The Croatian team SAMO4DM2 is developing an AI tool that would help family doctors identify patients with type 2 diabetes who may be at risk of not following their treatment, explain the reasons for that risk and suggest what to do next.
For a family doctor, the problem is rarely a lack of information. The harder task is deciding which information requires attention now.
A medical practice may care for many people living with type 2 diabetes, each with a different treatment history and pattern of contact with the healthcare system. Reviewing every record in search of early signs of poor adherence is difficult to reconcile with the pace of routine primary care.
Diabetes Navigator, developed by the Croatian team SAMO4DM2 during the AI4Health.Cro Innovation Challenge, is designed to narrow that field of attention.
The proposed artificial intelligence tool would review existing patient data once a week and identify the 5% to 10% of patients with type 2 diabetes considered most at risk of not following their treatment.
For each patient, the tool would provide three reasons behind the assessment, expressed in clear clinical language. It would then recommend a practical next step: sending an SMS, making a telephone call or arranging an appointment.
The system is also designed to follow the intervention and allow its details, together with the history of previous actions, to be transferred into the patient’s medical record.
The intention is to fit the tool into the existing workflow of a family medicine practice, using infrastructure and data that are already available. Doctors would receive a limited weekly overview rather than another continuous stream of information requiring review.
A problem rooted in everyday practice
The idea came from Ivica Škvorc, the team leader and a family physician with more than 20 years of experience. His motivation was grounded in everyday work with patients and in the need for a tool that could help doctors anticipate poor adherence among people receiving treatment for type 2 diabetes.
His interest is in developing digital health solutions that reduce the burden on healthcare professionals and help them concentrate on focused work with patients.
The rest of the team brought expertise from business organisation, software quality, systems architecture and technology delivery.
Lana Škvorc has more than 10 years of experience in business organisation and EU projects, with a background in information technology and business systems management. Her work has included controlling, marketing, entrepreneurship, professional education and coaching.
Siniša Sambol is a software quality and business analysis specialist with more than a decade of experience in the IT industry. His expertise includes quality-assurance automation, business analysis, technical transformation and web development.
Željko Sučić has more than 20 years of experience in software engineering, systems architecture, enterprise integration, business analysis and the management of development teams. His professional experience also includes the healthcare sector.
The division of roles was central to the development process. The physician defined the clinical parameters, while the technology specialists translated them into the structure of a digital tool.
According to Lana Škvorc, the project would have been considerably narrower had it been developed by people from only one professional field. The team’s breadth came from combining different forms of expertise and allowing each member to build on the work of the others.
Making an AI result understandable
A prediction has limited value in a medical practice unless a doctor can understand why it has been made and decide how to respond.
Diabetes Navigator was therefore designed to provide more than a risk score. For every patient appearing on the weekly list, the tool would show three specific factors contributing to the assessment. These would be translated into language intended to be clear to the clinician.
The system would then propose one of three actions: an SMS, a telephone call or a medical examination.
This structure reflects the team’s attempt to connect analysis with a manageable clinical response. The doctor would remain responsible for deciding what action to take, while the tool would help identify the patients who may require attention.
Its proposed use as a weekly radar also reflects the realities of family medicine. Rather than asking doctors to monitor all patients continuously, the system would direct their attention towards a smaller group identified from the available data.
What AI4Health.Cro added to the process
The AI4Health.Cro Innovation Challenge allowed the team to develop the concept in a structured setting and work with healthcare data in a secure environment for the first time.
The experience was intensive. The team had a limited period in which to develop the solution and had to revise the model several times before it operated in a sufficiently stable and focused way. The competition required more than the technical construction of a prototype. The team also worked on developing the business model and validating the idea.
Evaluation criteria helped define what the solution needed to demonstrate, while advice from the organisers and previous participants during the kick-off and meetup sessions supported the team as it refined both the product and the way it was presented.
The process taught the team to iterate quickly, maintain a clear focus and communicate the problem through a business concept that could be applied in practice.
This was important because Diabetes Navigator sits at the intersection of several requirements. It must respond to a clinical need, work within existing technical conditions and have a credible path towards implementation.
The competition created a framework in which those parts could be considered together rather than developed separately.
Team coordination was supported by a clear division of responsibilities. Each member worked within their own area of expertise, allowing clinical, technical and business questions to be addressed in parallel.
Working within the limits of the healthcare system
The team’s plans for further development begin with validation. The first step is to assess interest in the solution and determine what the healthcare system and its users need from such a tool. The team then plans to test Diabetes Navigator clinically in five to 10 medical practices while monitoring measurable outcomes.
This pilot would provide an opportunity to examine whether the weekly overview is useful in routine family medicine, whether the explanations are sufficiently clear and whether the proposed actions fit into existing workflows.
Alongside this work, the team intends to develop a proof of concept and build partnerships with sponsors and professional stakeholders.
It also plans to explore opportunities through EU projects and continued cooperation with the AI4Health.Cro ecosystem, to secure gradual and sustainable development.
Lana Škvorc described the team’s approach as an attempt to bring together the possibilities of digital technology and the actual conditions of the healthcare system.
The team sees both the potential and the limitations of that environment. Its ambition is to align them in a way that permits a realistic next step.
From prototype to pilot
The competition prize would be invested in further development and validation of the prototype. The team intends to use the funding for user testing, improvement of the tool’s functions and adaptation to the needs of patients and healthcare professionals.
The prize would also help accelerate development towards a pilot and possible application in the healthcare system.
At present, Diabetes Navigator remains a proposed tool awaiting further validation. The planned pilot in family medicine practices will be necessary to determine how it performs in routine work and whether it can support earlier, more focused responses to poor treatment adherence.
Its underlying idea is modest but consequential.
A family doctor cannot examine every patient record each week with the same intensity. A weekly system that identifies a smaller group, explains why those patients may be at risk and proposes a simple next action could help the doctor decide where attention is most urgently needed.
For SAMO4DM2, AI4Health.Cro provided the setting in which that idea could be developed using healthcare data, revised against defined evaluation criteria and shaped into a clearer clinical and business concept.
The next test will take place outside the competition: in medical practices, with healthcare professionals and patients, and through measurable results.