By Dr Carlotte Kiekens
Last week, the city of Krakow hosted the 25th European Congress of Physical and Rehabilitation Medicine (ESPRM 2026). Amidst the gathering of Europe’s leading clinicians and researchers in the field of physical and rehabilitation medicine and rehabilitation at large, the PREPARE project took centre stage to showcase a bold vision: a future where rehabilitation is not only person-centred but powered by federated data and artificial intelligence. Through two intensive sessions, our team shared the progress of our 7-million-euro Horizon Europe initiative, highlighting how we are turning complex health data into actionable clinical insights.
The Challenge: Why Data in Rehabilitation is “Different”
Rehabilitation is inherently complex. Unlike a single surgical procedure, it is a complex multimodal person-centred process. As Carlotte Kiekens highlighted in her opening remarks, while Clinical Decision Support Systems (CDSS) are common in other fields, rehabilitation lacks validated, large-scale prediction models. The PREPARE project (Personalised Rehabilitation via Novel AI Patient Stratification Strategies in Europe) was born to fix this. With 20 partners across nine countries, we are tackling the fragmentation of health data.
Session 1: PREPARE-REHAB SESSION – EXPLORING DATA-DRIVEN REHABILITATION RESEARCH IN EUROPE
A core theme of our first session was interoperability. How do you compare nine different clinical databases from across Europe?
Unlocking Health Data
Ruud Selles introduced the audience to the OHDSI Common Data Model (CDM). By using a federated learning approach, we don’t move sensitive patient data; instead, we move the algorithms to the data. This ensures privacy while allowing us to train AI on a massive, diverse scale.
A comparison of seven rehabilitation databases
Carlotte Kiekens shared a snapshot of the similarities and differences in structure and content among the rehabilitation databases and their included variables: measurements, modifiers and outcomes. The findings demonstrated significant variance in data types, primarily biomedical indicators, with more limited consideration of psychosocial and contextual factors. These results underscore the necessity of standardised, comprehensive data collection (including psychosocial & contextual factors) to facilitate the development of effective clinical decision support systems and personalised rehabilitation strategies. Then Stefano Negrini presented the Guideline for Intervention Description in Rehabilitation (GUIDE-Rehab) reporting tool, which was used to describe the various rehabilitation programs of the PREPARE-REHAB clinical partners, and allow the comparison of the various included interventions.
Standardizing the Language of Rehabilitation
One of the most significant hurdles discussed by Helena Burger, Stefano Negrini, and Ruud Selles is “OMOPping”, the process of mapping rehabilitation variables to the OMOP standard. Our findings revealed a critical gap: many rehab-specific concepts (like PROMs, various interventions for rehabilitation and long-term functional outcomes) are missing from current global vocabularies.
By using the WHO’s International Classification of Functioning, Disability and Health (ICF) framework, PREPARE-REHAB is working to extend these libraries, creating a new gold standard for how rehabilitation data is recorded globally.
Session 2: ARTIFICIAL INTELLIGENCE (AI) AND REHABILITATION
The second more general session, chaired by Carlotte Kiekens, moved from data structure to clinical application.
The first speaker was Giovanni Morone who presented the REHALISE initiative “Rehabilitation Enhanced by Artificial Intelligence” on Clinical Application of Artificial Intelligence in Rehabilitation. This project consists of a biannual living review investigating current clinical applications of AI in the rehabilitation field, in collaboration with the Cochrane Rehabilitation, Functioning, and Disability Thematic Group (CochraneRehab).
Also in this session Carlotte Kiekens briefly presented the PREPARE-REHAB project, followed by a series of oral presentations, exploring how AI is already beginning to outperform or support human decision-making:
- Classification & Coding: Linda Nieminen showcased how AI-powered ICF coding can handle complex health records with high accuracy, a task that is traditionally labor-intensive for clinicians.
- Clinical Accuracy: Stefano Negrini presented a fascinating comparison between AI and human agreement on classifying Cochrane Systematic Reviews, while Simona Dan explored how different prompt strategies affect AI performance in complex rehab scenarios.
- Diagnostics: Research presented by Aysegul Garip Honamlioglu and Mehmet Fatih Kocibar demonstrated that custom-trained AI models are becoming increasingly proficient in diagnosing hip pain and grading osteoarthritis, matching or even exceeding standard clinician-trained tools.
The Road Ahead: From Data to the Patient’s Bedside
The conclusion of our week in Krakow was clear: Data science is no longer a “future” concept in rehabilitation, it is the present. Through our nine demonstration pilots, the PREPARE-REHAB project is validating these AI tools in real-world settings. We are not just looking at statistics; we are looking at how these tools change clinical decisions, how clinicians adopt them, and how they improve the actual lived experience of the patient.
As we move forward, our goal remains to provide reliable, personalised care that considers the unique external circumstances and personal factors of every individual.
.
Posted by: Nayara Güércio (Trilateral Research)