基于强化学习框架的脓毒症抗生素多策略推荐模型  

A Multi-Policy Recommendation Model for Antibiotic Use in Sepsis Based on Reinforcement Learning Framework

作　　者：王嫄 刘安岐 盛梦茹 侯佳佳 赵婷婷[1] 于琦 WANG Yuan;LIU Anqi;SHENG Mengru;HOU Jiajia;ZHAO Tingting;YU Qi(College of Artificial Intelligence,Tianjin University of Science and Technology,Tianjin 300457,China;School of Management Shanxi Medical University,Taiyuan 030001,China)

机构地区：[1]天津科技大学人工智能学院,天津300457 [2]山西医科大学管理学院,太原030001

出　　处：《天津科技大学学报》2025年第2期71-80,共10页Journal of Tianjin University of Science & Technology

基　　金：国家自然科学基金项目(61976156);天津市科技特派员项目(20YDTPJC00560)。

摘　　要：脓毒症是全球人类几大死亡原因之一,而抗生素是脓毒症治疗的重要一环。近年来,研究人员认为医疗决策问题可以映射为马尔科夫决策过程,并使用强化学习方法进行治疗策略推荐。结合基于值函数和基于策略的强化学习方法构建多策略推荐的模型框架,对脓毒症治疗过程中抗生素的使用进行策略推荐。针对脓毒症患者的特征信息划分不同的决策区域,应用多策略模型进行个性化治疗建议。结果表明:多策略选择模型能够使患者预后良好的情况达到80.32%。通过统计分析决策轨迹和药物作用选择,模型能够提供符合临床实践的合理药物建议,推荐合适的抗生素组合,改善患者的预后效果。Sepsis is one of the leading causes of human death worldwide,and antibiotics are an important part of sepsis treatment.In recent years,researchers have considered medical decision-making problems as Markov decision processes and used reinforcement learning methods for treatment strategy recommendations.In this article we propose a multi-policy recommendation framework combining value-based and policy-based reinforcement learning methods for antibiotic use in sepsis treatment.Different decision regions are defined based on patient characteristic information,and the multi-policy model provides personalized treatment recommendations.The results show that our multi-policy selection model can achieve a good prognosis for patients in 80.32% of cases.Through statistical analysis of decision trajectories and drug action selection,our model can provide reasonable drug recommendations in accordance with clinical practice,and recommend appropriate antibiotic combinations to improve patient prognosis.

关 键 词：强化学习 医疗决策 脓毒症 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]