PreSISe-1 - Prehospital Decision Support for Identification of Risk of Sepsis

Important results from the project

The project aimed to develop and evaluate an AI-based decision support for early - prehospital - identification of sepsis risk. Different versions of this functionality were developed, tested, visualized and integrated into an existing decision support and record system (Aweria Prehospital) and evaluated by paramedics in full-scale simulations. The work actualized a large number of integrity-related, clinical, regulatory and legal issues which, based on the project´s work, were compiled in the report " AI as a medical device".

Expected long term effects

One of the AI models developed succeeded in distinguishing patients with severe sepsis from patients with milder ones (accuracy 74%, specificity 79%, sensitivity 61%). However, there was a lack of sufficient comparison data to train and test the algorithm for distinguishing patients with sepsis from other patients. The benefit of early indication that an increased risk of severe sepsis is present was assessed by ambulance nurses as great, and the proposed design on interpretation visualization and integration into an existing system was perceived as clear and easy to use.

Approach and implementation

The approach of developing algorithms from several datasets, visualizing and integrating into an existing clinical system, and tests in full-scale simulations worked well overall. Aggravating factors that also delayed the project were the introduction of the GDPR shortly before the project start, with subsequent uncertainty about how this would be applied, as well as the covid-19 pandemic, which meant that the simulations had to be reduced. However, the biggest challenges were quality and access to prehospital data, which is critical to address if AI is to benefit prehospital care.