Informs Annual Meeting 2017

TE07

INFORMS Houston – 2017

TE07

5 - Data-driven Management of Post-transplant Medications: An APOMDP Approach Alireza Boloori, Arizona State University, 505 W. Baseline Rd, Apt 2158, Tempe, AZ, 85283, United States, aboloori@asu.edu, Soroush Saghafian, Harini A. Chakkera, Curtiss B. Cook Patients after organ transplantations receive high amounts of immunosuppressive drugs (e.g., tacrolimus) to reduce the risk of organ rejection. However, this practice has been shown to increase the risk of New-Onset Diabetes After Transplantation (NODAT). We propose an ambiguous POMDP framework to generate effective medication management strategies for tacrolimus and insulin. Our approach increases the patient’s quality of life while reducing the effect of transition probability estimation errors. We also provide several managerial and medical implications for policy makers and physicians. 320C Kidney Exchange Sponsored: Health Applications Sponsored Session Chair: Burhaneddin Sandikci, University of Chicago, Chicago, IL, 60637, United States, burhan@chicagobooth.edu 1 - Incentives in Kidney Exchange Itai Ashlagi, Stanford University, Department of Management Sci. and Engr., Stanford, CA, 94305, United States, iashlagi@stanford.edu We provide empirical evidence for the inefficiency resulting in the current kidney exchange marketplace in the US and propose a novel mechanism that can implement the first best in equilibrium. 2 - Operation Frames and Clubs in Kidney Exchange Tuomas W. Sandholm, Carnegie Mellon University, Gates Center for Computer Science, Pittsburgh, PA, 15213, United States, sandholm@cs.cmu.edu, Gabriele Farina, John Dickerson Kidney exchanges use 2-cycles, 3-cycles, and chains. We propose generalizations. We allow more than one donation in exchange for an organ. We also support a donor willing to donate if any of a number of patients receive organs. We generalize these notions to organ clubs that are willing to donate outside if the club receives organs from outside according to specifications. We introduce operation frames for sequencing operations across batches. We present IP formulations for the clearing problems for these exchanges. Experimentally, in the single-donation setting, operation frames improve planning by 34%-51%. Allowing up to two donors to donate in exchange for an organ increases welfare further. 3 - A Non-asymptotic Approach to Analyzing Kidney Exchange Graphs Christopher Ryan, University of Chicago, 5807 S Woodlawn Ave, Chicago, IL, 60637, United States, chris.ryan@chicagobooth.edu, Yichuan Ding, Dongdong Ge, Simai He We propose a novel methodology to study kidney exchange. Taking the random graph model of kidney exchange introduced in Ashlagi, Gamarnik, Rees and Roth (2012), we propose a non-asymptotic approach based on a two-phase procedure where random walks are used to allocate chains, followed by allocation in cycles. Random walks preserve the probabilistic structure of residual graphs, greatly facilitating analysis. Our results complement previous findings in large (limit) graphs and also shows how prioritizing chains over easy-to-transplant cycles improves performance and we provide analytical bounds on the associated benefits. 4 - Matching and Thickness in Dynamic Markets Maximilien Burq, Massachusetts Institute of Technology, 77 Massachusetts Avenue, Cambridge, MA, 02139, United States, mburq@mit.edu, Vahideh Manshadi, Itai Ashlagi, Patrick Jaillet We study dynamic matching in an infinite-horizon stochastic networked market, in which some agents are a priori more difficult to match than others. Agents can match either bilaterally or indirectly through chains. We study the effect of varying the composition of the market. Counter-intuitively, we show that increasing the market size does not always result in less waiting for hard-to- match agents. We also provide precise bounds that allow us to theoretically and numerically compare different matching policies and priority schemes. TE06

322A Personalized Medical Decision-Making Sponsored: Health Applications Sponsored Session Chair: Gian Gabriel P. Garcia, University of Michigan, Ann Arbor, MI, 48105, United States, garciagg@umich.edu 1 - Mobile Health for Chronic Disease Management: A Change Point Detection Approach Anthony Bonifonte, Georgia Institute of Technology, 5940 Water Oaks Dr, Austell, GA, 30106, United States, ABonifon@gatech.edu Wearables and other mobile health devices have the potential to revolutionize the management of chronic disease. We present a Bayesian changepoint detection algorithm that indicates when a health characteristic has exceeded a threshold value. We demonstrate the performance of this algorithm on realistically simulated blood pressure data, and show improvement over baseline policies and simpler, more general changepoint procedures. This work demonstrates the enormous improvement in chronic disease screening and monitoring that can be achieved using data from mobile health technologies. 2 - Designing Personalized Weight Loss Interventions with Behavioral Analytics Yonatan Mintz, UC Berkeley, 1822 Francisco St., Apt 10, Berkeley, CA, 94703, United States, ymintz@berkeley.edu, Anil Aswani, Philip Kaminsky, Elena Flowers, Yoshimi Fukuoka In this talk, we develop a behavioral analytics framework which uses patient data to effectively design personalized weight loss interventions. Our framework utilizes a utility maximization model for patient behavior which combined with integer programming and Bayesian prediction allows us to create several personalized interventions, as well as aggregate these interventions into a cohort weight loss program. We then present simulation results which show that our method maintains efficacy while potentially reducing the associated person hours and cost of the intervention. 3 - A Stochastic Programming Approach to Determine Decision Boundaries for Medical Diagnosis Gian Gabriel P. Garcia, University of Michigan, 1823 Pointe Crossing Street, Apt 204, Ann Arbor, MI, 48105, United States, We use stochastic programming to determine decision thresholds for medical diagnosis risk models under uncertain data samples from a fixed population. These thresholds maximize sensitivity and specificity under constrained false positive and false negative rates. We show that sample average approximation solutions for these models are consistent estimators. We apply our method to concussion assessment and show that our thresholds dominate single decision thresholds under multi-criteria evaluation. 4 - Value and Design of Clinical Decision Support Systems in the Presence of Bias in Imaging-based Decisions Mehmet U.S. Ayvaci, University of Texas at Dallas, 800 W. Campbell Rd, Sm33, Richardson, TX, 75080-3021, United States, mehmet.ayvaci@utdallas.edu, Mehmet Eren Ahsen, Srinivasan Raghunathan A patient’s clinical risk information could bias a radiologist’s assessment of a mammogram. We examine the design and value of a clinical decision support system (CDSS) in the breast cancer diagnosis context. The CDSS assists a caring physician in her follow-up recommendation to a patient based on the radiologist’s assessment. We show that the CDSS can eliminate the adverse impact of bias if there is no variability associated with the bias-induced error in radiologist’s assessment. Using point estimates based on mammography practice and the medical literature, we show that a bias-aware CDSS can significantly improve the expected patient life years or the accuracy of decisions based on mammography. garciagg@umich.edu, Mariel Sofia Lavieri, Ruiwei Jiang, Steven P. Broglio, Michael McCrea, Thomas W. McAllister

411