Engagement of patients, families, and others has become essential to research, from study conception to dissemination [1-6]. An increasing body of evidence shows that engagement can positively influence the research process—shaping the research questions asked, improving research conduct (eg, recruitment, retention, and data collection), guiding the return of results, and more—all in ways that center patient and community perspectives [7-12]. Cancer and cancer genomics research are no exception [13]. Patients have a long history of advocacy and engagement in cancer research [14,15]. Engagement is thought to be critical to promoting representativeness, recruitment, retention, and trust (understood broadly to mean participants can rely on the research enterprise to protect their interests) in genomics research [16]. For instance, engagement has the potential to improve research relevance, promote recruitment and retention, enhance dissemination, and increase diversity among research participants [17].

A recent review involving major cancer research and cancer genomics programs (such as the National Cancer Institute [NCI] Cancer Moonshot Initiative’s Participant Engagement and Cancer Genome Sequencing [PE-CGS] Network) called for measuring effectiveness and comparing methods of engagement in cancer genomics research [18]. Despite progress in the measurement of engagement effectiveness [9,19-25], comparative effectiveness research on discrete engagement methods (eg, advisory panel methods, social media–based engagement discussions, and community engagement studios) is sparse and tends to focus on a single aspect of a method (eg, group composition or online vs in-person modalities) [26,27].

Understanding the most effective, efficient, and patient-centered methods of engagement may be particularly important for rare cancers, including adult lower-grade gliomas [28]. Although rare (affecting <1 per 100,000 people in the United States), they represent up to 20% of malignant brain tumors, affect relatively young adults, and are associated with high morbidity and mortality. Genomics research promises to provide new insights into low-grade glioma (LGG) diagnosis and treatment, including understanding the many potential genomic variations in tumor types [29-31]. Because lower-grade gliomas are rare, research must often recruit from multiple geographic locations, with variability in social, economic, technological, and clinical contexts [32]. This geographically dispersed community creates challenges for research—yet presents an opportunity for studying engagement methods.

The International Low-Grade Glioma Registry (LGG Registry) was established in 2016 at Yale University to conduct genomic and epidemiological research into risk factors and outcomes for LGG. Recently, a study of the molecular evolution of LGG, termed Optimizing Engagement in Discovery of Molecular Evolution of Low-Grade Glioma (OPTIMUM) and focused on persons with recurrent LGG in the LGG Registry, was funded by the NCI as part of the Cancer Moonshot Effort and PE-CGS network [33].

To date, the LGG Registry includes more than 700 registrants from around the world. Most participants define themselves as White, of non-Hispanic ancestry, and of relatively high educational and economic status. To reduce disparities in access or willingness to engage in LGG research, the LGG Registry has developed partnerships with people living with LGG, care partners, and experts [34,35]. As it remains unclear which engagement methods are the most effective, the LGG Registry explored 3 methods for engagement of the LGG community in genomic research: a research advisory council (RAC) and online discussions via Facebook (Meta Platforms, Inc) and Twitter (X Corp) social media.

The purpose of this study was to compare engagement processes and outcomes among 3 methods of engaging people living with LGG, their care partners, clinicians, researchers, and others in the planning, conduct, and dissemination of genomic research.

Funded as part of the NCI’s PE-CGS Network, OPTIMUM aims to enroll into the LGG Registry people diagnosed with LGG who have had 2 or more surgeries for glioma. OPTIMUM’s Engagement Optimization Unit—a required PE-CGS center component—aims to identify effective and feasible strategies for engaging people with LGG and others in the planning, conduct, and dissemination of genomic research. The OPTIMUM engagement optimization unit’s primary goal is to identify effective strategies for engaging the LGG community in LGG genomics research to inform LGG Registry recruitment, data collection procedures, and the return of results. We conducted an exploratory quasi-experimental multimethod study to compare 3 engagement methods; participants were not randomly assigned to the engagement method (but instead had voluntarily chosen their assignment), but we did structure engagement activities to be as similar as possible. Our level of assessment was the engagement method itself. We used project tracking documents, audio recordings, data gathered from social media platforms, and surveys to evaluate reach, engagement experience, and trust in research. Permission was requested and granted for publishing direct quotes reported in this paper from identifiable individuals.

An engagement method refers to “a set of tools, techniques, and processes that are used to enact all of the ‘high-level’ purposes of engagement: identify and convene partners, create reciprocal relationships (level the playing field), engage in bi-directional communication, elicit perspectives, and make decisions over time and in partnership” [36]. We compared 3 engagement methods for building relationships and gathering community input to inform research [37-40]. Engagement methods included facilitated discussions with the LGG Registry’s RAC; “tweet chats” in collaboration with the #BTSM (brain tumor social media) community (established in 2012; monthly tweet chats started in 2013) on Twitter; and interactive Facebook posts with the Oligodendroglioma/LGG Warriors (henceforth “Warriors”) private Facebook group (established in 2013), which includes primarily people living with LGG. Each engagement method involved 4 parallel engagement activities with each group, in the form of interactive discussions facilitated by the research team about topics relevant to the conduct of LGG genomics research.

Both social media groups existed before this project and were established by members of the brain tumor community, not the research team (Table 1). The RAC was established by the research team as part of the OPTIMUM project to inform optimization of LGG Registry recruitment, enrollment, and return of results strategies. RAC members were originally recruited in early 2022 from research team personal contacts, the LGG Registry contact list, and social media. The RAC consists of 25 people (including 19 people with LGG) purposefully selected to represent a range of community and scientific perspectives and to be demographically diverse (Multimedia Appendix 1). The RAC met once for an introductory call in February 2022 before the structured engagement activities described in the subsequent sections. Our overall goal was to minimize differences in how each parallel engagement activity was conducted across groups. All 3 methods were established at the time of our study; that is, the RAC, “tweet chats” with #BTSM, and Facebook “topic of the day” discussions were existing methods of engagement. First, for each engagement method, we held facilitated discussions on the same four topics: (1) trust and benefits of genomic research, (2) registry recruitment, (3) registry data collection, and (4) return of results. All engagement activities occurred during March through September 2022. Second, for each topic, we developed a facilitator’s agenda with near-identical prompts tailored to the engagement modality and population (Multimedia Appendix 2). There were typically 3 to 4 discussion prompts for each topic, aligned with 4 orienting research ethics concepts: autonomy, privacy, ownership, and relevance.

It is important to note the differences that may exist between these 3 methods that may create differences in the experience of participating in a given engagement activity. For instance, the Facebook Warriors group responded to prompts asynchronously, whereas the RAC and Twitter engagement activities occurred synchronously; this was done to respect the existing structure of the Warriors group, which included a “topic of the day” with most responses written within 24 to 48 hours. In addition, the free-flowing, unpredictable nature of these semistructured group discussions meant that impromptu prompts occurred in response to the discussion and were thus not the same across methods (as would be the case if we had used rigidly structured engagement activities). To facilitate the discussion, facilitators for each engagement activity prompted participants to clarify in their answers which prompt they were addressing in their answer.

RAC members consented to audio recording to allow for analysis. Tweet chats and Facebook posts included “transparency notices” indicating that the content of the discussions would be used to inform research priorities for the LGG Registry and that anyone who did not wish to be included in the analysis should not participate.

We manually tracked attendance in RAC meetings. Tweet chat participation data were generated using Symplur, a health care social media analytics company. Symplur provides downloadable spreadsheets with the total number of Twitter accounts that tweeted a particular hashtag in a given time frame, the number of tweets and mentions for each account, user-reported location, and Symplur-identified stakeholder category for each participating account. We manually counted the number of unique individuals participating (commenting or reacting to a Facebook “topic of the day” post) in each Facebook activity.

To describe the general characteristics of engagement activity participants, we used several data sources. RAC members completed a survey as part of the application process in which they self-reported race, ethnicity, gender, income, education, stakeholder type, type of LGG diagnosis, insurance type, and US geographic location. For the #BTSM community, we used Symplur data (which include stakeholder type, eg, patient, clinician, and care partner) and self-reported geographic location for all Twitter accounts that participated in @BTSMchat-hosted tweet chats on March 6, 2022, April 3, 2022, June 5, 2022, and August 7, 2022. Each “live” chat lasted 60 minutes in length, although due to the asynchronous nature of Twitter, participation data include tweets posted using the #BTSM hashtag during the chat and up to 12 hours after each live event; for the first chat only, we also included chats in the 12 hours before the live event. For Facebook group members, the administrator for the Warriors group provided group-level demographics (age, gender, and US-based) generated using Facebook’s Page Admin interface.

We used qualitative content analysis to compare the 3 engagement methods in terms of (1) the nature of the interactions between community members and the research team during engagement activities and (2) recommendations to the LGG Registry. All RAC meeting discussions were audio recorded and then professionally transcribed by a professional transcription service. Twitter transcripts were generated by Symplur, which included all tweet content, the associated Twitter account, time and date of the tweet, and data about participant demographics and stakeholder type (eg, patient, clinician, and care partner). Facebook transcripts were created through screenshot and PDF creation of all comments and reactions to the “topic of the day” posts. All transcripts were uploaded to the qualitative data management software, ATLAS.ti (version 23; ATLAS.ti GmbH).

Qualitative analysis was conducted by a team experienced in qualitative data analysis. Coding and analysis were conducted by 2 data analysts (CR and SGH) and 1 qualitative methodologist (JGB). The analysis was immersive and iterative, beginning with data collection and involving multiple passes through the data to identify deductive and inductive codes to represent the discussion topics (experiences, concerns, and interests of participants). To create the codebook, an inductive approach allowed ideas to emerge [41,42], and deductive codes were added based upon the discussion prompts. Content analysis was completed within and across engagement methods. Preliminary findings were identified for each method, followed by a comparison of the results to identify themes. In detail, rounds of team-based coding were completed by engagement method (in order: RAC, Twitter, and then Facebook) until all 3 engagement methods were fully analyzed across all of their engagement activities. Coders only coded data as linked to a prompt if it were clearly a response to it; data that were not clearly in response to a prompt were coded inductively. Then, the qualitative team compared the preliminary results by engagement method to identify similarities and differences between methods to determine the qualitative results (themes).

The qualitative team met regularly to debrief, refine the codebook, and ensure the codes were applied similarly across coders, thus helping in establishing trustworthiness of the analysis and results [43]. After initial coding calibration, team members double coded 80% of the transcripts to maintain calibration. To further ensure analytic rigor and reliability, the research team engaged in member checking by sharing findings back with participants to see if initial results reflected their experience and capture any missing important discussion points [44]. A reflexive framework guided all aspects of analyses (framing of the analysis, assigning codes, and emerging interpretations of data into themes) [45]. In the final analytic stage, we assessed similarities and differences in the nature of interactions and recommendations for building trust in genomic research and enhancing participation in the LGG Registry among the 3 engagement methods [35,36,38,40,44].

We chose to evaluate the engagement methods as both a “state” and a “process” [46]. This meant evaluating engagement activities (by having participants fill out surveys after an engagement activity and being specifically told to evaluate that activity) and the overall process of engagement after all 4 activities for each method (in this case, using the Patient Engagement in Research Scale [PEIRS-22]). To do so, at the end of the first and last engagement, participants were invited to complete an online survey evaluating their experience via Qualtrics (Qualtrics International Inc), with a target sample size of 20 survey respondents per method per administration (Multimedia Appendices 3 and 4). Surveys are reported in a manner consistent with the CHERRIES (Checklist for Reporting Results of Internet E-Surveys) checklist for web-based surveys [47] (Multimedia Appendix 5). RAC members received an email invitation via Qualtrics. For Facebook and Twitter, the group administrators posted an announcement about the opportunity to participate in a survey, indicating interested participants should direct message the study lead to receive a link to the survey. Engagement participants also received direct messages from the study lead or group administrator inviting them to complete the survey. Participants received a US $20 gift card for each survey they completed (earning up to US $40 if they completed both surveys).

The postengagement surveys included measures of engagement experience, trust in medical researchers, recollections of which LGG Registry team members were involved, self-reported costs to participate, and demographics. Engagement experience was assessed using 3 established survey measures and some de novo items. The Public and Patient Engagement Evaluation Tool (PPEET) [48] assesses engagement experience, processes, and perceived outcomes of engagement. The 9-item condensed Research Engagement Survey Tool (REST) [49,50] assesses 8 engagement principles based in community-based participatory research, including partner input, capacity building, equity, and trust among partners. The PEIRS [21] is a 22-item survey that has been validated, shortened [51], and even translated into other languages [52,53]. We chose the PEIRS-22 additionally because its items appear capable of evaluating discrete engagement activities (in contrast to the REST, which assesses more long-term, community-based partnerships). PEIRS-22 assesses the overall meaningfulness of engagement on a scale of from 0 to 100 and includes 7 subscales. These subscales include procedural requirements (ie, 7 items assessing team introductions, opportunities to contribute, ability to perform tasks, participation in decisions, receipt of updates, clear communication, and participants’ assessment of time, all on 5-point Likert scales), convenience (ie, 3 items assessing convenience in participating), contributions (ie, 3 items assessing engagement activity and participants’ perceptions of their contributions), team environment and interaction (ie, 2 items assessing perceptions of the team), support (ie, 2 items assessing support to participate), feeling valued (ie, 2 items assessing how participants thought they were valued), and benefits (ie, 3 items assessing how engagement activity participants benefitted from the experience). The 4-item Trust in Medical Researchers Scale was used to assess trust [54].

Given the low expected sample size for the surveys, these analyses were considered exploratory. Participation data and survey demographic data were analyzed using descriptive statistics (counts, frequencies, and percentages). To our knowledge, no standardized scoring system exists for the PPEET items, which range from “strongly disagree” to “strongly agree” on a 5-point Likert scale; consistent with standard survey practice, these were dichotomized, and the percentage of participants who strongly agreed or agreed with each item was presented. PEIRS-22 subscales were dichotomized using the cutoff values used by the scale’s designers during its validation [51], with respondents with scores indicating low meaningfulness contrasted with those reporting the engagement was moderately, very, or extremely meaningful. For categorical data, tests between groups within each time point were performed using the chi-square test; Fisher exact test was used in cases of low expected cell counts. Continuous survey responses were described using medians and IQRs when the distribution was highly skewed (ie, PEIRS-22 overall scale and REST) and using means and SDs when normally distributed (ie, Trust in Medical Researchers Scale). Comparisons between the 3 engagement methods at each time point were made using Kruskal-Wallis test and 1-way ANOVA. All statistical analyses were performed using SAS (version 9.4; SAS Institute), and P<.05 was considered significant.

This study was approved as exempt human participants research by the Colorado Multiple Institutional Review Board (protocol 20-1001). Participation in engagement activities and analysis of engagement discussion transcripts was not considered human participants research and did not require informed consent. Nonetheless, social media posts included transparency notices, indicating that the activity was being led by a group of researchers and the discussions held may be used to inform research. Only engagement experience surveys were considered human participants research. Survey participants received an information consent before completing the surveys. They were compensated with gift cards worth US $20 for survey completion. All survey data were stored in secure folders accessible only to the research team to protect privacy and confidentiality. Documentation of consent was waived.

All 25 RAC members participated in at least 1 of the 4 engagement activities reported here (n=24, 96%; n=21, 84%; n=21, 84%; and n=22, 88% participants in RAC meetings 1-4 respectively). There were 197 unique people who participated in tweet chats (n=72; n=37; n=81; n=77 participants in chats 1-4, respectively). Tweet chat participants were a mixture of community representatives, with 45% identified by Symplur, a social media analytics platform, as people with brain tumors (including, but not limited to, LGG). There were 133 unique people who participated in Facebook Warriors page discussions (58, 17, and 88, participants in Facebook post series 1, 2, and 4, respectively [post series #3 excluded; refer to the footnotes in Table 1]). On the basis of Facebook Page Admin statistics, 76% of the group members are aged between 25 and 54 years, 73% identify as women, and 65% live in the United States. Among those people who participated in the engagement activities, there were 92 completed postengagement experience surveys (52 postengagement 1; 40 postengagement 4; refer to Table 2 for survey respondent characteristics). As surveys were anonymous, it is not known how many of the survey respondents were the same across administrations. While not statistically significant due to the small “n” involved, the RAC appeared to have greater representation of those of Black or African American race and Hispanic ethnicity, as well as a more equal self-identified gender balance.

Results of the engagement experience surveys (n=52, postengagement 1; n=40, postengagement 4) are shown in Table 5 (PPEET), Table 6 (REST and Trust in Medical Researchers Scale), and Table 7 (PEIRS-22; postengagement 4 only). The PPEET items showed positive ratings of engagement experience overall for RAC participants, where all items were endorsed by at least 70% of respondents; ratings for social media methods were more variable, with the lowest ratings reported by Facebook engagement activity participants, where several items were endorsed by only 50% to 70% of participants. As shown in Table 5, there were statistically significant differences in engagement experience for several PPEET items: belief that participation was diverse (eg, at time point postengagement activity 1: RAC: 100%, Twitter: 91%, and Facebook: 58%; P=.004), belief that a wide range of views were expressed in engagement 1 (RAC: 84%, Twitter: 91%, and Facebook: 50%; P=.02), and belief that the activity achieved the stated objectives (RAC: 100%, Twitter: 81%, and Facebook: 67%; P=.03).

P values obtained from the Kruskal-Wallis test for REST and 1-way ANOVA for TMR. P value considered statistically significant at <.05.

There were no statistically significant differences between engagement approaches in the overall meaningfulness of engagement, as assessed by the REST or PEIRS-22 (Figure 1A). However, RAC members rated PEIRS-22 procedural requirements more positively (Figure 1B) and reported recalling a wider range of expressed views than social media participants (Table 5); only 6% (1/16) of RAC members assessed the procedural requirements as “low” in terms of meaningfulness, compared to 55% (6/11) of Twitter participants and 50% (1/2) of Facebook participants (P=.01).

This study was designed to compare advisory council versus social media–based methods for engaging patients and other community representatives in the planning and conduct of LGG genomic research. To our knowledge, this was the first study designed prospectively to evaluate differences between advisory council and social media–based research engagement methods. Qualitative analyses revealed minimal content differences in the insights generated by each engagement method; recommendations for how the OPTIMUM project may improve trust, promote participation in LGG genomic research, manage data collection, and return individual and research results were similar across engagement methods. However, the 3 methods exhibited differences in the number and types of people engaged, the nature of interactions between researchers and engagement activity participants, and how participants experienced the engagement process.

We observed differences in the PEIRS-22 procedural subscale between the 3 engagement methods, with the RAC rating engagement procedures more favorably; no differences were found in the other subscales. For the other subscales, this may be because the participating members of the research team were consistent across engagement methods (meaning differences in the team and environment were mitigated), all methods were considered convenient for individual participants (who volunteered to participate), and so on. For the procedural subscale, the fact that a RAC comes with an expectation of a standing advisory committee over time could mean that participants are more likely to feel introduced to the research team, have opportunities to contribute, and participate in decisions—all core elements of that subscale.

Importantly, we do not conclude that a RAC is inherently “better” than other methods; instead, we aim to highlight relevant differences. A RAC may provide consistent engagement with the same individuals, a relatively simple structure, and the ability to purposively select members for diversity. However, a RAC may not have the reach of social media engagement, which can engage more diverse individuals, and social media engagement may be less labor or resource intensive for learning important insights (as our study found similar qualitative themes). However, our findings suggest this is not a given; social media did not necessarily yield more diverse participants, and additional efforts may be needed to promote diversity and inclusivity for those methods of engagement. These relevant differences inform selecting the type of engagement activity based on the research question and population. For instance, a research topic that requires a higher level of scientific understanding may best plan to engage Twitter and RAC communities, as the quantity and quality of Facebook data may be limited. Likewise, Facebook and RAC engagement may better suit a research question best understood through narrative interaction between participants, as these engagements were where we observed participants organically responding to one another.

As expected, the advisory council method represented fewer total individuals than those engaged through social media. We had aimed to broaden both total numbers and the representativeness of those participating by using social media methods. Contrary to expectations, our results show that the RAC was more diverse in terms of race and ethnicity than those engaged through social media. The OPTIMUM project team translated findings from this study into several types of decisions. First, the results informed how we would continue to engage the LGG community throughout the remaining conduct and dissemination of the research, with both RAC and social media engagement continuing throughout the project period, although the RAC meets more regularly and is more involved in all aspects of study conduct and dissemination. Second, the results informed content for recruitment messages and channels (eg, an emphasis on recruitment via clinical settings rather than through online channels), content for the LGG Registry website and social media pages, and decisions to allocate more project resources to partnerships with additional clinical sites. The need for mitigating data and specimen collection burden—especially regarding access to electronic health records, pathology reports, and tumor samples—was a key insight.

Our study adds to the literature on the science of engagement in important ways. Although research shows that engagement can influence which questions are asked, how studies are conducted, and how findings are shared with relevant communities [1-12], few studies examine which specific engagement methods are most effective for particular purposes or populations. Prior studies have examined the effect of panel composition (eg, Delphi panels [55]) and compared the experience of patients versus researchers within a single method [26]. One previous study used surveys to examine online versus in-person focus group engagement regarding rural health care (not research); it found greater satisfaction among in-person participants and a lack of representativeness [27]. Others compared online voting, in-person focus groups, and mailed surveys in a low back pain data registry; using a qualitative evaluation, results showed that all methods generated similar research priorities but a better experience among in-person focus groups [56].

Unlike these previous studies, we used both qualitative and quantitative data (validated surveys and scales) prospectively to compare the 3 methods of engagement. Similar to these previous studies, we found few differences in the content of recommendations, but we did find differences in communication styles and how recommendations were expressed. Importantly, we found few quantitative differences in participants’ overall assessment of engagement methods or their overall trust in research. Together, these findings suggest that all 3 methods may have a place in OPTIMUM LGG genomic research engagement. Experience may be optimized when participants are able to choose methods with which they are comfortable.

Our results have implications for how to engage community in research more effectively and hypotheses for future research. Participants in the advisory council rated the procedural elements of engagement (eg, proper introductions, opportunities to contribute, bilateral communication, and whether the activity is worth one’s time) more highly. If social media engagement included more informal or unstructured time, these findings might change. Unexpected results showed that our RAC members were less likely to report feeling their views were heard. This finding could reflect the opinions of council members who felt uncomfortable speaking, the existence of more outgoing and outspoken personalities in a group, or different expectations among social media participants about what it means to feel heard. Third, despite the arguably broader reach of social media, advisory council participants perceived greater diversity of views. This could reflect the intentional recruitment efforts that RACs often involve, as ours did, or the persistence of a digital divide.

As a pilot exploratory study of the comparative effectiveness of engagement methods, we interpret our findings cautiously but can still offer several recommendations for researchers. First, our study demonstrates the feasibility of conducting an evaluation of engagement processes alongside engagement activities and the ability to detect differences even with relatively small numbers. Related to this, we found the engagement survey tools and items we used to be appropriate for these methods and this context, although adapted tools and items may be needed for other communities. Second, although qualitatively we found no differences in the content of recommendations among groups, we do not suggest they are therefore equal. Issues of cost, time, and expertise, for example, are likely to matter; a researcher already familiar with advisory panel methods is likely to do a better job with that method compared to social media formats, for example, and not all communities have active online communities.

Third, our findings suggest that the presumed advantages and disadvantages of each method are not fixed or guaranteed. Engagement via online social media sites did not inherently appear to increase reach or diversity; researchers may need to take extra steps to identify and recruit for diversity or use other methods to fill these gaps. Likewise, an advisory panel—despite engaging a small number of individuals—did not necessarily mean all voices were heard; a researcher will still need to use techniques within meetings (eg, use of a talking object that is passed around a real or virtual room) to allow all individuals an opportunity to contribute. Finally, a robust, comprehensive, and comparative assessment of engagement methods in different contexts and for different purposes is still an unmet need, particularly for underrepresented research groups and groups considered marginalized. Further research is needed to improve and validate engagement measures used (eg, in languages other than English). Patient partners (coauthors of this paper) emphasize that the results of this study highlight opportunities to engage the brain tumor community not only in genomic research but also in a wide range of other priority areas, such as quality of life research.

The principal limitation of this study was that this analysis was secondary to the primary purpose of the engagement activities—which was to establish bidirectional communication and build relationships with the LGG community. The fact that we used existing engagement structures via the RAC, Twitter, and Facebook meant that participants self-selected into engagement methods that may have matched their preferred style of communication. This approach is respectful of engagement principles; that we observed statistically significant differences and meaningful qualitative differences between groups is thus still important, and future research should explore the possibility and permissibility of randomized designs. Moreover, although we sought to evaluate the engagement state after specific activities in engagements 1 and 4 as well as the process from beginning to end using the PEIRS-22, the wording of items and participants’ own perceptions of them can be subject to interpretation. This too requires further research.

In addition, the data used for qualitative analysis were not gathered using standard qualitative research methods, such that the usual ability to follow up with a respondent to ask for clarification or more information about a statement was not present. The high degree of trust in medical research reported in postengagement surveys suggested that we failed to engage community members with less trust in research. The resulting recruitment messages may therefore underemphasize key points needed to enhance trust. For instance, those engaged suggested that noting the involvement of respected research institutions and naming scientists involved would enhance trust in the LGG Registry, but that might not be the opinion of those without previous good experiences with academic institutions. Although we partnered with the community in engaging Facebook participants, other strategies—different Facebook groups, different social media platforms, or nonsocial media–based engagement—may be required to gather more diverse perspectives.

Engagement of patients, families, and other community partners in research is an ethical imperative. To do so with authenticity requires that we evaluate engagement methods for their effectiveness across diverse contexts, and we must hold our engagement methods to the same rigorous standard of evidence as our research methods. Our study demonstrates the feasibility of comparative evaluation of engagement methods that can further inform engagement approaches. Future research should examine additional methods comparatively in different research settings and communities and for different purposes.