Informs Annual Meeting 2017

TD08

INFORMS Houston – 2017

TD06

characteristics and (2) keep the updated machine parameters close to the original plan. 2 - Automated IMRT Treatment Planning using Hierarchical Constrained Optimization Masoud Zarepisheh, Memorial Sloan Kettering Cancer Center, New York, NY, United States, zarepism@mskcc.org, Linda Hong, James G. Mechalakos, Margie A. Hunt, Gig S. Mageras, Joseph O. Deasy This study formulates IMRT treatment planning as a hierarchical constrained optimization (HCO) problem (also known as prioritized optimization). HCO prioritizes the clinical goals and optimizes them in ordered steps. We introduce a novel idea to speed up the optimization process by sparsifying the influence matrix and then solving an extra unconstrained optimization problem to compensate for the inaccuracy caused by sparsity. The commercial treatment planning system API is used to automate the entire workflow and our package is currently being used in clinical environment to automate the treatment planning process. 3 - Generating High-quality DMPO Seeds for Single- and Multiple-arc Treatment Plans using a Conformal Arc/VMAT Hybrid Planning Model Joshua T. Margolis, Clemson University, 277 Freeman Hall, Clemson University, Clemson, SC, 29634, United States, jtmargo@clemson.edu, Scott J. Mason, Weiguo Lu, Steve Jiang, Troy Long In radiation therapy, conformal arc therapy uses a coplanar moving gantry to continuously deliver radiation to cancerous tissue and has little difficulty in passing post-planning quality assurance. However, plan quality suffers when specific organ-at-risk sparing does not explicitly influence aperture shapes, such as in the more-involved volumetric modulated arc therapy (VMAT). We study a conformal arc/VMAT hybrid planning model use dose volume histogram-driven direct machine parameter optimization (DMPO) to find a locally optimal solution for a given set of apertures. We study how to generate high-quality DMPO seeds for single- and multiple-arc treatment plans. 4 - Optimization of Nonuniformly Fractionated Radiotherapy Treatments with Bounds on the Achievable Benefit David Papp, North Carolina State University, 3222 SAS.Hall, Campus Box 8205, Raleigh, NC, 27606, United States, dpapp@ncsu.edu, Melissa R. Gaddy, Jan Unkelbach, Sercan Yildiz Radiotherapy treatments are fractionated, meaning that the radiation dose is delivered over several days rather than in a single session. In current clinical practice, the same dose is delivered on each treatment day. We formulate an optimization model to assess the achievable benefit of altering the dose from day to day. The model is large-scale and nonconvex. Using clinical cases, we show that even locally optimal nonuniform treatments yield a substantial improvement over conventional ones. SDP relaxation bounds show that the computed locally optimal treatments are nearly globally optimal. 322B Disaster and Emergency Management Contributed Session Chair: Damitha Bandara, Albany State University, Albany, GA, United States, damitha.bandara@asurams.edu 1 - Distributionally Robust Optimization of an Emergency Medical Service Station Location Problem Kanglin Liu, PHD Candidate, Tsinghua University, Tsinghua University, 30 Shuangqing Road, Beijing, 100084, China, lkl15@mails.tsinghua.edu.cn, Zhihai Zhang Locations and capacities in the emergency medical service (EMS) system are key issues of humanitarian logistics. This paper proposes a distributionally robust model to optimize the location, number of ambulances and demand assignment in an EMS system by minimizing the expected total cost. The model guarantees a high probability of satisfying demand in the entire geographical area by introducing joint chance constraints, and identifies the total expected cost by a data-driven approach. The model is approximated as a tractable second-order cone problem and efficiently solved by an Outer Approximation approach. Extensive numerical results and a real case of Beijing show the validly of the program. TD08

320C Decision-making in Healthcare Sponsored: Health Applications Sponsored Session Chair: Vishal Ahuja, Southern Methodist University, Dallas, TX, 75275, United States, vahuja@smu.edu 1 - Inferring Doctor Decisions: Post-stent Patient Streaming using Personalized Medicine Kellas Ross Cameron, Boston University, 62 Bay State Road, Apt. #1, Boston, MA, 02215, United States, kellas@bu.edu, Nitin Joglekar, Nachiketa Sahoo, Jugnu Jain We develop a structural econometric model to analyze doctors’ choice of medication after cardiac stent surgery. The estimated model shows the role of patient’s medical, as well as economic conditions in their choice. The model can be used to improve personalized medication processes by streaming those who will benefit the most from a new genomic test-based treatment instead of assigning them to a conventional trial and error process. 2 - Investigating Steroid with drawal Strategies for Kidney Transplant Recipients Yann Ferrand, Clemson University, Department of Management, 101 Sirrine Hall, Clemson University, Clemson, SC, 29634, United States, yferran@clemson.edu, Christina M.Kelton, Vibha Desai, Teresa M. Cavanaugh, Jaime Caro, Jens Goeble, Pamela C. Heaton We study long-term complications associated with various steroid withdrawal strategies for kidney transplant recipients. We develop a model calibrated with an econometric study of patient data from a national registry to simulate the long- term course of these patients, and derive patient outcomes based on the characteristics of patient population. We focus part of the investigation on the effect of induction therapy. 3 - Designing Liver Allocation System using Neighborhoods Sanjay Mehrotra, Northwestern University, Dept of I.E./ M.S.C246 Tech Inst, 2145 Sheridan Road, Evanston, IL, 60208-3119, United States, mehrotra@iems.northwestern.edu We present approaches used to design a liver allocation system to address geographic disparity in liver transplantation. This design is being considered for public comments. We will also discuss the complexity of national public policy change that is model driven. 4 - An Analytics-driven Approach for Optimal Diabetes Screening Decisions Michael Hahsler, Southern Methodist University, Bobby B. Lyle School of Engineering, P.O. Box 750123, Dallas, TX, 75275, United States, mhahsler@lyle.smu.edu, Vishal Ahuja, Farzad Kamalzadeh, Michael Bowen About 10% of the US adult population have diabetes and almost 40% are at risk of developing diabetes. Screening can be used to identify these individuals. Population-based screening is an expensive task and thus prioritizing whom to screen and how often to screen becomes of interest from a decision-making point of view. We develop a Markov Decision Process (MDP) to model disease progression and find the optimal screening policy. For parameter estimation, we suggest the use of Hidden Markov Models. To produce individualized screening recommendations we incorporate predictive diabetes risk modeling. We will present results for electronic health record data from the Parkland Health & Hospital System. 322A Optimization in Radiation Therapy Sponsored: Health Applications Sponsored Session Chair: David Papp, North Carolina State University, Raleigh, NC, 27695, United States, dpapp@ncsu.edu 1 - DVH-driven Direct Machine Parameter Optimization for Adaptive IMRT and VMAT Radiation Therapy Treatment Planning Troy Long, University of Texas Southwestern Medical Center, Dallas, TX, United States, troy.long@utsouthwestern.edu, Steve Jiang, Joshua T. Margolis, Weiguo Lu In conventional fractionated radiation therapy, an overall treatment plan is developed with pre-treatment imaging and then delivered over a number of treatment fractions. However, patient geometry changes between treatment fractions, and physicians often will adapt the existing treatment plan to the new geometry. We study automatically generating adapted IMRT and VMAT plans that (1) attempt to reproduce the original plan’s dose-volume histogram (DVH) TD07

377