Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

MA63

INFORMS to become a trusted source of expertise. Led by INFORMS Director of Public Affairs & Marketing Jeff Cohen and Signal Group Executive Vice President Charles Cooper, this session will provide an update about INFORMS’ advocacy activities, including the INFORMS Government & Analytics Summit. The session will also include a discussion about the overall current state of Washington, an analysis of the 2018 mid-term election polls and predictions, and the practical impacts and implications of different election day outcomes. Panelist Charles Cooper, SIGNAL, Washington, DC, United States n MA62 West Bldg 103A Data-Driven Decision Making Sponsored: Data Mining Sponsored Session Chair: Milton Soto-Ferrari, Indiana State University, Scott College of Business, Terre Haute, IN 1 - Integrating CAP Domains into Analytics Courses: Methods & Examples Concetta A. DePaolo, Indiana State University, Scott College of Business, Terre Haute, IN, 47809, United States The Certified Analytics Professional (CAP«) designates competencies or domains for analytics professionals. This presentation will describe how these domains have been integrated into an undergraduate business analytics course. In addition, examples of case studies involving real-world data used to teach analytics process and methods will be shared. 2 - Frontier Estimation for Automated Machine Levels Outputs Sandeep Bhowmick, Indiana State University, Terre Haute, IN, United States As an alternative to data envelopment analysis, Frontier estimation methods have been found to be more efficient in estimating efficiency measures. Separating random noise component into stochastic and non-stochastic components is found useful in capturing efficiency components of production methods. This presentation shows an application of stochastic frontier estimation method to analyze efficiency values on an industry supplied data at automated machine level outputs. 3 - A Method for Systematic Monitoring of Treatment Receipt in Breast Cancer Patients Milton Soto-Ferrari, Indiana State University, 30 N. 7th St, Room F-211, Terre Haute, IN, 47807, United States This research aims to propose a method to monitor the clinical and non-clinical variables that influence the receipt and adherence to treatment in breast cancer patients. Building upon a Bayesian Network model for mining cancer registry data. The method presents a framework for the automatic monitoring of patients to predict radiation treatment receipt. This framework uses information from a regional cancer institution with a sample of 1922 patients from years 2009-2014. n MA63 West Bldg 103B Joint Session DM/AI: Deep Learning in Text Mining and Analytics Sponsored: Data Mining Sponsored Session Chair: Onur Seref, Virginia Tech, Blacksburg, VA 1 - Adverse Drug Events Detection and Recognition: A Deep Learning Based Approach Long Xia, Virginia Polytechnic Institute and State University, 880 W. Campus Drive, 2069 Pamplin Hall, Blacksburg, VA, 24061, United States, Drug safety profiling plays a significant role in medications decision-making. Analyzing the adverse drug events from patients’ generated data is an integral part of drug safety profiling. With the practical need for advanced text mining techniques for adverse drug events identification and extraction, we proposed a research framework which utilizes deep learning network as the platform, incorporates both word embeddings features and features obtained from existing models to complement feature coverage and comprehensiveness. By further incorporating human designed features, our approach achieved a significantly improved performance.

n MA60 West Bldg 102B HAS Session Sponsored: Health Applications Sponsored Session Chair: Van-Anh Truong, Columbia University, New York, NY, 10027, United States 1 - Continuum of Care for Reducing Readmissions: Balancing Pre- and Post-discharge Efforts Xiang Liu, University of Michigan, 1205 Beal Ave, Ann Arbor, MI, 48109, United States, Mariel Sofia Lavieri, Jonathan Helm, Ted Skolarus Hospital readmissions are burdensome and costly. We analytically study, under a Bundled Payment (BP) policy 1) how hospitals balance effort between the inpatient stay stage and the post-discharge stage; and 2) how a public healthcare funder designs a BP policy and readmission penalty program to incentivize hospitals to balance its efforts for readmissions reduction. We develop a novel Strengthen Then Maintain (STM) framework that is generalizable to a set of machine maintenance problems. We uncover novel managerial insights for the design of BP policies and readmission penalty programs. 2 - Optimizing Discharge Decisions in a Hematology Ward Mor Armony, New York University, 44 West 4th Street #8-62, New York, NY, 10012, United States, Galit Bracha Yom-Tov Following treatment hematology patients face an increased risk of developing an infection. If a patient remains at the hospital his risk of catching an infection is higher than at home, but once an infection develops the mortality risk at the hospital is lower than at home due to quick access to appropriate treatment. We study the problem of dynamically determining discharge times for hematology patients subject to capacity constraints. We show that in an overloaded operating regime this dynamic problem reduces to a static problem with a simple solution of two-class two-discharge thresholds. We characterize the specific optimal solutions under empirically driven time-to-infection distributions. 3 - Online Resources Allocation with Learning and Reoptimization Zhen Xu, Columbia University, New York City, NY, 10027, United States, Van Anh Truong We study a multi-period revenue management problem where a decision maker assigns each arriving customer to one of multiple products made from multiple resources with finite capacity. Every assignment of a customer to a product will generate a random reward. The objective is to jointly learn the mean reward function and maximize the expected revenue. We formulate the problem as a multi-armed bandit problem. We propose a natural and simple extension of the UCB family of algorithms. We show that by taking advantage of re-optimization techniques, our proposed algorithm achieves a regret of O(log3T), which significantly reduces the O(\sqrt(T)) bound. 4 - Toward a Genomic Liquid Biopsy Andrew A. Li, MIT, Cambridge, MA, United States, Jackie W. Baek, Vivek Farias, Chinmay Jha, Deeksha Sinha The cost of DNA sequencing has fallen 10,000x in the last ten years, and we are finally in sight of the silver bullet for cancer screening: an early-stage blood test. As sequencing can now be performed affordably on a tiny fraction of a genome, what remains is a massive variable selection problem. We provide an efficient algorithm, based on a decomposition at the gene level, that scales to full genomic sequences across thousands of patients. We contrast our selected variables against DNA panels from two recent, high-profile studies and demonstrate that our own panels achieve significantly higher sensitivities at the same cost, along with accurate discrimination between cancer types for the first time. n MA61 West Bldg 102C Advocacy and Elections: A Discussion about the INFORMS Advocacy Initiative and the 2018 Midterm Elections Emerging Topic: INFORMS Special Sessions Emerging Topic Session Chair: Jeff Cohen, INFORMS, 5521 Research Park Drive, Suite 200, Catonsville, MD, 21228, United States 1 - Advocacy and Elections: a Discussion about the INFORMS Advocacy Initiative and the 2018 Midterm Elections Jeff Cohen, INFORMS, 5521 Research Park Drive, Suite 200, Catonsville, MD, 21228, United States In 2018 INFORMS launched a new advocacy initiative to raise policymakers’ awareness and interest in the O.R. and Analytics fields, build strategic relationships between INFORMS and the policy community, and enable

141