ERC grants for research on AI in emergency care

There is immense potential for AI-based clinical decision-support systems, says Johan Sundström, cardiologist and professor of epidemiology.

Emergency doctors must make rapid life-changing decisions in a chaotic environment, based on limited information. Risks of important diagnoses being missed, or of patients having unnecessary examinations, waiting times and hospital admissions, are considerable. There is therefore immense potential for AI-based clinical decision-support systems, says Johan Sundström, cardiologist and professor of epidemiology.

“Artificial intelligence (AI) has an enormous capacity to master the ability to recognise patterns, by learning from millions more patients than an individual doctor can meet in a whole career.”

Focus and concerted strength

Sundström is receiving an Advanced Grant of SEK 25 million from the European Research Council (ERC) for the five-year project “Enhancing emergency department safety, efficacy and cost-effectiveness by artificial intelligence”.

“The grant means a tremendous lot. Now we can focus, concert our strength and get results more rapidly. Our work is interdisciplinary, and we have multiple scientific problems to solve. The idea is that, ultimately, we’ll help emergency doctors, and the grant will enable us to integrate every component of the work and test how it works in reality.”

The scientists are going to develop AI-driven algorithms that estimate risk and are trained on health data from up to six million patient visits to accident and emergency (A&E) departments. Parameters include, for example, electrocardiograms (ECGs), reasons for visits, vital indicators, previous disease history and self-reported symptoms.

“We’re also focusing on new types of data, and we’ve developed a tool for depicting one’s own pain. Pain is the most common reason for A&E visits, but it’s hard to describe and document.”

Quick decisions

Data of this type are suitable for AI, which is extremely good at pattern recognition. In particular, machines have the upper hand when it comes to rapidly processing huge quantities of data,” Sundström explains.

“As an A&E doctor, you may have just two minutes to read a patient’s medical records and get an idea of the situation before the ambulance arrives. It’s a very limited window for a person, but for AI two minutes is a long time.”

The plan is for the technology to have undergone clinical testing within the five-year project period. But there are some scientific problems to be resolved before this. For example, there are gaps or holes in the data collected from A&E departments. A patient admitted with a foot injury gets no ECG, for example.

Major scientific advance

Another challenge is to construct AI models that use varying types of data in a single system that can provide a coherent answer.

“These are difficult scientific challenges, so we’re excited about how well it’s going to succeed. But the ERC grants go to projects with high risk and high potential. If it goes well, the scientific advances will be quite big,” Sundström says.

The scientists who are to work on the project belong to his research group in clinical epidemiology at the Department of Medical Sciences. In part, the research will be conducted at the Anders Wiklöf Institute for Heart Research, which shares premises with AI4Research, Uppsala University's AI initiative.