INFORMS 2021 Program Book

INFORMS Anaheim 2021

WB23

2 - Evaluation and Mitigation of Estimation Bias of the Bayesian Beta-bernoulli Two-armed Bandit Problem with Binary Responses Amin Yarahmadi, Lancaster University, Lancaster, United Kingdom, Peter Jacko To derive a response-adaptive procedure as an optimal policy to a dynamic optimization problem, we consider the Bayesian Beta-Bernoulli finite-horizon two-armed bandit problem in which binary responses are modelled as Bernoulli random variables and the objective function is defined by maximising the Bayes- expected total number of patient successes in the trial which we call the patient benefit. We first evaluate the per-treatment bias of Maximum Likelihood Estimator (MLE) showing that it is unacceptably high and variable. We propose a new augmented estimator with the aim to mitigate the estimation bias. Using simulations we show that the proposed estimator can be tuned to provide results in which estimation bias and its variability is notably improved. We also investigate several novel modifications of the dynamics of the bandit problem to obtain further improvements. 3 - Adaptive Seamless Dose Finding Clinical Trials Amin Khademi, Clemson University, Clemson, SC, 29634, United States, Ningyuan Chen In this work, we study the adaptive design of dose-finding Phase II clinical trials, by simultaneously considering efficacy and toxicity. We formulate this problem as a non-parametric bandit problem and propose two policies. The first one is based on dose escalation principles and the second one is based on bisection search and UCB algorithms. We test the performance of these algorithms along with benchmarks on synthetic and real datasets. WB20 CC Room 203B In Person: Health Care I Contributed Session Chair: Hui Jia, University of Tennessee, Knoxville, TN, 37916, United States 1 - Detection of Blood Clots Within Pulmonary Microcirculation Following E-cigarette Exposure in Mice Reza Iranzad, University of Arkansas, Fayetteville, AR, United States, Xiao Liu, Margaret Bennewitz, Hunter Snoderly Edge detection of medical imaging data plays a vital role that helps radiologists, pathologists and improves diagnostic accuracy. In this context, an algorithm is demonstrated through the edge detection of imaging data collected during a medical experiment on mice lungs when exposed to E-cigarette. The edge detection algorithm uses feature engineering coupled with a tree-based ensemble model for medical imaging data. The process helps extract features from raw images and utilizes these features to improve edge detection performance. 2 - The Impact of Economic Insecurity on Covid-19 Mitigation Efforts Kellas Cameron, Assistant Professor, University of South Florida, Tampa, FL, United States, Deepti Singh Due to the way that the US federal government delegated the effort to mitigate the impact of the Covid-19 pandemic to state governments, we saw how various prevention methodologies significantly impacted both state infection rates and economic impacts. Different lockdown protocols, social distancing mandates, and mask requirements, implemented over the three waves of the pandemic were posited to have different impacts dependent on state culture, climate, and economic stability. Our work compares three economically similar states Texas, California, and Florida and demonstrates how a state’s view of economic insecurity significantly drove infections rates and economic recoveries. 3 - Hospital-physician Integration and Cardiac Surgery Outcomes: A U-shaped Relationship? Hui Jia, University of Tennessee, Knoxville, Knoxville, TN, United States We utilize patient-visit level information for Florida patients hospitalized for coronary artery bypass graft (CABG) to test hypotheses that posit a U-shaped association between integration and care outcomes such as patient length of stay (LOS), in-hospital mortality risk, and readmission risk. This study defines the level of integration between a hospital and its cardiovascular surgeons as the fraction of surgeons who operate only at that hospital. Our econometric analysis indicates that patient LOS and mortality risk are minimized at integration tipping points of 51% and 45%, respectively.

4 - Is Supply Chain Diversion Fueling the Opioid Crisis? Evidence From a Quasi-Experiment Jingwen Yang, PhD Candidate, University of Minnesota-Twin Cities, Minneapolis, MN, United States This study investigates the impact of supply chain diversion on prescription opioid abuse. It relies on the enactment of the Drug Supply Chain Security Act (DSCSA) as an “anti-diversion” exogenous shock that gives rise to a quasi-experimental design. Exploiting cross-state variations in opioid diversion exposure, the study finds that after the enactment of the DSCSA, states with higher initial prescription opioids seizure rates experience larger decreases in prescription opioid abuse. Such decreases are significant among the female, younger-aged and white populations. The study further explores a potential diversion mechanism and the most targeted prescription opioids. WB22 CC Room 204B In Person: Empirical and Behavioral Research in Service Operations General Session Chair: Hyun Seok (Huck) Lee, Korea University Business School, 97333-3235 1 - The Gatekeeper’s Dilemma: When Should I Transfer This Customer? Maqbool Dada, Johns Hopkins Carey Business School, Baltimore, MD, United States, Evgeny Kagan, Brett Hathaway In many service encounters front-line workers (often referred to as gatekeepers) have the discretion to attempt to resolve a customer request, or to transfer the customer to an expert service provider. We study the gatekeeper’s transfer decision analytically and experimentally. Our experimental results offer mixed support for rational model predictions and advance our understanding of cognitive capabilities and rationality limits on human server behavior in queueing systems. WB23 CC Room 204C In Person: Analytics in Service Operations General Session Chair: Lennart Baardman, University of Michigan, Ann Arbor, MI, 48103, United States 1 - Taylor Approximation of Data-Driven Inventory Policies for Distribution Systems with Feature Information Kevin Shang, Duke University, Fuqua School of Business, Durham, NC, 27708-9972, United States This paper studies a pricing problem for a single-server queue where customers arrive according to a Poisson process. For each arriving customer, the service provider announces a price rate and a system wait time, and the customer decides whether to join the queue and the duration of the service time. The objective is to maximize either the long-run average revenue or social welfare. We formulate this problem as a continuous-time control model whose optimality conditions form a set of delay differential equations. We develop an innovative method to obtain the optimal control policy, whose structure reveals new insights. In particular, the provider should compensate the customer by lowering the price when the wait time is sufficiently long. In a numerical study, we find that our revenue-maximizing pricing policy can also improve social welfare over the static pricing policy. 2 - Joint Product Ranking and Inventory Planning on Online Platforms Zijin Zhang, University of Michigan, Ann Arbor, MI, United States, Hyun-Soo Ahn, Lennart Baardman In e-commerce, page displays are real estate and how items are shown (e.g., vertical ranking) affects consumer behavior, hence demand. Data shows items placed in top positions substantially receive more clicks or scroll-locks. In this paper, we consider joint inventory planning and product ranking. Specifically, we study analytic models in three separate scenarios: order and rank only once, order and ranking changes over time, and learn to order and rank. We develop an optimal polynomial-time algorithm when order quantities and ranking decisions are made simultaneously. This algorithm and its variant are also proven to be asymptotically optimal when ranking changes over time. Furthermore, we extend it to dynamic settings: real-time optimization and online learning. By analyzing these two extensions, our algorithms remain strong both theoretically and practically.

153