Research project Data-driven precision medicine for improved infection control and antibiotic stewardship
Healthcare-associated infections and unnecessary antibiotic treatment are serious problems in healthcare. The project aims to develop AI models, including large language models, to help improve infection control and antibiotic stewardship.
In this project, we use machine learning to create individual risk profiles and predictive models. We aim to prevent healthcare-associated infections, which are infections that occur in hospitalized patients. By analyzing large amounts of electronic health records data (>19 million care episodes at Karolinska University Hospital), the goal is to develop models that can be used to reduce unnecessary antibiotic treatment and predict which patients will develop infections. To validate our research results, we will evaluate the predictive performance of the models on patients in Region Västerbotten, Sweden.
Models will be based on both structured data and clinical text using machine learning and large language models. We also develop methods to explain the models’ predictions with the aim of facilitating trustworthy AI in healthcare.
The project is at the forefront of research in data-driven patient safety work. It can lead to reduced antibiotic use, reduced incidence of healthcare-associated infections and reduced antibiotic resistance.
Project members
Project managers
Aron Henriksson
Senior Lecturer, Associate Professor
Pontus Naucler
Professor
Suzanne Ruhe-van der Werff
Anders Johansson
Professor
More about this project
The objective is to improve our understanding of why patients develop healthcare-associated infections and to develop innovative measures to prevent these infections. We follow individual patient trajectories within the hospital and develop individualized risk profiling using data science and machine learning methods, to reduce unnecessary antibiotics, antimicrobial resistance, and development of healthcare-associated infections. The project will advance methodological knowledge of data-driven patient safety research.
Specific aims:
- To investigate factors contributing to incorrect antibiotic use and to develop prediction models to avoid unnecessary antibiotics.
- To develop fully automated algorithms and prediction models for healthcare-associated infections.
- To investigate the effects of implementation of fully automated surveillance systems for healthcare-associated infections in real-world settings.
The impact of the project will be an increased understanding of the causes and trajectory of adverse events and how this can be used to develop decision support systems for prevention and improved patient safety. Also, it will move the field of AI applications on structured and free-text electronic health records data forward, including the use of large language models in healthcare.