题    目：Optimizing ICU Antibiotic Usage with Antimicrobial Resistance Considerations

嘉　宾：谢金贵 教授 慕尼黑工业大学

主持人：曾智宇 助理教授 上海交通大学安泰经济与管理学院

时　间：2026年9月2日（周三）10:00-11:30

地　点：上海交通大学安泰经济与管理学院包兆龙图书馆A507

内容简介：

Antibiotic overuse is widespread in intensive care units (ICUs), where prolonged therapy is often prescribed to reduce short-term treatment failure risk. Such practices contribute to antimicrobial resistance (AMR) and worsen future treatment outcomes. We study how health organizations should determine population-wide antibiotic stewardship guidelines, focusing on the optimal average duration of therapy measured in days of therapy (DOT) while accounting for the long-term impact of AMR. We develop a two-stage hierarchical optimization framework with a constrained Markov decision process (MDP) to jointly determine the optimal DOT target and the corresponding daily antibiotic discontinuation policy, with the objective of minimizing treatment failure risk. Using occupation-measure and dual analyses, we establish that the optimal lower-level policy has a state-dependent threshold structure, with randomization required in at most one state--epoch pair. We further show that when the DOT constraint is nonbinding, prolonged therapy may remain optimal under certain conditions, providing a theoretical explanation for overtreatment in practice. At the upper level, we show that the DOT optimization problem is convex and piecewise linear under mild conditions, yielding an optimal target that balances marginal treatment benefit against marginal resistance cost. Using real-world ICU data, we demonstrate that the proposed policy can simultaneously reduce both treatment duration and treatment failure rates. We further extend the framework to incorporate initial-state-dependent DOT targets, continuation costs, and multi-dosage optimization, and show that the main structural insights continue to hold in these more general settings. Our results provide a quantitative framework for designing antibiotic stewardship policies that balance short-term clinical outcomes with long-term AMR considerations. The threshold-based policies are interpretable and operationally implementable, while the optimized DOT targets provide practical guidance for health organizations seeking to control antibiotic overuse without compromising patient outcomes.

演讲人简介：

Prof. Dr. Jingui Xie is a W3 Professor at the Technical University of Munich (TUM) and holds the Dieter Schwarz Stiftung Associate Professorship of Business Analytics. His research focuses on data-driven decision making under uncertainty, with applications in healthcare operations and service systems. His work combines operations management, stochastic modeling, reinforcement learning, optimization, and causal inference to improve decision-making in high-stakes service systems. Professor Xie has published in leading journals including Management Science, Operations Research, Manufacturing & Service Operations Management, and Production and Operations Management.

欢迎广大师生参加！