Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

SC78

4 - Multifidelity Models for Buffer Allocation Problems Andrea Matta, Politecnico di Milano, Via La Masa 1, Milan, 20156, Italy, Ziwei Lin, Nicla Frigerio Buffer allocation problems in flow line require the estimation of the expected throughput of the system. Literature reports a rich set of optimization methods using either analytical methods or discrete event simulation as performance evaluation models. The two kinds of models are usually used sequentially at different levels of the optimization phase. This work explores the advantages and drawbacks of combining them into the buffer allocation problem. Different multi- fidelity regression models are used as surrogate models for estimating the expected system throughput in the buffer allocation problem. Results show that using low fidelity models helps to reduce the total computational effort. 5 - Simultaneous Optimization of Buffer Allocation, Routing, & Population in Closed Finite Queueing Networks James Smith, University of Massachusetts at Amherst, MA, United States, Sencer Yeralan Optimal routing in closed finite queueing networks is a challenging problem all by itself. When it is combined with optimal buffer allocation and the population of the closed network,we achieve a mixed integer nonlinear programming problem. A queueing performance decomposition approach combined with a sequential quadratic programming algorithm is the main vehicle for the analysis and synthesis of the problem. Various closed network topologies are utilized to demonstrate the efficacy of the methodology. n SC77 West Bldg 213A Decision Analytic Models at the Centers for Disease Control and Prevention Sponsored: Public Sector OR Sponsored Session Chair: Chaitra Gopalappa, University of Massachusetts, Amherst, MA, 01003, United States Co-Chair: Seyedeh Khatami 1 - Estimating HIV Transmission Rates Along the HIV Care Continuum in the United States Zihao Li, Centers for Disease Control and Prevention, Atlanta, GA, United States, Yao-Hsuan Chen, Chaitra Gopalappa, Paul Farnham, Stephanie Sansom Estimating HIV transmission rates is critical to understanding how to reduce disease spread. We used the Progression and Transmission of HIV (PATH), an agent-based model that replicates HIV transmission in the United States, to estimate 2015 transmission rates among persons at various stages of HIV care. The overall transmission rate in 2015 was 3.51 per 100 person-years. The rate was 0.07 per 100 person-years for persons in care and virally suppressed compared with 15.64 per 100 person-years for persons newly infected and undiagnosed. Viral suppression is associated with greatly reduced transmission. Thus, strategies to promote viral suppression may be very effective. 2 - Marginal Cost-effectiveness of Pre-exposure Prophylaxis Compared with Improving HIV Care and Treatment in the United States Nidhi Khurana, CDC, 1600 Clifton Road MS E-48, Atlanta, GA, 30333, United States, Paul G. Farnham, Katherine A. Hicks, Justin Carrico, Stephanie L. Sansom We used a dynamic, compartmental model of the sexually active U.S population to estimate the marginal cost-effectiveness of preventing HIV infection with pre- exposure prophylaxis (PrEP) compared with preventing HIV transmission through enhanced care and treatment of persons with HIV. We modeled the application of PrEP to persons at high risk of acquiring HIV and investigated its marginal cost- effectiveness for the entire population as well as for individual transmission groups (men who have sex with men, persons who inject drugs, and heterosexuals). We assessed cumulative costs, new infections, and quality- adjusted life-years from 2016 through 2020. 3 - Using System Dynamics to Investigate HIV Positive Testing Yield in Malawi Roma Bhatkoti, CDC, Atlanta, GA, United States, Andrew Auld, Mtemwa K. Nyangulu, George Bicego With the goal of ending the AIDS epidemic by 2030, UNAIDS set an ambitious target of diagnosing 90% of people living with HIV by 2020 as part of their 90-90- 90 framework. As HIV testing programs have expanded over the past few years to meet this target, it has been observed that HIV positive diagnostic yield (a critical factor for determining HIV testing targets) has decreased over time. In order to investigate this declining HIV diagnostic yield and to identify ways to potentially optimize HIV testing strategies, a simulation model was constructed using system dynamics depicting different population stocks. Publicly available data from UNAIDS and the Malawi HIV testing program were used to model yield.

4 - How to Reach Zero HIV Incidence in the US? A Reinforcement Learning Approach to Optimizing HIV Intervention Decisions Seyedeh Khatami, University of Massachusetts Amherst, Amherst, MA, United States HIV has been a persistent issue in the US. The National HIV/AIDS Strategy (NHAS) for the United States aims to reduce HIV incidence by 25%, by 2020, by increasing coverage on all key interventions that are independently evaluated to be cost-effective. Due to interactions between interventions and the dynamic changes in the epidemic over time, we propose to identify combinations of interventions that can optimally lead to zero incidence, formulated as a sequential decision-making problem. We formulate the problem as a Markov decision process (MDP) with multiple decision stages over a finite time horizon and solve using reinforcement learning, a simulation-based control optimization algorithm. n SC78 West Bldg 213B Location Models Sponsored: Location Analysis Sponsored Session Chair: Oded Berman, University of Toronto, Toronto, ON, M5S 3E6, Canada Co-Chair: Dmitry Krass, University of Toronto, Toronto, ON, M5S 3E6, Canada 1 - Benefit Maximizing Network Design in the Public Sector Robert Aboolian, California State University-San Marcos, 2771 Palmetto Drive, Carlsbad, CA, 92009, United States, Oded Berman Governments are involved in providing essential services, such as healthcare, transportation, education and utilities. Their mandate is to maximize the societal benefit by acting as agents of the public in contrast with the private firms’ mission to maximize profit. Many models in this area try to (re-)design the public service so as to maximize the number of people who will benefit from the program given a limited budget. These models do not consider the marginal benefit (savings in costs to tax payers by adding an extra unit of service capacity) provided. In this work, we determine the optimal number, locations and capacities of a network of facilities so as to maximize the overall benefit to the public. 2 - Uniform Price Approximation in Elastic Demand Location Models We consider a location model with stochastic customer demand where demand rate is affected by service quality (waiting) and price. We show that under certain assumptions, the optimal pricing policy is uniform, with identical price charged at every facility. We then apply this policy as a heuristic for the general case, deriving both theoretical and computational performance bounds. 3 - Locating Time-knapsacks for Optimising Operational Performance Mozart Batista de Castro Menezes, Kedge Business School - Bordeaux, Office: 1449, Talence, 33405, France, Diego Ruiz-Hernandez In this work we analyse the way a manufacturer’s fixed-costs should be allocated to each production order rather than the way they are currently allocated. Using that information, we introduce a facility location model where fixed-costs are incurred when opening a facility; time-knapsacks representing production lines are assigned to the facility, adding another layer of fixed-costs; and finally, production orders are allocated to knapsacks with corresponding variable costs. Using a real situation, resulting solutions have higher capacity utilisation, smaller and fewer facilities than current network design. Serving demand from closest facility is also less frequent. 4 - Multi-period Home Healthcare Provider Facility Location-allocation Problem Vahid Roshanaei, University of Toronto, 1706 35 Charles Street West, Toronto, ON, M4Y 1R6, Canada, Oded Berman, Opher Baron We study the facility location-allocation problem of a multi-period home healthcare provider in Toronto, Canada. We develop a mixed-integer programming model that considers (i) nurses’ flexibility where nurses of higher skills can perform the tasks of nurses with lower skills, (ii) patients’ continuity of care, and (iii) inter-facility resource-sharing. We demonstrate how inclusion of the above impact cost-savings. Dmitry Krass, University of Toronto, Rotman School of Management, 105 St George Street, Toronto, ON, M5S 3E6, Canada, Oded Berman

89