The core search strategy hinges on three core concepts—contributors, engagement practices, and LHSs—each of which is described by a range of terms. Many of these terms have overlapping meanings, subtle context-specific nuances (2, 53), and/or politicized interpretations (54, 55). Diversity of terms can lead to unclear meaning and fragmentation in the literature (56, 57). Defining terms and identifying potential synonyms increases clarity and consistency, and enables a more comprehensive search strategy (58). Drawing on prior knowledge, a list of search terms was drafted then revised based on the literature on patient engagement (47, 49, 50) and LHSs (2, 59), as well as review protocols on engagement-related topics (58, 60, 61). This list was refined with feedback from the SAG. Notably, SAG members pointed to established, if siloed and distinct, traditions of engagement within research and quality improvement. To increase the relevance of search results, the setting was restricted to terms specific to learning health systems, organizations, and/or networks. This approach is consistent with previous research (17, 59), while also allowing some flexibility to capture different approaches to LHS implementation at the meso- and macro-level. Advocacy-related terms were excluded, as the SAG did not see this as a core function of engagement within local health systems. Finally, taking a similar approach to Chrysikou et al. (58), terms were grouped into “domains of terms.” Domains of terms were mapped against the first three components of the PerSPEcTiF question framework: perspective (Per), setting (S), phenomenon of interest (P), environment (E), comparison (c), timing (Ti) and findings (F) (Table 1). This question framework accounts for context in complex interventions (62), and provided structure for subsequent data extraction and analysis.

Searches were performed for the terms included in each core domain—Perspective (contributors), Setting (LHSs in direct patient care settings), and Phenomenon (active engagement)—and were limited to the title, abstract and keywords. Where available, Medical Subject Headings (MeSH) terms were included. The initial search strategy was developed for PubMed-MEDLINE (Supplementary Material S2), and adapted for CINAHL, PsycINFO, and Embase. Databases were searched in December 2022. Search syntax translations and search results were tracked in Microsoft Excel (63) (Supplementary Material S3). Targeted keyword searches were used to hand search highly-relevant journals: Learning Health Systems, Health Expectations, Research Involvement and Engagement, Health Research Policy and Systems, and BMC Health Services Research. Additionally, MS shared a collection of peer-reviewed literature that was tagged for patient engagement in another LHS review (17, 59). To maintain a manageable sample, all grey literature was excluded.

The core search results, as well as peer-reviewed literature recommended by MS and/or identified during the initial search, were uploaded to Covidence, a web-based software platform, for deduplication and screening (64). Screening was complicated by the heterogeneity of the literature and, as a result, inclusion and exclusion criteria were refined through three rounds of title and abstract screening. To speed the process, studies likely to meet inclusion criteria were tagged “highly relevant” and moved directly to full-text screening. Gradually, through this familiarization process, a domain-based screening rubric was developed (Table 2).

Given the large number of papers meeting the refined inclusion criteria (n = 63), a sampling strategy was devised based on “thick description” (65, 66). Here, “thick descriptions” are loosely defined as those that offer sufficient detail to contribute to meaningful interpretations of the literature and, employing a common strategy, “thick descriptions” were contrasted against “thin descriptions” as relative measures of the level detail (67). All sources meeting the inclusion criteria were exported from Covidence (64) to Microsoft Excel (63), and assessed for richness using a 2 × 2 matrix based on the detail provided about the LHS and engagement activities. This sampling strategy allowed for the examination of contributor roles in different settings, contrasting different approaches to engagement. RG performed screening and sampling, with a small number of sources additionally screened and assessed for richness by TF and JW.

The literature was highly heterogenous (Table 3), with only a third of sources (n = 12) focused primarily on reporting engagement activities. In addition to providing rich detail, these sources surfaced tensions around representation (68), between individual experiences and population-level data (72, 73), trust and different ways of knowing (15, 74, 75), and also documented barriers and facilitators to engagement (15, 68, 72, 73, 75–78). Yet, half these sources offered scant contextual information on the LHSs (15, 73, 75–77, 79). This highlights the value—and relative scarcity—of “thick descriptions” of engagement within the LHS literature (even when selected for), and the separation of rich detail on engagement from contextual information about the LHSs where these activities take place. Conversely, most sources describe a particular LHS, with varying levels of detail about engagement reported as a part of a broader program of work (n = 20). A small number of sources tackling aspects of LHS design (n = 4) also met the inclusion criteria by including examples of engagement in patient care settings.

Most papers (n = 22) took the form of an experiential report that describes the firsthand experience of an LHS (68, 72, 74, 75, 78–95). Sometimes experiential reports were published explicitly as “use cases” (95), “real-world examples” (81), case studies (68, 92) or commentary (80), but often without providing specific methods [e.g., (78, 79, 92)]. Notably, only nine sources explicitly reported contributors as co-authors in-text and/or in the author information (15, 76, 77, 79, 80, 87, 96–98). While this suggests limited co-authorship by contributors, this may be an underestimate. For example, two sources used group authorship—CERTAIN Collaborative (72), “the pSCANNER team” (89)—where a full list of people involved was included at the end of the paper (whether those listed included contributors was unclear). Contributors may also identify based on professional roles within health systems but also have lived experience [e.g., (77)]. Still, this suggests that contributors may be infrequently involved in reporting engagement practices within LHSs.

All included sources were published in 2013 or later, with most sources (n = 31) published in 2017 or later. Nearly a third of sources (n = 10) were published in 2022, the year that the searches were performed. While the recency of the literature may improve its relevance to the current policy window, the sequential delays between the coining of LHS in the 2007 National Academy of Medicine (NAM) report (6) (the definition most cited amongst the sampled literature), the development of LHS models with an active role for patients, and the reporting of patient engagement activities may signal a tendency for engagement to be added in after implementation, developing over time as LHSs mature (72, 79, 82). Additionally, much of the literature is fragmented between journals with different topic areas, with 36 sources published in 26 journals. Only one source was published in a journal focused specifically on engagement (Health Expectations)—suggesting that the academic literature describing engagement activities within LHSs is relatively isolated from the broader peer-reviewed literature on engagement.

Theory is used in the literature to situate engagement within LHSs; guide engagement activities; organize collaboration; implement and scale interventions; and critically examine social processes (Table 4). Despite the existence of numerous patient engagement frameworks (48), engagement functions were generally described as a component of broader LHS models. Only two sources used engagement frameworks to guide activities (72, 76), with a further two sources highlighting engagement frameworks that helped orient publicly-funded health systems towards engagement over time, namely the Strategy for Patient-Oriented Research (SPOR) in Saskatchewan, Canada (80) and the Person-Centred Practice Framework in Sweden (78).

More commonly, organizational theory is used to guide collaboration, and implementation science frameworks used to implement and scale interventions. Notably, there is a small cluster of sources with overlapping authorship that describe the use of “Actor-Oriented Architecture” (88, 90, 99), developed by Fjeldstad et al. (100), to theorize aspects of collaboration. Amongst these, Murray et al. (88) emphasize that organizations can be designed to support people to prioritize and solve problems—reflecting distributed leadership as a mechanism for collaborative priority setting. Overall, there is a notable lack of critical analysis of social processes. Milligan et al. (97) offer a strong critique, arguing that the “rather narrow orientation toward clinical research and health service delivery neither adequately captures the breadth and depth of relevant theory from other research traditions…nor promotes a model that prioritizes nonclinical or tacit forms of knowledge” (p. 2). There are, however, a handful of exceptions drawing on critical social theory to challenge inequities arising from colonialism, social marginalization, and/or established hierarchies of knowledge (15, 95, 97).

Most sources reported engaging more than one type of contributor (n = 30). Patients or people with lived experience of a health condition (n = 34) were engaged most often, followed closely by family and informal caregivers (n = 29) (Figure 2). Engaging contributors alongside other stakeholders was seen as advantageous, for example, in describing the design of the Collaborative Chronic Care Network (C3N), Seid et al. (90) noted that “taking a multi-stakeholder perspective forced the design team to consider the new system not as a system for doctors or a system for patients, but rather as a system for people” (p. 7)—an approach aligned with the quadruple aim central to the value proposition for many LHSs (4) and fitting pragmatic approach common to POR (21).

Engaging with communities was less common (n = 12) but also provided the only two instances where patients were not engaged. Community engagement focused on addressing persistent health inequities rooted beyond healthcare systems (97, 101). Notably, the term “community” was used variably, often making it difficult to determine who was being involved. Perhaps because the concept is nebulous and the practicalities of engaging large groups challenging (90), community perspectives were typically accessed via individuals: community leaders (83), organizers (90), representatives and/or members (75, 82, 85, 95, 101)—raising the thorny issue of representation (75). There were less frequent references to broader community outreach (75, 93) and partnering with community organizations (95), and these approaches were most common in LHSs with strong traditions of participatory community-based research (75, 101). Nonetheless, “community” tends to be treated as cogent “categories of shared identity” [(101), p. 2], often with scant reference to tensions within and between groups or the potential for intersecting identities.

Community engagement was also characterized by more a relational approach. For example, Seid et al. (90) frame LHS models as capable of appealing to solidarity, and thus supporting a shift from “health care as a transaction to shared work” (p.5). However, this focus on shared enterprise does not necessarily challenge the underlying assumptions about health or ways of knowing within these systems. Milligan et al. (97) emphasize how Indigenous knowledge systems, including a relational approach to well-being, can be at odds with colonial health systems that are typically rooted in Western traditions and derived from a biomedical model that privileges individual patient autonomy. This epistemic relationality presents a more fundamental challenge for LHSs, and the ways in which contributors' expertise and knowledge is valued within them. Notably, references to “publics” were rare (n = 5), amorphous, and often presumed, e.g., Levin et al. (84) indicated “patient partners provided a welcomed perspective on concerns and questions relevant for members of the patient community and the greater public” (p. 5). As a result, publics were not considered a distinct category for analysis.

Finally, examining who was being engaged provided insight into the potential nature of their contributions. There was considerable variation in the terms used to describe contributors, which were grouped to reflect their perceived expertise (Table 5). While the emphasis is clearly on lived experience of patients and caregivers, less common terms highlight potential shifts, for example a growing emphasis on cultural and/or organizational knowledge and relationships [e.g., Elders and Knowledge Keepers (97), veterans (102)]. Despite relatively few search terms emphasizing the relationship to the LHS, several were identified in the literature, e.g., actors (95, 100), coproducers (77), or “improvers” (74)—highlighting experience with the LHS as a sought-after form of expertise.

Timing of engagement was difficult to discern and largely inferred from the description of engagement activities. It was generally possible to determine whether engagement first occurred before or after the initial LHS implementation, with an even split between LHSs that engaged contributors from the outset or during early implementation versus those that engaged after implementation (Figure 3). This breakdown suggests that patients are often not default players, despite suggestions otherwise [e.g., (99)].

The exception was large publicly-funded LHSs, most of which initiated some form of engagement prior to LHS implementation (73, 80, 84, 89, 97, 98, 102, 103). The connection appears more tenuous at other levels of implementation, but there may be complex relationships at play. For example, all four meso-level LHSs that initiated engagement prior to implementation tended to be nested in larger organizational settings with established engagement traditions (77, 93, 96, 99, 104). This suggests that timing of engagement could be influenced by the presence of ongoing relationships and existing capacity for engagement. However, it is difficult to tell how much these findings are impacted by lack of reporting about the timing of engagement and/or when a health system starts using LHS terminology to describe itself.

Contributors were engaged in 30 LHSs providing direct patient care (Supplementary Material S4). Most sources reported on LHSs operating in a single, high-income country: USA (n = 22), Canada (n = 6), United Kingdom (n = 3), or Sweden (n = 2). Exceptions included two sources offering examples from the US and Sweden (77, 99), and sources reporting on the Starzl Network for Excellence in Pediatric Transplantation, an international collaboration with partners in the US and Canada (87, 91), and ImproveCareNow (ICN), which has members in US, UK, Belgium, and Qatar (68, 74, 81, 90, 99, 105).

Clearly, engagement is happening in a variety of LHSs, operating in a range of care settings, serving diverse populations, and implemented at different levels, from smaller projects to large-scale networks. Half of the LHSs (n = 15) focused on specific diseases (68, 74, 81, 90, 99, 105) or services for specific disease (77, 99). Two LHSs focused on creating infrastructure: a learning network supporting discrete projects (83) and an LHS creating technical infrastructure to support data sharing between three large health networks (89). Additionally, some larger networks aimed to create unified standards of care across multiple organizations (87, 91). Nearly a third of LHSs (n = 7) were categorized as having more than one focus, for example Early Intervention Services for Psychosis in Québec is a disease-specific LHS implemented within a publicly-funded provincial health system (103).

Similarly, the boundaries between meso- and macro-levels of implementation are porous, for example some LHSs categorized as meso-level operate as single health systems but involve networks of facilities [e.g., (86, 94)]. Boundaries between LHSs are often grey. The Cincinnati Children's Hospital, an LHS early adopter in the US, acts as a common denominator for multiple LHSs, including operating internal services as LHS (Diabetes Centre), hosting larger networked LHS (ICN), and providing mentorship and support for LHS spin-offs (C3N). Nesting LHSs within a single health system was evident in other sources too, for example, the STRONG STAR Consortium (102) and Veteran Affairs Evidence Synthesis Program (73)—both positioned as LHSs, but taking different approaches to engagement within Veteran's Affairs (itself an LHS). These findings suggest that engagement is present in LHSs of all shapes and sizes, and points to opportunities to nest LHS approaches—providing learning opportunities within and between LHSs.

Unsurprisingly, given the variability of LHS implementation, several sources emphasized the need to tailor engagement to specific LHS settings (83, 94, 97, 101). Yet, there were no obvious connections between LHS focus or level of implementation and contributors' roles or the timing of engagement—with one notable exception. LHSs categorized as “publicly-funded, macro-level LHS” (n = 7) tended to focus on power-sharing and/or partnership as key motivators for engagement (73, 78, 80, 84, 89, 97, 98, 102, 103), as opposed to more pragmatic contributions to bounded projects and, as noted previously, all but one initiated engagement activities prior to LHS implementation (Figure 3). Analysing a diverse sample of LHSs helped to identify shared features, present across various settings, that differentiate LHSs from more siloed traditions of research and quality improvement (Table 6). These features impacted engagement—both positively and negatively—and thus represent important contextual considerations for designing patient-oriented LHSs and engagement practices within them.

Engagement activities (n = 192) were grouped into a taxonomy of engagement outlining seven functions (Table 7). All sources reported more than one type of activity. The most common engagement activities have strong traditions in patient-oriented research and/or quality improvement: advisory councils (n = 20); informal roles for individuals (n = 15); patient surveys (n = 16); training and education (n = 13); working groups (n = 13); creating resources (n = 9); providing testimonials (n = 7). These activities are consistent with the tendency for LHSs to build on internal research and quality improvement work [e.g., (75, 101)] or be designed de novo by partners drawing on existing traditions of engagement [e.g., (80, 90)]. Confirming the barrier noted by Menear et al. (4), priority setting activities were rare (n = 4). The lack of engagement in priority setting may be the product of delayed engagement, for example half of large networked LHSs reported on engagement activities occurring post-implementation (Figure 3), and is consistent with challenges noted in a CIHR SPOR evaluation (106). However, the relative rarity of these activities could also reflect a lack of reporting about timing of engagement and/or these functions may be assumed within references to other activities, e.g., governance committees.

The taxonomy of engagement provides practical examples of how contributors are being engaged in LHS, but also highlights shifts in engagement methods, particularly towards co-production (15, 76, 84, 88, 92, 95–98, 104) and, to a lesser extent, participatory methods (15, 95) that support community engagement. Existing approaches are evolving, e.g., the growing use of social media (18, 68, 73, 90, 107) and a shift from deficit-based training towards mentorship (74, 79, 103). However, newer approaches are less common, and the lines between old and new ways of engaging are often unclear. For example, in some instances, social media was used to involve patients in dialogue, e.g., the creation of an online community with “more than 150 people posting more than 1,700 messages” in a “private social media site” [(90), p. 5]. Conversely, social media channels were more commonly used to disseminate information [e.g., (18, 73)].

Nearly two-thirds of the 192 engagement activities were mapped to a position on the IAP2 Spectrum (n = 139) (Figure 4). The aggregate data suggests contributors' influence over decision-making remains limited in LHS settings, with only a small number of activities coded to “Empower” (n = 14). Roughly half coded along the low- and middle-spectrum: “Inform” (n = 11), “Consult” (n = 24) and “Involve” (n = 49). While the descriptions of these activities suggest that their influence over decision-making is relatively bounded, SAG members noted that these activities can still serve important functions and be appropriate, if aligned with engagement goals. More than a quarter of engagement activities were coded unclear (n = 53)—highlighting the need for more descriptive and evaluative reporting of engagement practices.

The breakdown of IAP2 Spectrum coding by activity provides a more nuanced view. Clearly design matters, with the same activity often occupying different places on the IAP2 Spectrum (Figure 5). This variation usually reflects differences between LHSs but, on rare occasions, the same type of activity occupied different places on the IAP2 Spectrum within the same LHS. These differences suggest that the nature of engagement cannot be taken for granted. For example, while co-production is billed as fundamentally collaborative, it was often invoked without details about the methods involved. As a result, about half of these general references were coded as unclear (84, 88, 92, 96, 104). Conversely, some references to co-production impacted decision-making sufficiently to be coded “Empower” (76, 95, 97), but the participatory methods associated with these approaches were coded as either “Involve” or “Collaborate”—suggesting co-production can also be more than the sum of its methodological parts.

The influence of the contextual features that impact engagement in LHSs (Table 6) is also apparent in the ways contributors are involved. For example, while governance committees often clearly described their decision-making authority (nearly all coded “Empower”), contributors may be ascribed influence over decision-making by virtue of their membership regardless of how much influence they exert. Similarly, influence over decision-making was far less clear for advisory groups than for working groups. Yet, decisions (albeit not usually at the system-level) tended to be made and implemented directly by working groups, perhaps an illustration of the assertion that distributed leadership can provide a path to meaningful engagement within LHSs [e.g., (75, 87, 95)].

The transition from informal to formal roles for individual contributors also provided opportunities for distributed leadership. Informal roles are more common (n = 15), but often had vague responsibilities and little or no reference to compensation. The scope of work for these roles tends to be more loosely defined (if at all) and/or their relationship to the broader organizations less clear. In contrast, four sources describe paid or funded formal roles for contributors (75, 86, 99, 102) that had clearer scopes (only one was coded unclear). Formal roles tended to move beyond narrative work and advisory functions to exert control over research, engagement, and/or knowledge translation activities, for example “Patient Supporters” in the Self-Dialysis Unit at Ryhov Hospital conceived the unit, co-designed the service, and trained patients to perform self-dialysis (99). Formal roles can also allow contributors to shape LHSs without assuming leadership positions, for example as “Community Research Associates” who combine research skills with community knowledge to perform a narrow range of tasks (75). These examples illustrate how contributors can be embedded within LHSs and guide the daily work of research and health system improvement.

Additionally, the active involvement of contributors in data-related functions appeared to be both constrained and evolving. Half of attempts to reflect contributor perspectives were consultative (n = 8) (18, 72, 74, 84–86, 96, 103), and accomplished primarily through patient/client surveys and standardized data collection tools (e.g., Patient-Reported Outcome Measures or, less frequently, Patient-Reported Experience Measures). There is a tension between bespoke or open measures that capture personal experience and standardized measures, which can be more cleanly collected, easily compared, and used to generate population-level data (72, 98). Knowles et al. (15) argue the need “to expand big data of informatics with the rich data of narrative and experience” (p. 112)—a position echoed in the call for data that is meaningful to both clinicians and patients (104) and capable of empowering people experiencing inequities (95). This process could involve engagement focused not only on what data is collected and how it is interpreted, but also by whom. There are indications that this transition is happening within LHSs through formal roles for contributors, for example Myers et al. (93) describe peer researchers gathering and interpreting data. There were also several engagement activities that actively involved contributors in data-related functions by helping to select and/or co-develop symptom questionnaires and survey tools (72, 83, 84), or through usability testing (103) and co-developing designing data dashboards (104).

Notably, despite excluding advocacy-related terms from the search syntax, advocacy activities were described in a small number of sources. For example, a petition jointly created by physicians and parents “to support prioritizing children for pediatric donor livers” received more than 15,000 signatures, demonstrating the “power of a unified voice” to “facilitate collection action” [(87), p. 423].

Ten distinct roles for contributors were synthesized, serving three broad functions within LHSs (Table 8). Contributors acted as System Shapers (designing and defining LHSs), Community and Capacity Builders (expanding and supporting LHSs), and Implementers (undertaking hands-on tasks) (Figure 6). Roles were flexible, varied, and often overlap with individuals sometimes performing several functions, along with the ability to come and go from roles. Three LHSs are associated with a single role for contributors (81, 85, 104); all other LHSs were associated with multiple roles.

For LHSs described in more than one source, e.g., ImproveCareNow (ICN) (68, 74, 81, 90, 99, 105), it was common for different sources to describe different roles. Differences in contributor roles between sources could be due to the focus of the publication but, in at least some cases, sources indicated roles that changed over time. For example David et al. (68) describes self-directed changes to the ICN Patient Advisory Council structure and membership engagement, and how this adaptive process helped “meet identified community needs” (p. 8). Therefore, roles described for each LHS are unlikely to be exhaustive but rather provide a snapshot of engagement and, just like engagement activities, offer opportunities to tailor engagement practices to specific LHS settings. There is considerable flexibility for those designing LHSs, both in addressing system needs and matching contributors with roles that suit their interests and skills.

In line with the dearth of activities coded to “Empower” on the IAP2 Spectrum, the most common roles related to the pragmatic “Implementers” function and are associated with a narrower scope of engagement—either as “Consultants” (n = 20 LHSs) or “Workers” (n = 18 LHSs). Conversely, the roles associated with “System Shapers” and “Community and Capacity Builders” suggest intriguing possibilities to reimagine engagement to take advantage of the unique features of LHSs (Table 6). For example, within the “System Shapers” category, “Leaders” did not necessarily have influence over system-level decision-making but were able to effect change through distributed leadership. For example, formal roles such as “funded investigators” (86) represent a devolution of powers akin to what Hughes and Duffy (108) describe as “user-led research,” where people with lived experience direct work including defining projects, seeking funding and publishing results. Defined to reflect distributed leadership, “Leaders” was the third most common role, present in nearly half of LHSs (n = 15 LHSs)—suggesting an expanding and possibly bottom-up influence for contributors over these dynamic health systems. Other examples include “Advocates” which reflect connections between LHSs and broader social context; “Service Designers” which reflects the embeddedness of contributors within clinical care settings; and “Reservoirs” which are well-suited to nested and networked approaches.

“Reservoirs,” where formal or informal affiliation connects contributors to LHSs, were also a key strategy for embedding contributors, and a way to balance between rapid change and potential longevity of LHSs (68, 74, 78–80, 90, 92, 95, 99, 105). Access to experienced contributors was seen as helpful for people new to engagement and a route to including more diverse perspectives (84). Reservoirs may also support flexibility, allowing contributors to take on and resign different roles over time. Practically speaking reservoirs were created via social processes (establishing and nurturing relationships), capacity development (training and mentoring), and organizational infrastructure, for example supportive policies, hiring engagement staff, formalizing contributor roles, routine communications (e.g., mailing lists), and patient registries. The impact of reservoirs depends on how they are used and maintained, and the ways in which contributors are called up. For example, patient registries can be passive pools occasionally dipped into (75, 78) or used to more proactively engage contributors, e.g., the Diabetes Centre Patient Registry, which is used to identify at-risk patients to involve and potentially support with an innovation fund (95). Having supportive organizational structures with engagement staff, e.g., dedicated office (76) or platform (80), is noted as a facilitator but warrants further exploration.

A closely related strategy involved embedding contributors as “Peer supporters,” whose involvement helped create engagement-ready environments. “Peer supporters” were involved in various peer functions, e.g., supporting engagement including through recruitment (75), onboarding (87), and helping other contributors to participate in or manage engagement activities and governance functions (79, 88, 89, 93). Peer functions were largely performed via groups, for example coalitions (80), collaboratives (76), networks (79, 98, 102). While mostly positioned as beneficial to the LHSs, contributors also expressed a desire for peer communities that provide forums to share experiences and support each others' engagement efforts (80). Peer functions were positioned near-universally as positive and supportive. For example, contributor involvement in recruitment is presented in a largely uncritical way with a focus on the benefits of attracting and supporting new contributors [e.g., (68, 75, 79, 80, 89)] and rarely problematized the potential bias for recruiting narrow perspectives. There was no discussion of what Vinson [(109), p. 3] calls “peer regulation”—a process of “orienting members’ action toward the stated values of a group”—and, by extension, the potential for “peer policing” or informal social control over contributors. More work is needed to understand how peers will relate to each other, as well as other health system actors, particularly those mediating peer functions such as engagement staff.