The CO-CEAL multi-year study began by finalizing the health topic on which to focus for each of our partner communities via listening sessions and kick-off meetings. Sessions were held in person and virtually (e.g., Zoom) and carried out in English, Spanish (directly or with interpretation), and Somali based on the community’s preference and co-facilitated by the project multi-PI and Community Connectors. We held these sessions in each partner community with several purposes in mind. First, we returned important results from several of the prior Common Surveys conducted in each community. Common Surveys have been, and continue to be, collected by all CEAL Regional Teams. Prevalence of chronic illness was a main focus in Common Survey 3 and each community’s results were shared with them, respectively. We also shared our longitudinal analysis of Common Survey waves 1–3 which showed a significant impact on COVID vaccination rates through exposure to the BCT messaging. Second, we shared the plans for the various phases of our multi-year study with each community to increase interest and engagement, and gave individuals an opportunity to reflect on how they wished to participate in the various phases. Finally, using the Common Survey 3 results, data from the Colorado Equity Compass¹ and health priority survey data from December 2023/January 2024, we confirmed with each community their chosen priority health topic of focus for the multi-year study.

Invitations to the kick-off/listening sessions were disseminated as broadly as possible with the support of our Community Connectors and prior BCT facilitators and participants, when appropriate. Word of mouth and social media were used to ensure as broad of representation as possible from each partner community.

After the community kick-off/listening sessions, GMB (10–12) was carried out in each partner community to map and identify the key factors contributing to the priority health topic as well as leverage points to be developed into an intervention. GMB is a participatory process that brings together eight to ten individuals from various sectors of the community with lived experience to surface factors as well as the inter-relationships that contribute to a problem of interest. Over a series of meetings, workgroup participants discussed the origins, underlying causes, and possible solutions to a problem of interest through guided group activities. The process resulted in each community workgroup developing a visual diagram that showed how different factors are connected to the prioritized health topic and guided the collaborative identification of actionable areas to intervene through an impact and feasibility prioritization process. This process involved the GMB teams reviewing evidence for interventions when selecting a particular factor for intervention. Later in the project, representatives from all GMB workgroups will come together to prepare a policy brief that summarizes the health priorities and proposed areas for intervention, especially highlighting recommendations for addressing structural factors that contribute to disparities in chronic disease and communities’ ability to resolve them.

Community Connectors helped recruit prospective GMB participants and co-facilitators, which were finalized with support from the Community Engagement lead and GMB academic team. Most (four of five) community workgroups used a co-facilitation model that paired an academic team member with a community member and had a community co-facilitator assistant to provide support (e.g., meeting logistics, notes, etc.). One community workgroup was led by two community member co-facilitators who also shared the assistant duties. The community chose this approach due to cultural and linguistic preferences and norms, and consulted with the academic GMB team, as needed. GMB facilitation training occurred both before formal GMB sessions began and between the meetings to allow for adaptability and feedback during the process. Based on the community’s preference, GMB meetings were convened in a combination of in-person and virtual sessions or fully virtually. Duration of the GMB process in each community ranged from 5 to 6 months. To bolster engagement and be mindful of various factors, the research team, in collaboration with Community Connectors and GMB facilitation teams, considered the participants’ needs and concerns with regards to meeting timing and venues that would be accessible and safe, and prepared for and debriefed from meetings pertaining to sensitive topics.

BCT constitutes the next phase of work in each partner community. The BCT groups will use the BCT process to devise or transcreate an existing intervention or interventions to address the factor(s) identified in the GMB process as key leverage points. These interventions could be similar to the community specific, culturally relevant informational messaging utilized successfully in our first three CEAL funding rounds, newly designed interventions, or they could be community specific, culturally relevant adaptations of existing evidence-based interventions.

In contrast to the BCT process used in our first three CEAL funding years, we will employ a traditional BCT process over 6–9 months rather than a rapid 8–10 week process (2). This longer BCT process will allow ample time for participants to learn about and refine the results of the GMB process and to design or adapt an existing intervention for their community. Participants in BCT function as research partners rather than research subjects and are the driving force behind the development of a product, message, and/or intervention that “works” for the intended users and communities. The BCT teams will also co-design the implementation and provide input into the Ecological Momentary Assessment (EMA) measures being used to evaluate the transcreated interventions.

The facilitation model for BCT will be similar to GMB, pairing an academic and community facilitator co-facilitation team and having a community co-facilitator assistant. BCT facilitation refresher training will be offered for the facilitation teams, along with longitudinal coaching of the teams. BCT co-facilitators and BCT assistants will be recruited in a variety of ways, including recommendations from Community Connectors and the Community Engagement lead, as well as, when suitable, re-engaging former BCT facilitators who continue to be a good fit for the role. BCT participants will be recruited in a collaborative manner with Community Connectors suggesting potential community partners to whom facilitation teams can reach out to invite. Additionally, former GMB participants may also be approached for participation. Each community will ideally have 12–14 community participants.

The CO-CEAL multi-year study includes an overarching advisory group to provide input from individuals from our communities and partner organizations that may not be intimately involved with aspects of the project. The advisory group will be comprised of 9 members representing key groups of influence including our 5 partner communities, the Trailhead Institute’s Regional Health Connectors, Rocky Mountain Prevention Research Center, The Care Collaborative (formerly PMCC/ECHO Colorado) and the Colorado Clinical and Translational Sciences Institute PACT Council. This group will meet quarterly with the multi-PIs and project staff to review and provide input on the project goals, operations and progress.

CO-CEAL’s various individual and organizational community partners and their roles are described in Table 3.

CO-CEAL plans to recruit 500 respondents across all participating communities, with sample sizes relative to the community’s size. EMA respondents will be recruited via Community Connector data collectors, as CO-CEAL has successfully done for past Common Surveys. Potential participants will be given information about the project (e.g., in person or via social media) and then complete an initial eligibility survey, followed by an online consent if they are eligible. Reloadable gift cards will be used to compensate EMA respondents $5 for each one-to-two-minute survey completed over a 1 week period. This dollar amount was determined given the brevity of the survey and aligns with our former Health Knowledge Monitoring and Response System (HKMRS) pilot project that had a similar survey structure (iHeard Colorado²). Power calculations are based on anticipated change in outcomes/behaviors both aggregated across all communities and for individual communities, as well.

Common Survey participants will be recruited as we have done in prior waves, utilizing Community Connector Data Collectors who have deep connections into our partner communities with known, trusted relationships. As in past waves, CO-CEAL anticipates recruiting 1,000 respondents for the first wave during the multi-year study and then returning to these same respondents in the second wave in a cohort approach. Participants will be compensated $50 per survey.

With input from the Community Connector teams, the research team will contact up to 25 facilitators and participants (approximately five from each community) from GMB and BCT to participate in individual interviews to understand their experiences and gain an in-depth perception of the group processes. We chose this sample size to ensure representation from a variety of roles and community perspectives and to achieve thematic saturation. We will interview participants until thematic saturation is reached (13, 14). Participants will be compensated $40 per interview.

EMA/mHealth technology tracks behavior across time and context, which allows for identifying momentary factors that may be associated with key outcomes. In the current study, we will employ EMA to assess the uptake and impacts of our interventions on communities at the individual level. EMA is a survey-based technique that is distributed via mobile, phone-based surveying. We have developed experience with weekly, mobile, phone-based surveying in iHeard Colorado. Based on EMA best practice (15), we will use signal contingent (15–17), EMA messaging that utilizes text messages with short surveys sent randomly throughout the day that participants respond to allowing for capturing individual behaviors, feelings, attitudes, and contextual factors (e.g., accessibility) as they unfold across time and context. We will assess key outcomes related to behaviors, access, etc. based on the community identified intervention targets from our GMBs and BCTs in each community. According to best practice, the EMA observation period for each participant will be 7 days, with 4–5 text messages sent throughout the day. This will allow for identifying patterns within and across day of participant outcomes of interest to the intervention. These assessments will be collected, as appropriate to the intervention target, up to daily during defined weeks during Years 2 and 4, prior, potentially during, and post intervention implementation and dissemination.

We will add similar, yet static measures from the EMA in each community to the Common Survey to assess exposure and changes in response to the interventions. Items to assess exposure to the interventions in each community will assess message awareness and cultural adaptation effects from the interventions designed in BCT. In prior Common Survey waves, we successfully used images of the transcreated intervention materials in each community to assess exposure. Exposure and awareness will be assessed via one adapted item which asks, “Have you recently seen a [priority topic] message on [distribution channel(s)] that shows [brief message description]?” with responses of, “Yes, definitely; Yes, probably; No; Do not know.” Additionally, the communication materials developed by the community will be evaluated for effectiveness using items related to knowledge and attitudes about the chosen health priority. We will also assess cultural adaptation through participants’ rating of the communication materials including perceptions of trustworthiness, personal relevance, personalization, cultural fit, appearance, and information—using a previously developed measure of cultural sensitivity. Items will be adapted from Singelis and colleagues’ measure of their two-dimensional theory of cultural sensitivity in health communication (18). The initial Common Survey wave will be administered prior to the intervention with items relevant to communities’ priority health topics and anticipated transcreated interventions in each community. The follow-up Common Survey will be administered in Year 4.

CO-CEAL will conduct community key informant interviews, gather field notes and hold debrief conversations with participating community team members from GMB and BCT. These efforts will offer insight into participants’ experiences in the processes and identify areas of improvement. Interviews will also help us understand the fidelity of the GMB and BCT processes and the level of cross-sector collaborations and community capacity building that was fostered.

CO-CEAL utilizes an NIH-provided survey called the Common Survey for the recurring pre/post standardized community surveys. CO-CEAL includes additional questions relevant to the intervention’s health topic and created materials. For the multi-year study, CO-CEAL will use Common Survey 4. Surveys will be administered by Qualtrics secure online survey system, which can be taken via mobile application on a phone, tablet, or computer; if not Qualtrics, EMA surveys will be administered via another secure platform. Identifying information will be collected to ensure ability to retain participants and conduct follow-up surveys; however, once collected identifiable information and survey response data will be stored separately. Surveys completed online via Qualtrics remain confidential and secure and will be accessible via login and password to research team members. Common Surveys may be conducted in person, by mail, online/video call, or phone with a Community Connector data collector while EMA surveys will use a text/SMS or similar platform.

Survey participants will be consented electronically in Qualtrics. Postcard consent will be obtained even with face-to-face with assistance from a member of the study team. If requested, participants will receive a written consent document and additional verbal consent will be assessed. Completion of the survey will then indicate consent.

If a Community Connector data collector collects survey response data by phone or video call, they will first verify the identity of the respondent and then enter their responses directly into the Qualtrics system to ensure the security and confidentiality of participants. Mailed surveys will be returned directly to the University of Colorado. No identifying information will be included in mailed surveys, only a study ID. For all aspects of the study, only members of the research team, including Research Associates, analyst, and Principal Investigators will have access to subjects’ identities/identifiers and audio-recordings. All such staff members will be asked to sign confidentiality agreements, agreeing not to reveal the names or other identifiers of study participants to anyone outside the research team. All personal contact information will be erased/destroyed once their participation in the research is completed.

Interview participants will be consented verbally and provided with a postcard consent. Interviews will be conducted via phone or virtual platform (e.g., Zoom). All interview data will be stored on a secure server. Audio recordings will be kept on the secure server until a transcript is created. Transcripts will be redacted of all identifiable information. Once a transcript is created, the recording will be deleted. Identifiable interview data will be kept confidential and will only be accessible to the study team. Deidentified interview data, in the form of redacted transcripts, will be shared with funders via a secure web-based platform.

All data will be synthesized for reporting purposes and no identifiable information will be shared in any reports. All members of the research team will keep all identifiable survey data confidential within the research team.

Though EMA uses longitudinal data trends, it is reasonable to assume the completeness and fidelity of those data may include a substantial amount of missingness (e.g., individuals may miss certain EMA data entry points), particularly in non-clinically controlled settings. Because of this, we use a more conservative F-test (via a mixed-ANOVA) for power calculations to account for the potential of substantial within-person missingness which allows us to compare both between groups (one community compared to another) within groups (change over time within a single community) and potential interactions.

Assuming 95% power, at least four time points, an intraclass correlation of 0.4 (also a conservative estimate), and six groups with unequal sample sizes (our latest sample sizes range from the smallest group of consisting of 4% to the largest group comprising 27%), we estimate that 426 individuals are needed to detect differences using a small effect size (Cohen’s d = 0.1). Using a final sample of 500 individuals allows for an approximate 15% attrition rate (a recent meta-analysis on 30 EMA studies estimated an average attrition rate of 17%) (19). We believe these calculations well insulate our study from potential missingness, attrition, and type II errors.

With respect to individual community level analysis, the above power calculation will also allow us to conduct a within groups investigation by community (e.g., detecting changes, relative to the outcome, within a single community). Given the smallest community in our previous work comprised 4%, we anticipate that given a target sample size of 500 will produce at least 20 individuals in our urban AI/AN community. Using the following assumptions of 95% power, intraclass correlations of 0.4, and 7 longitudinal data points (assuming we get at least one EMA data point per day), we would need 21 individuals in order to detect changes with a medium effect size (Cohen’s f = 0.3). While this gives us the minimum number of individuals for the smallest group, this also allows for more complex analyses with smaller effect sizes for the larger groups. For example, our largest group in our previous study comprised 27% of the sample, equating to a sample size of 135 in our 500 target sample. Given the same power calculation assumptions with the aforementioned AI/AN community, a sample of 135 would not only allow us to detect changes with a much smaller effect size (Cohen’s f = 0.12), but would also allow for more robust statistical testing for sensitivity analyses using covariates (e.g., age, education, income) and stratifications (e.g., changes among females, changes among those in certain age demographics, changes among those with certain health conditions).

EMA data presents complexities because of its nested structure: assessments (level 1) are nested within days (level 2), which are in turn nested within individuals (level 3). Given this nature, we plan to use the method described by Ma and colleagues which implements a linear- mixed effects model (LMM) with random effects at each level (20). This model analyzes both within-subject effects (e.g., longitudinal trends in an individual) and between-subject effects (e.g., community level comparisons). Both pre-intervention and post-intervention data will be used to analyze trends (e.g., slopes) in EMA measures across time. LMMs can also account for covariates within the model and are level specific (e.g., sex/age for level 1, day of the week, such as a weekend for level 2, etc.). All outcomes will be tested for normality, with adjustments made to skewed data (e.g., log transformations). Statistical significance will be assessed at the 0.05 level.

While LMMs will be used to estimate within- and between-subject trends in EMA outcomes, we will also employ interrupted time-series analysis to complement these models. Interrupted time-series is particularly suited for evaluating the impact of an intervention on outcomes over time by modeling level and slope changes between pre- and post-intervention periods (21). This approach is consistent with other EMA protocols in which interrupted time-series has been used to estimate causal effects in real-world, longitudinal settings. Lastly, beyond analyzing longitudinal trends post-intervention, we will also correlate EMA measures with common survey items to assess how changes in EMA trends are associated with social determinants of health, attitudes, and trust as well as exposure to any community level intervention. This can be done using the same LMM framework as described previously.

Given that communities will have the flexibility to choose tailored outcomes based on Common Survey 3, the above analyses will allow for flexibility in testing research questions given those chosen outcomes. For example, if a given community chooses a mental health indicator (e.g., the GAD-7 scale which assesses general anxiety disorder), the proposed analyses will allow for the testing of several research questions: (1) Are there changes/variability in anxiety levels in the week of data collection during the first assessment of EMA? (2) Are there changes/variability in anxiety levels in the week of data collection, post-intervention, during the second assessment of EMA? And (3) Is there a significant change in these slopes of anxiety, comparatively, from each EMA data assessment period (i.e., did the intervention between EMA periods significantly change the slopes of anxiety when comparing EMA1 to EMA2)? Other outcomes, depending on what individual communities choose, can be modified to fit these research questions to assess statistically significant changes.

As noted above, we will add equivalent measures from the EMA in each community to the Common Survey to assess changes in response to the interventions. Additionally, we will add items to assess exposure to the interventions in each community. In prior Common Survey waves, we successfully used images of the transcreated intervention materials in each community to assess exposure. The initial Common Survey baseline wave will be administered prior to the interventions and include items relevant to the priority health topics and anticipated transcreated interventions in each community.

The domains for the Common Survey include demographics, social determinants of health, information and trust. Aside from general descriptive statistics, we will conduct bivariate analyses comparing subgroup distributions (e.g., by language, community) to overall sample proportions or means, using chi-square or t-tests as appropriate. We will also create a data dashboard(s) to illustrate data on health behaviors, attitudes, and exposure to intervention materials.

We will conduct linear/logistic regression analyses (depending on the outcome) to explore the relationships between social determinants of health, information and trust and exposure to intervention materials on chronic disease outcomes. In our prior Common Survey samples we have seen sufficient power to observe significant odds ratios for health behavior differences across communities in vaccine uptake, as well as impacts of trust on vaccine behaviors. With an anticipated sample size that is similar in our future Common Survey samples we believe we will have sufficient power to observe pre-post differences in health behaviors from years 1–4.

The CO-CEAL multi-year study spans 4 years (Figure 1). The first year focuses on confirming health priority topics and project activities to set the stage for the remainder of the project. GMB spans the first 2 years to inform the intervention plans, followed by BCT to develop and/or modify existing interventions to address the health topics and leverage points identified in earlier stages. Surveys will be administered in year 2 (Common Survey) and early in year 3 (EMA) prior to interventions beginning. Community interventions will be carried out in year 3, followed by post-intervention surveys, data analysis, and other evaluation activities in year 4.