Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

TE63

2 - A Dynamic Call-in Control for Unpunctual and Impatient Customers Yunzhe Qiu, Washington University in St. Louis, 1 Brookings Dr., St. Louis, MO, 63130, United States, Jie Song Motivated by the increasing demand for health care service and uncertainty of patient arrivals, we propose a real-time dynamic appointment scheduling system with heterogeneous customers whose sensitivity to the waiting time are varied. We construct a continuous time Markov Decision Process in a finite horizon, and derive a myopic optimal policy with the switching-line based on the current information to minimize the total waiting cost. The main finding is the widely used FCFS policy is optimal within homogeneous customers but not optimal if heterogeneity is allowed. An empirical study based on the Pediatric clinic of the Peking University Third Hospital validates our results. 3 - A Simulation Optimization Approach to Nursing Home Capacity Planning with Residents Heterogeneity Xuxue Sun, University of South Florida, Tampa, FL, 33620, United States, Nazmus Sakib, Nan Kong, Chris Masterson, Hongdao Meng, Kathryn Hyer, Mingyang Li With rapid population aging and overwhelming acute care utilization at hospitals, nursing homes (NHs) are responsible for meeting with the excess demand and increasing acuity of older adults. Successful NH care preparedness becomes significant yet challenging, since NH residents suffer from diverse chronic diseases and functional disabilities, and the service need of each resident varies over time. This work proposes a simulation optimization approach to determining the optimal bed number and staff capacities subject to the heterogeneous service demand. 4 - Hospital Readmission Reduction Strategy Using Stochastic Programming Behshad Lahijanian, University of Florida, 303 Weil Hall, P.O. Box 116595, Gainesville, FL, United States, Michelle M. Alvarado The Hospital Readmission Reduction Program (HRRP) aims to reduce hospital readmissions by applying a financial penalty to hospitals whose readmission rates are worse than their peer group median. We develop a stochastic program with probabilistic constraints for the hospital’s optimal care strategy in response to HRRP. The model maximizes the hospital’s profit and utilizes probabilistic constraints to control uncertain readmission probabilities across all patients. Additionally, we explore the impact of penalty and incentive HRRP designs on the hospital’s profit and care strategies. n TE62 West Bldg 103A Joint Session DM/Practice Curated: Data Science in Health Care I Sponsored: Data Mining Sponsored Session Chair: Kaiye Yu, Tsinghua University, Beijing, 100084, China 1 - A Multi-task Warped Gaussian Process Learning for Predicting the Optimal Next Line of Therapy in Multiple Myeloma Emisa Nategh, PhD Student, University of Washington, Dept of Info Systems & Operations Management, Michael G. Foster School of Business, Seattle, WA, 98195-3226, United States, David Coffey, Hamed Mamani, Yingfei Wang The objective of this research was to develop a model to predict the sequence of therapy which results in the best progression free survival (PFS). In this paper, we formulate the next line of treatment PFS based on previous line of treatment PFS and its effect on overall survival as a multi-task learning problem in the context of warped Gaussian process. We propose a novel methodology by using composite kernel over tasks in our model to improve our result. We illustrate its practicability by applying it to sequencing of therapy in cancer patients. 2 - Making Predictions of Pediatric Outpatient Clinic Room Requirements to Improve Room Allocation Policies Tian He, Graduate Student, University of Pittsburgh, 3700 Ohara St, 1025 Benedum Hall, Pittsburgh, PA, 15261, United States, Kiatikun Louis Luangkesorn, Pranay Mohanty Children’s Hospital of Pittsburgh provides space for over 20 specialty outpatient pediatrics clinics on site. These clinics have been having issues with patient access, the ability of patients to schedule an appointment in a timely fashion. While the specialties state that lack of additional space prevents them from increasing capacity, the rooms are currently underutilized on the whole. We develop and apply predictive models based on data found in Electronic Health Records to identify on a session by session basis the presence of excess room which could be reallocated to services that require additional rooms to meet known demand.

3 - A Data Driven Analytical Framework for Hospital Readmission Prediction Kaiye Yu, Tsinghua University, Room 520, Shunde Building, Beijing, BJ100084, China, Xiaolei Xie The improvement effort to reduce hospital readmissions necessitates reliable risk prediction tool. We develop and validate an analytics framework for predictive modeling of all-cause readmissions using data-driven approaches. Dimensionality reduction, predictors identification, data sampling and imbalance data handling are performed. A mix-ensemble framework which is integrated by the base machine learning classifiers is proposed after the comparison of the different prediction techniques using hospital administrative data. Finally, the models are assessed and managerial insights are obtained. 4 - A Data Driven Analytical Framework for Hospital Readmission Prediction Kaiye Yu, Associate Professor, Tsinghua University, Beijing, China, Xiaolei Xie The improvement effort to reduce hospital readmissions necessitates reliable risk prediction tool. We develop and validate an analytics framework for predictive modeling of all-cause readmissions using data-driven approaches. Dimensionality reduction, predictors identification, data sampling and imbalance data handling are performed. A mix-ensemble framework which is integrated by the base machine learning classifiers is proposed after the comparison of the different prediction techniques using hospital administrative data. Finally, the models are assessed and managerial insights are obtained. n TE63 West Bldg 103B Joint Session DM/Practice Curated: Data Science for Forecasting Sponsored: Data Mining Sponsored Session Chair: Guangrui Xie, Virginia Tech, Blacksburg, VA, 24061, United States 1 - A Data-driven Forecasting Approach for Newly Launched Seasonal Products Tugba Efendigil, MIT, Cambridge, MA, United States, Vicky Wing Kei Chan, Majd Kharfan Demand forecasting is becoming a very complicated process in fashion industry due to the short product life-cycle, the obsolescence of the retail calendar, and the lack of information for newly launched seasonal items. Therefore, this study focuses on demand prediction with a data-driven perspective both leveraging machine-learning techniques and identifying significant predictor variables to help fashion retailers achieve better forecast accuracy. 2 - Sparse Reduced-rank Regression with Applications We present a sparse estimation of the reduced-rank multivariate regression model. Our estimation proposal minimizes the block-$\ell_1$ norm of the rows of the coefficient matrix subject to a sparsity inducing $\ell_\infty$ norm. Extensive simulation results will be presented to demonstrate effectiveness of the method. We apply the method to build a predictive model for the interest rates time series covering treasuries, corporate, term spreads and public-private spreads, from the FREDQD data. Theoretical properties of the proposed method will also be presented. 3 - An Adaptive Data Communication Scheme for Bandwidth Limited Residential Load Forecasting Guangrui Xie, Virginia Polytechnic Institute and State University, 112 Durham Hall, 1145 Perry Street, Blacksburg, VA, 24061, United States, Xi Chen, Yang Weng While adding new capabilities, the proliferation of distributed energy resources makes it challenging to provide reliable power and voltage forecast for operational planning purposes. We propose an integrated Gaussian Process-based method(IGP) for hourly electric load forecasting, which utilizes not only the data streams generated by the target customer but also those of relevant customers in the feeder system. An adaptive data communication scheme is further proposed to maintain the high forecast accuracy of IGP when a data communication bandwidth constraint is imposed in some feeders. The superior efficacy of IGP and the adaptive data comunication scheme is tested on various IEEE test cases. Haileab Hilafu, Assistant Professor, University of Tennessee, 916 Volunteer Blvd, Knoxville, TN, 37996, United States

401