Abstract Title of Dissertation COVID-19 VACCINE HESITANCY AND UPTAKE IN THE UNITED STATES CONSIDERED THROUGH THE LENS OF HEALTH BEHAVIOR THEORY Lauren Emily Kauffman, M.P.H. Doctor of Philosophy, 2024 Dissertation directed by Dr. Quynh Nguyen Associate Professor Department of Epidemiology and Biostatistics University of Maryland, College Park Given the low COVID-19 vaccine uptake rates in many areas of the United States1 despite their demonstrated safety and effectiveness,2 COVID-19 vaccine hesitancy and vaccination barriers continue to be critical areas of research in epidemiology and behavioral health science. This series of studies focuses on COVID-19 vaccine hesitancy and vaccination barriers, as they relate to vaccination intention and vaccine uptake, considered in the context of established health behavior theories. The first study is a systematic review of existing research on COVID-19 vaccine hesitancy using one or more health behavior theories as key components of the design or analysis. This study examined the types of theories that are most often used, how they are used, and where research gaps exist. The remaining two studies use data from the U.S. COVID-19 Trends and Impact Survey,3 a national cross-sectional survey. The second study investigates the association between recent feelings of anxiety or depression and vaccination intention, as well as between these feelings and identifying with specific vaccine hesitancy reasons. The third study examines vaccine hesitancy and barriers among those with chronic illness or disease, a particularly vulnerable population. Factor analysis was conducted using constructs from the Theory of Planned Behavior as a framework, and the results were used in a regression model to investigate the association between these underlying factors and vaccination intention. This research demonstrated the usefulness of the Theory of Planned Behavior, the Health Belief Model, and the 3 Cs Model in existing and future COVID-19 vaccine hesitancy research, as well as identified Protection Motivation Theory as a promising area for future research. Additionally, psychological states were demonstrated to be significantly associated with vaccine hesitancy, adjusting for demographic, socioeconomic, and time factors. Lastly, the Theory of Planned Behavior was found to be applicable to those unvaccinated and with chronic illness, as the construct factor scores developed were significantly associated with vaccine hesitancy (adjusting for the presence of specific chronic conditions and demographic, socioeconomic, and time factors). These associations were also consistently demonstrated in subgroup analyses of participants with specific chronic conditions. COVID-19 VACCINE HESITANCY AND UPTAKE IN THE UNITED STATES CONSIDERED THROUGH THE LENS OF HEALTH BEHAVIOR THEORY BY LAUREN EMILY KAUFFMAN, M.P.H. DISSERTATION SUBMITTED TO THE FACULTY OF THE GRADUATE SCHOOL OF THE UNIVERSITY OF MARYLAND, COLLEGE PARK, IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY 2024 ADVISORY COMMITTEE: ASSOCIATE PROFESSOR QUYNH NGUYEN, CHAIR PROFESSOR JAMES CAMPBELL (UMD SCHOOL OF MEDICINE) ASSOCIATE PROFESSOR TYPHANYE DYER ASSOCIATE CLINICAL PROFESSOR JENNIFER HODGSON ASSISTANT PROFESSOR TIANZHOU (CHARLES) MA PROFESSOR SANDRA QUINN, DEAN’S REPRESENTATIVE © Copyright by Lauren Emily Kauffman 2024 ii Acknowledgments This research is based on survey results from Carnegie Mellon University’s Delphi Group. Thank you to my PhD advisor and committee chair, Dr. Quynh Nguyen. Her mentorship, expertise, and encouragement have been invaluable to my academic success. Thank you to my long-time mentor and committee member, Dr. Typhanye Dyer. Her mentorship in teaching and research over these past six years has been instrumental to my success throughout my graduate studies and in my dissertation. Thank you to the members of my committee, Dr. Charles Ma, Dr. Sandra Quinn, Dr. James Campbell, and Dr. Jennifer Hodgson. I am deeply grateful for the time, effort, and care they have taken as members of my dissertation committee. Their expertise and guidance have been crucial in developing my research. Thank you to Dr. Rodman Turpin, who provided guidance and expertise in developing the systematic review manuscript. Thank you to the many members of the faculty and staff, my instructors, my peers, and my students in the School of Public Health. Their contagious enthusiasm for public health has been a steady, positive influence throughout my graduate school journey. Thank you to my friend of over 20 years, Lisa, who has always provided invaluable moral support and friendship. Thank you to my dog, Chloe, who is a warm, bright light in every day. Finally, this dissertation is dedicated to my parents. This would not have been possible without their unwavering belief in me as a student, as a researcher, and as a person, and their boundless practical and emotional support have meant everything to me. Thank you, all. iii Table of Contents Chapter 1: Introduction ............................................................................................................... 1 Background and Rationale .......................................................................................................... 1 Objectives and Research Questions ............................................................................................ 1 Theoretical and Conceptual Framework and Hypotheses ........................................................... 3 Innovation and Significance ........................................................................................................ 4 Chapter 2: Methods ...................................................................................................................... 6 Systematic Review ...................................................................................................................... 6 Study Design ............................................................................................................................... 6 Statistical approaches to test hypotheses ..................................................................................... 6 Model Specification .................................................................................................................... 7 Assessment of Confounding ....................................................................................................... 8 Assessment of Model Assumptions ............................................................................................ 8 Potential Biases and Limitations ................................................................................................. 9 Chapter 3: Manuscript 1 | Systematic Review of Health Behavior Theory Use in U.S. COVID-19 Vaccination Hesitancy-Focused Studies ................................................................ 12 Background ............................................................................................................................... 12 Specific Aims ............................................................................................................................ 21 Theoretical/Conceptual Framework .......................................................................................... 22 Methods ..................................................................................................................................... 23 iv Criteria for Selection and Preliminary Search ....................................................................... 23 Full Search ............................................................................................................................. 26 Results ....................................................................................................................................... 27 Bias Assessment .................................................................................................................... 32 Discussion ................................................................................................................................. 33 Study Strengths and Limitations ............................................................................................... 35 Human Subjects and Ethical Considerations ............................................................................ 36 Chapter 4: Manuscript 2 | Investigating the Effect of Frequency of Recent Anxious and Depressed Feelings on COVID-19 Vaccination Intention and Vaccine Hesitancy among Unvaccinated Individuals in the United States ........................................................................ 37 Background ................................................................................................................................... 37 Specific Aims ............................................................................................................................ 41 Theoretical/Conceptual Framework .......................................................................................... 43 Methods ..................................................................................................................................... 46 Data Source ............................................................................................................................ 47 Participants and Criteria for Selection ................................................................................... 49 Variables of Interest and Potential Measurement Issues ....................................................... 50 Power analysis ....................................................................................................................... 58 Statistical Analyses ................................................................................................................ 59 Results ....................................................................................................................................... 62 v Discussion ................................................................................................................................. 68 Human Subjects and Ethical Considerations ............................................................................ 72 Chapter 5: Manuscript 3 | Investigating COVID-19 Vaccine Hesitancy among the Unvaccinated Chronic Illness Population in the United States through the Lens of the Theory of Planned Behavior ...................................................................................................... 74 Background ............................................................................................................................... 74 Specific Aims ............................................................................................................................ 77 Theoretical/Conceptual Framework .......................................................................................... 78 Methods ..................................................................................................................................... 81 Data Source ............................................................................................................................ 81 Overall Study Design............................................................................................................. 82 Participants and Criteria for Selection ................................................................................... 82 Variables of Interest and Potential Measurement Issues ....................................................... 83 Statistical Analyses ................................................................................................................ 88 Results ....................................................................................................................................... 95 Discussion ................................................................................................................................. 98 Human Subjects and Ethical Considerations .......................................................................... 102 Chapter 6: Conclusions and Public Health Significance ....................................................... 103 Appendix .................................................................................................................................... 107 Supplemental Materials for Chapter 3 ................................................................................. 107 vi Supplemental Materials for Chapter 4 ................................................................................. 125 Supplemental Materials for Chapter 5 ................................................................................. 130 References .................................................................................................................................. 136 1 Chapter 1: Introduction Background and Rationale The World Health Organization identified vaccine hesitancy as one of ten major threats to global health back in 2019. At this time, the WHO indicated significant concern about vaccine hesitancy regarding its role in the reversal of progress in addressing vaccine-preventable diseases (such as measles) and as a potential barrier in the prospect of using vaccines to help eliminate diseases such as cervical cancer and wild poliovirus4. The COVID-19 pandemic has since brought the issue of vaccine hesitancy even more to the forefront of awareness of public health researchers and the general population. The issue is further compounded by issues of vaccine access and availability. Given the low COVID-19 vaccine uptake rates in many areas of the United States.1 COVID-19 vaccine hesitancy and barriers to vaccination have been and continue to be critical areas of research in epidemiology and behavioral science. Research into COVID-19 vaccine uptake and hesitancy has been, and will continue to be, inextricably linked to health behavior theory. Well-developed and extensively tested theories can provide solid frameworks for understanding and predicting health behavior in a variety of circumstances, and establishing the serviceability of a particular theory as a framework is a critical phase in COVID-19 vaccine hesitancy research. Objectives and Research Questions The three manuscripts presented here seek to address the topic of COVID-19 vaccine hesitancy and uptake through the lens of health behavior theory. The first manuscript is a systematic review of existing research articles on COVID-19 vaccine hesitancy that explicitly use health behavior theories and models in their design or analysis. The goal was to assess the 2 current status of health behavior theory in COVID-19 vaccine hesitancy research, including which theories or models are used most often, how effectively they are used, how integral they are to the research design or analysis, and where there are potential gaps in understanding and using health behavior theories in this context. The second manuscript is an analysis of national COVID-19 data from the United States, The Delphi Group at Carnegie Mellon University U.S. COVID-19 Trends and Impact Survey, in partnership with Facebook (hereafter referred to as the U.S. COVID-19 Trends and Impact Survey or U.S. CTIS).3 This analysis focuses on individual reasons for vaccine hesitancy and their relationship to unvaccinated participants’ emotional states of anxiety and depression. Existing health behavior research and multiple health behavior theories suggested that a person’s psychological state has an impact on how public health messaging is received. This research sought to demonstrate the effect of recent emotional and psychological states of anxiety and depression (not restricted to clinically apparent anxiety and depression disorders) on specific reasons for vaccine hesitancy as well as overall intention to get vaccinated. The third manuscript assesses the serviceability of a particular health behavior theory – the Theory of Planned Behavior – as it pertains to vaccination hesitancy and barriers among individuals who are particularly vulnerable to severe COVID-19 outcomes due to having one or more chronic illnesses such as cancer, heart disease, high blood pressure, asthma, chronic lung disease, and diabetes. This research involved conducting factor analysis, with factors based on the three main constructs of the Theory of Planned Behavior and assessing the significance of the calculated factor scores in predicting vaccination behavior and intention to get vaccinated. The goal of this research was to determine the suitability of the Theory of Planned Behavior for use 3 in designing more effective public health messaging and interventions for vulnerable chronic illness populations. Theoretical and Conceptual Framework and Hypotheses The systematic review manuscript focuses on many individual and inter-personal health behavior theories and models, including the Health Belief Model, the Theory of Reasoned Action, the Theory of Planned Behavior, the Reasoned Action Approach, the Integrated Behavioral Model, the Transtheoretical Model, Protection Motivation Theory, the 3 Cs Model for vaccine hesitancy, and the 5Cs Model for COVID-19 vaccine hesitancy. The hypotheses of the first manuscript were that, in the existing scientific literature, these theories and models would vary in frequency of use, level of influence on the study design and/or analysis, and usefulness for specific study populations. Manuscript 2 discusses more broadly how various items in the U.S. CTIS used in the manuscript analysis reflect common constructs from many health behavior theories and models, including the Theory of Reasoned Action, the Theory of Planned Behavior, the Integrated Behavioral Model, and the 3 Cs Model. The hypotheses of this manuscript were that, among unvaccinated participants, 1) recent feelings of anxiety, depression, and COVID-19-related worry were significantly associated with overall vaccine hesitancy, and 2) recent feelings of anxiety, depression, and COVID-19-related worry were significantly associated with specific vaccine hesitancy reasons (e.g., concern about side effects, not trusting the government, concern about safety of the vaccine with particular health issues, etc.). Manuscript 3 focuses specifically on the application of the Theory of Planned Behavior to the population of unvaccinated individuals with chronic health issues. Analyses included conducting factor analysis, with specific survey items restricted to loading on specific factors 4 based on the Theory of Planned Behavior, and logistic regression using these factor scores. The hypotheses of this manuscript were that 1) factor analysis would generate significant factor loadings based on the constructs of the Theory of Planned Behavior, and 2) these construct factor scores would be significant in a logistic regression model predicting vaccine hesitancy, controlling for chronic health issues. Innovation and Significance The overall goal of this research is to be better able to tailor COVID-19 vaccine U.S. public health messaging based on key characteristics of the unvaccinated population to improve the overall effectiveness of the messaging. This includes not only insight into specific demographics or specific vaccine hesitancy reasons, but insight into the emotional states of those individuals and the effect of the presence of chronic health conditions. This also includes insight into what particular health behavior theories may best be used in particular populations to develop interventions. Increasing the effectiveness of COVID-19 vaccination communications, messages, and interventions has the immediate goal of increasing vaccine uptake among eligible individuals and, ultimately, reducing the burden of COVID-19 morbidity and mortality in the United States. Additionally, insight into these complexities of vaccine hesitancy and confidence throughout the early stages of the COVID-19 health crisis (with increasing levels of vaccine availability over time to different groups of people) can provide valuable information for the rollout of future novel vaccines. The early steps, research, and vaccination roll-out strategies for COVID-19 were inherently based on information the scientific and policy communities had from their understanding of previous infectious disease outbreaks and previous vaccine rollouts. The research presented here will provide another valuable source of information that public health 5 professionals can apply not just to current COVID-19 vaccine hesitancy, but also to potential future vaccines. 6 Chapter 2: Methods Systematic Review The first manuscript is a systematic review of the use of individual- and interpersonal- level health behavior theories and models in COVID-19 vaccine hesitancy, confidence, and uptake research. The potential biases involved are publication biases, as not all relevant studies may have made it to publication, and the biases within the studies themselves. Bias assessment was conducted on each article included in the review, and a second reviewer participated in assessing the articles ultimately chosen for review. The remaining portion of this chapter will be describing manuscripts 2 and 3. Study Design The remaining two manuscripts used cross-sectional data from the U.S. COVID-19 Trends and Impact Survey.3 Manuscript 2 used data from February to August of 2021, and manuscript 3 uses data from May 2021 to February 2022. These data were chosen due to the changing nature of the survey – the combinations of variables needed for each respective set of research questions were specific to those time periods. Weights and strata were incorporated into the analyses for both manuscripts. Statistical approaches to test hypotheses The data analyses for manuscripts 2 and 3 were generated using SAS® software, Version 9.4 of the SAS System for Windows (© 2016). Rao-Scott Chi-Square tests were used for bivariate analyses, and hierarchical logistic regression was used for modeling the outcomes with increasing inclusion of groups of other relevant covariates and potential confounders. For manuscript 3, confirmatory factor analysis was conducted to generate factor scores based on 7 constructs from the Theory of Planned Behavior. These scores were then used in the logistic regression models. The U.S. CTIS used stratified random sampling, and the data included weights and strata for use in statistical analysis. The final weights have three components: 1) the design weights, which are based on population sizes of specific regions as there was disproportionate stratification by region, 2) non-response weights with inverse propensity score weighting based on user-provided information in the Facebook app (including gender, age, and country of residence), and 3) post-stratification weights which make the total of the weights (based on age, gender, and subnational-region) equal to benchmark data for each sub-national region in the U.S. The strata were based on sub-national region (states).5 While the above Facebook app data was used to calculate weights, the sociodemographic data used in the models (age, race/ethnicity, gender, education, and occupation) were obtained from users via the survey itself. Model Specification Logistic regression models (with sets of covariates added in a hierarchical way) were developed based on the key variables of interest (in the case of manuscript 2, feelings of anxiety, depression, and COVID-19-related worry, and in manuscript 3, factor scores calculated based on a correlation matrix between various survey items representing constructs from the Theory of Planned behavior) and potential confounders (see section below). Fully adjusted models for both manuscripts were adjusted for age, gender, race/ethnicity, education, occupation, and time. The models in manuscript 3 were additionally adjusted for the presence of chronic health conditions. 8 The logistic regression models were used to generate odds ratios – these are appropriate to determine the presence of significant associations (which is appropriate considering the research questions and hypotheses laid out in manuscripts 2 and 3), but these values should not be used to approximate relative risk as the high prevalence of the outcome would result in an inflated estimate.6 In future research, additional statistical methods (such as log Poisson regression or log binomial regression) could be attempted to calculate prevalence ratios (which would more closely approximate risk ratios), and alternative study designs that maintain temporality could be implemented to measure risk directly for this particular research question. Assessment of Confounding For both manuscripts, bivariate analyses were conducted with Rao-Scott Chi-Square tests due to the weighted and categorical nature of the data. Significant variables that were flagged as potential confounders in the literature review were included in the logistic regression models in a hierarchical fashion. The odds ratios for the key variables of interest were then compared between the unadjusted and adjusted models with the change-in-estimate method. Assessment of Model Assumptions Multiple sensitivity analyses were conducted in both manuscripts to assess the potential bias introduced by assumptions (see Tables 4.8 and 5.9).This included assumptions that participants not shown a particular survey item would have responded in a certain way had they been shown the item (i.e., people indicating that they definitely intended to get vaccinated did not have any vaccine hesitancy reasons) and assumptions that particular observations could be excluded (due to answers suggesting that the survey was not completed in good faith) without substantially changing the results. 9 Potential Biases and Limitations As is common for cross-sectional survey data, potential limitations and biases include selection bias, information bias, and a lack of temporality. While several participants were interviewed over an extended period of time, participants were not followed up with to see if their individual intention to get vaccinated and/or reasons for vaccine hesitancy changed over time. They were also not followed up with to determine if vaccination behavior matched the intention measured by the survey; therefore, the key outcome is based on intention to vaccinate and not vaccination behavior itself. The application of the findings to influence vaccination behavior through vaccination intention is therefore operating under the assumption that intention accurately predicts behavior. While intention is a component of behavior (as suggested by the widely successful use of theories such as the Theory of Planned Behavior that are based on this assumption), the extent to which the two are correlated may vary based on many factors,7 and the level of correlation between intention and behavior for COVID- 19 vaccination specifically is in need of further study. The sampling process permitted a user to be reconsidered for invitation to the survey thirty days or more after their last invitation – such responses were not linked using any identification methods.8 It is possible that individual participants were interviewed multiple times over the course of the study and recorded as separate, unrelated observations, potentially violating statistical assumptions that observations are independent of one another. Another limitation is that not all potentially relevant variables were collected by the survey; for example, political affiliation was not among the information gathered. An alternative option that may be associated with political affiliation was considered (a group-level variable 10 indicating state); however, as state-level political affiliation is not a close proxy for individual political affiliation, making this assumption has the potential to introduce bias. Selection bias is a possibility due to the method of selecting participants (via Facebook) and because the survey was available in only English, Spanish, French, Vietnamese, Brazilian Portuguese, and simplified Chinese. Non-Facebook users and participants who did not speak one of the aforementioned languages would not be included in the study sample. Non-response bias is also likely, as is probable with voluntary surveys. For the U.S. CTIS, approximately 1-2% of those invited actually took the survey. The survey weights provided in the dataset are intended to counteract the aforementioned issues, limiting the effect of such bias.8 Information bias is also a possibility, as participants provided their own information both on their Facebook profiles (used to calculate weights) and in the survey responses themselves. This leaves the potential for social desirability bias (limited by the anonymous nature of the survey) as well as “deliberate trolling.” This is seen in individuals providing false information (and bragging about doing so) due to beliefs about conspiracy and suppressing of information by scientists about the COVID-19 pandemic. This is also seen in responses provided as protests against individual questions (e.g., selecting the “other” gender option and providing transphobic responses when asked to self-identify). This was not expected to meaningfully affect results unless looking at small, specific demographic groups (which was not the case for these manuscripts).8 Finally, the results of manuscripts 2 and 3 are limited in their generalizability due to the time periods in which the specific survey waves were implemented. Both sets of results are based on the experiences of U.S. residents at a time when the vaccine was already available, and data collected from earlier time periods would likely show different effects. Additionally, manuscript 11 2’s results are based on data from an earlier stage of vaccine availability than manuscript 3. Since the situation changed significantly over time (certain groups were prioritized early in the process and other groups became eligible as coverage among these key groups increased and overall availability increased), the descriptive statistics and the effects demonstrated in different time periods were expected to vary. In other words, any COVID-19 vaccine research results based on a particular level of vaccine availability (e.g., limited availability, general population availability of initial doses, or booster availability) may not be applicable to other time periods. 12 Chapter 3: Manuscript 1 | Systematic Review of Health Behavior Theory Use in U.S. COVID-19 Vaccination Hesitancy-Focused Studies Background Significance Despite the demonstrated safety and effectiveness of COVID-19 vaccines,2 uptake of COVID-19 vaccines has varied greatly. As of May 2023, some states and territories (such as Maryland, Hawaii, Puerto Rico, Guam, American Samoa, Republic of Palau, the Northern Mariana Islands, New York, Vermont, Massachusetts, Connecticut, Rhode Island, and Maine) have primary series vaccination rates of 80% or higher, with the District of Columbia having the highest rate at 91.3%. Other states have much lower rates of primary series uptake, including Wyoming (53.2%), Alabama (53.3%), and Mississippi (53.8%). These rates drop drastically when only considering those who had the updated booster dose, such as 34.3% in Vermont and only 6.8% in Mississippi.1 In addition to vaccine access and availability considerations, vaccine hesitancy is a significant concern in areas with low vaccine uptake. Existing Knowledge Coronavirus disease 2019 (COVID-19) is caused by infection with severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). The first vaccine for prevention of COVID- 19, Pfizer-BioNTech COVID-19 Vaccine, received emergency use authorization in December of 2020 for people 16 years of age or older.9 Since then, more vaccines have received emergency use authorization or FDA approval, including the Pfizer-BioNTech COVID-19 primary series and bivalent booster vaccines and the Moderna COVID-19 primary series and bivalent booster vaccines for individuals aged 6 months and older, as well as the Janssen COVID-19 vaccine and the Novavax COVID-19 Adjuvanted vaccine for individuals aged 18 years and older.10-13 13 Vaccine Hesitancy The Sage Working Group on Vaccine Hesitancy developed a definition for the concept in 2015, stating that, “vaccine hesitancy refers to delay in acceptance or refusal of vaccination despite availability of vaccination services. Vaccine hesitancy is complex and context specific, varying across time, place and vaccines.”14 This term and associated definition cover the extent and scope of this concept, whereas other terms such as vaccine confidence, uptake, acceptance, activism, criticism, and refusal are related but nonequivalent concepts. These other concepts often overlap with vaccine hesitancy and each other but should avoid being used interchangeably when possible for the sake of clarity.14-16 For example, one may be vaccine-hesitant and ultimately accept the vaccine, whereas another may ultimately refuse the vaccine despite having confidence in its safety and efficacy. This establishes vaccine hesitancy as a psychological state of indecision or otherwise being undecided on vaccination, rather than a behavior or an affect. Cognition and affect influence vaccine hesitancy, and behavior is the usual outcome of interest related to vaccine hesitancy, but vaccine hesitancy itself is distinct. 16 In this document, “vaccine hesitancy" will be used to refer to the state of indecision but will also incorporate vaccine delay and refusal (given the skip-function structure of the questionnaire used to collect the data to be used in manuscripts 2 and 3). Vaccination intention will refer to one’s plan, at a given point in time, to either get vaccinated or stay unvaccinated in the future. Vaccination status will refer to vaccine uptake (whether a person is vaccinated at a specific point in time). Common Individual Health Behavior Theories Individual and interpersonal health behavior theories were the focus of this systematic review for two reasons. The first is that this is intended as a prelude; the secondary data analyzed 14 in manuscripts 2 and 3 (chapters 3 and 4) have an individual-level focus, and individual health behavior theories will be further explored in those manuscripts. The second is that the wide variety of health behavior theories that exist are virtually innumerable, especially considering the constant evolution of health behavior theory. The individual-focused health behavior theories chosen for this systematic review are commonly used, robust, and have withstood the test of time.17 It is important to note that there is no one perfect theory for all situations – each theory has its strengths and limitations that may result in an ideal fit in one situation and a poor fit in another. The theory constructs appear in bold in the following sections. Note that the constructs were sometimes collapsed into fewer, broader construct categories or broken up into subconstructs in practice, all of which were commonly found to be referred to in the literature as “constructs”. This largely accounts for the different numbers of constructs cited for a particular theory in different pieces of literature. Health Belief Model The Health Belief Model has a long history, starting in the 1950s18-20 but it is still tested and used today. The main idea of the Health Belief Model is that modifying factors (such as demographics, personality, and knowledge) affect individual beliefs, and that the collective individual beliefs plus cues to action affect individual behaviors. Of note are six main constructs: 1) perceived susceptibility of disease, 2) perceived severity of disease, 3) perceived benefits of health behavior, 4) perceived barriers to the health behavior, 5) self- efficacy, and 6) cues to action. The first five constructs constitute the “individual beliefs,” with cues to action outside of this affecting individual beliefs and affecting individual behaviors directly. Perceived threat may be considered the combination of perceived susceptibility and severity. Strengths of the Health Belief Model are its intuitiveness and its versality, as it can and 15 has been usefully applied in a variety of behavioral health situations. Limitations include the lack of an emotional component and a lack of understanding relationships between constructs (and potential mediator or moderator effects).20 Theory of Reasoned Action The Theory of Reasoned Action assumes that intention is the best or main predictor of health behavior. The first key constructs include behavioral beliefs (what one thinks will be the outcome if the behavior is performed) and evaluations of behavioral outcomes (whether they are seen as positive or negative outcomes), which affect “attitude.” Additional constructs include normative beliefs (perception of whether other individuals will approve of the behavior) and motivation to comply (how important those individuals’ approval is) which affect “subjective norm.” In this theory, “external variables”, such as personality and demographics, affect these four main constructs which directly affect “attitude” and “subjective norm,” which in turn affect “intention to perform the behavior”, which directly affects “behavior”. Strengths of this theory include its relative simplicity (four main constructs, although they may be considered two constructs with two subconstructs each), which is practical for use and implementation, as well as being a great fit for health behaviors that are almost exclusively under volitional control. Its limitations are the flip side of the coin – it may be overly simplistic in certain situations, and it may not be as useful in situations where there are significant factors outside of volitional control.7, 21 Theory of Planned Behavior The Theory of Planned Behavior includes the four constructs (two broader categories) from the Theory of Reasoned Action but includes two more (which, again, may be considered one construct category). Control beliefs (perception of barriers or facilitators to performing the 16 behavior) and perceived power (the assumed level of impact of those perceived barriers or facilitators) affect “perceived control.” Similar to the Theory of Reasoned Action, “external variables” affect the six main constructs, which respectively affect “attitude,” “subjective norm,” and “perceived control,” in turn affecting “intention to perform the behavior” which ultimately affects “behavior” 7, 21, 22. A strength of this theory is that it addresses a limitation of the Theory of Reasoned Action by including constructs that are outside of volitional control. However, it does still have limitations in that it does not include additional outside factors such as those included in the Integrated Behavioral Model.7 Integrated Behavioral Model The Integrated Behavioral Model shares many characteristics with the Theory of Reasoned Action and the Theory of Planned Behavior with some substitutions and additions. “Attitude” is affected by the constructs experiential attitude (similar to behavioral beliefs) and instrumental attitude (similar to evaluations of behavioral outcomes). “Perceived Norm” is affected by injunctive norm (similar to subjective norm) and descriptive norm (beliefs about whether others perform the behavior). “Personal Agency” includes perceived control and the added construct of self-efficacy (beliefs about one’s ability to overcome barriers). “Attitude,” “perceived norm,” and “personal agency” all directly affect “intention to perform the behavior,” thereby indirectly affecting behavior itself. Other additions compared to the previous theories are factors that do not affect behavior indirectly through intention, but rather affect behavior directly. These include knowledge and skill, salience of the specific behavior, environmental constraints, and habits. A strength of this theory is its comprehensiveness, including factors from other theories and combining them into one large theory.7 However, what it gains in comprehensiveness it also gains in complexity, making it more difficult to use and implement. 17 Transtheoretical Model (with Stages of Change) The Transtheoretical Model constructs include the stages of change, the processes of change, decisional balance, and self-efficacy. The stages of change refer to one’s intention to take action (and how soon) then the action itself (and how long it is performed). The six stages are precontemplation, contemplation, preparation, action, maintenance, and termination. The processes of change refer to how one goes through the stages of change; this includes consciousness raising, dramatic relief, self-reevaluation, environmental reevaluation, self- liberation, helping relationships, social liberation, counterconditioning, stimulus control, and reinforcement management. Decisional balance refers to the pros and cons of making the behavior change. Self-efficacy here is comprised of two components: confidence and temptation. An important distinction between this theory and those previously discussed is that it is a theory of behavioral change, rather than a theory of behavior. A strength of this theory is the inclusion of emotional factors (as represented by “dramatic relief” in the processes of change), while a limitation is that it may be less applicable to children and adolescents.23 Theories Added After Preliminary Search Some theories and models that were referenced in the literature that was added after a preliminary search (not among the original search criteria) were the 5 Cs of COVID-19 Vaccine Hesitancy,24 the Extended Parallel Process Theory,25 the Reasoned Action Approach,26 and Protection Motivation Theory.27 The Reasoned Action Approach was incorporated in the formal search, as it is a more recently developed, expanded version of the Theory of Planned Behavior with very similar constructs, and has been found to be useful in understanding health behavior.28 The 5 Cs of Vaccine Hesitancy was included in the formal search as it is largely based on the Theory of Planned Behavior and Health Belief Model.24, 29 The 3 Cs Model of Vaccine 18 Hesitancy was also included as it was a more general predecessor to the 5 Cs Model.14 Protection Motivation Theory27 was also included in the full search, as it were determined in the process of the search to be individually or interpersonally-focused and have a specific relevance to health behavior. The Extended Parallel Process Theory 25 was included in the full searches, but ultimately the article with this theory was not included as it was not a quantitative study (see Criteria for Selection). Reasoned Action Approach / Reasoned Action Model The Reasoned Action Approach, developed in 2011, is similar in structure to the Theory of Reasoned Action and the Theory of Planned Behavior in three of the main construct categories (attitude, perceived norm, and perceived behavioral control), but also includes the construct of actual control, including environmental factors, skills, and abilities. In this model, the actual control construct directly affects the relationship between intention (as formed by the three other constructs) and behavior.30 Protection Motivation Theory Protection Motivation Theory was developed in 1975 and has since undergone revision.31 It includes seven overall constructs grouped into three schemes, including: 1) threat appraisal, based on threat vulnerability, threat severity, and maladaptive rewards 2) coping appraisal, based on response efficacy, self-efficacy, and response cost, and 3) emotion (fear), the last of which mediates the relationship between severity/vulnerability and the overall outcome, protection motivation.32 19 3 Cs The “3 Cs” Model of Vaccine Hesitancy was developed by the SAGE (Strategic Advisory Group of Experts on Immunization) Working Group on Vaccine Hesitancy and published in 2015. The three categories in the 3 Cs Model include 1) confidence, 2) complacency, and 3) convenience.14 The first category, confidence, pertains to the trust one has in the vaccine itself (that it is safe and effective), confidence that those who provide the vaccine-related health services are reliable and competent, and trust that those who recommend vaccines (i.e., policymakers) have genuine, unadulterated motivations for doing so.14 This has some, but not complete, overlap with the Theory of Planned Behavior construct attitude and the Health Belief Model constructs perceived benefits and perceived barriers. The second category, complacency, has significant overlap with the Health Belief Model constructs of perceived susceptibility and perceived severity, as well as having relations to a construct common to many health behavior models, self-efficacy. Complacency here refers to how necessary the vaccine is perceived to be; if perceived risks are low, vaccination may not be considered important, particularly compared to other behaviors and responsibilities.14 The third category, convenience, speaks to perceived and/or actual barriers or lack thereof. Some examples of factors influencing convenience include real and/or perceived availability, accessibility, cost, cultural context, comfort, and health literacy.14 This construct has commonalities with the previously mentioned constructs of perceived barriers from Health Belief Model and self-efficacy from several health behavior theories. 20 5 Cs In 2021, the 3 Cs model was expanded and tailored specifically to COVID-19 vaccine hesitancy, adding the additional behavioral factors 4) communication and 5) context.29 The communication component focuses on the prevalence of misinformation that can generate and amplify fear, anxiety, confusion, and distrust. The 2021 “5 Cs” article suggests that this be addressed with open and genuine dialogue and community engagement and social media accountability. The context component refers to sociodemographic characteristics, including but not limited to religion, ethnicity, and socioeconomic status. (This takes from and expands on the cultural context component of the convenience category in the 3 Cs model).14 The “5 Cs” article proposes that messaging be provided through trusted healthcare services and respected community leaders, with an emphasis on cultural competence.29 Other theories, models, and frameworks that were identified but did not fit the qualifications of a health-related theory and/or an individual or interpersonal-focused theory included Ecological Theory and Social Determinants of Health. Importance of Health Behavior Theory in Epidemiology Health behavior theory has primarily been the domain of behavioral and community health, rather than epidemiology. However, much of the work in this area focuses, by design, on creating and developing interventions using relatively small sample sizes. Taking these established theories and incorporating them with epidemiological and biostatistical national survey methods at the outset of data analysis has the potential to bridge epidemiological findings related to vaccination and the development of wide-spread (national-level) interventions to affect vaccination as a behavior. 21 Gaps in Knowledge Health behavior theory has been applied to COVID-19 vaccination research; however, what theory is most applicable can vary based on population and time. Using an untested or poorly applicable theory to a specific situation or population of interest is akin using “a compass on Mars.”33 Particularly considering the heterogeneity among what may broadly be considered the “vaccine hesitant” population, the application of a variety of theories may be appropriate. Creating a single intervention based on one theory may be too generalized and may be a poor fit for many vaccine-hesitant individuals. This must additionally be balanced with practicality in terms of time, resources, and feasibility of getting the intervention to the intended recipients without also contacting unintended recipients for whom the intervention may be ineffective or counterproductive. Specific Aims Objective of the Project The objective of this first manuscript is to conduct a systematic review of the use of health behavior theory in literature describing or analyzing COVID-19 vaccine hesitancy or confidence studies that take place partially or entirely within the United States. Research Question and Hypotheses The research aims of interest are to determine the extent or frequency of use of each of the well-established health behavior theories mentioned in COVID-19 vaccine hesitancy research, to determine how influential the theory was in study development and/or analysis, and to determine the extent to which each theory was found to be useful for the given population of interest in each article. 22 It was anticipated that the older theories (such as Health Behavior Theory) would likely be more represented because they have had more time to disseminate in the scientific community and have been tested more extensively, suggesting credibility. It was also anticipated that more simplistic (i.e., fewer constructs) theories will be more represented due to their relative ease of implementation. This would include Health Behavior Theory, Theory of Reasoned Action, and Theory of Planned Behavior. It was hypothesized that studies that draw from multiple health behavior theories will see less influence from each theory referenced on the overall study, either having one largely influential theory and other less influential ones, or multiple somewhat influential theories depending on whether the authors chose one as the primary theory or chose to integrate the theories. Lastly, it was hypothesized that the utility of the theories will vary based on multiple factors such as the population of interest. A theory’s applicability may vary based on demographics and relevant social contexts,33 and a consistent level of utility across multiple population is not expected. Of interest was whether the theories are merely more or less useful, or if in some instances the use of a particular theory has no utility or is counterproductive to public health. Theoretical/Conceptual Framework The theoretical frameworks being studied in this systematic review include the Health Belief Model, the Theory of Reasoned Action, the Theory of Planned Behavior, the Integrated Behavioral Model, the Reasoned Action Approach, the Transtheoretical Model/Stages of Change, the Protection Motivation Theory, the 3 Cs Model, and the 5 Cs Model. The systematic review itself was conducted according to the updated (2020) PRISMA guidelines.34 23 Methods Overall Study Design This manuscript is a systematic review conducted based on the PRISMA guidelines.34 Five of the previously listed health behavior theories were included in the initial search, but additional relevant theories were added through an iterative process until no additional relevant theories were found in the articles returned by the search. Data Sources The databases PubMed,35 Scopus,36 Web of Science,37 Public Health,38 and Academic Search Ultimate39 were searched for relevant articles (the databases indicated as “best bets” by the University of Maryland University Libraries Database Finder in the subject of Public Health.40 Duplicates were identified and removed, first using EndNote software and then verified by the main researcher. Criteria for Selection and Preliminary Search The inclusion criteria for the literature in this systematic review were as follows: Screening criteria (based on title, abstract, and available keywords): 1. COVID-19 vaccination focus a. The situation that developed was unique but could provide valuable information for a future public health crisis. 2. “Vaccine hesitancy” or “vaccine confidence” mentioned, or a closely related concept. a. The intention is to focus on how or why people did or didn’t choose to get vaccinated – reporting vaccination rates alone was not considered to be sufficient for inclusion. 24 3. Mention of a particular health behavior theory or theories with individual and/or interpersonal focus. a. This includes but is not limited to Health Belief Model, Theory of Reasoned Action, Theory of Planned Behavior, Integrated Behavior Model, Transtheoretical Model, etc. An iterative process was implemented to accumulate additional relevant theories/models that fit in the same category as those already mentioned (individual-or interpersonal-level and health behavior-focused). The search process was modified until no new relevant theories appeared in any of the returned articles. b. There are theories that have individual or interpersonal levels within them but have a wider perspective. This perspective is valuable, but due to the research question at hand and the nature of the vaccine hesitancy data that is analyzed in manuscripts 2 and 3, a primarily individual and/or interpersonal theory focus is most appropriate. Theories that have a wider perspective can lack the necessary level of individual or interpersonal detail for this particular research question. As an example, ecological models have their strengths, including the consideration of multi-level interactions, but they also have the weakness of lack of specificity that must be supplemented by other theories (an ecological theory can be viewed as more of a framework for multiple, more specific theories at each level, or a “meta-model”). Therefore, using ecological models without the guidance of other theories, such as individual- and interpersonal-level theories, may result in difficulty developing testable hypotheses and specific direction for intervention development.41 4. Study participants include population(s) residing in the United States (with statistics available for U.S. participants separate from the entire sample in multi-national studies). 25 Assessment Criteria (full text review): 5. Includes original data collection and/or analysis (not a review). a. This could include descriptive and analytic studies, and primary data collection as well as secondary data analyses. b. This is to exclude any items that did not report novel data or data interpretations on a population. (Other systematic reviews that were performed using different criteria were also excluded.) 6. At least one named construct from the named theory is used to develop interventions, guide initial study design and research question development, develop survey questions or other methods of measurement, and/or guide analysis of data after collection. a. This criterion was specified to exclude studies that reference a health behavior theory in the introduction or discussion, but do not use that theory to guide the study in any clearly observable way. 7. Aspect(s) of vaccine hesitancy, confidence, intention, uptake, or behavior as a main variable of interest. (Note: these were originally to be categorized as exposures or outcomes, but that element was removed to allow for the inclusion of correlation-based studies with no clear exposure-outcome assignment and qualitative studies that may seek to identify themes rather than assess a specific exposure-outcome relationship.) A preliminary search was conducted in June of 2023 to assess the feasibility of the systematic review based on the existing parameters (in terms of article quantity) and to add search parameters as deemed relevant. Between the preliminary and the full search additional search parameters were added, including five theories and models (as previously mentioned) and the names of all U.S. states 26 and territories (rather than simply “United States” OR “U.S.” OR “US” OR “U.S.A.” OR “USA” as this might have excluded studies conducted in the U.S. that were particular to a region). Full Search The full search was conducted on January 3, 2024, and included the following search terms (see Table 3.1) which had to be included in the abstract, title, or keywords if database allowed searching by keywords. Table 3.1 Search terms 1. covid-19 or coronavirus or 2019-ncov or sars-cov-2 or cov-19 AND 2. vaccin* or immunization* AND 3. hesitan* OR confiden* OR refus* OR delay OR decline OR confiden* AND 4. “Theory of Planned Behavior” OR “TPB” OR “Theory of Reasoned Action” OR “TRA” OR “Integrated Behavior Model” OR “IBM” OR “Health Belief Model” OR “HBM” OR “Transtheoretical Model” OR “TTM” OR “Stages of Change” OR “5 Cs” OR “5Cs” OR “Reasoned Action Approach” OR “RAA” OR “3Cs” OR “3 Cs” AND 5. “United States” OR “U.S.” OR “US” OR “U.S.A.” OR “USA” OR “Alabama” OR “Alaska” OR “Arizona” OR “Arkansas” OR “California” OR “Colorado” OR “Connecticut” OR “Delaware” OR “Florida” OR “Georgia” OR “Hawaii” OR “Idaho” OR “Illinois” OR “Indiana” OR “Iowa” OR “Kansas” OR “Kentucky” OR “Louisiana” OR “Maine” OR “Maryland” OR “Massachusetts” OR “Michigan” OR “Minnesota” OR “Mississippi” OR “Missouri” OR “Montana” OR “Nebraska” OR “Nevada” OR “New Hampshire” OR “New Jersey” OR “New Mexico” OR “New York” OR “North Carolina” OR “North Dakota” OR “Ohio” OR “Oklahoma” OR “Oregon” OR “Pennsylvania” OR “Rhode Island” OR “South Carolina” OR “South Dakota” OR “Tennessee” OR “Texas” OR “Utah” OR “Vermont” OR “Virginia” OR “Washington” OR “West Virginia” OR “Wisconsin” OR “Wyoming” OR “American Samoa” OR “Northern Mariana Islands” OR “CNMI” OR “Guam” OR “Puerto Rico” OR “Virgin Islands” OR “USVI” OR “Micronesia” OR “FSM” OR “Marshall Islands” OR “RMI” OR “Palau” After the review of the items returned from the initial full search, the search was run again with (“Protection Motivation Theory” or “PMT”) for parameter #4, with the otherwise 27 same search parameters (numbers #1 - 3 and #5) and the same time constraints (published on or before Jan 3, 2024). For Protection Motivation Theory, two additional articles were returned and added to the list for screening and assessment. Results A total of 225 articles were returned by the search (see Figure 3.1). Of these, 140 were removed as duplicates, 33 did not meet the screening criteria, and 8 did not meet the assessment criteria, leaving 44 articles. Results of the search, screening, and assessment processes were assessed by a second reviewer, Rodman Turpin, Ph.D., M.S.1 Articles were then additionally restricted to those with quantitative analyses (n = 36), due to the structure of the three research questions proposed and how they could be feasibly assessed. Research Question 1 What is the extent or frequency of use of each of these health behavior theories in COVID-19 vaccine hesitancy research? A summary of the findings is available in Table 3.2 and Table 3.3. Among the articles included in the review, the Health Belief Model (HBM) was the most commonly used by far, considering both instances where HBM was used alone42-59 (18 articles) and instances where HBM was used in combination with other theories (6 articles).26, 60- 64 The second most frequently used theory was the Theory of Planned Behavior, used alone in 5 articles65-69 and in combination with other theories in 4 articles.60-63 Other theories, models, and frameworks were relatively rare, with some only appearing in combinations with other theories or models. These findings are consistent with the research hypothesis. HBM and TPB are both older, well-established theories that have relative simplicity compared to other models. It is 1 Assistant Professor, Department of Global and Community Health, College of Public Health, George Mason University. Second reviewer; provided guidance in study methodology design. 28 Figure 3.1. Flowchart of search, screening, and assessment processes. 29 Table 3.2. Theories or models by article (alone or in combination) Theory, Model, or Framework Number of Articles HBM alone 42-59 18 TPB alone 65-69 5 HBM & TPB 60-63 4 3 Cs alone 70-72 3 PMT alone 73, 74 2 TTM alone 75 1 HBM & TRA 64 1 HBM & RAA 26 1 RAA & IBM 76 1 Table 3.3. Overall frequencies of theories or models Theory, Model, or Framework Frequency HBM 24 TPB 9 3 Cs 3 RAA 2 PMT 2 TRA 1 TTM 1 IBM 1 possible that TRA was not as prevalent as expected because it was “supplanted” by TPB, which is based on TRA.7 Use of theory varied over time, with those studies with data collected by the end of 2020 tending to use the Health Belief Model and the Theory of Planned Behavior and studies conducted later using a wider variety of models, theories, and frameworks. Research Question 2 How influential was the theory in study development and/or analysis? This was determined based on two key elements: the influence of the theory (in other words, how many of the constructs were referenced) and the application of the theory (how closely did the study’s use HBM = Health Belief Model IBM = Integrated Behavioral Model PMT = Protection Motivation Theory RA = Reasoned Action RAA = Reasoned Action Approach TPB = Theory of Planned Behavior TRA = Theory of Reasoned Action TTM = Transtheoretical Model 30 of the constructs match the established definitions as discussed in the background section)? Influence/application was rated on a scale from (1) limited, (2) somewhat, (3) influential, and (4) very influential. Those rated as having limited influence and application were articles that used very few of the constructs and/or the study elements asserted to be based on the constructs bore little to no resemblance to the established definitions of the constructs. Those rated as “somewhat” either referenced most or all of the constructs (but applied them relatively loosely), referenced only some of the constructs but applied them very well, or some combination thereof. Those rated as “influential” used most of the constructs that were well applied, or they used all of the constructs but a small subset of them were relatively loosely applied. Those rated as “very influential” used all the constructs and had elements that clearly addressed each one that closely matched the established construct definitions. A summary of these ratings is available in Table 3.4, but a full list of the articles with population characteristics, time of data collection, sample sizes, and explanations of the reasons for the ratings is available in the Appendix in Tables A3.1 – A3.6. Table 3.4. Level of influence of the theory on the study development and/or analysis. HBM TPB 3Cs PMT RAA TRA IBM TTM Total 1 – Limited 5 1 . . . . . . 6 2 – Somewhat 4 1 . . . . . . 5 3 – Influential 11 1 1 2 2 1 1 1 20 4 – Very influential 4 6 2 . . . . . 12 Total 24 9 3 2 2 1 1 1 Often, the theory constructs were used to develop questionnaire items and/or select questionnaire items from preexisting survey tools. It was very common for studies to use most 31 but not all of the constructs in study design, or to use all of the constructs and have one or two of them loosely applied. The few studies where all the construct or construct categories were thoroughly applied use the Health Belief Model, Theory of Planned Behavior, or the 3 Cs Model. It was also not uncommon for articles to reference the Health Belief Model, but to only include one or two constructs or not clearly allocate any survey items or aspects of analysis to specific constructs. Research Question 3 To what extent was each theory found to be useful for the given population(s) of interest in each article? Utility was considered based on the significance of the items associated with the constructs. Usefulness/significance was rated as (1) limited, (2) somewhat, (3) useful, or (4) very useful. Note that the usefulness is affected by the level of influence and application from the previous section – if few constructs were mentioned or the constructs were loosely applied, the usefulness or significance would also be inherently low. In such an instance, since either the constructs were not used or they did not resemble the typical definition of the constructs, the potential usefulness of the theory if it had been influential and well applied was unknown. The rating of “limited” usefulness indicates that none of the constructs were meaningful or significant in the study conclusions. “Somewhat” useful indicates that at least one of the constructs was meaningful (up to two for theories with greater numbers of constructs) to the study’s conclusions. “Useful” indicates that some or a majority of the constructs were significant or meaningful, and “very useful” indicates that all constructs were significant or meaningful. A summary of these ratings is available in Table 3.5, but a full list of the articles is available in the Appendix in Tables A3.1 – A3.6. 32 Theories generally had some level of utility in each article, with each having at least one construct being meaningful to the overall results and conclusions. There were four articles where theories were very useful (i.e., all the constructs or construct categories were significant or meaningful). Two of these articles used the Health Belief Model; one article focused on the general adult population in the U.S.,45 and the other specifically focused on first responders in Kentucky.48 One of these articles used the Theory of Planned Behavior and focused on the student, faculty, and staff population at a residential research university in Northeastern Ohio.68 Lastly, one of these articles used the 3 Cs Model and focused on Black and White residents in Southeast Michigan.72 However, the particular piece(s) of context that made these theories a good fit for these situations (demographics, location, time, etc.) are difficult to discern. Table 3.5. Level of usefulness or significance of the theory on the study’s population based on the results and conclusions. HBM TPB 3Cs PMT RAA TRA IBM TTM Total 1 – Limited . . . . . . . . 0 2 – Somewhat 12 5 . . . . . . 17 3 – Useful 10 3 2 2 2 1 1 1 22 4 – Very useful 2 1 1 . . . . . 4 Total 24 9 3 2 2 1 1 1 Bias Assessment Articles were assessed for bias using JBI Critical Appraisal tools for cross-sectional and cohort studies,77 experimental studies,78 qualitative studies,79 and mixed method studies (combination of other appraisal tools as appropriate).77-80 Of the 36 articles, 31 were categorized as cross-sectional studies, 2 as cohort studies, 2 as mixed-methods studies, and 1 as an experimental study. 33 The bias assessment included the consideration of multiple factors, including but not limited to the clear definition of inclusion criteria, valid and reliable measures of exposure and outcome, identification of confounders, and appropriateness of statistical analysis method used. Each factor was to be given an assessment of “yes,” “no,” “unclear,” or “not applicable,” which ultimately was used for an overall appraisal result of "include,” “exclude,” or “seek further information.”77-80 While all 36 studies had strengths and limitations, none of the limitations found were considered significant enough to warrant the designation of “exclude” or “seek further information.” Discussion The Health Belief Model was the most used theory in addressing COVID-19 vaccine hesitancy. It was common for theories not to be fully applied to a particular study, with select constructs being focused on and others relatively or completely ignored. While this is unavoidable in some studies, such as in the case of secondary data analysis, studies should ideally include all the constructs or explicitly state the reasons for excluding specific constructs. Otherwise, this makes it difficult to compare the usefulness of the theories in these studies, as many of the pieces that were left out were what made each theory unique from the others (e.g., authors can “trim” a theory to fit their study and easily cite the influence of any number of theories with overlapping constructs). For studies that attempted to implement more than one theory, generally both studies had mild influence (with a few constructs from each, or with focus on the overlapping constructs). In one instance where both theories were fully implemented, the resulting usefulness of each of the theories was relatively low.63 Each of the theories was at least somewhat useful for its particular population, with some being more useful than others but most not reaching the threshold of having all significant 34 constructs. This suggests that, among the theories considered for COVID-19 vaccine hesitancy and uptake research, there is not necessarily a wrong or poorly fit theory for any particular population, but rather that the theories are generally useful but some are a very good fit for a particular situation. COVID-19 vaccination confidence and hesitancy research may benefit from additional consideration of the Protection Motivation Theory that is currently underutilized but has a similar level of complexity as the widely used Health Belief Model (in terms of number of constructs). Protection Motivation Theory has the unique construct of maladaptive response rewards (or maladaptive benefits) that is distinct from constructs in other theories such as perceived barriers, attitude, or behavioral beliefs in that it explicitly teases out reasons why individuals may not perform the health behavior specific to perceived positive consequences of not doing the behavior (rather than a broader concept that includes these reasons plus reasons related to convenience, complacency, or confidence and focuses on negative consequences of performing the behavior). This construct can include potential items such as the perception that not getting the COVID-19 vaccine saves the participant time and money, or that not getting the vaccine would mean the participant doesn’t have to worry about side effects74 (rather than the common framing in other theories’ constructs where these may be addressed as barriers: cost of the vaccine, or attitude: concern about side effects). The different framing of the question (from negative to positive) may affect how participants respond to items that are attempting to measure these reasons. This could be applied to a variety of different reasons that are typically framed as “barriers” (e.g., “not doing the behavior saves me time/means I can spend time doing other things that I enjoy” may elicit very different responses than the more familiar “I don’t have the time to perform this behavior”). 35 COVID-19 vaccine hesitancy research may benefit from revisiting theories such as the Reasoned Action Approach, Theory of Reasoned Action, Integrated Behavioral Model, or Transtheoretical Model but with a focus on full application, as many of the studies that used these theories did not use all the theory’s constructs. However, the constructs that were used tended to be significant, strongly suggesting the theories’ potential relevance and utility. Study Strengths and Limitations PRISMA method. The PRISMA method, or the Preferred Reporting Items for Systematic Review and Meta-Analyses method, was first published in 2009. An updated version was published in 2020.34 The PRISMA method provides standardized language for use in systematic reviews, a flow diagram to guide identification of relevant studies, and an extensive checklist on how systematic reviews should be reported and presented. This standardization of search methods and reporting based on years of use, evaluations, and revisions of the method lends an increased soundness and reproducibility to systematic reviews that use the PRISMA method. Publication bias. The use of the PRISMA systematic review process will result in the inclusion of all relevant published studies, but not all studies get published. Research that lacks novelty (such as confirmatory studies) or research that is not popular is less likely to be included in a systematic review, regardless of the thoroughness of the search. Search procedures. While effort was made to include a wide variety of search terms that would capture the literature that met the search criteria (COVID-19 vaccine hesitancy focus in the United States), it is possible that some literature was missed. This could be due to unusual phrasing or the use of less common synonyms for the criteria components, as well as references to specific study locations (such as cities or counties in the United States) in literature that does not otherwise mention the state or country in the title, abstract, or keywords. 36 Bias within found studies. A systematic review is dependent on the individual studies it is composed of, and those studies may themselves have a variety of biases such as selection or information biases. The PRISMA method includes a risk of bias assessment so that each study may be considered in an appropriate context, but this bias cannot be removed or ignored. The JBI Critical Appraisal tools 77-80 were used to assess bias, but each of these tools includes items with options of “yes,” “no,” “unclear,” or “not applicable.” Given that no study is completely free from bias and studies have unique research questions that may require different participant selection, measurement, and analysis approaches, studies were included in the review if there was demonstration of consideration for each main source of potential bias mentioned in the relevant JBI tool (such as selection of participants, validity and reliability of measures, addressing confounding, etc.). While qualitative studies were initially considered for inclusion in the review, the method of measurement outlined for two of the three research questions were best aligned with quantitative studies. Revision of the methods of measurement (and perhaps the research questions themselves) may be necessary to properly evaluate qualitative studies for their use of health behavior theory. Human Subjects and Ethical Considerations Because the method used is a systematic review of existing published scientific literature, personal identifiers were available to the investigator. Therefore, this does not fall under the category of human subjects research. 37 Chapter 4: Manuscript 2 | Investigating the Effect of Frequency of Recent Anxious and Depressed Feelings on COVID-19 Vaccination Intention and Vaccine Hesitancy among Unvaccinated Individuals in the United States Background Significance Many individuals in the U.S. delayed getting (or have yet to receive) a COVID-19 vaccine despite their availability. Thus, vaccine hesitancy is important to understand moving forward in an effort to communicate effectively with unvaccinated individuals who would be good candidates for COVID-19 vaccination (i.e., no medical contraindications). Clarifying not just the specific reasons why someone is vaccine hesitant, but also the corresponding emotional responses, is a critical step in developing effective communication and intervention materials. Emotional appeals, often used in public health messaging, have the potential to reduce or increase the effect of the messaging depending on not just the message contents, but on the characteristics of the message recipient. As an example, a fear-appeal based message (focused on risk severity and susceptibility) can be successful among certain audiences, but it also has the potential of unintended consequences such as increased anxiety and denial in others.81 Knowing the average emotional state (nervousness, anxiety, depression, etc.) of people with a specific vaccine hesitancy reason can allow for the development of carefully crafted messages to address, and avoid exacerbating, negative emotions associated with that specific reason. This has the potential to increase the effectiveness of COVID-19 vaccine messaging and interventions that take this information into account. 38 Existing Knowledge Anxiety and Depression during COVID-19 Rates of anxiety and depression were generally higher than usual during the pandemic. COVID-19 had a global effect that was disruptive to people’s lives and certainly had the potential to induce anxious and depressed feelings. A study assessing the impact of COVID-19 on the prevalence of anxiety and depressive disorders found that U.S. adults were over three times as likely to screen positive for one or both in mid-2020 compared to 2019.82 Another study, focused on Canadian and U.S. individuals in mid-2020, found high levels of anxiety that COVID-19 would infect their loved ones (88%), with the majority reporting significantly high levels of stress (63%), and many meeting the criteria for Generalized Anxiety Disorder and Major Depressive Disorder (31% and 29%, respectively), with younger age, female sex, and past mental health treatment being significant predictors.83 Psychological Factors and Health Behaviors The relevance of psychological states, such as anxiety and depression, have been demonstrated to be relevant in research on health behaviors or treatment compliance for other health conditions. Among dialyzed patients, anxiety and depression were found to be mediators of the relationship between health literacy and diet non-adherence, suggesting that attention be given to patients’ psychological distress to increase diet adherence.84 Among those with depression, the Medication Possession Ratio was demonstrated to be below the standard of adherence, with increasing comorbidity associated with lower adherence.85 39 Psychological Factors and Vaccine Hesitancy Since the beginning of the pandemic, it has been suggested that emotional or psychological states may play a significant role in vaccine hesitancy and, by extension, should be a significant consideration in vaccine communications and messaging.86 One study found that a very large proportion of variation in vaccination intention (about 70%) was explained by psychobehavioral factors (identified as distinct “personas” such as “COVID Skeptics” and “System Distrusters”), compared to a very small proportion of variation (1%) due specifically to demographic factors.87 Recent research among a small South Korean sample has suggested that a high score on the GAD-7 (Generalized Anxiety Disorder-7) could be a risk factor for vaccine hesitancy.88 In this research, depression, as measured by the Patient Health Questionnaire-9 (PHQ-9), was not a significant factor in the final model; however, this could perhaps be explained by the relatively small sample size (275) and the presence of eleven potentially overlapping individual psychological and personality variables included in the final model; in unadjusted bivariate analysis, the PHQ-9 score was borderline significant (p=0.08).88 A study analyzing a convenience sample in Germany in January 202189 found that COVID-19 and health-related anxiety and fears were positively associated with COVID-19 vaccine acceptance, whereas social and economic fears were negatively associated. The authors suggest that health-related fears are “functional” fears that promote public health compliance (by reducing the probability of the negative outcomes that are feared) whereas social and economic fears were “dysfunctional” in nature and worth further study (but potentially being explained by factors such as the absence of adaptive internal coping or external factors to buffer adverse economic or social consequences). More general depressive and anxiety symptoms were not significantly associated with vaccine acceptance in this study. Worth noting is that these 40 independent variables of interest were all considered separately in the analysis (Spearman rank correlations and Kruskal-Wallis tests), only controlling for age and educational level, so the effect of multiple types of anxiety and depression were not considered together in one model.89 In contrast, a study on the effect of Social Anxiety Disorder (SAD) and vaccination behavior90 found that those with SAD were more likely to receive or plan to receive a COVID-19 vaccine. These results lend credence to the authors’ theory that fear of negative evaluation from others may compel those with SAD to comply with the social norm (which, among many young populations such as at universities, trended towards getting vaccinated). It should also be noted that this was also a small research sample (84 individuals) regarding a specific type of anxiety in a specific population (southeastern U.S. university),90 so this may not apply to broader populations. Gaps in Knowledge A reasonable extension of existing research would be the hypothesis that recent anxious, nervous, or depressed feelings in general could also affect vaccine hesitancy, and specifically that they could increase or decrease the probability of identifying with specific vaccine hesitancy reasons, depending on the source of anxiety. More information is needed on the effect of anxious and depressed feelings on vaccine hesitancy and ultimately vaccine behavior and uptake, particularly in later months of the pandemic (2021 – 2022), in U.S.-based populations, and with larger sample sizes identified and recruited with probability sampling methods. Additionally, more research is needed on this topic regarding anxious and depressed feelings in general (as opposed to exclusively disorder-levels of anxiety and depression, since emotional states themselves may play a role in vaccine hesitancy, or health-related anxiety specifically). This 41 research proposes to address this gap using survey data from millions of U.S.-based individuals collected as part of the U.S. COVID-19 Trends and Impact Survey3. Specific Aims Objective of the Project The objective of this research is to determine the extent to which feelings of anxiety, nervousness, or being on-edge influence vaccination behavior and intention, as well as the extent to which they influence whether a person identifies with a specific vaccine hesitancy reason, such as being concerned about side effects of the vaccine, not knowing if the vaccine will work, the vaccine being against one’s religious beliefs, and so on. Identifying the role of feelings of anxiety in specific vaccine hesitancy reasons can aid in the creation of more effective public health communication for COVID-19 and for future novel infectious disease public health crises. Research Question and Hypotheses The two main sets of research questions are as follows: 1. Do recent feelings of nervousness, anxiety, and being on edge, feelings of depression, and/or COVID-19 worry significantly affect vaccination behavior or intention to get vaccinated? 2. Do recent feelings of nervousness, anxiety, and being on edge, feelings of depression, and/or COVID-19 worry significantly affect whether people identify with the following statement(s):  I am concerned about possible side effects of a COVID-19 vaccine. (1)  I am concerned about having an allergic reaction to a COVID-19 vaccine. (2)  I don't know if a COVID-19 vaccine will work. (3) 42  I don't believe I need a COVID-19 vaccine. (4)  I don't like vaccines. (5) / I don't like vaccines generally. (16)  My doctor has not recommended it. (6)  I plan to wait and see if it is safe and may get it later. (7)  I don't trust COVID-19 vaccines. (10)  I don't trust the government. (11)  It is against my religious beliefs. (15)  I have a health condition and am concerned about the safety of the vaccine for people with my condition. (12)  I am currently/planning to be pregnant and/or breastfeeding and do not want to get vaccinated at this time. (14) Two answer options, “I think other people need it more than I do right now” (8) and “I am concerned about the cost of a COVID-19 vaccine” (9) will not be considered since they do not describe vaccine hesitancy per se, but rather availability or perceived control. The hypotheses are: 1a. Higher frequency of recent feelings of nervousness, anxiety, and being on edge are significantly positively associated with being vaccine-hesitant (e.g., negatively associated with being vaccinated / intending to get vaccinated). 1b. Higher frequency of recent feelings of depression are significantly positively associated with being vaccine-hesitant (e.g., negatively associated with being vaccinated / intending to get vaccinated). 43 1c. COVID-19-related worry is significantly negatively associated with being vaccine-hesitant (e.g., positively associated with being vaccinated / intending to get vaccinated). 2a. Higher frequency of recent feelings of nervousness, anxiety, and being on edge are significantly positively associated with identifying with the vaccine hesitancy reasons listed above. 2b. Higher frequency of recent feelings of depression are significantly positively associated with identifying with the vaccine hesitancy reasons listed above. 2c. COVID-19-related worry is significantly negatively associated with identifying with the vaccine hesitancy reasons listed above. The basis for these hypotheses is existing research suggesting that certain types of anxiety (such as COVID-19 and health-related fears) are positively associated with COVID-19 vaccine acceptance and that other types of anxiety or depression are negatively associated with COVID-19 vaccine acceptance. Theoretical/Conceptual Framework This research question fits well with several existing models and frameworks. The Theory of Planned Behavior, Theory of Reasoned Action, and Integrated Behavioral Model According to the Theory of Planned Behavior, individual-level factors can directly affect the main constructs of attitude, subjective norm, and perceived control (see Figure 3.1). This is also true of the related Theory of Reasoned Action and Integrated Behavioral Model7 and also for the Health Behavior Model20, 91. It was hypothesized here that states of anxiety and depression (measured by how often one has felt that way over the past 5- or 7-day period) are 44 such individual-level factors that will significantly affect these main constructs as they pertain to COVID-19 vaccination beyond other common individual-level factors such as demographics. In U.S. COVID-19 Trends and Impact Survey, data was collected on feelings of nervousness, anxiety, and being “on-edge,” as well as on specific reasons for not having been vaccinated or not definitely intending to be vaccinated. Many of these vaccine hesitancy reasons can be allocated into one of the three main constructs of the Theory of Planned Behavior – more detail will be provided in Chapter 5’s Theoretical and Conceptual Framework section). The Three Cs of Vaccine Hesitancy and the Five Cs of COVID-19 Vaccine Hesitancy The vaccine hesitancy reasons from the U.S. CTIS can be allocated to some of the behavioral factors in the “3 Cs” Model of Vaccine Hesitancy and the “5 Cs of COVID-19 Vaccine Hesitancy,”29 with a heavy focus on confidence and trust (Table 3.1). The constructs of these models share some overlap with other health behavior theories as well. The “3 Cs” Model of Vaccine Hesitancy was developed by the SAGE (Strategic Advisory Group of Experts on Immunization) Working Group on Vaccine Hesitancy and published in 2015.14 The three categories in the 3 Cs Model include 1) confidence, 2) complacency, and 3) convenience. The first category, confidence, pertains to the trust one has in the vaccine itself (that it is safe and effective), confidence that those who provide the vaccine-related health services are reliable and competent, and trust that those who recommend vaccines (i.e., policymakers) have genuine, unadulterated motivations for doing so.14 This has some, but not complete, overlap with the Theory of Planned Behavior construct attitude and the Health Belief Model constructs 45 perceived benefits and perceived barriers. Some vaccine hesitancy items that fit into the category of confidence are: • “I am concerned about possible side effects of a COVID-19 vaccine.” • “I am concerned about having an allergic reaction to a COVID-19 vaccine.” • “I don't know if a COVID-19 vaccine will work.” • “I don't like vaccines.” / “I don't like vaccines generally.” • “I plan to wait and see if it is safe and may get it later.” • “I don't trust the government.” • “I don't trust COVID-19 vaccines.” • “I have a health condition and am concerned about the safety of the vaccine for people with my condition.” • “I am currently/planning to be pregnant and/or breastfeeding and do not want to get vaccinated at this time.” The second category, complacency, has significant overlap with the Health Belief Model constructs of perceived susceptibility and perceived severity, as well as having relations to a construct common to many health behavior models, self-efficacy. Complacency here refers to how necessary the vaccine is perceived to be; if perceived risks are low, vaccination may not be considered important, particularly compared to other behaviors and responsibilities.14 Some items that fit into this category are: • “I don't believe I need a COVID-19 vaccine.” • “My doctor has not recommended it.” 46 • “I think other people need it more than I do right now.” The third category, convenience, speaks to perceived and/or actual barriers or lack thereof. Some examples of factors influencing convenience include real and/or perceived availability, accessibility, cost, cultural context, comfort, and health literacy.14 This construct has commonalities with the previously mentioned constructs of perceived barriers from Health Belief Model and self-efficacy from several health behavior theories. A vaccine hesitancy item that fits with this category is “I am concerned about the cost of a COVID-19 vaccine.” In 2021, the 3 Cs model was expanded and tailored specifically to COVID-19 vaccine hesitancy, adding the additional behavioral factors 4) communication and 5) context.29 The communication component focuses on the prevalence of misinformation that can generate and amplify fear, anxiety, confusion, and distrust. The 2021 “5 Cs” article suggests that this be addressed with open and genuine dialogue and community engagement and social media accountability. The context component refers to sociodemographic characteristics, including but not limited to religion, ethnicity, and socioeconomic status. (This takes from and expands on the cultural context component of the convenience category in the 3 Cs model).14 The “5 Cs” article proposes that messaging be provided through trusted healthcare services and respected community leaders, with an emphasis on cultural competence.29 A vaccine hesitancy item from the U.S. CTIS that fits this description is “It is against my religious beliefs.” Methods The data analysis for this manuscript was generated using SAS® software, Version 9.4 of the SAS System for Windows (© 2016). 47 Data Source This research is based on The Delphi Group at Carnegie Mellon University U.S. COVID- 19 Trends and Impact Survey, in partnership with Facebook, hereafter referred to as the U.S. COVID-19 Trends and Impact Survey or U.S. CTIS.3 This was part of the largest public health survey ever conducted (CTIS), and the data has been used throughout the pandemic by the Institute of Health Metrics and Evaluation as part of their COVID prediction models, by Johns Hopkins University to study in-person schooling and household COVID-19 risk, by the Centers for Disease Control and Prevention to help inform their COVID-19 response, and by Nguyen et al. at University of Maryland to study vaccination, mask wearing, and vaccine hesitancy over time.92, 93 The U.S. COVID-19 Trends and Impact Survey includes 12 survey waves deployed from April 2020 to June 2022, with each wave implementing specific changes (including new items and answer options, reordering of items, and deleting out-of-date items and answer options as the COVID-19 situation evolved).92, 94 The data used in this research includes waves 8 and 10 (see Table 4.1), as these waves contained the vaccine hesitancy items of interest and consistently asked all of the exposure items of interest. (Note that while waves 11-13 technically included the exposure items of interest, participants were randomly assigned to one of two modules starting in wave 11. Two of the three key exposures (anxious and depressed feelings) were in one module and the other of the three key exposures (COVID-19 worry) was in the other module, so no participants who received these waves of the survey answered all three of the key exposure items, leading to the decision to restrict to the waves and times that had complete data on the critical variables.)94 48 Wave 11 was deployed as of May 20, 2021, but some respondents continued to receive and complete the Wave 10 version past this date.94 The analysis includes a time variable defined by survey responses completed March 10, 2021 and before (time 1) and March 11, 2021 and after (time 2). This was a noteworthy point in the timeline of COVID-10 in the United States, as it was 1) the one-year anniversary of the World Health Organization declaring a global Table 4.1. U.S. COVID-19 Trends and Impact Survey Wave Deployment Dates94 and Survey Completion Dates Wave Deployment Date Frequency Percent % 8 2021, Feb 8 474,829 38.33 10* 2021, Mar 2 764,102 61.67 Date survey completed Year Month 2021 Feb 445,251 35.94 Mar 473,056 38.18 Apr 207,284 16.73 May - Aug 113,340 9.15 Time stratification Time 1 up to Mar 10, 2021 639,577 51.62 Time 2 from Mar 11, 2021 599,354 48.38 * “9” was skipped to synchronize numbering with the University of Maryland international survey. pandemic, and 2) the date of the announcement by the U.S. presidential administration that all adults would soon be eligible for the vaccine and an announcement of plans to facilitate increased vaccine accessibility.95 These events are suspected to have had an impression upon individuals’ perception of COVID-19 and COVID-19 vaccines, which prompted the stratification (see Table 4.1). 49 Participants and Criteria for Selection The sampling frame included monthly Facebook users aged 18 years or older.5 For this study, the sample was restricted to those who answered the depression, anxiety, and COVID- worry items, indicated their intention to get vaccinated, and had FIPS data for the strata. (See Tables 4.2 and 4.3 for sociodemographics of the sample.) The decision was made to focus on unvaccinated participants because there would be fewer assumptions that would be difficult to substantiate. Including vaccinated participants would presume that 1) their levels of anxiety, depression, and worry were similar at the time of the survey as they were around the time they were deciding to get vaccinated, and 2) that none of them would have selected any of the specific vaccine hesitancy reasons had they been shown (vaccinated participants did not see these items). While those who answered “yes, definitely” to the vaccine intention question also did not see these items, these participants were still included because it was considered a less strenuous assumption that someone who at the time of the survey was definitely intending to be vaccinated was not vaccine hesitant, and also their recent feelings of anxiety, depression, and worry were measured at an appropriate time (compared to making these assumptions about someone who was vaccinated at some point in the past). A sensitivity analysis was conducted where these individuals were excluded from the analysis and this did not substantially change the results (see Table 4.8). In accordance with existing literature on this survey data,96 the decision was made to exclude observations selecting “prefer to self-describe” for gender (n=15,187, or ~1% of the sample) for the main analysis, as this was found to be closely associated with discriminatory answers and other responses suggesting that such observations did not represent a good faith effort to complete the survey (high prevalence of usually low-prevalence sociodemographics, 50 such as 22% indicating a doctorate degree for education and 29% indicating Hispanic and 28% indicating multiple/other non-Hispanic for race). In addition, participants who selected 10 or more of the 11 unique chronic condition items (n = 3,440, or less than 0.3% of the sample) were excluded for similar reasons (46% indicating 75+ years old for age, 59% indicating prefer to self- describe and 13% indicating non-binary for gender, 55% indicating a doctorate degree for education, and 57% indicating Hispanic and 23% indicating multiple/other non-Hispanic for race). Sensitivity analyses were conducted without these exclusions to assess the potential bias of excluding these observations (see Table 4.8). Variables of Interest and Potential Measurement Issues Outcome Variable The first outcome of interest is the binary variable vaccine hesitancy, split into “non- hesitant” and “vaccine hesitant.” The “Vaccine Hesitant = No” group includes those who had not yet received a COVID-19 vaccine but indicated “yes, definitely