Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

WB57

2 - Data-driven Modeling of a Dynamic and Heterogeneous Contact Network for Understanding the Transmission Behavior of Infectious Diseases Yuan Zhou, University of Texas at Arlington, Arlington, TX, United States Contacts are fundamentally linked to the propagation behavior of human-to- human transmitted diseases. Although several well-known structures of social network have been widely applied in the literature for establishing contacts, such as small-world and scale-free networks, an argument arises concerning the adequacy of such a network in capturing the public contacts that often take place in between individuals who are not socially affiliated. We develop a two-layer network framework for representing both social and public contacts. We will utilize both simulated demographical data that obtained from public data sources and synthetic data that generated for representing human mobility. 3 - A Multi-column Generation Approach for Radiation Therapy Treatment Planning Gazi Md Daud Iqbal, University of Maryland School of Medicine, Baltimore, MD, United States, Jay Michael Rosenberger, Hao Howard Zhang Both intensity modulated radiation therapy (IMRT) and volumetric modulated arc therapy (VMAT) delivery use multileaf collimator to shape the radiation beam in order to achieve modulation. Column Generation approaches have been proposed to generate these shapes (called apertures) to deliver the radiation therapy treatment. Due to large number of candidate columns (feasible apertures), column-generation-based algorithm is computationally expensive, which affects the achievable solution quality within a clinically acceptable time frame. Instead of adding columns one at a time, this research uses a multi-column generation approach to obtain deliverable apertures for both IMRT and VMAT. 4 - Adjusting for Time Varying Confounding with Multiple Treatment Variables Aera Leboulluec, University of Texas-Arlington, Arlington, TX, 76013, United States, Nilabh Ohol, Victoria C. P. Chen, Jay Michael Rosenberger Time varying confounding plays a critical role in longitudinal studies. In medical research, estimating an effect of treatment on an outcome of interest is biased due to presence of time varying confounders. This bias results in inconsistent treatment estimates. Most of literature on handling time varying confounding demonstrates the implementation of methods such as inverse probability of treatment weighting to estimate consistent estimates of a single treatment. This presentation extends this approach to multiple treatments and considers both uncorrelated and correlated treatments. Optimizing Health Policy and Medical Decision Making to Account for Patient Heterogeneity Sponsored: Health Applications Sponsored Session Chair: Diana Maria Negoescu, University of Minnesota, Minneapolis, MN, 55455, United States 1 - Better Infectious Disease Control through Patient Incentive Programs Sze-chuan Suen, University of Southern California, Los Angeles, CA, 90089-0193, United States, Diana Maria Negoescu, Joel Goh To reduce the number of patients failing to complete long antibiotic regimens, public health departments may offer patients financial incentives to remain on treatment. However, it is unclear how incentive programs should be structured for the greatest societal benefit when resources are scarce. We design an optimal incentive schedule in the context of disease transmission, patient incentive compatibility, and resource constraints. We then demonstrate its application to the case of tuberculosis treatment, where a complete treatment regimen can take over 6 months. 2 - Optimal Decision Making in a Markov Model with Parameter Uncertainty: The Case of Chronic Kidney Disease M. Reza Skandari, Imperial College Business School, South Kensington Campus, Ayrton Rd, Kensington, London, SW7 2AZ, United Kingdom, Steven Shechter We investigate a Markov decision process whose unknown transition parameters are revealed partially through state observation. Decisions are made as the state evolves. We use the model to study the optimal time to start preparing a type of vascular access for chronic kidney disease patients who will need dialysis. n WB57 West Bldg 101B

n WB54 North Bldg 232B BOM Session Sponsored: Behavioral Operations Management Sponsored Session

Chair: Xiaobo Zhao, Tsinghua University, Beijing, 100084, China Co-Chair: Wanshan Zhu, Tsinghua University, Beijing, 100084, China 1 - Forecast Information Sharing with Multiple Suppliers: Trust & Coordination Meng Li, Rutgers University, Camden, NJ, United States, Yang Zhang, Yue Li The retailer in a complementary goods supply chain considers whether to share its private demand forecast with its suppliers. According to the theoretical benchmark, share or not share does not make any difference, regardless of how many suppliers involved. However our experiment finds that, sharing the forecast may impede the profit of retailer as well as the performance of suppliers, because the heterogeneity in the agents’ trust results in a coordination failure. As a result, the retailer may prefer not to share the forecast when facing multiple suppliers. 2 - Impact of Market Information on Buying Decisions: A Behavioral Study Diana (Yan) Wu, University of Kansas, 1300 Sunnyside Avenue, Summerfield Hall Room 345, Lawrence, KS, 66045-7585, United States, Kay-Yut Chen We study how market information influences purchase behavior. Experiments are conducted in which buyers evaluate trade-offs between value uncertainties, likelihood of out-of-stock and price adjustments. We find decisions can be swayed by market information and develop a model to explain behaviors. 3 - Value of High-Quality Logistics: Evidence from a Clash between SF Express and Alibaba Ruomeng Cui, Emory University, Atlanta, GA, USA, Meng Li, Qiang Li. Consumers regard product delivery as an important service component that influences their shopping decisions on online retail platforms. Failing to ship products to customers in a timely and reliable manner will diminish customer experience and companies’ profitability. In this research, we explore how much customers value a high-quality delivery experience when shopping online. Our identification strategy exploits a natural experiment: a clash between SF Express —- the largest private logistics service provider with the highest reputation in both speed and reliability in China —- and Alibaba - the largest online retail platform in China. 4 - Decision Bias of Strategic Customers with Private Product-value Information: An Experimental Study Song Yanan, Tsinghua University, Beijing, China, Xiaobo Zhao, Wanshan Zhu We consider strategic customers who compete for limited stock of a product due to rationing. We conduct a laboratory experiment to study the customers’ decisions, and find that, less customers wait for discount than predicted when stock quantity is high; but slightly more customers wait when stock quantity is low. Behavior models reveal that this result is caused by the decision biases of bounded rationality and risk aversion. These findings imply that retailers should discreetly use rationing because the decision biases of strategic customers benefit retailers under light rationing, but hurt them under heavy rationing. n WB56 West Bldg 101A Joint Session HAS/Practice Curated: Data Science in Health Care Operations Research Sponsored: Health Applications Sponsored Session Chair: Jay Michael Rosenberger, University of Texas-Arlington, Arlington, TX, 76019, United States 1 - Parameter Adjustment for Pain Management Treatment Optimization Amith Viswanatha, The University of Texas at Arlington, Arlington, TX, 76013, United States Mixed Integer Programming (MIP) solvers like CPLEX have several parameters and finding the right parameter settings for a given problem formulation is a challenging task. In this research, we use the design of experiments (DOE) approach to study the influence of CPLEX parameters on the computation time of the pain management optimization model. The orthogonal array design is used to initially screen the significant parameters and a full factorial design is later employed for a detailed investigation on the influence of the selected parameters.

461