“The ICUdata project is quite ambitious,” says Paul Elbers, intensivist (main practitioner in Intensive Care) at Amsterdam UMC and one of the initiators of ICUdata.nl. “We are trying to bring together all data from all intensive care patients in the Netherlands. This data is collected routinely through the devices used for monitoring and life support.” This means that the project does not only go much further than the initial focus on the treatment of Covid patients. It is also being scaled up retroactively: “The idea is to share data from all ICU patients. So not just Covid patients,” says Elbers. “We work with datasets that go back at least 10 years in time.”

Analyzing large sets of IC data has enormous potential within IC departments. “We are finding out how best to provide mechanical ventilation for patients with severe pneumonia. Or which patients benefit most from blood transfusions after major trauma. Or who is best helped with which dose of antibiotics. Who can benefit from certain medications after cardiac arrest. Or… I mean, there are literally thousands of questions that these datasets can answer. The larger the data set, the better,” says Elbers.

Bringing AI to the bedside

While ICUdata.nl is concerned with collecting and sharing the data itself, Elbers is also concerned with what happens afterwards. He is co-chair of the Laboratory of Critical Care Computational Intelligence within the Intensive Care Medicine department of Amsterdam UMC. The goal, he says, is “to use the insights gained from ICU data to build models using advanced statistical techniques such as machine learning and AI, and then bring those models back to the bedside.” Doctors and nurses can then receive real-time advice on how to treat patients.

This is already happening at Amsterdam UMC, using the results of the analysis of the hospital’s own data. Elbers cites two examples of models that have been successfully deployed: “The first is an algorithm that helps us decide which patients should receive which dose of antibiotic – this has been successfully implemented over the past three years. The other is a collaboration with Pacmed – our long-term partner for AI – with whom we developed a model for re-uptake. When an intensivist considers sending a patient to the regular ward, we want to get a prediction about how likely the patient is to need intensive care treatment again. Then we might want to hold off on the resignation. Conversely, if you’re considering keeping a patient in intensive care, and the algorithm says it’s all right to discharge them to the regular ward. Then you could consider doing that sooner, which of course saves a lot of time and money. This has now been implemented on our IC since last month and will soon be tested in a pilot. Those are two current examples of AI at the bedside.” The more data you use for such models, the more powerful and reliable they become and the more potential there is for new applications. So the ICUdata project can have a huge impact.

“These datasets can literally answer thousands of questions”

The great potential of data sharing

Like so many recent developments in healthcare, this one was also given a kickstart by the corona crisis. “It became painfully clear that there was no structure for ICs to share data,” says Elbers. “And that this had to change very quickly. So we started collecting Covid-related data from the hospitals in record time. About 38 hospitals shared their data within two months.” While addressing all the usual regulatory and privacy protection concerns, including the GDPR, other more bureaucratic hurdles were tackled more efficiently than ever before. “People were in a can-do mode,” Elbers says. “Because everyone saw that this had to be solved anyway.”

Want to safely share data, while remaining in control? Read more about our initiative AMdEX

Once the data sharing process started, its potential became even more apparent. “It allowed us to show what our laboratory is capable of: combining this data meaningfully and extracting results from it,” says Elbers. This attracted the attention of Zorgverzekeraars Nederland, the umbrella organization of major Dutch health insurers. “They saw our efforts and said to us and the Dutch Association for Intensive Care: Can’t you do that for regular ICU patients?” To gain more insight, improve treatments and efficiency, and last but not least to be prepared for a possible next pandemic, ICUdata was awarded a €2 million grant.

Sound legal framework

With the emergency phase of the pandemic behind us, the processes have returned to a somewhat less frenetic pace (much to the chagrin of Elbers, who believes all this should have happened at least 15 years ago). As the size of the project increases, so do the regulatory implications. “We wanted to set up a legal framework, a governance framework that is sustainable for years to come,” says Elbers. Where previously all data sharing took place via Amsterdam UMC, a separate entity for managing the data has now been established, the ICUdata Foundation. New contracts with server providers and with participating hospitals must be assessed by many different parties. The privacy protection officers of all hospitals are involved, as is the Dutch Federation of University Medical Centers. “It all became high-level politics,” says Elbers. “We used to do this with three people, now about 150 people are involved.” But things are moving forward. The foundation will collaborate with five participating hospitals: OLVG Amsterdam, Amsterdam UMC, Erasmus Rotterdam, Leeuwarden Hospital and Radboud University Nijmegen. “Once those five have gone through the onboarding process, there is of course room for much more,” says Elbers.

One for all, all for one

In addition to privacy issues, there are other obstacles that often stand in the way of hospitals sharing their data, Elbers says. “Privacy is definitely number one, but the second is competition between hospitals and between researchers.” Hospitals are also wary of scrutiny, potentially misinterpreting data. “As far as I’m concerned, the advantages far outweigh the disadvantages,” says Elbers. “We’ve tried to get around these issues by making sure this is a real Three Musketeers project: one for all and all for one. Anyone who shares their data can get all data back and access to everyone’s data. This was a way of trying to build mutual trust. And now that trust has been built, it really pays off.”

This article previously appeared at Smart Health Amsterdam.