LUMC uses AI to predict infections after surgery
The LUMC introduces PERISCOPE, an advanced AI model that predicts the risk of infections after surgery. This tool helps doctors and nurses to act more quickly and thus better prevent complications in patients. By analysing important patient data, PERISCOPE provides accurate predictions and supports decision-making in clinical practice.
How does PERISCOPE work?
PERISCOPE is an AI model that analyses historical patient data to predict who is at risk of developing an infection after surgery. It works based on information such as previous infections, other conditions such as diabetes, and vital signs such as heart rate, weight, and blood pressure. By combining this data, the model calculates the risk of infection and displays it as a percentage and a risk category (low, medium, or high). This allows healthcare providers to quickly see which patients need extra attention and take more targeted action to prevent complications.
Background and validation
-
Research and development
The idea for PERISCOPE arose from surgeons' desire to better predict which patients are at risk of infection after surgery. PhD researcher Siri van der Meijden played a central role in this. Together with her team, she collected data from almost ten years of electronic patient records and determined which factors were important, such as previous infections, other conditions (e.g. diabetes) and vital parameters such as heart rate and blood pressure. After thorough analysis, PERISCOPE was developed and prepared for use in practice.
-
Testing and reliability
After development, PERISCOPE was thoroughly tested using data from more than 250,000 operations in three hospitals: the LUMC, Radboudumc and the East Limburg Hospital in Belgium. At the LUMC, five hundred predictions made by the AI model were compared with those made by doctors. This showed that PERISCOPE performed at least as well as experienced doctors and even scored better when doctors were uncertain or had less experience. This demonstrated that the model provides reliable support for decision-making in clinical practice.