Community engagement to reduce inequalities in health: a systematic review, meta-analysis and economic analysis

Identifying systematic reviews

We searched a range of registers, websites and databases for systematic reviews that discuss how some or all of their included studies contain interventions that utilise community engagement. The reviews were used to identify included primary studies that are relevant to the scope of this project; the systematic reviews themselves were not included in the syntheses in this project (see Study selection and eligibility criteria).

The systematic review registers, websites and databases that we searched were:

1. Database of Promoting Health Effectiveness Reviews (DoPHER). DoPHER is developed and maintained by the EPPI-Centre. It has focused coverage of systematic and non-systematic reviews of effectiveness in health promotion and public health worldwide. It currently contains details of > 2500 reviews of health promotion and public health effectiveness. All reviews are assessed and coded for specific characteristics of health focus, population group and quality (URL: http://eppi.ioe.ac.uk/webdatabases/Intro.aspx?ID=2, accessed 15 March 2013).

2. Cochrane Database of Systematic Reviews (CDSR). CDSR includes all Cochrane Reviews (and protocols) prepared by Cochrane Review Groups in The Cochrane Collaboration. As of Issue 5, 2011, CDSR includes 6641 articles: 4622 reviews and 2019 protocols (URL: www.thecochranelibrary.com/view/0/AboutTheCochraneLibrary.html#CDSR, accessed 15 March 2013).

3. Database of Abstracts of Reviews of Effects (DARE). DARE is developed and maintained by the Centre for Reviews and Dissemination (CRD) and currently contains > 21,000 systematic reviews. It is focused primarily on systematic reviews that evaluate the effects of health-care interventions and the delivery and organisation of health services. The database also includes reviews of the wider determinants of health such as housing, transport and social care when these impact directly on health, or have the potential to impact on health (URL: www.crd.york.ac.uk/CMS2Web/AboutDare.asp, accessed 15 March 2013).

4. The Campbell Library. The Campbell Collaboration’s library of systematic reviews includes reviews and protocols prepared by Campbell review groups under any of the six co-ordinating group themes: crime and justice, education, international development, methods, social welfare and review users (URL: www.campbellcollaboration.org/library.php, accessed 15 March 2013).

5. NIHR Health Technology Assessment (HTA) programme website. The HTA programme produces research about the effectiveness of different health-care treatments and tests for those who use, manage and provide care in the NHS. The HTA website houses all of the reviews published through the HTA programme in the HTA journal series and holds in excess of 550 titles (URL: www.hta.ac.uk/project/htapubs.asp, accessed 15 March 2013).

6. HTA database hosted by the CRD. This database currently holds > 10,000 summaries of completed and ongoing HTAs from around the world. Database content is supplied by the 52 members of the International Network of Agencies for Health Technology Assessment (INAHTA) and 20 other HTA organisations worldwide (URL: www.crd.york.ac.uk/crdweb/AboutHTA.asp, accessed 15 March 2013).

Identifying primary research through the Trials Register of Promoting Health Interventions and NHS Economic Evaluation Database

Searches of the systematic review resources were supplemented by searches of the TRoPHI database and the NHS Economic Evaluation Database (NHS EED).

The TRoPHI database includes focused coverage of trials of interventions in health promotion and public health worldwide. It covers both randomised and non-randomised controlled trials and currently contains details of > 4500 trials and is updated four times a year. This source was searched to ensure that relevant trials published outside of the time frame or scope of the reviews identified in the review databases listed in the previous section are detected.

Part of the TRoPHI data set was used in a comparison of randomised and non-randomised trials49 and we proposed to add additional studies from reviews that were carried out since this study. The approximately 300 studies in this data set were already classified using one of two data collection tools that capture detailed information about their methodology, participants, planning and process measures (if any), intervention and outcomes. [Although many of the studies were potentially relevant, surprisingly few passed both our ‘community engagement’ and ‘inequalities’ filter, and so the vast majority (> 99%) of studies synthesised came from our other searches.]

The NHS EED includes records of economic evaluations of health-care interventions, including cost–benefit, cost–utility and cost-effectiveness analyses; the database currently includes > 11,000 economic evaluations. The database is maintained through weekly literature searches that are conducted by the CRD.

Other search sources

The final component in our search strategy was contact with authors of identified studies. We contacted authors of a small number of key studies that were excluded on methodological grounds to ask them if they had outputs that would meet our inclusion criteria, or if they could provide further information about the study to assess its suitability for inclusion. ‘Key studies’ were large-scale UK-based evaluations, such as the Health Action Zones initiative.

Dates of searches

• DoPHER: 26 July 2011.

• TRoPHI: 16 August 2011.

• The Campbell Library, CDSR, DARE, HTA and NHS EED: 17 August 2011.

• Supplementary search of HTA journal in Web of Knowledge for papers published 2010–11: 18 August 2011.

Selecting reviews

The purpose of this stage was to identify reviews that might include relevant primary studies. The following criteria were applied to titles and abstracts of reviews:

• published after 1990 (the date cut-off set by other reviews on which we are building, e.g. Popay et al. 4)

• a systematic review (i.e. describe search strategies and inclusion criteria used)

• included outcome, economic, or process evaluation studies

• described at least one intervention potentially relevant to community engagement

• included at least one study in the results section

• written in English

• measured and reported health or community outcomes.

Studies were limited to the English language because of a lack of resources to translate documents. Each systematic review was assessed against these criteria in a stepwise fashion, such that any review excluded because it failed a criterion later in the list must have passed any preceding criterion. We were deliberately inclusive when considering the concept of community engagement at this stage to avoid missing any reviews that might include studies with community engagement even though it was not mentioned in the review’s abstract; as such, reviews that referred to community-level interventions were typically included except when this conflicted with the inclusion criteria above. Indeed, applying a consistent definition of community engagement across reviews was a challenge. In initial stages we relied on reviewers’ previous experience with the literature and developed written guidance on the definition within EPPI-Reviewer 4 as understandings about the concept emerged through group discussion about the interventions, providers, locations and study aims described across the reviews. It is possible that screening titles and abstracts of reviews may have missed some primary studies that had elements of community engagement. However, because of the reflective method used in consolidating our definition of community engagement early on in the screening process, this is more likely to be because of a lack of detail in the reference information available than because of systematic bias from reviewers.

We then retrieved the full-text copy of all reviews that passed these inclusion criteria. A brief screening of the full-text document was then conducted to check that the review was, in fact, systematic and that it included primary studies of relevance to our review. (Relevance at this stage was judged according to the criteria presented in Selecting trials for the map, although the criteria were not applied stepwise and were not recorded.) Potentially relevant primary studies were then added to the EPPI-Reviewer 4 database for the second stage of screening.

Selecting trials for the map

Once the final set of systematic reviews was obtained, we screened within each review to identify potentially relevant primary studies (trials). This involved scanning the evidence tables and reference lists of the reviews for relevant trials. We then located the abstracts for these trials.

The titles and abstracts of the trials identified during this process, plus those identified through TRoPHI and NHS EED, were then assessed for inclusion in the review. Studies were included if they met all of the following criteria:

• published after 1990 (the date cut-off set by other reviews on which we are building, e.g. Popay et al. 4)

• includes primary research, in that data have been collected during that study through interaction with or observation of study participants

• not a Master’s thesis

• includes outcome, economic, and/or process evaluations of interventions

• community engagement is the main approach of the intervention

• for outcome evaluations, study has a control or comparison group (i.e. it must be a controlled trial, either randomised or non-randomised)

• published in English.

Once all of the studies had been screened on their titles and abstracts, full reports were obtained for those that appeared to meet the criteria or for which there was insufficient information to be certain. The retrieved articles were then screened based on the full-text article.

Three additional criteria were applied at full-text screening which ensured that sufficient detail was present for critical appraisal to take place and that the study participants were relevant to our focus on inequalities:

1. study characterises study populations or reports differential impacts in terms related to social determinants of health that can be captured by the PROGRESS-Plus framework (this framework, which enables us to characterise different dimensions of potential disadvantage, is described in Chapter 3)

2. study reports health or health-related outcomes for the effectiveness evaluation and/or process data in a process evaluation

3. study was conducted in an Organisation for Economic Co-operation and Development (OECD) country (at the time of screening OECD member countries were Australia, Austria, Belgium, Canada, Chile, Czech Republic, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Iceland, Ireland, Israel, Italy, Japan, Korea, Luxembourg, Mexico, Netherlands, New Zealand, Norway, Poland, Portugal, Slovak Republic, Slovenia, Spain, Sweden, Switzerland, Turkey, UK and the USA).

Those studied that passed the inclusion criteria on the basis of full-text screening were included in the map of engagement models and contributed to the theoretical synthesis.

Selecting trials for in-depth review

With regard to the extraction of outcome data, we carried out a final sift of the studies to determine which would be included in the in-depth review. Some studies were excluded because, on closer inspection, they did not have an adequate (independent) control group or quantitative data from which we could calculate an effect size estimate for the meta-analysis. We also filtered out studies that did not report at least one of the following outcome types:

• health behaviours (e.g. smoking, food intake, physical activity)

• community outcomes (e.g. perceived increased access to services in the area)

• engagee outcomes (e.g. skills acquired by the engagees whilst engaged in the initiative).

Time constraints meant that we had to further narrow the scope of the studies for meta-analysis – it was simply impossible to extract outcome and risk of bias data for all 319 studies included in the map in the time frame available. It was decided, after consultation with the project advisory group, that a sensible way to identify a manageable set of studies was by focusing on health issues that were emphasised in the Marmot Review28 as being particularly problematic in terms of health inequalities in the UK. To achieve this, we identified any health priority areas in the Marmot Review in ‘health inequalities and the social determinants of health’ (see Chapter 2) and ‘policy objectives and recommendations’ (see Chapter 4) that mapped onto the health topics of the studies included in our review. These are presented in Table 1. Only studies categorised under those health topics were included in the in-depth review.

Data collection (extraction) process

Quality assessment

The outcome evaluations (controlled trials) were assessed for methodological quality using an adaptation of the Cochrane risk of bias assessment tool. 54 We examined the studies in a range of dimensions including methods of assignment, the comparison group type, the comparability of groups at baseline/methods of adjustment, attrition and selective reporting. In the meta-analysis, we tested to see whether effect size was associated with methodological quality.

The tool we used to assess the quality of the process evaluations was refined in a recent review55 and assesses whether or not steps were taken to minimise bias and error/increase rigour in sampling, data collection and data analysis; findings were grounded in/supported by the data; there was good breadth and/or depth achieved in the findings; and the perspectives of intervention participants were privileged. The findings from process evaluations that did not score well were included but a sensitivity analysis was conducted to assess their impact on the overall analysis, as findings that depend solely on the evidence of poorer quality process evaluations are more provisional than those coming from stronger evaluations.

We had planned to assess the quality of economic evaluation studies using the Consensus on Health Economic Criteria (CHEC) list. 56 In the event, no such evaluations were identified.

Theoretical synthesis (see Chapters 4 and 8)

The theoretical synthesis was the first analysis to be completed. This analysis is similar in some respects to Pawson’s57 work on realist synthesis and examines in particular the theories, mechanisms and contexts of community engagement. It does not, however, attempt to engage in causal reasoning, leaving this task to the meta-analysis. The theoretical synthesis has been split into two parts in the report: Chapter 4 (answering RQ1 and RQ2), which presents the range of models of community engagement that have been presented elsewhere and focuses in particular on the theories of change that underpin each model; and Chapter 8, which presents a broad conceptual framework that encapsulates the studies in the statistical and economic analyses as well as the models of engagement in Chapter 4.

The first synthesis in this review aimed to understand the range of models and approaches underpinning community engagement, and the mechanisms and contexts through which communities are engaged. These aims are a way to gain a conceptual understanding of community engagement, rather than a more aggregative systematic review approach of counting the number of studies representing various models, approaches, mechanisms and contexts. 58,59

The theory-building nature of these aims led the research team to use methods of study identification and synthesis more appropriate for conceptual analysis research synthesis. In this type of synthesis, searching aims to build an understanding of a particular phenomenon by gathering a number of articles that present different perspectives on that phenomenon. Once a sufficient range of ideas have been identified, studies that do not add anything new to the topic are put to one side; in effect, a saturation of perspectives has been reached. 60

Using our data set of included studies along with theoretical literature that we identified alongside them, we adopted a purposive search and inclusion strategy more appropriate to gathering concepts, rather than the more traditional approach of exhaustively accumulating all literature on the topic. 61 A set of included studies was created that had been identified by team members as being examples of each of the different community engagement strategies. These studies were then synthesised in two ways.

The team began with the conceptual framework for community engagement (see Figure 1). First, one reviewer read, summarised and extracted data on community engagement aspects from each study and then compared those findings to the team’s conceptual framework to see whether issues in each paper refuted, confirmed or added new information to the model. The conceptual framework was developed as new issues were discovered. This ‘rolling’ or ‘constant comparative’ method of synthesis has been used in previous EPPI-Centre reviews. 62–64 The emerging conceptual framework and the summaries were read and discussed by the research team, to ensure that the framework reflected the combined expertise and individual perspectives of the whole team. A draft of the conceptual framework was presented to the review’s advisory group and the final framework was revised based on their feedback.

Second, all papers identified before in-depth coding in the map (n = 559) were clustered using the Lingo3G text mining algorithm (Carrot Search s.c., Poznan, Poland);65 these clusters were then manually organised according to the area of health concern, type of community engagement, participants’ PROGRESS-Plus characteristics and country/geographical details. This information was then also used to refine the framework. Each study was then manually coded; each time data were extracted from a study, its mechanisms and contexts were compared with our conceptual framework and the framework checked for its adequacy. After the meta-analysis and economic analysis were completed, the framework was again revised to take account of their findings.

Meta-analysis

The meta-analysis (quantitative synthesis) uses various statistical methods to address RQ3–8, by testing whether any observed differences in the results of included studies might be associated with the type of community engagement they employed. This is reported in Chapter 5. Methods used are descriptive statistics, meta-analysis (homogeneity tests), analysis of variance (ANOVA) and meta-regression. 66 For the random-effects model analyses we followed the methods described in Lipsey and Wilson52 and used SPSS (SPSS Inc., Chicago, IL, USA) macros written by Wilson. 67 The risk of bias across studies was assessed as described in Quality assessment.

Analyses were conducted separately for the following outcome types:

• health behaviours at immediate post-test (e.g. fruit and vegetable intake)

• health behaviours at delayed follow-up

• health consequences at immediate post-test (e.g. cholesterol levels)

• participant self-efficacy related to the health behaviours, at immediate post-test

• social support related to the health outcomes, at immediate post-test

• engagee outcomes (e.g. skills, empowerment)

• community outcomes (e.g. perceived improved access to health services in the local area).

Possible moderating or confounding factors included:

• the community engagement theory of change

• whether it was a single or multicomponent intervention, and whether community engagement was evident in all components

• the Marmot Review28 priority health area

• characteristics of the intervention (setting, strategy, deliverer, duration)

• characteristics of the participants (age, PROGRESS-Plus health inequality group)

• whether the intervention was targeted or universal

• the potential for risk of bias and characteristics of the evaluation.

Moderators and confounders notwithstanding, we aimed to identify the amount of variance (if any) that is explained by different approaches to community engagement with participants within each review, each topic domain and finally across all studies in the meta-analysis. Specific aspects of the analysis, such as data cleaning (identification and treatment of outliers and skewed data), sensitivity analyses and assessment of publication bias, are reported in the results chapter (see Chapter 5) alongside the relevant results to facilitate understanding of the findings.

We chose to focus our reporting of the results on the trends in pooled effect size estimates, rather than between-group statistical significance – as is usually common in meta-analysis. Typical meta-analyses attempt to infer findings from the sample to a hypothetical population. This is problematic for our review because the issues that we are exploring – community engagement and health inequalities – are so broad and difficult to define that it is impossible to know exactly to what population the results of any inferential statistics would apply. Instead, we emphasise observed trends, to help disentangle some of the differences between the types of evidence we have collected. This can help us to understand what might occur in other similar studies not included in the review, but not in any one specific situation because (and as discussed in Chapter 8) the causal pathways are complex and potentially unique to each study.

Thematic synthesis of process evaluations

The thematic synthesis of process evaluations narratively described emerging themes and factors evident across the process evaluations. Our original plan was to conduct a framework synthesis68,69 of process evaluations using a tool that we had constructed based on an earlier synthesis of community engagement process evaluations. 4 We found, however, that, although the framework ‘worked’ for some aspects of the process evaluations that we were synthesising, it did not cover the range of issues that we were encountering. We therefore moved to a more open structure and applied a process evaluation data extraction tool developed and used in previous EPPI-Centre reviews. 55,70 The complete process evaluation data extraction tool is provided in Appendix 4.

The data extraction tool was used to assess 12 criteria, including data collection method (e.g. interviews, surveys); type of stakeholder who provided the process information; the timing of the process evaluation in relation to the intervention; methods and rigour of sampling, data collection and analysis; assessment of how grounded the data were in authors’ findings; assessment of the breadth and depth of findings; and extent to which the process evaluation privileged the perspectives and experiences of the public. An overall rating (low, medium or high) was given to each study in terms of the study’s methods and the usefulness of each study’s findings in drawing conclusions about what works, why and for whom.

Specific content about processes was coded using the following headings:

• acceptability of the intervention to the participants or providers

• accessibility/programme reach

• consultation/collaboration/partnerships

• programme content (e.g. use of incentives, fit between content and aims of intervention)

• costs (e.g. issues of sustainability)

• implementation (e.g. frequency, duration and amount of adherence to programme content)

• management/responsibility issues

• quality of programme materials

• skills and training of intervention providers

• other issues.

Within each of the process evaluation data extraction questions, responses from each study were summed and frequencies reported. Ratings of overall reliability and usefulness were determined by reviewing responses to each data extraction question. Findings from each data extraction category were assessed and compared to determine whether:

• lower rigour/lower usefulness studies came to similar or different conclusions to/from higher rigour/higher usefulness studies

• findings varied by publication date

• findings varied by health topic

• findings varied by the type of community engagement model used.

Economic analysis

The final component of our study, the economic analysis, answers RQ8 and RQ9 and investigates the resource implications of various approaches to community engagement. It also reports on the extent to which they have been evaluated in terms of their potential cost-effectiveness.

The data extracted from the studies were synthesised narratively. Two tools were developed specifically for this project to capture data on economic issues (e.g. sufficiency of funds) and resource utilisation, cost and cost–consequences (e.g. staff costs). These tools were then combined into one data extraction tool (see Appendix 5). Items covered the following domains:

• resourcing and cost breakdown

• economic consequences of interventions

• economic evaluation methods and findings

• availability or flow of funds

• sufficiency of funds

• securing additional funds

• financial sustainability

• linking investment in intervention with impact on outcomes

• sources of funds

• role of volunteers

• financial and economic incentives

• other issues.

The synthesis started with a consideration of trends across the studies in terms of costs, resource use, cost–consequences and economic evaluation methods used. We then explored the findings from individual economic evaluations. Finally, specific issues that we examined in more detail were the value and role of volunteering, the use of financial and other incentive mechanisms to influence rate of community engagement, gains or losses to human and social capital, and funding and sustainability issues.

We also aimed to make use of information extracted from our review of the literature and supplemental information from other studies and community engagement implementers to undertake some limited decision modelling to explore the potential cost–utility and cost-effectiveness of selected different approaches to community engagement.

Country

Table 2 details the countries in which the studies were carried out (note that the map was restricted to studies carried out in OECD countries as in the inclusion criteria). As is often the case in systematic reviews of public health research, significantly more studies are carried out in the USA than in any other country, although a reasonable number do hail from the UK.

Age groups

Table 3 shows the various age groups targeted by the interventions in the included studies. Proportionately, young people receive the most attention, although the interventions are targeted at all groups.

Sex of participants

Figure 5 depicts the relative numbers of interventions that were targeted at one sex only or at both males and females. The majority of participants were male in < 9% of the studies (n = 2893,157,163,214,227,237,246,260,270,291,292,294,304,305,326,344,361,365,372,383,385,386,394,414,417,428,429,431), whereas the majority were female in 30% of the studies (n = 11072–74,77–79,86,94,95,97,98,102,109,111,120,121,124–126,128–130,132,133,138,140,144,146,147,149–151,155,160–162,168,171,175,177,178,181,183,184,186,187,191,196,198,199,203,205–207,215,223,225,235,239,241,245,249–251,254,261,262,264,265,267,269,272,279,281,283,290,296,297,301,303,308,309,313,318,322,324,334,336,338,340,342,348,350,351,366,373,376,382,387,399,401,402,404,411,413,415,421,424,426,430). Just over half (57%) of the studies included mixed groups of males and females. 71,75,76,80–85,87–92,96,99–101,103–108,110,112–119,122,123,127,131,132,134–137,139,141–143,145,152–154,156,158,159,164–167,169,170,172–174,176,179,180,182,185,188–190,192–195,197,200–202,204,209,210,212,216–220,222,224,226,228,229,232–234,236,238,240,242,243,247,252,258,273–278,284–286,288,289,293,295,298,299,306,307,311,315–317,319,320,323,325,327,328,330,331,335–337,339,341,343,345,346,349,352,353,355,356,358–360,363,364,369–371,374,377–379,384,388,395–398,400,403,405–410,418,420,422,423,425,427

FIGURE 5.

Sex of participants in the studies (n = 319 studies).

Main PROGRESS-Plus characteristics

In framing the scope of this review we needed to identify, describe and categorise the range of social determinants that, either separately or in combination, influence the health of participants. 432 Given the way that different factors interact in different circumstances, socioeconomic status alone cannot be used to identify potentially vulnerable groups as ‘a person’s socio-economic position is shaped by their position in other structures of inequality’ (p.5). 433 As in previous reviews, socioeconomic position has been captured in terms of (1) broader determinants of health, using the mnemonic of PROGRESS (for Place of residence, Race/ethnicity, Occupation, Gender, Religion, Education, Socioeconomic position and Social capital) developed by Evans and Brown;434 (2) other characteristics that attract discrimination, such as age, disability and sexual orientation, drawing on Krieger’s work;435 and (3) features of relationships such as social exclusion within family or community contexts (e.g. smoking parents, being ‘looked after’ or excluded from school) or time-dependent circumstances (e.g. recovery from illness, or other life transitions) relevant to particular research. Together, these three ways of conceptualising the social determinants of health have been brought together in the PROGRESS-Plus tool. 436

Table 4 contains a breakdown of the main PROGRESS-Plus characteristics of the sample/study population. By some distance, the most frequent category is ethnicity, with 120 studies,71,78–82,85,88,94,96,97,99,103,105–110,112–114,126,127,131,132,140,145,146,150,151,157,161,164–166,173,174,180,182,190,193,194,196,197–199,201,205,210,215,223,224,227,232,236,239,241,243,246,247,249–251,254,258,262,264,265,267,269,275–277,279,283,284,291,292,294,298,299,308,309,311,312,317,318,322,325–327,330,335–337,342,343,348,351,353,359,363,365,366,373,382,386,387,395,400,404,406,409,411,413,422,425,426,428 109 of which were carried out in the USA. The most frequent category of disadvantage in the UK is socioeconomic position (eight studies100,118,149,159,217,240,286,339),followed by ethnicity (four studies157,284,308,355), the reverse of the pattern in the USA. The two studies185,403 in the USA falling into the occupation/employment status category all fell into the worker/employee subcategory rather than blue-collar/manual labourer, with one other study139 focusing on blue-collar workers in Sweden. The three studies338,350,376 in which gender was a cause of potential disadvantage were all focused on females.

The ‘other’ category in the ‘other vulnerable groups’ included at-risk families of newborns, multiethnic, underserved women, parents of children with chronic illnesses, parents of children with juvenile rheumatoid arthritis, pregnant and post-partum women at risk for child placement, pregnant women at risk from partner violence, vulnerable families, low-income mothers and children, phenylketonuric women, migrant farmworkers and male sex workers. These groups were often vulnerable to compounding biological or social influences simultaneously.

Table 5 adds a little more nuance to the picture. It shows, particularly in the US studies, that participants often suffered from more than one type of disadvantage and that socioeconomic position, education and place of residence join ethnicity as particularly significant types of disadvantage.

Addressing health inequalities

In the context of inequalities in health, the manner in which an intervention is delivered is significant. As discussed above, there are multiple and complex determinants of health in different groups, and interventions may therefore need to be carefully targeted; universal programmes risk affecting the population differentially and possibly therefore increase inequalities by benefiting relatively healthy groups disproportionately. Table 6 summarises the approaches taken in the interventions in the map, showing that the majority of interventions are indeed targeted or delivered to a specific PROGRESS-Plus group (237 studies71–75,77–84,86–89,92–96,98–105,107,109,111–113,117,119–126,128–133,137,138,140–142,144–153,155–157,159–172,174–184,186,187,191,193,194,196–199,201,203,205,207,210,214,216,218,219,222–225,227–229,233,237,239,240,242,243,245–247,249,250,254,258,260–262,264–267,270,272–279,281,283–286,288,290–293,296,297,299,301,303–306,308,309,311–313,318–320,322,324,327,328,330,335–338,340–343,345,348,350–354,356,358,359,361,363,366,370,372–374,376–378,382–386,388,394–396,398–404,407–409,413,415,417,418,422–425,427,429–431), rather than being delivered universally (48 studies90,91,108,110,114–116,118,127,134,136,143,154,188,190,192,195,200,204,212,217,220,226,232,238,259,269,289,295,307,314–317,325,326,331,334,339,346,349,364,379,405,410,420,421,426,428). Intervention evaluations could assess the impact on health inequalities in two ways: either by directly targeting a specific PROGRESS-Plus group, for example a minority ethnic group, or by looking at PROGRESS-Plus characteristics within those targeted groups (e.g. assessing differential impact by gender within a minority ethnic group).

Table 7 expands the above statistics to include the health topics addressed by each method. It shows the wide range of topics addressed by community engagement interventions and also where some of the clearest foci of activity are – and where there may be gaps in our knowledge. For example, within the smoking cessation and prevention literature, it is striking that cessation research is almost exclusively targeted at a PROGRESS-Plus group, but that smoking/tobacco prevention research is universal in nature. A health promotion perspective might suggest that more targeted work on prevention is needed to reduce the need for targeted cessation activity.

Table 8 shows where each of the interventions in the 319 studies was delivered. Some of these sites, such as residential care and the WIC (Women, Infants and Children) clinic, suit targeted interventions; others either suit universal interventions or need additional measures to distinguish disadvantaged groups, such as measures of workforce, school enrolment or neighbourhood populations. (The WIC scheme provides federal grants to US states for supplemental foods, health-care referrals and nutrition education for low-income pregnant, breastfeeding and non-breastfeeding post-partum women and to infants and children up to age 5 years who are found to be at nutritional risk. A WIC clinic is a health-care clinic available under this scheme.)

Table 9 details the person or people who delivered the public health intervention.

The final dimension of interventions captured in the map is intervention type (Table 10). By far the most common type of intervention involved education carried out by either a peer or a community member (10172,73,77,87,93,95,98,100,102,103,118,120,125,133,147,149,153,158,161–164,166,172,176,177,179,181,182,188,192,200,202,209,212,214,219,220,222,233,241,245,247,249,264,270,273,274,284,293,296–299,304,305,308,309,312–318,324,330,337,341,342,345,348,352,354,358,360,361,364,369,373,376,378,379,387,394,396–398,400,401,407,409,410,413,415,422–424,426,429,431 and 11175,79–81,83–85,90–92,94,96,101,103–106,109,112,113,115,117,118,122–124,128,131,132,134,136,137,142,144,145,147,148,150,152,154,158,165,172–174,177,179,182,190,193–195,197–199,201,204,206,207,209,210,212,217,218,222,235,238,243,251,254,262,265,266,272,274,276–279,283,285,288,290–292,298,301,315,316,319,320,322,323,325–327,331,336,338,346,350,354,364,366,377,385,406,407,417,424,430 studies respectively). A cross-tabulation of intervention type by provider is provided in Table 11.

In Table 12 we classified the studies according to the policy priorities identified in the Marmot Review,28 as described in Chapter 2. The final table in the map (Table 13) cross-tabulates topic against the main PROGRESS-Plus group in the study. It gives a good indication regarding the distribution of health issue and population group in the map.

Defining ‘community’

The commissioning brief for this project defined community engagement as ‘approaches to involve communities in decisions that affect them’. Mason et al. 40 have defined community engagement for health promotion as engaging groups of people who share geographies, interests or identities; these are the ‘groups with distinct health needs and/or demonstrable health inequalities’ of the commissioning brief. However, there are many different ways of understanding what a community is, and the boundaries between where communities stop and individuals begin can be blurred; other work has thus reviewed both patient and community engagement,438 as the rationale for involving both patients and communities in decisions that affect them can be similar. Indeed, it is possible for the terms ‘patient’, ‘public’ and ‘community’ to be used interchangeably in some contexts, and definitions of what the defining characteristics of a community are vary considerably. 438 Brenner and Manice437 stated: ‘community is a fluid concept; individuals may belong to multiple communities at any one time’ (p. 88). In 1984, Wandersman suggested that there are two broad conceptualisations of community, one being geographical in nature (‘a place’) and the other concerned with ‘a set of relationships and resources’ (cited in Kelly,439 p. 417). This definition is particularly useful – and visionary for its time – when one needs to include virtual communities within an overall understanding of community. Building on these ideas, we have found the following definition helpful in this review: ‘a group of people united by at least one but perhaps more than one common characteristic, including geography, ethnicity, shared interests, values, experience or traditions’ (p. 88). 437

This definition comes closest to encapsulating the range of communities that we have encountered in this review, as a ‘community’ may be no more than a group of people who share a similar characteristic, or it may be a highly organised and integrated group of people pursuing a shared goal.

The above conceptualisation of a community is very much a minimalist viewpoint; it does not require the community to function in any coherent way (as it may have been defined from outside, e.g. by health professionals). More demanding concepts of community have been summarised by the Clinical and Translational Science Awards (CTSA) Community Engagement Key Function Committee Task Force440 and include a systems perspective, which conceptualises different constituents such as individuals, schools and businesses as acting together, like a ‘living creature’; a social perspective, which emphasises the role that social networks play in the operationalisation of the community, and thus the importance of understanding where the network (i.e. community) needs strengthening; a virtual perspective, which focuses on online communities; and the individual perspective, which emphasises individuals’ sense of their community memberships and acknowledges that this may be at odds with researchers’ perspectives, and also the fact that an individual may belong to multiple communities.

Although we have adopted an open definition of community, it is important to note that most community engagement initiatives in this review do involve communities that function (to varying degrees) as integrated systems and that understanding the interplay of the different actors they contain may contribute to developing an appreciation of how they can be mobilised to improve people’s health. Having said this, those in the communities defined by those outside them may well self-identify themselves in quite different ways. The social perspective mentioned above is useful in this regard. Rather than conceptualising communities as being made up of people with shared attributes (e.g. healthy, high income, female), social network theory offers an alternative view which emphasises the relationships and ties that individuals have within their various communities. Some authors suggest that a resource stemming from these relationships is social capital. Although conceived differently by different scholars (see, for example, references 441–443), the general theory suggests that social networks characterised by reciprocity, trust and shared norms can facilitate the achievement of specific shared goals. 444 Social networks, however, can be either health promoting or harming depending on the nature of shared norms. Strong social ties that benefit members of one particular group may also be used to exclude others outside the group from sharing those resources (p. 15). 444 Although this leaves less room for individual agency, it can be a useful way of visualising community structure and identifying the different roles that people within a community might play.

The rationale for examining community engagement in the context of health inequalities needs some consideration at this point (see Chapter 1, The challenge of health inequalities in the UK, for a discussion of health inequalities). Although the use of community engagement to improve the health of disadvantaged groups has been recommended, as they tend to be more disenfranchised than other groups and lack power and control over their lives (e.g. references 15 and 38), individuals may be less able to take advantage of traditional health promotion messages because of this lack of control. Health improvement may thus be less successful in these areas and any universal community engagement intervention (covering both relatively advantaged and disadvantaged groups) risks increasing health inequalities. However, it may be that engaging communities in improving their health achieves the joint objectives of both improving health and making good some of the social gradient in ‘healthy’ community characteristics identified elsewhere. 28 A conceptual framework for reducing inequalities in health through community engagement thus needs to consider some of the challenges to engagement that may exist, along with some possible mechanisms for achieving participation where engagement may be relatively lacking. (We have given here functional definitions of ‘community’ that were operationalisable in this review. The term is of course contested on many levels; see, for example, Finlayson445 and Hillery446 and a recent report by Hamalainen and Jones447 for a critical analysis of the concept that is particularly useful for those working in community engagement.)

Thus, although we recognise that some of the reasons for engaging communities in decisions that affect them may overlap with patients’ involvement in decision-making, in the context of improving inequalities in health we have made an explicit differentiation between the two. The ‘engagement’ (which will be discussed next) needs to be with groups (communities) of individuals and, rather than being concerned with individual’s decision-making, we are concerned with collective decision-making on the part of service users. This has led us to including potential benefits and harms for communities in the range of outcomes that are considered – as well as individuals’ health.

Defining ‘engagement’

‘Engagement’ is a second contested concept that is often associated with terms such as ‘community development’, ‘participation’, ‘empowerment’, ‘competence’, ‘capacity’, ‘cohesiveness’ and ‘social capital’. 448,449 To locate this work within a developing corpus of literature, which is building our understanding of community engagement in the UK, we have employed the same definition as used in an earlier report that laid out a conceptual framework for community engagement. ‘Community engagement’ is therefore ‘an umbrella term encompassing a continuum of approaches to engaging communities of place and/or interest in activities aimed at improving population health and/or reducing health inequalities’ (p. 2). 14

Such a broad definition covers activities that range greatly in scope and impact, from the relatively unengaged exchange of information, through various models of citizen consultation to the delegation of power and control to community organisations, and therefore is particularly useful for investigating a diverse literature.

Other terms overlap conceptually but do not capture the issue of community engagement. These include ‘community participation’, ‘consultation’, ‘information’, ‘development’ and ‘empowerment’. It is important to note that ‘community participation’ and related terms (e.g. ‘community-based participatory research’) sometimes imply both a specific set of research methods or approaches and research that builds relationships with the goals of societal transformation/improved health. Thus, in academic circles, terms can have the joint meaning of being concerned with methods of service design and evaluation as well as being an end in themselves. The term ‘community participation’ is often used interchangeably with ‘engagement’ and, as WHO observes, the quantity of terms and the lack of precision with which they are employed can cause confusion. WHO450 defines ‘community participation’ as:

A process by which people are enabled to become actively and genuinely involved in defining the issues of concern to them, in making decisions about factors that affect their lives, in formulating and implementing policies, in planning, developing and delivering services and in taking action to achieve change.

p. 10

The above definition is considerably narrower than our definition of engagement, and demands a high degree of community involvement – and influence – over local activities. ‘Community engagement’ is, however, a difficult term to define, as it:

Is a rich concept that varies with its application and definition. The way that participation is defined also depends on the context in which it occurs. For some, it is a matter of principle; for others, practice; for still others, an end in itself.

p. 7451

Thus, ‘participation’ does not capture the idea that communities can participate to differing degrees.

Unlike the above, the term ‘community development’ has a clearer definition and a long history of use. It is defined by WHO450 as:

A way of working underpinned by a commitment to equity, social justice and participation that enables people to strengthen networks and to identify common concerns and supports people in taking action related to the networks. It respects community-defined priorities, recognizes community assets as well as problems, gives priority to capacity-building and is a key mechanism for enabling effective community participation and empowerment.

p. 11

Community consultation and information-sharing are perhaps more straightforward concepts. Community consultation suggests that people participate by being consulted or by answering questions. Agents who are external to the community lead in defining the problem and articulating solutions. Such a consultative process does not concede any sharing of decision-making and professionals are typically not obligated to take on board people’s views. 452 Participation through information-sharing on the other hand involves the flow of information between community member(s) and agents. Although often considered a less active form of participation, Cornwall453 argues that ‘transparency over certain kinds of information opens up the possibility of collective action in monitoring the consistency of rhetoric with practice’ (p. 272).

Possibly the most demanding form of community engagement, in terms of the delegation of power and control, is often termed ‘community empowerment’. This concept is discussed at length below and for the purposes of this section we use a definition from Empowerment and Poverty Reduction published by the World Bank:454 ‘Empowerment is the expansion of assets and capabilities of poor people to participate in, negotiate with, influence, control, and hold accountable institutions that affect their lives’ (p. vi).

Empowerment is often portrayed as the ‘ultimate’ form of engagement, as it requires the ceding of power and control to people who traditionally lack such power and control. Feminist perspectives on power and empowerment direct us towards the complexity of the relationship between engagement and empowerment. For example, Yoder and Kahn455 advance a gender-based conceptualisation of power, contrasting a female aim to seek ‘power to’ (i.e. personal empowerment) with a male aim of seeking ‘power over’ (i.e. domination and control). Others have argued that ‘empowerment is a process that changes social and political relationships between individuals and the communities and networks within which social and political power circulate’ (p. 45). 456 Therefore, when engaging communities the directions of these changes are important and considerations of inequalities in gender and ethnicity ought to be considered.

The interaction of these concepts suggest a hierarchical relationship in which each concept is a natural progression from the last. Participation is a clear prerequisite of empowerment and, as community development is a ‘key mechanism’ for both participation and empowerment, it follows that participation and empowerment are subsets of community development. In our ontology of engagement, ‘engagement’ is an umbrella term that encompasses all approaches and activities below it: community development, empowerment, information, consultation and participation.

A visual representation of these related concepts is given in Figure 6, showing that all terms are subsets of the umbrella term ‘engagement’ and that each approach is a necessary condition for its immediate subset.

FIGURE 6.

Set relationships in concepts around engagement.

Although the above hierarchy will fit many definitions, it is important to recognise that many people have different conceptualisations of engagement and, as this extended quotation from Stuart Hashagen42 shows, would see ‘engagement’ as involving significant partnership and empowerment:

Engagement suggests a different sort of relationship. It suggests that there is a ‘governance’ system and a ‘community’ system. To build the collaborative relationships on which a complex activity such as community planning would depend, it is necessary for the governance system to fully understand the dynamics of the communities with which it seeks to work, and to be prepared to adapt and develop structures and processes to make them accessible and relevant to those communities. In this way, the term engagement warns us against making assumptions about communities: it asks for a dialogue. It also implies that the development of the relationship itself will need to be a focus for attention: ‘government’ will need to engage with communities as well as asking communities to engage with it.

p. 2

As authors, we have discussed whether the term ‘engagement’ encapsulates the studies that we have been reviewing and whether alternative language, such as ‘participation’, ‘empowerment’ or ‘involvement’, might be better. However, as discussed above, although the language used around community engagement lacks precision, it is possible to identify key relationships and show that each has particular characteristics. Because the extent to which the community is ‘involved’, ‘engaged’ or ‘participating’ varies widely between the studies that we have identified, the umbrella term ‘engagement’ includes all of the ways of interacting with communities contained in the studies found in the review.

This section has been a relatively brief definitional overview to identify relationships between some of the key terms found in the studies that we have reviewed and provide a transparent description of the development of our definition. However, there are a wide range of theoretical models that underpin these different approaches. To develop a conceptual framework that encapsulates this diversity, we now review some of the main models of community engagement.

Models based on a utilitarian (pragmatic) or health systems perspective

One of the critical points of departure among those advocating community engagement approaches concerns ‘who’ initiates engagement and therefore who is deciding who the community is/are. Community members may engage with issues about their own or their families’ health on an individual basis, but efforts to address inequalities require a collective effort in which a community builds its own identity and engages public institutions, or members of public institutions reach out to the community that they have defined as relevant to the issue. The community may have pre-existing social relationships; or it may be formed to respond to a particular problem. Whether pre-existing or newly evolving, the community may define an issue or need that they wish to change; or others external to the community may identify the issue that requires change, and subsequently define the relevant community, which may consist of individuals or pre-existing groups (‘engagees’). Individuals or groups may thus also come to identify or imagine themselves as a community through the process of engagement. When members of the community define themselves, they may be communities of interest or populations with specific needs. When others define the community, the community could be a socioeconomically disadvantaged population, or ‘communities’ bound by geography for instance. 459 Thus, communities are defined in many different ways, and people living within those communities can identify themselves differently, depending on the context. 459

In the classical patient/consumer involvement model, rationales for health services to engage service users include:

• involving patients in the decision-making process to choose the most appropriate treatment for complex conditions and improve treatment delivery, patient experiences and patient outcomes

• involving patients because they have a right to be involved in decisions made about them

• sharing responsibility to reduce the scope for blame to be attributed on the health service and to provide opportunities for mistakes to be identified. 438

When broadened out to involve communities, many of the aforementioned rationales remain valid, although the model is still one of health (and other) services reaching out to engage particular (actual or potential) service users who they have identified as requiring assistance. In these situations, members of the community who are being engaged may identify themselves as being members of that community, or they may have no more in common with one another than the ‘one common characteristic’437 mentioned above.

The theory of change that accompanies such programmes might resemble that outlined in Figure 7. In this model, the need for intervention is usually identified by expert observation and the appropriate response devised within existing policy, practice and resource frameworks. The utility of involving stakeholders for the reasons given above is becoming widely accepted and people are often involved to differing degrees to improve acceptability and take-up. The key mechanism for improvement here is that the intervention is more effective than it otherwise would have been because of the incorporation of the perspectives of those targeted; the question to be explored later is therefore whether health outcomes increase in proportion to the amount that stakeholders are able to influence (or in some situations to control) the intervention.

FIGURE 7.

Theory of change for patient/consumer involvement.

An evaluation by Davis et al. 243 illustrates this model of engagement. The Pathways intervention was developed as a culturally appropriate obesity prevention programme for American Indian school children. In a bid to address nationally high rates of obesity amongst American Indian populations, researchers worked with community members and teachers to develop culturally appropriate curricula, family information packs and family events focused on obesity prevention. This intervention is an example of one that is developed out of a need identified by those outside of the community, in which community members ‘buy in’ to the idea of doing something to tackle an issue, but appear to have very little influence on the initiative’s development, management and evaluation.

A related, but different, model of engagement is the use of peers or lay community members to deliver the intervention. This may be combined with the above health services model, but the particular ‘active ingredient’ in this case is that, rather than being delivered by health professionals, etc., the intervention is delivered by peers/lay people and is more likely to be effective because peers/lay people have more credibility, are able to communicate more effectively and persuasively and are able to respond better to situations ‘on the ground’ than people who are less immersed in the community of interest. 460,461 (There is a veritable industry of peer-delivered health promotion – particularly in North America – which is supported by organisations such as the National Association of Peer Program Professionals; www.peerprogramprofessionals.org/.)

The theory of change that accompanies peer-/lay-delivered interventions is shown in Figure 8. This is a deliberately simplified version which emphasises the main difference from other approaches: the medium of delivery. For example, Anderson et al. 72 describe an evaluation of peer counselling on exclusive breastfeeding in a predominantly Latina low-income community in the USA. Local mothers were trained in breastfeeding counselling skills and provided antenatal, in-hospital and postnatal breastfeeding support and education to women. Initiation and exclusive breastfeeding rates were concurrently assessed. In this model, community members have a high degree of involvement in intervention delivery but do not contribute as partners to that intervention’s design or evaluation. Of course, there are many examples of interventions that have been designed in collaboration with the community of interest which are also delivered by peers, and the theory of change in this case would be a combination of the two. As South et al. 462 conclude, the relationships between peer/lay intervention providers and those that they are working with is complex, and the above theory of change treats the complex range of these interactions as a ‘black box’ within which many different processes take place: ‘social relationships are core to understanding lay health worker programmes and therefore analysis needs to take account of the capacity for community members to move within a spectrum of participation defined by increasing responsibility for others’ (p. 656).

FIGURE 8.

Theory of change for peer-/lay-delivered interventions.

Margaret Whitehead432 considerably extends the above model in her typology of actions to tackle social inequalities in health. Here, rather than a health problem being observed, a health inequality is the focus of attention. Whitehead identifies four categories of policies and interventions to tackle health inequalities: strengthening individuals, strengthening communities, improving living and working conditions and promoting healthy macro-policies. Although the model (outlined in Figure 9) is similar to the model above, in that the identification of the problem and (to differing degrees) its solutions are located outside the community, the typology has a strong emphasis on individual and community empowerment. Individual empowerment is usually framed around a deficit model, in which an individual’s lack of knowledge, skills, self-confidence, etc. is addressed by intervention. More recently, there has been an increasing emphasis on capability models, which recognise individuals’ skills and strengths and seek to remove barriers to their application in healthy lifestyle choices. In strengthening communities, the model addresses social isolation, acknowledging that people’s inability to be fully part of society can have serious impacts on their health. Two models here are put forward: horizontal and vertical. Horizontal interventions aim to strengthen connections within communities, the theory being that well-functioning communities will be able to influence their locality to create healthier conditions. Vertical interventions aim to improve connections between different levels on the social scale to improve inclusiveness and participation, the theory here being that ‘fostering solidarity throughout society produces a less divided society, one with smaller social inequalities and hence more equitable access to the resources for health’ (p. 474). 432

FIGURE 9.

The logic of health inequalities interventions.

The above comes mostly from a health services perspective, and is the starting point for many interventions (see below). However, models of community engagement also encompass community-led interventions, in which the community drives engagement and specifies the outcomes of interest and the means by which they are to be achieved. This model, often known as community ‘empowerment’, is outlined by Woodall et al. 463 Empowerment as a concept is discussed below, but here it is important to emphasise the genesis of action. Rather than being located within the health service, or as a result of engagement between the health system and the local community, the need for intervention can be identified within the community. This model often begins by outlining the growth from the concerns of an individual, which progress through increasingly large groups through to partnerships, which then result in social and political action (Figure 10). This is therefore speaking to the same issues that Whitehead432 identified in ‘strengthening communities’, the difference being that it is the communities who have mobilised themselves into action.

FIGURE 10.

Community empowerment as a continuum (p. 10464). Reproduced by permission of SAGE Publications, London, Los Angeles, New Delhi and Singapore from Laverack G. Health promotion practice: power and empowerment. London: Sage; 2004. © Laverack, 2004.

The issue of who identifies the need and who specifies the community goes to the heart of arguments about democracy and accountability that permeate the literature about community engagement (see the following section). They also matter from a theoretical and effectiveness perspective, because many of the arguments advanced in favour of community engagement build on the premise that, if people are ‘signed up’ to the intervention or programme that they are receiving, they are more likely to participate and better outcomes can result (see Chapter 1, The evidence base for community engagement). Understanding how and why people engage in community activity, and why they do not, is therefore a critical component of any coherent conceptual framework, and is also important for this review in terms of being able to identify which models of community engagement are associated with better health outcomes.

Models based on a community empowerment (or ideological) perspective

The previous section has touched on issues of power and empowerment, discussing how some intervention models begin from a health systems perspective, in which people are ‘engaged’ for a range of reasons, whereas in the Whitehead432 taxonomy individuals and communities are empowered to be in a position to impact on their lives and localities. Empowerment is concerned with the movement of control from those who currently are in possession of it to those who currently are not; in this sense it is a zero sum game, with the total amount of power remaining constant. Arnstein’s465 ‘ladder of citizen participation’ (Figure 11) is the oldest, and possibly the most widely known, taxonomy of empowerment. Rooted in a 1960s perspective of citizen involvement in urban planning, it contains eight levels of participation that are grouped into three ranks of citizen control: ‘non-participation’, ‘tokenism’ and ‘citizen power’. It shows how different models of participation are more, or less, empowering than others, beginning with essentially ‘non-participative’ ways in which those with power can reach out to those without, to ‘citizen control’, in which power to direct has been ceded completely. True participation begins only once power is delegated, with the other types of participation being dismissed as ‘tokenism’ and ‘non-participation’.

FIGURE 11.

Arnstein’s465 ‘ladder of citizen participation’. Reproduced by permission of Taylor and Francis Ltd from Arnstein S. A ladder of citizen participation. J Am Inst Plann 1969;35:216– 24. http://dx.doi.org/10.1080/01944366908977225

Many other engagement models draw on the ladder metaphor, beginning with minimal ‘interest’ in information-giving and ending with full citizen control. Hart466 differentiated the types of involvement in this framework further when looking at the community participation of children and youth. Wilcox5 drew on both of these theories in developing a theory of public involvement, but suggested that a further process of participation existed: from initiation, preparation and participation to continuation. Further, Wilcox recommended considering issues of control, power and purpose, understanding practitioners’ and stakeholders’ roles, committing to partnerships, generating community-based ideas and developing community confidence and skills. This conceptualisation of community participation has led to the development of specific toolkits that recommend steps to community participation based on underlying principles of what is important in engaging communities. 42,467,468

It is important to recognise the ethical and political dimension of the ladder; as well as representing ‘effective’ ways to involve the public in public policy (and improve the nation’s health), the top of the ladder represents the most democratic and egalitarian approaches towards public service, whereas the lower rungs are associated with authoritarianism and a lack of accountability.

Woodall et al. 463 are concerned with the utility of community empowerment to impact on health outcomes. They define empowerment ‘in its widest and most radical sense’ as being concerned with ‘combating oppression and injustice’ and being ‘a process by which people work together to increase the control they have over events that influence their lives and health’ (p. 9). They relate this to the vision for health promotion set out in the Ottawa Charter469 and related declarations for people to ‘increase control over, and to improve, their health’ (p. 1); that is, that a fundamental aspect of health promotion is people’s empowerment. However, they perceive a disconnection between the aims of health promotion and how it is investigated and practised. 448 They argue that empowerment has become a ‘buzz-word’ that is losing its connection with ‘its original roots as a radical social movement’ (p. 742). Citing Laverack and Wallerstein,470 they argue that a lack of clarity regarding the definition of the term ‘community empowerment’ has resulted in its dilution and its conflation with other terminology such as ‘community competence’, ‘capacity’, ‘cohesiveness’ and ‘social capital’ (p. 744). 448 Although these concepts have their place and their uses, the authors’ concern is that community empowerment is following wider moves in health promotion more broadly ‘toward a reductionist individualistic enterprise focused largely on behaviour change at an individual level, rather than a discipline that focuses on addressing social justice and wider power structures through social and structural change’ (p. 743). 448 Such a view of community empowerment thus fails to appropriately consider existing structural inequalities of economic, social or cultural power that may mediate levels of engagement. According to Laverack and Wallerstein470 then, empowerment is losing its power. Their argument about the term ‘empowerment’ is part of a wider discussion on the past, present and future of health services, in which community participation is seen as being marginalised in the undoubted progress that has been made in other areas of primary health care. 471 Although, to some extent, we have separated conceptually models that are centred on empowerment and those that emphasise other concepts, such as participation, it is clear that the language of empowerment is utilised in different ways and that not all of them adhere to an imperative to ‘combat oppression and injustice’, and argue instead that empowering communities is a sensible way to improve people’s health (a utilitarian perspective – or argument).

The reason that community empowerment is such a key issue is because it is concerned with the wider social determinants of health (and is therefore of critical importance in a review about inequalities in health) and has quite different theoretical foundations from more individualistic approaches. Although individual empowerment can result in individual improvements in health, this can be attained without engagement in any community activities. Interventions that facilitate individual empowerment leave the power structures that gave rise to health inequalities untouched. Interventions that empower communities are to be valued not (only) because they combat oppression, but because they change some of the fundamental dynamics of power that have led to those inequalities in health arising in the first place. The mechanisms of change that underpin these two approaches are quite different, and, although individual empowerment is not unimportant, if empowerment is located only within individuals ‘it overlooks change in the political and social context in which people live’ (p. 743). 448 In contrast, the rationale for empowering communities rather than (or as well as) individuals is that we might expect to see the capacity for communities to improve their own health increased; thus, engagement is to be conceptualised as an outcome in its own right for some interventions, as well as a mediator for improved health across many others.

An example of empowerment is provided in Andrews et al. 73,208 This study used community-based participatory research to develop a culturally sensitive smoking cessation intervention with African American women resident in low-income housing. The community was engaged throughout the project, from identifying a need to tackle tobacco use and approaching researchers for help, to collaborating by way of resource provision; neighbourhood forums to assist with data interpretation and dissemination; an advisory board of residents (60%), housing authority members, pastor, school official, health department, community clinic staff and community agencies; intervention planning and delivery; use of community health workers; and contributions to analysis and interpretation of the pilot study. Among the studies included in our review, almost all community engagement occurred in the interventions’ design or delivery, rather than in its evaluation. Andrews et al. 73,208 is a rare example that we located which also described community engagement in the evaluation of the intervention as well.

Bridging the utilitarian and social justice rationales for empowerment

Notwithstanding the concerns expressed by Woodall et al. 463 about the dilution of empowerment, this chapter will conclude with a new conceptual framework for community engagement that aims to encompass the breadth of perspectives and approaches so far discussed. This will not be the first such framework, however, and it owes much to the work of Jennie Popay in her earlier briefing paper for NICE14 and other consolidating work by WHO (e.g. reference 89).

Popay et al. 4,14,21,472 developed a framework for community involvement in public health, further developing the idea of levels of engagement but within a health services framework. They found that there is considerable complexity in the approaches to community engagement, in the levels of involvement and whether engagement occurs as a reaction to outside pressures or as an internally derived initiative.

Figure 12 outlines the first of two frameworks from this work, which is concerned with understanding different types of community engagement. This framework conceptualises the most powerful type of community engagement as being associated with control being devolved to the community; and the least powerful as lacking in empowerment and being concerned with information provision and exchange. Lying in between the two are consultation and the co-production of knowledge. Although the foundations underpinning each type of engagement lie on this continuum, they cannot be associated with the precise methods used (either singly or in combination), which include ‘citizens panels and juries, rapid appraisal techniques, neighbourhood committees, community forums, participatory evaluation/research and community champions’ (p. 6). 14

FIGURE 12.

A conceptual framework for community engagement.

Conceptually, Figure 12 is similar to the community engagement continuum described by the CTSA Community Engagement Key Function Committee Task Force440 and the ‘wheel of participation’ (Figure 13; this figure is outlined in the 2002 WHO paper Community Participation in Local Health and Sustainable Development,450 which is itself developed on the basis of Davidson’s 1998 paper entitled ‘Spinning the Wheel of Empowerment’473). The wheel denotes four cornerstones of participation – information, consultation, participation and empowerment – each with hierarchical levels and examples. Popay et al. 4,14,21,472 observe that it is not possible – or indeed appropriate – always to aim for the ‘top’ of the ladder of empowerment (e.g. the full citizens’ control shown in Figure 11) and that other ‘rungs’ or ‘sectors’ may contain the most suitable approach.

Both Popay and WHO use utilitarian arguments in support of community engagement (or participation), rather than highlighting the social justice and democracy angle as illustrated earlier by Arnstein465 and Woodall et al. 463 WHO makes the point that citizens and communities may argue that they have a right to have a say, but the justification for encouraging participation is made in terms of it being an effective method of increasing democracy, combating social exclusion, enhancing people’s well-being and quality of life through empowerment, mobilising resources, developing holistic and integrated approaches, achieving better decisions and more effective services and ensuring the ownership and sustainability of programmes. The context of the briefing paper by Popay14 was to inform the development of public health programme guidance, so the rationale for engagement is clearly located within a need to improve health and reduce health inequalities.

FIGURE 13.

The wheel of participation (adapted from Davidson473 by WHO450).

The final model reviewed here is the ‘pathways from community engagement to health improvement’ diagram in the Popay report (p. 9). 14 This builds on the previous models in that it has the familiar ordinal scale that leads from basic information provision through to community control, but it moves on from this to incorporate a theory of change, and suggests that certain models of community engagement are limited in terms of the extent to which they can impact social and health outcomes. Shown in Figure 14, this model is read from the bottom up, with community participation, empowerment and control increasing until the top level corresponds to full ‘community control’. The theory of change begins with changes to services as a result of the community engagement intervention (‘service outcomes’), through intermediate outcomes (social capital and increased empowerment and improved environment) to final impact on health outcomes. Critically, the model rules out the possibility of the lower levels of community engagement – ‘informing’ and ‘consultation’ – impacting on the intermediate social outcomes, and therefore they are capable of only relatively small improvements in health outcomes. As community participation, empowerment and control increase, so do impacts on the outcomes, with the largest improvements being associated with full community control. Popay14 notes that willing communities and public organisations and workers willing to share power and influence with communities are important (p. 12). She also noted that there are direct benefits and ‘dis-benefits’ to community members who participate;4 and that a lack of training, timing and administrative issues were barriers to successful community engagement. Power misuse by professionals and cultural stereotypes of community members by professionals were also noted to be barriers.

FIGURE 14.

Pathways from community engagement to health improvement.

Examples of intervention strategies in line with the above framework include those described in the National Evaluation of Local Strategic Partnerships undertaken through the UK Office of the Deputy Prime Minister. 356 Intended to join up services and community organisations in socioeconomically deprived areas of the UK, multiple locally-developed strategies were developed through partnerships between public sector, private, business, community and voluntary stakeholders. In these types of initiatives, the need for intervention is determined by outside stakeholders; responsibility and authority for addressing the issues are handed to the community members and service providers in that locale.

The studies

Of the 131 studies included in the meta-analysis,71–201 five (3.8%) were conducted in the UK,100,119,149,157,159 113 (86.3%) were conducted in the USA,72–99,101,102,104–106,108–118,120,121,123–128,130–133,135,137,138,140–142,144,146–148,150–155,158,160–163,165–181,183–201 five (3.8%) were conducted in Canada71,103,107,145,156 and eight (6.1%) were conducted in other OECD countries. 122,129,134,136,139,143,164,182 In terms of publication date, 63 (48.1%) were published in the 1990s,74,76,78,82,83,88,90,93,94,97,99,102–106,110,111,113,115,118,120,121,124,125,129,130,133,135–137,139–142,144,145,147–151,153,156,157,158,160,166,169–171,174,176,178,179,181,182,186,190–193,198 62 (47.3%) in the 2000s71–73,75,77,81,84–86,89,91,92,95,96,98,100,101,107–109,114,116,117,119,122,123,127,128,132,134,138,143,146,152,154,155,159,161–165,167,168,172,173,175,177,180,183–185,187–189,194–197,199–201 and six (4.6%) in 2010 or later. 79,80,87,112,126,131

The participants

In terms of health inequalities, most studies were classified as being primarily targeted at, or delivered to, ethnic minority groups (n = 56,71,77–82,85,86,88,94,96,97,99,103–110,112–114,126,127,131,132,140,145,146,150,151,155,157,161,164,165,173,174,176,180–182,188,190,191,193–199,201 42.7%), followed by socioeconomic position (n = 34,72,74,75,84,93,98,100,102,115,119,121,125,129,133–135,148,149,153,156,159,160,162,163,169,171,177–179,183–186,192 26.0%). There was also a large group of studies (n = 21,73,76,83,87,89,90,101,111,116,120,128,137,138,152,164,168,170,172,187,189,200 16.0%) that had multiple PROGRESS-Plus categorisations, as clear from the title and abstract; the majority of these represented a combination of ethnic minority group status and low-income and/or inner-city status. Most of the ethnic minority participants were classified as black, African American or Hispanic/Latino.

The database included a spread of participants across all age groups and most of the studies included participants from more than one age group. The majority of the studies included young people (aged 11–21 years; n = 79;72–76,81,82,87–92,95,97–102,104–110,112,113,115,117,118,120,122–125,127,129,130,133–135,138,140–144,146–149,152,153,155,157,158,160–163,167,171,173,175,177–179,181,183,184,186,188,192,194,195,200,201 60.3%) and/or adults (age 22–54 years; n = 65,72–74,77–82,85,90–92,95–98,100–103,105,109,112,113,120,122,125,126,128,129,132–135,137,138,140,143,146–151,155,157,160–164,171,172,175,177,178,181,183,184,186,187,194,196,199 49.6%). In total, 7971,75,76,80–85,87–92,96,99–101,103–108,110,112–119,122,123,127,131,132,134–137,139,141–143,145,148,152–154,156,158,159,164–167,169,170,172–174,176,179,180,182,185,188–190,192–195,197,200,201 (60.3%) studies had mixed sex samples, 4972–74,77–79,86,94,95,97,98,102,109,111,120,121,124–126,128–130,133,138,140,144,146,147,149–151,155,160–162,168,171,175,177,178,181,183,184,186,187,191,196,198,199 (37.4%) had predominantly female samples and three93,157,163 (2.3%) had predominantly male samples.

The interventions

The interventions were conducted over a range of health topics (Table 14). The most commonly targeted health issue was substance abuse (n = 18,75,88,99,107,108,111,118,123,124,127,141,142,158,167,173,179,192,200 13.7%), followed by cardiovascular disease (n = 14,80,82,90,106,126,135,143,154,156,166,170,193,196,199 10.7%).

The interventions were conducted in a range of settings and in many cases across multiple settings within an intervention (Table 15). The most common setting for interventions was in the community (n = 56,73,75,82,85–87,90–92,99,100,102–105,107,113,115,119,122–124,131,134–138,141–143,145,152–154,156,158,159,161,164,167–174,176,177,179,180,190,193,195,201 42.7%). Many interventions also included non-geographic settings, such as media tailored to the participants’ needs (e.g. tailored newsletters or information sheets; n = 53,74–77,79,80,83,85,87,90,91,93–97,101,102,109,110,114,115,117,121,125–128,131,133,135,140,145,146,149,154–156,160,162,163,171,177,178,180,183,184,190,197–201 40.5%) or mass media (n = 21,73,90,100,103,113,115,122,125,127,131,134,136,137,142,145,148,154,156,158,177,180 16.0%); such interventions could be delivered wherever the participant was located.

Most of the interventions included multiple intervention strategies (Table 16). The most common strategy was education provision (n = 105,72,73,75–96,98–115,117,118,120,122–128,131–134,136–139,142–145,147–156,158,159,161–166,172–174,176,177,179,181,182,185,187,188,190–201 80.2%). Advice (n = 71,71,72,74,77–79,82,83,87–89,81–93,95,96,98,99,101,103,107–112,118,121,123,124,128–134,139–144,146,147,149,155,156,160,163,168,171,174,175,177–182,184,186,187,189–191,193,197,198,200,201 54.2%), social support (n = 58,72–74,78,80,84,91–95,98,100,101,103,105–107,109,111,116,118,120,123,124,129,130,141,142,144,147–150,153,155,156,160–163,167,172,175,177,179,181–184,186–190,197,198,201 44.3%) and skill development training (n = 51,72,76,77,79,81,82,84,88,89,92,94,101,102,104,106,107,110,115,118,123,124,126,127,129,138,141,142,145,147,149,152,155,158,161–163,170,171,173,178,179,181,182,192–196,198–200 38.9%) were also common strategies.

The interventions were delivered by a variety of professionals and non-professionals, sometimes in collaboration with each other (Table 17). The most common intervention deliverers were non-peer community members (n = 58,75,79–81,83–85,90–92,94,96,98,103–106,109,112,113,115,116,118,122–124,128,130–132,134,136,137,142,144,145,147,148,150,152,154,157,158,165,167,172–174,177,179,182,190,193–195,198,199,201 44.3%) and peers (n = 49,72–74,77,87,93,95,97,100,102,103,118,120,121,125,129,130,133,135,140,141,146,147,149,152,153,158,160–164,166,168,169,171,175–179,181–184,186,188,192,200 37.4%).

The comparators and evaluation features

A variety of comparators were used in the intervention evaluations. The largest group of evaluations employed usual care comparators (n = 38,71,72,74,76,83,89,97–99,103,108,110,115,120,123,126,127,129,130,135,141,145–147,151,155,157,158,160,162,163,166,169,181,184,192,193,200 29.0%; Table 18). Thirteen (10.0%) of the studies directly compared the effectiveness of community engagement by having a comparison condition that differed from the intervention only in the involvement of community members. 84,88,93,97,98,109,129,157,181,183,184,186,187

Fifty-nine (45.0%) evaluations used randomisation methods to allocate participants to the intervention or comparison condition (Table 19). 71–73,77–80,82,83,85,86,95–98,101,105,106,108,116,117,120,123–129,135,138,141,142,144,146,147,150,153,155,158,161,162,167–169,172,173,175,181,183,184,187,189,190,192,194–197 Twenty-six (19.8%) of the evaluations conducted follow-up testing. 72,75,77–79,87,93,95,98,101,107,114,125,127,129,131,140,144,151,153,173,183,186,191,192,195

The effect size estimates

As described in the Chapter 2, studies could contribute more than one effect size estimate to the data set under the following conditions:

• when there were both immediate post-test and delayed follow-up measures, to test the persistence of effects over time

• when there were outcomes from different points in the causal pathway (see Figure 15)

• when there were measures of both engagees and public health intervention participants.

As a result, we calculated multiple effect size estimates for some studies: a total of 212 across the 131 studies. Of the 212 effect size estimates, 191 were calculated from post-test measurements and 21 were from follow-up measurements. The remainder of this section refers only to the 191 post-test effect size estimates. The follow-up measures are explored only in terms of the maintenance of intervention effects (see Maintenance of intervention effects).

Of the post-test effect size estimate, 82 out of 131 studies (62.6%) contributed only one effect size estimate and the mean number of effect size estimates per study was 1.47 (SD = 0.68). 71,72,74,75,83,84,88,89,91–93,95–97,99–109,115,117,118,120,121,123,124,126,128,133,135,137–142,144,146,149,151,152,155,157,159,162,163,169,171,173,174,176,178,180,181,183,184,186,188–193,196–198,200,201,230,259,266,287,335,390,391,403 Figure 16 shows the frequency of studies by number of effect size estimates.

FIGURE 16.

Frequencies of studies with one or more post-test effect size estimates.

Effect size estimates based on continuous data are represented by d, whereas log-odds ratio (LOR) is used to represent effect size estimates calculated from binary data. The 115 post-test effect size estimates based on continuous data ranged from d = −0.77 to d = 1.43. (The range of values after outliers was Winsorised to 2 SDs above or below the mean and data were reverse coded when necessary such that all positive values indicate that the effect favours the intervention group. Three post-test effect sizes were Winsorised: all three were > 2 SDs above the mean.) A positive d indicates that participants in the treatment group, on average, scored higher than those in the control group. An effect size estimate of d = 1.0 means that participants in the treatment group scored, on average, 1 SD higher than participants in the control group on the particular outcome measure. In total, 12 of the 115 continuous post-test effect size estimates were coded as ‘missing data’ because there was insufficient information to calculate an effect size estimate.

The 76 effect size estimates based on binary data ranged from LOR = −1.05 to LOR = 2.14. (The range of values after outliers was Winsorised to 2 SDs above or below the mean and data were reverse coded when necessary such that all values > 1 indicate that the effect favours the intervention group. Three post-test effect sizes were Winsorised: one low outlier and two high outliers.) A LOR effect size estimate > 0 indicates that participants in the treatment group were more likely to achieve the desired outcome than those in the control group, whereas LOR < 0 indicates that the desired outcome was more likely in the control group. Four of the 76 binary post-test effect size estimates were coded as missing data.

To enable synthesis of the effect size estimates across outcome types (i.e. continuous d effect size estimates and binary LOR effect size estimates), the LOR effect size estimates were converted to d effect size estimates using the procedures discussed in Chapter 2. Figure 17 presents the frequencies of effect size estimates (for health behaviour outcomes only), in which the effect size estimates consist of d-values and LOR-values that have been converted to d-values. This figure indicates the distribution of the effect size estimates and shows that the majority of the effect size estimates were > 0.

FIGURE 17.

Frequencies of effect size estimates for health behaviour outcomes.

Overall, are community engagement interventions effective?

Across all outcome types, interventions were effective (Tables 21 and 22). There were insufficient effect size estimates for community outcomes and engagee outcomes and so the effect size estimates could not be synthesised statistically; we present these effects in Table 21. It is clear from the available outcome data that there are benefits to the community and engagees; all five outcomes are positive and statistically significantly different from a null effect.

Table 22 presents the results for the outcomes: health behaviours, health consequences, participant self-efficacy and participant social support. From Table 22 we can see that the pooled effect size estimate across interventions is positive (indicating that the outcomes measured were in favour of the treatment group) and statistically significant (as indicated by the p-values and 95% CIs) for these four outcomes. The statistical significance of the pooled effect means that the effect size estimate is significantly different from a null effect, in which there are no differences between the intervention group and the comparison group.

There was, however, significant heterogeneity across the four categories of quantitative outcomes observed in Table 22. As such, we conducted moderator analyses to attempt to explain variation in the observed effectiveness of the interventions. These analyses are described in the following sections.

Research question 3: Which approaches to community engagement are associated with improved health outcomes among disadvantaged groups? How do these approaches lead to improved outcomes?

Research question 4: Which approaches to community engagement are associated with reductions in inequalities in health? How do these approaches lead to reductions in health inequalities?

Research question 5: Which types of intervention work best when communities are engaged?

Research question 6: Is community engagement associated with better outcomes for some groups than others? (In particular, does it work better or less well for children and young people?)

Research question 7: How do targeted and universal interventions compare in terms of community engagement and their impact on inequalities?

Sample size

In Overall, are community engagement interventions effective? (see Table 22), results of a homogeneity test revealed that, for all outcomes, there was significant variability amongst the effect size estimates. Throughout this chapter, we have attempted to explain some of this variability through moderator and predictor variable analyses. Although many trends emerged, we failed to identify variables that could explain the heterogeneity amongst the effect size estimates.

One phenomenon that appeared to be related to the effect size estimates was the size of the study, as indicated by the funnel plot in Figure 22. To explore whether the sample size might explain some of the variation in the effect size estimates, we conducted an unweighted meta-regression. [An unweighted model, in which the weight for all studies was set to 1, was used because including study weights in the model would inflate the observed relationship between sample size (the independent variable) and effect size (the dependent variable). This is because the inverse variance study weights used in meta-analysis are largely a function of sample size.] This model tested whether the log of the sample size of each study predicted the size of the effect for health behaviour outcomes. The results indicated that, although sample size was not a significant predictor of the effect size estimate (unstandardised regression coefficient = –0.10, standard error = 0.08), it did explain about 10% of the variance in the effect size estimates (as indicated by the model R²). As such, it is likely that sample size accounts for some of the heterogeneity observed amongst the effect size estimates.

Acceptability

Four studies243,375,381,477 examined issues of acceptability. In each case these studies simply measured participation rates and suggested that higher participation rates implied participant acceptability of the intervention. Three studies discussed different reasons for higher participation rates. Quinn et al. 375 indicated that participation rates may have been influenced by delivery of the interventions by community volunteers in particular. Davis et al. 243 and Grant et al. 477 suggested that participation rates might have been influenced by the way in which the community members (e.g. teachers, volunteers or leaders) approached their involvement in the intervention (i.e. with enthusiasm, supported or feeling confident). Conversely, Ritchie et al. 381 noted that community workers were less likely to address the issue of smoking with low-income community members because of fears of alienating them. Further, the same workers suggested that an approach to smoking cessation in a community must start from a more holistic view of health, raising self-esteem and reducing stress before addressing smoking behaviours.

Accessibility/programme reach

Five studies73,208,194,375,475,476 discussed issues related to the facilitation of accessibility or programme reach for participants, linking these to participation rates. The interaction between community members – study participants and those who supported the intervention in some way – appears to influence how widely a programme reaches. Quinn et al. 375 indicated that participation rates were high because lay health educators were able to recruit extensively from within their church congregations. Wilcox et al. 194 noted that programme awareness was associated with key outcomes. Participants who had spoken with a church member about physical activity during the intervention phase were more likely to be in an action- or maintenance-level stage of change; to be undertaking moderate physical activity; and to have increased fruit and vegetable consumption. When participants reported pastor support for physical activity and healthy eating interventions, a statistically significant higher level of physical activity was reported. Duggan et al. 476 found that assessments by programme staff (rather than hospital staff), conducted in person, may have increased the reach of a programme. In addition, they suggested that, when an index child had considerable identified need, either the family’s recognition of that need or the home visitor’s efforts to encourage that family to take up services influenced the family’s decision to accept an intervention.