How AI is used for care planning in the ICU

Background

Despite the digitalized world in which we live, planning and resource allocation in Swedish intensive care is often dependent on manual tools such as whiteboards, binders and Excel sheets. In Region Västerbotten, a need was identified to improve care logistics and free up time for care staff. Care burden and care time, two crucial parameters for planning, were assessed subjectively, which created inefficiency and uncertainty in resource allocation.

Solution

The region initiated an AI project in which large amounts of patient data from three ICU departments were collected and analyzed. With over 137,000 care events and 13,000 patient variables, an AI model was trained using Random Forest Survival Analysis, an algorithm suitable for predicting time-to-event. The model calculates care time and care burden for each patient and presents predictions in a web application updated daily at 7am.

The application shows:

• Discharge predictions via a traffic light system.

• How many patients are expected to remain in hospital tomorrow, and in the days to come.

• Care burden per patient.

Results

The AI model has proven to be highly accurate with up to 80% accuracy in individual discharge predictions. It has been validated against real-world data and compared to assessments from doctors and nurses, where the model outperformed in several respects. The result is more objective and efficient planning, enabling staff to focus on care instead of administration.

The project has also led to:

• Cost efficiency through better resource utilization.
• Improved quality of care through more accurate decisions.
• Scalability; today there are approximately 10 AI models in operation, with associated applications.

Future Vision

The region is now exploring the possibility of sharing its methods and models with other regions without compromising patient confidentiality. This could enable national collaboration and thus improve care for patients across the country

In our new system, technology supports, rather than replaces, clinical expertise, strengthening trust and improving everyday working conditions in the ICU.