Informs Annual Meeting 2017

WA09

INFORMS Houston – 2017

2 - Individualized Prediction of Chronic Disease Occurrence Qingpeng Zhang, City University of Hong Kong, Kowloon, Hong Kong, qingpeng.zhang@cityu.edu.hk, Haolin Wang, Eman Leung, Frank Y. Chen This research proposes a tensor factorization based model to predict the onset of chronic diseases and the co-occurrence of chronic and acute diseases using patients’ EHR record. With 2 year-worth of hospitalization records in a major hospital in Hong Kong, we demonstrate that the proposed model is able to predict the occurrence of new chronic diseases at individual level with superior accuracy. 3 - Two-stage Stochastic Programming Model for Resource Planning in Patient Centered Medical Homes Mohammad Hessam Olya, Wayne State University, 71 W. Hancock St, Apt 4, Detroit, MI, 48201, United States, mhmdhesam@gmail.com, Kai Yang One of the biggest challenges in PCMH healthcare delivery system is to find an efficient mechanism to design PCP patient panel where there is a portfolio of healthcare providers. Since the demand is unknown, a framework that incorporates workload prediction model into stochastic allocation is needed. In this study using a novel patient workload prediction model for generating different demand scenarios called multivariate hierarchical structured additive regression, a two-stage stochastic integer programming with integer recourse model for patient allocation is presented for assisting decision makers while minimizing the total cost of patient assignment to care providers at a given time. 4 - Analysis of the Infection Time from a Potential H7N9 Influenza Pandemic Outbreak Walter A.Silva Sotillo, University of South Florida, 4202 E Fowler Ave, Tampa, FL, 33620, United States, silvasotillo@mail.usf.edu, Mingyang Li, Tapas K. Das Avian influenza viruses have been affecting human populations for a long time. Since then, other mutations and reassortments of the influenza viruses have emerged causing pandemics. Researchers have developed early estimates of some of the epidemiological parameters to characterize H7N9 virus in China but infected persons appear everyday. In this research we analyzed the distribution that follows the infection time from a potential H7N9 influenza pandemic outbreak using an agent-based (AB) simulation model. We analyzed some continuous probability distributions and conclude that the lognormal distribution provides a better fit estimate to the time to be infected. 5 - A Multi-state Semi-Markov Model of Patient Flow in Long-term Care Network Hambisa Keno, BAE Systems, Burlington, MA, United States, hambisa.keno@gmail.com, Nan Kong, Steven Landry, Christopher M. Callahan Long-term care (LTC) represents an expensive segment of the healthcare system. It is essential to carefully characterize patient flow in the LTC network, with implications on model-based policy evaluation. We extend a length of stay based model to characterize patient flow across the network via an multi-state semi- Markov (MSSM) model. We investigate the order of the embedded Markov chain that best characterizes patient transition data. We limit the regression parameters in the model to casual variables through Bayesian network structure learning on care transitions. We diagnose sensitivity of the model parameters and compare goodness-of-fit of the model for different embedded Markov chains. 322A Recent Advances in Radiation Therapy Planning Sponsored: Health Applications Sponsored Session Chair: Gino J. Lim, University of Houston, Houston, TX, 77204, United States, ginolim@uh.edu 1 - Using Inverse Optimization to Evaluate Knowledge-based Planning Methods for Oropharyngeal Cancer Aaron Babier, University of Toronto, 153 Robert Street, Unit 2B, Toronto, ON, M5S.2K6, Canada, ababier@mie.utoronto.ca, Justin James Boutilier, Andrea L. McNiven, Timothy Chan Knowledge based planning (KBP) is often used to predict clinically acceptable dose volume histogram (DVHs), however the predictions may be undeliverable. We develop two KBP methods to predict DVHs for over 200 oropharyngeal cancer patients, and use an inverse optimization model to reverse engineer fluence deliverable maps from the predictions. To illustrate the impact that a prediction has on our pipeline, we compare the plans that were reverse engineered from each KBP method. Each plan is also compared to a clinical plan to validate our fluence maps and demonstrate the role inverse optimization can play in an automated planning pipeline. WA07

2 - Robust Hypoxia-based Radiation Plan Optimization Arkajyoti Roy, Bowling Green State University, 349 Business Administration Building, Applied Statistics and Operations Research, Bowling Green, OH, 43403, United States, aroy@bgsu.edu, Omid Nohadani Low oxygen concentration reduces radiosensitivity of tumor cells, adversely affecting radiation treatments. Reoxygenation leads to uncertain temporal changes in cell oxygenation, ranging from short-term fluctuations due to perfusion to gradual changes from angiogenesis. In this work, a time-dependent uncertainty set is employed to model the reoxygenation progression. A robust optimization model incorporating the uncertainty set is solved for a set of clinically realistic reoxygenation scenarios. The proposed method is evaluated in comparison to current clinical practice using a prostate case. 3 - Spatioteporally Integrated Radiotherapy with Treatment Length Update In a typical formulation of the spatiotemporally integrated radiotherapy, the number of treatment sessions (N) has to be decided prior to the beginning of the actual treatment. However, taking advantage of the new imaging techniques, a more dynamic treatment planning can adapt to the tumor’s response thoughout the treatment course and determine N on the fly. In this study, we propose a method to dynamically determine the optimal number of remaining sessions, and reoptimize the treatment plan in each session accordingly. Solutions obtained in this manner have more flexibility to the tumor’s evolution and could potentially result in higher biological effective dose. 4 - Optimization of Radiation Therapy Fractionation Dose Considering the Biological Effects Azin Khabazian, University of Houston, 5465 Braesvalley Dr. Apt 566, Houston, TX, 77096, United States, akhabazian@uh.edu, Gino J. Lim The goal in fractionation radiation therapy is to maximize tumor-damage while limiting toxic effects of radiation on nearby healthy anatomies. This is achieved through a sequential radiation treatment process in which the treatment plan can be modified using a regular feedback of measurements. In this study, we develop a chance-constrained programming (CCP) framework under a probabilistic description of dose delivery to optimize the fractionation dose by adjusting to tumor response in each fraction of treatment planning. We evaluate the model in the context of tumor coverage and normal tissue sparing and numerically demonstrate its superiority over the traditional treatment planning. 330A Nonprofit Operations Sponsored: Manufacturing & Service Oper Mgmt Sponsored Session Chair: Milind Sohoni, Indian School of Business, Hyderabad, 500032, India, milind_sohoni@isb.edu 1 - Improving Societal Outcomes in the Organ Donation Value Chain Priyank Arora, Georgia Institute of Technology, 800 W. Peachtree Street NW, Atlanta, GA, 30308, United States, priyank.arora@scheller.gatech.edu, Ravi Subramanian We examine a principal-agent problem in the cadaver organ donation value chain (ODVC) where the principal in our case is a social planner that has an overall quality-adjusted-life-year improvement objective. The agents include a non-profit organ procurement organization (OPO) with a volume-of-care objective and a profit-maximizing hospital (trauma center). While the majority of the healthcare operations management literature focuses on the demand side of the ODVC, we develop an analytical model to study the effects of contextual parameters and operational decisions of the supply-side entities (OPO and hospital) on their respective payoffs and on societal outcomes. 2 - Payment for Results - Funding Non-profit Operations Sripad K. Devalkar, Indian School of Business, Ac3 L1 Room 3114, Hyderabad, 500032, India, sripad_devalkar@isb.edu, Milind Sohoni, Neha Sharma Payment for results (PfR), with funding contingent on actual results delivered, has been proposed as a mechanism that allows donors to channel funds to more efficient NPOs. However, many NPOs are financially constrained and need an intermediary to provide the funds required to implement the project upfront. We compare the performance of a PfR mechanism involving a social planner, intermediary and NPO, with that of a traditional upfront funding mechanism involving only the social planner and NPO. For risky projects, we find that the social planner’s expected utility under traditional funding can be higher compared to when using PfR. Ali Ajdari, University of Washington, Seattle, WA, 98199, United States, ajdari@u.washington.edu, Archis Ghate WA09

443