Informs Annual Meeting 2017

SD44

INFORMS Houston – 2017

SD45

3 - The Business of Healthcare: Physician Integration in Bundled Payments

360D JFIG Paper Competition I Sponsored: Junior Faculty JFIG Sponsored Session Chair: Jose Luis Walteros, University at Buffalo, SUNY, 413 Bell Hall, Buffalo, NY, 14260, United States, josewalt@buffalo.edu 1 - JFIG Paper Competition I Jose Luis Walteros, University at Buffalo, SUNY, 413 Bell Hall, Bu, NY, 14260, United States, josewalt@buffalo.edu The 2017 JFIG paper competition features paper submissions from a diverse array of talented junior faculty members. The prize committee evaluated submissions based on the importance of the topic, appropriateness of the approach, and significance of the contribution. After careful review, the prize committee selected a group of finalists to present their research in one of the two JFIG sessions. For information on the finalists and their papers, please refer to the online program. 2 - Optimal Signaling Mechanisms in Unobservable Queues David Lingenbrink, Cornell University, Ithaca, NY, United States, d.lingenbrink@gmail.com, Krishnamurthy Iyer We consider the problem of optimal information sharing in the context of a service system. In particular, we consider an unobservable single server queue offering service at a fixed price to a Poisson arrival of delay-sensitive customers. The service provider can observe the queue, and may share information about the state of the queue with each arriving customer. The customers are Bayesian and strategic, and incorporate any information provided by the service provider into their beliefs about the queue size before deciding whether to join the queue or leave without obtaining service. We pose the following question: which signaling mechanism should the service provider adopt to maximize her revenue? We establish that, in general, the optimal signaling mechanism requires the service provider to strategically conceal information from the customers to incentivize them to join. In particular, we show that a signaling mechanism with two signals and a threshold structure is optimal. Furthermore, for the case of linear waiting costs, we obtain analytical expressions for the thresholds of the optimal signaling mechanism. Finally, we prove that the optimal signaling mechanism under the optimal fixed price can achieve the revenue of the optimal state-dependent pricing mechanism. This suggests that in settings where state-dependent pricing is not feasible, the service provider can effectively use optimal signaling to achieve the optimal revenue. Our work contributes to the literature on Bayesian persuasion in dynamic settings, and provides many interesting directions for extensions. 3 - Optimization of Tree Ensembles Tree ensemble models such as random forests and boosted trees are among the most widely used and practically successful predictive models in applied machine learning and business analytics. Although such models have been used to make predictions based on exogenous, uncontrollable independent variables, they are increasingly being used to make predictions where the independent variables are controllable and are also decision variables. In this paper, we study the problem of tree ensemble optimiza- tion: given a tree ensemble that predicts some dependent variable using controllable independent variables, how should we set these variables so as to maximize the pre- dicted value? We formulate the problem as a mixed-integer optimization problem. We theoretically examine the strength of our formulation and exploit the structure of the problem to develop two large- scale solution methods, one based on Benders decom- position and one based on iteratively generating tree split constraints. We test our methodology on a real data set in the domain of drug design, where we show that our methodology can efficiently solve large-scale instances to near or full optimality and outperforms solutions obtained by heuristic approaches. Moreover, specific to the drug design context, our approach can be used to efficiently identify candidate compounds that optimally trade-off predicted performance and novelty with respect to existing, known compounds. Velibor Misic, UCLA Anderson School of Management, 110 Westwood Plaza, Suite B406, Los Angeles, CA, 90095, United States, velibor.v.misic@gmail.com

Turgay Ayer, Georgia Institute of Technology, School of Industrial and Systems Engineering, Groseclose 417, Atlanta, GA, 30332, United States, ayer@isye.gatech.edu, Jan Vlachy, Mehmet U.S. Ayvaci, Srinivasan Raghunathan Under the prevailing fee-for-service payments, incentives of hospitals and physicians are misaligned, leading to large inefficiencies. Bundled payments unify payments to the hospital and physicians and are expected to encourage care coordination and reduce costs. However, as hospitals differ in their relationships with physicians in influencing care level of physician integration, it remains unclear what spectrum of physician integration will facilitate bundling. We study (1) the impact of the level of integration between the hospital and physicians in the uptake of bundled payments, (2) the consequences of bundling with respect to overall care quality and costs/savings. 4 - Hospital Quality, Medical Charge Variation, and Operational Performance: Implications for Bundled Payment Reform Models Seokjun Youn, Texas A&M.University, 320 Wehner Building, 4217 Texas A&M.University, College Station, TX, 77840-4217, United States, syoun@mays.tamu.edu, Gregory R. Heim, Subodha Kumar, Chelliah Sriskandarajah The premise of a bundled payment is to aggregate a number of care delivery services around an episode of care, and to set a priori reimbursement irrespective of actual cost. As such, the bundled payment aims to concomitantly “decrease variability,” resulting in reduced and predictable costs. However, a fundamental question, whether the lower variability in practice can leads to better operational performances, is unexplored. We tackle this issue preceded by measuring practice variation based on fee-for-service medical charges of inpatients in NY and FL. We provide implications for designing incentives of bundled payment programs. 360C Pro Bono Analytics Sponsored: Public Sector OR Sponsored Session 1 - Pro Bono Analytics Panel Session David T. Hunt, Oliver Wyman, One University Square, Suite 100, Princeton, NJ, 08540, United States, david.hunt@oliverwyman.com Pro Bono Analytics (PBA) is an INFORMS program to match analytics professionals who are willing to volunteer their skills with nonprofit organizations that would benefit from analytical techniques. This panel session will begin with a brief description of some actual PBA projects, and will then focus on a discussion by volunteers and nonprofit representatives about the experiences and challenges of introducing analytics solutions into nonprofit organizations. The discussion will include defining the project outcome, overcoming data problems, and implementing the project results. 2 - Panelist Vineet Madasseri Payyappalli, University at Buffalo, 38 Embassy Square Apt 7, Tonawanda, NY, 14150, United States, vineet.mp86@gmail.com 3 - Panelist Adam Clark, U.S. Department of Defense, Washington, DC, United States, adamclark.usa@gmail.com 4 - Panelist Scott Arthur, DePelchin Children’s Center, Houston, TX, United States, sarthur@sohmission.org 5 - Panelist Jason Lau, DePelchin Children’s Center, Houston, TX, jlau@depelchin.org SD44

116