PROTOCOL: Use of community participation interventions to improve child immunisation in low‐ and middle‐income countries: A systematic review and meta‐analysis

Immunisation remains one of the most cost‐effective interventions to prevent and control life‐threatening infectious diseases. A study estimating the health and economic impact of routine immunisation for the period 2001–2020, found that introduction and/or increased coverage of vaccines are projected to avert over 14 million deaths, 350 million cases of illness, 8 million cases of long‐term disability and 700 million disability adjusted life years (DALYs) (Ozawa et al., 2017). Yet rates of routine vaccination of children in low‐ and middle‐income countries (LMICs) are strikingly low or stagnant, leading to a high disease burden and high infant and child mortality. In 2017, an estimated 19.9 million infants worldwide were not reached with routine immunisation services. Around 60% of these children live in 10 low‐and‐middle income countries, including Ethiopia, India, Nigeria and Pakistan (“Immunization coverage,” 2018). Furthermore, national averages on immunisation coverage often obscure the underlying disparities within countries, and as a result, inequalities in immunisation often go unobserved or are underreported (Restrepo‐Méndez et al., 2016). Therefore, there is an urgent need for interventions that improve immunisation of children in these countries. A number of innovative approaches to this problem have arisen in recent years. These approaches include strategies and technologies that enhance communities' participation in the planning, delivery, monitoring and uptake of routine vaccinations of children. These strategies have received considerable attention from funders, researchers, and practitioners (UNICEF, n.d.; WHO, 2008). However, there is at present a dearth of rigorous and systematic evidence about the effectiveness and cost‐effectiveness of these interventions. There is therefore a need to make such evidence available to guide policymakers and public health practitioners in making informed decisions about these interventions. This review will systematically analyse the literature on this topic, conduct causal chain analysis to identify key facilitators and inhibitors of intervention success, and perform cost‐effectiveness analysis to assess the value‐for‐money of enhancing community participation to improve immunisation of children.


| The problem, condition or issue
Immunisation remains one of the most cost-effective interventions to prevent and control life-threatening infectious diseases. A study estimating the health and economic impact of routine immunisation for the period 2001-2020, found that introduction and/or increased coverage of vaccines are projected to avert over 14 million deaths, 350 million cases of illness, 8 million cases of long-term disability and 700 million disability adjusted life years (DALYs) (Ozawa et al., 2017). Yet rates of routine vaccination of children in low-and middle-income countries (LMICs) are strikingly low or stagnant, leading to a high disease burden and high infant and child mortality. In 2017, an estimated 19.9 million infants worldwide were not reached with routine immunisation services.
Around 60% of these children live in 10 low-and-middle income countries, including Ethiopia, India, Nigeria and Pakistan ("Immunization coverage," 2018). Furthermore, national averages on immunisation coverage often obscure the underlying disparities within countries, and as a result, inequalities in immunisation often go unobserved or are underreported (Restrepo-Méndez et al., 2016). Therefore, there is an urgent need for interventions that improve immunisation of children in these countries.
A number of innovative approaches to this problem have arisen in recent years. These approaches include strategies and technologies that enhance communities' participation in the planning, delivery, monitoring and uptake of routine vaccinations of children. These strategies have received considerable attention from funders, researchers, and practitioners (UNICEF, n.d.;WHO, 2008). However, there is at present a dearth of rigorous and systematic evidence about the effectiveness and cost-effectiveness of these interventions.
There is therefore a need to make such evidence available to guide policymakers and public health practitioners in making informed decisions about these interventions. This review will systematically analyse the literature on this topic, conduct causal chain analysis to identify key facilitators and inhibitors of intervention success, and perform cost-effectiveness analysis to assess the value-for-money of enhancing community participation to improve immunisation of children.

| The intervention
The review will focus on all interventions that use community participation or engagement to improve routine immunisation coverage of children. multiple villages. For the purposes of our review, a community comprises a group of people who are served by a particular primary health facility (e.g., a subdistrict-level health centre). Thus, communities encompass a wide range of stakeholders, including caregivers, health service providers, and influential community members such as religious or other traditional leaders. Therefore, our review will include any intervention that is directed towards any of the above types of community members. Interventions that target higher levels of the health system-such as a programme to improve state-level officials' use of immunisation-related data-will be excluded.

| What is community participation?
We define community participation as a process wherein communities are included, to various degrees, in planning, decision-making and implementation of activities that directly impact them (Stuart, 2017). The Spectrum of Public Participation, developed by the International Association for Public Participation (iap2.org), identifies five levels of public (community) participation, ranging from informing communities to empowering them ( Table 1).
The levels defined above refer to degrees of participation rather than types of intervention activities-indeed, these participation levels cut across intervention activities. For example, an intervention to inform or educate community members about the importance of immunisation could engage communities at the inform level (if communities are merely told that an informational campaign will be introduced in their area), at the empower level (if community members direct and manage all aspects of designing and implementing the informational campaign), or at any level in between. Similarly, interventions to motivate community members or increase health provider capacity could engage communities at any of these levels. In addition, a programme might also engage different community members at different levels (for different components of the same programme) or the level of engagement for the same set of participants may change over the course of a programme (Hardy, 2015).
However, it is important to note that the participation spectrum does not assume one level of participation to be better than the others or imply that the deeper the level of engagement, the better.
In other words, the spectrum does not classify the participation levels into a hierarchy. Relatedly, the quality of engagement is distinct from the level of involvement-that is, although the spectrum is arranged from less to more substantive, the fact that interactions are more substantive does not always mean they are of higher quality. Thus, an intervention at one of the higher levels of involvement may still be deficient in the way it incorporates community input and feedback because of poor design or implementation. For example, an intervention could make efforts to understand community concerns and consider them in decision-making, and thus be classified as an involvelevel programme. But those efforts to understand may be weak, or community concerns may be considered but ultimately given minimal weight in decision-making.
It is also important to note that the spectrum aims to capture a wide range of ways that intervention implementers can interact with communities. Although simply informing a community that an intervention is being implemented is arguably not participatory (and hence, interventions at the inform level are arguably not "community participation" interventions), we rely on the Spectrum of Public Participation because it is an established framework for understanding community engagement. In particular, including interventions at the inform level in our review will allow us to assess the hypothesis, implicit in the design of many community-engagement interventions, that informing communities is an insufficient level of engagement to effect change.

| How the interventions might work
A 2015 3ie scoping paper (Sabarwal et al., 2015) systematically mapped the literature on immunisation interventions and also consulted several programme managers and policy experts to understand why community participation could be the key to improving immunisation outcomes for children in areas where the coverage has stagnated or declined or that are hard to reach. The findings from the scoping study indicate that working with or engaging communities could help develop an understanding of the context, target population, problems and barriers, and lead to identification of contextually relevant solutions and desired outcomes, and mobilising community support for them. In addition,

Level of involvement Description
Inform Reaching out to/informing the community about or as a part of an intervention

Consult
Consulting the community as part of a process to develop an intervention/programme, or build community awareness and understanding around an intervention

Involve
Involving the community through a range of mechanisms to ensure that issues and concerns are understood and considered as part of the decision-making process

Collaborate
Collaborating with the community by developing partnerships to formulate and deliver an intervention Empower Shared leadership/empowering the community to make decisions and to implement and manage change.
Source: adapted from International Association for Public Participation (2018) and Auditor-general's report, Government of Victoria, Australia (2017).
existing literature on behaviour change (Bicchieri & Xiao, 2009;Reynolds, Subašić, & Tindall, 2015) highlights the importance of recognising that individuals usually function under the influence of social norms, that is, they respond to their peers and community, and while activities such as information and education campaigns might have some influence, individuals might feel bound by collective decisions (Riedy, 2012). The role of peers and of social norms in shaping attitudes towards vaccination is particularly important given that vaccine hesitancy has been documented in countries of all income levels (although it takes very different forms in different countries; see Dubé, Gagnon, Nickels, Jeram, & Schuster, 2014 not know when and where to take their children for immunisation, while at the same time the delivery of services is irregular. Additionally, the same participation process may also be leveraged to seek more substantive inputs from community members, including the frontline health workers, for designing the mode and content of reminders, targeting the hardest-to-reach households, and for mobilising the community at large. This example highlights how different intensity levels of community participation could give rise to the same activity involving reminders, which in turn could lead to improvement in access to immunisation services and an eventual increase in immunisation rates. Importantly, the theory of change linking community participation to immunisation outcomes does not specify a precise causal mechanism whereby specific intervention activities yield specific behavioural responses and outcomes. Rather, the theory assumes that different barriers to immunisation exist in different communities, and that different categories of activities will be best suited to addressing those barriers. The underlying logic of using community participation to increase immunisation coverage is that engaging communities can be an effective way to (a) identify activities that will address the immunisation barriers extant in that community, and (b) implement those activities in ways that are appropriate for the local context. beneficiary/community-and provider-oriented. These activities lead to outputs and outcomes related to improvement in demand for, delivery of, and access to immunisation services. This ultimately leads to improved timely uptake of vaccination(s) and an increase in full F I G U R E 1 Simplified intervention theory of change immunisation rates. As noted above, the precise causal mechanisms that begin with community engagement and end with improved immunisation rates will differ depending on the level of participation utilised and the nature of the activities selected; consequently, the theory of change does not depict the full details of each involvement level's mode of action. However, note that the theory does rely on the assumption that in order to produce the desired results, specific expertise, commitment, and resources will be required to engage the community effectively at the intended level of participation. The details of the theory of change-and in particular, how different levels of involvement will trigger different causal sequences-will be further developed in the course of the review and evidence synthesis.  Rowe et al., 2018). As these specific barriers are largely absent in HICs, reviews in which much of the evidence comes from HICs are of limited relevance to the needs of policymakers in LMICs.
Below we list 10 existing reviews and briefly document why they do not already accomplish the goals of our proposed review; see "References" section for full references.  ours. In addition, the literature in this area is expanding rapidly, so we anticipate that there will be a substantial amount of new evidence to be included in our review. 1 Moreover, we plan to look at outcomes beyond DPT3 in our review and use an analysis framework focusing on levels of community participation which is different from the one used in this review. In addition, we aim to assess whether there is any evidence of adverse effects of community participation interventions.

| METHODOLOGY
Our review will follow Cochrane and Campbell Collaboration guidelines on conducting SRs. We will conduct a broad, systematic search, apply consistent criteria to the screening of studies for inclusion, and, where appropriate, synthesise studies through metaanalysis. We will present the findings of the review in forest plots and a "Summary of Findings" table.
Items retrieved through the search will be imported to EPPI-Reviewer 4 and independently screened by two trained reviewers at the title/abstract and full-text levels. At the title/abstract level, we will use EPPI-Reviewer's machine learning and text mining features to prioritise citations that are likely to be included and to exclude items that are very unlikely to be included (O'Mara-Eves, Thomas, McNaught, Miwa, & Ananiadou, 2015;Thomas, McNaught, & Ananiadou, 2011). We will begin by manually screening 1,000 random records to train the machine learning algorithm, and have EPPI-Reviewer order the remaining records according to their probability of inclusion as determined by EPPI-Reviewer's algorithm. We will then set the EPPI-Reviewer algorithm to learn continuously as we screen more records, and to continually re-order remaining records according to their probability of inclusion. Thus, as we screen from the top of the list, records with low probabilities of inclusion will continue to be pushed further to the bottom of the list. When we reach a point where we exclude 500 consecutive records, we will automatically exclude the remaining titles (i.e., those with an even lower probability of inclusion than the 500 consecutive exclusions). We will, however, screen a random sample of the automatically excluded records to verify the accuracy of the algorithm.
Full-text screening will be conducted by two independent reviewers, with disagreements resolved through discussion (and the lead author making final decisions on any contested cases). We will document the screening process so that a PRISMA flow chart and a include studies with the following designs: • Randomised controlled trials • Regression discontinuity designs • Instrumental variables estimation • Statistical matching (e.g., propensity score matching) • Difference-in-differences • Fixed effects estimation • Interrupted time series analysis We include both experimental and quasiexperimental designs because randomisation is often impractical or unethical in the context of development interventions. The above designs are all established methods for drawing causal inferences about an intervention's effects (Shadish, Cook, & Campbell, 2002). There will be no inclusion restrictions based on study publication status.
We will not impose any restrictions based on language of publication. Between members of the review team and consultants, we will have the capacity to screen and extract data from studies in Chinese, Spanish, Portuguese and French. If our search retrieves potentially relevant studies in other languages, we will attempt to find qualified reviewers to screen and, if they meet our inclusion criteria, extract data from these studies.
The review will include only studies that measure the effectiveness of an intervention, rather than studies of efficacy. Because studies fall along a continuum from efficacy to effectiveness, rather than fitting cleanly into one category or the other (Singal, Higgins, & Waljee, 2014;Thorpe et al., 2009), we will not have precise determinative criteria to distinguish efficacy from effectiveness studies. Rather, we will use a set of questions to guide the decision of whether to classify a study as an efficacy trial (and hence ineligible for inclusion). The more a study exhibits the characteristics described in the questions below, the stronger will be the case for classifying it as an efficacy trial. 5 Reviewers will be trained to flag any unclear cases for further review and discussion, with the lead author making final decisions on any contested cases.

| Research objective
(1) Does the study aim to determine how an intervention (e.g., technology, treatment, procedure or service) functions under ideal conditions, as opposed to how the intervention functions under "real-world" conditions (i.e., approximating the conditions that would inhere in a large-scale rollout of the intervention)?

| Population
(2) Are the participants likely unrepresentative of the general population? For example, are strict inclusion and exclusion criteria used to enrol a relatively homogenous population, which may limit the generalisability of the results?

| Providers
(3) Is the intervention primarily delivered by researchers, rather than people who might be expected to deliver the intervention under large-scale rollout conditions (e.g., health workers, community leaders, or NGOs)?
Because the distinction between efficacy and effectiveness is a subtle one, we do not anticipate that reviewers will be able to determine which category a study falls into based on the title or abstract. Thus, only the full-text screening protocol will include questions related to this distinction.

| Types of participants
For this review, we will include studies evaluating programmes targeted at rural, peri-urban and urban populations to increase immunisation of children under five in LMICs (where a country's income status is determined by its World Bank classification at the time an intervention was carried out). All interventions that do not focus on immunisation outcomes for children (e.g., vaccination campaigns targeting adults) will be excluded. We will also exclude studies from high income countries, as the evidence from these countries will have limited applicability to the contextual factors prevalent in LMICs. As for the participants, while our review will focus on interventions targeting community members, interventions need not target the community as a whole; it is sufficient that an intervention targets any subgroup of a community as defined above.
When only a subset of the participants/beneficiaries in a study are eligible (e.g., a study evaluating an intervention's effects on both childhood and adult vaccination), we will include the study only if the study reports outcomes separately for the eligible participants/beneficiaries.

| Types of interventions
Any intervention involving some degree of community participation, where our definitions of "community" and "participation" are given above, will be eligible for inclusion. The types of relevant interventions include, but are not limited to: Interventions to promote communication and community dialogue about routine immunisation • Interventions to promote community participation in the design of activities to encourage routine immunisation • Interventions to promote community participation in designing and implementing routine immunisation delivery services • Interventions to improve monitoring and accountability mechanisms for routine immunisation-related activities • Peer recognition and monetary incentives to improve routine immunisation outcomes for children Interventions that do not engage communities as defined above (i.e., people served by a given primary health facility) will be excluded. Also excluded will be interventions that do not involve any of the participation levels in the Spectrum of Community Participation (e.g., those that do not even inform communities about an intervention). 6

| Types of outcome measures
To be included, studies must assess at least one of the primary or secondary outcomes identified below.

Primary outcomes
• Full immunisation coverage (FIC) • Antigen specific immunisation coverage (e.g., DPT3 or Penta 3 coverage, measles coverage) • Timely uptake of vaccines and key downstream effects (e.g., morbidity and mortality) of the primary outcomes. In addition, although some outcome measures might differ in degree (full, partial or no routine immunisation) or might be a subset of each other (dropout after DPT1/Penta 1 will be categorised as partially immunised), the outcomes are wide ranging to account for the differences and preferences of authors in reporting them.
Both official health records (including vaccination cards) and parent recall will be considered acceptable measures of immunisation coverage. If a study reports both types of measures, we will extract data on both. We will note which measure the authors believe to be more reliable and treat this as the default "true" measure of the outcome. However, we will conduct sensitivity analysis to determine whether using a different coverage measure significantly affects the results of our analysis.
Although there are standardised tools for measuring constructs like parental attitudes towards vaccination (e.g., Oladejo et al., 2016;Shapiro et al., 2018), these have not yet been universally adopted or validated for populations of interest to this review (e.g., rural populations in LMICs). Therefore, we will defer to authors' definitions of these constructs and their instruments for measuring them.

| Duration of follow-up
Because the timeline for administering most childhood vaccinations is short (most should be administered in the first 14 weeks after birth), there will be no restrictions on duration of follow-up.

| Types of settings
As noted above, studies must have been conducted in a LMIC to qualify for inclusion. Apart from this, there will be no restrictions on types of settings. We anticipate that most studies will cover rural areas, but peri-urban and urban settings are also eligible for inclusion.
Taking into account all of the above criteria, our scoping work has identified the following studies as ones that would likely meet our criteria and be included in the review:

| Search strategy
We will conduct a comprehensive search to identify as much relevant literature as possible. Appendix 1 includes a draft set of search terms to be used in electronic searches, and an example full search strategy for a key database (MEDLINE). The precise search strings and logic (e.g., index terms, truncation operators) will be adapted for each database. The review will document the full search strategy, including the specific strings used for each database. We will ensure that information on errata is included in results downloaded from electronic databases (hosts), so that errata can be consulted for included studies.
We will conduct electronic searches of the following databases of published sources: • MEDLINE • CAB Global Health

• Campbell Library
To identify relevant grey literature, we will search the following databases and websites (some of which contain a mixture of published and grey literature): • Google Scholar • Eldis • GAVI

• Epistemonikos
• Innovations for Poverty Action (IPA) • Abdul Latif Jameel Poverty Action Lab (J-PAL) • 3ie Impact Evaluation Repository • 3ie Systematic Review Repository • Registry of International Development Impact Evaluations (RIDIE) • Global Development Network • World Bank Development Impact Evaluation (DIME) and Impact

Evaluation Policy Papers
• Inter-American Development Bank • Center for Global Development • Center for Effective Global Action (CEGA) • DFID Research for Development (R4D) • USAID In addition, we will be working with a team of researchers from Hong Kong University, who are experienced in information retrieval, to translate the search strategy and conduct searches of the following databases: • Web of Science TM-Chinese Science Citation Database SM Once we have identified studies to be included, we will review references in included studies for additional sources, as well as search (using Google Scholar) for sources that cite included studies.
Although the review will include only primary studies, we will also conduct a separate search for other relevant SRs. We will have a screening protocol for these reviews to identify those relevant to our research questions, and we will check the references of these reviews for additional sources. 7 We will also share our included studies list with experts in the field and request them to identify any omissions.
In addition to an assessment of the effectiveness of immunisation interventions, our review will also include cost-effectiveness analysis (Dhaliwal, Duflo, Glennerster, & Tulloch, 2011), as well as 7 Note that, simultaneously with this SR, we will be completing an evidence gap map (EGM) on the same topic that will include systematic reviews. Thus, the reviews whose references we check will be those that are included in our EGM.
causal chain analysis (Kneale, Thomas, Bangpan, Waddington, & Gough, 2018) to identify facilitators and inhibitors of intervention effectiveness. To identify sources relevant to these analyses, we will conduct a targeted search for additional information on the interventions included in the effectiveness review, including cost information, process evaluations, and feasibility studies. We will use citation tracking to identify companion papers to included impact evaluations, as well as database and website searches using names of programmes and authors of included studies. We will also contact authors and implementing agencies to request additional project information if required.
We will conduct the first round of search shortly after the protocol is approved. Shortly before we begin to write up our findings in a SR report, we will re-run the main electronic searches to identify any relevant papers that have been published or indexed while the SR was in progress. We will screen the newly-retrieved studies and those that meet the eligibility criteria will be included in the review.

| Description of methods used in primary research
We anticipate that most studies will use either a randomised controlled trial or difference-in-differences methods, but we will include studies employing any of the designs enumerated above in "Types of study designs".

| Criteria for determination of independent findings
Where more than one paper or report is identified on a single study, we will choose one as the "main" paper. A study may be designated as the main paper because it reports on our primary (as opposed to secondary) outcomes of interest or because it contains more detailed reporting of results. When multiple papers report different results on an identical outcome, we will contact the authors to enquire about the differences and choose the results that more accurately reflect the impact of the intervention as relevant to our research questions.
If contacting the author does not yield a clear decision, we will use results from the paper with the latest publication date. Other papers from a study (besides the main paper) will be considered "secondary reports." We will use these secondary reports to provide additional information about that one study, which may include outcome measures not reported in the main paper or supplementary information about implementation or costs.
Where information is collected on the same intervention for different outcomes at the same or different periods of time, we will extract information on the full range of outcomes over time. We will identify the most common follow-up period and include the follow up measures that match this most closely in the meta-analysis. Where multiple outcomes are reported from different specifications, we will select the specification with the lowest risk of bias in attributing impact, for example, the most appropriately specified outcomes equation. When studies include multiple outcome measures to assess related outcome constructs (e.g., measuring access to immunisation services through both number of days when health workers are present at health facilities and number of days when vaccines are stocked), we will select the outcome that appears to most accurately reflect the outcome construct of interest.
3.5 | Critical appraisal 3.5.1 | Assessment of risk of bias in experimental and quasiexperimental studies (Review Questions 1-2) We will assess risk of bias based on categories of bias recommended by the Cochrane Non-Randomised Studies Group and procedures recommended by Waddington et al. (2017). These tools have been developed to assess the risk of bias for a range of quasiexperimental studies, as well as experimental studies. We will assess risk of bias based on the following criteria, coding each paper as "Yes," "Probably Yes," "Probably No," "No" and "Unclear" according to how they address each domain:

Attrition
The risk of bias for each included study will be conducted by two independent reviewers, with disagreements resolved through discussion (and the lead author making final decisions on any contested cases). We will report the results of the assessment for each of the assessed criteria for each study, and note the source of each judgement (e.g., page numbers of included studies where information is reported).
In addition, we will attempt to explore if there are systematic differences in outcomes between primary studies with different risk of bias. If meta-analysis is feasible, we will conduct sensitivity analysis to assess the robustness of the results to the risk of bias in included studies.

| Critical appraisal of qualitative studies, process evaluations and project documents (Review Questions 3 and 4)
To address Review Questions 3 and 4, we will be drawing on a broader range of sources beyond impact evaluations, including JAIN ET AL.

| 9 of 20
qualitative studies, process evaluations and project documents. We will adopt different approaches to appraise the different types of documents, as outlined below.
We will assess the quality of included qualitative studies and descriptive quantitative studies using an adapted version of the Critical Appraisal Skills Programme checklist (CASP, 2018) developed by Snilstveit et al. (2015), making judgements on the adequacy of reporting, data collection, presentation, analysis and conclusions drawn. The draft checklist is included in Appendix 2, which may be adapted following further piloting.
For critical appraisal of process evaluations, we will draw on and adapt several resources. Rigorous process evaluations require reliable data from a representative sample, so assessment of sampling and methods of data collection are obvious issues to consider (Snilstveit et al. 2015), along with discussions of the assumptions underlying the evaluation and the transferability of results (Jimenez et al., 2018). We will draw on the above sources, along with existing guidelines for process evaluations (Scriven, 2015) and the CASP checklist to create a quality appraisal checklist suited to process evaluations.
Project documents provide information about planned, ongoing or completed interventions and we will be using them in our review to get sufficient information about the context and design of the interventions included in our review. As this information is descriptive we will not formally appraise the quality of such documents.

| Critical appraisal of cost evidence (Review Question 5)
For appraisal of cost evidence we will mainly draw on the approach taken in Doocy and Tappis (2017). They adapt two guides to the use and appraisal of cost evidence; the German Federal Ministry for Economic Cooperation and Development's Tools and Methods for Evaluating the Efficiency of Development Interventions (Palenberg, 2011) and the Campbell Collaboration Economic Methods Policy Brief (Shemilt et al., 2008).

| Details of study coding categories
Studies that meet the inclusion criteria will go through a coding procedure to: record information about the study's publication, setting, and design; extract data on reported results; and assess risk of bias. All coding will be done by two independent, trained coders, with discrepancies resolved through discussion (and the lead author making final decisions on any contested cases). Table 2 presents a draft of the coding fields and instructions.

| Statistical procedures and conventions
We will use forest plots to display the distribution of effect sizes and confidence intervals for similar outcomes across included studies.
Where studies are sufficiently similar in terms of their interventions, populations, and designs, we will use meta-analysis to pool and analyse results. As recommended by the Cochrane Handbook of Systematic Reviews (section 9.4.4.1), we will weight studies in the metaanalyses using the Mantel-Haenszel method. Given that contextual factors are likely to vary significantly across studies, we plan to use random-effects meta-analysis. Where outcome measures are dichotomous (i.e., vaccination received or not), we will extract data and calculate log-transformed risk ratios to derive standardised effect sizes across studies. For any continuous outcomes (e.g., knowledge about or attitudes towards immunisation), we will extract data and calculate the standardised mean difference (SMD), to which we will apply the Hedge's g transformation. We will conduct separate metaanalyses for different immunisation-related outcomes (e.g., DPT3 coverage and FIC). Meta-analysis results will be presented in forest plots. We will check the forest plots against results from included studies to check that data have been extracted and coded appropriately (e.g., direction of effect has not accidentally been reversed).
When data cannot be extracted in a usable fashion from included papers, we will contact the authors to request the necessary data. If this fails, studies without usable data will be included in the review even if they cannot be included in the meta-analyses.
Where cluster-randomised trials have appropriately adjusted for cluster effects (e.g., by using clustered standard errors), we will rely on the effect sizes and standard errors reported in the studies. Where authors have not appropriately adjusted for clustering, we will make adjustments as recommended in the Cochrane Handbook for Systematic Reviews (Higgins, Eldridge, & Li, 2019) and related literature (e.g., Donner & Klar, 2002). When there are multiple treatment arms being compared to a single control arm (e.g., Treatment A vs. control and Treatment B vs. control), we will include only intervention and control groups that meet eligibility criteria. When multiple arms of the same study meet eligibility criteria, we will divide the control group sample size by the number of treatment arms to avoid double counting.
We will use the Q, τ 2 , and I 2 statistics to analyse the heterogeneity of results across studies, and attempt to explain observed heterogeneity through moderator analysis, including subgroup analysis and, if applicable, meta-regression. We will conduct moderator analyses to investigate sources of heterogeneity. We will use the framework of Lipsey (2009) which classifies moderators into three broad categories of extrinsic, methodological and substantive characteristics. We aim to include extrinsic variables such as funder and publication date; methodological variables such as study design and risk of bias; and substantive variables such as participant characteristics (gender, socioeconomic status), and context (country, geographical setting). While the exact set of moderators will be determined as we learn more about the characteristics of included studies, at a minimum we will (again, assuming meta-analysis is feasible at all) present meta-analyses that stratify by study design (experimental vs. quasiexperimental), by risk of bias, and by level of community involvement. In addition, we hope to be able to analyse the role of quality of engagement as a moderator, though we may be limited by the detail of reporting in included studies. Equity description Open answer-provide a description of how the study considers gender and equity, and for which population to corroborate answers above (include page numbers where relevant) Adverse effects Input any potential adverse effects of the intervention that the study reports on. If study does not report on any potential adverse effects, code as "Not applicable" Evaluation design The general evaluation design (i.e., experimental or quasiexperimental)

Evaluation method
The specific method used to evaluate impact (mainly applicable to quasiexperimental studies-regression discontinuity, instrumental variables, etc.)

Unit of observation
The unit(s) of observation/analysis used in the study, for example, individual, household, village, and so forth Programme name Input the name of the project/programme being evaluated (if any) Implementing agency Input the name of the agency or agencies implementing the programme Programme funding agency Input the name of the agency or agencies funding the programme (note: this is not the same as organisations that fund the research of the evaluation) Research funding agency Input the name of the agency or agencies funding the research (note: this is not the same as organisations that fund the programme) Number of study arms The number of separate study arms (e.g., control + two versions of treatment = 3 arms).
Control condition description Briefly describe the "usual practice" that the control group receives while the treatment group is receiving the intervention T A B L E 2 (Continued)

Coding field Description/coding instructions
Intervention A: quality of engagement Specify whether the quality of engagement with the community was low, medium, high or cannot be determined from the reporting in the study Intervention B: primary targets Who are the primary targets of the intervention (i.e., those who have substantive interactions with programme implementers from outside the community)? For example, caregivers, health workers, or community leaders. If only one treatment arm, code as "Not applicable" Intervention B: secondary targets Who are the secondary targets of the intervention (i.e., those who, as part of the intervention's design, have substantive interactions with the primary targets). For example, caregivers who receive information from community leaders, who are the primary targets as defined above. If there are no secondary targets, or if there is only one treatment arm, code as "Not applicable" Intervention B: activities Briefly describe the activities that comprise the intervention. If only one treatment arm, code as "Not applicable" Intervention B: level of involvement Specify the intervention's level of involvement of the community (Inform, Consult, Involve, Collaborate or Empower). This should be the intervention's level of involvement as it was actually implemented (regardless of how it was designed). If only one treatment arm, code as "Not applicable" Intervention B: quality of engagement Specify whether the quality of engagement with the community was low, medium, high, or cannot be determined from the reporting in the study. If only one treatment arm, code as "Not applicable"

Add rows for additional interventions as needed
Take-up of intervention Does the study report on the level of intervention take-up?
Knowledge/awareness/attitudes Does the study report on changes in participants' knowledge, awareness, or attitudes?
Access to immunisation services Does the study report on changes in participants' access to immunisation services?
Health-seeking behaviour Does the study report on changes in participants' health-seeking behaviour?
Timely uptake of vaccinations Does the study report on changes in timely uptake of vaccinations?
DPT3/penta3 coverage Does the study report on coverage of DPT3 or penta3 vaccines?
Measles coverage Does the study report on coverage of measles vaccination?
Antigen-specific coverage Does the study report on immunisation coverage for other specific antigens?
DPT1-DPT3/penta1-penta3 dropouts Does the study report on rate of dropouts from DPT1 or penta1 to DPT3 or penta3?
OPV1-OPV3 dropouts Does the study report on rate of dropouts from OPV1 to OPV3 Morbidity Does the study report on morbidity outcomes?

Mortality
Does the study report on mortality outcomes?
Gender Does the study report on differential impacts by gender?
Socioeconomic status Does the study report on differential impacts by socioeconomic status (e.g., wealth quantile)?
Rural versus peri-urban versus urban Does the study report on differential impacts by rural/peri-urban/urban population setting?
Other vulnerable group Does the study report on differential impacts by any other vulnerable group? More generally, as we continue to refine our theory of change, we will identify key pathways in the causal chain and identify study characteristics that could plausibly make a difference to the functioning of these pathways.
If there are more than 10 studies reporting the same outcome that can be combined in a meta-analysis, we will use funnel plots to examine risk of bias due to missing results (Page, Higgins, & Sterne, 2019).
To incorporate cost data and to undertake cost effectiveness analysis, we will largely follow the methodology in Dhaliwal et al. (2011). We expect that most of this data will not be found in the published papers, so we anticipate having to go back to authors to collect this data. Costs will include the costs of administration, targeting, training, monitoring, materials, users, and so forth. On impact size, we will focus on one outcome at a time (e.g., full immunisation), but also make sure to indicate if the programme affected other outcomes. Insignificant impacts will not be included in the costeffectiveness analysis.
We will use Review Manager 5 to complete and submit the review, while using Stata 14 for all statistical analyses, including the metan command for meta-analyses.

| Treatment of qualitative research
We will use qualitative research to supplement the findings of our quantitative synthesis. While we will not seek out all qualitative studies relating to community engagement and immunisation in LMICs, we will look for qualitative studies to provide additional information about the context and the implementation of interventions If not reported, code as "not specified" Outcome 1: Control group "yes" The number of participants in the control group who attained the reported outcome (e.g., received DPT3) Outcome 1: Control group "no" The number of participants in the control group who did not attain the reported outcome (e.g., received DPT3) Outcome 1: Treatment A "yes" The number of participants in Treatment A who attained the reported outcome (e.g., received DPT3) Outcome 1: Treatment A "no" The number of participants in Treatment A who did not attain the reported outcome (e.g., received DPT3) Outcome 1: Treatment B "yes" The number of participants in Treatment B who attained the reported outcome (e.g., received DPT3). If there is only one treatment arm, code as "Not applicable" Outcome 1: Treatment B "no" The number of participants in Treatment B who did not attain the reported outcome (e.g., received DPT3). If there is only one treatment arm, code as "Not applicable" Outcome 1: Treatment C "yes" The number of participants in Treatment C who attained the reported outcome (e.g., received DPT3). If there are fewer than three treatment arms, code as "Not applicable" Outcome 1: Treatment C "no" The number of participants in Treatment C who did not attain the reported outcome (e.g., received DPT3). If there are fewer than three treatment arms, code as "Not applicable" Add rows for additional outcomes as needed Cost per additional child fully immunised Input the reported cost per additional child fully immunised. If not reported, code as "not applicable" Cost per additional child not dropping out from DPT1 or penta1 to DPT3 or penta3 Input the reported cost per dropout prevented. If not reported, code as "not applicable" Cost per DALY averted Input the reported cost per DALY averted. If not reported, code as "not applicable" Cost per additional death averted Input the reported cost per additional death averted/life saved. If not reported, code as "not applicable" included in the quantitative synthesis. This may include feasibility studies, stakeholder analyses, formative evaluations, process evaluations, and project reports, among other documents. These sources will provide key inputs to our analysis of the facilitators and inhibitors of effective community engagement interventions.

SOURCES OF SUPPORT
This work is supported by the International Initiative for Impact Evaluation (3ie).

DECLARATIONS OF INTEREST
The International Initiative for Impact Evaluation (3ie) provides funding and technical assistance for seven ongoing impact evaluations of community engagement interventions for immunisation as a part of its immunisation evidence programme. This technical assistance includes, but is not limited to: reviewing study designs, analysis plans, and data collection instruments; advising research teams on how to improve study components and address challenges that arise during the course of the evaluation; and supporting grantees in engaging with stakeholders to promote uptake and use of evidence generated by the evaluations.
As members of 3ie staff, authors Monica Jain, Mark Engelbert, Marie Gaarder, and Avantika Bagai have all had varying levels of involvement in reviewing proposals for these evaluations and providing research teams with technical assistance. These SR authors therefore have a vested interest in the success and prominence of these studies, which fall under the scope of the proposed SR. This presents a conflict of interest in that the authors may apply different standards to these studies when reviewing them for inclusion in the SR or deciding how much weight to give them in the analysis.
However, there are several procedural safeguards and transparency measures in place that mitigate the risk this conflict of interest imposes. First, all candidate studies, including those funded by 3ie, will undergo a rigorous multi-step screening process, including review at the title, abstract, and full-text levels. To qualify for inclusion in the SR, a study must be judged to meet the inclusion criteria by two independent screeners who have reviewed the full text of the study. The screening protocol will be made publicly available, as will the protocol for analysing included studies.
Moreover, the screening for this SR will be conducted by independent consultants who, while paid by 3ie, will not be 3ie staff and will not have provided technical assistance or interacted with authors of 3ie-funded studies. Finally, the authors will, upon request, provide full records of each round of screening, detailing the studies that were excluded and included at the title, abstract, and full-text levels.
The authors have no financial interests in this area and have not published any prior reviews on the topic.

PRELIMINARY TIMEFRAME
We plan to submit the review in or around February, 2020.

PLANS FOR UPDATING THE REVIEW
Contingent on funding, this review will be updated two years after the original review is published. Monica Jain will lead the update. If for any reason we are unable to carry out this update, we will notify the International Development Coordinating Group and, if requested, help identify a suitable team of authors to update the review.
or Belize or Bhutan or Bolivia or Bosnia or Herzegovina or Her-