Abstract
Sepsis is a leading cause of death in the ICU. It is a disease requiringcomplex interventions in a short period of time, but its optimal treatmentstrategy remains uncertain. Evidence suggests that the practices of currentlyused treatment strategies are problematic and may cause harm to patients. Toaddress this decision problem, we propose a new medical decision model based onhistorical data to help clinicians recommend the best reference option forreal-time treatment. Our model combines offline reinforcement learning withdeep reinforcement learning to address the problem that traditionalreinforcement learning in healthcare cannot interact with the environment,enabling our model to make decisions in a continuous state-action space. Wedemonstrate that, on average, the treatments recommended by the model are morevaluable and reliable than those recommended by clinicians. In a largevalidation dataset, we found that patients whose actual doses from cliniciansmatched the AI's decisions had the lowest mortality rates. Our model providespersonalized, clinically interpretable treatment decisions for sepsis that canimprove patient care.