Researchers at the Icahn School of Medicine at Mount Sinai have developed an AI-based tool that could improve blood sugar management in patients recovering from heart surgery. The tool, called GLUCOSE, uses a type of artificial intelligence known as reinforcement learning to recommend personalized insulin doses in real time.
After cardiac surgery, patients are vulnerable to dangerous fluctuations in blood sugar levels. Traditional insulin dosing protocols often struggle to adapt to the rapidly changing conditions in the ICU. GLUCOSE addresses this by learning from real-world ICU data and tailoring recommendations to each patient’s unique situation.
In tests, the model matched or outperformed experienced doctors at keeping glucose levels within a safe range—even though it only used current patient data, while doctors had access to full histories. Importantly, the researchers stress that the tool is designed to support, not replace, clinicians.
GLUCOSE was trained using conservative and distributional reinforcement learning, ensuring it offers safe and reliable advice. The team envisions future integration into electronic health records to deliver insulin guidance automatically.
Although the tool does not yet account for nutrition, its success using limited data highlights its potential to improve safety and efficiency in ICU care. The ultimate goal is to enhance provider decision-making and improve outcomes for high-risk patients.
Article written by Technology network team
19/05/2025
Source:
Mount Sinai