INFORMS 2021 Program Book

INFORMS Anaheim 2021

MC22

3 - Implications of Climate Change for Decarbonized Electricity System Planning: Examples from California Brian Tarroja, Professional Researcher, University of California, Irvine, Irvine, CA, United States Regional electricity systems are evolving to incorporate more zero-carbon energy resources and transform the infrastructure that underpins such electric grids. To ensure that these efforts are successful, these must be adapted to account for how climate change affects regional electricity supply, demand, and infrastructure. Here, we provide examples of how climate change affects decarbonized electricity system planning through effects on hydropower generation, water availability for thermally-based electricity resources, and building electricity demand. California is used as an example due to its combination of susceptibility to drought, temperature extremes, and policies to decarbonize their electricity system. Further, we explore the effectiveness of different solutions to mitigate undesirable impacts on decarbonized electricity system planning. MC19 CC Room 203A In Person: Emergency Medicine Operations? General Session Chair: Saharnaz Mehrani, The George Washington University, Washington, 20052, United States 1 - A Machine-learning Framework for Addressing Emergency Department Crowding Problem Abdulaziz Ahmed, University of Minnesota-Crookston, Crookston, MN, United States, Omer Ashour We develop a machine learning framework for predicting whether a patient is hospitalized or discharged based on the patient’s information. Such information is vital signs, demographic data, and the complaints a patient presents while arriving in an emergency department. 2 - Dynamic Coordination of Exams in a Radiology Practice Saharnaz Mehrani, University of Connecticut, Storrs, CT, United States, Miao Bai, David Bergman, Carlos Henrique Cardonha We study dynamic coordination of exams on multiple diagnostic machines in a radiology practice with inpatients, outpatients, and patients from emergency department. There is stochasticity in both patients’ itinerary and radiologic care. There are different costs per unit wait time associated with different patient urgencies to receive radiologic care. Our goal is to minimize total expected wait cost in one day. We formulate the problem as a Markov decision process and adopt an approximate dynamic programming algorithm to solve it. We evaluate the performance of our model and algorithm on a real-world problem and show that the resulted policy outperforms two baseline heuristic policies. MC20 CC Room 203B In Person: Data Analytics in Opioids Use/Misuse General Session Chair: Sujee Lee, Soongsil University, Seoul, 06978, Korea, Republic of 1 - Data-driven Models for Identifying Risk Factors Leading to Opiate Abuse Jinha Lee, Bowling Green State University, Bowling Green, OH, 43403-0154, United States, Arthur Yeh, Qizhen Lan, Jung Im Choi, Hyojung Kang Drug addiction, abuse, and overdose deaths have become the most pressing public health issue in the U.S. Understanding drug abuse and overdose patterns from a geo-spatial framework can empower communities to develop strategy for responding to the drug abuse based on where incidents take place. Therefore, it is imperative to identify behavioral and socioeconomic factors that affect community level and subsequently develop a geo-spatial model to assess drug abuse risk. The essential idea is to harness the potential of data analytics to identify the behavioral geo-spatial model that identifies risk factors in local communities given drug abuse-related socioeconomic features, and to develop optimal data-drive strategies and guidelines for minimizing abuse. 2 - A Machine Learning Integrated Opioid Prescription Optimization Framework Sujee Lee, Soongsil University, Seoul, Korea, Republic of, Philip A. Bain, Jingshan Li We propose a framework that integrates machine learning and optimization models to determine the optimal amounts of opioids in the initial and subsequent prescriptions. For this purpose, the amounts of opioids consumed by total joint placement (TJR) patients in SSM Health, Madison, WI were investigated through patient surveys. In the framework, the machine learning model is trained to estimate the opioid demand level for each patient. Then, the proposed optimization model minimizes the expected opioid leftovers as well as the

number of opioids refills to determine the optimal amount of opioid prescription for each demand level. The resulting prescription decisions are compared with the current practice in SSM Health. The results prove that the model can help reduce opioid leftovers, without increasing the burden of hospitals and patients.

MC21 CC Room 204A In Person: Recent Advances in Multistage Stochastic Programming General Session Chair: Harsha Gangammanavar, Southern Methodist University, Dallas, TX, 75275, United States 1 - Multistage Stochastic Programming with Optimal Stopping Rui Peng Liu, Georgia Institute of Technology, Atlanta, GA, United States The theory of optimal stopping finds interesting applications in house selling, one- armed bandit, option trading, etc. In this talk, we present a general formulation of multistage stochastic programming that incorporates optimal stopping. This formulation can be solved, as usual, by writing down the Bellman equation and applying dynamic programming. Our focus will be on applications and computational aspects of the formulation. 2 - Risk Bounds and Rademacher Complexity in Batch Reinforcement Learning Yaqi Duan, Princeton University, Princeton, NJ, United States We consider batch Reinforcement Learning (RL) with general value function approximation. Our study investigates the minimal assumptions to reliably minimize Bellman error, and characterizes the generalization performance by (local) Rademacher complexities of general function classes. Concretely, we view the Bellman error as a surrogate loss for the optimality gap, and prove: (1) In double sampling regime, the excess risk of Empirical Risk Minimizer (ERM) is bounded by the Rademacher complexity of the function class. (2) In the single sampling regime, sample-efficient risk minimization is not possible without further assumptions, regardless of algorithms. However, with completeness assumptions, the excess risk of FQI and a minimax style algorithm can be again bounded by Rademacher complexities. (3) Fast statistical rates can be achieved by using localization. MC22 CC Room 204B In Person: Healthcare Policy and Regulation General Session Chair: Anqi Wu, University of Illinois, Champaign, IL, 61821, United States 1 - The Impact of Uncertainty Avoidance Culture on Patient Engagement During the Covid-19 Pandemic Kellas Cameron, Assistant Professor, University of South Florida, Tampa, FL, 33602, United States, Lu Kong National Cultures have played an undeniable role in how different countries have been able to effectively tackle the Covid-19 pandemic. We posit the countries that exhibit a higher uncertainty avoidance index had better responses due to their population’s willingness to adapt to new social and health directives, as this risk averse nature has been associated with higher patient engagement. Our model demonstrates how these dimensions of national culture as opposed to individual preferences directly impacted the implementation of novel mitigation processes to lower infections rates. We outline ways in which those in health operations can leverage these cultural norms to maximize positive patient health outcomes. 2 - Healthcare Reimbursement Policy Impact on Multiple-provider Readmission Reduction Programs Jon M. Stauffer, Texas A&M University, College Station, TX, 77843, United States, Jonathan Eugene Helm, Kurt M. Bretthauer We examine the transition from Fee-for-Service to alternative reimbursement plans, such as bundled payments and the Hospital Readmission Reduction Program, and how this impacts the motivation for providers to reduce readmissions. Results show that bundled payment plans do motivate cost- effective readmission reduction effort from hospitals, but if post-discharge providers are included in the gain-sharing contracts they may perform insufficient or excessive effort. We discuss two redesigned structures (single-controlling provider and risk-adjusted) for bundled payment plans to address these readmission reduction effort misalignment issues.

63