Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

TD12

4 - Supply Failure Probability in Pharmaceutical Supply Chains under Input-model Uncertainty Canan Gunes Corlu, Boston University, Alp Akcay, Tugce Martagan We consider a pharmaceutical manufacturer who sources a customized product with unique attributes from a set of unreliable suppliers. We model the likelihood of a supplier to successfully deliver the product via Bayesian logistic regression and study the impact of input uncertainty; i.e., the uncertainty that is due to the estimation of logistic regression parameters from limited amounts of historical data, on the overall supplier failure probability. We investigate how the input- model uncertainty changes with respect to the characteristics of the historical data and the product attributes. n TD12 North Bldg 126A Dynamic Decision Making in OM Sponsored: Manufacturing & Service Oper Mgmt Sponsored Session Chair: Hao Zhang, University of British Columbia, Vancouver, BC, V6T 1Z2, Canada 1 - An ADP Approach to Dynamic Matching with Applications in Kidney Exchange Fan You, University of Colorado-Boulder, 995 Regent Drive, Boulder, CO, 80302, United States, Dan Zhang, Doug Popken We study dynamic matching problems where the objective is to maximize the total matchings over a finite planning horizon with random arrivals and depar- tures. We start by formulating the problem as an MDP, which suffers from the curse of dimensionality. Subsequently, we apply the approximate linear program- ming approach. We derive an upper bound as well as a control policy from the solution of the ALP, and test numerically on a set of dynamic kidney exchange instances taken from the literature. 2 - Optimizing Remote Maintenance Decisions under Imperfect Failure Predictions Guanlian Xiao, Eindhoven University of Technology, Eindhoven, Netherlands, Alp Akcay, Lisa M. Maillart, Geert-Jan Van Houtum We consider a critical component which deteriorates according to a three-state discrete time Markov chain, with two unobservable operational states, and a self- announcing failed state. A remote monitoring center periodically receives imperfect signals and suggests maintenance actions to a local service organization. We build a partially observable Markov decision process that takes the binary signals as input, and minimizes the discounted total corrective and preventive maintenance costs. We characterize the optimal policy by at most one critical threshold, and provide a closed-form critical solution. We also show the impact of imperfectness of signals on the objective values. 3 - Analytical Solution to a Partially Observable Machine- Maintenance Problem Weihua Zhang, University of British Columbia, 6335 Thunderbird Crescent, Vancouver, BC, V6T 2G9, Canada, Hao Zhang We study a classic machine maintenance problem in which the state of the machine changes according to a hidden Markov process under regular production. The state can be observed upon inspection and can also be reset through replacement. The objective is to find a production-inspection- replacement policy that minimizes the expected discounted cost over an infinite horizon. In contrast to the standard Bayesian framework, we adopt a recently established dual framework to derive an exact, analytical solution to this problem. The solution carries a cyclic structure and highlights the core behavior, which explains the puzzling non-monotone optimal policy widely acknowledged in the literature. 4 - Managing Queues with Static Delivery Guarantees Mehdi Hosseinabadi Farahani, PhD, The University of Texas at Dallas, Richardson, TX, 75080, United States, Milind Dawande, Ganesh Janakiraman We study the problem of managing queues in online food ordering services where customers, who place the order online and pick up at the store, are promised a common due-date lead time. The objective is to minimize the total earliness and tardiness costs incurred by the customers. Due to the difficulty of implementing sophisticated policies in practice, we investigate the performance of easily- applicable heuristic policies. Our results show that a simple policy, that starts serving a customer as soon as the time remaining to the due date falls below a threshold, yields near-optimal solutions.

n TD13 North Bldg 126B Data-Driven Models in Healthcare Sponsored: Manufacturing & Service Oper Mgmt/Healthcare Operations Sponsored Session Chair: Vishal Ahuja, Southern Methodist University, Southern Methodist University, Dallas, TX, 75275, United States 1 - Utilizing Data Driven Decision Support Systems to Reduce Readmission Rates for Patients with Congestive Heart Failure J K. Srinivasan, Boston University, Boston, MA, United States, Kellas Ross Cameron Physicians currently attempt to identify CHF patients that are likely to be readmitted within 30 days. However, the readmission rate for these patients remains over 20%. Data-driven decision support systems can provide an additional tool to understand which patients are at high-risk of readmission. We create a statistical model to utilize patient-specific information to not only more accurately identify the patients most likely to be readmitted, but also why - whether for condition-related reasons or not. This allows physicians to suggest patient-specific readmission prevention strategies. 2 - Being on the Productivity Frontier: Identifying Triple Aim Performance Hospitals Sriram Venkataraman, University of South Carolina, Department of Management Science, Moore School of Business, Columbia, SC, 29208, United States, Aleda Roth, Anita L. Tucker, Jon A. Chilingerian Hospital decision-makers may face trade-offs that make it difficult to obtain the trifecta of high performance on clinical quality, patient experience, and technical efficiency. Using datasets from 2010 and 2012 and data envelopment analysis, we identify more than 40 triple aim performance hospitals among U.S. acute care hospitals with 200 beds or more. Further, we find that the percentage of physicians employed by the hospital has a positive and significant relationship with triple aim performance, and that bed utilization rate has a positive relationship with technical efficiency but a negative relationship with clinical quality and patient experience performance. 3 - Reduction of Patient and Surgery Preparation Time in the Operating Room Yann Ferrand, Clemson University, Clemson, SC, United States, Brandon Lee, Dee San, Kevin M. Taaffe, Lawrence Fredendall, Amin Khoshkenar, Anjali Joseph, Scott Reeves This paper used a set of cameras to record all the activities in the operating room during 26 surgical procedures and the room preparation periods. Four of these were selected for intensive analysis of the room turnaround and surgery preparation times. All activities by all staff in the room were coded and placed into a Gantt chart format timeline. The timeline was used to identify potential bottleneck resources and how slack resources could begin activities in parallel with the bottleneck resources to reduce the overall length of the turnaround time. 4 - Quality Improvement Spillovers: Evidence from the Hospital Readmissions Reduction Program Mohamad Soltani, University of Wisconsin-Madison, 4284A, 975 University Ave., Madison, WI, 53706, United States, Robert Batt, Hessam Bavafa In this study, we look at the impact of the Hospital Readmissions Reduction Program in the US hospitals to examine (1) the extent to which hospitals were able to achieve 30-day readmission reductions for target patients and (2) the extent to which non-target patients experienced “spillover improvements in 30- day readmissions. We also study 31-60-day readmissions and length of stay as possible mechanisms by which hospitals can achieve improvements in 30-day readmissions. Taken together, our results show that this policy has been effective and has generated significant beneficial spillovers in quality improvement.

350