This study complied with the Declaration of Helsinki and received ethical approval from the University College London Research Ethics Committee (reference number 24235/001) on 19 June 2023.

Before enrolment, research participants received a participant information sheet, had the study procedure explained, and were informed of their rights to withdraw at any time. Written informed consent was obtained from all participants.

Data were pseudoanonymized using an alphanumeric code. The code list and consent forms were stored separately from the anonymized data in locked cabinets (paper) and on a password-protected, encrypted institutional drive (electronic), accessible only to the core research team.

Participants received shopping vouchers for their time: £20 (GBP £1=US $1.27 as of Jan 1, 2024) for 45-minute usability or focus group sessions (in Discovery, Define, and Design phases) and GBP £80 in total for the 1-week Deliver phase testing (GBP £25 for onboarding, GBP £25 for postsession feedback, and GBP £30 for app use), in line with National Institute for Health and Care Research (NIHR) payment guidance [39].

PPI and HCI experts participated in an advisory capacity and were not classified as research participants; written consent was therefore not required. PPI received £20 per hour reimbursement per NIHR recommendation [39].

We drew on established methodological frameworks to structure the development. The Medical Research Council framework for developing complex interventions guided the overall approach [40]. In line with this framework, we developed a logic model at the outset to specify ENHANCE’s core components, proposed mechanisms of action, and expected outcomes (Multimedia Appendix 1). We used the Double Diamond model to structure development activities across the Discovery, Define, Design, and Delivery phases, supporting iterative exploration and refinement of the intervention [41] (Figure 1).

As shown in Figure 1, during the discovery phase, we identified candidate components for each risk-factor module through an evidence scan and conducted usability testing of existing apps and materials to explore how older adults engage with them, informing ENHANCE’s design decisions. The Define phase translated these insights into guiding principles, informed by the person-based approach, to steer key design decisions [37]. This phase also produced detailed blueprints for risk-factor modules, cognitive training games, and coaching structure. During the Design phase, wireframes and prototypes were iteratively refined through usability testing until a stable, high-fidelity version was achieved, and a coaching manual was also designed. The Delivery phase involved a 1-week user-testing period to evaluate the feasibility and usability of the intervention.

The GUIDED (Guidance for Reporting Intervention Development) framework checklist for reporting the development process is provided in Multimedia Appendix 2 [42].

Coproduction involved an interdisciplinary research team (clinical academics, risk-factor experts, and app developers), research participants, patient and public involvement and engagement (PPI) contributors, and human–computer interaction (HCI) experts. Research participants and PPI provided lived experience and end-user feedback, while the research team and HCI experts contributed technical, clinical, and design expertise.

Research participants were community-dwelling UK adults aged 50 years or older with no severe impairment affecting basic tablet use. We used purposive and snowball sampling to include underserved communities. The research team identified recruitment venues in London and Staffordshire through food banks, religious centers, community centers serving underserved communities, and neighborhood shopping malls in relatively deprived areas, using the Index of Multiple Deprivation (IMD) [43]. Researchers visited sites, distributed leaflets, and screened interested individuals for eligibility. Additional participants were recruited through existing participants and community center staff.

PPI were adults aged 50 years or older, recruited via researcher networks, participants’ referrals, or prior involvement in dementia prevention research.

HCI experts in user experience and interface design were recruited through the research team’s own professional networks.

As coproduction primarily used usability testing, sample sizes followed established principles whereby 5-10 users identified approximately 85%-95% of usability problems [44,45]. Both participants and PPI were included in the calculation, as both contributed end-user perspectives as intended users of the app.

For the Discovery phase, 5 participants were required per risk factor area (5 × 9 module materials = 45). For cognitive training games, a larger sample of 50 was predetermined, anticipating 5-10 distinct game types each requiring separate evaluation (5 × 10 games = 50). This gave a total of 95 participants.

For the Define and Design phases, up to 3 iterative cycles were planned per phase, with a minimum of 10 participants per cycle (30 per phase).

For the Delivery phase, 10 participants were predetermined for a single 1-week usability testing iteration, identifying approximately 95% of major usability issues [44].

In the following, we report the coproduction methods according to the Discovery, Define, Design, and Delivery phases of the Double Diamond model.

The Define phase translated discovery findings into concrete design decisions for ENHANCE.

The Delivery phase aimed to finalize and refine the intervention through real-world use before feasibility testing.

Participants were recruited across 11 venues in London and Staffordshire, including food banks, community centers (serving individuals experiencing poverty, specific religious groups, and ethnic minority communities), and a shopping center. PPIs were recruited through researcher networks, many of whom had previously taken part in our other aging research projects, while HCI experts were recruited through our professional networks. In total, 199 unique individuals contributed to coproduction activities: 162 research participants, 33 PPIs, and 4 HCI experts. Participation spanned across 4 phases—Discovery (n=108), Define (n=54), Design (n=40), and Deliver (n=10) (Figure 1)—with some individuals contributing to more than 1 phase.

Participants had a mean age of 65 (SD 9, range 50-87) years, and 65% (106/162) were female; detailed demographic characteristics are presented in Table 1. The largest ethnic group was those who identified as Asian (78/162, 48%), followed by White (47/162, 29%), and Black (27/162, 17%). The mean IMD was 3 (SD 2, range 1-9). Most participants lived in socioeconomically deprived areas, with 79% (110/139) in the bottom 40% of IMD areas. Almost half (68/150, 45%) were retired, and over half (66/131, 50%) had not attained qualifications beyond General Certificate of Secondary Education (an exam taken at the minimum school-leaving age in the United Kingdom). English was the first language for around half (71/154, 46%), and only 44% (68/155) of participants stated that they felt comfortable or very comfortable using technology.

Although demographic data were not collected for the PPI or HCI experts, PPIs were generally female, aged older than 50 years, with diverse ethnic and educational backgrounds. HCI experts were primarily PhD students and postdoctoral researchers in their late 20s to mid-30s, specializing in HCI and app development.

We coproduced ENHANCE, a coach-supported digital intervention targeting modifiable dementia risk factors. This process resulted in an evidence-informed, theory-based, and user-centered intervention designed to reduce dementia risk, with particular attention to accessibility for underserved older adults. By prioritizing accessibility for groups at higher dementia risk—including individuals who are digitally excluded and often underrepresented in prevention research—the intervention was designed to be usable by a broad older adult population. Using a structured, multistage development process, we integrated evidence from the literature, behavioral theories, and coproduction activities with 199 contributors to develop ENHANCE. This approach generated detailed insights into the design and usability needs of underserved groups and informed the development of both a high-fidelity app prototype and a comprehensive coaching package. Together, these components form a complete intervention ready to be tested in forthcoming feasibility and full trials.

We developed ENHANCE as an app-based, coach-supported multidomain program targeting 10 dementia risk factors. The intervention includes 7 adaptive cognitive training games accessible to all participants and tailored risk-factor modules. Each module lasts 6 weeks and incorporates educational videos, check-in questions, and “Specific, Measurable, Achievable, Relevant, and Time-Bound” goals to support behavioral change and monitoring. Rewards include visual progress indicators, coach encouragement, and a virtual meadow that participants can plant and grow. Throughout development, scientific evidence, user feedback, and practical considerations informed design decisions.

First, 6-week modules enable early, observable progress and support habit formation, with habit strength typically plateauing within this timeframe [58]. Short-term improvements were designed to enhance motivation and self-efficacy, sustaining behavior change beyond the module period. This was further supported by PPI and participant feedback, emphasizing the value of early visible progress. For risk factors where substantial physiological change within 6 weeks is unlikely (eg, diabetes), modules target achievable self-management behaviors—building skills, establishing routines, and initiating health care contact—placing users on a path toward longer-term change.

Second, we embedded accessibility features throughout ENHANCE, shaped directly by coproduction activities with older adults. This addressed a gap in existing digital interventions where such features are rarely incorporated [27]. We prioritized images over text to reduce reliance on reading ability, avoiding color-dependent game design to accommodate users with color blindness or macular degeneration, subtitled all videos to support hearing impairment, provided adjustable volume, and paired all in-app feedback with both visual and audio cues. Delivering via tablet further supported easier navigation and reduced visual strain.

Cultural accessibility considerations included a nature-based visual theme found to be broadly acceptable across genders and cultures, video content representing diverse ethnicities and religious preferences (eg, food recipes), and exercise options ranging from seated to higher-intensity alternatives. The word bank of a word-based game was curated using culturally universal and commonly understood words. Number-based games were deliberately framed without any reference to “maths,” as feedback from our earlier user testing indicated that such framing was off-putting for some participants.

Digital accessibility considerations included phone number registration in place of email and a simplified login system without repeated password entry. Game design varied in type and difficulty to accommodate diverse educational backgrounds, gaming experience, and physical abilities, with motor demands kept deliberately low. But some accessibility enhancements identified during coproduction—including expanded language options and voice guidance with text-to-speech functionality—could not be realized within the current development scope and remain priorities for future iterations.

Third, we designed each module to address 1 risk factor at a time, supported by early usability testing in which participants preferred focusing on 1 manageable change at a time. This is grounded in the behavioral change model, which posits that tackling multiple changes simultaneously risks exceeding an individual’s capability [38] and the paradox of choice [59], whereby too many options increase cognitive load and decision fatigue. Focusing on a risk factor, therefore, supports clearer goal formation and sustained engagement.

Fourth, ENHANCE was designed to offer tailoring and choice, allowing participants to select which risk factors to address, in what order, and how—for example, choosing specific exercises and frequency. This gave users control and ensured the intervention felt relevant rather than prescriptive, supported by BCT evidence that engagement is greater when individuals choose their own method [38,51]

Fifth, we adopted a front-loaded human coaching model, informed by evidence that digital behavior-change interventions achieve higher engagement when paired with human support [36], and that intensive early contact builds confidence, skills, and familiarity with the intervention, supporting longer-term adoption [60]. The model was also designed to be scalable as coaches were nonspecialists with a variety of health care or related backgrounds, reducing reliance on expensive, highly trained clinicians. The brief fortnightly remote coaching contacts minimized the ongoing coach burden. The in-app messaging made the coach accessible and reduced reliance on scheduled contact. The structured manuals and training materials allow replication and fidelity of delivery. However, whether this coaching model is effective or cost-effective remains to be established through a full RCT, though evidence from similar coaching models in lifestyle interventions suggests potential [61,62].

In ENHANCE, we conducted extensive qualitative work with underserved populations to inform the refinement of our app. Our findings broadly align with those reported in the PRODEMOS development work on the digital needs of underserved groups [64]. Our participants highlighted coaching, gamification features, adaptive cognitive training games, and videos as important for initial and sustained engagement and valued positive framing and personalized content. Conversely, complicated navigation, text-heavy interfaces, and overly technical content were perceived as key barriers. This study also adds further insight by suggesting that focusing on 1 risk factor at a time may help prevent user overload and that initially simple, familiar games can support sustained app use. These insights may help guide the design of future digital tools that enhance long-term engagement in underserved groups.

Studies of app-based interventions frequently identified negative attitudes toward technology and privacy concerns as barriers to engagement among older people or underserved populations [36,68,69]. Notably, such themes were absent from our usability testing findings. This may reflect the trust cultivated through in-person contact with a consistent coach and credible research team, or growing societal familiarity with app-based technologies [70], though it is also possible that our think-aloud methodology, which directed participants’ attention toward specific interface interactions, made broader attitudinal concerns less likely to surface.

While underserved groups shared common preferences and barriers, coproduction revealed meaningful differences across subgroups. Ethnic minority participants more frequently requested culturally matched content. For example, broadening dietary recommendations to include vegetarian recipe options. Those from lower socioeconomic backgrounds prioritized affordability, addressed by ensuring all food and exercise content featured budget-friendly and equipment-free options. These findings illustrate a broader principle: designing explicitly for underserved subgroups often improves the intervention for all.

Our coproduction process generated several practical lessons. First, in our coproduction activities, communications between participants and developers were mostly indirect—participant feedback was gathered by researchers and then relayed to the app developer. While this ensured a structured information flow, it could introduce delays and occasional misunderstandings. Direct involvement of users in coproduction workshops would likely strengthen user-centered design by enabling real-time discussion of user needs, technical constraints, and feasible solutions. Second, obtaining meaningful early-stage feedback was challenging when designs were presented abstractly, particularly for less digitally experienced users—a difficulty also noted in PRODEMOS [64]. This improved markedly once clickable prototypes were introduced, enabling richer and more actionable feedback. Third, establishing clear guiding principles at the outset helped anchor the project’s direction, serving as a consistent reference point and reducing design drift. Finally, systematic documentation—including structured records of feedback, design decisions, changes implemented, and changes deferred with reasons—was essential for maintaining clarity and accountability. A shared online platform accessible to all stakeholders, using plain language throughout, supported inclusive decision-making.

The ENHANCE study has several strengths. First, development was theory-driven and evidence-based, with key decisions—including intervention approaches, module cadence, and cognitive games selection—underpinned by robust evidence and user needs. Second, the development was systematic—beginning with a program theory and logic model, followed by evidence scans and usability testing of existing digital materials, with content developed iteratively from resulting design principles. Third, extensive coproduction involved 162 end users from underserved backgrounds, a level of inclusion rarely achieved in digital dementia-prevention research. Fourth, ENHANCE combines scalable app-based delivery with personalized coach support, a model particularly effective for socioeconomically disadvantaged users who may face digital exclusion [36].

This study also has limitations. First, most research participants were recruited in London and Staffordshire, therefore involving few people living in rural areas, limiting the generalizability of our findings. Second, we conducted a rapid, pragmatic evidence scan rather than a full systematic review to identify intervention components for each dementia risk factor module. Therefore, some relevant interventions may have been overlooked.

Third, the app is currently available only on tablets—a deliberate design choice to improve usability, but one that limits accessibility, given that mobile phones are more commonly owned. Although tablets were provided during this study, this may pose challenges for real-world implementation. Future iterations should explore a responsive design approach that adapts across devices, while still maintaining accessibility for older adults.

Fourth, the app is currently available only in English, and during usability testing, participants expressed a desire for versions in other languages (eg, Hindi). While we used simplified language and minimized text to support users with lower literacy, individuals with no English literacy may still face barriers.

Fifth, some design decisions were influenced by practical constraints rather than evidence or user preference alone. Development timelines and technical limitations meant not all coproduction suggestions could be incorporated. For example, a full redesign of the reward system and repositioning of the library icon. These refinements will be prioritized.

Sixth, a formal accessibility audit against a recognized standard, such as Web Content Accessibility Guidelines (WCAG) 2.1 AA [71], was not conducted during development. Instead, we adopted an empirical, user-centered approach, iteratively identifying and resolving usability barriers through repeated testing with users with low digital literacy, guided by HCI expert collaborators—an approach well-suited to surfacing real-world barriers that a standardized checklist might miss. A Google Accessibility Scanner audit confirmed satisfactory touch target sizes, clickable items, contrast ratios, and content labeling, though some interface texts were not fully exposed to screen readers. A formal WCAG 2.1 AA audit will be prioritized in future iterations.

Finally, video streaming delays caused by large file sizes over mobile data connections were identified during testing and addressed iteratively through video compression, SIM card upgrades, and configuring the app to predownload and cache content when connected to the internet. However, mobile data costs and limited offline functionality may present barriers to wider implementation. Future iterations will prioritize minimizing data usage and enabling offline access from the outset.

Despite the structured development process, uncertainties remain regarding engagement and adherence beyond the 1-week testing period, coach workload in routine practice, and sustainability of behavioral change postintervention. These will be examined in our upcoming feasibility trial, assessing 3-month engagement, usability, and preliminary efficacy, and subsequently in the definitive RCT with embedded process evaluation.

While our coaching model was designed with scalability in mind, human coaching presents challenges for cost-effective deployment at scale. We intentionally limited coach contact time and included only 1 face-to-face meeting, balancing cost-effectiveness with accessibility for users. Cost-effectiveness will be formally examined in the RCT. Given the substantial economic burden of dementia, even a modest effect from a relatively low-cost intervention could prove cost-effective.

In the longer term, AI integration may reduce human burden while maintaining intervention quality. We have already automated lower-intensity tasks—such as reminders, progress summaries, and tailored in-app feedback—while retaining human contact for feedback, motivational support and goal setting. While evidence on AI coaching remains mixed [72,73], incorporating empathy, personalization, and behavior change theory may improve its effectiveness. Future research could explore hybrid human–AI models that progressively reduce human burden without compromising engagement or outcomes [73].

The ENHANCE intervention was developed using an evidence-informed, theory-driven, and coproduced process to ensure that its design aligned with the needs, preferences, and capabilities of underserved midlife and older adults who are at higher risk of dementia but often underrepresented in prevention research. By involving 162 end users throughout every stage of development, ENHANCE represents, to our knowledge, one of the most comprehensively coproduced digital dementia-prevention interventions to date. The resulting app uses a unified meadow-themed design and incorporates cognitive training games alongside tailored dementia risk factor modules delivered through educational videos and weekly check-in questions. Our usability testing demonstrated high satisfaction, strong engagement, and the feasibility of the intervention. ENHANCE is now ready to progress to a feasibility trial, and if successful, a definitive RCT to assess sustained engagement, implementation feasibility, and its potential to reduce dementia risk at scale.