STRENGTHS AND LIMITATIONS OF THIS STUDY

For country vaccination willingness, we included only studies with national representative samples.

For under-resourced communities’ vaccination willingness, we included studies with purposive samples.

We compared countries’ vaccination willingness with official country-level national reports.

Official country-level reports about uptake among under-resourced communities were limited.

We could not compare vaccination willingness with real-world vaccine uptake statistics among under-resourced communities.

Introduction

Cumulative excess death from the COVID-19 pandemic made it a leading global cause of death between 2020 and 2021.1 Universal vaccination played a significant role transitioning into post-pandemic life.2 COVID-19 vaccines were developed and authorised in record time; as of April 2023, 70% of the world population received at least one COVID-19 vaccine dose. However, vaccine uptake is complicated; it involves more than simply making vaccines available. For instance, inequitable vaccine distribution possibly contributes to the 2.8-fold difference in vaccine coverage between high-income countries (HICs) and low-income countries (LICs).3 Whereas vaccine uptake in HICs was 81%, vaccine uptake in LICs was 29%.4

Countries with strong public health systems and economic resources achieved some early success vaccinating populations, yet people from historically, socially or economically under-resourced communities, such as people who experience homelessness, people from ethnic and racial minorities, as well as people with immigration or refugee experience, possibly remained unvaccinated for complex reasons. Regarding vaccination willingness and uptake among people from ethnic minority groups, Raizai et al 5 6 identified several structural aspects resulting from a mistrust of government and public health bodies: systemic racism and discrimination at societal and healthcare system levels, histories of unethical studies, as well as under-representation of people from ethnic and racial minority groups in health, drug and vaccine trials. Distrust in medical institutions from inappropriate care and mistreatment also impacted vaccination willingness among people from socially or economically under-resourced communities, such as members of indigenous communities or racial minority groups as well as among incarcerated individuals.7–9

Additionally, local barriers to access vaccinations and individual vaccine hesitancy played roles explaining vaccine uptake differences within and among countries.3 Nonetheless, structural access barriers and individual vaccine hesitancy possibly share common pathways, which complicate disentangling their effects in vaccination uptake.10 For instance, in a systematic review of barriers, facilitators and vaccine hesitancy with included studies about mainly HICs, they found individuals from minority ethnic groups concurrently experience more access barriers along with higher vaccine hesitancy and lower vaccine uptake when compared with individuals from majority ethnic groups and non-migrants.11 Therefore, a debate is ongoing about the true proportion of hesitancy and vaccine refusal among unvaccinated individuals in HICs. Although individual vaccination willingness is not under discussion, an understanding about vaccination willingness and vaccine uptake possibly informs health policies more reliably, identifies access barriers to vaccines, facilitates vaccination campaign planning and enhances uptake, eventually.

Generally, marginalisation and vaccine uptake in HICs has been scarcely described in the literature. We performed a realist synthesis to evaluate COVID-19 vaccine acceptance and its determinants among people from under-resourced communities in HICs. We compared data collected from a specific systematic review with real-world statistics to study the general evolution of vaccination rates—from hypothetical acceptance before the widespread rollout of vaccination programmes—until December 2021, 1 year after the first vaccine was available and when presumably, most HIC populations could be vaccinated. In addition, we compared hypothetical vaccination willingness between the general population and under-resourced communities in HICs.

MethodsStudy design and sources of data

We conducted a quantitative realist synthesis on the prevalence of vaccine acceptance among the general population from HICs. We followed the Realist And Meta-narrative Evidence Syntheses: Evolving Standards quality and publication standards and reporting guidelines.12 We also report our findings according to the statement on Preferred Reporting Items for Systematic Reviews and Meta-Analyses.13 We defined vaccination willingness as the proportion of participants willing or intending to receive a vaccine before vaccines were available. We defined vaccine uptake as the real proportion of the population with complete vaccination as reported by each country until November 2021.

A medical information specialist searched three electronic databases: PubMed, Embase and Dimensions ai. For informal sources, and to add possibly relevant articles where the search terms only appear in the full text of an article, we also screened the first 200 hits of a Google Scholar search. The detailed search strategy is available in the section 1 of the online supplemental material. We sought peer-reviewed scientific literature published before 30 November 2022. Different descriptors were used for each component of the search, for surveys investigating COVID-19 vaccine attitudes among adult populations from HICs before COVID-19 vaccine rollout. We used the World Bank database to classify countries of origin according to income at the time of data collection (US$12 536 or more gross national income per capita in 2019). We defined the study to include surveys reporting quantitative data on populations willing to be vaccinated when vaccines became available. We included surveys meeting the following criteria: (1) conducted in 2020–2021 among adult populations before vaccine rollout campaigns; (2) reported prevalence of vaccination willingness via questionnaires; (3) peer-reviewed; (4) performed probabilistic sampling; and (5) reported results for general populations and/or under-resourced communities.

To mitigate the risk of bias, for country vaccination willingness, we included only studies with national representative samples. For under-resourced communities' vaccination willingness, we also included studies with purposive samples.

We excluded studies of unrepresentative participants from general populations, such as people with particular conditions or health statuses—like people with diabetes or pregnant people—or particular occupations—like healthcare workers or university students. We excluded articles with incomplete information, systematic reviews and meta-analyses, and reports from meetings or congresses.

We provide details for our study selection and data extraction methods in section 1 of the online supplemental material. When multiple records included data from the same country, we extracted data from all of them and calculated country-specific pooled prevalence and used the pooled prevalence as the value to compare further with real-world statistics of vaccine uptake.

Study outcomes

For each country, outcomes of interest included (1) the proportion of people willing to be vaccinated according to results of the systematic review (primary outcome: vaccination willingness/acceptance); and (2) the proportion of vaccinated people according to the real-world data statistics (secondary outcome: vaccine uptake).

Data selection and extraction

Two reviewers independently screened all records and verified included and excluded studies by using REDCap (Vanderbilt University, Nashville, Tennessee, USA). We report identification, exclusion and inclusion of studies in the online supplemental figure 1. One reviewer extracted data using a pre-piloted extraction form, and a second reviewer verified the extracted data. Extracted variables were included, yet were not limited to sample size, study design, publication date, survey date, country and study population composition, community type, age, vaccine hesitancy, vaccine acceptance and vaccine refusal (section 1.d of the online supplemental material). We extracted all proportions as reported. For the realist synthesis, we obtained available country-specific data from multiple sources.14 15 We provide sources of information and definitions for country-specific variables in section 1.d of the online supplemental material.

Patient and public involvement

Patients or the public were not involved in the design, or conduct, or reporting, or dissemination plans of our research.

Potential bias assessment

Two independent reviewers assessed the risk of bias for each study using the checklist for prevalence studies from Hoy et al; we assessed each question independently and calculated scores, as recommended by checklist developers.16 However, we did not use total scores in analyses. Instead, we grouped questions into categories according to the bias domain they addressed.17 We analysed risk of selection bias and risk of non-response bias as potential sources of heterogeneity among studies. We provide potential bias assessment results in online supplemental table 1.

Statistical analysisData synthesis

We estimated the pooled prevalence of vaccination willingness and 95% CI using random-effects models. We used the ‘metaprop’ function from the ‘meta’ package in R (V.3.5.1) to synthesise and display findings from included studies in forest plots. For overall summary estimates, we calculated prediction intervals to represent the likely range of proportions obtained in subsequent studies conducted in similar settings.18 We quantified statistical heterogeneity using the I² statistic. Heterogeneity was classified according to the most recent version of the Cochrane Handbook: 0–40% might not be important; 30–60% may represent moderate heterogeneity; 50–90% may represent substantial heterogeneity; 75–100% considerable heterogeneity. However, in meta-analyses of prevalence, heterogeneity according to the I² statistic is expected to be substantial and possibly not discriminative.19 Therefore, we also calculated prediction intervals to describe the expected range of estimates.

Sensitivity analyses

We performed sensitivity analyses. First, we used the influence function in the ‘metafor’ package to compute outliers and influential case diagnostics, including externally standardised residuals and leave-one-out estimates of heterogeneity. Second, we investigated the impact of selection bias as a potential source of heterogeneity by means of meta-regression.

Real-world data analysis

After synthesising information from included studies, we compared results for each country with real-world data statistics concerning vaccination uptake. In addition, we identified how different country characteristics and policies (online supplemental table 2) in each country could be associated with vaccination uptake. Specifically, we selected four components to examine separately: percentage of populations older than 65 years; social spending as a percentage of gross domestic product (GDP); healthcare spending as a percentage of GDP; and stringency index (Oxford COVID-19 Government Response Tracker index) at the start date of vaccine rollout campaigns in each country since we thought them most likely associated with vaccine uptake among general populations.14

Results

After deduplication, we identified 3349 potentially relevant citations. After initial screening based on titles and abstracts, we selected full texts of 214 articles for detailed evaluation (online supplemental figure 1). After full-text assessment, we excluded 152 citations. We provide the complete list of excluded references and reasons for exclusion in section 1c of the online supplemental material. We included the remaining 62 articles that reported vaccination willingness before vaccine rollout at the country level.

General characteristics of included studies

We provide detailed characteristics of included studies in table 1. Overall, studies included 299 769 individuals from 18 HICs. Among the 62 included references, 45 studies reported results for general populations and 17 studies reported results for at least one under-resourced community. We calculated the weighted average of exported mean ages from each study; the mean age was 47.5 years. The proportion of women ranged from 16% to 93% among studies including patients from both sexes. Two studies reported including only men.20 21 Study sample sizes conducted among general populations ranged from 316 to 63 266 and study sample sizes conducted among under-resourced communities ranged from 83 to 18 474.

Table 1

General characteristics of included studies

Since reporting vaccination willingness via questionnaire was an inclusion criterion, all studies used validated questionnaires or questionnaires developed specifically for studies.

General characteristics of the included countries

We present detailed characteristics of included countries in online supplemental table 2. Country populations ranged between 2.6 million (Qatar) and 332 million (USA). Median population was 11.1 million (IQR: 7.9–67). Median percentage of populations older than 65 years was 19 (IQR: 16.8–22.2), and median value for life expectancy was 81.5 years (IQR: 81–83). With respect to economic indicators related to public policy, median social spending as a percentage of GDP was 25 (IQR: 18–29); median healthcare spending as a percentage of GDP was 10.3 (IQR: 8.7–11.3). We determined two median indicators of inequality: poverty gap 0.29 (IQR: 0.26–0.33) and gender wage gap 15 (IQR: 6–19), respectively.

Proportion of people from general populations reporting vaccination willingness before vaccine rollout

Among general populations, the summary proportion of vaccination willingness (figure 1) was estimated across all study settings as 67% (95% CI 61% to 72%, 45 studies). 45 studies reported vaccine acceptance among general populations: Australia (three studies)22–24; Austria (one study)25; Canada (two studies)26 27; Croatia (one study)28; Denmark (one study)29; France (five studies)30–34; Germany (one study)35; Greece (one study)36; Ireland (one study)37; Israel (one study)38; Italy (four studies)39–42; Japan (five studies)43–47; Portugal (one study)48; Qatar (one study)49; Belgium (one study)50; the UK (seven studies)51–57; and the USA (nine studies).58–66

Figure 1

Random-effects meta-analysis of COVID-19 vaccine acceptance in the general population. For each study, boxes and horizontal lines correspond to the respective point estimate and accompanying 95% CI. The size of each box is proportional to the weight of that study result in the fixed-effects model. The red diamond represents the 95% CI of the summary pooled estimate of the effect and is centred on pooled prevalence of vaccine acceptance. Heterogeneity estimate of I2 accompanies the summary estimate. Studies are ordered by the proportion of acceptance.

Proportion of people from under-resourced communities reporting vaccination willingness before vaccine rollout

The summary proportion of vaccination willingness for studies conducted among people from under-resourced communities (figure 2) was estimated as 52% (95% CI 0.46% to 0.57%, 17 studies). The 17 studies reporting vaccine acceptance in under-resourced communities included four studies among people experiencing homelessness67–70; two studies among people using illicit and unprescribed drugs71 72; three studies among lesbian, gay, bisexual and transgender populations21 73 74; two studies among incarcerated populations20 75; two studies among refugee and undocumented migrant populations76 77; and one study for each one of the following: indigenous population,9 a rural community,78 a Latino population79 and a black American population.7 In the cumulative meta-analysis from sensitivity analyses, we found a trend towards acceptance according to dates of data acquisition ranging from 32% in early pandemic stages to 52% during late pandemic stages before vaccine rollout (section 5c of the online supplemental material).

Figure 2

Random-effects meta-analysis of COVID-19 vaccine acceptance in special populations. For each study, boxes and horizontal lines correspond to the respective point estimate and accompanying 95% CI. The size of each box is proportional to the weight of that study result in the fixed-effects model. The red diamond represents the 95% CI of the summary pooled estimate of the effect and is centred on pooled prevalence of vaccine acceptance. Heterogeneity estimate of I2 accompanies the summary estimate. Studies are ordered by the proportion of acceptance.

Proportion of vaccine uptake from real-world country statistics 1 year after vaccine rollout

The summary proportion of vaccine uptake from included countries was estimated as 73% (95% CI 0.69% to 0.76%, 18 countries). In general, the proportion of vaccine uptake for each country was higher than vaccination willingness before vaccine rollout (online supplemental table 3), except for Croatia (−15%), Denmark (−3%) and the USA (−8%). In the cumulative meta-analysis, we did not observe an effect from date of vaccine approval on vaccine uptake at the end of 2021 (section 6 of the online supplemental material). However, in meta-regression analyses (section 6 of the online supplemental material), vaccine uptake increased according to the proportion of the population older than 65 years (OR=1.8, 95% CI 1.04 to 3.1) and decreased at higher stringency index values (OR=0.8, 95% CI 0.69 to 0.94).

DiscussionMain findings

Our realist synthesis involves data from 62 studies and 18 countries; we contribute to knowledge about the prevalence of vaccine acceptance among general populations and people from under-resourced communities. Additionally, we compared proportions of expected vaccine uptake from studies conducted before vaccines were available with the real uptake from the end of December 2021. To our knowledge, ours is the first systematic and realist review comparing vaccination willingness and vaccine uptake using real-world statistics among general populations with people from under-resourced communities in HICs.

The countries included in the study represented 70% of the HIC world population. Most countries showed higher vaccine uptake compared with the reported vaccination willingness in studies conducted before the vaccine rollout. For all studies among general populations, the proportion of vaccination willingness was 67% (95% CI 62% to 72%). In real-world settings, the overall proportion of vaccine uptake among countries was 73% (95% CI 69% to 76%). However, the scope of this study is limited in exploring possible explanations for lower-than-expected rates of vaccine uptake in Croatia, Denmark and the USA. For all the other countries, the real-world uptake was consistently higher than the reported willingness before rollout.

It is worth noting that some studies not included in our meta-analysis that evaluated the willingness to receive the vaccine when the vaccination rollout had already started in their country may have reported higher rates of willingness to receive the vaccine compared with the country’s real uptake.80 However, this should not be interpreted as an overestimation since such willingness was estimated on the unvaccinated fraction of the population instead of the total population of the country who was completely unvaccinated only before the rollout.

The pooled proportion from studies reporting vaccination willingness among under-resourced communities before rollout was 52% (95% CI 0.46% to 0.57%). Official country-level reports about vaccine uptake among under-resourced communities were too limited so we could not compare vaccination willingness before rollout with real-world uptake statistics among under-resourced communities after vaccine rollout.

Findings in context

The proportion of vaccination willingness among people from under-resourced communities was consistently lower than the proportion of vaccination willingness among people from populations in total. Existing evidence suggests people from ethnic minority groups7 and indigenous communities reasonably distrust medical institutions from experiences of differential care and mistreatment.8 9 Mistrust of institutions and governments was reported as the most common reason to delay vaccine uptake among ethnic minority groups,7 indigenous communities8 9 and incarcerated people.75 Experiences of discrimination, stigma and barriers to access were reported as possible explanations for lower prevalence of vaccine acceptance among people from sexual and gender minority groups.81

Despite the lack of official data on real-world uptake among under-resourced communities, some studies have reported lower vaccine uptake compared with the general population. For instance, a study among healthcare workers in the UK found that vaccine uptake was 58.5% among South Asian and 36.8% among black ethnic minority groups, compared with 70% in white healthcare workers.82 Another analysis of patient primary care records in the UK found lower vaccine uptake among different ethnic groups (black 68%, white 96%) and to a lesser extent, among different levels of deprivation (most deprived 91%, least deprived 97%).83

Recent evidence provides initial insights about overcoming barriers to vaccination uptake. For instance, multicomponent interventions with tailored communication of risks of remaining unvaccinated and benefits of becoming vaccinated,84 community-based action and engagement of religious and community leaders, dialogue to understand reasons for mistrust in government and public health bodies, as well as provision of access to convenient vaccination in collaboration with community-based and trusted health institutions.85

We suggest future studies compare trajectories of vaccination willingness with vaccine uptake among under-resourced communities. We also recommend future research link findings of trajectories with context-specific actions to address barriers to vaccine uptake among people from under-resourced communities. Ultimately, more research is needed to better understand vaccine uptake and the joint interactions among barriers, unwillingness, hesitancy, postponement or other unknown aspects driving vaccine uptake. The identification of necessary adjustments needed to improve vaccination uptake among different groups may inform future vaccination programmes.

Strengths and limitations

Studies reporting prevalence served as important sources of evidence during the COVID-19 pandemic and helped researchers understand factors related to the disease and inform policies. However, prevalence estimates from individual studies and pooled prevalence estimates from our meta-analyses may have been affected by selection and reporting biases.17 Nevertheless, our inclusion criteria attempted to reduce such risks of bias, and we performed multiple sensitivity analyses that provided insights into possible sources of heterogeneity. A strength of the realist approach is the use of diverse sources of information. In the specific context of COVID-19 vaccine acceptance, the fact that countries have reporting systems in place to keep population-based statistics made it possible to assess the real-life counterpart of the studies.86

Conclusion

Our systematic and realist review highlights COVID-19 vaccine uptake in HICs generally exceeded expressed vaccination willingness before vaccine rollout and vaccination willingness tended to be lower among under-resourced communities, when compared with total populations living in HICs. Our review emphasises the importance of realist reviews for assessing vaccine acceptance and particularly the need for more specific real-world statistics on vaccine uptake among under-resourced communities as well as the importance of context-specific actions to promote vaccine uptake and reporting.

Data availability statement

Data are available upon reasonable request. All data relevant to the study are included in the article or uploaded as supplemental information. All data relevant to the study are included in the article or uploaded as supplemental information. Also, any other data are available upon reasonable request.

Ethics statementsPatient consent for publication

Not applicable.

Ethics approval

Not applicable.

Acknowledgments

We thank the Lindenhof Foundation for their financial support of our research group. We also thank Kristin Marie Bivens for her editorial work and guidance on our manuscript.