Face‐to‐face interventions for informing or educating parents about early childhood vaccination

Summary of findings for the main comparison.

Face‐to‐face interventions directed to parents for informing or educating parents about early childhood vaccination, as compared with control
Patient or population: parents of preschool‐aged children or expectant parents Settings: clinics, antenatal classes, or the mother's home Intervention: face‐to‐face information or educational interventions Comparison: control (no education, other education, or control not described)
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	No of Participants (studies)	Quality of the evidence (GRADE)	Comments
	Assumed (baseline) risk	Corresponding (intervention) risk
	Control (no face‐to‐face information or education)	Face‐to‐face information or education
Vaccination status Final time point (3, 6, or 12 months post‐intervention)	55 per 100¹	66 per 100 (57 to 75)	RR 1.20 (1.04 to 1.37)	3004 (7 studies)	⊕⊕⊝⊝ low²	The results for this outcome were variable, so the true result may be substantially higher or lower than this estimate.
Knowledge or understanding (Different measures used by the studies: knowledge of vaccine‐preventable diseases, vaccines, contraindications to vaccination, or a combination, on varying scales) Final time point (3 or 6 months post‐intervention)³		The mean knowledge score in the intervention group was 0.19 standard deviations higher (0.00 to 0.38 standard deviations higher)		657 (4 studies)	⊕⊕⊕⊝ moderate⁴	One further study (Quinlivan 2003) of 124 participants did not report individual group mean scores as data were skewed. The authors reported that there were no significant differences in the knowledge scores of the intervention and control groups (MD 0.85, 95% CI: ‐0.06 to 1.76). A standard deviation of 0.2 represents a small difference between groups (based on Cohen’s effect sizes; Cohen 1988).
Attitudes or beliefs (Different measures used by the studies: perceived severity of vaccine‐preventable diseases, perceived necessity of vaccines, or a combination, on varying scales) Final time point (3 or 6 months post‐intervention)		The mean score for perceived severity of diseases, necessity of vaccines, or both, in the intervention group was 0.03 standard deviations higher (0.20 standard deviations lower to 0.27 standard deviations higher)		292 (3 studies)	⊕⊕⊝⊝ low⁵	A standard deviation of 0.2 represents a small difference between groups (based on Cohen’s effect sizes; Cohen 1988).
Intention to vaccinate 1 to 7 (1 = definitely do not intend to vaccinate; 7 = definitely do intend to vaccinate) 3 months post‐intervention	The mean intention to vaccinate in the control group was 5.58 (SD 2.13)⁶	The mean intention to vaccinate in the intervention groups was on average 1.17 points higher (0.51 to 1.81 points higher)		179 (2 studies)	⊕⊕⊝⊝ low⁷
Adverse effects (anxiety associated with intervention) 20 (low anxiety) to 80 (high anxiety) 3 months post‐intervention	The mean anxiety in the control group was 33.9 (SD 13.53)	The mean anxiety in the intervention group was 1.93 points lower (7.27 lower to 3.41 higher)		90 (1 study)	⊕⊕⊝⊝ low⁸	The study authors note that a normal score is in the range of 34 to 36 on this scale.
Parent experience of the intervention	No studies measured this outcome.
Cost (personnel, supplies, travel, office space, and orientation costs of intervention)	Effect of intervention was very uncertain. A single study reported that the estimated mean cost of usual care per fully immunised child was USD 1587, or USD 1273 for children defined as high‐risk.⁹ The estimated additional cost per fully immunised child per intervention was approximately 8 times higher than usual care for all children, and 4 times higher for high‐risk children.			365¹⁰ (1 study)	⊕⊕⊝⊝ low¹¹	Did not include potential costs arising from non‐immunisation. Costs calculated per individual took account of indirect case management costs (frequency and time taken for case managers to complete client‐related contact tasks).
The basis for the assumed risk* (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: Confidence interval; RR: Risk Ratio
GRADE Working Group grades of evidence High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.
¹Assumed risk based on median control group risk across studies ²We downgraded the evidence for this outcome for risk of bias (‐1). One trial was at unclear risk for sequence generation, two trials were at high, and one at unclear risk of bias for allocation concealment. We also downgraded for inconsistency (‐1) because, while the nature of the interventions and participants were relatively similar across studies, there was considerable statistical heterogeneity that was not easily explained (I² = 85%, Chi² P < 0.00001). ³One study measured this outcome at 15 months postpartum. The intervention included multiple sessions at varying times, so the time between the final session and outcome assessment was not known. ⁴We downgraded the evidence for this outcome for risk of bias (‐1). One trial was at unclear risk for sequence generation and one trial was at high risk of bias for allocation concealment. ⁵We downgraded the evidence for this outcome for indirectness (‐1). There are many aspects of attitudes that are important to decision making. The specific attitudes measured in this outcome were only part of what could be measured, and therefore this outcome was a somewhat incomplete or indirect indication of attitudes. We also downgraded for risk of bias (‐1), because lack of blinding may have impacted this subjective outcome. ⁶Assumed risk based on control group score from Jackson 2011, as this was the only validated scale. ⁷We downgraded the evidence for this outcome for imprecision (‐1). The sample size for this outcome was relatively small and the effect estimate showed potentially appreciable benefit (i.e. the CI of the pooled effect estimate crossed 0.5). We also downgraded for risk of bias (‐1), because lack of blinding may have impacted this subjective outcome. ⁸We downgraded the evidence for this outcome for risk of bias (‐1), because only one study contributed data and it was at unclear risk of bias for sequence generation. We also downgraded for imprecision (‐1) because the total sample size was small (N = 135 participants). ⁹High‐risk subgroup defined as those children who received only 3/5 or fewer well‐child visits. ¹⁰Includes high‐risk subgroup (86 participants). ¹¹We downgraded the evidence for this outcome for risk of bias (‐1), because it was at high risk of bias for allocation concealment. We also downgraded for indirectness (‐1), because the evaluated intervention was multi‐component and included telephone reminders in addition to education.

Background

available in

This review originated as a part of the Communicate to Vaccinate (COMMVAC) project (2010‐2016), which sought to build evidence for communication interventions related to childhood vaccination (Lewin 2011). This topic was selected through an international priority setting exercise because face‐to‐face communication is widely used around the world, can be implemented in a range of settings and can be adapted or tailored for different populations. A second review topic was also selected from this exercise (information or education aimed at communities (Saeterdal 2014)).

Since 2013, when the review was originally published, there have been a number of developments in the field of vaccination communication research. Most notably, research related to vaccine hesitancy has highlighted the relevance and importance of face‐to‐face communication in countries of all income levels (Ames 2017; Henrikson 2015; Jarrett 2015; Leask 2015; Opel 2013). Therefore, in recognition of the progress of the field and the potentially wider audience for this review, we have revised the Background to strengthen the explanation of the intervention's theoretical underpinnings and situate this review within the landscape of evidence published since 2013. Some information from the original review, such as the detailed definition of the terms 'inform' and 'educate', is now presented in the supporting Appendices (Appendix 1).

Description of the condition

Early childhood vaccination is an essential public health practice, carried out in every country in the world and saving two to three million lives every year (WHO 2016). In addition to reducing preventable premature mortality, the practice of vaccination is cost‐effective, increasing parent or carer productivity, and lowering health care costs by reducing childhood morbidity (Ozawa 2012; WHO 2018). Childhood vaccination also has a broader social benefit of promoting health equity by ensuring the distribution of health within populations, and contributes to the public good by supporting disease containment (Luyten 2016).

Nevertheless, over 19 million children per year do not receive all the recommended basic vaccines (WHO 2018), and coverage levels fluctuate in individual countries or populations, and for specific vaccines (Nandy 2016; Smith 2008; UNICEF 2014). While the burden of vaccine‐preventable disease is greatest in low‐ and middle‐income countries (LMICs; Oyo‐Ita 2016; Sutter 2006), outbreaks occur all over the world, including in countries with relatively high coverage (e.g. Belgium, Italy, Ireland, Australia, USA; Barrett 2016; Filia 2017; Grammens 2017; Townsville‐Mackay 2013; Zipprich 2015). Therefore, improving and maintaining global childhood vaccination rates is an ongoing public health goal, prioritised by major international health strategies and agreements, such as the UN Millennium Development Goals (UN Millennium Project 2006), the WHO‐UNICEF Global Immunization Vision and Strategy (WHO 2009), and the Global Vaccine Action Plan 2011 to 2020 (WHO 2012).

The major factors influencing vaccination coverage relate to 1) availability of the vaccine, 2) awareness of recommendations, 3) accessibility of the services, 4) acceptance of the vaccines, and 5) activation to trigger the action of getting a vaccine (Thomson 2016). Face‐to‐face interventions to inform or educate parents specifically target two of these factors: awareness and acceptance.

Enhancing awareness of vaccination and its importance is critical to address barriers associated with limited or incorrect understanding of the value of vaccines, the risks of vaccine‐preventable diseases, or the vaccine schedule itself, which evolves and changes over time (Esposito 2014). People are unable to seek on‐time vaccination if they do not know where to go, how to access services, when or which vaccines are due, or why they are important.

Vaccine acceptance and the issue of vaccine hesitancy have become a major global focus in recent years (WHO SAGE Working Group on Vaccine Hesitancy 2014a). On the continuum from complete vaccine acceptance to total refusal, hesitant individuals are more likely to refuse or delay some or all vaccines (Dubé 2016). People may feel hesitant because they don't believe vaccines are necessary ‐ a conundrum created by vaccination is that as its uptake becomes more widespread, the diseases it prevents become less visible, and people can feel less urgency to vaccinate (Smith 2015). Vaccine hesitancy is also influenced by concerns about the safety of vaccines, highlighting the critical importance of appropriate responses to safety‐related events or rumours (WHO SAGE Working Group on Vaccine Hesitancy 2014). Although more commonly discussed and researched in high‐income countries (HICs), vaccine hesitancy is a worldwide phenomenon (Dubé 2014), with notable anti‐vaccine campaigns and safety scares taking hold in the UK (Brown 2012), Pakistan (Murakami 2014), and Nigeria (Kaufmann 2009). Ensuring that parents can access, understand, and judge the quality of information sources is critical, as misinformation and rumours can take root and spread particularly rapidly in a population with a high degree of hesitancy towards vaccines (Salmon 2015).

Successful vaccination programmes rely on people having appropriate information, and sufficient knowledge, awareness, and acceptance of vaccination to make the decision to participate (Dubé 2016; Shahrabani 2009; Thomson 2016; Zyngier 2011). Interventions to inform or educate may not necessarily be sufficient to change behaviour in all cases, but this does not negate their importance (Dubé 2015; Leask 2011). Ensuring that people are informed and knowledgeable about their health is a UN‐codified human right and a principal tenet of patient‐centred care (Dwamena 2012; Hill 2011; Rodriguez‐Osorio 2008; United Nations 2008; WHO 2007). Information is fundamental to valid consent, and extensive qualitative evidence indicates that parents want more or different information about vaccines (Ames 2017).

Description of the intervention

Information and education can be provided in various ways, but this review focuses specifically on face‐to‐face communication interventions (Kaufman 2017b). Face‐to‐face communication can be interactive, adaptable, and efficient, and already accompanies the delivery of nearly every vaccine injection or drop around the world. It allows for real‐time dialogue, during which parents are able to explain their concerns or preferences and ask personally‐relevant questions. It can take place in one‐on‐one interactions, or with small groups of people. Face‐to‐face communication is particularly useful when people are semi‐literate, or not literate in either their language or the majority language of their country.

Face‐to‐face information or education can be delivered by a range of individuals (e.g. volunteers, advocates, peers), but it is particularly relevant in the context of the healthcare encounter. Healthcare providers (e.g. doctors, nurses, community health workers, Indigenous health workers) are the primary source of information for parents about routine childhood vaccination (Freed 2011; Jones 2012). Parents trust the recommendations of their providers over other sources of information about vaccination, and communication that is respectful and builds trust can help hesitant parents work through their concerns (Freed 2011; Jones 2012; Leask 2015). Conversely, poor provider communication experiences can cause parents to feel confused, disrespected, or mistrustful, and potentially make them less likely to return for their child's next vaccination appointment (Brown 2010; Leask 2012). Face‐to‐face delivery of vaccination information or education to parents is clearly influential, but the most effective messages and communication styles are not yet established. Some evidence supports a participatory communication approach for increasing vaccine acceptance (Opel 2015), while other research suggests a more presumptive communication style may be more effective at increasing uptake (Brewer 2017).

In addition to one‐on‐one discussions with providers, face‐to‐face interventions to inform or educate may include oral presentations, individual or group classes or seminars, information sessions, or home outreach visits. Such interventions may be undertaken on their own, or combined with other interventions (e.g. telephone contact). This review focused on face‐to‐face communication with parents or guardians of preschool‐aged children, either individually or in groups. Interventions directed to adolescents were not included, as our focus was on early childhood vaccines. Interventions directed to communities are addressed by a separate review (Saeterdal 2014). Interventions directed to healthcare providers, such as professional education or communication training interventions, were also outside the scope of this review.

After a thorough investigation of the definitions of the words 'information' and 'education' (see Appendix 2), we determined that there was no reliable way to distinguish between the two across studies. A recent comprehensive table of definitions of education‐ and training‐related terms developed by the World Federation of Public Health Associations working group on Public Health Education and Training supports the conclusion that 'education' and 'information' are largely considered coterminous (WFPHA 2017). Therefore, we included any trial evaluating an intervention that aimed to inform, educate, or both, in this review.

How the intervention might work

While creating a generally informed and health literate populace is an important public health goal, most information or educational interventions aim to change or support specific health behaviours. The interventions in this review targeted the parental health behaviour of vaccinating their young children on time, and with all recommended vaccines.

Behaviour change is complex, and a number of theories outline potential factors and antecedents that may predict or influence an individual's behaviour (see Appendix 2 for an overview of several relevant theories). The Health Belief Model (HBM) is a key behaviour change theory that is particularly relevant for vaccination, as it was developed to explain people's participation in preventive health programmes, including vaccination (Champion 2008; Janz 1984). According to the HBM, the decision to take action ‐ or not ‐ is a product of a number of variables: perceived severity (of vaccine‐preventable diseases (VPDs)); perceived susceptibility (of one's child to VPDs); perceived benefits (of vaccination); perceived barriers (to getting vaccinated); and self‐efficacy (perceived ability to act). Each of these variables may be influenced by an information or educational intervention.

Several other health behaviour change theories (e.g. the Theory of Reasoned Action (TRA), Theory of Planned Behaviour (TPB), Integrated Behavior Model (IBM) share the idea that the most important determinant of behavioural action is behavioural intention (Ajzen 1985; Ajzen 1991; Montaño 2008). These theories suggest that intention is based on a combination of attitudes (i.e. how a person feels about the behaviour), subjective norms (i.e. social pressure related to the behaviour), and perceived behavioural control (i.e. whether the person feels able to act). The IBM also explicitly recognises the importance of a person having sufficient knowledge and skills to perform the behaviour. Again, information or educational interventions can potentially impact vaccination behaviours by targeting any of the contributing factors identified in these models. For instance, learning about vaccine safety and efficacy may foster positive attitudes towards vaccines. Interventions could increase parents' perceived behavioural control by helping them to understand how, where, and when to vaccinate their children.

Interventions that intend to inform or educate do not necessarily mean just passively giving people information. While this may be adequate when the primary barrier to vaccination uptake is a lack of knowledge or a lack of access to necessary information, there is considerable evidence to suggest that interventions that just provide information are not necessarily sufficient to change behaviour (Goldstein 2015; Ryan 2014). In some cases, this may be because external barriers still exist, such as barriers to availability of the vaccine and access to the service. In other cases, people may not actually receive the intended information, they may not be able to understand it, they may not trust it, or it may be inaccurate. For certain individuals with firmly‐rooted opposing beliefs, receiving factual information may actually backfire, and further entrench their opposing views (Nyhan 2014; Nyhan 2015). Therefore, it is important to recognise that information or educational interventions can involve much more than information provision ‐ they can be interactive, tailored, and targeted. Increasingly, interventions are designed with a theoretical underpinning, and are supported by qualitative data (Corace 2016; Jones 2014; Leask 2012). An appropriately designed information or educational intervention can potentially influence not only parents' knowledge, but also their attitudes, perceptions about their peers, sense of self‐efficacy, intention to vaccinate, and ultimately, their vaccination behaviours.

Why it is important to do this review

Interventions whose manifest purpose is to inform or educate people about vaccination are extremely common, reflecting a focus on information‐giving and the provider‐parent communication encounter as potential strategies to influence vaccination decision making. Despite this, few trials and no systematic reviews of randomised controlled trials (RCT) have evaluated the effects of face‐to‐face communication, and most research is qualitative or observational. In addition, many trials in the vaccination communication area do not measure key intermediate outcomes that help decision‐makers understand whether, and how, an intervention works (Kaufman 2016). Therefore, there is an urgent need to establish a rigorous evidence base that considers the full range of relevant outcomes to guide implementation decisions and practice guidelines (Leask 2014).

This review's comprehensive, global approach includes face‐to‐face communication that aims to address both awareness and acceptance of vaccination. While much of the research about vaccine hesitancy and provider‐parent communication focuses exclusively on HIC settings, face‐to‐face communication is a globally‐relevant strategy that is widely used in LMICs as well as HICs. This topic was identified as important by vaccination programme managers, policymakers, researchers and other stakeholders from LMICs in deliberative forums that informed the original review (Lewin 2011). It is critical to determine the effects of face‐to‐face interventions ‐ including their content, format and timing ‐ so that valuable resources can be allocated appropriately.

Relationship to other reviews

Since this review's original publication in 2013, there have been a number of new or updated systematic reviews published on related topics. In Table 1, we briefly summarised several relevant non‐Cochrane reviews. These reviews generally differ from ours in that they included a broader range of intervention types or study designs; they did not incorporate risk of bias assessment in the presentation of their results; they included vaccinations for adolescents or adults as well as children; or they focused exclusively on LMICs. While there was some variation in the conclusions drawn by the authors of these reviews, most agreed that there was a shortage of high‐quality RCT evidence on the effects of information or educational interventions for childhood vaccination.

Table 1. Summary of related non‐Cochrane reviews

Review	Title	Review focus and main findings	Relationship / key differences
Bright 2017	A systematic review of strategies to increase access to health services among children in low and middle income countries	Focus: effectiveness of interventions aimed at increasing access to health services for children aged 5 years and below in LMICs. Supply side interventions included: delivery of services at or closer to home and service level improvements (e.g. integration of services). Demand side interventions included: educational programmes, text messages, and financial or other incentives. Included studies: 57 RCTs and non‐randomised controlled trials Findings: health promotion or educational programmes were the most commonly evaluated intervention in the review. Education delivered by healthcare workers generally showed a positive impact on healthcare utilisation or immunisation uptake, but mixed effectiveness when delivered by community health workers	Broad range of interventions addressing health topics beyond vaccination and limited to LMICs.
Connors 2017	Provider‐parent communication when discussing vaccines: a systematic review	Focus: determining effective communication practices for providers and parents, and establishing whether collaborative and participatory communication between parents and providers impact parents’ views on vaccination. Included studies: 9 studies, mostly descriptive and qualitative, with 1 RCT Findings: evidence was limited and low quality. Themes of trust and support suggested the value of participatory communication, though opposing evidence in favour of presumptive communication also existed.	Included qualitative studies, found similar dearth of high‐quality evidence on effectiveness. The RCT tested an intervention directed to physicians, so was not included in our review.
Harvey 2015	Parental reminder, recall and educational interventions to improve early childhood immunisation uptake: a systematic review and meta‐analysis	Focus: effectiveness of remind, recall, and educational interventions for childhood vaccination uptake Included studies: 28 controlled studies, 13 focusing on reminders and 17 on parental education Findings: postal and telephone reminders combined showed the greatest positive effect as a reminder intervention. Educational interventions may be more effective in LMICs. Discussion‐based education was more effective than written education.	Several studies appearing in our review also appeared in the meta‐analysis of educational interventions in this review, but there were additional non‐face‐to‐face educational interventions included in this review.
Jarrett 2015	Strategies for addressing vaccine hesitancy – a systematic review	Focus: identify, describe, and assess potential effectiveness of strategies to reduce vaccine hesitancy Included studies: 166 peer‐reviewed and 15 grey literature records reporting an evaluation of an intervention. There was no restriction on study design Findings: three intervention categories: dialogue‐based, incentive‐based, or reminder‐ and recall‐based interventions. Most interventions were dialogue‐based. The most effective strategies were multi‐component and tailored to specific populations and specific problems.	Broad range of included interventions, beyond face‐to‐face communication. Many non‐RCT study designs were included in this review.
Odone 2015	Effectiveness of interventions that apply new media to improve vaccine uptake and vaccine coverage	Focus: effectiveness of interventions that apply new media to promote vaccination uptake and increase vaccination coverage Included studies: 19 studies, most from the USA. 13 experimental and 6 observational. Findings: interventions included Facebook, SMS, YouTube videos, apps, software, email, and targeted websites. Some evidence that SMS, some websites or web portals, and computerised reminders may increase vaccination rates.	Included vaccines for children, adolescents, and adults. No face‐to‐face communication interventions, only new media
Sadaf 2013	A systematic review of interventions for reducing parental vaccine refusal and vaccine hesitancy	Focus: effectiveness of interventions to decrease parental vaccine refusal and hesitancy toward recommended childhood and adolescent vaccines Included studies: 30 studies, primarily before‐and‐after intervention design with some RCTs, non‐RCTs and evaluation studies. Most from USA Findings: three intervention categories: passage of state laws, implementation of school‐ and state‐level laws, and parent‐centred information or education. Education, particularly short pamphlets, most commonly studied, with heterogeneous formats and mixed effects. Overall, no convincing evidence on effective strategies. Authors noted few studies measured hesitancy‐relevant outcomes, and most evidence was of low quality.	Broad range of included interventions, beyond face‐to‐face communication
Williams 2014	What are the factors that contribute to parental vaccine‐hesitancy and what can we do about it?	Focus: barriers to vaccination and strategies to address vaccine hesitancy Included studies: seven studies, largely RCTs or CRCTs. Not all specifically recruited hesitant parents Findings: current data did not support one method for communicating with hesitant parents over another. Wide range of interventions used	Included interventions beyond face‐to‐face, also included adolescent (HPV) as well as childhood vaccines.

In addition to the relevant non‐Cochrane reviews outlined in Table 1, this review is also closely related to three Cochrane reviews (Ames 2017; Oyo‐Ita 2016; Saeterdal 2014). Authors of each of these reviews include members of the COMMVAC project team.

The Oyo‐Ita 2016 review considered all interventions evaluated in LMICs to improve vaccination coverage. This included interventions such as incentives, education, supportive supervision, or outreach directed at parents, caregivers, and communities. The review found fourteen studies conducted in LMICs, thirteen of which were assessed to be at high risk of bias. The majority involved a communication element (e.g. health education, home visits, information campaigns). There was generally low‐ to moderate‐certainty evidence that health education and information campaigns improved vaccination coverage in LMIC settings. Our review is global in scope and focuses on face‐to‐face communication only, while the Oyo‐Ita review included all intervention types, but focused on studies conducted only in LMICs. Three studies appeared in both reviews, but these studies were contextualised in different ways, based on the scope of the respective reviews (Bolam 1998; Usman 2009; Usman 2011).

Saeterdal 2014 included vaccination communication interventions aimed at communities, which we did not include in this review. The authors defined 'community‐aimed' interventions as 1) interventions directed at a geographic area, 2) interventions directed to groups of people who shared at least one common social or cultural characteristic, or both. Our review focused on interventions directed to parents, individually or in groups, whereas Saeterdal 2014 included interventions directed to groups, which may or may not have included parents. The authors found low‐certainty evidence from two cluster‐RCTs that interventions aimed at communities may improve attitudes towards vaccination, and probably increased vaccination uptake under some circumstances, when compared with routine practices. This review is currently being updated.

The Ames 2017 review included qualitative studies exploring parents' and caregivers' views and experiences of communication about childhood vaccinations, and the influence of communication on their decision making. Ames 2017 used thematic analysis to synthesise data from 38 studies, primarily from high‐income countries. The authors had high confidence in the evidence for the following findings: parents wanted more information than they receive; parents wanted balanced information about benefits and harms of vaccination; parents viewed healthcare workers as important sources of information; and parents found it difficult to know what information sources to trust, or how to find unbiased information. The authors compared their qualitative findings with the results of the trials included in the originally published version of this review (Kaufman 2013), and in Saeterdal 2014. They determined that most of the interventions evaluated in the trials addressed at least one aspect of communication that was important to parents (e.g. tailoring information to parent needs).

Objectives

available in

Methods

available in

Criteria for considering studies for this review

Types of studies

Randomised controlled trials (RCTs) and cluster‐RCTs. We excluded all studies rated at a high risk of bias on the random sequence generation criteria of the 'Risk of bias' tool, because we considered these studies to be quasi‐RCTs.

Types of participants

Communication interventions about childhood vaccination are often complex because multiple participant groups are involved in the delivery and receipt of the intervention. The intervention is delivered to one group (parents) to promote vaccination, which is administered to another group (children). The planning and implementation of the intervention or the vaccination programme itself is addressed by a third group (program organisers). The three participant groups were:

Children: infants (less than 1 year) or preschool‐aged children (1 to 5 or 6 years). We only included RCTs with school‐aged children if the main focus of the intervention was vaccines whose primary series began in infancy or preschool‐aged children.
Parents: parents, guardians, or others fulfilling the parental role, alone or in groups, targeted to receive face‐to‐face information or education, and who had at least one child due or overdue for childhood vaccinations. We also included participants who were expectant parents, individuals or couples currently pregnant, considering adoption, or otherwise expecting to become guardians of a child. The intervention could have been directed to parents individually or in groups.
Vaccine program organisers: anyone involved in the planning or implementation of vaccination programmes or interventions

Types of interventions

Face‐to‐face communication interventions directed to parents to inform or educate them about routine childhood vaccinations. Such interventions describe or impart information about some feature of routine childhood vaccination with the purpose of changing parent knowledge, beliefs, attitudes, behaviour, or self‐efficacy. The type of content that may be covered includes information about: vaccine‐preventable diseases (e.g. symptoms, prevalence, transmission, severity); vaccines (e.g. delivery method, dose, ingredients, schedule, risks or side effects, benefits); or vaccine service delivery (e.g. where to go to receive vaccinations, costs, clinic opening hours, services to assist with access). Despite potential heterogeneity of intervention content or intensity across studies, we agreed that interventions with the purpose of informing or educating parents about vaccination were sufficiently similar to conceptually group for analysis in this review.

We included interventions delivered by anyone, including physicians, nurses, midwives, health visitors, or other healthcare professionals; trained volunteers; lay health workers; members of the community; or peers.

The term 'routine vaccinations' means all routine childhood vaccines outlined by the WHO (WHO 2018). Human papillomavirus vaccine (HPV) was excluded, because it is delivered to adolescents. This review focused only on interventions about early childhood vaccines because interventions related to vaccines for older children may be significantly different in nature (e.g. they may target children directly, or have a reduced focus on the parent's role in decision making).

We included face‐to‐face interventions conducted in cluster‐RCTs in the context of a mass vaccination campaign if it was possible to isolate and report the effects of the face‐to‐face communication interventions delivered to parents for the vaccination of young children or infants from the larger campaign. Similarly, we included multi‐component interventions with a face‐to‐face element if the outcomes of the face‐to‐face intervention alone could be determined from the reported data. For example, trials of multi‐component interventions involving face‐to‐face education plus reminders or health service access assistance measured both vaccination status and knowledge (Quinlivan 2003; Wood 1998). We could not isolate the effect of the face‐to‐face intervention on the vaccination rate, but we could attribute any changes in knowledge to the face‐to‐face educational intervention alone. We may also have been able to determine the effect of a face‐to‐face intervention if a trial measured outcomes at multiple or staged time points (or both), including assessment of the effects of one intervention component or type before the addition of other intervention types. We did not consider Interventions to be multi‐component if a secondary form of information or education (e.g. a pamphlet) was provided along with the face‐to‐face intervention that was specifically described as supplementary or supporting.

In this updated review, we tightened our inclusion criteria with regard to complex, multi‐topic, early parenting education and information interventions (e.g. well‐child appointments or home visits). Such interventions generally cover a broad range of issues, which may be tailored to the individual participant. It is often unclear whether and what kind of vaccination information is discussed, or whether every participant receives the same information or education related to vaccination. These trials may measure vaccination uptake or status, but this may be used as a general health access indicator, and does not necessarily confirm that vaccination was a topic of discussion (or indeed a focus of the intervention). This review focused on face‐to‐face information or education that was specifically about vaccination, so our updated process for determining inclusion or exclusion of such studies was as follows:

If the trial specifically described the vaccination content in the intervention, and it met other inclusion criteria, we included it.
If the trial briefly mentioned that vaccination was a topic covered by the intervention, but did not describe the content in any detail, we included it only if it also measured a vaccination‐specific knowledge, attitude, or intention outcome (indicating that vaccination was covered to some degree within the content of the intervention).
We excluded trials that only mentioned vaccination briefly or not at all, and measured only vaccination status, on the grounds that the intervention was not primarily an information or educational intervention about vaccination.

We welcome contact from any authors who believe their studies may have been erroneously excluded based on our interpretation of the trial report.

This updated review addressed two comparisons:

Face‐to‐face interventions directed to parents versus control (usual care or passive intervention, i.e. non‐face‐to‐face information or education, or no intervention),
Face‐to‐face intervention A versus face‐to‐face intervention B.

We reduced the comparisons from the original review, which considered the effects of the intervention when directed to individual parents or to groups of parents. There was no clear evidence to suggest that education delivered in a group setting was likely to work differently from education delivered to individuals, and so we felt that this comparison was less informative for end users of the review.

We included face‐to‐face interventions designed to inform or educate, which may have included oral sessions, lectures, one‐on‐one or group classes or seminars, information sessions, home visits, or outreach sessions.

We did not include community‐directed interventions, as these were considered in the Saeterdal 2014 review.

Types of outcome measures

Primary outcomes

Children: vaccination status of child (i.e. vaccination status up‐to‐date, or receipt of one or more vaccines, as defined by study authors); outcome domain: vaccination status and behaviours
Parents: knowledge or understanding of vaccination; outcome domain: knowledge or understanding
Parents: attitudes or beliefs about vaccination; outcome domain: attitudes or beliefs
Parents: intention to vaccinate child; outcome domain: attitudes or beliefs
All categories: adverse effects; outcome domain: any

Secondary outcomes

Parents: parent experience of intervention (e.g. satisfaction, assessment of communication); outcome domain: communication delivery and design
Vaccine programme managers: cost of implementing intervention; outcome domain: cost

Justification of outcome measures

Our recent research to define and prioritise core outcome domains for the evaluation of vaccination communication interventions informed the selection of outcomes for this review (Kaufman 2017; Kaufman 2017a). First, we developed a taxonomy of potential vaccination communication outcomes that were derived from trials, non‐vaccination health communication studies, and focus groups with stakeholders (parents, healthcare providers, researchers, and policymakers; Kaufman 2017). This taxonomy organised outcomes into eight domains: 1) knowledge or understanding, 2) attitudes or beliefs, 3) vaccination status and behaviours, 4) communication delivery and design, 5) community participation, 6) decision making, 7) health status and well‐being, and 8) cost.

Using a Delphi survey, we asked representatives from each stakeholder group to rate the relative importance of each of these outcome domains when evaluating a communication intervention to inform or educate about vaccination (Kaufman 2017a). The top four domains for this type of intervention, according to stakeholders, were 'knowledge or understanding', 'attitudes or beliefs', 'vaccination status and behaviours', and 'communication delivery and design'. Therefore, we ensured that outcomes from each of these domains were captured by this review.

While changes in knowledge are not always directly linked to changes in health behaviours, we included knowledge as a primary outcome because improving knowledge was the stated purpose of many programmes and interventions (Ryan 2014). Particularly with complex communication interventions, it was important to measure intermediate outcomes that reflected the intervention's purpose, in addition to endpoint outcomes, such as vaccination status (Craig 2008; Moore 2015; Petticrew 2011; WHO SAGE Working Group on Vaccine Hesitancy 2014). Doing so could help to unpack how and whether an intervention worked or where it broke down.

In our taxonomy of outcomes, intention to vaccinate fell under the domain of 'attitudes or beliefs'. However, we decided to include two separate outcomes associated with this domain; one broadly defined as 'attitudes or beliefs', and the more specific outcome of 'intention to vaccinate'. Intention and attitudes are separate determinants of behaviour in most behaviour change theories (see How the intervention might work), with intention more directly preceding behaviour change. Changes in intention, but not behaviour, may indicate the presence of external barriers to vaccination (daCosta 2005). In comparison, changes in attitudes may be particularly relevant for identifying subtle shifts in vaccine acceptance or hesitancy that are not reflected by changes to either intentions or behaviours.

We included the outcome 'parent experience of the intervention', because the way in which communication is delivered and received can substantially impact its overall effectiveness. For instance, parents cited poor communication experiences with healthcare providers made them less likely to consider or undertake vaccination (Leask 2012; Leask 2015).

While it was not prioritised in the Delphi survey, we also included cost as an outcome in this review. The cost of implementing an intervention is particularly important to record, if it is measured, to improve equity in healthcare delivery and to increase the global applicability of research evidence. Cost is an important factor for decision and policy makers, so we included it, where reported. We included adverse events to capture any potential negative effects of the interventions ‐ for example, anxiety or distress ‐ and because it is Cochrane policy for systematic reviews to consider adverse effects (Loke 2011).

We did not include or exclude studies on the basis of whether the chosen outcomes were measured or reported.

Search methods for identification of studies

Electronic searches

We searched the following sources:

Cochrane Central Register of Controlled Trials (CENTRAL; 2017, Issue 7) in the Cochrane Library (searched 3 July 2017);
MEDLINE Ovid (2012 to July 3 2017);
Embase Ovid (2012 to July 3 2017);
CINAHL EBSCO (Cumulative Index to Nursing and Allied Health Literature; 2012 to July 3 2017);
PsycINFO Ovid (2012 to July 3 2017).

We tailored strategies to each database and reported them in the appendices section of the review (Appendices 4 to 9). We had no language or date restrictions.

In this update, we did not search two databases previously searched for the original review: Global Health (CAB) and Global Health Library (WHO). For a full list of the searches run in the original review, see Appendix 3. These databases had a very low yield of relevant RCTs, and we determined that any studies indexed in these databases would also be found via other databases and resources.

Searching other resources

We searched all ongoing trials in ClinicalTrials.gov (searched July 2017) and the WHO International Clinical Trials Registry Platform (ICTRP) for any ongoing trials registered since 2012 (searched July 2017). We contacted authors to obtain further information or eligible data, if available.
We searched for grey literature in OpenGrey (http://www.opengrey.eu/; searched July 2017). The Grey Literature Report, a database searched in the original review, has been discontinued, so we did not search it for this update.
We searched the reference lists of all included papers, and any key papers in the field. We also searched the ISI Web of Science (both the Social Science Citation Index and the Science Citation Index; searched July 2017), and Google Scholar (searched August 2017) for papers that cited the studies included in the review. We contacted authors of included studies and vaccination experts and asked for additional references.

Data collection and analysis

Selection of studies

We combined all electronic database search results in Endnote and removed duplicate records. Two review authors (JK and LW) independently screened all titles and abstracts to assess which studies met the selection criteria. These authors also independently screened all ongoing trials, grey literature, and studies identified through reference list or reverse citation searching. We excluded studies that clearly did not relate to the selection criteria. We retrieved the full text of all studies determined to be potentially relevant. Two review authors (JK and LW) independently screened these studies for inclusion. They resolved disagreements through discussion with another review author (RR). We reported studies excluded at this stage in the screening process in the 'Characteristics of excluded studies' table. We provided citation details and any available information about ongoing studies, and collated and reported details of duplicate publications, so that each study (rather than each report) was the unit of interest in the review. We reported the screening and selection process in an adapted PRISMA flow chart (Stovold 2014).

Data extraction and management

Two review authors (JK and LW) independently extracted the data from included studies and discussed any disagreements with a third author (RR) until consensus was reached. We used a revised version of the data extraction form used for the original review, which combined features of the template developed by the Cochrane Consumers and Communication Review Group (CCC 2016), the Cochrane equity checklist (Ueffing 2012), and the COMMVAC extraction template. The data extraction form included the following components: details of study, participant characteristics, country and health system features, setting, intervention, intervention quality, co‐interventions, risk of bias, outcomes, and study conclusions. We reported this information for each included study in the 'Characteristics of included studies' table. One review author (JK) entered all data into Review Manager 5 (RevMan 5; RevMan 2014); a second review author (LW or RR) independently checked for accuracy against the data extraction sheets.

Assessment of risk of bias in included studies

We assessed and reported on the methodological risk of bias of included studies in accordance with the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011a), and the guidelines of the Cochrane Consumers and Communication Review Group (Ryan 2013), which recommend the explicit reporting of the following individual elements for RCTs: random sequence generation, allocation sequence concealment, blinding (participants, personnel), blinding (outcome assessment), completeness of outcome data, selective outcome reporting, other sources of bias, such as contamination. We considered blinding separately for different outcomes where appropriate (e.g. blinding may affect subjective outcomes differently from objective ones), but made an overall judgment of the risk of bias for this element. For each criteria, we described the relevant information provided by the authors and judged each criteria as being at high, low, or unclear risk of bias, as set out in the judging criteria provided in Higgins 2011. We deemed studies to be at the highest risk of bias if they scored high or unclear on either sequence generation or allocation concealment (based on growing empirical evidence that these factors are particularly important in influencing risk of bias; Higgins 2011).

For cluster‐RCTs, we assessed the risk of bias associated with an additional factor: selective recruitment of cluster participants (Ryan 2013).

In all cases, two authors (JK, LW or RR) independently assessed the risk of bias of included studies, and resolved any disagreements by discussion, to reach consensus. We contacted study authors for additional information about the included studies, and for clarification of the study methods, as required. We incorporated the results of the 'Risk of bias' assessment into the text of the review, through systematic description and commentary about each of the elements, and in standard tables, leading to an overall assessment of the risk of bias of included studies, and a judgment about the internal validity of the review's results.

Measures of treatment effect

For dichotomous outcomes (e.g. vaccination status), we analysed data based on the number of events and the number of people assessed in the intervention and comparison groups. We used these to calculate the risk ratio (RR) and 95% confidence intervals (CI).

For continuous measures (e.g. knowledge or understanding, attitudes, or beliefs), we analysed data based on the mean, standard deviation (SD), and number of people assessed, for both the intervention and comparison groups, to calculate the mean difference (MD) and 95% CI. If the MD was reported without individual group data, we had intended to use this to report the study results. If more than one study measured the same outcome using different scales, we calculated the standardised mean difference (SMD) and 95% CI using the inverse variance method in RevMan 5.

Vaccination status was defined slightly differently across studies (e.g. receipt of single or multiple vaccines; 'up to date' status for complete schedule). We accepted the definition of vaccination status used by study authors. We recorded these in the 'Characteristics of included studies' tables.

Where studies used two separate methods to measure the same outcome (e.g. two measures of parental knowledge), or measured two different outcomes that could be considered part of the same outcome category (e.g. receipt of one vaccine and completion of all vaccines, which both fall under 'vaccination status'), we adopted the approach outlined by Brennan and colleagues (Brennan 2009).

We selected the primary outcome identified by the study authors that correlated to our stated outcomes of interest.
If no primary outcome was specified, we selected the one specified in the sample size calculation.
If there was no sample size calculation, we ranked the reported effect estimates and selected the outcome with the median effect estimate. When there was an even number of outcomes, we included the outcome whose effect estimate was ranked n/2, where n was the number of outcomes.

Several studies measured multiple aspects of parent attitudes or beliefs (e.g. perceived severity of diseases, perceived benefit of vaccines, necessity of vaccines, self‐efficacy). We compared the descriptions of these outcomes across studies, and selected the outcome from each that was most similar, in terms of what was measured and the direction of the scale used. We described all measures of attitudes or beliefs reported in each study in Table 2.

Table 2. Additional measures of parent attitudes

Study	Outcome	Scale used
Jackson 2011	Attitude towards MMR One item: ‘For me to give my child the combined MMR vaccine at the recommended ages would be…’	1 (extremely negative attitude) to 7 (extremely positive attitude)
	Necessity beliefs Four items assessing the necessity of MMR: e.g. ‘Without the combined MMR vaccine, my child could get very ill from measles, mumps, or rubella’	Each item scored from 1 to 5 and summed, for total outcome scale of 4 (not at all necessary) to 20 (very necessary) We used this item in the meta‐analysis.
	Concern beliefs Four items assessing concerns about MMR: e.g. ‘Giving my child the combined MMR vaccine worries me’	Each item scored from 1 to 5, and summed for total outcome scale of 4 (not at all concerned) to 20 (very concerned)
Saitoh 2013	Perceived severity (HBM) Two items: e.g. ‘The vaccine‐preventable diseases are serious diseases.’	Each item scored from 1 to 5 (strongly disagree to strongly agree), and summed for a total outcome scale of 2 to 10. We used this item in the meta‐analysis.
	Perceived susceptibility(HBM) One item: ‘My baby is not very likely to get vaccine preventable diseases.’	Each item scored from 1 to 5 (strongly disagree to strongly agree), and summed for a total outcome scale of 1 to 5.
	Perceived benefits (HBM) Four items: e.g. ‘The vaccines for babies will prevent my baby from getting sick with a vaccine‐preventable disease.’	Each item scored from 1 to 5 (strongly disagree to strongly agree), and summed for a total outcome scale of 4 to 20
	Perceived barriers (HBM) Five items: e.g. ‘The vaccines will make my baby sick,’ ‘The vaccines do not prevent the vaccine‐preventable diseases.’	Each item scored from 1 to 5 (strongly disagree to strongly agree), and summed for a total outcome scale of 5 to 25
	Self‐efficacy (HBM) Two items: e.g. ‘I feel comfortable getting the vaccines for my baby.’	Each item scored from 1 to 5 (strongly disagree to strongly agree), and summed for a total outcome scale of 2 to 10
	Perceived behavioral control (IBM) One item: ‘I have control over whether or not my baby gets the vaccines.’	Each item scored from 1 to 5 (strongly disagree to strongly agree), and summed for a total outcome scale of 1 to 5
	Social norm (injunctive) Four items: e.g. ‘Most people important to me think I should get my baby the vaccines.’	Each item scored from 1 to 5 (strongly disagree to strongly agree), and summed for a total outcome scale of 4 to 20
	Social norm (descriptive) Two items: e.g. ‘I know other people my age who got their baby the vaccines.’	Each item scored from 1 to 5 (strongly disagree to strongly agree), and summed for a total outcome scale of 2 to 10
Saitoh 2017	Perceived severity (HBM) Two items: e.g. ‘Vaccine preventable diseases are serious diseases.’	Each item scored from 1 to 5 (strongly disagree to strongly agree), and summed for a total outcome scale of 2 to 10 We used this item in the meta‐analysis.
	Perceived susceptibility(HBM) One item: ‘My baby is not very likely to get vaccine preventable diseases.’	Each item scored from 1 to 5 (strongly disagree to strongly agree), and summed for a total outcome scale of 1 to 5
	Perceived benefits (HBM) Five items: e.g. ‘The vaccines for babies will prevent my baby from getting sick with vaccine‐preventable diseases.’ ‘If my baby receives his or her vaccines, it will help protect my friends and family from getting vaccine‐preventable diseases.’	Each item scored from 1 to 5 (strongly disagree to strongly agree), and summed for a total outcome scale of 5 to 25
	Perceived barriers (HBM) Six items: e.g. ‘I feel uncomfortable because vaccines are painful or uncomfortable for my baby,’ ‘Vaccines are too expensive.’	Each item scored from 1 to 5 (strongly disagree to strongly agree), and summed for a total outcome scale of 6 to 30
	Perceived behavioral control (IBM) One item: ‘I have control over whether or not my baby gets the vaccines.’	Each item scored from 1 to 5 (strongly disagree to strongly agree), and summed for a total outcome scale of 1 to 5
	Social norm (injunctive) Three items: e.g. ‘Most people important to me think I should have my baby vaccinated.’	Each item scored from 1 to 5 (strongly disagree to strongly agree), and summed for a total outcome scale of 3 to 15
	Social norm (descriptive) Two items: e.g. ‘I know other people my age who had their babies vaccinated.’	Each item scored from 1 to 5 (strongly disagree to strongly agree), and summed for a total outcome scale of 2 to10

Where studies recorded outcome data at multiple time points, we reported the data from the final follow‐up, because this time point may capture people who were delayed in obtaining vaccination directly following the intervention. We extracted outcome data recorded at other time points and reported these in Additional tables 8 to 12.

Unit of analysis issues

The two cluster‐RCTs used intraclass correlation coefficient (ICC) to adjust their sample sizes to account for clustering, and conducted appropriate analysis, but did not report their effective sample sizes. Therefore, we recalculated effective sample sizes based on information reported in each study, and divided the reported sample size by the design effect (Higgins 2011). In Jackson 2011, the authors used longitudinal analysis because they felt that the participant and cluster numbers were too small to use multilevel modelling. Saitoh 2017 used a hierarchical linear mixed‐effects model to account for repeated measures, by adjusting for interactions between groups, time, and groups x time (as fixed‐effect) and participants (as random‐effects).

We calculated the design effect (DE), using the ICC reported in each study (0.05) and the number of clusters in each study. For each reported outcome, we divided the original sample size by the DE to establish the effective sample size (Higgins 2011; McKenzie 2016). For the dichotomous outcome of vaccination status, we also divided the number of events occurring in each group by the DE. We reported the adjusted sample sizes in all meta‐analyses, and reduced the weightings given to these studies.

Saitoh 2017 reported a DE of 1.95, based on an average cluster size of 20. There were 100 participants in the intervention group and 88 in the control group, which we adjusted to 51 (intervention) and 45 (control) for all outcomes.

Jackson 2011 reported a DE of 1.5, based on an average cluster size of 11. There were 68 in the intervention group and 67 in the control group, which we adjusted to 45 (intervention) and 45 (control) for all outcomes.

Dealing with missing data

Where necessary, we contacted study authors for missing outcome data, missing study‐level participant characteristics, or missing summary data. We had intended to impute missing summary data where possible, and report any assumptions in the results tables. We investigated the effect of our choice of any imputed data, including ICCs, on the pooled effect estimate through sensitivity analyses.

Assessment of heterogeneity

We assessed statistical heterogeneity through visual inspection of forest plots, and the Chi² test for heterogeneity. We quantified the heterogeneity using the I² statistic; however, we did not set a threshold for acceptable heterogeneity. As Pigott and Shepperd explain, complex interventions are much more likely to appear heterogeneous, but this does not necessarily mean they are not comparable (Pigott 2013). The decision to meta‐analyse or not should ultimately be made by the reviewers. In this review, the interventions were very tightly defined, even though there could be variability in design or delivery. The target population was parents in all instances, regardless of setting. Therefore, while we anticipated some variability, we expected that the intervention would be working in largely the same way across included populations. We discuss the potential sources and implications of heterogeneity in our analysis, and assessment of the certainty of the evidence, using GRADE methodology.

Assessment of reporting biases

We assessed reporting bias qualitatively, based on the characteristics of the included studies (e.g. if only small studies were identified that indicated positive findings in favour of face‐to‐face information or educational interventions, or if information that we obtained from contacting experts in the area and the authors of retrieved studies suggested that there were unpublished studies). We did not find sufficient RCTs (at least 10) to construct a funnel plot to formally investigate small study effects or publication bias (Higgins 2011).

Data synthesis

We presented summary statistics for each of our outcomes in table form (see Table 3; Table 4; Table 5; Table 6; Table 7). These tables include data for each study group and the timing of outcome assessments.

Table 3. Vaccination status outcome data

Session length	Outcome	Study	Timing of outcome assessment (days/months)	Intervention group		Control group		Notes
Session length	Outcome	Study	Timing of outcome assessment (days/months)	Observed (n)	Total (N)	Observed (n)	Total (N)	Notes
Shorter sessions (1 to 10 min)	Up‐to‐date immunisation status at 3 months of age for Hib, HBV, and PCV7 vaccines	Saitoh 2013	3 months post‐intervention	24	70	3	36	One 10‐minute session Intervention participants = combined prenatal intervention arm and postnatal intervention arm
	Up‐to‐date immunisation status at 6 months of age for Hib, PCV13, DTaP‐IPV, HBV, and rotavirus	Saitoh 2017	6 months post‐intervention	22	51	21	45	Three 5‐minute sessions Figures adjusted to account for clustering (original intervention N = 100; control N = 88)
	Receipt of DPT3	Usman 2009	90 days post‐intervention	242	375	205	375	One 2‐ to 3‐minute session
	Receipt of DPT3	Usman 2011	90 days post‐intervention	228	376	149	378	One 2‐ to 3‐minute session
Longer sessions (11 min+)	Receipt of MMR vaccine	Jackson 2011	3 months post‐intervention	18	19	18	25	One 2‐hour session Figures adjusted to account for clustering (original intervention N = 29; control N = 37)
	Up‐to‐date status at 12 months of age (1 dose BCG, 3 doses HepB, 3 doses OPV, 3 doses DTP, 1 dose MR, 1 dose JEV)	Hu 2017	12 months post‐intervention	376	418	359	433	One 15‐minute session
	Up‐to‐date immunisation status at 3 months of age (BCG, > 2 doses diphtheria and pertussis vaccine, OPV)	Bolam 1998	3 months post‐intervention	Groups A + B = 179	Groups A + B = 205	Groups C + D = 169	Groups C + D = 198	One 20‐minute session Groups A + B = education at birth Groups C + D = no education at birth
	Up‐to‐date immunisation status at 6 months of age (BCG, 3 doses diphtheria and pertussis vaccine, OPV) time point not included in meta‐analysis	Bolam 1998	6 months post‐intervention	Group B = 100	Group B = 104	Group D = 91	Group D = 97	One 20‐minute session Group B = education at birth only Group D = no education

Table 4. Knowledge outcome data

Outcome	Study	Scale used	Timing of outcome assessment (days/months)	Intervention group			Control group			Notes
Outcome	Study	Scale used	Timing of outcome assessment (days/months)	mean/mean change	standard deviation	N	mean/mean change	standard deviation	N	Notes
Knowledge of immunisation schedule (multi‐component intervention – did not use immunisation status outcome)	Quinlivan 2003	0 to 10	6 months postpartum			62			62	'There were no significant differences in knowledge of infant vaccination schedules between the two groups at the antenatal (intervention: median 0 (IQR 0.1); control: 0 (0.2); P = 0·54) or postnatal assessments (4 (1.6); 2 (1.4); P = 0·08). The unadjusted mean difference in knowledge between the two groups was 0.85 points (95% CI –0.06 to 1.76)'
Knowledge of immunisation contraindications (multi‐component intervention – did not use immunisation status outcome)	Wood 1998	0 to 3	At initial interview, and interview following conclusion of intervention 15 mo later	2.7	1	185	2.6	1	180
Basic knowledge and knowledge of diseases	Saitoh 2013	0 to 13 VPD^a score	100 days post‐intervention	prenatal intervention 4.5	prenatal intervention 2.8	prenatal intervention 34	4.6	3.8	36	We combined the separate scores for VPD knowledge and basic knowledge into a single knowledge score, and data from the two intervention arms were combined into single intervention group in the meta‐analysis
		0 to 13 VPD^a score	100 days post‐intervention	postnatal intervention 4.5	postnatal intervention 2.9	postnatal intervention 36	4.6	3.8	36
		0 to 10 basic knowledge score	100 days post‐intervention	prenatal intervention 3.4	prenatal intervention 1.8	prenatal intervention 34	1.9	1.9	36
		0 to 10 basic knowledge score	100 days post‐intervention	postnatal intervention 2.6	postnatal intervention 1.5	postnatal intervention 36	1.9	1.9	36
Basic knowledge and knowledge of diseases	Saitoh 2017	0 to 13	1 month postpartum time point not included in meta‐analysis	9.8	2.1	51	9.8	1.9	45	4 questions about vaccine‐preventable diseases, 5 basic immunisation knowledge questions, and a question asking participants to identify four recommended vaccines from a list of 12 Figures adjusted to account for clustering (original intervention N = 100; control N = 88)
Basic knowledge and knowledge of diseases	Saitoh 2017	0 to 13	6 months postpartum	10.7	1.6	51	10.5	2.0	45
Knowledge of MMR	Jackson 2011	0 to 11	1 week post‐intervention time point not included in meta‐analysis	8.22	1.5699	45	7.83	1.2709	45	Figures adjusted to account for clustering (original intervention N = 68; control N = 67)
Knowledge of MMR	Jackson 2011	0 to 11	3 months post‐intervention	7.3	1.5632	45	7.08	1.183	45
Knowledge about vaccines, infectious diseases, contraindications, schedule	Bjornson 1997	16‐question test, scores reported for individual questions	immediately following intervention			128			99	test scores broken down by question rather than by person. Authors could not provide mean scores. This study was described narratively in the review.

^aVPD: vaccine‐preventable diseases

Table 5. Attitudes outcome data

Outcome	Study	Scale used	Timing of outcome assessment (days/months)	Intervention group			Control group			Notes
				mean/mean change	standard deviation	N	mean/mean change	standard deviation	N
Parents’ beliefs about the necessity of MMR	Jackson 2011	4 to 20	1 week post‐intervention time point not included in meta‐analysis	17.63	2.7267	45	17.18	2.8288	45	SD calculated from CI provided in study. Figures adjusted to account for clustering (original intervention N = 68; control N = 67) Four items (1 = strongly disagree to 5 = strongly agree) assessed parents’ beliefs about the necessity of the MMR vaccine, e.g. ‘Without the combined MMR vaccine, my child could get very ill from measles, mumps or rubella’ additional attitude measures reported separately
			3 months post‐intervention	17.43	2.3136	45	17.21	3.0748	45
Perceived severity of vaccine‐preventable disease	Saitoh 2013	2 to 10	100 days post‐intervention	prenatal int 8.6	prenatal int 1.3	prenatal int 34	8.2	1.5	36	Two items (1 = strongly disagree to 5 = strongly agree) assessing parents’ perceptions about the severity of vaccine‐preventable diseases In the meta‐analysis, we combined the data from the two intervention arms into a single attitude score for all intervention participants Additional attitude measures reported separately
				postnatal int 7.9	postnatal int 2	postnatal int 36
Perceived severity of vaccine‐preventable disease	Saitoh 2017	2 to 10	1 month post intervention time point not included in meta‐analysis	8.3	1.6	51	7.8	1.5	45	Two items (1 = strongly disagree to 5 = strongly agree) assessing parents’ perceptions about the severity of vaccine‐preventable diseases Figures adjusted to account for clustering (original intervention N = 100; control N = 88) Additional attitude measures reported separately
			6 months post intervention	8.3	1.8	51	8.3	1.6	45

Table 6. Intention to vaccinate outcome data

Outcome	Study	Scale used	Timing of outcome assessment (days/months)	Intervention group			Control group			Notes
Outcome	Study	Scale used	Timing of outcome assessment (days/months)	mean/mean change	standard deviation	N	mean/mean change	standard deviation	N	Notes
Intended MMR choice	Jackson 2011	1 to 7	1 week post‐intervention time point not included in meta‐analysis	6.03	1.4873	45	5.44	1.9679	45	SD calculated from CI provided in study. Figures adjusted to account for clustering (original intervention N = 68; control N = 67). Three items measured on a 7‐point scale e.g. ‘I intend to give my child the combined MMR vaccine at the recommended ages’ (definitely do not to definitely do). Responses averaged over the three items.
Intended MMR choice	Jackson 2011	1 to 7	3 months post‐intervention	6.34	1.4047	45	5.58	2.1319	45
Intent to immunise	Saitoh 2013	1 to 4	100 days post‐intervention	3.7288	0.4484	59	3.4	0.4983	30	Four‐point scale where 1 = no, 2 = undecided, 3 = yes, for a specific vaccine, and 4 = yes. Study authors presented the data for this outcome with both intervention arms combined. Rather than providing mean and SD, the authors reported the number of participants who selected each scale rating. For the meta‐analysis, we transformed these data into mean and SD, with N = total respondents for this question.

See: Summary of findings for the main comparison

Table 7. Adverse effects outcome data

Outcome	Study	Scale used	Timing of outcome assessment (days/months)	Intervention group			Control group			Notes
				mean/mean change	standard deviation	N	mean/mean change	standard deviation	N
Anxiety	Jackson 2011	20 to 80	1 week post‐intervention time point not included in meta‐analysis	30.89	11.9396	45	33.78	13.7341	45	SD calculated from CI provided in study. Figures adjusted to account for clustering (original intervention N = 68; control N = 67). Six items on a 4‐point scale (not at all to very much) e.g. ‘I feel calm’, ‘I am tense’. Positive items reverse scored, and all six items were summed. The total score was multiplied by 20/6. A normal score is 34 to 36.
			3 months post‐intervention	31.46	12.2701	45	33.39	13.5291	45

The decision to conduct a meta‐analysis was based on an assessment of whether the participants, setting, interventions, and outcome measures in the included trials were similar enough to draw meaningful conclusions from a statistically pooled result. Our consideration also depended on the availability of data from two or more primary studies for pooling. Due to the variability in the populations and interventions of the primary studies, we used a random‐effects model. One included cluster‐RCT was not included in the meta‐analysis because it did not report usable data (Bjornson 1997). Where possible, we pooled relative risks and calculated associated 95% confidence intervals to measure the effects of:

Face‐to‐face interventions directed to parents versus control (usual care or passive intervention, i.e. non‐face‐to‐face information or education, or no intervention), or
Face‐to‐face intervention A versus face‐to‐face intervention B.

We described the findings in the text of the review, with consideration of the potential impact of bias on the size or direction of the effect, the degree of heterogeneity and its possible sources, and their relevance to practice. Where we were unable to conduct a meta‐analysis, we grouped the data based on the comparison and outcome domain. Within each category, we presented the data in table format, and narratively described the results, grouped by outcome.

Subgroup analysis and investigation of heterogeneity

We did not plan any subgroup analyses at the protocol or original review stages. However, in this update, we made a post hoc decision to conduct two formal subgroup analyses to investigate potential sources of statistical heterogeneity in studies reporting vaccination status. The subgroups were relevant to practice and implementation, and were related to the delivery of the intervention (length of intervention session), and the number of vaccines received. Length of intervention delivery has important time and cost implications for decision makers and providers. We considered a subgroup analysis for the number of vaccines received following advice from vaccination experts, based on the possibility that receiving one vaccine may be less demanding, as an outcome, than receiving several.

Sensitivity analysis

We did not find enough studies to undertake a sensitivity analysis based on 'Risk of bias' assessment as planned. If we had found more studies, we had planned to remove those at the greatest risk of bias from the analyses. We also had intended to conduct sensitivity analyses to check the effects of imputed data (including imputation of ICC values), and to compare fixed‐effect and random‐effects analyses, in the event that small study effects were identified by funnel plots.

Assessing the certainty of the evidence

Two authors (JK, RR) independently assessed the certainty of the evidence, using the GRADE criteria. We assessed and reported the certainty of the evidence for each outcome, assessed against concerns of risk of bias, inconsistency, imprecision, indirectness, and publication bias, based on the methods described in chapter 11 of the Cochrane Handbook for Systematic Reviews of Interventions (Schünemann 2011).

We prepared summary of findings Table for the main comparison to present the results for each of the major primary outcomes, including potential harms, as outlined in the Types of outcome measures section. We converted results into absolute effects when possible, and provided a source and rationale for each assumed risk cited in the table.

Consumer participation

Those with a potential interest in this review include vaccine programme managers, policy makers, practitioners, and parent interest groups. In the peer review process for the protocol and the review, we sought external referees reflecting these interests (Kaufman 2012; Kaufman 2013). In addition, the outcomes selected in this review reflect the findings of a multi‐stage body of work that sought the views and outcome preferences of parents, healthcare providers, policy makers, and researchers (Kaufman 2017; Kaufman 2017a).

Results

Description of studies

Results of the search

In total, we identified 8247 new records through electronic database searches and 1416 records from other sources (Figure 1). After removing duplicates, we screened the title and abstract of 7177 reports, excluding 7141. We gathered 33 full‐text articles that appeared to be relevant (31 from databases and 2 from citation searches), as well as 3 records of ongoing studies (see Ongoing studies). Of these, we excluded 29 articles and records (see 'Characteristics of excluded studies' table).

Figure 1

Study flow diagram: 2018 review update

We excluded one previously included study, due to the decision to exclude multi‐topic, early parenting interventions from this update (Bartu 2006). We also included one previously excluded study, based on our revised assessment of the informational and educational nature of the intervention (Jackson 2011).

Included studies

We included a total of 10 trials in this update. In addition to the six trials from the original review, we added four new trials described in seven papers (see 'Characteristics of included studies' table). We identified all the new trials through database searches. We found three ongoing studies by our searches of trial registries (JPRN‐UMIN000012575; NCT02666872; NCT02984007). We contacted the authors of all included studies to clarify methods or to request additional data or intervention details. Some study authors responded with information on most, but not all queries, and one study author did not respond.

Study design

Three studies were cluster‐randomised controlled trials (RCT; Bjornson 1997; Jackson 2011; Saitoh 2017). The remaining seven were RCTs.

Sample size

Five studies involved between 100 and 250 participants, and five had more than 400 participants, including three studies with over 750 participants (Hu 2017; Usman 2009; Usman 2011). In two cluster trials, the effective sample sizes were calculated to be less than 100 (Jackson 2011; Saitoh 2017).

Participants and setting

The interventions were directed to mothers and parents, or expectant mothers or parents, and largely took place in clinics, hospitals, or other healthcare settings. Two studies targeted mothers for whom additional barriers to accessing vaccination exist (adolescent mothers, mothers of low socioeconomic status) (Quinlivan 2003; Wood 1998). Most studies took place in high‐income countries: Australia, Canada, China, England, Japan (two studies), and the USA, with the exception of Nepal, and Pakistan (two studies). Study features and settings are outlined in Table A.

Table A: Study features and settings

		Bjornson 1997	Bolam 1998	Hu 2017	Jackson 2011	Quinlivan 2003	Saitoh 2013	Saitoh 2017	Usman 2009	Usman 2011	Wood 1998
	Study design	cluster‐RCT	RCT	RCT	cluster‐RCT	RCT	RCT	cluster‐RCT	RCT	RCT	RCT
Location	Country	Canada	Nepal	China	England	Australia	Japan	Japan	Pakistan	Pakistan	US
Location	Income level	high	low	high	high	high	high	high	lower mid	lower mid	high
Setting	Healthcare centre (e.g. clinic, hospital, EPI centre)		X	X	X		X	X	X	X
	Other				childcare centres
	Parents' home		X			X					X
	Antenatal classes	X

Interventions

Six studies had a single intervention arm and a single control arm (Bjornson 1997; Hu 2017; Jackson 2011; Quinlivan 2003; Saitoh 2017; Wood 1998; Table B). The face‐to‐face education in Bjornson 1997 was actually the control arm, which was compared to an educational video intervention. For the purposes of this review, that designation was reversed. Four studies had additional intervention arms. There were two intervention groups in Saitoh 2013 (Group 1: prenatal education; Group 2: postnatal education). We combined data from these into a single intervention group (receiving education either pre‐ or postnatally). There were three relevant intervention arms in Bolam 1998 (Group A: education at birth and three months after birth; Group B: education at birth only; Group C: education at three months after birth only). The outcome data we included in this review was that recorded at three months post‐natally, when Groups A and B had both received education at birth (combined into a single intervention group) and Groups C and D (control) had not. Usman 2009 and Usman 2011 measured identical interventions delivered to populations in different areas (urban and rural). Each featured one relevant intervention arm (centre‐based education only), and two arms that were not relevant to this review (re‐designed immunisation card with or without centre‐based education).

Table B: Intervention features

		Bjornson 1997	Bolam 1998	Hu 2017	Jackson 2011	Quinlivan 2003	Saitoh 2013	Saitoh 2017	Usman 2009	Usman 2011	Wood 1998
Relevant arms	1 intervention arm	x		x	x	x		x	x	x	x
	2 intervention arms						x Grp 1: prenatal Grp 2: postnatal
	3 intervention arms		x Grp A: birth + 3 mo postnatal Grp B: birth only Grp C: 3 mo postnatal only^a
Control	No education or not described		x	x			x no info about vaccines of interest (only on vaccines required by law)		x	x
Control	Other education	x video			x MMR info leaflet	x routine support and info		x leaflet with general vaccination info			x health passport with general vaccination info
Frequency	Once	x	x Grps A + B^a	x	x		x		x	x
	Twice
	3+					x 6 times		x 3 times			x ˜ 4 times
Timing	Prenatal	x		x			x Grp 1: 34 to 36 wks' gestation	x 34 to 36 wks' gestation
	At birth, immediately postnatal		x Grps A + B: before postpartum hospital discharge^a				x Grp 2: 3 to 6 days post‐delivery	x 3 to 6 days post‐delivery
	> 1 week postnatal				x when child eligible for 1^st or 2^nd MMR dose	x 1 wk, 2 wks, 1 mo, 2 mo, 4 mo, and 6 mo postpartum		x at 1 mo well‐baby checkup	x at DPT1 visit (approx 6 wks of age)	x at DPT1 visit (approx 6 wks of age)	x 6 wks, 3.5 mo, and 5.5 mo of age
Duration	Short (1 to 10 min)	8 min					10 min	5 min	2 to 3 min	2 to 3 min
Duration	Long (11+ min)		20 min	15 min	2 h	1 to 4 h					approx 22 min
Group or individual	Group	x			x
Group or individual	Individual		x	x		x	x	x	x	x	x
Content	Immunisation only	x		x	x		x	x	x	x
Content	Multiple child health topics		x			x					x

^aThis review reports the 3‐month time point, where Groups A + B (intervention received once) were compared to Groups C + D (no intervention). At the 6‐month time point, reported in the study but not used in the meta‐analysis, Group A had received the intervention twice (birth and 3 months postpartum), Group B received it once (birth only), and Group C received it once (3 months postpartum only). Complete details for all intervention arms are available in Additional Tables.

Comparison 1: Face‐to‐face interventions directed to parents versus control

All included studies assessed face‐to‐face interventions directed to parents, but there were variations in their intensity, content, length, and control group intervention.

Seven studies evaluated a single intervention session, and three evaluated multi‐session interventions. Most interventions focused exclusively on delivering vaccination information, but interventions in three studies addressed a wider range of child health topics (e.g. breastfeeding, oral rehydration; Bolam 1998; Quinlivan 2003; Wood 1998).

There was considerable variation in the length of individual educational sessions evaluated across studies. Sessions were short (ten minutes or less) in five studies, and longer than 15 minutes in the remaining studies, including very long sessions ‐ over an hour ‐ in two studies (Jackson 2011; Quinlivan 2003).

In four studies, it appeared there was no education provided to the control group (Bolam 1998; Hu 2017; Usman 2009; Usman 2011). In one study, control group participants received routine information, but only about those vaccines required by law (Saitoh 2013). Because the intervention focused on different vaccines that were not legally required (Hib, HBV, and PCV7), we categorised the control group for this study as not receiving any education. Control groups in three studies received printed educational materials describing MMR (Jackson 2011), or general routine vaccination information (Saitoh 2017; Wood 1998). In one study, control participants received some routine vaccination information, but the format in which it was provided was not described (Quinlivan 2003). The control group in Bjornson 1997 received an educational video covering the same topics as the face‐to‐face education.

Comparison 2: Face‐to‐face intervention A versus face‐to‐face intervention B

No studies compared different types of face‐to‐face interventions.

Outcomes

Vaccination status

Please see Appendix 4 for the glossary of vaccination acronyms.

Vaccination status was measured in nine of the ten included studies. However, in two studies that assessed multi‐component interventions, the effect of the face‐to‐face intervention on changes to vaccination status could not be isolated from the effects of other components of the intervention. Therefore, we could not report on this outcome for these studies (Quinlivan 2003; Wood 1998).

Of the remaining seven studies, four measured appropriate or up‐to‐date vaccination for multiple vaccines at a particular age, and three measured receipt of a single vaccine (MMR or DPT). Outcomes were assessed in different ways. The final (or only) measurement time point for most studies was approximately three months post‐intervention, but Saitoh 2017 measured this outcome at six months and Hu 2017 at 12 months. Bolam 1998 measured vaccination status at both three and six months. We reported the three‐month time point because this allowed for the comparison of the greatest number of participants (i.e. the combination of intervention Group A and Group B (education at birth) compared with intervention group C and control group D (no education)).

Table C: Vaccination status outcome features

		Bolam 1998	Hu 2017	Jackson 2011	Saitoh 2013	Saitoh 2017	Usman 2009	Usman 2011	Wood 1998	Quinlivan 2003
Outcome	Vaccination up‐to‐date	x	x		x	x			x (data unusable)	x (data unusable)
Outcome	Receipt of single vaccine		x^a	x	x^a	x^a	x	x
Outcome timing	3 months	x		x	x (100 days post int)		x	x
	6 months	x^b				x				x
	12 months		x
	15 months								x
Assessment tool	Parent interview	x
	Review of records		x				x		x
	Questionnaire /test/ survey			x	x	x			x	x
	Not described							x

^aSecondary vaccination outcome reported by study but not used in meta‐analysis

^bAdditional time point reported by study but not used in meta‐analysis

Knowledge or understanding of vaccination

Five studies contributed usable data for this outcome. The final assessment was at three or six months post‐intervention. Wood 1998 measured knowledge at 15 months postpartum, but this was a multi‐session intervention that included a variable number of home visits, and the time between final session and outcome assessment was not stated. All data were collected through pre‐ and post‐tests (written or oral); higher scores corresponded to greater knowledge.

Jackson 2011 measured knowledge of MMR on a scale from 0 to 11. Quinlivan 2003 measured knowledge of the immunisation schedule using a scale ranging from 0 to 10. Saitoh 2013 and Saitoh 2017 both measured knowledge in similar ways. The knowledge assessment in Saitoh 2017 comprised four questions about vaccine‐preventable diseases, five basic immunisation knowledge questions, and a question asking participants to identify four recommended vaccines from a list of 12 (combined for a total score of 13). We received clarification from the study authors around the scales used in Saitoh 2013. Participants in this study were scored separately for correct identification of vaccine‐preventable diseases (0 to 13 points) and basic immunisation knowledge (0 to 10). We combined the scores for these two tests in our analysis ‐ this is explored further in the Effects of interventions section. Saitoh 2013 also included a self‐reported knowledge test, which was not an objective measure of knowledge and was not included in this review. Wood 1998 used two tests to measure knowledge of the immunisation schedule and contraindications. Data from the immunisation schedule test were reported inconsistently, so for the purposes of this review, only scores relating to knowledge of contraindications for immunisation could be used. The test featured three questions with a scale of 0 to 3.

Two studies did not contribute usable data on knowledge or understanding to the review (Bjornson 1997; Hu 2017), and the data from Quinlivan 2003 were skewed, so were not included in the meta‐analysis. Bjornson 1997 assessed knowledge of vaccines, infectious diseases, contraindications and the immunisation schedule using a 16‐question test, administered immediately before and directly following the delivery of the intervention session. The authors reported the percentage of participants who got each of the 16 questions correct on each test, but did not provide the mean scores of the intervention and control clusters, so the data could not be used in this review. We contacted the study authors, but they were unable to provide additional data, due to the age of the study.

In Hu 2017, the authors stated that they measured participant knowledge of the vaccine schedule and vaccine policy six months after the intervention was delivered. Participants were asked to correctly select 10 vaccines for a total of five points, and correct vaccine policy answers were worth up to six points. The aggregated knowledge score ranged from 0 to 11 and was dichotomised as seven points or higher, or less than seven points. However, the supplemental material provided with the published trial suggested that the knowledge test was not an objective measure of knowledge. The list of vaccines included only ten, and the answer options were 'Yes, I know' or 'I do not know'. These were also the answer options for the vaccine policy questions, e.g. 'Do you know the location for immunisation registry?' (Yes I know / I do not know). We contacted the authors to clarify the intent of these scales and to confirm that there had been no errors of translation, but we were unable to reach them. We determined that this measurement scale could not objectively assess knowledge, so we did not include these data in the review.

Table D: Knowledge or understanding outcome features

		Jackson 2011	Quinlivan 2003	Saitoh 2013	Saitoh 2017	Wood 1998	Hu 2017	Bjornson 1997
Outcome	Knowledge	x	x	x	x	x	x (data unusable)	x (data unusable)
Outcome timing	Immediately post‐intervention							x
	1 week	x^a
	1 month				x ^a
	3 months	x		x (100 days post int)			x
	6 months		x		x
	15 months					x
Assessment tool	Parent interview					x
Assessment tool	Questionnaire, test, or survey	x	x	x	x		x	x

^aAdditional time point reported by study but not used in meta‐analysis

Attitudes or beliefs

Three studies measured parent attitudes or beliefs towards vaccines in general (Saitoh 2013; Saitoh 2017), or towards MMR in particular (Jackson 2011). The attitude assessment tools for Saitoh 2013 and Saitoh 2017 were very similar, and were both based on aspects of the Health Belief Model and the Integrated Behavioral Model. The tool measured eight separate attitude outcomes, using 21 statements (Saitoh 2013), or 20 statements (Saitoh 2017). The eight outcome categories were: perceived severity, susceptibility, benefits, barriers, self‐efficacy, behavioural control, and injunctive and descriptive social norms. Participants rated each statement on a five‐point Likert scale from one (strongly disagree) to five (strongly agree). The tool was translated from English to Japanese. The translation was confirmed and the tool piloted prior to use, but was not described as validated.

Jackson 2011 measured three attitude outcomes: attitude towards MMR, necessity of MMR beliefs, and concern about MMR beliefs. The tool assessed attitude towards MMR with one question, scored on a seven‐point scale ('For me to give my child the combined MMR vaccine at the recommended ages would be ...' 1 extremely bad to 7 extremely good). This item had demonstrated reliability and validity. The authors adapted an existing questionnaire (Beliefs about Flu Vaccination Questionnaire) to assess beliefs, though this tool was not validated for MMR (Bekker 2003). Parents recorded their necessity beliefs using four items (e.g. 'Without the combined MMR vaccine, my child could get very ill from measles, mumps or rubella'). Concern beliefs were assessed with another four items (e.g. 'Giving my child the combined MMR vaccine worries me'). These items were scored on five‐point scales. Scores were summed for each type of belief, for a total score of 4 to 20 for necessity and for concern (higher scores indicated stronger beliefs about the necessity for, or concern about, MMR).

Although three studies measured attitudes or beliefs related to vaccination, these were measured in a variety of ways both within each study and across studies (Table 2). We reviewed the descriptions of the outcome measures to identify the most similar attitude outcome across studies: perceived severity (Saitoh 2013; Saitoh 2017) and necessity beliefs (Jackson 2011).

Table E: Attitudes or beliefs outcome features

		Jackson 2011	Saitoh 2013	Saitoh 2017
Outcome	Attitudes	x	x	x
Outcome timing	1 week	x^a
	1 month			x^a
	3 months	x	x (100 days post int)
	6 months			x
Assessment tool	Questionnaire, test, or survey	x	x	x

^aAdditional time point reported by study but not used in meta‐analysis

Intention to vaccinate

Two studies measured parents' intention to vaccinate at three months post‐intervention (Jackson 2011; Saitoh 2013). Both studies used pre‐ and post‐intervention questionnaires or surveys to record the data. In Jackson 2011, parents reported their intended choice about vaccinating their child with the first or second dose of MMR. They rated three items on a seven‐point scale, and the scores were averaged to produce a score from 0 to 7 (from 'do not intend to vaccinate', to 'do intend to vaccinate'). These items had demonstrated reliability and validity, according to the authors. In Saitoh 2013, parents answered one question about their intention to vaccinate using a four‐point scale (1 = no, 2 = undecided, 3 = yes, for a specific vaccine, and 4 = yes).

Table F: Intention to vaccinate outcome features

		Jackson 2011	Saitoh 2013
Outcome	Intention to vaccinate	x	x
Outcome timing	1 week	x^a
Outcome timing	3 months	x	x (100 days post int)
Assessment tool	Questionnaire, test, or survey	x	x

^aAdditional time point reported by study but not used in meta‐analysis

Adverse effects

Anxiety associated with the intervention was the only adverse event measured in any studies. Anxiety was measured by Jackson 2011 at one week post‐intervention and three months post‐intervention (final follow‐up). The scale included six items (e.g. 'I feel calm' or 'I am tense'), each scored on a four‐point scale, from 'not at all', to 'very much'. Positive items were reversed, and the scores for all six items were combined and then multiplied by 20/6 for a total score of 20 (very low anxiety) to 80 (very high anxiety). The authors report that a normal score was 34 to 36.

Secondary outcomes

No studies measured parent experience of the intervention.

The cost of implementing the intervention was measured by Wood 1998. See Effects of interventions for further details about the cost‐effectiveness calculations used.

Excluded studies

We excluded 29 studies at the full‐text stage (see 'Characteristics of excluded studies' table). The reasons for exclusion were:

the study was not an RCT or cluster‐RCT (N = 10);
the intervention was a maternal outreach or support intervention, where the immunisation content was not specified or consistently delivered (N = 5);
the study did not report on a face‐to‐face information or educational intervention (N = 5);
the face‐to‐face intervention was part of a multi‐component intervention or a mass media campaign, in which the effect of any face‐to‐face information or education could not be isolated (N = 4);
the intervention was not targeted to parents (N = 4); or
the intervention was not relevant to immunisation (N = 1).

One study included in the original review was excluded from this update, as it was a maternal outreach or support intervention with unclear immunisation content (one of the five studies excluded for reason 2, above; Bartu 2006).

Risk of bias in included studies

The risk of bias for all included studies is summarised in Figure 2 and Figure 3.

Figure 2

Risk of bias summary: review authors' judgements about each risk of bias item for each included study

Figure 3

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies

Allocation

We determined a priori that sequence generation and allocation concealment were the most significant and influential risk of bias attributes. We judged nine out of ten studies at low risk of bias for sequence generation, while Jackson 2011 was unclear. For allocation concealment, three were at high risk (Usman 2009; Usman 2011; Wood 1998) and we judged one as at unclear risk (Hu 2017). Through contact with study authors, we confirmed that appropriate allocation concealment had been performed in one study (Saitoh 2013).

Blinding

Due to the overt nature of face‐to‐face interventions, blinding participants and personnel was not possible in any of the included studies, and we assessed all as high risk of bias for these factors. Outcome assessors were adequately blinded in two studies (Bolam 1998; Quinlivan 2003). Assessors were not blinded in Usman 2009 or Usman 2011, but we judged the risk of bias to be low, because the only outcome assessed was receipt of vaccine, which is not a subjective outcome. The remainder were at high or unclear risk of detection bias due to lack of blinding.

Incomplete outcome data

We only assessed two studies at low risk of bias for incomplete outcome data, either because the outcome was assessed immediately following intervention delivery (Bjornson 1997), or because attrition was relatively low and the authors used an intention‐to‐treat analysis (Saitoh 2017). Two studies had a high risk of attrition bias because those lost to follow‐up were excluded from the analysis, and the dropout rates were relatively high (25% (Bolam 1998), and 32% (Hu 2017)).

We judged six studies as being at unclear risk of attrition bias: large numbers of participants failed to receive the intervention (Jackson 2011; Wood 1998); all non‐respondents were classified as not receiving the vaccine (Usman 2009; Usman 2011); substantially more participants withdrew from one group than the other without explanation (Quinlivan 2003; Saitoh 2013).

Selective reporting

We rated only one study at low risk of bias for this domain, because the outcomes planned in the trial registration record were all reported in the published study (Saitoh 2013). We rated one study at unclear risk of bias because the trial registration record included the proposed outcome of cost, which was not reported in the published study (Saitoh 2017). All other studies were at unclear risk of bias for this domain because protocols or trial registration records were not available.

Other potential sources of bias

We judged four studies to be at low risk of other bias due to low risk of contamination and comparability of groups at baseline (Bolam 1998; Hu 2017; Saitoh 2013; Saitoh 2017).

The authors of one cluster‐RCT did not account for the effects of clustering in the analysis, so we rated this study as high risk of bias for this attribute (Bjornson 1997). Two studies reported some significant baseline demographic differences between the intervention and control groups, for which we rated them as unclear risk of bias (Quinlivan 2003; Usman 2011). In Jackson 2011, baseline characteristics were similar and contamination across study arms was unlikely, but the authors note that the leaflet they provided to all arms may have been more thorough than the actual usual care in the region, meaning they may have been comparing two different decision‐support interventions (educational session plus leaflet versus leaflet) rather than an intervention with a control. We judged the potential risk of bias arising from this as unclear. There was insufficient information to judge if other potential sources of bias were present in Usman 2009 and Wood 1998 so we also rated them at unclear risk of bias.

Selective recruitment or participants (cluster‐RCTs)

We determined two cluster‐RCTs to be at low risk of bias for selective recruitment of participants (Bjornson 1997; Jackson 2011), while one was at high risk because participants were recruited after clusters were randomised (Saitoh 2017).

Effects of interventions

We present an outline of the main findings for each outcome in summary of findings Table for the main comparison.

We did not find sufficient RCTs to investigate reporting bias or small‐study effects using funnel plots. However, our qualitative assessment indicated that there was low risk of reporting bias for all outcomes, as the smaller studies showed no strong evidence of an effect.

Comparison 1: Face‐to‐face interventions directed to parents versus control

Primary outcomes

Vaccination status

Seven studies compared face‐to‐face information or educational interventions with control, with the vaccination status measured either three, six, or twelve months after the intervention (Bolam 1998; Hu 2017; Jackson 2011; Saitoh 2013; Saitoh 2017; Usman 2009; Usman 2011). Face‐to‐face communication to inform or educate parents may improve vaccination status (risk ratio (RR) 1.20, 95% confidence interval (CI) 1.04 to 1.37; 3004 participants; low‐certainty evidence).

Communication interventions are frequently complex, and it can be difficult to determine whether interventions are sufficiently similar to compare. While the level of statistical heterogeneity was high (Chi² = 40.68, df = 6 (P < 0.00001); I² = 85%), we decided to pool the data for these studies, acknowledging that the length of the educational sessions and the number of vaccines delivered were potentially important differences between them. We conducted post hoc formal subgroup analyses to investigate these two potential sources of heterogeneity (Analysis 1.1 and Analysis 1.2).

For analysis stratified by session length, we categorised sessions that lasted 10 minutes or less as 'short' (Analysis 1.1; Saitoh 2013; Saitoh 2017; Usman 2009; Usman 2011), and those lasting longer than 10 minutes as 'long' (Bolam 1998; Hu 2017; Jackson 2011). In Australia, the average clinical consultation time is 15 minutes, and healthcare providers need time to both deliver the intervention and the vaccine (Britt 2016). Therefore, we determined that 10 minutes was the maximum time an intervention could take without encountering potential scheduling and reimbursement implications. The test for subgroup differences was not significant (P = 0.15). Heterogeneity of the stratified studies remained relatively high, so we were unable to conclusively state whether session length was a major contributing factor to the heterogeneity in study effects.

The second analysis grouped studies that measured receipt of a single vaccine (Analysis 1.2; Jackson 2011; Usman 2009; Usman 2011) and those that measured receipt of multiple vaccines (Bolam 1998; Hu 2017; Saitoh 2013; Saitoh 2017). This decision was based on advice from vaccination experts, and reflects the fact that receiving one vaccine is less demanding, as an outcome, than receiving several. The test for subgroup differences was significant (P = 0.04). There was no appreciable difference in heterogeneity.

Knowledge or understanding of vaccination

Moderate‐certainty evidence from four studies (657 participants) suggested that face‐to‐face information or education probably improved parent knowledge slightly, compared with control (standardised mean difference (SMD) 0.19, 95% CI 0.00 to 0.38; Analysis 1.3; Jackson 2011; Saitoh 2013; Saitoh 2017; Wood 1998). Knowledge was measured by two separate scales in Saitoh 2013 (knowledge of vaccine‐preventable diseases and basic immunisation knowledge), which we combined into a single score by summing the mean scores from each scale and calculating revised standard deviation (SD) by squaring the SD for each scale, summing them and taking the square root. Statistical heterogeneity between these studies was low (Chi² = 3.86, df = 3 (P = 0.28); I² = 22%).

We were unable to include the outcome data from Quinlivan 2003 in our meta‐analysis because they were given as medians rather than means. Contact with the authors confirmed that this was because the data were skewed. In the published article, the authors reported that there were no significant differences between knowledge scores of the intervention and control groups at the antenatal or postnatal assessments (mean difference (MD) 0.85, 95% CI: ‐0.06 to 1.76).

Attitudes or beliefs

Three studies (292 participants) reported data on several measures of attitudes or beliefs. For the purpose of this review, we selected the most comparable measures of attitudes across the studies ‐ two on perceived severity of vaccine‐preventable diseases (Saitoh 2013, Saitoh 2017), and one on beliefs about the necessity of the MMR vaccine (Jackson 2011). Face‐to‐face information or educational interventions may lead to little or no change in parent attitudes or beliefs about disease severity or vaccine necessity (SMD 0.03, 95% CI ‐0.20 to 0.27; Analysis 1.4; low‐certainty evidence). Heterogeneity was not detected (Chi² = 0.08, df = 2 (P = 0.96); I² = 0%), but all were small studies with a small combined total number of participants.

Intention to vaccinate

Low‐certainty evidence from two studies (179 participants) suggests that face‐to‐face information or education delivered to parents may slightly increase parents' intention to vaccinate (SMD 0.55, 95% CI 0.24 to 0.85; Analysis 1.5; Jackson 2011; Saitoh 2013). Heterogeneity was not detected (Chi² = 0.81, df = 1 (P = 0.37); I² = 0%).

Adverse effects (anxiety)

Only one study (90 participants) measured this outcome (Jackson 2011). Face‐to‐face information or educational interventions may lead to little or no change in parents' anxiety (MD ‐1.93, 95% CI ‐7.27 to 3.41; low‐certainty evidence; Analysis 1.6).

Secondary outcome

Cost

Only one included study reported data on costs of implementing an intervention (Wood 1998). The certainty of this evidence was judged to be very low, and consequently we are very uncertain of the effects of the intervention on cost outcomes. This study reported a full cost‐effectiveness analysis of a case management intervention to promote childhood immunisation, compared with usual practice. Total costs were estimated, based on costs of personnel, supplies, travel, office space, and orientation costs of the intervention only. Costs calculated per individual took account of indirect case management costs (based on the frequency and time taken for case managers to complete client‐related contact tasks) in the same proportion as direct costs incurred by the intervention.

The total cost of the intervention was calculated to be USD 293,662 (price year not stated). For the intervention group (N = 185), the mean cost per participant was calculated as USD 1587, and the additional cost of bringing one non‐immunised child fully up‐to‐date with immunisations by one year was estimated to be USD 12,022. This study did not consider potential costs arising from non‐immunisation (e.g. costs of contracting a disease).

Analysis was also conducted on a high‐risk subgroup of those receiving the intervention, defined by the authors as those children who had received only three or fewer well‐child visits (of a total of five visits). In this subgroup (n = 86), the mean cost per participant was estimated as USD 1273, with the additional cost of bringing one non‐immunised child fully up to date with immunisations by one year calculated as USD 4546.

The study authors reported resource use, but this was expressed in terms of monetary costs rather than physical resource quantities. They noted that the cost‐effectiveness data may underestimate the true cost‐effectiveness of the intervention, due to start‐up and evaluation costs associated with the delivery of the intervention over a short time period. However, there are several factors that may limit the conclusions that can be drawn from this study: the cost‐effectiveness analysis did not consider effects of the intervention on subsequent care, nor of changes to the costs of usual practice over time. Cost data were presumably reported in US dollars, with no information provided on the price year. Given that the study was published in 1998, the costs and cost‐effectiveness of such a case management intervention may be substantially different in today's terms. All outcomes and costs associated with the study were measured at 12 months, with no longer‐term data reported, which limits the conclusions that can be drawn. Finally, the intervention delivered by this study was much more complex, and therefore potentially resource‐intensive, than many of the other interventions included in this review.

Parent experience of the intervention

No included studies reported this outcome.

Comparison 2: Face‐to‐face intervention A versus face‐to‐face intervention B

No included studies reported this comparison.

Discussion

available in

Summary of main results

This systematic review found some evidence, of low to moderate certainty, suggesting that face‐to‐face interventions to inform or educate parents about childhood vaccination may improve children's vaccination status, probably slightly improves parents' knowledge or understanding of vaccination, and may slightly increase intention to vaccinate. These interventions may lead to little or no difference in parent attitudes or anxiety related to the intervention. There was uncertain evidence related to the cost of implementing interventions, and no evidence available from the included studies regarding parent experience of the intervention.

Overall completeness and applicability of evidence

This review used a comprehensive search strategy, which followed Cochrane search methods, and was not restricted by language or publication status. We included only randomised controlled trials (RCT), though there tend to be fewer trials in this area than there are for clinical or biomedical topics, due to the challenges of conducting RCTs of complex public health interventions. This meant that relatively few studies met the inclusion criteria, but by definition, those that did should have been of relatively high quality compared to other study designs (Higgins 2011). Implementing communication interventions can be costly and time‐consuming, so it is important for policy makers and programme managers to be able to base their decisions on rigorous, high‐quality evidence, wherever possible. We excluded quasi‐RCTs, but recognise that excluding studies for inadequate randomisation can be seen as a form of research waste. Future updates of this review may consider quasi‐RCTs and additional study designs as eligible for inclusion.

The global applicability of the evidence included in this review is uncertain. Two studies were conducted in Japan, where some vaccines are required and funded by the government, but some are not (Saitoh 2013; Saitoh 2017). The coverage rate for those vaccines that are not required and funded is substantially lower, in part because this policy implies that these vaccines are optional, or less important. In this context, an intervention that increases awareness by providing information and education may show greater effect than in a country where awareness is high, but vaccine hesitancy is a problem (e.g. the UK, where Jackson 2011 took place). Three of the ten included studies were conducted in low and middle income countries (LMIC), which may also have implications for the applicability of the findings.

We did not include multi‐component interventions where the effects of the face‐to‐face intervention could not be isolated, which resulted in the exclusion of many studies that included a face‐to‐face element. It is possible that face‐to‐face interventions may be more or less effective, depending on the additional interventions with which they are combined. However, assessing the effects of a single intervention is important to allow decision makers to determine whether it is worthwhile, cost‐effective, or both, to implement a particular intervention alone; or whether it may be beneficial to combine interventions, and if so, which components to combine (Ryan 2014).

All of the new studies included in this update measured a much broader range of outcomes than those in the original review, which is encouraging, as it suggests increased recognition of the importance of considering intermediate outcomes, like knowledge, attitudes, and intention to vaccinate. In the original review, only two studies measured knowledge, and none measured any other intermediate outcomes. Unpacking the effects of complex interventions requires the measurement of a range of outcomes that are able to detect potential impacts along the causal chain (Breuer 2015; Craig 2008; Moore 2015). Measuring outcomes, such as knowledge, attitudes, and intention to vaccinate ‐ in addition to endpoint outcomes related to vaccination status ‐ is necessary to determine which parts of complex communication interventions are effective, and to identify whether interventions are impacting vaccine hesitancy. While the breadth of outcomes was greater in the newer studies, there were still some issues. The tools and scales used to measure attitudes were highly variable, making it difficult to compare the data for this outcome across studies. Also, no studies assessed parents' experiences of the intervention, such as satisfaction with, or acceptability of the intervention. Qualitative research has highlighted that parents' perceptions about their communication encounters can impact their decisions around vaccination (Ames 2017; Brown 2010). Similarly, cost is a critical factor affecting implementation decisions made by policymakers and programme managers, but cost was only reported in one study (Wood 1998). As this study was from 1998, and was limited to a specific complex intervention (case management) in a low‐income population in a high‐income country, it is unlikely to be generalisable to other settings.

Quality of the evidence

The certainty of the evidence was moderate to low for each outcome. All studies had limitations in design. We downgraded evidence where the contributing studies were at high or unclear risk of bias for sequence generation (Jackson 2011), or allocation concealment (Hu 2017; Usman 2009; Usman 2011; Wood 1998).

We also downgraded the certainty of the evidence for vaccination status, due to inconsistency, which was clear from the high level of statistical heterogeneity. The certainty of the evidence for intention to vaccinate and adverse effects was downgraded for imprecision, due to the wide confidence intervals for the included studies. We downgraded the certainty of the evidence for attitudes due to indirectness, because the specific measures for this outcome (perceived diseases severity and vaccine benefits) were only part of what could be measured, and what was relevant to parents and other decision makers. We did not downgrade any outcomes for issues of publication bias.

Potential biases in the review process

We pooled the results of studies reporting vaccination status, despite high levels of statistical heterogeneity. This may have introduced some bias because the actual effect size might be substantially larger or smaller than that of the pooled (mean) result. We conducted post hoc formal subgroup analyses to investigate two potential implementation‐relevant sources of this heterogeneity (length of intervention session and number of vaccines received). However, heterogeneity remained largely unexplained in both cases. As more studies are added in future updates of this review, we will reconsider this analysis and investigate the effects of relevant explanatory factors identified from the existing research.

Another possible source of bias is our selection of comparable measures of attitudes or beliefs. The data used for the purposes of analysis were selected from more than one measurement of attitudes and beliefs available per included trial. We cannot rule out the possibility that other researchers might choose alternative measurements from the same studies, and so reach different conclusions about the effects of the intervention on this outcome. This is relatively likely, as there is an absence of agreement over the most appropriate ways of measuring attitudes and beliefs in the field of vaccination.

All other methods and literature searching for this update did not deviate from the processes outlined in the original review, and should be relatively free from bias.

Agreements and disagreements with other studies or reviews

The findings of this review are largely consistent with those of other recent reviews, which suggest that educational interventions, face‐to‐face communication, or a combination, may have some positive impact on vaccination status and parent knowledge or attitudes. The Cochrane Review of interventions for improving childhood vaccination coverage in LMICs assessed 14 studies, six of which evaluated the effects of health education. The review authors found moderate‐certainty evidence that health education at village meetings or at home probably improved coverage (Oyo‐Ita 2016). Another Cochrane Review looked at community‐aimed interventions to inform or educate about early childhood vaccination. The authors included two studies, which found that community‐aimed interventions probably increased vaccination uptake in some circumstances, and may have improved knowledge and attitudes towards vaccination (Saeterdal 2014).

Two non‐Cochrane reviews considered a broader range of intervention types, but also highlighted the relevance of face‐to‐face educational interventions. According to Jarrett 2015, who reviewed strategies for addressing vaccine hesitancy, dialogue‐based interventions were the most commonly used. Harvey 2015 assessed parental reminder, recall, and educational interventions, finding that discussion‐based education was more effective than written education.

One recent review aimed to identify the effectiveness of specific provider‐parent communication techniques, but found insufficient high‐quality evidence on which to draw a conclusion (Connors 2017). Our review was similarly limited by the shortage of trial evidence on this topic.

In future updates of this review, we may compare our findings with those of RCTs involving face‐to‐face interventions for recommended adult vaccinations, to provide relevant context and insight.

Figure 1

Study flow diagram: 2018 review update

Figure 2

Risk of bias summary: review authors' judgements about each risk of bias item for each included study

Figure 3

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies

Analysis 1.1

Comparison 1 Face‐to‐face education versus control or non‐face‐to‐face education (all), Outcome 1 Vaccination status (stratified by length).

Analysis 1.2

Comparison 1 Face‐to‐face education versus control or non‐face‐to‐face education (all), Outcome 2 Vaccination status (stratified by number of vaccines delivered).

Analysis 1.3

Comparison 1 Face‐to‐face education versus control or non‐face‐to‐face education (all), Outcome 3 Knowledge.

Analysis 1.4

Comparison 1 Face‐to‐face education versus control or non‐face‐to‐face education (all), Outcome 4 Attitudes ‐ necessity.

Analysis 1.5

Comparison 1 Face‐to‐face education versus control or non‐face‐to‐face education (all), Outcome 5 Intention to vaccinate.

Analysis 1.6

Comparison 1 Face‐to‐face education versus control or non‐face‐to‐face education (all), Outcome 6 Adverse effects (anxiety).

Face‐to‐face interventions directed to parents for informing or educating parents about early childhood vaccination, as compared with control
Patient or population: parents of preschool‐aged children or expectant parents Settings: clinics, antenatal classes, or the mother's home Intervention: face‐to‐face information or educational interventions Comparison: control (no education, other education, or control not described)
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	No of Participants (studies)	Quality of the evidence (GRADE)	Comments
	Assumed (baseline) risk	Corresponding (intervention) risk
	Control (no face‐to‐face information or education)	Face‐to‐face information or education
Vaccination status Final time point (3, 6, or 12 months post‐intervention)	55 per 100¹	66 per 100 (57 to 75)	RR 1.20 (1.04 to 1.37)	3004 (7 studies)	⊕⊕⊝⊝ low²	The results for this outcome were variable, so the true result may be substantially higher or lower than this estimate.
Knowledge or understanding (Different measures used by the studies: knowledge of vaccine‐preventable diseases, vaccines, contraindications to vaccination, or a combination, on varying scales) Final time point (3 or 6 months post‐intervention)³		The mean knowledge score in the intervention group was 0.19 standard deviations higher (0.00 to 0.38 standard deviations higher)		657 (4 studies)	⊕⊕⊕⊝ moderate⁴	One further study (Quinlivan 2003) of 124 participants did not report individual group mean scores as data were skewed. The authors reported that there were no significant differences in the knowledge scores of the intervention and control groups (MD 0.85, 95% CI: ‐0.06 to 1.76). A standard deviation of 0.2 represents a small difference between groups (based on Cohen’s effect sizes; Cohen 1988).
Attitudes or beliefs (Different measures used by the studies: perceived severity of vaccine‐preventable diseases, perceived necessity of vaccines, or a combination, on varying scales) Final time point (3 or 6 months post‐intervention)		The mean score for perceived severity of diseases, necessity of vaccines, or both, in the intervention group was 0.03 standard deviations higher (0.20 standard deviations lower to 0.27 standard deviations higher)		292 (3 studies)	⊕⊕⊝⊝ low⁵	A standard deviation of 0.2 represents a small difference between groups (based on Cohen’s effect sizes; Cohen 1988).
Intention to vaccinate 1 to 7 (1 = definitely do not intend to vaccinate; 7 = definitely do intend to vaccinate) 3 months post‐intervention	The mean intention to vaccinate in the control group was 5.58 (SD 2.13)⁶	The mean intention to vaccinate in the intervention groups was on average 1.17 points higher (0.51 to 1.81 points higher)		179 (2 studies)	⊕⊕⊝⊝ low⁷
Adverse effects (anxiety associated with intervention) 20 (low anxiety) to 80 (high anxiety) 3 months post‐intervention	The mean anxiety in the control group was 33.9 (SD 13.53)	The mean anxiety in the intervention group was 1.93 points lower (7.27 lower to 3.41 higher)		90 (1 study)	⊕⊕⊝⊝ low⁸	The study authors note that a normal score is in the range of 34 to 36 on this scale.
Parent experience of the intervention	No studies measured this outcome.
Cost (personnel, supplies, travel, office space, and orientation costs of intervention)	Effect of intervention was very uncertain. A single study reported that the estimated mean cost of usual care per fully immunised child was USD 1587, or USD 1273 for children defined as high‐risk.⁹ The estimated additional cost per fully immunised child per intervention was approximately 8 times higher than usual care for all children, and 4 times higher for high‐risk children.			365¹⁰ (1 study)	⊕⊕⊝⊝ low¹¹	Did not include potential costs arising from non‐immunisation. Costs calculated per individual took account of indirect case management costs (frequency and time taken for case managers to complete client‐related contact tasks).
The basis for the assumed risk* (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: Confidence interval; RR: Risk Ratio
GRADE Working Group grades of evidence High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.
¹Assumed risk based on median control group risk across studies ²We downgraded the evidence for this outcome for risk of bias (‐1). One trial was at unclear risk for sequence generation, two trials were at high, and one at unclear risk of bias for allocation concealment. We also downgraded for inconsistency (‐1) because, while the nature of the interventions and participants were relatively similar across studies, there was considerable statistical heterogeneity that was not easily explained (I² = 85%, Chi² P < 0.00001). ³One study measured this outcome at 15 months postpartum. The intervention included multiple sessions at varying times, so the time between the final session and outcome assessment was not known. ⁴We downgraded the evidence for this outcome for risk of bias (‐1). One trial was at unclear risk for sequence generation and one trial was at high risk of bias for allocation concealment. ⁵We downgraded the evidence for this outcome for indirectness (‐1). There are many aspects of attitudes that are important to decision making. The specific attitudes measured in this outcome were only part of what could be measured, and therefore this outcome was a somewhat incomplete or indirect indication of attitudes. We also downgraded for risk of bias (‐1), because lack of blinding may have impacted this subjective outcome. ⁶Assumed risk based on control group score from Jackson 2011, as this was the only validated scale. ⁷We downgraded the evidence for this outcome for imprecision (‐1). The sample size for this outcome was relatively small and the effect estimate showed potentially appreciable benefit (i.e. the CI of the pooled effect estimate crossed 0.5). We also downgraded for risk of bias (‐1), because lack of blinding may have impacted this subjective outcome. ⁸We downgraded the evidence for this outcome for risk of bias (‐1), because only one study contributed data and it was at unclear risk of bias for sequence generation. We also downgraded for imprecision (‐1) because the total sample size was small (N = 135 participants). ⁹High‐risk subgroup defined as those children who received only 3/5 or fewer well‐child visits. ¹⁰Includes high‐risk subgroup (86 participants). ¹¹We downgraded the evidence for this outcome for risk of bias (‐1), because it was at high risk of bias for allocation concealment. We also downgraded for indirectness (‐1), because the evaluated intervention was multi‐component and included telephone reminders in addition to education.

Table 1. Summary of related non‐Cochrane reviews

Review	Title	Review focus and main findings	Relationship / key differences
Bright 2017	A systematic review of strategies to increase access to health services among children in low and middle income countries	Focus: effectiveness of interventions aimed at increasing access to health services for children aged 5 years and below in LMICs. Supply side interventions included: delivery of services at or closer to home and service level improvements (e.g. integration of services). Demand side interventions included: educational programmes, text messages, and financial or other incentives. Included studies: 57 RCTs and non‐randomised controlled trials Findings: health promotion or educational programmes were the most commonly evaluated intervention in the review. Education delivered by healthcare workers generally showed a positive impact on healthcare utilisation or immunisation uptake, but mixed effectiveness when delivered by community health workers	Broad range of interventions addressing health topics beyond vaccination and limited to LMICs.
Connors 2017	Provider‐parent communication when discussing vaccines: a systematic review	Focus: determining effective communication practices for providers and parents, and establishing whether collaborative and participatory communication between parents and providers impact parents’ views on vaccination. Included studies: 9 studies, mostly descriptive and qualitative, with 1 RCT Findings: evidence was limited and low quality. Themes of trust and support suggested the value of participatory communication, though opposing evidence in favour of presumptive communication also existed.	Included qualitative studies, found similar dearth of high‐quality evidence on effectiveness. The RCT tested an intervention directed to physicians, so was not included in our review.
Harvey 2015	Parental reminder, recall and educational interventions to improve early childhood immunisation uptake: a systematic review and meta‐analysis	Focus: effectiveness of remind, recall, and educational interventions for childhood vaccination uptake Included studies: 28 controlled studies, 13 focusing on reminders and 17 on parental education Findings: postal and telephone reminders combined showed the greatest positive effect as a reminder intervention. Educational interventions may be more effective in LMICs. Discussion‐based education was more effective than written education.	Several studies appearing in our review also appeared in the meta‐analysis of educational interventions in this review, but there were additional non‐face‐to‐face educational interventions included in this review.
Jarrett 2015	Strategies for addressing vaccine hesitancy – a systematic review	Focus: identify, describe, and assess potential effectiveness of strategies to reduce vaccine hesitancy Included studies: 166 peer‐reviewed and 15 grey literature records reporting an evaluation of an intervention. There was no restriction on study design Findings: three intervention categories: dialogue‐based, incentive‐based, or reminder‐ and recall‐based interventions. Most interventions were dialogue‐based. The most effective strategies were multi‐component and tailored to specific populations and specific problems.	Broad range of included interventions, beyond face‐to‐face communication. Many non‐RCT study designs were included in this review.
Odone 2015	Effectiveness of interventions that apply new media to improve vaccine uptake and vaccine coverage	Focus: effectiveness of interventions that apply new media to promote vaccination uptake and increase vaccination coverage Included studies: 19 studies, most from the USA. 13 experimental and 6 observational. Findings: interventions included Facebook, SMS, YouTube videos, apps, software, email, and targeted websites. Some evidence that SMS, some websites or web portals, and computerised reminders may increase vaccination rates.	Included vaccines for children, adolescents, and adults. No face‐to‐face communication interventions, only new media
Sadaf 2013	A systematic review of interventions for reducing parental vaccine refusal and vaccine hesitancy	Focus: effectiveness of interventions to decrease parental vaccine refusal and hesitancy toward recommended childhood and adolescent vaccines Included studies: 30 studies, primarily before‐and‐after intervention design with some RCTs, non‐RCTs and evaluation studies. Most from USA Findings: three intervention categories: passage of state laws, implementation of school‐ and state‐level laws, and parent‐centred information or education. Education, particularly short pamphlets, most commonly studied, with heterogeneous formats and mixed effects. Overall, no convincing evidence on effective strategies. Authors noted few studies measured hesitancy‐relevant outcomes, and most evidence was of low quality.	Broad range of included interventions, beyond face‐to‐face communication
Williams 2014	What are the factors that contribute to parental vaccine‐hesitancy and what can we do about it?	Focus: barriers to vaccination and strategies to address vaccine hesitancy Included studies: seven studies, largely RCTs or CRCTs. Not all specifically recruited hesitant parents Findings: current data did not support one method for communicating with hesitant parents over another. Wide range of interventions used	Included interventions beyond face‐to‐face, also included adolescent (HPV) as well as childhood vaccines.

Table 1. Summary of related non‐Cochrane reviews

Table 2. Additional measures of parent attitudes

Study	Outcome	Scale used
Jackson 2011	Attitude towards MMR One item: ‘For me to give my child the combined MMR vaccine at the recommended ages would be…’	1 (extremely negative attitude) to 7 (extremely positive attitude)
	Necessity beliefs Four items assessing the necessity of MMR: e.g. ‘Without the combined MMR vaccine, my child could get very ill from measles, mumps, or rubella’	Each item scored from 1 to 5 and summed, for total outcome scale of 4 (not at all necessary) to 20 (very necessary) We used this item in the meta‐analysis.
	Concern beliefs Four items assessing concerns about MMR: e.g. ‘Giving my child the combined MMR vaccine worries me’	Each item scored from 1 to 5, and summed for total outcome scale of 4 (not at all concerned) to 20 (very concerned)
Saitoh 2013	Perceived severity (HBM) Two items: e.g. ‘The vaccine‐preventable diseases are serious diseases.’	Each item scored from 1 to 5 (strongly disagree to strongly agree), and summed for a total outcome scale of 2 to 10. We used this item in the meta‐analysis.
	Perceived susceptibility(HBM) One item: ‘My baby is not very likely to get vaccine preventable diseases.’	Each item scored from 1 to 5 (strongly disagree to strongly agree), and summed for a total outcome scale of 1 to 5.
	Perceived benefits (HBM) Four items: e.g. ‘The vaccines for babies will prevent my baby from getting sick with a vaccine‐preventable disease.’	Each item scored from 1 to 5 (strongly disagree to strongly agree), and summed for a total outcome scale of 4 to 20
	Perceived barriers (HBM) Five items: e.g. ‘The vaccines will make my baby sick,’ ‘The vaccines do not prevent the vaccine‐preventable diseases.’	Each item scored from 1 to 5 (strongly disagree to strongly agree), and summed for a total outcome scale of 5 to 25
	Self‐efficacy (HBM) Two items: e.g. ‘I feel comfortable getting the vaccines for my baby.’	Each item scored from 1 to 5 (strongly disagree to strongly agree), and summed for a total outcome scale of 2 to 10
	Perceived behavioral control (IBM) One item: ‘I have control over whether or not my baby gets the vaccines.’	Each item scored from 1 to 5 (strongly disagree to strongly agree), and summed for a total outcome scale of 1 to 5
	Social norm (injunctive) Four items: e.g. ‘Most people important to me think I should get my baby the vaccines.’	Each item scored from 1 to 5 (strongly disagree to strongly agree), and summed for a total outcome scale of 4 to 20
	Social norm (descriptive) Two items: e.g. ‘I know other people my age who got their baby the vaccines.’	Each item scored from 1 to 5 (strongly disagree to strongly agree), and summed for a total outcome scale of 2 to 10
Saitoh 2017	Perceived severity (HBM) Two items: e.g. ‘Vaccine preventable diseases are serious diseases.’	Each item scored from 1 to 5 (strongly disagree to strongly agree), and summed for a total outcome scale of 2 to 10 We used this item in the meta‐analysis.
	Perceived susceptibility(HBM) One item: ‘My baby is not very likely to get vaccine preventable diseases.’	Each item scored from 1 to 5 (strongly disagree to strongly agree), and summed for a total outcome scale of 1 to 5
	Perceived benefits (HBM) Five items: e.g. ‘The vaccines for babies will prevent my baby from getting sick with vaccine‐preventable diseases.’ ‘If my baby receives his or her vaccines, it will help protect my friends and family from getting vaccine‐preventable diseases.’	Each item scored from 1 to 5 (strongly disagree to strongly agree), and summed for a total outcome scale of 5 to 25
	Perceived barriers (HBM) Six items: e.g. ‘I feel uncomfortable because vaccines are painful or uncomfortable for my baby,’ ‘Vaccines are too expensive.’	Each item scored from 1 to 5 (strongly disagree to strongly agree), and summed for a total outcome scale of 6 to 30
	Perceived behavioral control (IBM) One item: ‘I have control over whether or not my baby gets the vaccines.’	Each item scored from 1 to 5 (strongly disagree to strongly agree), and summed for a total outcome scale of 1 to 5
	Social norm (injunctive) Three items: e.g. ‘Most people important to me think I should have my baby vaccinated.’	Each item scored from 1 to 5 (strongly disagree to strongly agree), and summed for a total outcome scale of 3 to 15
	Social norm (descriptive) Two items: e.g. ‘I know other people my age who had their babies vaccinated.’	Each item scored from 1 to 5 (strongly disagree to strongly agree), and summed for a total outcome scale of 2 to10

Table 2. Additional measures of parent attitudes

Table 3. Vaccination status outcome data

Session length	Outcome	Study	Timing of outcome assessment (days/months)	Intervention group		Control group		Notes
Session length	Outcome	Study	Timing of outcome assessment (days/months)	Observed (n)	Total (N)	Observed (n)	Total (N)	Notes
Shorter sessions (1 to 10 min)	Up‐to‐date immunisation status at 3 months of age for Hib, HBV, and PCV7 vaccines	Saitoh 2013	3 months post‐intervention	24	70	3	36	One 10‐minute session Intervention participants = combined prenatal intervention arm and postnatal intervention arm
	Up‐to‐date immunisation status at 6 months of age for Hib, PCV13, DTaP‐IPV, HBV, and rotavirus	Saitoh 2017	6 months post‐intervention	22	51	21	45	Three 5‐minute sessions Figures adjusted to account for clustering (original intervention N = 100; control N = 88)
	Receipt of DPT3	Usman 2009	90 days post‐intervention	242	375	205	375	One 2‐ to 3‐minute session
	Receipt of DPT3	Usman 2011	90 days post‐intervention	228	376	149	378	One 2‐ to 3‐minute session
Longer sessions (11 min+)	Receipt of MMR vaccine	Jackson 2011	3 months post‐intervention	18	19	18	25	One 2‐hour session Figures adjusted to account for clustering (original intervention N = 29; control N = 37)
	Up‐to‐date status at 12 months of age (1 dose BCG, 3 doses HepB, 3 doses OPV, 3 doses DTP, 1 dose MR, 1 dose JEV)	Hu 2017	12 months post‐intervention	376	418	359	433	One 15‐minute session
	Up‐to‐date immunisation status at 3 months of age (BCG, > 2 doses diphtheria and pertussis vaccine, OPV)	Bolam 1998	3 months post‐intervention	Groups A + B = 179	Groups A + B = 205	Groups C + D = 169	Groups C + D = 198	One 20‐minute session Groups A + B = education at birth Groups C + D = no education at birth
	Up‐to‐date immunisation status at 6 months of age (BCG, 3 doses diphtheria and pertussis vaccine, OPV) time point not included in meta‐analysis	Bolam 1998	6 months post‐intervention	Group B = 100	Group B = 104	Group D = 91	Group D = 97	One 20‐minute session Group B = education at birth only Group D = no education

Table 3. Vaccination status outcome data

Table 4. Knowledge outcome data

Outcome	Study	Scale used	Timing of outcome assessment (days/months)	Intervention group			Control group			Notes
Outcome	Study	Scale used	Timing of outcome assessment (days/months)	mean/mean change	standard deviation	N	mean/mean change	standard deviation	N	Notes
Knowledge of immunisation schedule (multi‐component intervention – did not use immunisation status outcome)	Quinlivan 2003	0 to 10	6 months postpartum			62			62	'There were no significant differences in knowledge of infant vaccination schedules between the two groups at the antenatal (intervention: median 0 (IQR 0.1); control: 0 (0.2); P = 0·54) or postnatal assessments (4 (1.6); 2 (1.4); P = 0·08). The unadjusted mean difference in knowledge between the two groups was 0.85 points (95% CI –0.06 to 1.76)'
Knowledge of immunisation contraindications (multi‐component intervention – did not use immunisation status outcome)	Wood 1998	0 to 3	At initial interview, and interview following conclusion of intervention 15 mo later	2.7	1	185	2.6	1	180
Basic knowledge and knowledge of diseases	Saitoh 2013	0 to 13 VPD^a score	100 days post‐intervention	prenatal intervention 4.5	prenatal intervention 2.8	prenatal intervention 34	4.6	3.8	36	We combined the separate scores for VPD knowledge and basic knowledge into a single knowledge score, and data from the two intervention arms were combined into single intervention group in the meta‐analysis
		0 to 13 VPD^a score	100 days post‐intervention	postnatal intervention 4.5	postnatal intervention 2.9	postnatal intervention 36	4.6	3.8	36
		0 to 10 basic knowledge score	100 days post‐intervention	prenatal intervention 3.4	prenatal intervention 1.8	prenatal intervention 34	1.9	1.9	36
		0 to 10 basic knowledge score	100 days post‐intervention	postnatal intervention 2.6	postnatal intervention 1.5	postnatal intervention 36	1.9	1.9	36
Basic knowledge and knowledge of diseases	Saitoh 2017	0 to 13	1 month postpartum time point not included in meta‐analysis	9.8	2.1	51	9.8	1.9	45	4 questions about vaccine‐preventable diseases, 5 basic immunisation knowledge questions, and a question asking participants to identify four recommended vaccines from a list of 12 Figures adjusted to account for clustering (original intervention N = 100; control N = 88)
Basic knowledge and knowledge of diseases	Saitoh 2017	0 to 13	6 months postpartum	10.7	1.6	51	10.5	2.0	45
Knowledge of MMR	Jackson 2011	0 to 11	1 week post‐intervention time point not included in meta‐analysis	8.22	1.5699	45	7.83	1.2709	45	Figures adjusted to account for clustering (original intervention N = 68; control N = 67)
Knowledge of MMR	Jackson 2011	0 to 11	3 months post‐intervention	7.3	1.5632	45	7.08	1.183	45
Knowledge about vaccines, infectious diseases, contraindications, schedule	Bjornson 1997	16‐question test, scores reported for individual questions	immediately following intervention			128			99	test scores broken down by question rather than by person. Authors could not provide mean scores. This study was described narratively in the review.
^aVPD: vaccine‐preventable diseases

Table 4. Knowledge outcome data

Table 5. Attitudes outcome data

Outcome	Study	Scale used	Timing of outcome assessment (days/months)	Intervention group			Control group			Notes
				mean/mean change	standard deviation	N	mean/mean change	standard deviation	N
Parents’ beliefs about the necessity of MMR	Jackson 2011	4 to 20	1 week post‐intervention time point not included in meta‐analysis	17.63	2.7267	45	17.18	2.8288	45	SD calculated from CI provided in study. Figures adjusted to account for clustering (original intervention N = 68; control N = 67) Four items (1 = strongly disagree to 5 = strongly agree) assessed parents’ beliefs about the necessity of the MMR vaccine, e.g. ‘Without the combined MMR vaccine, my child could get very ill from measles, mumps or rubella’ additional attitude measures reported separately
			3 months post‐intervention	17.43	2.3136	45	17.21	3.0748	45
Perceived severity of vaccine‐preventable disease	Saitoh 2013	2 to 10	100 days post‐intervention	prenatal int 8.6	prenatal int 1.3	prenatal int 34	8.2	1.5	36	Two items (1 = strongly disagree to 5 = strongly agree) assessing parents’ perceptions about the severity of vaccine‐preventable diseases In the meta‐analysis, we combined the data from the two intervention arms into a single attitude score for all intervention participants Additional attitude measures reported separately
				postnatal int 7.9	postnatal int 2	postnatal int 36
Perceived severity of vaccine‐preventable disease	Saitoh 2017	2 to 10	1 month post intervention time point not included in meta‐analysis	8.3	1.6	51	7.8	1.5	45	Two items (1 = strongly disagree to 5 = strongly agree) assessing parents’ perceptions about the severity of vaccine‐preventable diseases Figures adjusted to account for clustering (original intervention N = 100; control N = 88) Additional attitude measures reported separately
			6 months post intervention	8.3	1.8	51	8.3	1.6	45

Table 5. Attitudes outcome data

Table 6. Intention to vaccinate outcome data

Outcome	Study	Scale used	Timing of outcome assessment (days/months)	Intervention group			Control group			Notes
Outcome	Study	Scale used	Timing of outcome assessment (days/months)	mean/mean change	standard deviation	N	mean/mean change	standard deviation	N	Notes
Intended MMR choice	Jackson 2011	1 to 7	1 week post‐intervention time point not included in meta‐analysis	6.03	1.4873	45	5.44	1.9679	45	SD calculated from CI provided in study. Figures adjusted to account for clustering (original intervention N = 68; control N = 67). Three items measured on a 7‐point scale e.g. ‘I intend to give my child the combined MMR vaccine at the recommended ages’ (definitely do not to definitely do). Responses averaged over the three items.
Intended MMR choice	Jackson 2011	1 to 7	3 months post‐intervention	6.34	1.4047	45	5.58	2.1319	45
Intent to immunise	Saitoh 2013	1 to 4	100 days post‐intervention	3.7288	0.4484	59	3.4	0.4983	30	Four‐point scale where 1 = no, 2 = undecided, 3 = yes, for a specific vaccine, and 4 = yes. Study authors presented the data for this outcome with both intervention arms combined. Rather than providing mean and SD, the authors reported the number of participants who selected each scale rating. For the meta‐analysis, we transformed these data into mean and SD, with N = total respondents for this question.

Table 6. Intention to vaccinate outcome data

Table 7. Adverse effects outcome data

Outcome	Study	Scale used	Timing of outcome assessment (days/months)	Intervention group			Control group			Notes
				mean/mean change	standard deviation	N	mean/mean change	standard deviation	N
Anxiety	Jackson 2011	20 to 80	1 week post‐intervention time point not included in meta‐analysis	30.89	11.9396	45	33.78	13.7341	45	SD calculated from CI provided in study. Figures adjusted to account for clustering (original intervention N = 68; control N = 67). Six items on a 4‐point scale (not at all to very much) e.g. ‘I feel calm’, ‘I am tense’. Positive items reverse scored, and all six items were summed. The total score was multiplied by 20/6. A normal score is 34 to 36.
			3 months post‐intervention	31.46	12.2701	45	33.39	13.5291	45

Table 7. Adverse effects outcome data

Comparison 1. Face‐to‐face education versus control or non‐face‐to‐face education (all)

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Vaccination status (stratified by length) Show forest plot	7	3004	Risk Ratio (M‐H, Random, 95% CI)	1.20 [1.04, 1.37]

1.1 Short (1 to 10 min)	4	1706	Risk Ratio (M‐H, Random, 95% CI)	1.32 [1.01, 1.73]
1.2 Long (11+ min)	3	1298	Risk Ratio (M‐H, Random, 95% CI)	1.07 [1.00, 1.16]
2 Vaccination status (stratified by number of vaccines delivered) Show forest plot	7	3004	Risk Ratio (M‐H, Random, 95% CI)	1.20 [1.04, 1.37]

2.1 Single vaccine	3	1548	Risk Ratio (M‐H, Random, 95% CI)	1.33 [1.11, 1.61]
2.2 Multiple vaccines	4	1456	Risk Ratio (M‐H, Random, 95% CI)	1.06 [0.95, 1.18]
3 Knowledge Show forest plot	4	657	Std. Mean Difference (IV, Random, 95% CI)	0.19 [0.00, 0.38]

4 Attitudes ‐ necessity Show forest plot	3	292	Std. Mean Difference (IV, Random, 95% CI)	0.03 [‐0.20, 0.27]

5 Intention to vaccinate Show forest plot	2	179	Std. Mean Difference (IV, Random, 95% CI)	0.55 [0.24, 0.85]

6 Adverse effects (anxiety) Show forest plot	1	90	Mean Difference (IV, Random, 95% CI)	‐1.93 [‐7.27, 3.41]

Comparison 1. Face‐to‐face education versus control or non‐face‐to‐face education (all)