Global Predictors of COVID-19 Vaccine Hesitancy: A Systematic Review

Background: vaccine hesitancy is defined as a delay in the acceptance or refusal of vaccination, even though immunisation is a determinant in reducing the mortality and morbidity associated with Coronavirus Disease 2019 (COVID-19). Aim: to identify and analyse the predictors of COVID-19 vaccine acceptance and/or hesitancy. Methods: a systematic review according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) criteria. Keywords: vaccine and (COVID or SARS) and (acceptance or acceptability or willingness or hesitancy or refusal) and (multivariate or regression) and (questionnaire or survey) and national. Databases/resources: PubMed, DOAJ, SciELO and b-on. Timeframe: March 2020–2022. Inclusion criteria: general population, questionnaire-based, calculation of a multivariate regression model and national studies. Quality assessment: application of the National Heart, Lung, and Blood institute (NHLBI) tool. Results: a total of 37 studies were selected, whose overall rate was fair. The most predominant predictors of vaccine hesitancy were a lower perceived risk of getting infected, a lower level of institutional trust, not being vaccinated against influenza, lower levels of perceived severity of COVID-19, or stronger beliefs that the vaccination would cause side effects or be unsafe. Discussion and conclusion: the identified predictors can be used to design tailored health policies and/or public health interventions, or to evaluate subjects’ vaccine hesitancy.

The COVID-19 pandemic also generated significant macroeconomic losses, such as a reduction in Gross Domestic Product (GDP) and increased microeconomic expenses across multiple countries (e.g., an increase of USD 2082.65 ± 345.04 to USD 2990.76 ± 545.98 for the patients admitted to an Intensive Care Unit (ICU) and negative financial outputs: job loss, the inability to meet financial obligations or essential needs, or using savings to meet financial obligations) [5,6].

PICOS Criteria Inclusion Criteria Exclusion Criteria
P: participants The general population (at least one country).
Studies specifically on specific subgroups (e.g., healthcare professionals or people with a certain disease, such as diabetes or asthma) were excluded.

I: intervention
Questionnaire-based studies, i.e., administration of a questionnaire/survey to collect participants' opinion/perception about COVID-19 vaccine hesitancy/acceptance.
Studies about other topics and/or that were not questionnaire-based.
C: comparison When applicable (e.g., studies carried out in more than one country).
O: outcomes To estimate the predictors of vaccine acceptance or vaccine hesitancy through a multivariate regression model.
Studies not estimating the predictors of vaccine acceptance or vaccine hesitancy through a multivariate regression model.

S: study design
Descriptive national studies, or descriptive studies enrolling at least 500 participants from the general population and carried out at a national level.
Studies enrolling less than 500 participants from the general population.

Followed Methodology per Defined Study Aim 2.2.1. Study Objectives 1 and 2
The study objectives (i) and (ii) were developed based on study findings (Appendix A). All data were collected in a MS Excel file, with quantification of the global occurrences/frequencies of predictors, identification of contradictory predictors and calculation of the predictors per region (subgroup analysis).

Study Objective 3
Two meta-analyses and systematic reviews on the same topic were identified in PubMed, b-on, SciELO, DOAJ and Cochrane Library on 2 May 2022. The keywords and search methodology were related to those applied in the present systematic review (see Section 2.3). The covered timeframe, the main findings and the conclusions of these reviews are presented in Table 2 [19,20].
The present systematic review was not classified as an update to a previous systematic review, based on the following definition: "an update asks a similar question with regard to the participants, intervention, comparisons, and outcomes (PICO) and has similar objectives; thus, it has similar inclusion criteria" [23]. Neither of the two previous similar/related systematic reviews on the same topic used the same inclusion or exclusion criteria as this study (see Tables 1 and 2) [19,20].
The most common predictors of vaccine acceptance were as follows: safety, efficacy, side effects, effectiveness and conspiracy beliefs (Asian countries); side effects, trust in vaccine and social influence (Europe) and information sufficiency, political roles and vaccine mandates (United States).
* is a bolean operator used in some browsers to automatically screen for related/derived words.

Keywords, Search Strategy, and Timeframe
The selected keywords were as follows: vaccine and (COVID or SARS) and (acceptance or acceptability or willingness or hesitancy or refusal) and (multivariate or regression) and (questionnaire or survey) and national. Singular keywords were used instead of plurals to cover the maximum number of related words (i.e., singulars and plurals). No filters were applied, except for PubMed, where the defined timeframe was previously defined. The keywords were conveniently selected based on the study objectives. For instance, "COVID or SARS" were used as keywords instead of "COVID-19 or SARS-CoV-2", with the aim of identifying more studies. As recommended, a broad and heterogeneous set of keywords were defined to increase the study precision and accuracy [22].
Particularly, the following stream of keywords: "vaccine and (COVID or SARS) and (acceptance or acceptability or willingness or hesitancy or refusal) and (multivariate or regression) and (questionnaire or survey) and national" was browsed in the main screening tool of each one of the selected databases/resources. Results were rechecked using the individual streams of keywords, such as "COVID and acceptance and multivariate and questionnaire and national" or "COVID and acceptance and regression and questionnaire and national". All possible combinations of keywords were searched.
The timeframe was defined as being between the beginning of the pandemic (11 March 2020) and March 2022 to identify the highest number of works/papers.

Screened Databases/Resources and Dates of Data Collection
Four databases/resources were screened (PubMed, DOAJ, SciELO and b-on), because they comprise a high number of journals [24][25][26][27]. B-on provides access to thousands of journals through diverse resources, such as Academic Search Complete, Current Contents (ISI) or Web of Science. The full list of b-on contents can be consulted at https://www.bon.pt/en/collections/ (accessed on 13 August 2022) [27]. SciELO was selected to cover Brazilian, Spanish, Latin American and Portuguese papers [26]. In general, the journals covered in these resources are peer-reviewed.

Data Collection Methodology and Quality Control Assessment
First, all abstracts were read. Studies were immediately excluded if they were not compliant with the inclusion criteria. Second, the preselected papers were downloaded, sequentially numbered and archived. Third, the preselected papers were fully read to Vaccines 2022, 10, 1349 5 of 39 recheck their compliance with the inclusion/exclusion criteria. Fourth, data were collected in a tabular format according to Appendix A, i.e., a standardised data extraction form was used.
All data were at least double-checked, considering that only one author carried out the present systematic review. However, errors in the identification and selection of papers were not expected, since the inclusion criteria were very simply and clearly elaborated. After printing, highlighting and reading all papers, the most relevant sections were directly copied and pasted to Appendix A (data collection). These copied and pasted sentences were rephrased, except for the study objectives. Study objectives were integrally copied and pasted, to maintain the original version. Duplicates were manually identified. When applicable, data were transported to an Excel file to compute/calculate descriptive statistics (e.g., sum and average number of participants or frequencies of predictors). The adopted methodological procedures were intended to safeguard the quality and validity of data extraction.

Quality Assessment of the Selected Papers
The impact factor from the Journal Citation Reports (JCR) and the quartile according to Scimago Journal & Country Rank of the selected papers are presented in Appendix A [28,29]. Two different metrics were conveniently defined to constitute a robust and diversified indicator of the journals' quality.
Additionally, a mapping analysis of the most impactful keywords (i.e., related and repeated keywords; the minimum considered number of occurrences of a keyword was two) of the selected papers was carried out to recheck/confirm the main topics covered by the selected papers. A mapping analysis of the authors of the selected papers was also carried out to find possible connections/collaborations between them. Both mapping analyses were carried out with VOSviewer version 1.6.18, which has been developed by Nees Jan van Eck and Ludo Waltman at Leiden University's Centre for Science and Technology Studies (CWTS) [30]. VOSviewer version 1.6.18 was downloaded and used by the study author to carry out the present mapping analysis in Lisbon, Portugal (July 2022).
The National Heart, Lung, and Blood institute (NHLBI) quality assessment tool for Observational Cohort and Cross-Sectional Studies was applied to assess/check the methodological quality (risk of bias) of the selected papers [31,32]. This tool was developed by the NHLBI in 2013 and is available for free at https://www.nhlbi.nih.gov/health-topics/ study-quality-assessment-tools (accessed on 13 August 2022). The tool was administered in accordance with the instructions displayed on the National Institutes of Health (NIH) website [32] by the study author in Lisbon, Portugal (July 2022); some scoring adaptations were created, as below described. The original questions of the NHLBI quality assessment tool are presented below (three types of replies are possible, yes, no and other, with "other" assuming the following options: CD, cannot determine; NA, not applicable; NR, not reported) [32].

1.
Was the research question or objective in this paper clearly stated? 2.
Was the study population clearly specified and defined? 3.
Was the participation rate of eligible persons at least 50%? 4.
Were all the subjects selected or recruited from the same or similar populations (including the same time period)? Were inclusion and exclusion criteria for being in the study prespecified and applied uniformly to all participants? 5.
Was a sample size justification, power description or variance and effect estimate provided? 6.
For the analyses in this paper, were the exposure(s) of interest measured prior to the outcome(s) being measured? * (No) 7.
Was the timeframe sufficient so that one could reasonably expect to see an association between exposure and outcome if it existed? * (No) 8.
For exposures that can vary in amount or level, did the study examine different levels of the exposure as it related to the outcome (e.g., categories of exposure, or exposure measured as a continuous variable)? ** (not applicable)  9.
Were the exposure measures (independent variables) clearly defined, valid, reliable and implemented consistently across all study participants? 10. Was the exposure(s) assessed more than once over time? (i.e., Was the questionnaire administered at least two times in different moments?/exposure or evaluation of independent variable). 11. Were the outcome measures (dependent variables) clearly defined, valid, reliable and implemented consistently across all study participants? 12. Were the outcome assessors blinded to the exposure status of participants? ** (not applicable) 13. Was loss to follow-up after baseline 20% or less? (i.e., % of participants who did not reply to the second questionnaire; if applicable) 14.
Were key potential confounding variables measured and adjusted statistically for their impact on the relationship between exposure(s) and outcome(s)?
According to the instructions of the NIH website, the exposure is the independent variable (i.e., willingness to get a COVID-19 vaccine), and the replies to questions 6 and 7 should be "no" for cross-sectional studies (*) [32].
Questions 8 ("If the exposure can be defined as a range (examples: drug dosage, amount of physical activity, amount of sodium consumed)") and 12 (Blinding means that outcome assessors did not know whether the participant was exposed or unexposed) were classified as not applicable to the selected studies (**) and were excluded. Questions 6 and 7 were also excluded, since they are not applicable to cross-sectional studies, and all evaluated studies (n = 37) are cross-sectional.
Scores were previously defined, as follows: 1 for yes; 0 for no; 0.5 for features not reported (i.e., if it is not possible to check if authors have carried out (or not) a certain evaluation/procedure); and 0.5 for the option "cannot determine", since published data were insufficient to calculate/check a certain feature.
The paired questions were as follows: questions 4.1(Were all the subjects selected or recruited from the same or similar populations (including the same time period)?) and 4.2 (Were inclusion and exclusion criteria for being in the study prespecified and applied uniformly to all participants?), and questions 10 (Was the exposure(s) assessed more than once over time?) and 13 (Was loss to follow-up after baseline 20% or less?); the scoring criteria were: 0.5 for yes; 0 for no; 0.25 for features not reported; and 0.25 for the option "cannot determine". Questions 4.1 and 4.2 were paired (analysed together) to respect the original version of the tool, and questions 10 and 13 were paired (analysed together), since they are inter-related.

Dates of the First Administered Vaccine per Each Studied Country
The dates of the first administered vaccine in each studied country/region can be consulted in Table 3. The presented dates were used to determine if the selected studies were carried out before or after the beginning of COVID-19 vaccination (see Appendix A).

Selected Studies
Globally, 37 papers were selected: 16 PubMed and 21 bon, 0 DOAJ and 0 SciELO papers (Figure 1). Vaccines 2022, 10, x FOR PEER REVIEW Figure 1. PRISMA 2020 flow diagram for new systematic reviews [55]: studies about the p of COVID-19 vaccine hesitancy/acceptance, which were specifically calculated through mu regression models.
The selected studies were published between 2020 and 2022:2020 (n = 7), 202 and 2022 (n = 7). The average number of respondents was at least 868,742 (SD = Figure 1. PRISMA 2020 flow diagram for new systematic reviews [55]: studies about the predictors of COVID-19 vaccine hesitancy/acceptance, which were specifically calculated through multivariate regression models.
Detailed information on the author, year, geographic region, screened database, paper metrics, study aim, sample size (number of participants that completed the study, i.e., valid respondents), main sociodemographic characteristics, methods, date of administration of questionnaire/survey, main findings and a brief discussion and conclusion are described in Appendix A.

Quality Control Assessment Paper Metrics and Mapping Analysis of the Journals from the Selected Papers
Looking at the journals of the 37 selected papers, only 5 journals did not have an impact factor JCR (Social Sciences, Population Medicine, Cureus and Journal of Pharmaceutical Policy and Practice) [93]. However, all the journals without an impact factor JCR were indexed in at least one of the following resources: PubMed, Web of Science or Scopus. Excluding the five journals without impact factors, the descriptive statistic of the impact factor JCR was: maximum 6.461; minimum = 2.221; average = 4.176; and SD = 1.098). The quartiles according to the Scimago Journal & Country Rank were as follows: Q1 (n = 29); Q2 (n = 4); Q3 (n = 1); and without attribution of a quartile (n = 3).
Regarding the keywords mapping analysis [30], from the 190 identified keywords (n = 37 selected papers), 52 keywords met the threshold (i.e., the minimum considered number of occurrences of a keyword was two) (Figure 2). maximum 459,235; and minimum 615). Logistic regression models were used in 28 (75.7%) out of 37 of the selected studies (Appendix A).
Detailed information on the author, year, geographic region, screened database, paper metrics, study aim, sample size (number of participants that completed the study, i.e., valid respondents), main sociodemographic characteristics, methods, date of administration of questionnaire/survey, main findings and a brief discussion and conclusion are described in Appendix A.

Quality Control Assessment
Paper Metrics and Mapping Analysis of the Journals from the Selected Papers Looking at the journals of the 37 selected papers, only 5 journals did not have an impact factor JCR (Social Sciences, Population Medicine, Cureus and Journal of Pharmaceutical Policy and Practice) [93]. However, all the journals without an impact factor JCR were indexed in at least one of the following resources: PubMed, Web of Science or Scopus. Excluding the five journals without impact factors, the descriptive statistic of the impact factor JCR was: maximum 6.461; minimum = 2.221; average = 4.176; and SD = 1.098). The quartiles according to the Scimago Journal & Country Rank were as follows: Q1 (n = 29); Q2 (n = 4); Q3 (n = 1); and without attribution of a quartile (n = 3).
Regarding the keywords mapping analysis [30], from the 190 identified keywords (n = 37 selected papers), 52 keywords met the threshold (i.e., the minimum considered number of occurrences of a keyword was two) ( Figure 2).
Overall, 233 authors were involved in the 37 selected papers, of which only 19 authors were connected to each other ( Figure 3) (mapping analysis of authors).

NHLBI Quality Assessment Tool
The global rating of the 37 selected studies was 82.1% (fair) (273.5 out of 333 points) ( It is important to note that excluding two questions-(i) Was the exposure(s) assessed more than once over time? (i.e., Was the questionnaire administered at least two times in different moments?) and (ii) Was loss to follow-up after baseline 20% or less? (e.g., % of participants who did not reply to the second questionnaire)-changed the overall rate of the 37 selected studies to good (91.2%). Additionally, the participation rate was not reported in 20 of the 37 studies, which compromised the evaluation of the item: participation rate of eligible persons (at least 50%). Was the exposure(s) assessed more than once over time? (i.e., Was the questionnaire administered at least two times in different moments?/exposure or evaluation of independent variable) 2 18.5 10.8 *

NHLBI Quality Assessment Tool
The global rating of the 37 selected studies was 82.1% (fair) (273.5 out of 333 points) (  Were key potential confounding variables measured and adjusted statistically for their impact on the relationship between exposure(s) and outcome(s)? 37 37 100 Was loss to follow-up after baseline 20% or less? 37 37 100 * Poor scores (below 80%); ** the participation rate was not reported in 20 out of the 37 studies.
It is important to note that excluding two questions-(i) Was the exposure(s) assessed more than once over time? (i.e., Was the questionnaire administered at least two times in different moments?) and (ii) Was loss to follow-up after baseline 20% or less? (e.g., % of participants who did not reply to the second questionnaire)-changed the overall rate of the 37 selected studies to good (91.2%). Additionally, the participation rate was not reported in 20 of the 37 studies, which compromised the evaluation of the item: participation rate of eligible persons (at least 50%).
In general, COVID-19 vaccination started in December 2020 or January 2021 in developed countries. The exceptions regarding the selected studies were as follows: February 2021 (Japan and South Korea); March 2021 (Lebanon, South Africa and Trinidad and Tobago) and May 2021 (Pakistan) ( Table 3). The findings from the 37 selected studies were analysed together, since almost all the selected studies were carried out before the beginning of the COVID-19 vaccination period in the country/region (n = 25), and the remaining studies started a few months after the beginning of the vaccination campaigns. Additionally, data from COVID-19-vaccinated individuals were only collected in one study (Hao et al., 2022) [65]; i.e., vaccine hesitancy was quantified through the intention or likelihood of getting a COVID-19 vaccine in the remaining selected studies. Finally, the question(s) and/or scale(s) used to quantify the intention to get a COVID-19 vaccine were not equal/standardised between the 37 selected studies. Thus, it was not possible to directly compare the findings from the studies carried out before or after the beginning of the COVID-19 vaccination period campaigns, and/or involving both periods.

Previous Similar/Related Systematic Reviews: Commonly Selected Studies
Globally, only 7 (18.9%) of the 37 selected studies were common between the present systematic review and the two previous similar/related systematic reviews [19,20]. The common studies were as follows:
These findings confirm the relevance of the present work since most of the selected studies were not selected/identified in the two previous similar systematic reviews. The  [19,20], which supports the validity of the present systematic review.

Global Predictors of Vaccine Hesitancy
The global predictors of vaccine hesitancy (i.e., predictors identified in at least two studies) are presented in Table 5. The identification of the global predictors of vaccine hesitancy was based on the findings of the multivariate models from the 37 selected studies. Higher education/tertiary education (if spending more time using social media)/postgraduate 3 8.1 Unemployed/being self-employed, unemployed or unable to work due to a long-term illness or disability 3 8.1 Lower level of worry regarding health risks/complacency in health/not declaring to be concerned about one's own health and the health of next of kin 3 8.1 Lower general vaccine knowledge/lower general vaccine knowledge index 2 5.4 People who do not have a positive view of the COVID-19 vaccine/lack of confidence in the COVID-19 vaccine 2 5.4 * Similar/equal predictors variables were aggregated, but the original designations were maintained to ensure study precision; ** number of occurrences in the 37 studies; *** 100% (n = 37).

Predictors of Vaccine Hesitancy Identified in Just One Study (Single Occurrences)
Predictors of vaccine hesitancy identified in just one study (2.7% of the 37 selected studies) were as follows: student, retired, age (no additional information), unstable job status, higher income, no religious affiliation, lower socioeconomic status and working in the informal sector, experienced worsening health status during pandemic, not internationally traveling in 2020, less trust in science, declaring to not know about the non-natural origin of the virus, not believing in the seriousness of the COVID-19 situation, less or no fear of COVID-19, infrequent users of traditional media, relying upon social media for virus information/people who report social media as their main source of news, people who use social media for 3 or more hours daily, less positive general COVID-19 vaccination beliefs and attitudes, fewer beliefs about social acceptability of a COVID-19 vaccine, a lower perceived need for vaccination, lower proportion of their family or close friends have already received it, opinion of healthcare provider/if respondents thought their healthcare provider would not recommend vaccination, lower cues to action (triggers for receiving COVID-19 vaccination by four items, including recommendations by the government, physicians, family members and friends, respectively), lower knowledge scores for SARS-CoV-2 infection, assigning importance to the vaccine's country of origin, living in a certain region, residing in cities, no family member died or admitted to ICU for COVID-19, having already tested positive for COVID-19, not having hypertension, having a food allergy, not being available to pay these vaccines, not being prosocial, the belief that the government restrictions were too lenient, the frequency of socialising prior to the pandemic (goes out frequently), shorter duration of action (less than 12 or 18 months), number of shots (more than one injection), production place (non-imported vaccines) and higher price.

Contradictory Predictors of Vaccine Hesitancy
The contradictory predictors of vaccine hesitancy (i.e., contrary explanatory variables/predictors of vaccine hesitancy/acceptance) are presented next.

Lower Level of Education vs. Higher Level of Education
A lower level of education was a predictor of vaccine hesitancy (n = seven), as determined by studies from multiple countries (one study) [58], the USA (one study) [60], the UK (one study) [70], Croatia (one study) [88], Latin America (two studies) [89,90] and South Africa (one study) [92]. A higher level of education was a predictor of vaccine hesitancy (n = three), as determined by studies from the European Union (one study) [57]; Australia (one study) [74]; and the United Arab Emirates (Asia) (one study) [76].

Not Residing in Cities or Living in Smaller Settlements vs. Residing in Cities
Not residing in cities or living in smaller settlements was a predictor of vaccine hesitancy (n = five), as determined by studies involving multiple countries (two studies) and studies from [57,58] Lebanon (Asia) (one study) [85]; Croatia (European Union) (one study) [88]; and Brazil (Latin America) [90]. Residing in cities was a predictor of vaccine hesitancy in one study (China) [69].

Unemployed vs. Employed
Being unemployed was a predictor of vaccine hesitancy (n = three), as determined by studies involving multiple countries (two studies) [57,58] and a study from the United Arab Emirates (Asia) (one study) [76]. Being employed was a predictor of vaccine hesitancy (n = two), as determined by studies from the USA (one study) [60] and Jordan (Asia) (one study) [80].

Most Frequent Predictors of Vaccine Hesitancy per Country or Region
The most frequent predictors of vaccine hesitancy per country or region (i.e., predictors/variables identified in at least 35% of studies) are presented in Table 6. The two studies reporting results from different regions were not included in the analysis of Table 6 [56,58].

Discussions
Globally, the selected studies covered a broad number of countries/regions (USA, China, UK, Australia, Asiatic countries, European Union, Latin America and the Caribbean and South Africa), and accounted for the participation of at least 868,742 respondents, although around half of the participants were from the USA. Logistic regression models were adopted in almost all studies, supporting an adequate comparison of study results and a precise calculation of variables/predictors. In general, only around half to two-thirds of the respondents of the selected studies declared themselves to be willing to get a COVID-19 vaccine, with higher declaration rates of acceptance in China, Australia, Indonesia and Brazil at >80% [67,68,73,78,90], and lower declaration rates of acceptance in Jordan, Israel, Qatar and Hong Kong at <45% [80][81][82]84], which confirms that vaccine hesitancy remains a global public health and economic problem, implying a direct negative impact on the mitigation of COVID-19 and on the control of mortality and morbidity. The ideal universal goal of COVID-19 vaccine coverage is at least 70%, which is especially relevant for the high-priority groups (e.g., geriatric patients) [4,94].
In general, the journals of the selected papers presented a high impact factor JCR, and the majority were from quartile one (Q1) according to the Scimago Journal & Country Rank, which seems to support the quality, validity and relevance of the selected studies. The quality of the selected papers was also confirmed by application of the NHLBI tool [32]. Additionally, the mapping analysis of keywords carried out with VOSviewer confirmed an inter-relationship between the selected keywords within the scope/aims of the present systematic review, supporting the confirmation of the study validity. In contrast, the mapping analysis of authors carried out with VOSviewer confirmed a limited number of connections (only 19 authors), which seems to indicate the need for more international collaborations, regarding the present topic [30]. Overall, the evaluated journal metrics and the mapping analysis carried out with VOSviewer were classified as external quality indicators of the selected studies.
Thus, the findings of the present systematic review seem to represent a valid and relevant contribution to understanding and comprehending the most predominant factors of COVID-19 vaccine hesitancy at a global level, which remains a hot topic.

Global Predictors of Vaccine Hesitancy
Considering that the contradictory predictors of vaccine hesitancy (e.g., males vs. females) are likely to be influenced by the social, political and/or cultural characteristics of a particular country, the global predictors of vaccine hesitancy were discussed once these were excluded.
The top five most predominant predictors of vaccine hesitancy (after exclusion of the contradictory predictors) were: lower perceived risk of getting infected, a lower level of institutional trust, not being vaccinated against influenza, lower levels of perceived severity of COVID-19 infection and stronger beliefs that the vaccination would cause side effects or be unsafe. Importantly, three out of these five predictors were also identified as relevant factors in the two previous related/similar systematic reviews and meta-analyses by  [19,20], which also supports the accuracy of the present systematic review.
However, other relevant factors have been identified in the present systematic review (i.e., lower perceived risk of getting infected and lower levels of perceived severity of COVID-19 infection), which may raise the awareness of other relevant explanatory variables/predictors of vaccine hesitancy in future research. For instance, belonging to a certain political party and having lower levels of perceived effectiveness/efficacy of a COVID-19 vaccine were also relevant predictors of vaccine hesitancy (after excluding the contradictory predictors). These findings are of the utmost relevance, since vaccination against influenza is not frequent in developing countries. Additionally, some contradictory predictors of vaccine hesitancy, such as belonging to a minority group or having a low income, may be more relevant for developing countries than for the developed ones. Overall, the number of studies carried out in developing countries was very limited. Thus, the top predictors are more likely to be explanatory variables of vaccine hesitancy in developing countries.
The subjects' misperceptions, distress and scepticism can lead to COVID-19 vaccine hesitancy. Pervasive misinformation on COVID-19 (e.g., fake news or the spread of imprecise or false information about COVID-19 on social media, such as Facebook or other networks) is likely to explain some of the most relevant predictors of COVID-19 vaccine hesitancy identified here [95][96][97][98].
Additionally, the top five predictors identified here are covered in previous explanatory models of vaccine hesitancy (e.g., the 3C model and/or the 2017 increasing vaccination model) [14,16]. In general, these top five predictors are mainly related to the subjects' opinions/perceptions, except for not being previously vaccinated against influenza (or being less willing to get a flu vaccine). Thus, health professionals or social workers should understand subjects' opinions/perceptions about COVID-19 risks, severity and transmission, as well as the safety issues of COVID-19 vaccines, to clarify and explain potential misinterpretations. The intervention and/or consultation of health professionals was previously identified to aid in avoiding subjects' vaccine hesitancy [56,99].
Other frequent predictors of vaccine hesitancy were being Republican or Conservative; having lower levels of perceived effectiveness/efficacy of COVID-19 vaccines; being single or non-married; having children at home; accepting vaccine conspiracies; having a lower level of worry about health risks; not shielding; being less likely to wear a mask; or having no family or friends with a previous COVID-19 infection (predictors reported in three or more studies). The preference for a certain political party was predominantly reported in the USA [59][60][61][62]64,65], although a study from South Korea also identified this predictor [86]. Thus, political parties/leaders should clarify the importance of vaccination to avoid the refusal of vaccines by their affiliates or sympathisers. Single or non-married people may be more worried about their health state since they live alone. People with children at home may believe that they will get COVID-19 through their children and therefore do not need to be vaccinated. People who accept vaccine conspiracies are less likely to trust in vaccines. A lower level of worry about health risks, not shielding and being less likely to wear a mask seem to indirectly support a lower perception of the potential risks of COVID-19, while the lack of COVID-19 infections in family or friends also seems to minimise the perceptions of the potential risks of this virus. However, these hypotheses should be demonstrated in future social studies by region/country.

Single Occurrences
Single occurrences of predictors of vaccine hesitancy (i.e., predictors only reported/identified in one study) were very diverse/heterogeneous. These predictors should be evaluated in future research. It was not possible to check whether these variables have been tested in the construction of the administered questionnaires, since methodologies around the design, development or validation of questionnaires/surveys were incompletely described in the 37 selected studies. Thus, research methodologies relating to the development and validation of study questionnaires/surveys (i.e., collection tools) should be reported in more detail in future studies.

Contradictory Predictors of Vaccine Hesitancy
Some of the contradictory predictors (i.e., contrary/opposite variables/predictors, which explain vaccine hesitancy in at least two countries/regions, such as females vs. males, or minorities vs. major ethnic groups) can be classified as potentially unexpected, as follows: • Older people are more vaccine-hesitant than younger people, since the former are expected to be more susceptible to COVID-19 infection; • Subjects with a higher level of education are more vaccine-hesitant than less educated subjects, since better-educated people are expected to be more prepared to understand the relevance of immunisation; • Employed subjects are more vaccine-hesitant than unemployed subjects, since employed people are theoretically more likely to have more public contact; • Citizens living in cities are more vaccine-hesitant than those living in rural areas or in smaller settlements (with a lower population density), since personal contact is less likely in rural areas.
These contradictory predictors of vaccine hesitancy are likely to be explained by political and/or cultural differences between countries/regions, although more studies are required to confirm this. The identification of contradictory predictors confirms the need to check the most prevalent explanatory variables of vaccine hesitancy per country/region, such as an eventual universal tool to determine subjects' vaccine hesitancy based on the top five predictors, as the constitution of universal predictors of vaccine hesitancy is necessarily related to some limitations.

Most Frequent Predictors of Vaccine Hesitancy per Country or Region
The most frequent predictors of vaccine hesitancy per country or region were variable (Table 6). For instance, the impact of belonging to a certain political party (i.e., parties other than the Democratic Party) on subjects' vaccine hesitancy was predominately identified as an explanatory variable in the USA [59][60][61][62]64,65]. Thus, political leaders, affiliates and/or collaborators from parties other than the Democratic Party should ideally explain the relevance of immunisation to mitigate and control the COVID-19 pandemic in the USA. In opposition to the political preference of subjects, being a female was the most frequent predictor of vaccine hesitancy in almost all other reported regions/countries [60][61][62]64,67,69,[74][75][76][77][80][81][82][83]88,89], which reinforces the importance of understanding females' motivations about COVID-19 vaccination. Women are likely to benefit from health professional counselling and training on COVID-19 immunisation.
Younger people are expected to be more vaccine-hesitant than older individuals, since COVID-19 is less likely to be severe in the first group, although this variable was only identified as a frequent predictor of vaccine hesitancy in some countries/regions (e.g., the USA, some Asiatic countries and South Africa) [58,62,63,65,69,71,73,[79][80][81]86,88,92]. Thus, younger people should be encouraged to get the COVID-19 vaccine to facilitate achieving herd immunity, mitigating the pandemic, controlling mortality, lethality and/or morbidity due to COVID-19. Contrary to expectation, older age was identified as one of the most frequent predictors of vaccine hesitancy in two studies from Latin America and/or the Caribbean, which is a concrete example of social and cultural variations between countries [89,90]. People from older groups can be less health-literate or educated in some countries (e.g., Mexico and Brazil), which may explain their higher levels of vaccine hesitancy in these nations.
Minorities and lower income were reported as the most explanatory variables of vaccine hesitancy for both the USA and South Africa [59][60][61][62][63][64]92], while a lower level of education was identified as one of the most frequent predictors of vaccine hesitancy in Latin America and South Africa [89,90,92]. These variables seem to be correlated: people from minority groups usually earn lower salaries, are less favoured and less educated than the general population. People from minority groups can also have language and/or health literacy limitations, which can limit their understanding of COVID-19 vaccination. These groups could also benefit from tailored vaccination campaigns and/or more proactive healthcare interventions (e.g., physicians, nurses, pharmacists, social workers or others) [99].
Subjects' perceptions about COVID-19 were only reported as the most predominant predictors of vaccine hesitancy in some countries (e.g., a perceived low risk of COVID-19 infection was identified in China and South Africa, and a lower perception of the severity of having COVID-19 was identified in Asiatic countries) [67,69,76,80,83,84,92]. Perceptions about COVID-19 vaccines were only reported as the most predominant predictor in three countries (perceived safety risks of COVID-19 vaccines in China and the UK, and perceived low efficacy of COVID-19 vaccines in China and South Africa) [66][67][68]92]. In this sense, subjects' perceptions about COVID-19 vaccines and/or disease seem to be more relevant in some countries than in others as explanatory factors of subjects' vaccine hesitancy. Health professionals should evaluate patients' perceptions of COVID-19 and COVID-19 vaccination during consultations to help in the immunisation decision-making process [56].
The adequacy, perception and/or awareness of COVID-19 information (e.g., COVID-19 vaccination) were identified as the most frequent predictors of vaccine hesitancy in the UK and South Africa [71,72,92], which reinforces the need to develop clear and comprehensible written and oral health materials on COVID-19-related topics. This type of health material should be evaluated through usability tests, with the involvement of citizens from different sociodemographic backgrounds.
A lack of trust of public authorities/government or health systems (e.g., healthcare professionals) was reported as the most predominant predictor of vaccine hesitancy in Asiatic countries and in some European Union member states [76,79,83,84,87,88]. Thus, studies to understand the reasons why citizens do not trust in health institutions or healthcare professionals are highly recommended. The most advanced health systems in the world are among the member states of the European Union, in contrast to some Asiatic countries, although citizens motivations were not evaluated.
Not being vaccinated against influenza (UK) [71,72], older people (Latin America) [89,90], people with children at home (some European Union member states) [57,88], not having a partner (South Africa) [92] and not living in cities or living in city suburbs/smaller settlements (some European Union member states) [57,88] were examples of individual variables identified as the most relevant predictors of vaccine hesitancy in just one country/region, which strengthens the relevance of investigating potential social discrepancies within countries/regions. These findings support the development and implementation of studies on the present topic for individual countries/regions aiming to define tailored political health measures, since socio-demographic predictors are likely to vary with time.
According to the findings of the systematic review of Roy et al. (2022) [20], the most common predictors of vaccine hesitancy per country/region were: safety, efficacy, side effects, effectiveness and conspiracy beliefs (Asian countries); side effects, trust and social influence (Europe); and information sufficiency, political roles and vaccine mandates (United States). Globally, only two factors were common between the systematic review of Roy et al. (2022) [20] and the present systematic review: one in member states of the European Union (trust) and one in the USA (political roles), which confirms the likely variation in explanatory factors with time and/or an eventual lack of precision or sensitivity in the implemented studies in some countries/regions (e.g., potential limitations on the development and/or validation of study questionnaires/surveys about vaccine hesitancy/acceptance).

Study Strengths and Contribution to the State-of-the-Art on Vaccine Hesitancy
The selection criteria of the present systematic review were based on studies that use multivariate logistic regression models, different regions/countries and large samples (at least 868,742 subjects participated in the selected studies). Thus, the predictors of vaccine hesitancy were likely to be precisely and accurately determined in the present work. The same or similar selection criteria were not applied in previous similar or related systematic reviews. Considering that the predictors of vaccine hesitancy are expected to longitudinally vary within the same country or between different countries, health authorities are required to regularly evaluate vaccine hesitancy, especially in countries with high rates of vaccine hesitancy.
Moreover, the contradictory predictors are exhaustively explained, and a detailed list of predictors of vaccine hesitancy is presented. These findings will support the development of new international studies on vaccine hesitancy and evaluation tools. In opposition to the published studies, both the most and least relevant predictors of vaccine hesitancy are described and explained. The methodologies and questionnaires of the selected studies are heterogeneously defined. Therefore, the present systematic review will contribute to strengthening and improving the methods of future research on vaccine hesitancy.
Finally, at least two out of the five most relevant predictors (i.e., the participants' lower perceived risk of getting infected and the lower perceived severity of COVID-19 infection) were not previously identified as the most relevant, reinforcing the need to evaluate subjects' perceptions of COVID-19 risks. The fact that three out of the five most relevant predictors of vaccine hesitancy were common between the present systematic review and the two previous similar/related systematic reviews (i.e., a lower level of institutional trust, not being vaccinated against influenza and stronger beliefs that the vaccination would cause side effects or be unsafe) seems to be an external validity indicator of the quality of the present systematic review.
Overall, the findings of the present systematic review clearly contribute to enhancing the state of the art on vaccine hesitancy.

Risk of Biases and Limitations of the Present Systematic Review
There is ultimately a risk of selection bias, since more screened databases (e.g., Scopus and Embase) and keywords could have been used, and a risk of interpretation bias, since the impact of vaccination campaigns on subjects' opinions/perceptions of vaccine acceptance/hesitancy have not been specifically evaluated (e.g., the findings from studies that started after and/or before the beginning of COVID-19 vaccination were analysed together).
There is risk of data collection bias, as data were only collected, checked and rechecked by just one author, which is not in line with the 2020 PRISMA recommendations [22]. However, all data were at least double-checked, and an external validation was carried out, given that three out of five of the most relevant predictors of vaccine hesitancy were equal/common with those identified in two previous similar/related systematic reviews [19,20]. This achievement seems to confirm the study accuracy and validity. Additionally, other methodological procedures were adopted to avoid data collection imprecisions, such as the adopted data collection methodologies. For instance, study objectives were purposively not rephrased to avoid any inaccuracy (Appendix A), although the remaining information was rephrased to avoid plagiarism.
Moreover, automation tools, such as SRDR, Distiller SR or Dedoose, were not used to automatically collect and process information [100][101][102]. To avoid potential bias in the analyses, more authors or automatic tools could have been involved in the design and execution of the present systematic review (e.g., to reconcile divergent points of view and to achieve consensus).
Subgroups analysis and sensitive analysis are not the same, although both contribute to study quality and validity. Besides the evaluation of the most frequent predictors of vaccine hesitancy per country or region and the evaluation of study findings after excluding contradictory predictors of vaccine hesitancy (subgroups analysis), a specific sensitivity analysis was not carried out in the present systematic review. Considering that the global rating of the NHLBI quality assessment tool was 82.1% (fair), an eventual exclusion of specific studies/findings based on the risk of bias did not seem to be relevant/applicable.
Additionally, the definition of quantitative thresholds or cut-off values is not possible, since qualitative predictors of vaccine hesitancy were calculated based on a multivariate regression model in the present systematic review. For instance, specific thresholds can be defined for certain biological parameters (e.g., glycaemia) in sensitive analysis of clinical studies.
A possible sensitivity analysis could have been based on the exclusion of studies applying different data collection tools (e.g., surveys using different questions or scales) or utilising different types of administration (e.g., self-administered, presential, online questionnaires, etc.). However, this evaluation was not performed, because the questionnaires from the selected studies were very heterogeneously designed (e.g., different questions, scales/scoring or administration types). Moreover, questionnaires were insufficiently described in some of the selected studies.
A protocol was not previously registered, such as in the Open Science Framework (OSF) repository [103]. However, the initial protocol has not been changed, all methodological details are described here, and the present systematic review does not involve clinical trials or epidemiological studies of medicines (i.e., there are no ethical issues).

Limitations and Risk of Biases of the Selected Studies
The selected studies were heterogeneously designed. For instance, the evaluated variables/collected data were variable between different studies (e.g., subjects' perceptions of COVID-19 or opinion of COVID-19 vaccines were only evaluated in some studies).
The type of collected data and questionnaires were variable between the selected studies, including different collected topics and the order of questions. The different order of questions may have introduced question order bias [104,105]. Thus, some studies may not be precise or sensitive. Studies on the present topic were only developed in some countries/regions (i.e., USA, China, UK, Australia, some Asiatic countries, the European Union, three countries from Latin America and the Caribbean and South Africa). The potential disadvantages of COVID-19 vaccination were not discussed in detail (e.g., the impact of adverse drug reactions on patients' health, limited efficacy and/or number of required shots). The questionnaire/survey construction and validation should be clearly described in the methods, since questionnaire-based methodologies were, in general, insufficiently reported in the selected studies.
Additionally, studies based on web-based surveys may have introduced bias into the results, since the profile and sociodemographic characteristics of the non-participants/respondents may have been imprecisely collected (e.g., data collection in social networks). Studies applying triangulation methodologies were limited (e.g., administration of additional tools, such as health literacy, numeracy or cognitive evaluation tools, tools to evaluate subjects' quality of life, social life, satisfaction and/or wellbeing (before and after the beginning of the pandemic), evaluation of knowledge about COVID-19 or vaccines, perception evaluations or simulations of decisions based on imaginary scenarios). Some variables were only quantified in some studies, such as the number/proportion of vaccinated individuals, perception of the safety and efficacy of COVID-19 vaccines and opinions about the severity of COVID-19, among other factors/variables. The sociodemographic variables were not distributed in a balanced way in all selected studies (e.g., the proportion of females/males or education level), although multivariate models normalise the contribution of individual variables in study findings. Longitudinal studies were lacking in most countries.
The selected studies were carried out in three timeframes (before and after the beginning of COVID-19 vaccination and/or involving both timeframes), which may have influenced the study findings. Moreover, when comparing the studies carried out before (first group) and after (second group) the beginning of COVID-19 vaccination, the willingness to get a COVID-19 vaccination was above 50% in all studies in the second group. Importantly, few studies from the first group reported a willingness to get a COVID-19 vaccine below 50% [80,84] (Appendix A). In this sense, COVID-19 hesitancy may have tended to decrease following the start of COVID-19 vaccination. However, data are not conclusive, since studies were generally only carried out in just one period, and there was a discrepancy between the type of questionnaires/collection tools used in the selected studies. The present findings reinforce the need to design and implement more prospective and/or retrospective studies on COVID-19 vaccine hesitancy.
According to the findings of the NHLBI tool, a selection bias may have occurred in some of the selected studies, since the participation rate and the profile of non-respondents (e.g., sociodemographic characteristics) were not reported in most studies. Information about study representativeness, power description or variance was only provided in around half of the selected studies, although at least 500 participants were enrolled. On the other hand, authors of the selected studies may have calculated the sample size based on power description or variance but may have opted not to publish this information. As sample size calculations were not reported in many studies, it is not possible to exclude the occurrence of study bias because of imprecise sample calculations. In general, questionnaires were only administered once. Thus, participants' perceptions on the willingness to get a COVID-19 vaccines were not checked/followed during a certain timeframe in almost all selected studies.

Future Research and Purposed Intervention Measures
The present research should involve more countries/regions. More studies and international collaborations are recommended to understand the contradictory predictors of vaccine hesitancy (e.g., females vs. males or minorities vs. major ethnic groups). An international tool to evaluate vaccine hesitancy could be developed and validated, based on the most frequent predictors of vaccine hesitancy reported in the present systematic review, for instance. High-quality health-related information about COVID-19 vaccines and immunisation should be developed and evaluated through usability tests in groups of participants from different sociodemographic backgrounds. Social studies are recommended to understand why citizens do not trust public authorities/government or health systems, or why they believe in conspiracy theories. Health authorities should develop detailed guidelines about the development and validation of data collection tools to characterise subjects' vaccine hesitancy (e.g., surveys or questionnaires), preferably per country/region. Sensitivity analyses are recommended in future research.
Health communication strategies seem to be a factor in understanding and reducing vaccine hesitancy. For instance, mass media, apps, video games and/or social networks can divulge high-quality information on the risk of being infected with COVID-19, the severity of COVID-19 infection, or the safety of COVID-19 vaccines, since these topics are associated with the most relevant predictors of vaccine hesitancy.
Additionally, health professionals (physicians, pharmacists, or nurses) can try to understand the subjects' perceived risks of becoming infected with COVID-19, institutional trust, influenza vaccination state (vaccinated against influenza or not), perceived severity of COVID-19 infection or opinions about the safety of COVID-19 vaccines, in order to understand the motives behind vaccine hesitancy. If applicable, health professionals should clarify subjects' doubts and inaccurate perceptions/opinions. In general, health professionals' counselling and intervention can effectively contribute to mitigating subjects' vaccine hesitancy [99]. National information campaigns are also recommended, covering all community pharmacies and/or health centres for instance, since health professionals are prepared to clarify citizens' doubts. Ideally, health professionals should receive training on communication strategies to effectively convey all health messages. These health messages should be adapted/tailored according to the level of the subjects' health literacy.

Conclusions
Globally, 37 questionnaire-based studies on the predictors of vaccine hesitancy were selected, which were estimated through multivariate regression models. The selected studies covered a significant number of countries/regions and respondents. These studies complied with satisfactory internal quality measures (e.g., the outputs of the NHLBI assessment tool) and external quality standards (e.g., bibliographic metrics or mapping analysis), which seems to confirm the relevance/quality of the present systematic review. However, the implementation of similar studies is still lacking in many countries at a global level.
The five most predominant predictors of vaccine hesitancy (after exclusion of contradictory predictors) were: a lower perceived risk of getting infected, lower levels of institutional trust, not being vaccinated against influenza, a lower perceived severity of COVID-19 infection and stronger beliefs that the vaccination would cause side effects or be unsafe. These findings were accurately determined through the application of multivariate regression models, which support their possible application in international research (e.g., the validation of a future tool to quickly detect COVID-19-vaccine-hesitant subjects). These top five predictors can also be used to design tailored health policies and/or public health interventions.
Diverse contradictory predictors have been identified (i.e., contrary variables/predictors, which explain vaccine hesitancy in at least two countries/regions, such as females vs. males, or minorities vs. major ethnic groups). Additionally, the profile of the most frequent predictors of vaccine hesitancy also varied between analysed countries/regions. Globally, these findings confirm the likely existence of social cultural differences between countries/regions and the need to check the profile of these predictors per country/region. However, while two previously identified similar/related systematic reviews did not apply the same inclusion/exclusion criteria (e.g., the obligatory calculation of multivariate logistic regression modes to estimate the predictors of vaccine hesitancy/acceptance), they reported some common frequent predictors of vaccine hesitancy [19,20]. These findings seem to confirm the validity and relevance of the present systematic review and, indirectly, the accuracy of the implemented research. A lower perceived risk of getting infected and a lower perceived severity of COVID-19 infection were among the most frequent explanatory predictors of vaccine hesitancy, especially in the present systematic review.
The selected studies covered a broad number of countries/regions (i.e., USA, China, UK, Australia, Asiatic countries, member states of the European Union, Latin America and the Caribbean and South Africa) and enrolled a significant number of participants, which strengthens the relevance of the study conclusions.

Conflicts of Interest:
The author declare no conflict of interest. Explanatory variables of vaccine hesitancy: lower efficacy (less than 70% or 90%); shorter duration of action (less than 12 or 18 months); more adverse events, number of shots (more than one injection); production place (non-imported vaccines); and price (higher price). 65% were willing to vaccinate, with 27% being in the "maybe" category. Explanatory variables of vaccine hesitancy (respondents were more likely to be in the COVID-19 vaccine "maybe" than "yes" group): perceived COVID-19 to be less severe; had less trust in science; were less willing to get a flu vaccine and were female. Explanatory variables of vaccine hesitancy (respondents were more likely to be in the COVID-19 vaccine "no" than "maybe" group): they did perceive COVID-19 to be a hoax * Only some of the collected sociodemographic variables are reported here due to space limitations; ** studies that not exclusively evaluate predictors of vaccine acceptance and/or hesitancy.