2015 Informs Annual Meeting

MD44

INFORMS Philadelphia – 2015

MD43 43-Room 103A, CC

3 - A Low-cost Method for Multiple Disease Prediction Mohsen Bayati, Assistant Professor, Stanford Graduate School of Business, 655 Knight Way, Stanford, CA, United States of America, bayati@stanford.edu, Andrea Montanari, Sonia Bhaskar Recently, in response to the rising costs of healthcare, companies have been investing in programs to improve the health of their workforce. These programs aim to reduce the incidence of chronic illnesses and require a low-cost screening to detect individuals with a high risk of developing such diseases. We offer a multiple disease prediction procedure that maximizes the predictive power while minimizes the screening cost. Our method is based on multi-task learning from machine learning. MD42 42-Room 102B, CC Joint Session MSOM-Health/HAS: Operations Research/Management for Public Health: Data-Driven and Dynamic Decision-Making Sponsor: Manufacturing & Service Oper Mgmt/Healthcare Operations Sponsored Session Chair: Soroush Saghafian, Harvard University, 79 JFK Street, Cambridge, MA, 02138, United States of America, Soroush.Saghafian@asu.edu 1 - New Data-driven Approach to Safety and Risk Management in ICUs Retsef Levi, J. Spencer Standish (1945) Professor of Operations Management, Sloan School of Management, MIT, 100 Main Street, BDG E62-562, Cambridge, MA, 02142, United States of America, retsef@mit.edu, Patricia Folcarelli, Yiqun Hu, Jeffrey Adam Traina, Daniel Talmor We develop an innovative system approach to safety in ICUs. The approach is based on the innovative concept of risk drivers, which are states of the ICU, its environment and its staff that affect the likelihood of harms, as well as an innovative aggregated measure of the ‘burden of harm’. Using real data we develop statistical models that identify risky states in the ICUs of a major academic medical center. 2 - Developing Optimal Biomarker-Based Prostate Cancer Screening Policies Christine Barnett, University of Michigan, 1205 Beal Ave., Ann Arbor, MI, United States of America, clbarnet@umich.edu, Brian Denton, James Montie Recent advances in the development of new biomarker tests, which physicians use for the early detection of cancer, have the potential to improve patient survival by catching cancer at an early stage. We describe a partially observable Markov decision process (POMDP) to compute near optimal prostate cancer screening strategies. We present results based on Monte Carlo simulation to compare the policies developed using our approximated POMDP methods with those recommended in the medical literature. 3 - Optimizing Hepatitis C Screening and Treatment Allocation Strategy Yuankun Li, University of Washington, Seattle, WA, United States of America, yuankunl@uw.edu, Zelda Zabinsky, Hao Huang, Shan Liu Chronic hepatitis C (HCV) is a significant public health problem affecting 2.7-3.9 million Americans. The U.S. healthcare systems are ramping up combined HCV screening and treatment efforts, but screening and treatment programs are very costly. We design the optimal HCV screening and treatment allocation strategies in the next 10 years under yearly budget constraint from a national perspective. The method includes simulation optimization using adaptive probabilistic branch and bound. 4 - A Robust POMDP Framework for the Management of Post-transplant Medications Alireza Boloori, PhD Student Of Industrial Engineering, Arizona State University, 699 S Mill Avenue, Office # 313, Tempe, AZ, Patients after organ transplantations receive high dosages of immunosuppressive drugs (e.g., tacrolimus) to reduce the risk of organ rejection. However, this practice has been shown to increase the risk of New-Onset Diabetes After Transplantation (NODAT). We propose a robust POMDP framework to generate effective medication management strategies for tacrolimus and insulin. Our approach increases the patient’s quality of life while reducing the effect of transition probability estimation errors. 85282, United States of America, aboloori@asu.edu, Curtiss B. Cook, Soroush Saghafian, Harini A. Chakkera

Empirical Revenue Management Sponsor: Revenue Management and Pricing Sponsored Session Chair: Dan Zhang, University of Colorado at Boulder, 995 Regent Dr, Boulder, United States of America, Dan.Zhang@colorado.edu 1 - Would You Like to Upgrade to a Premium Room? An Empirical Analysis on Standby Upgrades Ovunc Yilmaz, PhD Student, University of South Carolina, 1014 Greene St, Columbia, SC, 29208, United States of America, oyilmaz@email.sc.edu, Mark Ferguson, Pelin Pekgun Standby upgrades, where the guest is only charged if the upgrade is available at the time of arrival, is one technique that has become increasingly popular in the hotel industry. Working on a data set from a major hotel chain, we analyze the linkage between guest attributes, hotel characteristics and guest decision-making for standby upgrades through an empirical study. 2 - Analytics for an Online Retailer – Demand Forecasting and Price Optimization at Rue La La Kris Johnson Ferreira, Harvard Business School, Morgan Hall 492, Boston, MA, 02163, United States of America, kferreira@hbs.edu, David Simchi-levi, Bin Hong Alex Lee We present our work with Rue La La, an online retailer who offers limited-time discounts on designer apparel. One of their main challenges is revenue management for new products. We use machine learning to build a demand prediction model, the structure of which poses challenges on creating a pricing policy. We develop theory around multi-product price optimization and use this to create and implement a pricing decision support tool. Field experiment results show significant increases in revenue. 3 - A Model to Estimate Individual Preferences using Panel Data Gustavo Vulcano, NYU, 44 West Fourth St, Suite 8-76, New York, NY, 10012, United States of America, gvulcano@stern.nyu.edu, Srikanth Jagabathula In a retail operation, customer choices may be affected by stockout and promotion events. Given panel data with the transaction history of each customer, we use a general nonparametric framework in which we represent customers by partial orders of preferences. Numerical experiments on real-world panel data show that our approach allows more accurate, fine-grained predictions for individual purchase behavior compared to state-of-the-art existing methods. 4 - Estimation of Arrival Rates and Choice Model with Censored Data Anton Kleywegt, Georgia Tech, 755 Ferst Drive NW, Atlanta, GA, 30332, United States of America, anton@isye.gatech.edu Revenue management models with customer choice behavior include two types of parameters: (1) customer arrival rates and (2) choice parameters. Revenue managers usually have censored arrival data only, because no-purchase data are not included. For both homogenous and nonhomogeneous Poisson arrivals we give necessary and sufficient conditions for the arrival rates and choice parameters to be identifiable with such censored data, and we give algorithms for parameter estimation, with numerical results with airline data MD44 44-Room 103B, CC Pricing and Information in Innovative Business Models Sponsor: Revenue Management and Pricing Sponsored Session Chair: Jose Guajardo, University of California Berkeley, 545 Student Services Bldg #1900, Berkeley, CA, 94720-1900, United States of America, jguajardo@berkeley.edu 1 - Information Provision Policies in Developing Countries: Heterogeneous Farmers and Market Selection Chen-Nan Liao, National Taiwan University, No.1,Sec. 4, Roosevelt Rd., Taipei City, Taiwan - ROC, chennan@berkeley.edu, Ying-ju Chen, Chris Tang We examine the impact of information provision policies on farmer welfare in developing countries where heterogeneous farmers lack relevant information for making market (or crop) selection. We show that the optimal information provision policy may call for limited dissemination, and the government can implement it while overcoming perceived unfairness by providing information to all farmers at a nominal fee. We also examine issues including information dissemination via a for-profit company.

245