Bridging AI and Primary Care: AI-assisted Health System Integration Model for the Inclusion App

Rwanda, like other developing nations, still struggles to fully incorporate rehabilitation services into its healthcare system, especially at the primary care level. Rehabilitation is offered in public, private, and some non-governmental organization (NGO)-supported health facilities (Kumurenzi et al., 2022). Public facilities mainly provide rehabilitation at secondary and tertiary levels, such as district hospitals, referral hospitals, and university teaching hospitals (Kumurenzi et al., 2022). Rural areas, characterized by poor health infrastructure and geographical isolation, often face the challenge of mountainous terrain, hindering access to healthcare. Rehabilitation services are confined to district hospitals and not available at health centres, health posts, or within the community, which are the primary points of care for rural populations (Dukuzimana et al., 2025). These challenges in accessibility and availability of rehabilitation services, were the main motivators to develop a community-based digital rehabilitation application, the Inclusion App, in Project Inclusion (Physitrack, n.d.). The application allows community health workers (CHWs) to easily share basic exercise programs for the most common conditions (Oduor & Aartolahti, 2024).

Project Inclusion

Since 2023, the project has revolutionised rehabilitation service delivery by using community-based and digital methods. Sixty-eight CHWs have successfully provided basic rehabilitation services to around 13,000 individuals with common musculoskeletal conditions such as lower back and osteoarthritis, along with tailored exercises for the elderly in Rwanda.

The CHWs received training on basic musculoskeletal conditions, identifying and managing red flags, digital and general health literacy (Oduor & Aartolahti, 2024). They were also given monthly internet access and, if needed, basic smartphones to support their work. A retrospective study revealed that this approach could cut the cost of rehabilitation services by as much as seven times compared to traditional facility-based care (Kempers et al., 2025). The approach also improved the CHWs’ skills, boosting their standing and sense of community.

The Inclusion App’s success highlights that equipping CHWs with digital tools is both practical and essential for advancing equitable rehabilitation. By strategically integrating AI-assisted digital rehabilitation into primary care, the future of community-based rehabilitation will be more adaptable, accessible, and patient-focused.

An AI-assisted Health System Integration model

Despite Project Inclusion’s beneficial outcomes, various challenges remain, including a lack of advanced understanding of musculoskeletal conditions, limited time available for regular training sessions, inconsistent follow-ups using the Inclusion App because of other commitments, and a lack of consistent data collection and monitoring.

To address these issues, incorporating AI support alongside the Inclusion App offers a promising solution. AI can transform rehabilitation by improving evaluation, screening, goal setting, follow-up and monitoring as well as prevention. AI tools learn from data to aid in clinical and rehabilitation procedures, customizing and complementing traditional rehabilitation strategies (Davenport & Kalakota, 2019; Lanotte et al., 2023; Rowe et al., 2022). AI-guided digital rehabilitation has the potential to be more scalable, efficient, and sustainable, thereby bridging gaps in knowledge, access, and continuity of care.

However, rehabilitation services – spanning physical, psychological, and social dimensions – remain fragmented and difficult to access in a coordinated way. This fragmentation poses a challenge not only for service users but also for providers and policymakers seeking scalable, inclusive solutions. Discussions with officials from Rwanda’s Ministry of Health have emphasized integrated solutions that tackle multiple challenges, rather than single-issue applications. Given that multidisciplinary rehabilitation is so comprehensive, a more integrated approach is required.

Recognizing this gap, a multidisciplinary team convened for a one-week workshop in Rwanda to co-create a new model for integrating rehabilitation into primary care. The goal was to create a digital rehabilitation (DR) integration framework using artificial intelligence (AI), customised for specific local needs and designed to integrate with the national health management information system (HMIS). At the heart of this effort is the AIRe Platform (Jamk University of Applied Sciences, n.d.), envisioned as an open, adaptable and interoperable system to deliver personalized rehabilitation, and feed valuable data back into national health systems.

Training AIRe

The AIRe Platform operates through two interconnected layers: AIRe Talk for rehabilitation service users and AIRe Hub for professionals. While AIRe Talk enables rehabilitation users to interact naturally with AI in their own language and receiving personalized guidance and support, AIRe Hub allows professionals, such as physiotherapists and occupational therapists, to train and customize the AI for specific target groups and contexts.

Training and customizing AIRe does not require any coding skills. Thanks to intuitive no-code functionalities, professionals can define thematic areas and guide AIRe by prompting it toward specific topics around which conversations with users should revolve. These topics include, for example, the key components of community-based rehabilitation and the World Health Organisation’s package of interventions for rehabilitation (World Health Organization, 2023). Once a theme is identified in the dialogue between the user and AI, AIRe knows how to respond and what content or instructions to offer. With upcoming developments, professionals will also be able to enrich AIRe’s knowledge base by adding validated articles and trustworthy materials for each theme. This allows AIRe to draw on reliable, expert-approved content when engaging in conversations with users ensuring that guidance is both contextually relevant and professionally grounded.

However, it is important to acknowledge the potential of technology, particularly AI, to either worsen or improve existing inequalities in healthcare. AI-driven systems risk worsening health inequalities because of biased data or unequal distribution of newly developed services, especially within primary care settings, where equitable access is crucial for underserved communities (Allen et al., 2024).

Piloting the Integrated Model

On the final day of the workshop, three CHWs tested AIRe Talk, the AI assistant integrated into the AIRe Platform, and the integrated exercise programs from the Inclusion App. This pilot project aimed to assess AIRe Talk’s overall usability, the precision of its Kinyarwanda translations, and its practical usefulness within real-world community settings. During the testing phase, the “thinking aloud” method was used to gain insights into users’ cognitive processes while interacting with AIRe Talk (Ericsson & Simon, 1980; Noushad et al., 2024).

At the end of the testing, each CHW completed a brief online survey to assess their attitudes, emotions, and the usability of AIRe Talk. The survey included three self-reported questionnaires: the Information Technology Attitude Scales for Health (ITASH) (Ward et al., 2007), the Positive and Negative Affect Schedule (PANAS) (Watson et al., 1988), and the System Usability Scale (SUS) (Brooke, 1996). The ITASH questionnaire was modified by substituting “ICT” with “AI” or “AI solutions.”

CHWs’ Experiences of the AI Assistant (AIRe Talk)

CHWs found AIRe Talk intuitive and easy to use, especially on mobile phones. This conversational interface was described as natural and empowering, helping CHWs feel more confident when supporting end users. Instead of referring questions to external professionals, they felt equipped to provide trustworthy, immediate advice. One participant noted that AIRe “teaches and empowers,” enhancing their role in the community.

The platform’s Kinyarwanda support was generally accurate, though minor translation errors occurred. Users were still able to understand the intended meaning, and the AI clarified instructions when asked. The conversation flow was smooth, though some noted that multiple questions at once could be overwhelming. The system also provided symptom warnings and step-by-step guidance, contributing to a sense of reliability.

Clinical Relevance and Real-World Use

CHWs tested AIRe Talk with actual client cases, including arthritis, low back pain, and finger stiffness. The platform correctly identified the conditions and provided appropriate exercise programs. One user was impressed by AIRe’s ability to recognize arthritis in Kinyarwanda and to suggest relevant knee osteoarthritis exercises. AIRe Talk also offered content on insomnia and mental health, demonstrating its potential for holistic support.

The results from the questionnaires reinforced the insights gained during the group discussion. The ITASH scores averaged 66, suggesting a mostly positive attitude towards AI solutions. On the positive PANAS, the CHWs’ average score was 49/50, indicating a strong positive emotional response. The average negative PANAS score was just 10, the lowest possible score, suggesting minimal negative emotions. These findings suggest CHWs experienced predominantly positive feelings while using AIRe Talk. Additionally, the average SUS score was 92, signalling that users found AIRe Talk exceptionally user-friendly.

Reflections and Group Insights

During a follow-up group discussion, the CHWs expressed strong appreciation for the platform. They felt AIRe could empower them and boost their confidence when they did not have immediate access to healthcare professionals. They appreciated being able to provide accurate information instantly, without delays or needing to make referrals.

Suggestions for improvement included adding audio functionality in Kinyarwanda to make it easier to use and build trust, especially for those with low literacy. They stated that voice commands could be more efficient and allow more users to be helped simultaneously when compared to typing/text-based input. Challenges such as internet and device access were identified as external hurdles, independent of AIRe.

One CHW shared how AIRe helped her realize that previous exercise prescriptions were not always right for the client. The platform’s symptom-based recommendations could help her make better assessments and decisions.

Importantly, the CHWs did not see AIRe as a threat. They viewed it as a tool to enhance their abilities, especially in supporting vulnerable individuals. They also suggested integrating other tools, like the Inclusion App, to help with follow-up visits and collaborative decision-making.

The pilot demonstrated that the AIRe Platform has strong potential to support community-based rehabilitation. It empowers CHWs, improves confidence, and enhances the accuracy of care. With further development such as additional content for different conditions, improved Kinyarwanda translation and audio features, AIRe could become a key asset in scaling equitable rehabilitation access.

Acknowledgements

We would like to acknowledge the valuable support provided by the University of Rwanda and its Centre of Excellence in Biomedical Engineering and E-Health (CEBE), whose facilities and collaboration made this workshop possible. The University of Rwanda/CEBE is a long-term partner of Jamk University of Applied Sciences, working together to advance research, innovation, and capacity building in digital rehabilitation in Rwanda and beyond.

We would also like to thank Jean Damascene Bigirimana, Flora Munezero, and Jean d’Amour Niyonkuru for their helpful comments, discussions, and other assistance during the workshop.