Brief

Researchers have developed a new framework called Clinical Narrative-informed Preference Rewards (CN-PR) to learn reward functions for reinforcement learning in healthcare. This method extracts trajectory-level preferences from clinical narratives, such as discharge summaries, to supervise the learning process. The CN-PR framework was evaluated in dynamic sepsis treatment, showing that policies learned with this approach led to improved recovery outcomes and comparable mortality rates to existing methods. AI

Researchers have developed DeepEN, a deep reinforcement learning framework designed to personalize enteral nutrition for critical care patients. Trained on data from over 11,000 ICU patients, DeepEN generates tailored 4-hourly targets for calories, protein, and fluids. The framework demonstrated a significant reduction in estimated mortality and improved metabolic stability compared to standard clinical practice. AI