INFORMS 2021 Program Book

INFORMS Anaheim 2021

MD17

MD17 CC Room 202A In Person: DAS Flash Talks I Flash Session Chair: Kyle J. Hunt, University at Buffalo, Buffalo, NY, 14086, United States 1 - Is it Better to Elicit Quantile or Probability Judgments to Estimate a Continuous Distribution? Elicit Both to Create an Inner Crowd Asa Palley, Indiana University, Bloomington, IN, 47405-1703, United States, Saurabh Bansal Two elicitation methods can be used to gather a probability distribution from an expert: (i) quantile judgments for a set of fixed probability values, or (ii) probability judgments based on a set of fixed variable values, but a consensus on which format yields more accurate distribution estimates has not been reached. We conduct 8 experiments with 1,456 participants and 30,870 distribution judgments to answer this question. We find that averaging distributions constructed from quantiles and visual probability elicitation tools delivers superior accuracy and calibration of confidence intervals. We recommend eliciting judgments via both formats and combining the resulting distributions. 2 - On The Adoption of New Technology to Enhance Counterterrorism Measures: Attacker-defender Games Kyle J. Hunt, University at Buffalo, Buffalo, NY, 14086, United States, Jun Zhuang To address the adaptive and dynamic threats that terrorists bring to our global society, it is imperative that counterterrorism agencies continue to improve their capabilities. To this end, one important challenge that has received scarce attention in the research community is the adoption of new counterterrorism technologies. Given that the adoption of new technology is common in practice, it is essential to study the strategic advantages of doing so. To help fill the gap in the literature, in this work, we develop attacker-defender models in which a defender seeks to adopt new technology, and an adversary seeks to attack a target. MD18 CC Room 202B In Person: Discrete Optimization General Session Chair: Giacomo Nannicini, IBM T.J. Watson, Yorktown Heights, NY, 10598, United States 1 - Political Districting to Minimize Cut Edges Austin Buchanan, Oklahoma State University, Stillwater, OK, 74078-5017, United States, Hamidreza Validi When constructing political districting plans, prominent criteria include population balance, contiguity, and compactness. The compactness of a districting plan, which is often judged by the “eyeball test,” has been quantified in many ways. This paper considers the number of cut edges, which has recently gained traction in the redistricting literature as a measure of compactness because it is simple and reasonably agrees with the eyeball test. We study the stylized problem of minimizing the number of cut edges, subject to constraints on population balance and contiguity. With the MIP techniques proposed in this paper, all county-level instances in the USA (and some tract-level instances) can be solved to optimality. Our techniques extend to minimize weighted cut edges (e.g., to minimize district perimeter length) or to impose compactness constraints. 2 - Dual Bounds From Decision Diagram-based Route Relaxations: An Application to Truck-Drone Routing Willem-Jan Van Hoeve, Carnegie Mellon University, Pittsburgh, PA, United States, Ziye Tang We propose iterative algorithms to compute dual bounds motivated by connections between decision diagrams (DDs) and dynamic programming (DP) models used for pricing in branch-and-cut-and-price algorithms. We apply techniques from the DD literature to generate and strengthen novel route relaxations for obtaining dual bounds without column generation. Our approaches are general and can be applied to various vehicle routing problems where DP models are available. We apply our framework to the traveling salesman with drone problem, and show that our algorithms produce dual bounds competitive to those from the state-of-the-art, and can scale to larger problem instances.

MD19 CC Room 203A In Person: Data, Learning, and Decision-Making in Healthcare Management General Session Chair: Danqi Luo, Stanford Graduate School of Business, Stanford, CA, 94305-7216, United States 1 - Modeling HIV/AIDS for Urban Centers in California to Inform Disease Control Goals Sze-chuan Suen, University of Southern California, Los Angeles, CA, 90089-0193, United States, Anthony Nguyen In 2019, the federal government launched a new initiative called Ending the HIV Epidemic: A Plan for America (EHE) with the goals of reducing the number of new HIV infections by 75% by 2025 and 90% by 2030. However, it is unclear how this should be achieved, as public health departments can allocate resources over a variety of interventions: growing adoption of pre-exposure prophylaxis (PrEP), increasing new diagnoses, improving retention in antiretroviral therapy (ART), etc. To inform this effort, we partner with state and local public health decisionmakers in California to build a microsimulation model of HIV/AIDS for Los Angeles, San Diego, and San Francisco using city-specific data. We use the model to examine intervention portfolios and assess what level and combination of strategies might be needed to achieve EHE goals. 2 - Conformalized Survival Analysis Zhimei Ren, Stanford University, Stanford, CA, United States, Emmanuel Candès, Lihua Lei Existing survival analysis techniques heavily rely on strong modelling assumptions and are, therefore, prone to model misspecification errors. In this paper, we develop an inferential method based on ideas from conformal prediction, which can wrap around any survival prediction algorithm to produce calibrated, covariate-dependent lower predictive bounds on survival times. In the Type I right-censoring setting, when the censoring times are completely exogenous, the lower predictive bounds have guaranteed coverage in finite samples with only the i.i.d. data assumption. Under a more general conditionally independent censoring assumption, the bounds satisfy a doubly robust property. The validity and efficiency of our procedure are demonstrated on synthetic data and real COVID-19 data from the UK Biobank. MD22 CC Room 204B In Person: Data Driven Applications In Healthcare Operations General Session Chair: Sandeep Rath, University of North Carolina at Chapel Hill Kenan Flagler, Chapel Hill, NC, 27599, United States 1 - Inference of Arrival Intensity in a Hospital Network From Count Data Qianli Xu, Purdue University, West Lafayette, IN, United States, Harsha Honnappa, Pengyi Shi We consider the arrival prediction problem of a hospital network. We capture patient movements within the hospital network with a multi-station queueing network, where the arrival process is time-varying and follows a doubly stochastic Poisson process. We propose frameworks to infer the arrival intensity process with count data. The first one is based on variational inference and the second one expectation-maximization. In a numerical experiment, we find both our methods perform well for queueing networks with a CIR arrival intensity process. 2 - Data-Driven Surgical Tray Optimization Nishanth Mundru, UNC Kenan-Flagler, Chapel Hill, NC, United States, Vinayak V. Deshpande, Sandeep Rath Surgical procedures account for over 60% of the operating cost of a hospital. However, on average less than 20-30% of reusable instruments supplied to surgeries are used. Using actual surgical instrument usage at a large multi- specialty hospital, we formulate a data-driven mathematical optimization model for surgical tray configuration and assignment, to reduce costs of unused instruments. We develop a solution methodology that scales to thousands of surgeries, thousands of instruments, and hundreds of surgical trays. We validate our model with an expert-recommended solution for a subset of trays and find that our model-based solution leads to 20\% lower overage and 21\% lower underage.

78