Informs Annual Meeting 2017

SD07

INFORMS Houston – 2017

2 - ED Physician Staffing via a Multiclass Multistage Network: Opening the Blackbox Caglar Caglayan, Georgia Institute of Technology, 765 Ferst Drive , Room 321 ISyE Main Building, Atlanta, GA, 30332, United States, ccaglayan6@gatech.edu, Yunan Liu, Mustafa Y. Sir, Turgay Ayer, David Nestler, Kalyan Pasupathy We propose an “intuitive”, “realistic” and “tractable” emergency department (ED) staffing model. More specifically, we analyze a multi-class multi-stage (MCMS) queuing network with time-varying arrivals where the objective is to determine the optimal number of ED physicians, stabilizing tail probability of delay for each patient class and treatment stage. We apply our proposed method to real data from Mayo Clinic for numerical analysis and test its performance by a realistic simulation model under different routing rules. We further analyze the staffing of “efficiency-driven queue without abandonments” since this structure represents most of the queue stations appearing in the MCMS network. 3 - Improving Patient Transport Services through Crew Flexibility Craig Froehle, University of Cincinnati, 2925 Campus Green Drive, Cincinnati, OH, 45221-0130, United States, craig.froehle@uc.edu, David James Rea, Elizabeth Powell Patient transport services, where stable patients are moved between facilities using a variety of crew and vehicle types, are a largely unexplored part of healthcare operations. We build a discrete-event simulation model based on a dataset of ~30,000 patient transports to identify preferential policies regarding crew flexibility and vehicle location. In addition to system-level operational performance, a financial analysis also permits assessment of the policies’ overall cost-effectiveness. 4 - Introducing Swing Shifts to Dynamically Respond to Emergency Department Workload Uncertainty David L.Kaufman, University of Michigan - Dearborn, 19000 Hubbard Drive, Fairlane Center South, Dearborn, MI, 48126, United States, davidlk@umich.edu, Kalyan Pasupathy, Daniel Cabrera, Mustafa Y. Sir A fundamental problem of emergency care is matching resources to uncertain patient demands. Staffing allocation decisions require good matching with workloads but also consider the needs of emergency providers at very high risk of burnout. Mayo Clinic Emergency Department recently introduced a “swing shift,” which allows physicians to leave early depending on a workload threshold. While popular, swing shifts introduce several challenges: How to design a threshold mechanism? What is the optimal length of the furlough? When should these shifts start and what is their impact? We introduce an effective and tractable data- driven optimization model for a complex stochastic problem. 322A Modeling and Optimization for Cancer Screening and Treatment Sponsored: Health Applications Sponsored Session Chair: Jennifer Mason Lobo, University of Virginia, Charlottesville, VA, 22902, United States, jem4yb@virginia.edu 1 - A Dynamic Model for Understanding Long-term Trends in Prostate Cancer Screening Ozge Karanfil, Harvard T.H. Chan School of Public Health, 651 Huntington Avenue, FXB Building, 643E, Boston, MA, 02115, United States, karanfil@hsph.harvard.edu, Hazhir Rahmandad, Jack Homer, John D. Sterman Whereas guidelines for routine screening should be based on medical evidence, evidence often has relatively little impact on practice, which has led to ongoing controversy and conflict over guidelines. There are significant variations in clinical practice, including evidence of both over and underscreening. To explain the patterns of overscreening, fluctuations, and low adherence to guidelines, we develop an explicit broad boundary feedback theory tested in a formal model. The model presents an extended case study specific to PSA screening, including realistic presentations for test specifics, natural progression of the disease, and respective changes in population size and composition. 2 - Personalized Modeling for Assessing HPV Vaccination Strategies for Females Shengfan Zhang, University of Arkansas, 4207 Bell Engineering Center, Department of Industrial Engineering, Fayetteville, AR, 72701, United States, shengfan@uark.edu, Fan Wang Human papillomavirus (HPV) is the most common sexually transmitted infection in the United States. Currently, catch-up vaccines are recommended for males and females through age 21 and age 26, respectively, if they did not get vaccinated previously. In this research, we hypothesize that the cut off age of catch-up vaccine should be determined based on every single woman’s risk characteristics, rather than a one-size-fit-all age 26. We develop a discrete-event SD07

simulation model, incorporating patient-specific HPV risk estimation, to evaluate multiple clinical consequences after a woman get vaccinated based on a number of personal attributes. 3 - A Cost-effectiveness Analysis of Adjuvant Paclitaxel and Trastuzumab Regimen in Early Node-negative, Her2-positive Breast Cancer Ali Hijjar, University of Wisconsin-Madison, 1513 University Avenue, 3233 Mechanical Engineering Building, Madison, WI, 53706, United States, hjaar@wisc.edu, Mehmet Ergun, Murtuza Rampurwala, Oguzhan Alagoz Patients with early node-negative, HER2-positive breast cancer do not receive standard treatment since most were excluded from adjuvant trastuzumab trials. In this study, we investigated the cost-effectiveness of a novel one-year adjuvant paclitaxel and trastuzumab therapy that shows a decrease in the relapse rate and toxicity adverse events using a Markov model. We performed sensitivity and probabilistic analyses to evaluate the impact of uncertainty on model results. 4 - Optimal Liver Cancer Surveillance in Hepatitis C-infected Population Qiushi Chen, Massachusetts General Hospital, 101 Merrimac St., Liver cancer incidence rates have increased sharply in the United States. Although surveillance for liver cancer has shown survival benefit, the optimal surveillance policy remains unknown. To identify the most cost-effective surveillance policy, we develop a mixed-integer programming-based framework which allows to formulate practical policy structures. Our numerical results find that (1) the optimal surveillance interval should depend on patient’s stage of hepatitis C and age, and (2) expanding surveillance to earlier stage of hepatitis C improves the cost-effectiveness of HCC surveillance. 10th FL, Apt D3, Boston, MA, 02114, United States, qchen@mgh-ita.org, Turgay Ayer, Jagpreet Chhatwal 322B Stochastic Models in Health Systems Engineering Sponsored: Health Applications Sponsored Session Chair: Anil Aswani, UC Berkeley, San Francisco, CA, 94103, United States, aaswani@berkeley.edu Co-Chair: Philip Kaminsky, University of California-Berkeley, Berkeley, CA, 94720-1777, United States, kaminsky@berkeley.edu 1 - Data-driven Appointment Scheduling Tugce Gurek, University of California, Berkeley, 6141 Etcheverry Hall, Berkeley, CA, 94720, United States, tugce.gurek@berkeley.edu, Philip Kaminsky We focus on appointment scheduling of stochastic tasks on a highly utilized server where the task processing durations are challenging to estimate. Appointment scheduling involves both sequencing tasks, and setting estimated start times of those tasks. Tasks types are known prior to the appointment date, but task duration data is initially limited so duration estimates are continuously updated. Our goal is to develop a data-driven approach to sequencing and scheduling these tasks that also integrates expert knowledge into the scheduling decision. 2 - Improving Predictions of Pediatric Surgical Durations with Supervised Learning Zhengyuan Zhou, 160 Comstock Circle - Unit 106002, Stanford, CA, 94305, United States, zyzhou@stanford.edu We consider a model of multi-agent online learning under imperfect information, where the reward structures are given by a general continuous game. After introducing a general equilibrium stability notion, called variational stability, we examine the well-known online mirror descent learning algorithm and show that the iterates converge to variationally stable Nash equilibria under a variety of imperfect feedback. We then present two applications: non-convex stochastic optimization and power management in random wireless networks. We conclude with a discussion on further extensions. 3 - Data-driven Incentive Design in the Medicare Shared Savings Program Auyon Siddiq, 1749 Oxford St, Unit 22, Unit 22, Berkeley, CA, 94709, United States, auyon@berkeley.edu, Anil Aswani, Zuo-Jun Max Shen The Medicare Shared Savings Program (MSSP) was created to control escalating Medicare spending by incentivizing providers to deliver healthcare more efficiently. We formulate the MSSP as a principal-agent model and propose a new type of contract that includes a performance-based subsidy for the provider’s investment. We show that the proposed contract dominates the existing MSSP contract by producing a strictly higher expected payoff for both Medicare and the provider. We then present a maximum likelihood approach for estimating the parameters of the principal-agent model as well as the potential increase in savings under the proposed contract. SD08

102