Graz University of Technology, the University of Graz and the Medical University of Graz have jointly developed an interactive system that automatically adapts evidence-based medical information to patients’ prior knowledge and needs.

A medical diagnosis raises many questions, but in everyday hospital life there is often little time for detailed discussions. Information brochures or websites usually provide standardised content that hardly takes individual prior knowledge and needs into account. This is why Graz University of Technology (TU Graz) (Institute of Visual Computing), the University of Graz (Institute of Psychology) and the Medical University of Graz (Institute of General Medicine and Evidence-based Health Services Research) have jointly conducted research on new ways of providing patients with relevant information using the example of diabetes. Under the coordination of Tobias Schreck from the Institute of Visual Computing at TU Graz, the researchers have developed an adaptive information system that personalises medical knowledge. The aim was to present scientifically proven content in such a way that patients can understand and categorise it more easily, thereby creating a better understanding of diagnoses and therapeutic approaches.

Adaptive instead of static

The system, called A+CHIS (Adaptive Consumer Health Information System), automatically adapts information. It recognises how much depth of detail a person needs and presents content accordingly, for example as a simple keyword cloud, clear infographic or in-depth technical text.

The basis for this is what is known as multidimensional adaptivity. The system analyses anonymised interaction data such as mouse movements or scrolling behaviour in order to detect cognitive overload at an early stage and dynamically adapt the display. Using data from a study conducted in the project with 250 participants, the research team investigated how interaction patterns can be used to reliably detect when information is perceived as too complex or overwhelming. “Our aim was to communicate medical evidence in such a way that it is really understood and not just read,” explains Tobias Schreck.

AI as an evidence-based dialogue partner

Another building block is the use of large language models. The AI-supported components provide support as digital advisors. The University of Graz has contributed to the field of information processing by providing a cognitive psychology perspective. Psychologist Michael Bedek summarises the research work as follows: “For example, we have investigated how content can be presented in a way that is easier to understand. We have also examined what expectations people have of a platform and how biases can be avoided.” According to the scientist, people often seek confirmation of their own hypotheses rather than information that contradicts them.

To ensure the quality of the medical content, A+CHIS only uses materials that have been reviewed by the Medical University of Graz according to defined quality criteria. The AI uses this for its summaries and suggestions in order to reduce the risk of so-called “hallucinations” by large language models.

Open source and outlook

Although the system was initially developed for information on diabetes, it can be applied to any conceivable medical topic. The team is making the project results available as open-source code. By doing so, the researchers are creating a scientifically sound basis for future digital health information systems, for example in hospitals, doctors’ surgeries or insurance companies. A follow-up project will start in spring 2026 in which research will be carried out on the effective communication of trustworthy health information together with citizen scientists. In the long term, the researchers want to transfer the principles of adaptive information transfer to other areas of education. The aim is to make complex knowledge generally more comprehensible, more individualised and more effective.