Dynamic Care Unit Placements Under Unknown Demand with Learning
研究医院在患者需求未知且反馈延迟时，如何动态分配护理单元床位，提出一种学习型算法，用真实数据验证其能降低非计划再入院率。
Problem definition: Care units are the facilities where admitted hospital patients receive treatment and monitoring services. This paper studies the problem of deciding which patients to place into the various available care units at any time. To determine placements in practice, hospitals rely on clinicians to discern a patient’s care needs and appropriately trade-off between future demand and limited bed availability. Making the right decisions remains challenging because patients are heterogeneous, and demand is uncertain. Methodology/results: We develop a dynamic resource allocation algorithm to decide unit placements by learning the care needs of different patient types. We model hospital beds as reusable resources and assume decision feedback is not immediately available, but rather delayed for an unknown and random length of time. Lastly, we consider the demand to be unknown and allow patient arrivals to be arbitrarily sequenced for robustness. The applicability of our algorithm is demonstrated with real-patient data from a hospital collaboration, where we evaluate our proposed approach using unplanned readmission rates as the performance metric. From extensive simulations, our results suggest the proposed algorithm tends to outperform several greedy benchmarks as well as a hospital benchmark model. A theoretical performance guarantee for our algorithm is provided to complement the case study. Managerial implications: This paper contributes new insights into designing dynamic decision-making models for hospital admissions operations. Our work presents a simple but effective data-driven support tool to help clinicians trade-off between available bed capacity and a patient’s care needs when making care unit placements. We also demonstrate how our algorithm can support the reduction of unplanned readmissions through improved placement decisions. Funding: This work was supported by National Science Foundation Graduate Research Fellowship Program [Grant DGE 1256260]. Partial support for this research was provided to the first-author (A. Dean) by the National Science Foundation Graduate Research Fellowship under Grant DGE1841052. Any opinion, findings, and conclusions or recommendations expressed in this material are those of the authors(s) and do not necessarily reflect the views of the National Science Foundation. Supplemental Material: The online appendix is available at https://doi.org/10.1287/msom.2022.0260 .