2015 Informs Annual Meeting

WA21

INFORMS Philadelphia – 2015

3 - Solving A Real-world Snow Plow Optimization Problem: An Integrated Solution Approach Joris Kinable, Post-doctoral Researcher, Carnegie Mellon University, 5000 Forbes Avenue, Pittsburgh, PA, 15213, United States of America, jkinable@cs.cmu.edu, Stephen F. Smith, Willem-jan Van Hoeve Each year, many northern cities are faced with significant expenditures pertaining winter road maintenance. Snow plowing constitutes a significant part of these costs. This work presents an integrated, adaptive solution approach for a real- world snow plow optimization problem. The large number of routing and scheduling constraints render this problem particularly hard to solve. The performance of our solution approach is demonstrated on data from the city of Pittsburgh (USA). 4 - On Solving Quadratic Assignment Problems in Wireless Communications Hans Mittelmann, Arizona State University, Box 871804, Tempe, AZ, United States of America, mittelmann@asu.edu In digital wireless communications optimal index assignment leads to difficult quadratic assignment problems. Those are standard QAPs for single transmissions or for sequential multiple transmissions. They become higher dimensional QmAPs when simultaneously optimizing over several retransmissions. We report on the exact and approximate solution of such problems that arise in practice. Health Care Operations Sponsor: Health Applications Sponsored Session Chair: Qiushi Chen, chenqiushi0812@gatech.edu 1 - Can an Early Warning Score Predict Patients’ Hospital Length of Stay and Mortality? Nasibeh Azadeh-fard, PhD Candidate, Virginia Tech, 544 Whittemore Hall, Virginia Tech, Blacksburg, VA, 24061, United States of America, nasibeh@vt.edu, Jaime Camelio, Navid Ghaffarzadegan The Modified Early Warning Score (MEWS) is used in hospitals to quickly predict and prevent catastrophic events such as death. The prediction power of MEWS, however, is an empirical question. We analyze effectiveness of MEWS in a major hospital in the US over six months for a sample of 1021 patients. We find that MEWS modestly predicts hospital length of stay and death, while physicians’ specific characteristics and their subjective assessments are much better predictors of health outcomes. 2 - Routing Patients to Community Health Services to Maintain Patient Access after Facility Merger Aaron Ratcliffe, Assistant Professor, University of North Carolina at Greensboro, 438 Bryan Building, P.O. Box 26170, Greensboro, NC, 27402, United States of America, aaron.ratcliffe@uncg.edu Merging the facilities dedicated to a health service may allow for cost savings in terms of economies of scale and other efficiency improvements at the expense of poorer access to services for patients. We develop a queueing network model to examine how a social planner should route heterogeneous patient classes to community health resources to improve patient access in the absence of a previously dedicated facility. 3 - Integrated Staff and Room Scheduling for Surgeries: Methodology and Application Sandeep Rath, PhD Candidate, UCLA Anderson, B501 Gold Hall, UCLA Anderson, Los Angeles, CA, 90024, United States of America, Sandeep.Rath.1@anderson.ucla.edu, Kumar Rajaram We consider the problem of minimizing resource usage and overtime costs across multiple parallel resources such as anesthesiologists and operating rooms at a large multi-specialty hospital. We develop a two stage optimization program with recourse. We develop a data driven robust optimization method that solves large- scale real-sized versions of this model close to optimality. We validate and implement this model as a decision support system at the UCLA Ronald Reagan Medical Center. WA21 21-Franklin 11, Marriott

4 - Arbovirus Risk Maps in Texas Xi Chen, University of Texas at Austin, Austin, TX 78712, Austin, TX, United States of America, carol.chen@utexas.edu, Nedialko Dimitrov Dengue fever and Chikungunya virus two key mosquito-borne diseases in Texas. To focus state resources, public health officials need to identify the geographic risk areas for these diseases. We consider thousands of possible risk models, based on maximum entropy methods, combined with data on the transmission vectors, environmental, and socio-economic factors. We select the best model empirically, using historical Texas Dengue data. The final model is in use by Texas health officials. WA23 23-Franklin 13, Marriott Stochastic Modeling and Analysis with Applications Sponsor: Applied Probability Sponsored Session Chair: Jing Dong, Northwestern University, 2145 Sheridan Road, Tech C210, Evanston, United States of America, jing.dong@northwestern.edu Co-Chair: Jose Blanchet, Associate Professor, Columbia University, 500 W 120th St., Mudd Building, IEOR, 3rd Floor., New York, NY, 10027, United States of America, jose.blanchet@columbia.edu 1 - Stationarity and Interchange of Limits in Heavy Traffic Analysis We develop a streamlined approach for justifying the heavy traffic stationary approximation of stochastic processing networks. First, we demonstrate that the stability of a deterministic dynamic complementarity problem is sufficient for both the diffusion limit and pre-limit networks to have stationary distributions. Then, given an additional mild condition, we show the convergence of stationary distributions of pre-limit networks to that of the diffusion limit. 2 - Resource Allocation in Bike Sharing using Coupling and Linear Programming Shane Henderson, Professor, Cornell University, Rhodes Hall, Ithaca, NY, 14853, United States of America, sgh9@cornell.edu, David Shmoys, Eoin O’mahony We propose an optimization problem that allocates bike racks and bikes to stations across a city. The objective is a transient performance measure from a continuous-time Markov chain. We show that the objective possesses a (joint) discrete convexity property that allows for efficient solution via linear programming. The proof uses a combination of geometrical arguments and coupling theory. The results are illustrated using Citibike data in NYC. 3 - Tail Analysis Without Tail Information: A Worst-case Perspective Henry Lam, Assistant Professor, University of Michigan, 1205 Beal Ave., Ann Arbor, MI, 48109, United States of America, khlam@umich.edu, Clementine Mottet One common bottleneck in tail modeling is that, due to their very nature, tail data are often very limited. Rather than using conventional parametric fitting, we will describe a robust alternative that is based on a worst-case analysis under the geometric premise of tail convexity. We demonstrate that the worst-case convex tail behavior is either extremely light-tailed or extremely heavy-tailed, and construct numerical schemes to distinguish between the two cases and find the worst-case tail. 4 - On the Stability of Matching Queues Pascal Moyal, Université de Technologie de CompiËgne, Rue du Dr Schweitzer, Compiègne, 6023, France, pascal.moyal@utc.fr, Ohad Perry Consider a model in which, to each node of a graph G is associated an arrival process, and any entering item associated to node k either leaves the system if it finds in line, another item corresponding to a neighbor of k, or is stored in queue. Using fluid analysis, we investigate the stability of such matching models, which are of increasing practical importance. We show that, aside for a specific class of graphs, they can always be unstable even under a natural necessary stability condition. Hengqing Ye, Associate Professor, Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong - PRC, lgtyehq@polyu.edu.hk, David D. Yao

380