By Peter Witten, AAU Communication and Public Affairs
Photo: Colourbox

For ICU patients, large fluctuations in blood glucose levels can be life-threatening, especially for those with diabetes.

Now, a research project from Aalborg University/AAU shows that, with the help of AI and existing data, it is possible to more quickly predict which patients are at risk of experiencing dangerously high or low blood sugar levels. This enables doctors to intervene in time, potentially saving more lives.

Better predictions

“We have shown that our AI model can improve the prediction of blood glucose levels in ICU patients by 5-7 percent compared to the best existing prediction models,” says Associate Professor Arijit Khan from the Department of Computer Science at AAU.

Together with PhD student Mohammad Hadi Mehdizavareh, also from Computer Science, and Associate Professor Simon Lebech Cichosz from the Department of Health Science and Technology, he is behind the research project.

The AI system has analyzed data from over 200,000 ICU stays across 208 hospitals in the United States.

“ICU patients are constantly monitored using various equipment, such as blood pressure and heart rate monitors. At the same time, doctors and nurses observe them closely. We have looked at all available information, and with AI, the data is used to predict what will happen to the patients in the next hour, the next five hours, and so on,” explains Arijit Khan.

800 million affected

The World Health Organization (WHO) estimates that over 800 million people worldwide have diabetes, the vast majority with type 2. If they experience high or low blood sugar levels while admitted to an ICU, it increases the risk of complications and death.

The AI system developed by the three AAU researchers can analyze patient data in real time using information that is already available - for example from blood pressure measurements, lab results, and observations and records made by doctors and nurses.

Powerful tool

"Although we tested the system on blood glucose prediction, the same approach can be adapted for other medical challenges and for data from hospitals in different countries. This flexibility makes it a powerful tool that can be tailored to support many types of decisions in intensive care,"  says PhD student Mohammad Hadi Mehdizavareh. 

"While the results are very promising, we still need to ensure that AI models integrate seamlessly into the clinical workflow and that their predictions are transparent and understandable to doctors and nurses. Only then can we fully realize their potential to improve patient outcomes." adds Associate Professor Simon Lebech Cichosz from the Department of Health Science and Technology.

Millionfunding

The AAU researchers also plan to investigate whether the new AI system can be used to predict patients’ length of stay at the hospital and their risk of blood clots.

The research project spans five years and is funded by the Novo Nordisk Foundation with a grant of 10 million DKK.