key: cord-0729330-4weg291o
authors: Robinson, Eric; Jones, Andrew; Lesser, India; Daly, Michael
title: International estimates of intended uptake and refusal of COVID-19 vaccines: A rapid systematic review and meta-analysis of large nationally representative samples
date: 2021-02-06
journal: Vaccine
DOI: 10.1016/j.vaccine.2021.02.005
sha: 491e3799242d122db996de3ce45f80d56c4c5038
doc_id: 729330
cord_uid: 4weg291o

Background Widespread uptake of COVID-19 vaccines will be essential to extinguishing the COVID-19 pandemic. Vaccines have been developed in unprecedented time and quantifying levels of hesitancy towards vaccination among the general population is of importance. Methods Systematic review and meta-analysis of studies using large nationally representative samples (n≥1000) to examine the percentage of the population intending to vaccinate, unsure, or intending to refuse a COVID-19 vaccine when available. Generic inverse meta-analysis and meta-regression were used to pool estimates and examine time trends. PubMed, Scopus and pre-printer servers were searched from January-November, 2020. Registered on PROSPERO (CRD42020223132). Findings: Twenty-eight nationally representative samples (n = 58,656) from 13 countries indicate that as the pandemic has progressed, the percentage of people intending to vaccinate and refuse vaccination have been decreasing and increasing respectively. Pooled data from surveys conducted during June-October suggest that 60% (95% CI: 49% to 69%) intend to vaccinate and 20% (95% CI: 13% to 29%) intend to refuse vaccination, although intentions vary substantially between samples and countries (I2 > 90%). Being female, younger, of lower income or education level and belonging to an ethnic minority group were consistently associated with being less likely to intend to vaccinate. Findings were consistent across higher vs. lower quality studies. Interpretation: Intentions to be vaccinated when a COVID-19 vaccine becomes available have been declining globally and there is an urgent need to address social inequalities in vaccine hesitancy and promote widespread uptake of vaccines as they become available. Funding: N/A

The COVID-19 pandemic has resulted in more than 1 million deaths worldwide from March to October, 2020 1 and is likely to continue to have far reaching impacts on healthcare systems 2 . The development of vaccines against COVID-19 has been occurring at unprecedented speed, and as of November, 2020, there are multiple candidate vaccines in the final stages of testing 3 . The success of any vaccination programme is dependent on the proportion of the population willing to be vaccinated and based on recent estimates it is likely that up to three quarters of the population may require vaccination to bring an end to the pandemic 4, 5 .

Early in the pandemic a small number of studies surveyed adults to gauge public willingness to be vaccinated against COVID-19 and although a number of studies were reliant on non-representative convenience samples, the majority of the populations sampled intended to vaccinate [6] [7] [8] [9] . For example, a cross country survey found a relatively high average level of intended vaccination (72%), although sample sizes of individual countries were low (N = ~600-800) and may not provide accurate nationally representative coverage 7 .

However, as the pandemic has evolved there have been reports of widespread misinformation about COVID-19 10 , distrust in government 11 and public concerns about the safety of COVID-19 vaccines given their rapid development 12 , all of which may have affected vaccine uptake. It is also unclear whether vaccine acceptability will be sociodemographically patterned. Disadvantaged minority groups have previously been shown to be less likely to intend to be vaccinated for influenza 13, 14 , although a systematic review concluded that other demographic patterning of previous influenza vaccination programmes is inconsistent 15 . Given that current evidence on socio-demographic patterning of COVID-19 vaccination intentions is lacking, it will be important to understand how vaccination intentions differ within and across countries to inform measures to improve public acceptability and uptake of vaccination programmes.

We conducted a rapid systematic review and meta-analysis of large nationally representative study samples to address the current lack of consensus on; i) the proportion of the population willing to be vaccinated against COVID-19 when a vaccine become available and how this differs across countries, ii) whether vaccination intentions have declined as the pandemic has progressed, iii) socio-demographic inequalities in intended vaccine uptake.

To inform mass COVID-19 vaccination programmes, we used rapid systematic review methodology 16 . Rapid reviews provide timely evidence synthesises whilst maintaining the rigour of traditional systematic reviews 17 using expedited review processes 17,18 , such as limiting databases searched or number of reviewers (e.g. cross-checking of a proportion of extraction as opposed to independent extraction by a second author).

We included studies that measured intentions to be vaccinated against coronavirus in nationally representative samples of the general public. Published journal articles and pre-prints were eligible for inclusion. News articles reporting on opinion polls (with no corresponding scientific report including methodology used) were not eligible. The review is registered on PROSPERO (CRD42020223132).

To be eligible studies were required to have used a sampling approach designed to be nationally representative on key population demographics of the country (e.g. gender, education level), such as quota sampling or random probability sampling. Sampling error is problematic when examining prevalence estimates in small sample sizes 19 . Because sampling error tends to be minimal at sample sizes of ≥1000 19 , as is considered common practice for nationally representative surveys 20 , we limited eligibility to studies with a sample of n≥1000 from the same country. Studies that collected non-general public samples (e.g. healthcare professionals, parents, students) were not eligible. Studies that used non-representative sampling (e.g. convenience sampling, snowball sampling) were not eligible and studies that lacked sufficient methodological information to determine sampling used were ineligible.

To be eligible studies were required to include a question that measured intentions/willingness to use a vaccine for COVID-19 when one becomes available (e.g. 'I would use a vaccine for COVID-19 when it becomes available'). Studies that exposed participants to information designed to alter vaccine intentions (e.g. experiments comparing how public health messages may improve vaccine intentions) were not eligible. Studies that compared willingness to take different types of vaccine (e.g. varying hypothetical effectiveness/price) were ineligible, unless they also included a questionnaire item measuring general willingness to take a COVID-19 vaccine.

Outcome: Studies were required to report the proportion of participants responding to the different response options for the vaccine intention question (e.g. Yes vs. No, Willing vs. Unsure Vs. Unwilling, Likely vs. Undecided vs. Unlikely).

Article identification strategy. During November 2020, we searched PUBMED and Scopus (2020 onwards) for published articles in peer reviewed journals. We used the following search terms: (COVID OR coronavirus OR SARS-COV-2) AND (Vaccine OR Vaccination) AND (Inten* OR willing* OR attitud* OR hypothetical). We also searched three pre-print servers; Open Science Framework (which includes 30 other preprint archives, including PsychArxiv), MedrXiv and the Social Science Research Network (SSRN). One author conducted the initial title and abstract screening to exclude unrelated articles and a second author checked 25% of this (there were no discrepancies). A single author conducted full-text screening to determine eligibility and a second author cross-checked all eligibility decisions.

For all eligible articles identified through searches, we used forward citation tracking (Google Scholar) and our knowledge of existing research to identify any further articles.

Data extraction. For each study one author extracted information, all extraction was crosschecked by a second author and disagreements were resolved through discussion. We extracted the following information; bibliographic information, country, sampling procedure (e.g. quota vs. probability), sample size, month of survey, measure of vaccination intentions and results. We extracted results based on response options, resulting in % choosing response options indicative of definitely or probably yes, % definitive or probable no, % unsure (if measure allowed for the latter). If studies had multiple waves of data collection, we extracted results from the most recent wave. We also extracted information on whether studies reported results of analyses examining demographic predictors of vaccination intentions that were commonly examined (i.e. at least 5 studies) in studies. To be eligible for extraction, demographic predictors were required to be examined adjusting for other demographics (i.e. zero-order correlations were not eligible) in order to be confident of independent effects. We prioritised extraction of results of analyses that examined vaccine intentions (i.e. reference of category of yes vs. other (unsure/no combined). If this analysis was not available, in order of priority we favoured extracting demographic results for analyses examining yes vs. no, then yes vs. unsure.

We considered methodological factors of studies that may impact on results by increasing risk of bias. Quota-based samples were rated as being higher in risk of bias than probability-based sampling, as the latter tends to be a more accurate/representative sampling approach 21, 22 . We also reasoned that studies not having yet undergone formal peer review (unpublished pre-prints) may increase risk of bias and coded for this. For studies examining demographic predictors, we considered relatively smaller sample sizes (n<2500) as higher in risk of bias due to concerns over small numbers of cases in analyses when examining sub-groups. For example, with n ≥ 2500, for minority sub-groups (e.g. ~2% of the population, such as Black people in the UK) there would be expected to be a minimum of 50 cases in analyses, as opposed to only 20 cases with n=1000. Studies of demographic predictors that adjusted for attitudinal predictors (e.g. attitudes towards vaccination) of vaccine intentions were considered higher in risk of bias, as inclusion may mask associations between demographics and vaccination intentions. For example, if demographic patterning of vaccination intentions is caused in part by attitudinal differences, including attitudinal measures as predictor variables of intentions (alongside demographic factors) would reduce and 'mask' an association between a demographic factor and vaccination intentions.

We meta-analysed proportions of the samples ((Total sample N / 100) * % reporting) reporting: i) intending to vaccinate, ii) unsure, and iii) not intending to vaccinate. Analysis was performed using the 'metafor' package in R. We used a logit transformation on raw proportions in random effects, generic inverse variance meta-analyses with a restricted maximum-likelihood estimator. Transformations were conducted using the 'escalc' function in the metafor package. Back-transformed (inverse) logit values are presented in the text, whilst raw proportion data is presented in forest and funnel plots to aid interpretation. Heterogeneity was assessed with the I 2 statistic. We conducted metaregressions to examine the relationship between outcomes and month of study data collection (treating month as a continuous variable, e.g. March = 1, April = 2). We conducted leave-one-out analysis to examine the stability of the pooled estimates and identify any influential samples in the main analyses and meta-regressions. We reasoned that studies which did not use an 'unsure' or 'undecided' response option (e.g. responses grouped into yes vs. no, as opposed to yes vs. unsure vs. no) may affect vaccine acceptance and rejection estimates, so we examined if pooled estimates of intended vaccination and intended refusal differed between these two types of study. Furthermore, in sensitivity analyses we examined whether the inclusion of an 'unsure/undecided' response option moderated any effects of month of study data collection on intention estimates and the effect of month of study data collection separately in studies with vs. without an 'unsure/undecided' response option.

To address risk of bias, we conducted sub-group analyses to examine whether results differed between studies using quota vs. probability sampling and pre-prints vs. journal articles. We limited these analyses to studies which collected data early in the pandemic (March-May) to account for the majority of probability samples (3/4) and journal article sample (12/12) studies being conducted in this period. Sub-group analysis compared studies that did vs. did not allow for an 'unsure' response option. We also examined potential publication bias and small study effects (see online supplementary materials). Demographic predictors were measured and/or analysed differently across studies, so for each demographic we reported the proportion of studies finding evidence vs. no evidence of significant (p < .05) relationships with vaccination intentions, and whether results were similar when limited to studies that had larger sample sizes and did not adjust for attitudinal predictors in analyses. Publication and risk of bias analyses. There was minimal evidence of publication bias and leave out one analyses indicated limited variation in estimates (online supplementary materials document). We found no evidence that results differed between samples reported in journal articles vs. pre-prints. There was minimal evidence of differences in findings between studies using probability vs. quota sampling, with the exception of 'unsure' responses being lower in quota samples (see online supplementary materials). Table 2 .

In 12/14 studies older adults were significantly more likely to report intending to vaccinate than younger, one study found no effect of age and in one study young adults (<25 years) were more likely than middle aged adults, but not older adults. In 9/14 studies males were more likely to intend to vaccinate than females (no significant association in five studies).

Higher education level was associated with intending to vaccinate in 7/14 studies (no association in seven). White ethnic groups were more likely to vaccinate in 7/11 studies (no association in four). Higher income was associated with intending to vaccinate in 8/9 studies and in one study there was no association. Presence of a health condition was examined in five studies and was non-significant in n=4 and not having a health condition was associated with intending to be vaccinated in one study. When analyses were limited to the five higher quality studies (large sample size, no inclusion of attitudinal predictors) the role of demographic factors was more consistent; 5/5 studies found that older adults, males and higher education levels were associated with increased likelihood of intending to vaccinate. Similarly, 3/4 studies found that being white and 4/4 found that those on higher income were more likely to intend to vaccinate. Presence of a health condition was non-significant in n=2.

Results of this systematic review and meta-analysis of 58,656 participants drawn from 28 large nationally representative study samples across 13 countries indicates that the percentage of the population intending to be vaccinated when a COVID-19 vaccine becomes available has declined markedly across countries as the pandemic has progressed. Numbers reporting that they will refuse a vaccine have increased over time and a substantial proportion of adults now intend to refuse a vaccine, when available (June-October estimate = 20%). There is also consistent socio-demographic patterning of vaccination intentions; being female, younger, of lower income or education level and belonging to an ethnic minority group are associated with a reduced likelihood of intending to be vaccinated when a vaccine become available. Emerging evidence suggests that both exposure to misinformation about COVID-19 10, 24 and public concerns over the safety of vaccines 25 may be contributing to the observed declines in intentions to be vaccinated, and this highlights the need for measures to address public acceptability, trust and concern over the safety and benefit of approved vaccines. As well as observing declines in intentions to vaccinate over time in our main analyses, we also note intentions differed markedly by country. When sampled during a similar period early on in the pandemic (March-April), 91% of adults in China reported intending to be vaccinated, compared to 76% of adults in France. However, the most recent estimate (September-October) in France is now 52% and similar to the US (54%).

Across studies males were more likely to report intending to vaccinate than females and this is consistent with previous research examining influenza vaccine hesitancy and some studies identifying higher uptake among males 26 . However, there is also recent evidence of greater influenza vaccine uptake among females in both the US and Hong Kong 27, 28 . A range of social and contextual factors may explain gender patterning of vaccination intentions (e.g. current or planned pregnancy). It will therefore be important to understand why females show lower intentions to vaccinate against COVID-19 and whether uptake shows a similar gender patterning. There was also very consistent socioeconomic patterning of vaccination intentions: lower income or education and ethnic minorities were less likely to intend to vaccinate. There was no evidence in any studies reviewed that presence of a chronic health condition was associated with increased vaccination intentions, even though these groups are at increased risk of dying from COVID-19 2 9. Measures are required to maximise vaccine uptake in vulnerable and disadvantaged groups who have already been disproportionately affected by the pandemic, such as those from lower income and ethnic minority groups 29,30 .

Strengths of the present research are that we limited evidence synthesis to study designs that allow for accurate estimates of population level intentions with minimal sampling error, as small studies of non-representative samples are likely to provide biased estimates. A large proportion of the included studies used quota (as opposed to probability-based sampling) and

were pre-prints yet to be peer reviewed (as opposed to published journal articles). However, the type of sampling method used (quota vs. probability) had minimal impact on intentions estimates and that studies reported in pre-prints produced similar effect estimates as peerreviewed journals. Analyses also accounted for studies using different response formats and findings were similar. However, studies that included an 'unsure/undecided' response option had lower estimates of the proportion of the sampled population intending to vaccinate compared to studies with only two response options (e.g. 'yes' vs. no'), but there was no difference for the proportion of sampled populations indicating they would not vaccinate.

These findings indicate that studies which do not include an 'unsure' response option overestimate vaccination intentions and that 'unsure' response options may be primarily driven by participants who are unsure but more likely to intend to be vaccinated than not get vaccinated.

We used a rapid systematic review approach and this resulted in us surveying only two online database for published articles and rather than two authors independently assessing article eligibility and extraction, one author was responsible with cross-checking by a second author. However, we retained a number of features of best practice for systematic review and meta-analysis (e.g. assessing risk of bias, publication bias and thorough searching of grey literature). There were fewer studies later (as opposed to earlier) in the pandemic and this may affect the reliability of the time trends analyses. However, we found evidence of declining vaccination intentions when data was analysed using meta-regression (continuous month-of-year variable), when comparing study estimates from early in the pandemic were consistent 25, 31 . As included studies examined nationally representative samples we do not know whether a similar pattern of results would be expected among other population groups (e.g. healthcare workers) and research is required to address these questions 32,33 .

Findings of the present research are limited to studies that met eligibility criteria for having used large and nationally representative sampling and this tended to be developed western countries.

Conclusions. Intentions to vaccinate when a COVID-19 vaccine becomes available have been declining globally and there is an urgent need to address social inequalities in vaccine hesitancy and promote widespread uptake of vaccines as they become available.

Data sharing. Study data files and analysis code are openly available on the Open Science Framework; https://osf.io/hj4ds/ Contributors. The study was designed by ER. Data was collected by ER, AJ, IL and MD.

AJ and ER analysed the data. The first draft was written by ER. All authors critically revised the manuscript and agree to be accountable for all aspects of the work.

Role of the funding source. There was no funding source for this study.

All authors report no conflicts of interest. ER has previously received funding from the American Beverage Association and Unilever for projects unrelated to the present research. 

COVID-19 Weekly Operational Update. 2020. Accessed 30/11

The impact of the COVID-19 pandemic on cancer deaths due to delays in diagnosis in England, UK: a national, population-based, modelling study

Vaccine Efficacy Needed for a COVID

Coronavirus Vaccine to Prevent or Stop an Epidemic as the Sole Intervention

Will an imperfect vaccine curtail the COVID-19 pandemic in the

Acceptability of a COVID-19 vaccine among adults in the United States: How many people would get vaccinated?

A global survey of potential acceptance of a COVID-19 vaccine

Towards intervention development to increase the uptake of COVID-19 vaccination among those at high risk: Outlining evidencebased and theoretically informed future intervention content

Acceptance of covid-19 vaccination during the covid-19 pandemic in china

Susceptibility to misinformation about COVID-19 around the world

The Cummings effect: politics, trust, and behaviours during the COVID-19 pandemic

Attitudes Toward a Potential SARS-CoV-2 Vaccine: A Survey of U

Determinants of adults' intention to vaccinate against pandemic swine flu

Socioeconomic status, demographics, beliefs and A(H1N1) vaccine uptake in the United States

Barriers of Influenza Vaccination Intention and Behavior -A Systematic Review of Influenza Vaccine Hesitancy

What are the best methodologies for rapid reviews of the research evidence for evidence-informed decision making in health policy and practice: a rapid review

Rapid evidence assessment: increasing the transparency of an emerging methodology

Education-Resources/Election-Polling-Resources/Margin-of-Sampling-Error-Credibility-Interval.aspx. 2020. 20. Gallup. How polls are conducted. Accessed 30/11

Comparing the Accuracy of RDD Telephone Surveys and Internet Surveys Conducted with Probability and Non-Probability Samples

The effect of survey mode and sampling on inferences about political attitudes and behavior: Comparing the 2000 and 2004 ANES to Internet surveys with nonprobability samples

Confidence and Receptivity for COVID-19 Vaccines: A Rapid Systematic Review

Anti-vaccine attitudes and risk factors for not agreeing to vaccination against COVID-19 amongst 32,361 UK adults: Implications for public health communications

Willingness to vaccinate against COVID-19 in the US: Longitudinal evidence from a nationally representative sample of adults from

The intersection of sex and gender in the treatment of influenza

A retrospective analysis of gender-based difference in adherence to influenza vaccination during the 2018-2019 season

Household characteristics and influenza vaccination uptake in the community-dwelling elderly: a cross-sectional study. Preventive medicine reports

All authors report no conflicts of interest. ER has previously received funding from the American Beverage Association and Unilever for projects unrelated to the present research

Acknowledgments. ER's time was part-funded by the European Research Council and their support is gratefully acknowledged.