OLVG and Maasstad are going to use artificial intelligence on the IC

Thanks to artificially intelligent discharge software from software company Pacmed, the doctors in the intensive care unit of the OLVG hospital in Amsterdam have a new assistant.

This helps the doctor decide whether or not a patient can be discharged from the ICU. A kind of second opinion, “which can encourage a reconsideration,” says Reinier Crane, intensivist (specialized IC doctor) in the OLVG.

Since last summer, the model has been used in the Amsterdam UMC as part of the research phase. It has now left that phase and the OLVG and the Maasstad Hospital in Rotterdam have purchased the first licenses. The software has now been in use at the OLVG for two weeks. The Maasstad Hospital will also start using it in the coming weeks.

First, the doctor decides, says Crane, then he can test a decision against the software. It indicates the likelihood that a patient will return there within two weeks or die upon discharge from the ICU. Usually that chance is 5 to 10 percent. If the percentage is much higher, “we will not send that person away quickly,” says Crane.

Essential

The software also gives the doctor insight into which factors are taken into account in the software’s decision. This is essential, because the doctor’s assessment may differ from that of the software for a reason. Crane: “That works both ways: you may not have thought of something yourself and the model points to it. There may be a new problem emerging that the model already recognizes, and we don’t yet. But it is also possible that the model, based on the data, says that someone is ready for the less intensive nursing ward, and we ourselves see that someone still has too little muscle for that.”

Read also: this report about the discharge software in the Amsterdam UMC

According to Pacmed co-founder Wouter Kroese, it is important for doctors to realize that the software does not provide a complete answer to the question of whether someone can be discharged from the ICU. “That decision is still with the doctor. Our model only says how likely it is that someone dies shortly after discharge or has to be readmitted. But it can also miss data: a doctor sees the color on the patient’s cheeks.”

The AI ​​model makes its prediction based on all data in the electronic files of recently admitted patients in the relevant hospital. Doctors must place that prediction in the context of the patient. In the case of a patient on kidney dialysis (where a kidney is not working properly and the blood is artificially purified of waste products), the software will include in the assessment that the kidney function is insufficient. Crane: “As a doctor, I can ignore that, because the patient is already being treated for that. You can’t explain that to the model.”

AI software can prevent patients from being in ICU for an unnecessarily long time

According to both Crane and Kroese, the IC is the ideal place to implement artificially intelligent software for the first time. The software makes judgments based on a lot of data – which is continuously collected in the ICU, because all patient values ​​are continuously monitored: blood pressure, heart rhythm, respiration.

According to Crane, one of the advantages is that the software can prevent patients from spending unnecessarily long periods in the ICU. That is bad for the patient – ​​“an ICU admission is drastic and causes a lot of unrest” – but also expensive: a day in the ICU costs between 2,000 and 3,000 euros. Moreover, just like everywhere in healthcare, there are also staff shortages in the ICUs. If fewer beds are occupied unnecessarily, hopefully that will lower the pressure, says Crane. This also applies to readmissions: these can also be prevented, precisely by keeping a patient with a high risk of this in the ICU for a while.

So far – in the past two weeks – no situations have occurred in the OLVG in which the model judged differently from the doctor. In the research phase at Amsterdam UMC, they were “fortunately”, says Kroese. “It is nice that the software and the doctors are on the same page. But ultimately we want to provide new insights to improve the quality of care.”