A Dosing Strategy Model of Deep Deterministic Policy Gradient Algorithm for Sepsis

Abstract

BackgroundA growing body of research indicates that the use of computerized decision support systems can better guide disease treatment and save resources. An artificial intelligence (AI) clinical decision system may be a useful tool to predict how different dose combinations affect survival outcomes in patients with sepsis. We boldly use Deep Deterministic Policy Gradient (DDPG) algorithm to improve the sepsis AI clinical decision system developed by MIT and IMPERIAL College London (ICL) team. We believe that although the system has been verified by data, due to the limitations of the core algorithm, it cannot meet the needs of clinical use, it could also be that the model of training is not stable. The purpose of this study was to establish a more effective AI clinician decision-making system to guide sepsis treatment.MethodsWe chose the same data model as the MIT team for experiments. these datasets are from the publicly available Multiparameter Intelligent Monitoring in Intensive Care (MIMIC-III v1.4) database. We utilized hospitalization information for 38,600 adult patients over 15 years of age。we adopted DDPG (a strategy-based reinforcement learning method) to build an artificial intelligence clinician model.ResultsCompared with the sepsis AI clinician decision model developed by MIT based on DNQ algorithm, the sepsis AI clinician decision model developed by DDPG algorithm had a faster convergence rate, was closer to the clinician decision, and had a higher survival rate.ConclusionsThe AI clinician decision system based on DDPG algorithm can better help the ICU (Intensive Care Unit ) clinicians to cope with the changes in the condition of sepsis patients, reduce the workload, and provide a reference dosage.