Although AI has become indispensable worldwide, the recently published AI Monitor Hospitals shows that application in Dutch hospitals is still in its infancy. It often remains limited to small-scale projects or pilots, especially in the field of radiology or operations. The UMC Utrecht shows that it can be done differently. The hospital not only develops AI, but also takes the step to healthcare practice.
During Zorg & ict 2025, innovation manager Ilse Kant of UMC Utrecht talked about the Utrecht approach. "A few years ago, we were working very fragmented with AI within the UMC. We wanted to bring that more together and connect. We then set up the 3AI ecosystem, which focuses on data, research and implementation - DRI for short."
This approach is deeply intertwined with education, research and healthcare practice at UMC Utrecht and Utrecht University. "The three pillars are an accessible IT and data infrastructure, AI development including testing and evaluation, and ultimately implementation in practice."
AI readiness
Yet the road to real impact in healthcare is proving long. The AI Monitor Hospitals 2025 shows that only 20 percent of hospitals have one or more components of AI readiness in order. More than half even score below par on essential factors such as AI awareness, knowledge, data management and available resources. Kant: "Too little is landing in practice. Development is often focused on basic scientific research, whereas the goal should be to actually make a difference in healthcare."
The UMC Utrecht therefore opts for a structured approach with practical tools, such as an innovation funnel for applied AI and a 'Guide AI', for healthcare professionals struggling with AI applications. All of these elements are bundled into a quality system. "We hope this removes noise about which laws and regulations apply," Kant said. "But also that it creates speed. Thinking carefully about the goals at the front end creates clarity and direction." The key to sustainable innovation? "Don't stay in love with the solution you create, but with the problem you want to solve. What are you solving and for whom?"
Administrative lighting
According to Kant, the European AI Act provides an important impetus toward the responsible use of AI. She sees opportunities in three areas in particular. First, administrative relief: AI can support healthcare providers in capturing data in patient records and retrieving information. Second, AI offers opportunities for smarter use of people and resources, such as predicting patient flows and efficiently deploying beds and staff. Third, AI contributes to better patient care by supporting medical decisions and higher quality diagnostics and prognostics.
Predicting no shows
Specifically, UMC Utrecht is already showing impactful applications. For example, AI is being used to predict 'no shows' in the outpatient clinic. "We have now implemented this in all 62 outpatient clinics," Kant says. "The prediction helps us reduce the number of no shows by 25 percent. And the beauty: we have shared the solution open source on GitHub. Only when others also adopt it do you really make an impact."
Another example is the automatic generation of draft discharge letters. "Doctors often spend an hour per letter on this," Kant explains. "AI helps by creating a first version. The doctor only has to edit it. This is how we take away work pressure." This solution has been available since May 1 and will also be shared open source. A pilot with the AI dismissal letter showed that 35 percent of professionals use the concept immediately, while others use it more as a mnemonic, inspiration or monitoring tool.
A third example of AI deployment is consultation preparation for outpatient appointments. This time-consuming task can be alleviated by AI. The solution is currently under development. Work is also underway on "smart alerting," in which nurses monitor patients remotely. Predictive models help here to estimate when clinical deterioration is imminent. This project is also in the pilot phase.
Application in practice
That implementation is essential is also stressed by neonatologist Daniel Vijlbrief. "The real value of AI only becomes apparent after implementation, when professionals start using, questioning and improving it."
To make that happen, UMC Utrecht has set up five specialized AI Health Labs, focusing on regenerative and molecular medicine, imaging, prevention and AI methodology, among others. These labs collaborate with ten Utrecht University AI labs and companies in the region. This collaboration is housed in the broader EWUU AI Hub, in which TU Eindhoven and Wageningen University & Research also participate.
Support care goals
According to Kant, successful AI implementation depends on clear prioritization, collaboration with healthcare providers and patients and deployment of transparent review frameworks. "Make sure you deploy AI where it supports healthcare goals. Actively involve your end users and develop together. Then you really connect with practice."
She also stresses the importance of collaborating with partners such as the NFU and industry, responsible use of large language models such as ChatGPT, and establishing governance structures and quality systems. To deliver on the promise of AI, governance support is crucial. "Make sure your directors are up to speed on what is needed," Kant advises. "Attracting and retaining talent and expertise is just as important as having a good model."
She recommends that fellow institutions read ZonMw' s recently published AI Signaling Report and join the AI Implementation in Healthcare knowledge network.
Guidelines needed
Finally, Kant warns against scaling up AI too quickly and unchecked. "We need guidelines for AI in healthcare. You have to deal carefully with risks, laws and regulations, privacy and ethics." In the National Guideline for Quality AI in Care, developed in part under the leadership of UMC Utrecht, these aspects have now been elaborated. The UMC works on the basis of this guideline. "We can be proud of that."
