 1.37 1.37 1.36 1.36
Aged 75-84 3.21*** 3.14*** 3.11*** 3.15*** Aged 85+ 16.34*** 16.56*** 16.16*** 16.54*** Female 1.02 0.96 0.96 0.94
RE: Ref - Non-Hispanic White
Non-Hispanic Black 2.26*** 2.01** 2.16*** 2.06** Hispanic 0.62 0.64 0.62 0.63
Non-Hispanic Other 1.17 1.09 1.12 1.11
Education: Ref - Less than High School
High School 0.65* 0.66* 0.64* 0.65* Some College 0.42*** 0.43*** 0.42*** 0.42*** College and Higher 0.31*** 0.32*** 0.31*** 0.32*** BMI: Ref - Normal
Overweight 0.65* 0.62** 0.64* 0.63* Obese I 0.76 0.70 0.75 0.73
Obese II 0.70 0.64† 0.69 0.67†
Cumulative Packs Smoked 1.02*** 1.01*** 1.02*** 1.02*** Total Drinks Weekly 0.97† 0.97† 0.97† 0.97†
Insomnia Symptoms 1.47* 1.45* 1.46* 1.45* Pseudo R2 0.27 0.27 0.27 0.27
Note † p<0.1 * p<0.05, ** p<0.01, *** p<0.001
All models are adjusted for patch/plate and cell types. Odds ratios are reported. CCA – Cell Cycle Arrest; MD – Macromolecular Damage; SASP – Senescence-Associated Secretory Phenotype; Sum Score – Senescence Summary Score. Supplemental Table 4.6. Results from Logistic Regression Models Assessing the Associations
between Senescence Scores and 4-Year Mortality Adjusted for PC GrimAge AA (N=3,575)
N=3,575 CCA MD SASP Sum Score
Gene Expression Score 0.68 2.71† 1.43 1.83
PC GrimAge AA 1.18*** 1.16*** 1.17*** 1.17*** Age: Ref - Aged 55-64
Aged 65-74 1.48 1.46 1.46 1.46
Aged 75-84 3.93*** 3.81*** 3.84*** 3.85***

94
Aged 85+ 22.12*** 21.89*** 21.90*** 22.05*** Female 1.52* 1.40* 1.42* 1.41* RE: Ref - Non-Hispanic White
Non-Hispanic Black 2.02** 1.85** 1.93** 1.89** Hispanic 0.65 0.67 0.66 0.67
Non-Hispanic Other 1.24 1.18 1.21 1.20
Education: Ref - Less than High School
High School 0.72† 0.72† 0.71† 0.71†
Some College 0.49*** 0.49*** 0.48*** 0.49*** College and Higher 0.40*** 0.40*** 0.39*** 0.40*** BMI: Ref - Normal
Overweight 0.65* 0.64* 0.65* 0.65* Obese I 0.87 0.81 0.85 0.84
Obese II 0.74 0.69 0.72 0.71
Cumulative Packs Smoked 1.01* 1.01* 1.01* 1.01* Total Drinks Weekly 0.97* 0.97* 0.97* 0.97*
Insomnia Symptoms 1.37† 1.36† 1.36† 1.36†
Pseudo R2 0.29 0.29 0.29 0.29
Note † p<0.1 * p<0.05, ** p<0.01, *** p<0.001
All models are adjusted for patch/plate and cell types. Odds ratios are reported. CCA – Cell Cycle Arrest; MD – Macromolecular Damage; SASP – Senescence-Associated Secretory Phenotype; Sum Score – Senescence Summary Score. Supplemental Table 4.7. Results from OLS Regression Models Assessing the Associations
between Senescence Scores and Multimorbidity (N=3,580)
N=3,580 CCA MD SASP Sum Score
Gene Expression Score 0.02 0.21*** 0.12*** 0.17*** Age: Ref - Aged 55-64
Aged 65-74 0.11*** 0.11*** 0.11*** 0.11*** Aged 75-84 0.15*** 0.15*** 0.14*** 0.15*** Aged 85+ 0.16*** 0.15*** 0.15*** 0.16*** Female -0.04* -0.05* -0.05** -0.06** RE: Ref - Non-Hispanic White
Non-Hispanic Black 0.06** 0.05** 0.06** 0.05** Hispanic 0.01 0.01 0.01 0.01
Non-Hispanic Other 0.03 0.02 0.03 0.02
Education: Ref - Less than High School
High School -0.07* -0.06* -0.07* -0.07* Some College -0.05† -0.05† -0.05† -0.05†
College and Higher -0.12*** -0.11*** -0.12*** -0.12*** BMI: Ref - Normal
Overweight 0.00 -0.01 0.00 -0.01
Obese I 0.09*** 0.08*** 0.09*** 0.09*** Obese II 0.13*** 0.12*** 0.13*** 0.12*** Cumulative Packs Smoked 0.14*** 0.13*** 0.14*** 0.13*** Total Drinks Weekly -0.08*** -0.08*** -0.08*** -0.08***
Insomnia Symptoms 0.10*** 0.09*** 0.10*** 0.10*** Adjusted R2 0.14 0.16 0.15 0.16
Note † p<0.1 * p<0.05, ** p<0.01, *** p<0.001
All models are adjusted for patch/plate and cell types. Beta coefficients are reported.

95
CCA – Cell Cycle Arrest; MD – Macromolecular Damage; SASP – Senescence-Associated Secretory Phenotype; Sum Score – Senescence Summary Score. Supplemental Table 4.8. Results from OLS Regression Models Assessing the Associations
between Senescence Scores and Multimorbidity Adjusted for PC GrimAge AA (N=3,580)
N=3,580 CCA MD SASP Sum Score
Gene Expression Score 0.01 0.17*** 0.08** 0.13*** PC GrimAge AA 0.22*** 0.20*** 0.21*** 0.20*** Age: Ref - Aged 55-64
Aged 65-74 0.12*** 0.12*** 0.12*** 0.12*** Aged 75-84 0.16*** 0.15*** 0.16*** 0.16*** Aged 85+ 0.17*** 0.16*** 0.16*** 0.16*** Female 0.02 0.01 0.01 0.01
RE: Ref - Non-Hispanic White
Non-Hispanic Black 0.04* 0.04† 0.04* 0.04* Hispanic 0.02 0.01 0.02 0.01
Non-Hispanic Other 0.03 0.02 0.03 0.02
Education: Ref - Less than High School
High School -0.05† -0.05† -0.06† -0.05†
Some College -0.03 -0.03 -0.03 -0.03
College and Higher -0.08** -0.07* -0.08** -0.08** BMI: Ref - Normal
Overweight 0.00 0.00 0.00 0.00
Obese I 0.11*** 0.09*** 0.10*** 0.10*** Obese II 0.13*** 0.11*** 0.12*** 0.12*** Cumulative Packs Smoked 0.06** 0.06** 0.06** 0.06** Total Drinks Weekly -0.09*** -0.09*** -0.09*** -0.09***
Insomnia Symptoms 0.09*** 0.09*** 0.09*** 0.09*** Adjusted R2 0.17 0.18 0.17 0.18
Note † p<0.1 * p<0.05, ** p<0.01, *** p<0.001
All models are adjusted for patch/plate and cell types. Beta coefficients are reported. CCA – Cell Cycle Arrest; MD – Macromolecular Damage; SASP – Senescence-Associated Secretory Phenotype; Sum Score – Senescence Summary Score; AA – Age Acceleration; PC – Principal Component. Supplemental Table 4.9. Results from OLS Regression Models Assessing the Associations
between Senescence Scores and Biological Age Acceleration (N=2,660)
N=2,660 CCA MD SASP Sum Score
Gene Expression Score 0.02 0.29*** 0.16*** 0.21*** Age: Ref - Aged 55-64
Aged 65-74 -0.08*** -0.09*** -0.09*** -0.09*** Aged 75-84 -0.04 -0.05* -0.05* -0.05* Aged 85+ -0.01 -0.02 -0.02 -0.02
Female -0.04† -0.05* -0.06* -0.06** RE: Ref - Non-Hispanic White
Non-Hispanic Black 0.10*** 0.09*** 0.10*** 0.09***

96
Hispanic 0.05† 0.04† 0.05† 0.04†
Non-Hispanic Other 0.04* 0.04* 0.04* 0.04†
Education: Ref - Less than High School
High School -0.01 -0.02 -0.02 -0.02
Some College -0.08* -0.08* -0.09* -0.08* College and Higher -0.17*** -0.16*** -0.17*** -0.17*** BMI: Ref - Normal
Overweight 0.01 0.00 0.01 0.01
Obese I 0.06* 0.04 0.06* 0.05†
Obese II 0.15*** 0.13*** 0.14*** 0.13*** Cumulative Packs Smoked 0.11*** 0.10*** 0.11*** 0.11*** Total Drinks Weekly -0.09*** -0.09*** -0.09*** -0.09***
Insomnia Symptoms 0.05* 0.05* 0.05* 0.05* Adjusted R2 0.18 0.21 0.19 0.20
Note † p<0.1 * p<0.05, ** p<0.01, *** p<0.001
All models are adjusted for patch/plate and cell types. Beta coefficients are reported. CCA – Cell Cycle Arrest; MD – Macromolecular Damage; SASP – Senescence-Associated Secretory Phenotype; Sum Score – Senescence Summary Score. Supplemental Table 4.10. Results from OLS Regression Models Assessing the Associations
between Senescence Scores and Biological Age Acceleration Adjusted for PC GrimAge AA
(N=2,660)
N=2,660 CCA MD SASP Sum Score
Gene Expression Score 0.01 0.23*** 0.10** 0.15*** PC GrimAge AA 0.31*** 0.29*** 0.30*** 0.29*** Age: Ref - Aged 55-64
Aged 65-74 -0.08** -0.08*** -0.08*** -0.08*** Aged 75-84 -0.03 -0.04† -0.04† -0.04†
Aged 85+ 0.00 -0.01 -0.01 -0.01
Female 0.05* 0.04 0.04 0.03
RE: Ref - Non-Hispanic White
Non-Hispanic Black 0.08** 0.07** 0.08** 0.07** Hispanic 0.05* 0.05* 0.05* 0.05* Non-Hispanic Other 0.04* 0.04* 0.04* 0.04* Education: Ref - Less than High School
High School 0.00 0.00 0.00 0.00
Some College -0.05 -0.05 -0.06 -0.05
College and Higher -0.11** -0.11** -0.11** -0.11** BMI: Ref - Normal
Overweight 0.02 0.01 0.02 0.02
Obese I 0.08** 0.06* 0.07** 0.07** Obese II 0.14*** 0.12*** 0.14*** 0.13*** Cumulative Packs Smoked 0.00 0.01 0.01 0.01
Total Drinks Weekly -0.11*** -0.10*** -0.11*** -0.11***
Insomnia Symptoms 0.04† 0.03 0.04† 0.04†
Adjusted R2 0.23 0.25 0.24 0.24
Note † p<0.1 * p<0.05, ** p<0.01, *** p<0.001
All models are adjusted for patch/plate and cell types. Beta coefficients are reported.

97
CCA – Cell Cycle Arrest; MD – Macromolecular Damage; SASP – Senescence-Associated Secretory Phenotype; Sum Score – Senescence Summary Score.

98
Chapter 5: Conclusion and Discussion
The three empirical chapters (Chapters 2-4) in this dissertation contributed to the
literature of biodemographic research on aging. The first two chapters add longitudinal and
international evidence to the current discussion on the trajectory of clinical-level physiological
biomarkers and their associations with health outcomes. Chapter 2 examines the 8-year CMR
trajectory of a nationally representative sample of Americans with a mean age of 63 who
survived from 2006/2008 to 2014/2016. No significant age-related change in total CMR is found, and the CMR rate of change does not differ across age, gender, racial/ethnic, education, and
smoking population sub-groups. However, individual biomarkers used to construct the total
CMR do have different, even opposite trajectories – HbA1c, WC, and PP increased and DBP, RHR, and TC decreased. Medication contributes to the decrease in DBP, RHR, and TC. The
trajectory of individual biomarkers does differ across population sub-groups. The increase in
HbA1c is particularly significant among the non-Hispanic Black population, indicating a need
for better chronic condition screening and management, and the increase in WC is notable
among all education groups who completed high school, indicating a risk for central obesity and
a need for healthier lifestyle. Chapter 3 further shows that both a higher CMR level and an
increasing CMR over time while aging are associated with worse cognitive function among older
Americans (the association between CMR increase and cognitive function is no longer
statistically significant after accounting for education). Specifically, those with high-risk SBP, PP, HbA1c, and CRP values have significantly worse cognitive function (adjusted for age and
gender). However, interestingly, despite having an overall lower cognitive function score, among
older Chinese, the associations between CMR/individual biomarkers and cognitive function

99
cannot be found. The discrepancy in findings between the countries highlights the potential
influence of epidemiological and developmental contexts on health and aging. The insignificant
results in China might be due to the unique social pattern of CMR and that older Chinese might
have experienced a more gradual rise in CMR throughout their adulthood. After the investigation of CMR, which is a biomarker at the clinical level, Chapter 4
switches the focus to the cellular-/molecular-level and tests the potential of several RNA-based
biomarkers in measuring cellular senescence – one of the fundamental biological mechanisms of
aging. In our US nationally representative sample, the level of secretory phenotype (indicated by
the SASP score) is significantly higher at older ages while cell cycle arrest (indicated by the
CCA score) is significantly lower. In general, women and non-Hispanic Blacks have a higher
level of cellular senescence, and obese individuals are subject to higher levels of macromolecular
damage (MD score) and secretory phenotype (SASP score). Overall, the senescence RNA scores
are significantly associated with multimorbidity, biological age, and a set of epigenetic aging
measures, except for the CCA. The level of MD is associated with significantly higher odds of 4- year mortality. Based on the results, these senescence RNA scores capture multiple dimensions
of aging in a useful way. Notably, MD, SASP, and the senescence summary score are still
significantly associated with multimorbidity adjusted for the epigenetic aging measure, indicating that measuring cellular senescence adds to our understanding of physiological
dysregulation and age-related diseases. Altogether, the empirical chapters once again highlight
the value of biomarkers in understanding how social and biological factors interact and affect the
process of health and aging. The current study emphasizes the utilization of biomarkers, but the results also restate the
relative importance of social factors in explaining health outcomes – demographic,

100
socioeconomic, and behavioral variables are significantly associated with mortality (Chapter 4), multimorbidity (Chapter 4), cognitive function (Chapter 3), physiological risk indicated by
CMR (Chapter 2 & 3) or biological age (Chapter 4), and more fundamental aging measures
such as epigenetic aging measures and cellular senescence RNA scores (Chapter 4). As
previously discussed, the importance of social factors has been repeatedly reported in the aging
literature. When putting demographic characteristics, Crimmins (2020) found that when putting
social hallmarks of aging and various biological aging measures into the same model, social
factors (demographics, SES, childhood health and hardship & adult trauma, psychological, and
behaviors) explained a greater proportion of the variance in health outcomes (functional
difficulties, multimorbidity, cognitive dysfunction, and mortality) than biological and/or genetic
factors. The strong influence of social/behavioral factors on aging has implications for future
research. On one hand, as Crimmins pointed out in the paper, better ways to measure all
mechanisms related to the social hallmarks of aging are needed. For instance, future work may
aim at better-incorporating information from early life stages to examine the “long arm of
childhood” – the long-term effect of early-life factors on health and aging. On the other hand, the
strong influence of social factors means that other than the importance of technological
advancement and innovative medical/pharmaceutical interventions, policy interventions targeting
disparity, discrimination, resource distribution, education, health literacy, health behaviors, and
social support per se could potentially make a huge impact on the improvement of population
health and extension of health span. In Chapter 3 specifically, the comparison between the US and China reveals the
potential influence of the broader context and the value of harmonized survey designs and
measures. By far, cross-country studies are still often limited by the availability of data and are

101
underappreciated as a valuable approach to understanding human aging. However, in the future, this is likely to change thanks to the innovation in data collection and harmonization. For
instance, studies from more countries, especially from low-and-middle-income countries
(LMICs), are joining the International Family of Health and Retirement Studies (Gateway to
Global Aging, https://g2aging.org/home) and harmonization networks such as the Harmonized
Cognitive Assessment Protocol (HCAP, https://hcap.isr.umich.edu/). The collection of
biomarkers and the longitudinal follow-up of biomarker collection are also on the agenda of
many population surveys around the globe. In the future, researchers will be able to tackle a wide
range of questions using data that were previously not accessible. Conducting global aging
survey/research has multiple merits. First, it helps improve equity in the representation of data
and evidence, especially in terms of expanding the literature from WEIRD societies to LMICs. Second, it helps examine unique factors that only exist under certain contexts, such as a
rudimental level of education and a high level of pollution. Third, it serves as a good tool to
evaluate the consistency of mechanisms across diverse contexts, just like the aim of Chapter 3. Fourth, it helps identify and reduce bias in research instruments and measures. Using Chapter 3
as an example, the word recall test needs to be modified in different countries – the word list
needs to be somewhat redesigned to be culturally comprehensible and acceptable, and at the
same time comparable and translatable across cultures based on certain criteria such as
commonness and the average syllables of the words. Last but not least, it provides opportunities
for validating the measures developed under certain contexts. This is particularly important for
measures developed based on data science approaches such as machine learning algorithms. The current analyses in Chapter 4 and the planned future projects based on it align well
with the Geroscience approach. The emerging field of Geroscience continuously calls for the

102
identification of the underlying mechanisms of aging and multidisciplinary collaboration. One of
the key concepts of Geroscience is treating aging instead of specific chronic diseases or
conditions as the outcome of interest. While chronic diseases and conditions occur in some
individuals, aging occurs in all individuals, so it is a more universal mechanism. And because
aging is one of the main risk factors for all chronic diseases and conditions, it is also a more
fundamental mechanism. Hence, treating aging itself as the main risk factor for downstream
health outcomes and addressing the underlying biology of aging could potentially have a better
payoff than fighting separate diseases (Sierra, 2016; Barzilai et al., 2018). Another key concept
of Geroscience is that the translation of biological aging studies needs to be conducted under the
social context and informed by behavioral/social science models (Moffitt, 2020; Crimmins, 2020). Conclusions from biological and preclinical research based on cells, tissues, and
organisms in laboratory settings provide important evidence to our understanding of human
aging, but might not be directly applicable to human individuals who live and age in complex
real-world contexts. Hence, the leap “from lab to life” requires social science models based on
population-level data where biological and social factors interact and jointly affect population
health. Both of the two key concepts discussed above call for the development and testing of
biomarkers measuring the fundamental mechanisms of aging at the population level, which is
what Chapter 4 aims to do. As a potential future direction, it will be intriguing to see how the
biomarkers of underlying aging mechanisms perform in younger populations among whom
chronic diseases and conditions are not even present. Such research could possibly reveal how
early-life social, behavioral, and environmental exposures affect aging. For example, based on
the DNA methylation data from 600 children and adolescents aged 8-18 years who participated
in the Texas Twin Project, Raffington et al. (2021) find that children from more disadvantaged

103
families, who identify as Latinx (compared to Whites), who have higher BMI, and more tobacco
exposure exhibit faster DunedinPoAm-measured pace of aging. Aging is a gradual and highly complex process jointly influenced by a wide range of
intertwined biological and social-behavioral factors. This process is both universal and distinct –
It is a shared experience of all human beings across their life span, while the aging experience
can be highly contextual and variable across individuals. Hence, understanding and modeling
aging requires a comprehensive grasp of both social and biological information, as well as an
awareness of both common underlying mechanisms and potential disparities. This dissertation
attempts to add to our understanding of aging-related health changes and outcomes using high- quality population-level data that recently became available. To conclude, aging is a gradual process that is both universal and highly variable. We are
currently on a long but promising journey of understanding a set of highly complex mechanisms
in an increasingly precise and comprehensive way. Innovation in the collection, selection, and
construction of both biomarkers and social characteristics are both needed. The collaboration
between disciplines and across countries will also provide new opportunities to answer important
questions that were not even accessible in the past. All these efforts could provide evidence for
interventions aiming at increasing health span and improving the aging experience in general.

104
References
Ailshire, J., & Carr, D. (2021). Cross-national comparisons of social and economic contexts of
aging. The Journals of Gerontology Series B: Psychological Sciences and Social Sciences, 76(Supplement_1), S1-S4. Alley, D. E., Crimmins, E. M., Karlamangla, A., Hu, P., & Seeman, T. E. (2008). Inflammation
and rate of cognitive change in high-functioning older adults. The Journals of
Gerontology Series A: Biological Sciences and Medical Sciences, 63(1), 50-55. Anstey, K. J., Ashby-Mitchell, K., & Peters, R. (2017). Updating the evidence on the association
between serum cholesterol and risk of late-life dementia: review and meta-analysis. Journal of Alzheimer's Disease, 56(1), 215-228. Anstey, K. J., Cherbuin, N., Budge, M., & Young, J. (2011). Body mass index in midlife and
late-life as a risk factor for dementia: a meta-analysis of prospective studies. Obesity
Reviews, 12(5), e426-e437. Aversa, Z., Atkinson, E. J., Carmona, E. M., et al. (2023). Biomarkers of cellular senescence in
idiopathic pulmonary fibrosis. Respiratory Research, 24(1), 1-13. Barnard, M. E., Wang, X., Petrick, J. L., Zirpoli, G. R., Jones, D., Johnson, W. E., & Palmer, J. R. (2023). Psychosocial stressors and breast cancer gene expression in the Black
Women’s Health Study. Breast Cancer Research and Treatment, 1-4. Beck, D., de Lange, A. M. G., Pedersen, M. L., Alnæs, D., Maximov, I. I., Voldsbekk, I., ... &
Westlye, L. T. (2022). Cardiometabolic risk factors associated with brain age and
accelerated brain ageing. Human Brain Mapping, 43(2), 700-720. Belsky, D. W., Caspi, A., Corcoran, D. L., Sugden, K., Poulton, R., Arseneault, L., ... & Moffitt, T. E. (2022). DunedinPACE, a DNA methylation biomarker of the pace of
aging. Elife, 11, e73420. Belsky, D.W., Caspi, A., Houts, R., Cohen, H. J., Corcoran, D. L., Danese, A., ... & Moffitt, T. E. (2015). Quantification of biological aging in young adults. Proceedings of the National
Academy of Sciences, 112(30), E4104-E4110. Bird, C. E., Seeman, T., Escarce, J. J., Basurto-Dávila, R., Finch, B. K., Dubowitz, T., ... & Lurie, N. (2010). Neighbourhood socioeconomic status and biological ‘wear and tear’in a
nationally representative sample of US adults. Journal of Epidemiology & Community
Health, 64(10), 860-865. Bird, C. M., Webb, E. K., Cole, S. W., Tomas, C. W., Knight, J. M., Timmer-Murillo, S. C., ... &
Torres, L. (2024). Experiences of racial discrimination and adverse gene expression
among black individuals in a level 1 trauma center sample. Brain, Behavior, and
Immunity, 116, 229-236.

105
Brown, M. K., & Naidoo, N. (2010). The UPR and the anti-oxidant response: relevance to sleep
and sleep loss. Molecular neurobiology, 42, 103-113. Bussian, T. J., Aziz, A., Meyer, C. F., Swenson, B. L., van Deursen, J. M., & Baker, D. J. (2018). Clearance of senescent glial cells prevents tau-dependent pathology and cognitive
decline. Nature, 562(7728), 578-582. Butler, J., Kalogeropoulos, A. P., Georgiopoulou, V. V., Bibbins-Domingo, K., Najjar, S. S., Sutton-Tyrrell, K. C., ... & Psaty, B. M. (2011). Systolic blood pressure and incident
heart failure in the elderly. The Cardiovascular Health Study and the Health, Ageing and
Body Composition Study. Heart, 97(16), 1304-1311. Caliri, A. W., Tommasi, S., & Besaratinia, A. (2021). Relationships among smoking, oxidative
stress, inflammation, macromolecular damage, and cancer. Mutation Research/Reviews in
Mutation Research, 787, 108365. Campisi, J., Andersen, J. K., Kapahi, P., & Melov, S. (2011). Cellular senescence: A link
between cancer and age-related degenerative disease? Seminars in Cancer Biology, 21(6), 354-359. Campisi, J., & Robert, L. (2014). Cell senescence: role in aging and age-related
diseases. Aging, 39, 45-61. Campisi, J. (2013). Aging, Cellular Senescence, and Cancer. Annual Review of Physiology, 75, 685-705. Carroll, J. E., Cole, S. W., Seeman, T. E., Breen, E. C., Witarama, T., Arevalo, J. M., ... & Irwin, M. R. (2016). Partial sleep deprivation activates the DNA damage response (DDR) and
the senescence-associated secretory phenotype (SASP) in aged adult humans. Brain, Behavior, and Immunity, 51, 223-229. Carvalho, L. M. de, Wiers, C. E., Manza, P., et al. (2019). Effect of alcohol use disorder on
cellular aging. Psychopharmacology, 236(11), 3245-3256. Chanti-Ketterl, M., Stebbins, R. C., Obhi, H. K., Belsky, D. W., Plassman, B. L., & Yang, Y. C. (2022). Sex differences in the association between metabolic dysregulation and cognitive
aging: The Health and Retirement Study. The Journals of Gerontology: Series A, 77(9), 1827-1835. Chen, B., Jin, X., Guo, R., Chen, Z., Hou, X., Gao, F., ... & Zhang, Y. (2016). Metabolic
syndrome and cognitive performance among Chinese≥ 50 years: a cross-sectional study
with 3988 participants. Metabolic Syndrome and Related Disorders, 14(4), 222-227. Chinta, S. J., Woods, G., Demaria, M., Rane, A., Zou, Y., McQuade, A., ... & Andersen, J. K. (2018). Cellular senescence is induced by the environmental neurotoxin paraquat and
contributes to neuropathology linked to Parkinson’s disease. Cell Reports, 22(4), 930-940.

106
Chow, H. M., Shi, M., Cheng, A., Gao, Y., Chen, G., Song, X., ... & Herrup, K. (2019). Agerelated hyperinsulinemia leads to insulin resistance in neurons and cell-cycle-induced
senescence. Nature Neuroscience, 22(11), 1806-1819. Cole, S. W. (2014). Human social genomics. PLOS Genetics, 10(8), e1004601. Cole, S. W. (2019). The conserved transcriptional response to adversity. Current Opinion in
Behavioral Sciences, 28, 31-37. Cole, S. W., Levine, M. E., Arevalo, J. M., Ma, J., Weir, D. R., & Crimmins, E. M. (2015). Loneliness, eudaimonia, and the human conserved transcriptional response to adversity. Psychoneuroendocrinology, 62, 11-17. Cole, S. W., Shanahan, M. J., Gaydosh, L., & Harris, K. M. (2020). Population-based RNA
profiling in Add Health finds social disparities in inflammatory and antiviral gene
regulation to emerge by young adulthood. Proceedings of the National Academy of
Sciences, 117(9), 4601-4608. Coppé, J. P., Desprez, P. Y., Krtolica, A., & Campisi, J. (2010). The Senescence-Associated
Secretory Phenotype: The Dark Side of Tumor Suppression. Annual Review of Pathology, 5, 99-118. Crimmins, E., Faul, J., Kim, J. K., & Weir, D. (Published December, 2017). Documentation of
blood-based biomarkers in the 2014 Health and Retirement Study. Health and Retirement
Study. Retrieved from https://hrs.isr.umich.edu/publications/biblio/9387
Crimmins, E. M., Saito, Y., Kim, J. K., Zhang, Y. S., Sasson, I., & Hayward, M. D. (2018). Educational differences in the prevalence of dementia and life expectancy with dementia:
Changes from 2000 to 2010. The Journals of Gerontology: Series B, 73(suppl_1), S20- S28. Crimmins, E. M., Thyagarajan, B., Kim, J. K., Weir, D., & Faul, J. (2021). Quest for a summary
measure of biological age: The Health and Retirement Study. GeroScience, 43(1), 395- 408. Crimmins, E. M. (2015). Lifespan and healthspan: Past, present, and promise. The Gerontologist, 55(6), 901-911. Crimmins, E. M. (2020). Social hallmarks of aging: Suggestions for geroscience research. Ageing Research Reviews, 63, 101136. Crimmins, E. M., & Vasunilashorn, S. M. (2016). Biodemography: adding biological insight into
social, economic, and psychological models of population and individual health change
with age. In Handbook of Aging and the Social Sciences (pp. 55-75). Academic Press. Crimmins, E. M., Johnston, M., Hayward, M., & Seeman, T. (2003). Age differences in allostatic
load: An index of physiological dysregulation. Experimental Gerontology, 38(7), 731- 734.

107
Crimmins, E. M., Johnston, M., Hayward, M., & Seeman, T. (2003). Age differences in allostatic
load: an index of physiological dysregulation. Exp Gerontol, 38(7), 731–734. Crimmins, E. M., Kim, J. K., & Seeman, T. E. (2009). Poverty and biological risk: the earlier “aging” of the poor. The Journals of Gerontology Series A: Biological Sciences and
Medical Sciences, 64(2), 286-292. Crimmins, E. M., Thyagarajan, B., Kim, J. K., Weir, D., & Faul, J. (2021). Quest for a summary
measure of biological age: The health and retirement study. GeroScience, 43, 395-408. Crimmins, E. M., Faul, J., Kim, J. K., & Weir, D. (2015). Documentation of Biomarkers in the
2010 and 2012 Health and Retirement Study. Ann Arbor, MI: Survey Research Center, University of Michigan. Crimmins, E. M., Faul, J., Kim, J. K., & Weir, D. (2017). Documentation of blood-based
biomarkers in the 2014 Health and Retirement Study. Ann Arbor, MI: University of
Michigan, Survey Research Center. Crimmins, E., Faul, J., Kim, J. K., et al. (2020). Documentation of DBS blood-based biomarkers
in the 2016 Health and Retirement Study (Version 1). Ann Arbor, MI: Survey Research
Center, University of Michigan. Crimmins, E., Faul, J., Kim, J. K., Guyer, H., Langa, K., Ofstedal, M. B., & Weir, D. R. (2013). Documentation of Biomarkers in the 2006 and 2008 Health and Retirement Study. 2013. Crimmins, E., Guyer, H., Langa, K., Ofstedal, M. B., Wallace, R., & Weir, D. (2008). Documentation of physical measures, anthropometrics and blood pressure in the Health
and Retirement Study. HRS Documentation Report DR-011, 14(1-2), 47-59. Crimmins, E., Kim, J. K., & Vasunilashorn, S. (2010). Biodemography: New approaches to
understanding trends and differences in population health and mortality. Demography, 47(Suppl 1), S41-S64. Crook, Z., Booth, T., Cox, S. R., et al. (2018). Apolipoprotein E genotype does not moderate the
associations of depressive symptoms, neuroticism and allostatic load with cognitive
ability and cognitive aging in the Lothian Birth Cohort 1936. PLOS One, 13(2), e0192604. D’Amico, D., Amestoy, M. E., & Fiocco, A. J. (2020). The association between allostatic load
and cognitive function: A systematic and meta-analytic review. Psychoneuroendocrinology, 121, 104849. Everson, C. A., Henchen, C. J., Szabo, A., & Hogg, N. (2014). Cell injury and repair resulting
from sleep loss and sleep recovery in laboratory rats. Sleep, 37(12), 1929-1940. Faul, J. D., Kim, J. K., Levine, M. E., Thyagarajan, B., Weir, D. R., & Crimmins, E. M. (2023). Epigenetic-based age acceleration in a representative sample of older Americans:

108
Associations with aging-related morbidity and mortality. Proceedings of the National
Academy of Sciences, 120(9), e2215840120. Fielding, R. A., Atkinson, E. J., Aversa, Z., White, T. A., Heeren, A. A., Achenbach, S. J., ... &
LeBrasseur, N. K. (2022). Associations between biomarkers of cellular senescence and
physical function in humans: Observations from the lifestyle interventions for elders
(LIFE) study. Geroscience, 44(6), 2757-2770. Franceschi, C., & Campisi, J. (2014). Chronic inflammation (Inflammaging) and its potential
contribution to age-associated diseases. The Journals of Gerontology: Series A, 69(Suppl_1), S4-S9. Franklin, S. S., & Wong, N. D. (2016). Pulse pressure: how valuable as a diagnostic and
therapeutic tool? Journal of the American College of Cardiology, 67(4), 404-406. Fredrickson, B. L., Grewen, K. M., Algoe, S. B., Firestine, A. M., Arevalo, J. M., Ma, J., & Cole, S. W. (2015). Psychological well-being and the human conserved transcriptional
response to adversity. PLOS One, 10(3), e0121839. Geronimus, A. T., Hicken, M., Keene, D., & Bound, J. (2006). "Weathering" and age patterns of
allostatic load scores among blacks and whites in the United States. American Journal of
Public Health, 96(5), 826-833. Gong, P., Liang, S., Carlton, E. J., Jiang, Q., Wu, J., Wang, L., & Remais, J. V. (2012). Urbanisation and health in China. The Lancet, 379(9818), 843-852. Gorgoulis, V., Adams, P. D., Alimonti, A., Bennett, D. C., Bischof, O., Bishop, C., ... & Demaria, M. (2019). Cellular senescence: defining a path forward. Cell, 179(4), 813-827. Gross, A. L., Li, C., Briceño, E. M., Rentería, M. A., Jones, R. N., Langa, K. M., ... & Kobayashi, L. C. (2023). Harmonisation of later-life cognitive function across national contexts:
results from the Harmonized Cognitive Assessment Protocols. The Lancet Healthy
Longevity, 4(10), e573-e583. Haghani, A., Arpawong, T. E., Kim, J. K., Lewinger, J. P., Finch, C. E., & Crimmins, E. (2020). Female vulnerability to the effects of smoking on health outcomes in older people. PLOS
One, 15(6), e0234015. Haider, A. W., Larson, M. G., Franklin, S. S., & Levy, D. (2003). Systolic blood pressure, diastolic blood pressure, and pulse pressure as predictors of risk for congestive heart
failure in the Framingham Heart Study. Annals of Internal Medicine, 138(1), 10-16. Hartaigh, B. Ó., Allore, H. G., Trentalange, M., McAvay, G., Pilz, S., Dodson, J. A., & Gill, T. M. (2015). Elevations in time-varying resting heart rate predict subsequent all-cause
mortality in older adults. European Journal of Preventive Cardiology, 22(4), 527-534. Henrich, J., Heine, S. J., & Norenzayan, A. (2010). The weirdest people in the world?. Behavioral and Brain Sciences, 33(2-3), 61-83.

109
Higgins-Chen, A. T., Thrush, K. L., Wang, Y., Minteer, C. J., Kuo, P. L., Wang, M., ... & Levine, M. E. (2022). A computational solution for bolstering reliability of epigenetic clocks:
Implications for clinical trials and longitudinal tracking. Nature aging, 2(7), 644-661. Hobbs, K. A., Mann, F. D., Cole, S. W., & Krueger, R. F. (2021). Big Five personality and
CTRA gene expression: Lack of association in a midlife sample of US adults (MIDUS- Refresher). Personality and Individual Dif erences, 169, 109908. Horvath, S. (2013). DNA methylation age of human tissues and cell types. Genome Biology, 14(10), 1-20. Hou, X. (2008). Urban—rural disparity of overweight, hypertension, undiagnosed hypertension, and untreated hypertension in China. Asia Pacific Journal of Public Health, 20(2), 159- 169. Hu, P., Lee, J., Beaumaster, S., et al. (2020). Cognitive function and
cardiometabolic-inflammatory risk factors among older Indians and Americans. Journal
of the American Geriatrics Society, 68, S36-S44. Hughes, D., Judge, C., Murphy, R., Loughlin, E., Costello, M., Whiteley, W., ... & Canavan, M. (2020). Association of blood pressure lowering with incident dementia or cognitive
impairment: a systematic review and meta-analysis. JAMA, 323(19), 1934-1944. Ji, H., Kim, A., Ebinger, J. E., Niiranen, T. J., Claggett, B. L., Merz, C. N. B., & Cheng, S. (2020). Sex differences in blood pressure trajectories over the life course. JAMA
Cardiology, 5(3), 255-262. Johnson, S. C., Cavallaro, F. L., & Leon, D. A. (2017). A systematic review of allostatic load in
relation to socioeconomic position: Poor fidelity and major inconsistencies in biomarkers
employed. Social Science & Medicine, 192, 66-73. Juul Rasmussen, I., & Frikke-Schmidt, R. (2023). Modifiable cardiovascular risk factors and
genetics for targeted prevention of dementia. European Heart Journal. Karlamangla, A. S., Singer, B. H., McEwen, B. S., Rowe, J. W., & Seeman, T. E. (2002). Allostatic load as a predictor of functional decline: MacArthur studies of successful aging. Journal of Clinical Epidemiology, 55(7), 696-710. Karlamangla, A. S., Singer, B. H., & Seeman, T. E. (2006). Reduction in allostatic load in older
adults is associated with lower all-cause mortality risk: MacArthur Studies of Successful
Aging. Psychosomatic Medicine, 68(3), 500–507. Kim, J. K., Faul, J., Weir, D. R., & Crimmins, E. M. (2024). Dried blood spot based biomarkers
in the Health and Retirement Study: 2006 to 2016. American Journal of Human
Biology, 36(2), e23997.

110
Kim, J. K., & Crimmins, E. M. (2020). Blood pressure and mortality: joint effect of blood
pressure measures. Journal of Clinical Cardiology and Cardiovascular Therapeutics, 2(1). Kim, J. K., Ailshire, J., & Crimmins, E. (2019). Twenty year trends in cardiovascular risk among
men and women in the United States. Aging Clinical and Experimental Research, 31(1), 135-143. Kitayama, S., Akutsu, S., Uchida, Y., & Cole, S. W. (2016). Work, meaning, and gene regulation:
Findings from a Japanese information technology firm. Psychoneuroendocrinology, 72, 175-181. Kohrt, B. A., Worthman, C. M., Adhikari, R. P., Luitel, N. P., Arevalo, J. M., Ma, J., ... & Cole, S. W. (2016). Psychological resilience and the gene regulatory impact of posttraumatic
stress in Nepali child soldiers. Proceedings of the National Academy of Sciences, 113(29), 8156-8161. Kruman, I. I., Henderson, G. I., & Bergeson, S. E. (2012). DNA damage and neurotoxicity of
chronic alcohol abuse. Experimental Biology and Medicine (Maywood), 237(7), 740-747. Kusters, C. D. J., Klopack, E. T., Crimmins, E. M., Seeman, T. E., Cole, S., & Carroll, J. E. (2023). Short sleep and insomnia are associated with accelerated epigenetic age. Psychosomatic Medicine. Published online June 27, 2023. López-Otín, C., Blasco, M. A., Partridge, L., Serrano, M., & Kroemer, G. (2023). Hallmarks of
aging: An expanding universe. Cell, 186(2), 243-278. Lakka, H. M., Laaksonen, D. E., Lakka, T. A., Niskanen, L. K., Kumpusalo, E., Tuomilehto, J., & Salonen, J. T. (2002). The metabolic syndrome and total and cardiovascular disease
mortality in middle-aged men. JAMA, 288(21), 2709-2716. Lee, J., Phillips, D., Wilkens, J., & Gateway to Global Aging Data Team. (2021). Gateway to
Global Aging Data: Resources for cross-national comparisons of family, social
environment, and healthy aging. The Journals of Gerontology Series B: Psychological
Sciences and Social Sciences, 76(Supplement_1), S5-S16. Lee, S. H., Choi, I., Choi, E., Lee, M., Kwon, Y., Oh, B., & Cole, S. W. (2020). Psychological
well-being and gene expression in Korean adults: The role of age. Psychoneuroendocrinology, 120, 104785. Lei, X., Smith, J. P., Sun, X., & Zhao, Y. (2014). Gender differences in cognition in China and
reasons for change over time: evidence from CHARLS. The Journal of the Economics of
Ageing, 4, 46-55. Levine, M. E., Harrati, A., & Crimmins, E. M. (2018). Predictors and implications of accelerated
cognitive aging. Biodemography and Social Biology, 64(2), 83-101.

111
Levine, M. E., Lu, A. T., Quach, A., Chen, B. H., Assimes, T. L., Bandinelli, S., ... & Horvath, S. (2018). An epigenetic biomarker of aging for lifespan and healthspan. Aging (albany
NY), 10(4), 573. Levine, M. E. (2013). Modeling the rate of senescence: Can estimated biological age predict
mortality more accurately than chronological age? Journals of Gerontology Series A:
Biomedical Sciences and Medical Sciences, 68(6), 667-674. Levine, M. E., & Crimmins, E. M. (2014). Evidence of accelerated aging among African
Americans and its implications for mortality. Social Science & Medicine, 118, 27-32. Levine, M. E., Crimmins, E. M., Weir, D. R., & Cole, S. W. (2017). Contemporaneous social
environment and the architecture of late-life gene expression profiles. American Journal
of Epidemiology, 186(5), 503-509. Li, C., & Lumey, L. H. (2019). Impact of disease screening on awareness and management of
hypertension and diabetes between 2011 and 2015: results from the China health and
retirement longitudinal study. BMC Public Health, 19, 1-8. Liu, Z., Yang, H., Chen, S., Cai, J., & Huang, Z. (2019). The association between body mass
index, waist circumference, waist–hip ratio and cognitive disorder in older
adults. Journal of Public Health, 41(2), 305-312. Liu, Z., Kuo, P. L., Horvath, S., Crimmins, E., Ferrucci, L., & Levine, M. (2018). A new aging
measure captures morbidity and mortality risk across diverse subpopulations from
NHANES IV: A cohort study. PLOS Medicine, 15(12), e1002718. Lu, A. T., Quach, A., Wilson, J. G., Reiner, A. P., Aviv, A., Raj, K., ... & Horvath, S. (2019). DNA methylation GrimAge strongly predicts lifespan and healthspan. Aging (albany
NY), 11(2), 303. Mancia, G., De Backer, G., Dominiczak, A., Cifkova, R., Fagard, R., Germano, G., ... &
Zamorano, J. L. (2007). 2007 Guidelines for the management of arterial hypertension:
The Task Force for the Management of Arterial Hypertension of the European Society of
Hypertension (ESH) and of the European Society of Cardiology (ESC). European heart
journal, 28(12), 1462-1536. Mann, F. D., Krueger, R. F., Clouston, S., & Cole, S. (2020). Demographic correlates of
inflammatory and antiviral gene expression in the study of Midlife in the United States
(MIDUS). Biodemography and Social Biology, 66(3-4), 236-249. Merkin, S. S., Karlamangla, A., Roux, A. V. D., Shrager, S., & Seeman, T. E. (2014). Life
course socioeconomic status and longitudinal accumulation of allostatic load in adulthood:
Multi-Ethnic Study of Atherosclerosis. American Journal of Public Health, 104(4), e48- e55.

112
Moriya, J., Minamino, T., Tateno, K., Okada, S., Uemura, A., Shimizu, I., ... & Komuro, I. (2010). Inhibition of semaphorin as a novel strategy for therapeutic
angiogenesis. Circulation research, 106(2), 391-398. Mitchell, U. A., Ailshire, J. A., & Crimmins, E. M. (2019). Change in cardiometabolic risk
among blacks, whites, and Hispanics: findings from the Health and Retirement Study. The Journals of Gerontology: Series A, 74(2), 240-246. Mosley, W. J., Greenland, P., Garside, D. B., & Lloyd-Jones, D. M. (2007). Predictive utility of
pulse pressure and other blood pressure measures for cardiovascular outcomes. Hypertension, 49(6), 1256-1264. Muñoz-Espín, D., & Serrano, M. (2014). Cellular senescence: from physiology to pathology. Nature Reviews Molecular Cell Biology, 15(7), 482-496. Musi, N., Valentine, J. M., Sickora, K. R., Baeuerle, E., Thompson, C. S., Shen, Q., & Orr, M. E. (2018). Tau protein aggregation is associated with cellular senescence in the brain. Aging
Cell, 17(6), e12840. Narbutas, J., Van Egroo, M., Chylinski, D., et al. (2019). Cognitive efficiency in late midlife is
linked to lifestyle characteristics and allostatic load. Aging (Albany NY), 11(17), 7169. Ng, M., & Hazrati, L. N. (2022). Evidence of sex differences in cellular senescence. Neurobiology of Aging, 120, 88-104. Nyunoya, T., Monick, M. M., Klingelhutz, A., Yarovinsky, T. O., Cagley, J. R., & Hunninghake, G. W. (2006). Cigarette Smoke Induces Cellular Senescence. American Journal of
Respiratory Cell and Molecular Biology, 35(6), 681-688. O’Keeffe, L. M., Kuh, D., Fraser, A., Howe, L. D., Lawlor, D. A., & Hardy, R. (2019). Age at
period cessation and trajectories of cardiovascular risk factors across mid and later life: a
prospective cohort study. BioRxiv, 639518. Ogrodnik, M., Zhu, Y. I., Langhi, L. G., Tchkonia, T., Krüger, P., Fielder, E., ... & Jurk, D. (2019). Obesity-induced cellular senescence drives anxiety and impairs
neurogenesis. Cell metabolism, 29(5), 1061-1077. Oi, K., & Haas, S. (2019). Cardiometabolic risk and cognitive decline: The role of
socioeconomic status in childhood and adulthood. Journal of Health and Social Behavior, 60(3), 326-343. Oksuzyan, A., Crimmins, E., Saito, Y., O’Rand, A., Vaupel, J. W., & Christensen, K. (2010). Cross-national comparison of sex differences in health and mortality in Denmark, Japan
and the US. European Journal of Epidemiology, 25, 471-480. Olivieri, F., Marchegiani, F., Matacchione, G., et al. Sex/gender-related differences in
inflammaging. Mechanisms of Ageing and Development, 211, 111792.

113
Ortiz, K., Garcia, M. A., Briceño, E., et al. (2020). Glycosylated hemoglobin level, race/ethnicity, and cognition in midlife and early old age. Research in Human Development, 17(1), 20- 40. Pastor-Barriuso, R., Banegas, J. R., Damin, J., Appel, L. J., & Guallar, E. (2003). Systolic blood
pressure, diastolic blood pressure, and pulse pressure: an evaluation of their joint effect
on mortality. Annals of Internal Medicine, 139(9), 731-739. Pegueroles, J., Jiménez, A., Vilaplana, E., et al. (2018). Obesity and Alzheimer's disease, does
the obesity paradox really exist? A magnetic resonance imaging study. Oncotarget, 9(78), 34691. Peters, R., Xu, Y., Antikainen, R., Beckett, N., Gussekloo, J., Jagger, C., ... & Anstey, K. J. (2021). Evaluation of high cholesterol and risk of dementia and cognitive decline in older
adults using individual patient meta-analysis. Dementia and Geriatric Cognitive
Disorders, 50(4), 318-325. Pole, A., Dimri, M., & Dimri, G. (2016). Oxidative stress, cellular senescence and ageing. AIMS
Molecular Science, 3(3). Prince, M., Wimo, A., Guerchet, M., Ali, G. C., Wu, Y. T., & Prina, M. (2015). World
Alzheimer Report 2015. The Global Impact of Dementia: An analysis of prevalence, incidence, cost and trends (Doctoral dissertation, Alzheimer's Disease International). Program on Global Aging, Health & Policy, University of Southern California. (Accessed Sep 30, 2023). Graphs and Table. Gateway to Global Aging Harmonized Data. https://g2aging.org/data-explorer
Prospective Studies Collaboration. (2002). Age-specific relevance of usual blood pressure to
vascular mortality: a meta-analysis of individual data for one million adults in 61
prospective studies. The Lancet, 360(9349), 1903-1913. Ravona-Springer, R., Moshier, E., Schmeidler, J., Godbold, J., Akrivos, J., Rapp, M., ... & Beeri, M. S. (2012). Changes in glycemic control are associated with changes in cognition in
non-diabetic elderly. Journal of Alzheimer's Disease, 30(2), 299-309. Ridker, P. M. (2003). C-reactive protein. Circulation, 108(12), e81-e85. Riessland, M., Kolisnyk, B., Kim, T. W., Cheng, J., Ni, J., Pearson, J. A., ... Wang, W. (2019). Loss of SATB1 induces p21-dependent cellular senescence in post-mitotic dopaminergic
neurons. Cell Stem Cell, 25(4), 514-530. Rodríguez-Fernández, J. M., Danies, E., Martínez-Ortega, J., & Chen, W. C. (2017). Cognitive
decline, body mass index, and waist circumference in community-dwelling elderly
participants: Results from a nationally representative sample. Journal of Geriatric
Psychiatry and Neurology, 30(2), 67-76.

114
Ross, K. M., Cole, S. W., Carroll, J. E., & Schetter, C. D. (2019). Elevated pro-inflammatory
gene expression in the third trimester of pregnancy in mothers who experienced stressful
life events. Brain, Behavior, and Immunity, 76, 97-103. Santos, C. Y., Snyder, P. J., Wu, W. C., Zhang, M., Echeverria, A., & Alber, J. (2017). Pathophysiologic relationship between Alzheimer's disease, cerebrovascular disease, and
cardiovascular risk: A review and synthesis. Alzheimer's & Dementia: Diagnosis, Assessment & Disease Monitoring, 7, 69-87. Schafer, M. J., Zhang, X., Kumar, A., Atkinson, E. J., Zhu, Y. I., Jachim, S., ... & LeBrasseur, N. K. (2020). The senescence-associated secretome as an indicator of age and medical risk. JCI Insight, 5(12). Schmitz, N., Deschênes, S. S., Burns, R. J., et al. (2018). Cardiometabolic dysregulation and
cognitive decline: Potential role of depressive symptoms. The British Journal of
Psychiatry, 212(2), 96-102. Seccareccia, F., Pannozzo, F., Dima, F., et al.; Malattie Cardiovascolari Aterosclerotiche Istituto
Superiore di Sanita Project. (2001). Heart rate as a predictor of mortality: The MATISS
project. American Journal of Public Health, 91(8), 1258–1263. Seeman, T. E., Crimmins, E., Huang, M. H., Singer, B., Bucur, A., Gruenewald, T., ... & Reuben, D. B. (2004). Cumulative biological risk and socio-economic differences in mortality:
MacArthur studies of successful aging. Social Science & Medicine, 58(10), 1985-1997. Seeman, T. E., McEwen, B. S., Rowe, J. W., & Singer, B. H. (2001). Allostatic load as a marker
of cumulative biological risk: MacArthur studies of successful aging. Proceedings of the
National Academy of Sciences, 98(8), 4770-4775. Seeman, T., Merkin, S. S., Crimmins, E., Koretz, B., Charette, S., & Karlamangla, A. (2008). Education, income and ethnic differences in cumulative biological risk profiles in a
national sample of US adults: NHANES III (1988–1994). Soc Sci Med, 66(1), 72–87. Serrano, M., & Muñoz-Espin, D. (2021). Cellular Senescence in Disease. Academic Press. Sha, T., Cheng, W., & Yan, Y. (2018). Prospective associations between pulse pressure and
cognitive performance in Chinese middle-aged and older population across a 5-year study
period. Alzheimer's Research & Therapy, 10, 1-1. St. Sauver, J. L., Weston, S. A., Atkinson, E. J., Mc Gree, M. E., Mielke, M. M., White, T. A., ... & LeBrasseur, N. K. (2023). Biomarkers of cellular senescence and risk of death in
humans. Aging Cell, 22(12), e14006. Sun, J. K. L., Wu, D., Wong, G. C. N., Lau, T. M., Yang, M., Hart, R. P., ... & Chow, H. M. (2023). Chronic alcohol metabolism results in DNA repair infidelity and cell
cycle-induced senescence in neurons. Aging Cell, 22(2), e13772.

115
Svensson, T., Sawada, N., Mimura, M., Nozaki, S., Shikimoto, R., & Tsugane, S. (2019). The
association between midlife serum high-density lipoprotein and mild cognitive
impairment and dementia after 19 years of follow-up. Translational Psychiatry, 9(1), 26. Tchkonia, T., Morbeck, D. E., von Zglinicki, T., et al. (2010). Fat tissue, aging, and cellular
senescence. Aging Cell, 9(5), 667-684. Toms, R., Bonney, A., Mayne, D. J., Feng, X., & Walsan, R. (2019). Geographic and area-level
socioeconomic variation in cardiometabolic risk factor distribution: A systematic review
of the literature. International Journal of Health Geographics, 18(1), 1-2. Waki, T., Tanaka-Mizuno, S., Takashima, N., Takechi, H., Hayakawa, T., Miura, K., ... & Dodge, H. H. (2020). Waist circumference and domain-specific cognitive function among non- demented Japanese older adults stratified by sex: results from the Takashima cognition
study. Journal of Alzheimer's Disease, 73(3), 887-896. Waldstein, S. R., Rice, S. C., Thayer, J. F., Najjar, S. S., Scuteri, A., & Zonderman, A. B. (2008). Pulse pressure and pulse wave velocity are related to cognitive decline in the Baltimore
Longitudinal Study of Aging. Hypertension, 51(1), 99-104. Wang, Y. Q., Jia, R. X., Liang, J. H., et al. (2019). Dementia in China (2015–2050) estimated
using the 1% population sampling survey in 2015. Geriatrics & Gerontology
International, 19(11), 1096-1100. Wei, J., Yin, X., Liu, Q., Tan, L., & Jia, C. (2018). Association between hypertension and
cognitive function: A cross-sectional study in people over 45 years old in China. The
Journal of Clinical Hypertension, 20(11), 1575-1583. Wu, Q., Ailshire, J. A., Kim, J. K., & Crimmins, E. M. (2021). Cardiometabolic risk trajectory
among older Americans: Findings from the Health and Retirement Study. The Journals of
Gerontology: Series A, 76(12), 2265-2274. Yang, Y., & Kozloski, M. (2011). Sex differences in age trajectories of physiological
dysregulation: Inflammation, metabolic syndrome, and allostatic load. The Journals of
Gerontology Series A: Biological Sciences and Medical Sciences, 66A(5), 493-500. Yue, Z., Nie, L., Zhao, P., Ji, N., Liao, G., & Wang, Q. (2022). Senescence-associated secretory
phenotype and its impact on oral immune homeostasis. Frontiers in Immunology, 13, 1019313. Zhang, P., Kishimoto, Y., Grammatikakis, I., Gottimukkala, K., Cutler, R. G., Zhang, S., ... &
Mattson, M. P. (2019). Senolytic therapy alleviates Aβ-associated oligodendrocyte
progenitor cell senescence and cognitive deficits in an Alzheimer’s disease model. Nature
Neuroscience, 22(5), 719-728. Zhang, Y., Guan, Y., Shi, Z., Yue, W., Liu, S., Liu, S., ... & Ji, Y. (2019). Sex differences in the
prevalence of and risk factors for cognitive impairment no dementia among the elderly in

116
a rural area of Northern China: a population-based cross-sectional study. Neuroepidemiology, 52(1-2), 25-31. Zhang, Y. S., & Crimmins, E. M. (2019). Urban–rural differentials in age-related biological risk
among middle-aged and older Chinese. International Journal of Public Health, 64, 831- 839. Zhao, E., & Crimmins, E. M. (2022). Mortality and morbidity in ageing men: Biology, Lifestyle
and Environment. Reviews in Endocrine and Metabolic Disorders, 23(6), 1285-1304. Zhao, Y., Crimmins, E. M., Hu, P., Shen, Y., Smith, J. P., Strauss, J., ... & Zhang, Y. (2016). Prevalence, diagnosis, and management of diabetes mellitus among older Chinese: results
from the China Health and Retirement Longitudinal Study. International Journal of
Public Health, 61(3), 347-356. Zhao, Y., Hu, Y., Smith, J. P., Strauss, J., & Yang, G. (2014). Cohort profile: The China Health
and Retirement Longitudinal Study (CHARLS). International Journal of Epidemiology, 43(1), 61-68. Zhao, Y., Strauss, J., Yang, G., Giles, J., Hu, P., Hu, Y., ... & Wang, Y. (2013). China health and
retirement longitudinal study–2011–2012 national baseline users’ guide. Beijing:
National School of Development, Peking University, 2, 1-56. Zhou, Y., Huang, J., Zhang, P., Tong, J., Fan, F., Gou, M., ... & Hong, L. E. (2021). Allostatic
load effects on cortical and cognitive deficits in essentially normotensive, normoweight
patients with schizophrenia. Schizophrenia Bulletin, 47(4), 1048-1057. Zissimopoulos, J. M., Tysinger, B. C., St. Clair, P. A., & Crimmins, E. M. (2018). The impact of
changes in population health and mortality on future prevalence of Alzheimer’s disease
and other dementias in the United States. The Journals of Gerontology: Series B, 73(suppl_1), S38-S47. Zou, H., Li, Z., Tian, X., & Ren, Y. (2022). The top 5 causes of death in China from 2000 to
2017. Scientific Reports, 12(1), 8119.

Abstract (if available)

Abstract Incorporating biomarkers into social science aging research helps understand the risk for age-related health changes and explains how social and environmental factors get translated to health. However, the existing literature can benefit from longitudinal and cross-country evidence and new biomarkers to measure more fundamental mechanisms of aging. Based on the comparable nationally representative longitudinal surveys with biomarkers on older adults in the US and China, the current dissertation contributes to the literature by depicting the with-in-person trajectory of cardiometabolic risk (CMR) among older Americans, comparing the cross-sectional and longitudinal associations between CMR and cognitive function between the US and China, and testing a group of RNA-based cellular senescence measures at the population level, showing their associations with social-behavioral factors and age-related health outcomes. The results show that the overall CMR can be fairly stable during 8 years of aging among older Americans. Medication use contributes to the decrease in blood pressure, heart rate, and total cholesterol. However, glycosylated hemoglobin increases, especially faster among non-Hispanic Blacks, indicating a need for better blood sugar management; Waist circumference also increases, especially faster among people with higher education levels, indicating a need for a healthier lifestyle. Both a higher CMR level and a CMR increase are associated with worse cognitive function and faster cognitive decline among older Americans but not Older Chinese. In China, those with the highest socioeconomic status (SES) seem to have both higher CMR and better cognitive health. Among them, having a higher CMR is not as harmful as it is for their lower-SES counterparts in terms of cognitive health, but a rapid rise in CMR is additionally harmful. The results also show that among older Americans, older ages, female sex, being non-Hispanic Black, and being obese are linked to a higher level of cellular senescence. Cellular senescence correlates highly with epigenetic aging. It is also associated with health outcomes including physiological dysregulation and multimorbidity, with and without adjustment for epigenetic aging, and these associations are mainly driven by macromolecular damage and senescence-associated secretory phenotype. So, the RNA-based cellular senescence measure used in this dissertation measures aging and adds to our understanding of physiological dysregulation and age-related diseases in a useful way.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

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

PDF

Self-perceptions of Aging in the Context of Neighborhood and Their Interplay in Late-life Cognitive Health

PDF

Social determinants of physiological health and mortality in China

PDF

Identifying and measuring cell-intrinsic and cell-extrinsic factors influencing aging

PDF

Internet communication use, psychological functioning and social connectedness at older ages

PDF

Statistical algorithms for examining gene and environmental influences on human aging

PDF

Cognitive health and self-rated memory in later life: linkages to race/ethnicity, multimorbidity, and survival

PDF

Estimating survival in the face of pain: evidence from the health and retirement study

PDF

Is stress exposure enough? Race/ethnic differences in the exposure and appraisal of chronic stressors among older adults

PDF

Health differences between the elderly in Japan and the United States

PDF

Age integration in late life: sociodemographic & psychosocial correlates of intergenerational-only, peer-only, and age-integrated social networks

PDF

Essays on health and aging with focus on the spillover of human capital

PDF

Marital quality, gender, and biomarkers of disease risk in the MIDUS cohort

PDF

Longitudinal assessment of neural stem-cell aging

PDF

Examining the longitudinal influence of the physical and social environments on social isolation and cognitive health: contextualizing the role of technology

PDF

Investigating brain aging and neurodegenerative diseases through omics data

PDF

Lead exposure and health outcomes in older adults

PDF

Characterization of senescent cell heterogeneity using cell culture models

PDF

Three essays on health & aging

PDF

Residential care in Los Angeles: policy and planning for an aging population

Asset Metadata

Creator Wu, Qiao (author)

Core Title Biomarkers of age-related health changes: associations with health outcomes and disparities

School Leonard Davis School of Gerontology

Degree Doctor of Philosophy

Degree Program Gerontology

Degree Conferral Date 2024-05

Publication Date 04/30/2024

Defense Date 02/22/2024

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

Tag aging,biodemography,biomarker,cardiometabolic,cellular senescence,cognition,OAI-PMH Harvest,population health

Format theses (aat)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Crimmins, Eileen (committee chair), Ailshire, Jennifer (committee member), Kim, Jung Ki (committee member)

Creator Email qiaowu@usc.edu,qiaowu0831@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC113911908

Unique identifier UC113911908

Identifier etd-WuQiao-12867.pdf (filename)

Legacy Identifier etd-WuQiao-12867

Document Type Dissertation

Format theses (aat)

Rights Wu, Qiao

Internet Media Type application/pdf

Type texts

Source 20240503-usctheses-batch-1146 (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.

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