The Prof Ibsen Lecture at this year’s Euroanaesthesia conference addressed the theme of innovation and artificial intelligence. Professor Andre Dekker, from Maastricht University Medical Center and Maastro Clinic in The Netherlands, presented the topic Artificial Intelligence (AI) for Better Healthcare.
Anaesthetists today face several challenges: managing increasingly complex patients, adhering to tighter schedules and navigating workforce shortages while being under pressure to improve outcomes and optimise resources. AI offers promising solutions with the potential to enhance risk prediction, enable precision dosing, and streamline workflow management. It can support clinical judgment and enable safer, more personalised patient care.
Yet, the path from promise to practice is far from straightforward, as highlighted by Prof Dekker. Effective AI depends on access to high-quality, interoperable data, but healthcare data remains fragmented, siloed, and subject to stringent privacy regulations. Many AI algorithms operate as black boxes, raising concerns about transparency, bias, and trustworthiness. To address these challenges, rigorous model validation, explainable AI techniques, and federated learning approaches are essential to ensure confidentiality, fairness, and clinical oversight.
According to Prof Dekker, AI can help healthcare in two fundamental ways: first, by improving efficiency and automating administrative tasks, supporting workflows, and aiding in triage or diagnostics, particularly in imaging. These are the areas where healthcare can expect the most immediate gains. However, the bigger area of focus should be on how AI can improve the effectiveness of care and help clinicians make better decisions for individual patients. This often involves predicting a patient’s future health state based on potential interventions, something humans are not particularly good at but where AI excels.
When used appropriately, AI can relieve clinicians of repetitive and burdensome tasks, allowing them more time for meaningful patient interaction. AI can also objectively predict patient-specific outcomes, which facilitates shared decision-making and ensures treatment is tailored to the individual’s needs.
AI can also improve safety in the perioperative setting. It can outperform humans in predicting adverse events like intraoperative hypotension, difficult airways, or awareness under anaesthesia. It can also detect life-threatening conditions such as anaphylaxis or malignant hyperthermia faster and more reliably. That said, AI systems must also be transparent about their limitations, and frameworks need to be in place to ensure effective human oversight.
Prof Dekker has played a leading role in advancing the FAIR (Findable, Accessible, Interoperable, and Reusable) data principles in healthcare. One of his most notable contributions is the development of the Personal Health Train (PHT), a federated learning infrastructure that allows data to remain securely within hospitals while still enabling collaborative learning, effectively addressing ethical and privacy concerns. A critical aspect of the PHT is the requirement for hospitals to make their data FAIR, essentially making the concept actionable and beneficial rather than theoretical.
The main hurdle to the widespread adoption of FAIR principles in healthcare is the time and resources required to make data FAIR. Fortunately, advancements in supportive technologies, including AI, and legislation such as the European Health Data Space (EHDS) initiative are making this process easier and more accessible.
Prof Dekker’s AI models are widely used in oncology. One of the most exciting developments is the Dutch model-based indication system for proton therapy. This system relies exclusively on AI-generated predictions to determine treatment eligibility and reimbursement, with no clinician input. It represents a paradigm shift in how care decisions are made. Imaging-heavy fields like radiology, radiation oncology, pathology, ophthalmology, and dermatology will likely benefit first due to the volume and standardisation of their data. Following that, signal-based specialties—neurology, anaesthesiology, cardiology—will see gains. Additionally, foundation models, including large language models like GPT, will ease the administrative burden across all disciplines and support decision-making in narrative-driven specialties.
Source: Euroanaesthesia 2025
Image Credit: iStock
