Data-driven patient safety precision medicine: accelerating discovery and clinical use

Healthcare-associated infections (HAI) are the most common adverse events and are estimated to attribute to 1500 deaths and 750 000 extra hospital days yearly in Sweden at a cost of 6.5 billion SEK. At least 10-20% of HAI could be avoided if risk patients were identified early.

The aim of the project is to investigate individual and structural risk factors for HAI and to develop fully-automated predictive screening algorithms for early detection of HAI risk.

Also, we will develop electronic decision support systems for early detection coupled with structured interventions, and investigate user-experience and work-flow integration of these systems as well as effects on patient outcome.

Experiments are performed using state-of-the-art machine learning techniques and natural language processing in a unique research database consisting of structured and free text data from >19 million healthcare episodes at Karolinska University Hospital. End-user tests and clinical trials will be performed at hospitals in Region Västerbotten and Stockholm.

In the short-term the project will increase the understanding of the causes and trajectory of adverse events, primarily HAI, and how this can be used to develop decision support systems for prevention and improved patient safety.

In the long-term the project will move the field of AI application on structured and free-text electronic health record data forward and contribute to how these systems interact with healthcare users.