Keeping people at the centre of AI guideline navigators
Author: Felix Muehlensiepen1
1. Brandenburg Medical School Theodor Fontane
As a health services researcher working on digital health and the use of artificial intelligence (AI) in clinical care, I keep running into the same uncomfortable paradox: the right treatment for a patient often already exists in writing, agreed by experts, and backed by years of evidence, yet it still never reaches them. This happens far more often than we would like to admit. Clinical guidelines for chronic conditions can run to hundreds of pages and change constantly. Even the most committed clinicians cannot keep up with all of them, and study after study shows that only a minority of people with chronic illness actually receive the care their guidelines recommend.
This gap between evidence and everyday practice is what drew me to GUIDE-AI, a four-year European research project that began in January 2026, and which I am part of as lead of the User Experience in Digital Health lab at Brandenburg Medical School Theodor Fontane, responsible for the work package on people-centred innovation and interest holder engagement. My focus in the project is on potential user perspectives on AI-guided decision support, so that the tools we build close rather than widen the existing gap. My interest in the human side of these systems, rather than the algorithms alone, shapes how I'll tell the GUIDE-AI story in this blog.
What is GUIDE-AI?
GUIDE-AI (Guiding Your Treatment with Current Evidence) started from a simple question: Can generative AI help close the gap between what guidelines recommend and what patients actually receive?
The project is building tools called guideline navigators, which use large language models to help doctors quickly find the treatment recommended for a particular patient. The same technology will also produce plain-language explanations for patients, so they understand why a treatment has been chosen for them.
GUIDE-AI focuses on four common chronic conditions where the evidence-to-practice gap is well documented: heart failure with a reduced ejection fraction, chronic kidney disease, chronic obstructive pulmonary disease, and asthma. Alongside these, the consortium is developing an exploratory navigator for inflammatory bowel disease, to test how well the approach carries over to other conditions. All throughout, the project will remain independent of any single AI vendor or model.
A European consortium
My group is one of close to 20 partners in GUIDE-AI, with Charité Universitätsmedizin Berlin coordinating the whole. Sitting in those consortium meetings, what strikes me most is how differently guideline-based care and trust in AI play out from one place to the next: a recommendation that is routine in a German university hospital can be hard to act on in a smaller clinic elsewhere, with different data, staff, and rules. University hospitals, research institutes, patient organisations, and technology companies from Germany, Italy, the Netherlands, Estonia, Slovenia, Switzerland, the United Kingdom, and Israel all meet around the same table for exactly that reason: no single team could see the whole picture on its own.
Why technology alone is not the answer
It is easy to assume that a clever tool will be used simply because it works. Yet, in my research on digital health, I have watched the opposite happen again and again: well-built technologies go quietly unused, not because the engineering is poor, but because they do not fit into how clinicians actually work, or speak to what patients need or worry about. That is the failure I most want us to avoid through the work on GUIDE-AI. It is also at the heart of this year’s World EBHC Day theme, Evidence and AI: People at the Centre, which underscores that AI in healthcare is not only a technical challenge but also a human one.
Start with people, not the model
My own part of GUIDE-AI begins with people rather than with the model. Specifically, the work package that my team leads at Brandenburg Medical School has a clear job in this consortium: to bring the voices of clinicians and patients into the design of the navigators. It is important that we intervene in this way before the technology is locked down, not after. Therefore, before a single navigator is finalised, we are running a mixed-methods study with the people who will live with these tools. The aim of this study is to identify the questions that will ultimately decide whether a navigator is ever really used. Three groups matter most to us:
- Doctors: What would make them trust an AI suggestion? When would they overrule or ignore it?
- Patients: Do they want AI involved in decisions about their care, and how should that involvement be explained to them?
- Administrators and payers: What is needed for these tools to fit real healthcare systems safely, fairly, and sustainably?
Listen before building
Through a combination of interviews and surveys, we seek to explore the hopes, doubts, and ‘red lines’ that are usually not taken into account until a tool is already in use. These insights will then feed directly into how the navigators are designed, tested, and evaluated, alongside questions of privacy, regulation, and clinical safety.
Later, the navigators will be tested on existing data and then in a prospective, randomised study measuring concrete outcomes: whether more patients actually receive guideline-recommended treatment, and how often an AI analysis changes a treatment decision.
Early reflections
The project has only just started, so it is too early to say that real lessons have been learned. Yet a few principles are already shaping how we work. For a start, we keep asking who is missing from the conversation, because the patients least likely to receive guideline-based care are often the ones least likely to be asked about new AI tools. Furthermore, we treat trust as something we have to earn from clinicians and patients and then measure, not something we can assume. And we try to keep sight of the real aim: helping more patients receive treatment that works for them, rather than building a clever tool for its own sake.
Key messages
- Too many patients with chronic conditions still miss out on the care that guidelines recommend. That is the gap that GUIDE-AI is built around
- An AI navigator will only help if doctors and patients are willing to use it, which is why we are studying their views from the very start
- For us, keeping people at the centre means asking a simple question: Does the tool actually lead to better care?
References
CORDIS: EU Research Results. (2026). GUIDE-AI: Guiding your treatment with current evidence. https://doi.org/10.3030/101253015
JBI. (2026). World evidence-based healthcare day 20 Oct 2026. https://worldebhcday.org/
To link to this article - DOI: https://doi.org/10.70253/NWRW3483
Links to additional resources
https://www.ihi.europa.eu/projects-results/project-factsheets/guide-ainnovative Health Initiative
Conflict of interest
GUIDE-AI is funded by the Innovative Health Initiative (IHI), a public–private partnership between the European Union and European life-science industry associations. The consortium is coordinated by Charité Universitätsmedizin Berlin and brings academic, clinical, and patient partners together with the industry partners AstraZeneca, GlaxoSmithKline, and Takeda. The corresponding author is an academic researcher at Brandenburg Medical School and declares no personal or financial conflicts of interest.
Disclaimer
The views expressed in this World EBHC Day Blog, as well as any errors or omissions, are the sole responsibility of the author and do not represent the views of the World EBHC Day Steering Committee, Official Partners or Sponsors; nor does it imply endorsement by the aforementioned parties.
