2015 Informs Annual Meeting

TB23

INFORMS Philadelphia – 2015

TB21 21-Franklin 11, Marriott Bundled Payment Systems Sponsor: Health Applications Sponsored Session Chair: Danny Hughes, Harvey L. Neiman Health Policy Institute, 1891 Preston White Drive, Reston, VA, 20191, United States of America, dhughes@neimanhpi.org 1 - Optimizing Provider Decisions under Bundled Payments Brenda Courtad, University of Cincinnati, 2925 Campus Green Dr, Cincinnati, OH, 45221, United States of America, courtabl@mail.uc.edu When moving from fee-for-service to bundled payments the providers focus shifts from revenue generating to cost reducing. We develop a partially observable Markov decision process to aid providers in deciding which interventions to implement to reduce costs. 2 - Increasing Healthcare Value in Accountable Care Organizations through Incentive Redesign Christian Wernz, Virginia Tech, Industrial and Systems Engineering, Blacksburg, VA, 24060, United States of America, cwernz@vt.edu, Hui Zhang, Barry Barrios, Danny Hughes ACOs are incentivized by the Centers for Medicare and Medicaid Services (CMS) to lower costs while meeting quality standards. We determined how CMS’ incentive program can be improved, and how ACOs can optimally distribute incentives among their members. Using multiscale decision theory, we performed an in-depth analysis of computed tomography (CT) scanner investments and use in ACO hospitals, calibrated the model with Medicare data, and show how CT scan cost can be lowered and outcomes improved. 3 - Mitigating Small Provider Financial Risk under Prospective Bundled Payment Systems Danny Hughes, Harvey L. Neiman Health Policy Institute, 1891 Preston White Drive, Reston, VA, 20191, United States of America, dhughes@neimanhpi.org, Jeremy Eckhause Retrospective bundled payment models, which cover all medical services within an episode of care, usually include stop loss provisions to manage financial risk. As payments shift to prospective bundled payments, the mechanisms to manage these stop loss provisions may no longer exist. We develop nonlinear programming models to develop pricing strategies that address the inherent higher risk to smaller providers under such a payment system. 4 - Bundled Payments: The Roles of Organization and Diagnosis Turgay Ayer, Georgia Institute of Technology, 765 Ferst Dr., Medicare has started contracting with healthcare providers for bundled payments. However, most providers do not have experience with the risks and opportunities for such payment mechanism. We propose a game-theoretic model to capture the power dynamics between physicians and the hospital under various patient pathways. We use the model to generate hypotheses and test these hypotheses using real data. Atlanta, GA, 30332, United States of America, ayer@isye.gatech.edu, Mehmet Ayvaci, Jan Vlachy

2 - Statistical Guarantees for Individualized Rank Aggregation Sahand Negahban, Yale University, 24 Hillhouse Ave, New Haven, CT, 06510, United States of America, sahand.negahban@yale.edu We study a version of rank aggregation known as collaborative ranking. In this problem we assume that individual users provide us with pairwise preferences and from those preferences we wish to obtain rankings on items that the users have not had an opportunity to explore. We provide a theoretical justification for a nuclear norm regularized optimization procedure. 3 - Inference in High-dimensional Varying Coefficient Models Mladen Kolar, Assistant Professor, Chicago Booth, 5807 South Woodlawn Avenue, Chicago, IL, 60637, United States of America, mkolar@chicagobooth.edu, Damian Kozbur We focus on the high-dimensional linear varying-coefficient model and develop a novel procedure for estimating the coefficient functions. Our procedure works in a high-dimensional regime, under arbitrary heteroscedasticity in residuals, and is robust to model misspecification. We derive an asymptotic distribution for the normalized maximum deviation of the estimated coefficient function and demonstrate how these results can be used to make inference in high- dimensional dynamic graphical models. 4 - Elementary Estimators for High-dimensional Statistical Models Eunho Yang, IBM T.J. Watson, P.O. Box 218, Yorktown Heights, United States of America, eunho@cs.utexas.edu We propose a class of closed-form estimators for sparsity-structured high- dimensional models. Our approach builds on observing the precise manner in which the classical MLE breaks down under high-dimensional settings. We provide a rigorous statistical analysis that shows that our simple estimators recovers the same asymptotic convergence rates as those of computationally expensive L1-regularized MLEs. We corroborate statistical performance, as well as computational advantages via simulations. TB23 23-Franklin 13, Marriott New Advances in Production Planning and Scheduling Cluster: Stochastic Models: Theory and Applications Invited Session Chair: Jingshan Li, Professor, 1513 University Ave, Madison, WI, 53706, United States of America, jli252@wisc.edu 1 - Coordination in Multi-product Manufacturing Systems: Modeling and Analysis Cong Zhao, Research Assistant, University of Wisconsin-Madison, 1513 University Ave, Room 3235, Madison, Wi, 53706, United States of America, czhao27@wisc.edu, Ningxuan Kang, Li Zheng, Jingshan Li Multi-product systems are common in today’s manufacturing process. Effective coordination between products in such systems is important in operation. We study a two-product geometric manufacturing system and derive closed-form expressions of performance measures. An optimal allocation policy of buffer thresholds is developed, and the monotonicity of optimal buffer size with respect to machine parameters is investigated. The managerial insights to achieve optimal production control are discussed. 2 - Level Scheduling in Automotive Assembly Lines and its Effect on the Consumption of Ressources Heinrich Kuhn, Professor, Catholic University of Eichstaett- Ingolstadt, Supply Chain Management & Operations, Auf der Schanz 49, Ingolstadt, 85049, Germany, heinrich.kuhn@ku-eichstaett.de, Dominik Woerner Level scheduling approaches in sequencing of assembly lines are used as substitutional model for the underlying economic and sustainable objectives since a leveled distribution of materials requirements does not necessarily contribute directly to these objectives. We conduct a case study at a major German automotive company selecting relevant part families whose consumption is currently unequal distributed by an extensive simulation study. 3 - A New Scenario Based Sales and Operations Planning Model Nico Vandaele, Professor, KU Leuven, Naamsestraat 69 Box 3555, Leuven, 3000, Belgium, nico.vandaele@kuleuven.be, Catherine Decouttere, Gerd Hahn, Torben Sens We apply a scenario-based approach to the sales and operations planning process where both model-based and non-model based Key Performance Indicators are taken into account. This allows us to balance customer service, derived from aggregate order lead times, and relevant costs of operations when determining volume/mix decisions for internal and external production. An industry-derived case example with distinct outsourcing options is used to highlight the benefits of the approach.

TB22 22-Franklin 12, Marriott Learning High-dimensional/ Sparse Models Sponsor: Applied Probability Sponsored Session

Chair: Varun Gupta, University of Chicago Booth School of Business, Chicago, IL, United States of AmericaVarun.Gupta@chicagobooth.edu 1 - Robust Methods for High-dimensional Regression Po-ling Loh, Assistant Professor, University of Pennsylvania, 3730 Walnut St, 466 Jon M. Huntsman Hall, Philadelphia, PA, 19104, United States of America, loh@wharton.upenn.edu We discuss new methods for robust regression in high-dimensional settings. Our procedures draw upon classical approaches in robust statistics, designed for scenarios where the number of parameters is fixed and the sample size grows to infinity — however, these methods may be adapted to perform robust inference in high dimensions, as well. We also prove that the robust estimators, which involve minimizing nonconvex functions, may nonetheless be optimized to desirable accuracy.

291