The data doctor who wants to use artificial intelligence to improve healthcare
When can a patient be discharged from intensive care? Artificial intelligence can help answer this and other questions. But using such revolutionary technology in healthcare is not easy, as Anne de Hond has discovered.
Doctors want to transfer patients from an intensive care unit (ICU) to another department at exactly the right moment. Neither too soon, because then the patients are more likely to die or have to return to the ICU. Nor too late, because a stay in the ICU is very expensive and no one wants to be there for any longer than necessary. For her PhD research at the LUMC, De Hond studied how artificial intelligence can help with this decision.
De Hond used an existing algorithm and tested it with data from the LUMC to ensure it would work properly there. This meant data on the patient’s vital signs such as temperature, heart rate and blood pressure. Once it has been trained, the model will be able to be used in the hospital’s ICU. This does not mean of course that patients will no longer receive bedside visits from their doctors. ‘It’s a support tool. The final decision is always up to the doctor’, De Hond explains.
Trust in new technology
This is why doctors are closely involved with the ICU algorithm and an algorithm that De Hond developed together with LUMC’s IT department. This new algorithm predicts whether patients in the emergency department waiting room will be admitted or allowed to return home. ‘This will enable us to triage in the waiting room and anticipate whether a patient will be admitted to hospital’, says De Hond. ‘This will speed up the throughput.’
‘A lot of work is needed to convince doctors that artificial intelligence can be a good tool.’
The waiting room algorithm is now available in emergency departments, which De Hond says is a major milestone. ‘It is very difficult to implement a new technology such as this.’ First, it is difficult to get the data to train artificial intelligence because this data is often stored in different places. Second, it takes a lot of manpower from IT professionals, technicians and clinicians alike to implement and maintain. ‘We often underestimate that’, says De Hond. ‘And the technology has to be extremely secure because it processes sensitive personal information.’
Third, doctors have to be willing to put their trust in artificial intelligence. ‘We find that some clinicians are really eager to experiment with us whereas others are very critical’, says De Hond. The algorithm at the LUMC’s emergency department, for example, is not yet in routine use. ‘It will take a lot of work to convince doctors that it can be a good tool.’
Beware of the hype
Although De Hond is excited about the possibilities offered by artificial intelligence, she does not want to get too caught up in the hype. ‘In healthcare, we are all incredibly eager to develop new innovations, but I think we should also invest our energy in techniques that already exist, in putting them into practice. It takes an awful lot of testing to investigate whether a new technique is safe and works well.’
Close to patient care as a data doctor
Once she has completed her PhD, De Hond will continue researching artificial intelligence in healthcare as an assistant professor at UMC Utrecht. ‘I always wanted to work in healthcare and am glad that having studied econometrics I am now getting close to patient care as a sort of data doctor. Using revolutionary techniques to improve healthcare feels really satisfying.’
Text: Dagmar Aarts
Photo: Josje Deekens