While there are plenty of pilot projects, far too few AI innovations survive the three so-called 'valleys of death'. Despite their promise, these innovations often fail to reach the market, preventing them from proving their value. Pieter Jeekel, chair of the Health and Care working group, and Niels Bolding, Technology Lead at Health-RI, discuss the obstacles and the key solution: ensuring governance of generic facilities to guide AI innovations through these three valleys of death.
AI offers plenty of opportunities to improve healthcare, such as reducing administrative burdens and faster diagnoses. And these are just a few examples.
To keep healthcare accessible and affordable in the coming years, AI innovations are essential. The AI Coalition for the Netherlands and Health-RI, in collaboration with Zorgverzekeraars Nederland and many other stakeholders, are advocating for national measures and direction to guide many more innovations successfully through the three valleys of death.
Valley of death 1: Regulation
From concept to market entry
Before large-scale pilots and market analysis are possible, an AI innovation must pass through the first valley of death: the maze of rules surrounding the Medical Device Regulation (MDR). Since August 2024, the AI Act, the European regulation for AI, also applies. Pieter Jeekel: "Complying with the MDR is a challenge for every e-health innovation. And the AI Act doesn't make it any easier. It's good that there are European rules now, but the threshold has only become higher. Innovators often don't know exactly what requirements they need to meet. And even if they do know, it is difficult to find a suitable notified body."
Many different parties are involved in the regulation and associated obstacles, including the Ministry of Health, Welfare and Sport, Economic Affairs, Health Innovations NL, Care for Innovation, health insurers, Health-RI and the AIC4NL.
"The big problem is that no one is really responsible for the whole thing. We need to appoint one person in charge in the Netherlands with a mandate on behalf of all parties. This person can investigate which tools can be deployed, which sandboxes we can set up, which notified bodies we should contract and which support tools we should deploy."
Niels Bolding adds, "A central ELSI desk can be the starting point to pool these questions, answer them or direct generic solutions."
Valley of death 2: Data availability
From pilot to first paying customer
Those who have steered their innovation through the first valley of death and comply with regulations are one step ahead. Niels Bolding: "You've got a working product and a CE mark. But now it's really just beginning. How do you market your product commercially? And how do you make sure that users are willing to pay for it?"
Many innovations fail at this stage because they lack relevant data to train their algorithms. Bolding: "AI without data is an empty shell. But as an outsider, you don't have access to real patient data, but you do need it."
In this context, Bolding mentions the AI4Health action plan drawn up by the Dutch AI Coalition and Health-RI in cooperation with Zorgverzekeraars Nederland. In it, data availability is central. "It is about making large amounts of quality data accessible according to FAIR standards. In addition, we need to balance the risks and incentives for sharing data, organize distributed access and ensure representative and inclusive data."
According to Jeekel, a data-AI platform could be the answer: "Such a data platform should obviously involve the parties who are already working on data availability. Health-RI plays a key role in this, but initiatives such as CumuluZ could also contribute." He points to international examples: "They're already working hard on this abroad. For example, the Mayo Clinic in the U.S. has made agreements with several hospitals to share data. They also want to cooperate with us, but we have different laws and regulations. Still, it 's certainly possible in the Netherlands and Europe."
Valley of death 3: AI readiness
From first customer to sustainable solution
Once an innovator has both passed regulation and trained the application with data, the third challenge follows: real-world application. Its success depends on the AI readiness of healthcare organizations. Jeekel: "The pressure on healthcare is enormous, so things really need to change. But that's not easy. Implementing e-health has always been difficult, and AI innovations also create new issues around privacy and ethics."
Healthcare organizations need sufficient knowledge and capacity to assess, implement and use AI applications. By including AI integration as an essential part of their strategy, administrators can create a clear vision that inspires departments to work together. If automation experts then implement these applications and healthcare professionals develop the skills to work effectively with these technologies, the transition from pilots to widespread adoption can be accelerated. This will enable us to reap the full benefits of AI innovations for both healthcare professionals and the Dutch healthcare system.
As more and more healthcare organizations become AI-ready, they'll have a strong foundation to implement the innovations that survived the first two valleys of death. Collaboration is crucial here, Jeekel says. "Healthcare organizations can learn a lot from each other. We can support them with information, education and expertise so that they become truly AI-ready, which allows them to look beyond point solutions and implement systems that fit their way of working."
Jeekel and Bolding argue in favor of collaboration and coordination in close coordination with VWS to help AI innovations get through these valleys faster and better, with the intended outcome being more impact in healthcare.
