We formulated our content-related eligibility criteria according to the Participants–Concept–Context (PCC) mnemonic, recommended for scoping reviews by the Joanna Briggs Institute [21]. Participants were defined as professionals working in health promotion – either in public health, primary care, or other sectors – and citizens or communities. The first concept we defined was IC, which was considered as collaboration on health promotion between professionals from two or more sectors, as defined under Participants. The second concept was CP, which we considered as the involvement of citizens and communities in developing health promotion interventions or policies. Health promotion was defined as interventions and policies that aim to foster healthy behaviors within a supportive social and physical environment. The context was defined as local settings such as a neighborhood, community, or district. Studies were eligible if they mentioned facilitators and barriers for IC and/or CP. Only original research published in English or Dutch in 2012 or later was included. Although our protocol did not explicitly exclude non-empirical publications, full-text screening revealed a substantial volume and diversity of empirical studies. We estimated that non-empirical publications would have limited added value and decided to include only original research. The date range was selected to ensure relevance to current policy and practice, and to balance comprehensiveness with feasibility. A detailed overview of the inclusion and exclusion criteria is listed in Supplementary Appendix I.

We employed a three-step search strategy. First, we conducted a limited search in PubMed and Embase to identify highly relevant articles or so-called “key papers.” In these papers, we identified keywords in title, abstract, and index terms. Second, we conducted a search in PubMed, Embase, PsycInfo, CINAHL, Scopus, Global Health, Sociological Abstracts, and Social Services Abstracts, using the keywords and index terms identified in step 1. The search string for this search was built in two parts. The first part, about IC, consisted of four sets of search terms. The second part, about CP, consisted of three sets of search terms. Both parts were based on the PCC-mnemonic. The two parts of the search strategy were combined using OR. The search terms are listed in Table 1, the complete search strings for all databases are presented in Supplementary Appendix II. The search was conducted on 23 March 2023. A search update was conducted on 19 February 2025. Lastly, after study selection, we screened the reference lists of all included full-text articles for additional studies.

The search results were uploaded to Zotero 6.0.36, and duplicate records were removed. Inter-rater reliability sessions were conducted to ensure consistency between reviewers. Three reviewers (JS, LV, and JB) independently reviewed 25 records, after which they discussed discrepancies until consensus was reached. This process was repeated with two reviewers (JS and LV), with the third reviewer (JB) being consulted when consensus could not be reached.

The two reviewers then conducted title and abstract screening. For this process, we used ASReview 1.5, an open-source tool that applies active machine learning to continuously prioritize relevant records based on previous screening decisions. We followed the SAFE procedure as described by Boetje and Van de Schoot [22], which consists of four phases: Screen, Apply, Find, and Evaluate.

In the Screen phase, 1% of records were screened and labeled as “relevant” or “irrelevant” to train the active learning model. Based on the results of the Screen phase, the number of relevant records in the total dataset was estimated.

During the Apply phase, the two reviewers independently screened the dataset. Instead of following the stopping criterion described in our protocol – no relevant records identified in the last 100 records – we adopted the stopping criteria described in the SAFE procedure. This means we screened the dataset until the following four stopping criteria were met:

All key papers were labeled as relevant.

In cases of doubt, the full-text article was reviewed to ensure that the active learning model was adequately trained.

During the Find phase, a deep learning model was employed to ensure that no relevant records were missed. The labeling decisions from the Screen phase were used as prior knowledge to train the deep learning model, which then re-ordered the remaining unlabeled records. The reviewers then continued screening until no extra relevant records were identified in the last 50 records. After the Find phase, the reviewers compared their independently labeled records. Discrepancies were discussed, with the third reviewer consulted when consensus could not be reached.

The datasets of both reviewers were then combined into one file, with which the Evaluation phase was conducted. The active learning model for this phase was trained using the 10 highest- and lowest-ranked records from the Find phase. The third reviewer then screened the previously labeled records to identify any relevant ones that had been excluded. Screening was stopped when no extra relevant records were identified in the last 50 records.

For the updated search conducted in February 2025, one reviewer (JS) screened new titles and abstracts in ASReview. All full-text articles included in the initial search were used as prior knowledge. Screening was stopped when no extra relevant records were identified in the last 50 records.

Full-text articles were uploaded and screened using Zotero. The two reviewers independently screened all articles. Disagreements about inclusion were resolved through discussion, and when necessary, the third reviewer was consulted. Finally, the reference lists of all included articles were screened to identify additional relevant studies.

We charted data from all included studies using the data charting form presented in Supplementary Appendix III. This is a revised version of the form presented in the study protocol, as we identified excessive overlap between categories. We charted data on study characteristics, facilitators, and barriers. After data charting, we conducted a thematic analysis of facilitators and barriers for IC and CP using NVivo 14 [23]. Text related to facilitators and barriers was coded and discussed iteratively by two researchers (JS and LV) to further refine the codes. Then, the two researchers reviewed the codes and their relationships, leading to the creation of categories and themes.

A flow chart of the study selection process is presented in Figure 1. The numbers in this flow chart are the combined numbers from the initial search and the search update.

The initial search, conducted in March 2023, yielded 23,143 records. After removing duplicates, 12,986 records were imported into ASReview. In the Screen phase, 129 records were screened and labeled as ‘relevant’ or ‘irrelevant’ to train the active learning model. Based on the results of the Screen phase, the estimated number of relevant records in the total dataset was 2,059, indicating that a minimum of 4,118 records needed to be screened in the Apply phase to meet stopping criterion 2. Title and abstract screening using the SAFE procedure yielded 239 articles for full-text screening. After full-text screening, 41 studies were included in our review. The search update in February 2025 resulted in 22 additional potentially relevant records after title and abstract screening, of which three were included after full-text screening, yielding a total of 44 included studies.

Table 2 presents an overall summary of the study characteristics. The characteristics of the individual studies are provided in Supplementary Appendix IV. The included studies were published between 2012 and 2025, with the majority appearing between 2018 and 2020. In most studies, the focus of health promotion was healthy weight, food, and/or obesity (n = 26). Most studies (n = 21) employed mixed methods or a qualitative design (n = 18).

We summarized the results of our data charting and analysis in accordance with the study objective: to map facilitators and barriers for IC and CP in community-level health promotion. The results are presented as a narrative summary, supported by an illustrative framework (Figure 2).

Our main findings are organized into six themes. For each theme, we describe the factors it encompasses. These factors can act as facilitators, barriers, or both, and they relate to IC, CP, or both. When factors apply to both IC and CP, their specific interpretation may vary across these contexts. Table 3 provides an overview of themes, facilitators (F), and barriers (B) for IC and CP in the included studies.

Of the 44 studies we included, 16 reported facilitators and/or barriers for IC, 16 reported facilitators and/or barriers for CP, and 12 reported both. Overall, facilitators are reported more often than barriers. Relational dynamics, Essential people at the table, and Resources are the most frequently reported themes for IC (96%, 93%, and 93%, respectively) and for CP (100%, 96%, and 96%). Shared purpose and Operationalization are reported more frequently for IC (89% and 79%) than for CP (67% and 48%). Personal drives are reported more frequently for CP (48%) than for IC (11%). Figure 3 provides a visual comparison of reported themes in facilitators and barriers for IC and CP.

Overall, the six themes we identified were similar to previous reviews on IC [10, 12, 13, 17, 18], with one exception: no previous reviews report on the personal drives of participants. A possible explanation for this is that these reviews focused on IC alone, while the personal drives in our scoping review were mainly related to CP. At the factor level, communication, leadership, financing, organizational support, and shared goals are commonly identified. Factors not often identified are engagement, organizational benefits, and the visibility of results; these are only described by Leenaars et al [17]. Attention to tools, methods, and the design of meetings and activities to contribute to capacity building are reported by Rechel [10] and Martin-Misener et al [13]. Rechel [10] included the review by Martin-Misener et al [13], which explains their similarity in reported factors.

The importance of community engagement is described in three of the reviews on IC [10, 13, 18]. However, we found no previous reviews on CP in health promotion. CP is often defined as participation in health promotion activities, not in the development of activities or policies.

Visualizing our results in a framework proved valuable for interpreting the outcomes of our study. When comparing our framework to existing models, we found substantial resemblance with the Center for Community Health and Evaluation (CCHE) collaborative model [25, 67]. Other relevant models were less applicable. For example, both the HALL framework [68] and the framework by San Martín-Rodríguez et al. [69] – which guided the review by Martin-Misener et al. [13] – categorize determinants of collaboration across three levels: systemic/institutional, organizational, and interactional/interpersonal. This classification enables a structured analysis of the multi-level dynamics that influence collaboration. However, we identified several dimensions that are underrepresented or absent in these original models compared to our findings, for example, the presence of essential people at the table and the role of personal drives. The latter may be explained by the fact that these models focus on collaboration between primary care and public health, not on integrating CP into IC. We did find this critical role of stakeholder inclusivity in the CCHE model, which places the community and equity at the center of collaboration.

Although our themes broadly align with those in the CCHE model, several distinctions emerged. First, we identified “Personal drives” as a separate theme to capture intrinsic motivation for collaboration. Second, we split “Adequate structure and support” into three subthemes to more precisely locate facilitators and barriers. Finally, we incorporated leadership within the broader theme of “Relational dynamics,” emphasizing the relational soft skills that underpin effective collaboration.

Several limitations of our scoping review should be acknowledged. Firstly, we found that the definitions of concepts of interest to our review, such as health promotion, CP, and public health, varied considerably between studies, or it was unclear how these concepts were defined in the studies. This made it challenging to decide whether a study met our eligibility criteria, possibly resulting in the omission of relevant studies. For example, in some cases, it was difficult to determine whether CP indicated participation in the development of health promotion policies and activities, in the implementation of these activities, or participation in the activities themselves.

Secondly, we used ASReview for the title and abstract screening. This proved valuable for efficiently managing the large number of records our search yielded, as it significantly reduced the number of records that needed to be screened. However, a consequence of this approach is that we did not screen all titles and abstracts, which may have resulted in the omission of relevant studies.

Finally, it is worth noting that this review was designed to map facilitators and barriers for IC and CP; exploring how these facilitators and barriers relate to each other was beyond its scope.

Our findings and framework can support policymakers and stakeholders at various levels of the public health workforce seeking insight into the factors that influence IC and CP, as well as how to initiate, sustain, evaluate, and optimize collaboration processes.

At the professional level, when initiating a community-level health promotion collaboration, it is essential to involve all relevant stakeholders from the outset, including citizens. Establish a shared vision and common goals collectively to ensure that the interests of the various partners and the needs of the community are met. Meaningful CP requires treating citizens as equal partners, actively listening to their concerns and priorities, and fostering the exchange of diverse perspectives. This not only enhances the relevance of results but also contributes to building mutual trust. Participants’ differing motivations need to be recognized as well. Professionals typically participate on behalf of their organizations within a formalized structure governed by rules and agreements. In contrast, residents are personally invested, as the collaboration concerns their living environment and their daily lives.

At the organizational level, roganizational support is a key factor: IC can be challenging and is unlikely to succeed without adequate support in terms of time, money, commitment, and ownership.

In general, allowing sufficient time for the transition to a collaborative working structure is essential to foster sustainable and effective partnerships.

With our scoping review, we mapped six overarching themes concerning facilitators and barriers for IC and CP in community-level health promotion: Shared purpose, Operationalization, Relational dynamics, Adequate structure and support, Essential people at the table, and Personal drives. Understanding the drivers and mechanisms of IC and CP can support evidence-informed decisions at various levels of public health collaboration. Our framework can serve as a tool for evaluating and improving the diverse aspects of the IC and CP processes.