Worldwide Vaccination Willingness for COVID-19: A Systematic Review and Meta-Analysis

Countries across the globe are currently experiencing a third or fourth wave of SARS-CoV-2 infections; therefore, the need for effective vaccination campaigns is higher than ever. However, effectiveness of these campaigns in disease reduction is highly dependent on vaccination uptake and coverage in susceptible populations. Therefore, this systematic review and meta-analysis estimated the vaccination intention and identified determinants of willingness and hesitancy. This study updates the existing body of literature on vaccination willingness, and was conducted according to the PRISMA guidelines. PubMed was searched for publications, selecting only studies published between 20 October 2020 and 1 March 2021, in English, with participants aged >16 years of age. The search identified 411 articles, of which 63 surveys were included that accounted for more than 30 countries worldwide. The global COVID-19 vaccination willingness was estimated at 66.01% [95% CI: 60.76–70.89% I2 = 99.4% [99.3%; 99.4%]; τ2 = 0.83]. The vaccination willingness varied within as well as between countries. Age, gender, education, attitudes and perceptions about vaccines were most frequently observed to be significantly associated with vaccine acceptance or refusal.


Introduction
Vaccine hesitancy is defined as the "delay in acceptance or refusal of vaccination despite availability of vaccination services" [1]. Rises in cases of vaccine-preventable diseases have been related to declines in vaccine coverage. A recent narrative review showed that the majority of measles outbreaks in the United States and Canada were associated with unvaccinated individuals [2,3]. In 2019, the World Health Organization (WHO) declared vaccine hesitancy as amongst the top ten threats to global health [4]. A previous systematic review studying the global perspectives on hesitancy from 2014 found that determinants of, and factors relating to, hesitancy are highly context-specific; varying across place, time, as well as the type of vaccine [5].
Meanwhile, the contemporary COVID-19 pandemic has emphasized the importance of a global effort to obtain disease control, in which vaccination plays a key role. Vaccination uptake, however, is highly dependent on vaccine acceptance, and is ultimately instrumental in achieving herd immunity: a critical threshold for disease control. COVID-19 vaccination campaigns have been initiated since December 2020 [6]; however, concerns about vaccination hesitancy had been raised even prior to the start of these campaigns. In multiple surveys, individuals expressed their worries about the safety of the vaccine, due to its rapid development [7]. Other studies reported that individuals deemed the vaccine unnecessary, because they believed they were already immune [8].
Even before the approval and roll-out of COVID-19 vaccines, multiple studies have investigated vaccination willingness and/or hesitancy within several populations. In their Table 1. PubMed search string, adapted from Lin, Tu and Beitsch [13] (continued on the next page). Vaccination willingness was defined as the proportion of participants willing or intending to receive a vaccine (if or when available). Responses indicating a positive tendency towards vaccinate acceptance, or a negative tendency on a question regarding vaccine refusal or hesitancy, were considered as vaccination-willing (see full overview of categorization of response options in Table 2). Table 2. Survey question responses categorized as vaccination-willing.

Questioned Responses
Vaccination willingness yes/yes, I will be vaccinated/yes if it's free/strongly agree/agree/yes certainly/yes possibly/definitely/probably/definitely planning/probably planning/yes (only when the vaccine is provided for free)/yes (willing to pay for a vaccine)/very likely/somewhat likely/probably yes/completely agree/somewhat agree/definitely yes/quite a bit/extremely likely/8-10 (very likely)/I probably would want to receive it/I definitely would want to receive it/high 6-7 Vaccination refusal or hesitancy probably not/certainly not/accepting of a COVID- 19 vaccine To gain insight into the determinants of vaccination willingness, determinant analyses were performed for all selected articles. Initially, original questionnaires were checked to retrieve information on potential determinants. If survey questionnaires were unavailable, the full text article was read to retrieve information on measurements and determinants. Measurements were categorized as: significantly associated with the outcome (vaccine acceptance or refusal), not significantly associated with the outcome, unclear if a relationship with the outcome was investigated, or a relationship with the outcome was not investigated. Determinants were included when the association was investigated with a multivariable regression model. Only if the article did not present results of a multivariable regression, were estimates of univariable analyses considered.

Quality Assessment
To critically appraise the methodological quality of every study included in the systematic review and meta-analysis, the Joanna Briggs Institute (JBI) Critical Appraisal Tool Checklist for Prevalence Studies was used [15]. This checklist was deemed to fit the aim of the systematic review best. The checklist consisted of nine questions on the methods (e.g., sample selection, sample size, valid and reliable measurements) and applied analyses of a study. The appraisal tool aids in the determination of what articles should be either included or excluded, or whether more information should be sought [16].

Data Analysis
To estimate global vaccination willingness, a random-effect meta-analysis of single proportions was performed. Similar meta-analyses were performed to estimate vaccination willingness per, continent, if multiple surveys were conducted.
A random intercept logistic regression model (or generalized linear mixed model, GLMM) was used for the meta-analysis, after logit transformation of the data. The withinstudy variation was estimated with the Clopper-Pearson 95% confidence interval (CI) and the between-study variation was estimated with the maximum likelihood estimator for tau 2 . The Higgin's and Thompsons's I 2 was used to assess heterogeneity. Statistical analyses were performed, and the forest and funnel plots were created using R Statistical Software version 4.0.5. [17]. Microsoft Excel was used for determinant assessment.

Search Results
The data search identified 411 articles. One article was found to be an author's correction on another article, and was therefore merged with the original [18]. After assessing eligibility based on the title and abstract or the full text, 40 articles were included in the final selection. Amongst these, 4 articles were found to contain results of multiple surveys. Lazarus [20]. Lastly, the study of Wang et al. performed two surveys with different samples at two different time points [21]. This meant that the results of a total of 63 surveys were included in the meta-analysis ( Figure 1).

Study Characteristics
The vast majority of studies had a cross-sectional design, where data and measurements were retrieved via a telephone or online survey. Some studies recruited participants via existing databases, and others conducted snowball sampling (through social media, or email) as the main recruitment method. Surveys were performed between March 2020 and December 2020, with the most surveys performed in June (n = 10) [18,[22][23][24][25][26][27][28][29][30]. One study was conducted after the initiation of vaccination campaigns. However, campaigns had not rolled out; therefore, vaccines were not yet available in the countries of the surveyed participants [19]. The sample size varied from 113 participants [31] to 5114 participants per survey [32]. In two studies, participants were asked a question on vaccine hesitancy/refusal, instead of acceptance [20,33].

Study Characteristics
The vast majority of studies had a cross-sectional design, where data and measurements were retrieved via a telephone or online survey. Some studies recruited participants via existing databases, and others conducted snowball sampling (through social media, or email) as the main recruitment method. Surveys were performed between March 2020 and December 2020, with the most surveys performed in June (n = 10) [18,[22][23][24][25][26][27][28][29][30]. One study was conducted after the initiation of vaccination campaigns. However, campaigns had not rolled out; therefore, vaccines were not yet available in the countries of the surveyed participants [19]. The sample size varied from 113 participants [31] to 5114 participants per survey [32]. In two studies, participants were asked a question on vaccine hesitancy/refusal, instead of acceptance [20,33].
The continent-specific vaccination willingness in Asia, Europe, and North America were close to the global willingness ( Supplementary Material, Figures S1.1-1.6)

Determinant Assessment
The second aim of the review was to gain insight into determinants of vaccination willingness and/or hesitancy. An overview of the most frequently studied determinants is presented in Table 4. Out of the 40 included articles, 12 did not perform univariable or multivariable regression analyses with vaccination willingness as outcome. Therefore, these were not included in Table 4. Although only two articles questioned vaccine hesitancy/refusal rather than acceptance, multiple papers changed the reference category to study determinants of hesitancy/refusal or lower vaccine acceptance. We observed that age, gender, education and attitudes around vaccines were most frequently significantly associated with either vaccine acceptance or refusal amongst the surveys that studied these determinants.
All 28 papers studied gender and age. In 16 out of 28, a significant association was found between gender and vaccine acceptance or hesitancy. Age was found to be significantly associated with vaccine acceptance or hesitancy in 13 papers, and was borderline significant in 1 study [42]. Associations were found for ethnicity (in some articles also specified as race), but the majority (10 out of 12) of these articles were on populations in the United States or the United Kingdom.
Education was studied in 25 articles, of which 14 found a (borderline) significant association. All 10 papers that reported on the benefits, harms and/or barriers of vaccines and/or other beliefs and attitudes about vaccines reported significant associations with vaccination willingness or hesitancy. Out of the 9 papers that studied partisan or political preferences, 8 studies found a significant association with acceptance or hesitancy. Although 14 articles investigated living location (either urbanity or geography) in a regression analysis, only 5 found a (borderline) significant association.

Risk of Bias
The methodological quality and risk of bias was checked for all 40 included articles using the JBI checklist for prevalence studies. A large proportion of studies used suboptimal recruitment methods (e.g., convenience and snowball sampling via social media), which may have led to a non-representative sample of the national population, although no studies were excluded.
To assess publication bias, a funnel plot was made (Supplementary Material, Figure S2). Out of the 63 included surveys, 42 were found to be outliers compared to the mean estimated global vaccination willingness. Table 4. Selected articles (sorted by first author) and studied determinants of vaccine acceptance or hesitancy and/or refusal.

Family composition
Children in household / household size Suspected or laboratory COVID-19

SES indicators
Education R PEER REVIEW 13 of 20 ographics Living location Suspected or laboratory COVID-19 9, x FOR PEER REVIEW 13 of 20

Sociodemographics
Living location

Family composition
Children in household / household size

Health status
Risk group or conditions Suspected or laboratory COVID-19

Family composition
Children in household / household size

Health status
Risk group or conditions

Experiences with COVID-19
Suspected or laboratory COVID-19 Suspected or laboratory COVID-19 9, x FOR PEER REVIEW Table 4. Selected articles (sorted by first author) and studied determinants of vaccine acceptance or hesitancy and/or refusal.

Sociodemographics
Living location

Family composition
Children in household / household size

Health status
Risk group or conditions

Family composition
Children in household/household size •

Sociodemographics
Living location

Family composition
Children in household / household size

Health status
Risk group or conditions Suspected or laboratory COVID-19

Family composition
Children in household / household size

Health status
Risk group or conditions Suspected or laboratory COVID-19

Sociodemographics
Living location

Family composition
Children in household / household size

Health status
Risk group or conditions  phics Living location

Sociodemographics
Living location

SES indicators
Education ; and significantly associated with vaccine acceptance ( ); significantly associated with vaccine refusal or hesitancy ( ). Papers indicated with (-) have only presented univariable regressions; therefore, the symbols indicate the association found in the univariable regression analyses. * in living location indicates that a geographical location rather than urbanity was measured.

Discussion
This systematic review and meta-analysis of 63 surveys estimated a global vaccination willingness of 66.01% [95% CI: 60.76-70.89%]. Additionally, several determinants were significantly associated with vaccine acceptance and hesitancy. Age, gender, education as well as attitudes and beliefs about vaccination were observed to be most frequently related to the respondent's intention to receive or refuse a vaccine when available.
However, differences in acceptance rates between and within continents and countries were observed. This estimated global vaccination willingness of 66.01% is lower than the estimates presented by Wang et al. [14]. Based on their systematic review including studies up to November 4, 2020, they reported a pooled vaccine acceptance rate of 73.31% [95% CI: 70.52-76.01%]. Multiple studies have suggested that vaccine acceptance changes over time. Sallam et al. reported vaccine acceptance to vary from 56.9% in April 2020 to 75.4% in June in the United States [58]. The multi-country survey of Neumann-Böhme et al. [9] reported a 20% decline in acceptance across multiple European countries during their study period.
Similar to the study of Wang et al. [14], the observed variation in our review across studies due to heterogeneity was high. Wang et al. found an I 2 value of 98.8% (no 95% CI presented), compared to 99.4% in our study. To account for heterogeneity, a random-effects model was used. Differences in time, place, and between populations were expected, but could not be explained due to clinical differences, for example, nor could they be investigated using subgroup analyses [59]. The heterogeneity observed in our study could not be eliminated in the continental analyses, with I 2 values for all stratified analyses exceeding 95%.
In the determinant assessment, we observed significant associations with gender and vaccination hesitance or acceptance in multiple studies. Prior to the COVID-19 pandemic, a review reported that females were less likely to accept vaccination compared to males for tetanus, diphtheria, and pertussis (Td(ap)) and influenza [60]. This has been confirmed for COVID-19 [21,23,25,26,29,30,32,34,36,41,42,[61][62][63]. Unfounded rumors about vaccines having detrimental effects on fertility could explain why vaccination uptake and intention is lower in women. During the pandemic, similar concerns have been raised by (young) women as a reason for hesitancy [64].
Attitudes and perceptions, including perceptions on the benefits, harms and barriers of vaccines, were significantly associated with acceptance and hesitance. In line with the expectation, those with positive attitudes and perceptions of the COVID-19 vaccine or vaccines in general were more likely to accept and less likely to refuse, based on the findings of our review [14,19,25,26,35,43,49,52,55], as well as in other studies [61,65,66]. More specifically, and related to the rumors on infertility, concerns about the efficacy and possible side-effects seem to play a major role in the intention to vaccinate [61,65,66].
Healthcare system distrust as well as vaccine hesitancy are suggested to be closely related to health literacy [67]. Heidari et al. [68] emphasized the need for an intersectional gender approach to vaccine development and deployment. They recognize health literacy as one of the issues creating gender-related barriers to vaccination. Additionally, access to healthcare may influence vaccination willingness. The study of Bianconi et al. has discussed the potential role of gender inequalities in the access to healthcare facilities in COVID-19 testing and diagnostics [69]. Investigating similar barriers and inequities in healthcare (facility) access could be of interest during vaccine deployment [68].
To the best of our knowledge, this is the most recent systematic review and metaanalysis on vaccination willingness and hesitancy including related determinants. Previous studies have included papers published up to November 2020.
Our systematic review and meta-analysis is, however, subject to some limitations. To obtain the largest possible number of studies, we included studies with many different types of sampling methods. The majority of included studies used convenience sampling when recruiting participants. It is unclear how sampling methods may have influenced vaccination willingness, and whether these results may have been different if sampling would have been more purposeful. Concerns could be raised about minority and under-represented groups that may not be reached through convenience and snowball sampling methods [70].
In addition, although we did not apply geographical restrictions during the inclusion of articles, the final selection did show an overrepresentation of (Eastern) Asian and European studies. A study in nine low-and middle-income countries (LMICs) showed acceptance ranging between 23% (Benin) and 89% (Brazil) when the available vaccine was 90% effective and ranging between 48% (Benin) and 94% (Brazil) with a vaccine effectivity of 95% [71]. In their 2012 editorial, Thomson and Watson stated that vaccine adoption = access + acceptance [72].
Finally, in our determinant assessment, we could not provide full insights into the direction of associations. This is in line with earlier research which showed that determinants of vaccine hesitancy are highly context-specific [5]. This, together with high heterogeneity found in our study, as well as in the review by Wang et al. [14], imply substantial differences between (sub)populations. Therefore, strategies focusing on removing vaccination barriers should follow an approach that is in line with determinants that are relevant within their specific setting and population.
Future studies should therefore focus on the determinants observed in this study (age, gender, education as well as attitudes and beliefs about vaccination), yet with particular attention to context-specific conditions. This may reveal insights into (sub)population-specific vaccination hesitancy and thus help vaccination programmes to increase vaccination coverage.