Artificial intelligence Two hospitals formally give algorithms an advisory role in medical decisions.

Liza van Lonkhuyzen, January 28, 2020

https://www.nrc.nl/nieuws/2020/01/28/arts-krijgt-advies-van-algoritmen-a3988515

Photo: Flip Franssen

In the Reshape center, the innovation department of the UMCRadboud, employees monitor patients remotely.

The computer will advise doctors: can a patient be safely ‘discharged’ from intensive care?

The Amsterdam UMC, location VUmc, is putting this into practice now that the medical device regulator has given researchers the green light. Artificial intelligence is used to calculate the likelihood that a patient will end up in intensive care again after a transfer or die. Doctors can be advised in this way before they choose who is healthy enough to be accommodated in other departments.

The algorithm looks at two hundred factors. This includes, for example, laboratory results and notes from nursing staff. Doctors can choose to read out the most important factors in the advice so that they can understand what the computer is based on.

Protocol

Along with the approval for the VUmc, another hospital also took an important step with an ambitious data project. In the Radboudumc in Nijmegen, computer scores that say something about the health of patients become part of the protocol. The hospital wants to see problems coming before the patients feel sick.

The Nijmegen hospital has set up a former laboratory room as a kind of control room. There, on monitors, a lot of information is kept from the patients of two nursing departments: whether they are sitting upright in bed, lying down, how their heart rate is, their temperature, breathing, and so on.

With this data stream – which comes from small boxes on the wrist of patients and stickers on the body – an algorithm searches for connections. For example, if the heart rate rises and the respiratory rate increases, a respiratory tract infection could play a role. The computer searches for signals of blood poisoning, among other things.

Now that research from the hospital shows that continuous monitoring in itself already results in one third fewer transfers to intensive care, the way is clear to investigate whether even more can be gained with the ‘predictive’ computer scores. Doctors and nurses are required by the new protocol to examine the computer’s alarm signals.

Artificial intelligence

In the healthcare sector, much attention is being paid to the possibilities of artificial intelligence. Many hospitals and care institutions do experiments. The Leiden University Medical Center, for example, is working on a system that automatically recognizes patient complaints based on speech recognition and artificial intelligence in the doctor’s office and eventually even records them in the electronic patient file. The Pacmed data company, which also worked on the intensive care algorithm, developed software that should help estimate which men are more likely to have tumors after prostate cancer treatment.

There are many questions to be asked about this trend. For example, the predictions of the computer depend on the underlying data. If data is not entered correctly in the computer, it may come to the wrong conclusion. In addition, an algorithm only looks at what is measurable. What a doctor sees himself, in the doctor’s office or at the bedside, must be just as important, experts say.