Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

POSTERS

28 - Land-use Optimization of a Watershed for Nutrient Reduction Under Stochastic Weather Conditions Gorkem Emirhuseyinoglu, Iowa State University, Ames, IA, United States, Sarah M. Ryan Agricultural runoff causes nutrient loads in waterways and creates a hypoxia zone downstream, which threatens oceanic life. We build a land-use optimization model to minimize the cost for a watershed to meet target reductions in nitrate and phosphorus levels under stochastic weather conditions. Results are illustrated using an online tool. 29 - Stochastic Programing Models to Plan for Distributed Generation Under Wind Power Volatility We present non-linear stochastic integer programming models to find the optimal wind turbine capacities and locations that minimize costs of adopting renewable energy considering loss-of-load probability, thermal constraints, and volatilities on wind turbine power generation. The models solve exactly and through simulation optimization. The models include scenarios that represent the wind speed across the turbine blades over the different operational phases of the wind turbines. Wind speed data collected for Wellington, New Zealand and Rio Gallegos, Argentina permit to estimate the probabilities associated to the scenarios and illustrate the benefits of the stochastic models. 30 - Stochastic Dynamic Markov Decision Process on Airport Security Checkpoint Demand Shifting Nigel Pugh, PhD Student, North Carolina A&T State University, Greensboro, NC, 27409, United States, Hyoshin Park Airport Delays known to cost airport operation of U.S. airlines billions of dollars each year. In this paper, we focus on reducing airport congestion from both passenger and airport operations perspectives. This is accomplished through incentivizing the passenger to switch arrival times to encourage a system optimal. The demand is then used as an input for worker assignment problem of switching airline workers based on current demand. A Markov Decision Process is proposed to maximize reward by making best decision of whether to switch airline worker security checkpoint location depending on the current demand of the congestion state. 31 - Data Analytics in Agricultural Business Hieu Pham, Syngenta, Slater, IA, 50244, United States Advancing analytical techniques in agriculture are vital to satisfying increased food demand. At the forefront of the agricultural data science, revolution is Syngenta with innovative machine learning algorithms (sparse biclustering) and novel optimization models (portfolio selection) to mitigate world hunger. 32 - A Novel Dynamic Routing Framework for Shared Mobility Services Yue Guan, Massachusetts Institute of Technology, Cambridge, MA, 02139, United States Shared Mobility on Demand (MoD) services like Uber, Lyft, and Didi have shown that a continuum of solutions can be provided between traditional private transport and public mass transit. Here we propose a novel shared MoD framework which generates a dynamic route for multi-passenger transport, using a new concept of space window that introduces a degree of freedom to help reduce system cost in designing the optimal route. An Alternating Minimization based algorithm is developed. Its analytical properties are characterized. Detailed computational experiments are carried out that demonstrate its advantages in computational efficiency and optimality compared to standard optimization solvers. 33 - An Agent-based Simulation Model for Assessing a Charging Infrastructure Design for a Public Electric Vehicle Charging Network Long Zheng, University of Louisville, Louisville, KY, 40217, United States, long.zheng@louisville.edu, Lihui Bai We present an agent-based simulation model for performance assessment of the infrastructure design of a public EV charging network. The model, comprised of three agents (EV, battery and charging station), is capable of handling heterogeneous trips in an urban traffic network (e.g., commute, shopping). Further, the model considers a mix of charging sources: public, home charging and workplace charging, available to the EV fleet. The model can be used in conjunction with an optimization model for an optimal design of the charging infrastructure at both strategic and operational levels. 34 - A New Markov Processes Methodology for Identifying Screening Strategies for Breast Cancer in Low- and Middle-income Countries: Application to Peru Vijeta Deshpande, University of Massachusetts, Amherst, MA, 01002, United States, Chaitra Gopalappa, Shifali Bansal Peru is currently in preparation of a national æinvestment case’ for the prevention of premature mortalities from non-communicable diseases including breast cancer. This is informed by quantitative economic analysis of current and future health interventions. We apply a Markov process method for Clara Novoa, Associate Professor, Texas State University, San Marcos, TX, 78666, United States, Temitope Runsewe, Jordan Givens, Tongdan Jin

parameterization of a natural cancer progression model for conducting an economic analysis of interventions for Peru. We also apply a Markov decision process model for identifying optimal number of mammography screenings and ages to screen, under different willingness to pay per life-year saved to model the trade-offs in costs and impacts. 35 - Prediction of Inpatient Length of Stay Using Bootstrap Aggregating Machine Learning Algorithm Sohong Chakraborty, California State University, Los Angeles, CA, 90032, United States, Shilpa Balan, Divya Pakhale Hospitals face the daunting task to provide timely patient care. This research aims at improving the inpatient admission rate by predicting the length of stay in an inpatient care facility. A National Inpatient Sample (NIS) data for the year 2013 is considered for this study, which is available through the Healthcare Cost and Utilization Project (HCUP). We predicted the length of stay of patients using the Bagging or Bootstrap aggregating algorithm. The independent variables considered for this study are age, diseases, diagnosis categories, time of admission, hospital charges and emergency department services. The correlation coefficient for the inpatient length of stay is found to be 0.842. 36 - A Latent Markov Model for Predicting Return to Work for Injured Workers Suyanpeng Zhang, University of Michigan, Ann Arbor, MI, 48109, United States, Haozhu Wang, Mukai Wang, Brian T. Denton, Jenna Wiens, Jon Seymour We present a data-driven approach based on the use of latent Markov models for predicting the probability of returning to work for workers injured in the workplace. We show how models fit to observational longitudinal data on injuries and follow-up procedures and treatment can be used to classify patients based on sequential observations over time. We demonstrate the proposed approach using longitudinal data from a large national data set. 37 - Stochastic Optimization Algorithms for Robust Medical Decision Making Healthcare is increasingly reliant on mathematical models, such as Markov Decision Processes, to monitor disease progression and improve patient care. To address the abundance of parameter ambiguity from clinical data, the Multi- Model Markov Decision Process (MMDP) was invented. We designed a custom branch-and-bound algorithm to solve large, practical MMDPs that was orders of magnitude faster than commercial mixed-integer programming software, for a set of test cases. In addition, we implemented various objective functions allowing for different risk preferences towards parameter ambiguity. We present a relevant medical decision-making case study to illustrate our approach. 38 - An Analytic Approach to Incorporate the Six Aims for Quality in the Analysis of Trauma Care Services Lucy Aragon, Wichita State University, Wichita, KS, 67260, United States, Laila Cure, Karen Schieman The Institute of Medicine proposed six aims to guide healthcare quality improvement efforts. However, most healthcare improvement programs still evaluate quality along one aim at a time, effectiveness. This research proposes an analytic approach to incorporate all six aims in the evaluation of healthcare quality. A trauma care setting is used to investigate data requirements, develop the methodology and evaluate its implications. 39 - Impact of Referral Strategy on a Medical Network Shao-Jen Weng, Associate Professor, Tunghai University, Taichung City, 407, Taiwan, Ping-Wen Huang Hospital emergency departments have been becoming more congested, and thus one issue of interest is how to better allocate limited medical resources and make more effective medical care referral strategy decisions. This study thus uses the system simulation to design and simulate a medical system for analyzing different kinds of referral strategies to enhance medical bed usage efficiency in each hospital, and promote the quality of service offered by shortening the waiting times of patients with regard to getting a medical bed. 40 - Multi-method Simulation Modeling of the Hospital Readmission Reduction Program Arlen Dean, Arizona State University, Tempe, AZ, 85281, United States, Michelle Alvarado The Hospital Readmission Reduction Program (HRRP) was enacted to penalize hospitals with poor readmissions. A multi-method simulation model is created to capture the interaction of hospitals and the insurer under the HRRP. The goal is to improve healthcare quality by studying how a payment adjusting policy affects hospital decision making. Vinayak Ahluwalia, University of Michigan, Ann Arbor, MI, 48109, United States, Charmee Kamdar, Lauren N. Steimle, Brian T. Denton

341