2016 INFORMS Annual Meeting Program

MB23

INFORMS Nashville – 2016

MB21 107A-MCC Applications of Stochastic Models in Medical Decision Making Problems Sponsored: Health Applications Sponsored Session Chair: Mohammad Reza Skandari, University of British Columbia, #420 2053 Main Mall, Vancouver, BC, V6T-1Z2, Canada, reza.skandari@sauder.ubc.ca 1 - Patient-centered HIV Viral Load Monitoring Strategies In Resource-limited Settings Viral load (VL) testing is the most critical monitoring tool for assessing the effectiveness of treatment in HIV patients. The optimal frequency of VL monitoring remains unknown, despite it being the costliest routine monitoring tool for HIV in Sub-Saharan Africa. We formulate a model parameterized using person-level longitudinal data to simulate adherence behavior and disease dynamics over time, and to develop monitoring schedules that adapt to patient characteristics. We then evaluate the total costs and quality-adjusted life years achieved by monitoring VL at fixed intervals (status quo), as well as at variable intervals based on an individualized risk assessment of virologic failure. 2 - Timing The Use Of Breast Cancer Risk Information In Biopsy Decision Making Mehmet Ayvaci, University of Texas at Dallas, Jindal School of Management, Dallas, TX, United States, ayvaci@stanford.edu, Mehmet Eren Ahsen, Srinivasan Raghunathan, Zahra Gharibi Available clinical evidence is inconclusive on whether radiologists should use the patient risk profile information when interpreting mammograms. On the one hand, risk profile information is informative and can improve radiologists’ performance, but on the other hand, it may impair their judgment by introducing biases in mammography interpretation. Therefore, it is important to assess whether and when profile information use translates into improved outcomes. We model the use of profile information in mammography using a decision theoretic approach and explore the value of profile information. 3 - Developing Near-optimal Biomarker-based Prostate Cancer Screening Strategies Christine Barnett, University of Michigan, Ann Arbor, MI, United States, clbarnet@umich.edu Brian Denton Recent advances in the development of new biomarker tests, which physicians use for the early detection of cancer, have the potential to improve patient survival by catching cancer at an early stage. We describe a partially observable Markov decision process (POMDP) to compute near-optimal prostate cancer screening strategies. We present results based on Monte Carlo simulation to compare the policies developed using our approximated POMDP methods with those recommended in the medical literature. 4 - Optimizing Breast Cancer Diagnostic Decisions While Minimizing Overdiagnosis Sait Tunc, University of Wisconsin-Madison, Madison, WI, United States, stunc@wisc.edu, Oguzhan Alagoz, Elizabeth S Burnside Although the early diagnosis of breast cancer saves millions of lives every year, overdiagnosis of breast cancer may cause harm without benefit. We propose a large-scale MDP that uses multi-dimensional cancer risk vectors to incorporate cytologic grade to the breast cancer diagnostic decision problem and concomitantly reduce the overdiagnosis. We present efficient algorithms to find the exact solution to the given large-scale MDP, and introduce upper bounds to further improve the computational performance. MB22 107B-MCC Policy Evaluation from Operations to Public Health Invited: ORinformed Healthcare Policies Invited Session Chair: Diwakar Gupta, University of Minnesota and National Science Foundation, Minneapolis, MN, United States, guptad@umn.edu 1 - Facilitating Early Diagnosis Of Tuberculosis In India Sarang Deo, Indian School of Business, sarang_deo@isb.edu High incidence of TB in India is driven by long diagnostic delay resulting from poor practices of unorganized private providers, who are often patients’ first point of contact. We develop an operational model of patients’ diagnostic pathways and calibrate it using data collected from household surveys. We use it to estimate the impact of new technology and improved provider behavior on reduction of Diana Maria Negoescu, University of Minnesota, negoescu@umn.edu, Heiner Bucher, Eran Bendavid

diagnostic delay. We also develop a stylized economic model of private providers and estimate the monetary incentive required to achieve reduction in diagnostic delay. These models have informed the design of a large pilot program funded by the Gates Foundation in two Indian cities of Mumbai and Patna. 2 - Casualty Distribution To Hospitals In The Aftermath Of Mass-casualty Events Nilay T Argon, University of North Carolina, Chapel Hill, NC, 27514, United States, nilay@unc.edu, Alex Mills, Serhan Ziya Following a disaster, emergency responders must transport a large number of casualties to hospitals by limited transportation resources. Based on a Markov decision process formulation, we develop heuristic policies that use limited information on travel times and congestion levels to determine how to allocate ambulances to casualty locations and which hospitals should be the destination for those ambulances. By means of a realistic simulation study, we show that the proposed heuristics provide substantial improvement in the expected number of survivors, even when only limited information about the system state is available. 3 - Hospital-physician Gainsharing Contract Design Diwakar Gupta, University of Minnesota, Minneapolis, MN, United States, guptad@umn.edu, Mili Mehrotra, Xiaoxu Tang Participation in the bundled payments for care improvement (BPCI) initiative has provided hospitals the ability to gainshare with physicians. We formulate a model to study the contracts that hospitals could offer physicians based on their historical as well as ongoing performance improvement. Physicians have private information about their costs of achieving different improvement targets. Physicians may choose to either enter the gainsharing agreements with the hospital or continue to operate under the fee-for-service schedule. We characterize the optimal contracts and analyze the distribution of the gains within a game-theoretic setting. Chair: Bruce L Golden, University of Maryland-College Park, 1, Simpsonville, MD, 2, United States, bgolden@rhsmith.umd.edu Co-Chair: Sean Barnes, Univ of Maryland-College Park, 4352 Van Munching Hall, University of Maryland, College Park, MD, 20742, United States, sbarnes@rhsmith.umd.edu 1 - Understanding Emergency Department Jumper Behavior: Actionable Insights From Claims Data Using Machine Learning Xia (Summer) Hu, University of Maryland - College Park, College Park, MD, 20740, United States, xhu64@umd.edu Sean Barnes, Margret Bjarnadottir, Bruce L Golden Emergency Department (ED) “jumpers” refers to patients whose ED consumption levels have changed drastically over consecutive periods (e.g. frequent to non- frequent, or vice versa). Based on yearly insurance claim records, we leverage various learning algorithms to predict ED jumpers, whose behaviour are usually difficult to capture using traditional methods. Further, we analyze the characteristics of jumpers via clustering based on Bayesian Information Criteria. Based on this analysis, we provide actionable insights about preventable ED usage and risk management. 2 - Impact Of State And Federal Policy Changes By Socioeconomic Status On Emergency Medicine Practice In Maryland David Anderson, CUNY Baruch, davidryberganderson@gmail.com, Edward Andrew Wasil, Bruce L Golden, Laura Pimentel, Jon Mark Hirshon, Fermin Barrueto We study the effect of the implementation of the Affordable Care Act (ACA) and a Global Budgeting Revenue (GBR) structure for hospital reimbursement on the operations of Maryland emergency departments. Using a 24-month longitudinal dataset of monthly ED performance, we find that ACA/GBR implementation leads to a decrease in admission rate, increased revenue capture by hospitals, a decrease in the percent of uninsured patients, and a small increase in volume. Further, we find that all of the changes are more pronounced at hospitals with patient populations coming from lower socioeconomic status zip codes. MB23 108-MCC Healthcare Analytics: Collaborations with Practitioners Sponsored: Health Applications Sponsored Session

155