Learning Optimal Treatment Strategies for Sepsis Using Offline Reinforcement Learning in Continuous Space

Sepsis is a leading cause of death in the ICU. It disease requiring complex interventions short period time, but its optimal treatment strategy remains uncertain. Evidence suggests that practices currently used strategies are problematic and may harm to patients. To address this decision problem, we propose new medical model based on historical data help clinicians recommend best reference option for real-time treatment. Our combines offline reinforcement learning deep solve problem traditional field due inability interact with environment, while enabling our make decisions continuous state-action space. We demonstrate that, average, treatments recommended by more valuable reliable than those clinicians. In large validation dataset, find out patients whose actual doses from matched made AI has lowest mortality rates. provides personalized clinically interpretable sepsis improve patient care.