Informs Annual Meeting 2017

MD40

INFORMS Houston – 2017

MD41

3 - Predictive Modeling for Toxicities in Head and Neck Cancer Patients

352F Health Care, Modeling and Optimization Contributed Session Chair: Salar Ghamat, Ivey Business School, London, ON, Canada, sghamat.phd@ivey.ca 1 - Modeling Kidney Transplantation Decisions: Optimal Regulatory Oversight Zahra Gharibi, SMU, Bobby B. Lyle School of Engineering, Dallas, TX, 75205, United States, zgharibi@smu.edu, Michael Hahsler, Mehmet Ayvaci Regulatory oversight with report card programs are designed to evaluate transplant centers’ performance, but it is controversial whether behavioral responses to such programs improve patient outcomes. I propose a stochastic model that determines the socially optimal kidney transplant decision and calculates social loss under no oversight. I introduce a game theoretical solution to derive optimal regulatory oversight implemented as a citation policy which minimizes social loss. 2 - The Interaction of Outbreaks and the Healthcare System: Closing the Feedback Loops for Disease Modeling Craig Bakker, Pacific Northwest National Laboratory, 902 Battelle Blvd, Richland, WA, 99352, United States, craig.bakker@pnnl.gov, Matthew Robert Oster, Brent Daniel, Romarie Morales Rosado, Tim McPherson Healthcare services affect and are affected by disease outbreaks. We have developed a disease model that incorporates the public health system and explicitly considers the two-way interaction between that system and the disease. Here, we model the baseline spread of an outbreak along with the potential impact of the outbreak on the public health workforce. We then explore the potential impacts of various treatment options, resource constraints, and interruptions on the outbreak’s evolution. Finally, we discuss healthcare supply chain issues and how to incorporate them into our model. 3 - A Machine Learning Approach to Personalization from Observational Data 360A Additive Manufacturing - Data Analytics and OR II Invited: Advanced Manufacturing Invited Session Chair: Alaa Elwany, Texas A&M University, Engineering, College Station, TX, 77843, United States, elwany@tamu.edu 1 - Locating AM Hubs in the United States via Hybrid Manufacturing Guha Manogharan, Pennsylvania State University, University Park, PA, United States, gum53@engr.psu.edu, Michael G.Kay The ever-growing applications of Additive Manufacturing (AM) in the production of low volume- high value metal parts can be attributed to improved AM processing capabilities and complex design freedom. However, secondary post- processing using traditional processes such as machining, grinding, heat treatment and hot isostatic pressing, i.e. Hybrid Manufacturing is required to achieve Geometric Dimensioning and Tolerancing (GD&T), surface finish and desired mechanical properties. It is often challenging for most traditional manufacturers to participate in the rapidly evolving supply chain of direct digital manufacturing (DDM) through in-house investments in cost prohibitive metal AM. This research investigates a system of strategically-located AM hubs which can integrate hybrid- AM with the capabilities and excess capacity in multiple traditional manufacturing facilities. Using North American Industry Classification System (NAICS) data for machine shops in the US, an uncapacitated facility location model is used to determine the optimal locations for AM hub centers based on: (1) geographical data, (2) demand, and (3) cost of hybrid-AM processing. Results from this study have identified: (a) candidate US counties to build AM hubs, (b) total cost (fixed, operational and transportation) and (c) capacity utilization. It was found that uncapacitated facility location models identified demand centroid as the optimal location and was affected only by AM utilization rate. The constrained p-Median model has identified 22 AM hub locations as initial sites for Nishanth Mundru, Massachusetts Institute of Technology, Operations Research Center, E40, Cambridge, MA, 02139, United States, nmundru@mit.edu, Dimitris Bertsimas We consider the problem of assigning the best treatment option out of a finite number of choices to a patient/user depending on his/her observed features. Given historical observational data of patients which consists of their features, assigned treatments and their corresponding outcomes, we learn the optimal treatment policy for each patient in an offline setting. Using an optimization and machine learning approach, we demonstrate that our method outperforms the commonly used regress-and-compare methods on real data sets. MD42

Wei Jiang, Research Assistant, Department of Civil Engineering, Johns Hopkins University, Latrobe Hall, 3400 N. Charles Street, Baltimore, MD, 21218, United States, wjiang16@jhu.edu, Pranav Lakshminarayanan, Zhi Cheng, Xuan Hui, Sauleh Ahmad Siddiqui, Ilya Shpitser, Russell H. Taylor, Todd McNutt Head and neck cancer patients experience acute and long-term toxicities such as weight loss, dysphagia, and xerostomia due to radiotherapy. Oncospace is a learning health system developed at Johns Hopkins University that is accumulating a large volume of patients’ treatment data (imaging and radiation dose distributions) and clinical data (toxicities and demographic data). We build a machine learning model to predict head and neck cancer patients’ toxicities and investigate the influence of the radiation dose distribution in specific organs. The goal is to better understand where we can improve our delivery of radiation to reduce toxicity while maintaining adequate treatment of cancer. 4 - Studying Length of Stay and Early Readmissions in General Surgery Service Line: A Data Analysis Approach Wareef Al Najjar, Binghamton University, Binghamton, NY, United States, walnajj1@binghamton.edu, Khalid Aram Patient Length of Stay (LOS) and hospital readmissions should be given attention by healthcare organizations. LOS and readmission rates have been used to measure efficiency and effectiveness of care respectively. In this study, we focus on LOS in general surgery service line and its association to readmissions and other factors in a hospital in middle Georgia. First, we identified Diagnosis- Related Groups that have major contribution to average LOS based on some factors. Then we applied data analysis techniques and built a predictive model which will predict readmission cases earlier in their stay, to help in providing early interventions to avoid readmissions, while shortening their stay. 5 - A Least Squares Monte Carlo Approach to Appointment Scheduling under Patient Cancellation and No-show Behavior Jason Xiaotian Dou, PhD Student, University of Colorado Boulder, 995 Regent Dr, Boulder, CO, 80302, United States, xido4892@colorado.edu, Manuel Laguna, Dan Zhang Patient cancellation and no-show behavior is a major challenge in appointment scheduling. A stochastic dynamic programming model for appointment scheduling is introduced. We develop a least squares Monte Carlo approach to tackle the problem and show promising performance via extensive numerical experiments. 352E Daniel H. Wagner Prize Competition III Invited: Daniel H. Wagner Prize Competition Invited Session Chair: Patricia Neri, SAS Institute, Inc., 104 Grandtree Ct., Cary, NC, 27519, United States, patricia.neri@sas.com 1 - Outcome-Driven Personalized Treatment Design for Managing Diabetes Eva Lee, Georgia Tech, Industrial & Systems Engineering, Ctr for Operations Research in Medicine, Atlanta, GA, 30332-0205, United States, evakylee@isye.gatech.edu This work is joint with Grady Health Systems and Morehouse School of Medicine. We present a drug-effect-based personalized approach to improve treatment outcome for diabetic patients. First, a pharmacokinetic and pharmacodynamics (PK/PD) model is established to uncover drug effect based on analysis of anti- diabetic drug dosage and the blood glucose level recorded in the titration period of each patient. This personalized evidence is then utilized within a treatment plan model that optimizes the glycemic control and drug dosage. The optimized plans provide better glycemic control while using less drug. 2 - Dispatch Optimization in Bulk Tanker Transport Operations Ted L.Gifford, Schneider National Incorporated, All modes of freight transportation are subject to flow imbalances that impact the efficiency of asset utilization. The use of Mathematical Programming optimization models has a rich history of application to this problem. We address a particularly difficult variant of this problem that occurs in bulk chemicals transport. This difficulty is created by a large volume of activity, requiring 1,000 tractors and 1,600 tankers, coupled with complex product-sequencing constraints, tanker wash and preparation processes, and driver work rule constraints. To address this problem the engineering group at Schneider has designed and implemented a multi-phase, multi-dimensional matching algorithm and associated software system that enables business planners to leverage optimized solution recommendations. MD40 PO Box 2545, Green Bay, WI, 54306-2545, United States, giffordt@schneider.com, Tracy Opicka, Ashesh Sinha, Daniel Vanden Brink, Andy Gifford, Robert Randall

250