The use of AI in hospitals is at risk of collapsing due to fragmentation. This warning comes from Wouter Kroese, founder of AI pioneer Pacmed. “In the future, healthcare providers will use dozens, and perhaps even hundreds, of AI-based insights and predictions. If all of those will be point solutions, AI-solutions will become unaffordable, unimplementable, and unmanageable.”
Kroese’s plea does not come out of nowhere. In 2014, Pacmed developed its first plans to improve healthcare using data and artificial intelligence (AI). Ten years later, Kroese and his colleagues can look back on several noteworthy collaborations, including with Amsterdam UMC and Santeon, but also on a number of—what Kroese calls—“painful statistics.” “We have been involved in dozens of different development projects. The outcomes were almost always promising and the participants enthusiastic, but implementation and actual use rarely followed. A promising initial algorithm is just a fraction of the work needed to create a usable and scalable product.”
Painful Conclusion
According to Kroese, the slow adoption of AI is closely related to the fragmented organization of healthcare. As a result, hospitals and departments have a strong tendency to seek their own solutions. However, point solutions in medical AI are, in Kroese’s view, a dead end. “In the past few years, we have managed to bring AI to patients twice. Our software has proven capable of generating predictions and insights that help improve specific medical decisions. However, in both cases, it turned out to be neither a viable nor scalable product. It’s time for a harsh and painful conclusion for Pacmed: medical AI decision support is not viable as a point solution.”
EPD Integration
Scale is, according to Kroese, essential to address some fundamental issues regarding the application of AI in healthcare. He thinks, among other things, about data quality and the limited accessibility of data sources, such as the electronic patient dossier (EPD). “We naively started processing, harmonizing, validating, enriching, and standardizing all ICU data, but it takes an enormous amount of time to connect them. And we haven’t even talked about integrating with the various and differently configured EPD systems.”
A Real Shame
EPD integration is just one of the technical hurdles. There are also challenges such as data extraction, CE certification, continuous data monitoring, and cloud hosting. Together, these make costs “far too high for a point solution,” says Kroese. “We’ve paid for lessons learned in this regard over the past ten years. We obviously had to start somewhere. A point solution is a logical starting point. But the fact that, ten years later, we are still starting over again in hundreds of places in the Netherlands is a real shame. In the future, healthcare providers will use dozens, if not hundreds, of AI-based insights and predictions. If all of that consists of point solutions, it becomes unaffordable, unimplementable, and unmanageable.”
AI Hype
Nevertheless, this is the trend Kroese observes. “All over the country, new initiatives are emerging. It’s a shame that they are so fragmented and that those involved often start completely anew. We are spending millions of euros and thousands of scarce hours on AI in the Netherlands. It is crucial that these investments actually contribute to making the healthcare sector sustainable, which is under enormous pressure. Otherwise, we risk inflating an AI hype. If that bubble bursts, we will lose time and attention that AI will desperately need in the coming years.”
Reuse of Data
Pacmed aims to set a good example by consistently linking further activities to benchmarks such as value, impact, accountability, and scalability. “Together with hospitals, we are developing Pacmed Critical, our software for timely and safe discharge from the ICU, into an integral and ICU-wide product that brings together all valuable information based on the needs of healthcare providers,” says Kroese. “This means, for example, that we use the same data for different purposes. We will also facilitate the development of AI for hospitals and help ICUs bring their AI ambitions, research, and projects to the bedside.”
According to Kroese, ongoing activities at OLVG prove that this is possible. “In just a few months, we have been able to develop and roll out a new functionality for better utilization of capacity, while it took us seven years to develop our discharge software.”
.png)
