Informs Annual Meeting 2017

SA15

INFORMS Houston – 2017

5 - A Study of Co-occurrence of Multiple Chronic Conditions using Online Healthcare Forums Juxihong Julaiti, PhD Student, Penn State University, 445 Waupelani Drive, Apt J1, State College, PA, 16801-4445, United States, jpj5196@psu.edu In recent years participation in online forums to discuss health related issues is increasing. We study the on-line forums to study the relationship between multiple chronic conditions and resources allocation. We use linear, logistic regression, random forest to derive the relationships. The results indicate that: 1) patients with cancer and depression are more likely to participate in online forums; 2) patients with costly chronic conditions are more likely to use online forums; 3) whether a patient has cardiovascular disease, arthritis are significant variable to predict if the patient will have asthma etc. 6 - Dispensing Medical Countermeasures in Public Health Emergencies via Home Health Agencies and Points of Distribution Ashlea Bennett Milburn, University of Arkansas, 4207 Bell Engineering Center, Fayetteville, AR, 72701, United States, ashlea.bennett@uark.edu, Anna Hudgeons, Charleen McNeill A major concern regarding emergency preparedness at the state government level involves the handling and dispensing of the Strategic National Stockpile of medicinal supplies. This research utilizes simulation modeling to determine if using home health agencies (HHAs) in addition to Points of Distribution (PODs) to dispense SNS supplies can provide aid in a more timely fashion than using PODs alone. The comparative effectiveness of the two dispensing methods is demonstrated via a case study of pandemic influenza in Northwest Arkansas. 332D Analytics and Incentives for Efficient Healthcare Delivery Sponsored: Manufacturing & Service Oper Mgmt, Healthcare Operations Sponsored Session Chair: Hamsa Sridhar Bastani, Stanford University, Stanford, CA, 94305, United States, hsridhar@stanford.edu 1 - Yardstick Competition for Hospital Queues Nicos Savva, London Business School, London, United Kingdom, nsavva@london.edu, Tolga Tezcan, Ozlem Yildiz In this paper, we first show that the hospital reimbursement system currently used in practice does not incentivize hospitals to reduce waiting times. We then propose a modification which can achieve socially optimum investment without placing an onerous informational burden on hospital payers. 2 - Misaligned Incentives in Kidney Exchange Itai Ashlagi, Stanford University, Department of Management Sci. and Engr., Stanford, CA, 94305, United States, iashlagi@stanford.edu Growth of live-donor kidney exchange has stagnated although an increasing number of transplant centers are facilitating exchanges through large national exchanges, in isolation or in small consortiums with other centers. Using a novel dataset we document significant fragmentation and heterogeneity in participation by centers, an adversely selected and hard to match national exchange pool, and inefficient transplants outside the largest exchange. Using a simple supply and demand model we estimate the inefficiency and in the current market and calculate a rewards scheme that would be implemented by the optimal mechanism. 3 - Exploiting the Natural Exploration in Contextual Bandits Hamsa Sridhar Bastani, IBM Thomas J. Watson Research Center, Yorktown Heights, NY, United States, hamsab@wharton.upenn.edu, Mohsen Bayati, Khashayar Khosravi The contextual bandit literature focuses on an exploration-exploitation tradeoff because exploration-free greedy policies may yield poor performance in general. However, greedy policies are desirable when experimentation is costly or unethical (e.g., clinical settings). We posit a sufficient set of additional assumptions under which a greedy policy is asymptotically optimal. Next, we present Greedy-First, a new algorithm that uses observed data to determine whether to follow a greedy policy or to explore. This algorithm is asymptotically optimal without our additional assumptions, and significantly reduces experimentation in simulations. SA14

4 - Dynamic Bandit Approach for Personalized Treatment Yonatan Mintz, UC Berkeley, San Francisco, CA, United States, ymintz@berkeley.edu, Anil Aswani, Philip Kaminsky, Elena Flowers, Yoshimi Fukuoka Personalized fitness tracking devices enable the implementation of data driven exercise programs intended to treat sedentary lifestyles. Despite the prevalence of these devices, many individuals have a hard time adhering to the recommended programs since they may be incompatible with an individual’s schedule or provide ineffective motivation. In this talk, we address this problem by leveraging the data and infrastructure of mobile fitness tracking devices to personalize exercise programs for participating individuals. We develop a multi-armed bandit approach to adaptively learn each participant’s exercise preferences to personalize their exercise programs and increase adherence. 332E Demand Learning Sponsored: Manufacturing & Service Oper Mgmt, Supply Chain Sponsored Session Chair: Yiangos Papanastasiou, University of California Berkeley, Berkeley, CA, 94720, United Kingdom, yiangos@haas.berkeley.edu 1 - Dynamic Selling Mechanisms for Product Differentiation and Learning N. Bora Keskin, Duke University, Fuqua School of Business, 100 Fuqua Drive, Durham, NC, 27708-0120, United States, bora.keskin@duke.edu, John R.Birge We consider a firm that designs a menu of vertically differentiated products for a population of customers with heterogeneous quality sensitivities. The firm faces an uncertainty about production costs. We characterize the structure of the firm’s optimal dynamic learning policy and construct simple and practically implementable policies that are near-optimal. 2 - Bayesian Dynamic Learning and Pricing with Strategic Customers Xi Chen, New York University, 44 W. 4th St, NYU. KMC Room 8-50, New York, NY, 10012, United States, xchen3@stern.nyu.edu, Zizhuo Wang In this talk, we study such a learning problem when the customer is aware of the seller’s policy, and thus may behave strategically when making a purchase decision. We propose a randomized Bayesian policy (RBP), which updates the posterior belief of the customer in each period with a certain probability. We show that the seller can learn the customer type exponentially fast with the RBP even if the customer is strategic, and the regret is bounded by a constant. We also propose policies that achieve asymptotically optimal regrets when only a finite number of price changes is allowed. 3 - Dynamic Pricing with Online Product Reviews Dongwook Shin, Hong Kong University of Science and Technology, Kowloon, Hong Kong, dwshin@ust.hk Dongwook Shin, Columbia University, New York, NY, 10027, United States, dwshin@ust.hk, Assaf Zeevi We investigate how the presence of product reviews affects a dynamic-pricing monopolist who strives to maximize its total expected revenue. To formulate the problem in tractable form, we first study a fluid model, which is a good approximation when the volume of sales is large. In the context of the fluid formulation, the optimal pricing policy can be characterized in a closed-form, which is leveraged to design pricing policies that perform well with respect to the underlying revenue maximization problem. Lastly, we discuss the impact of product reviews on learning when the monopolist operates without knowing the demand function a priori. 4 - A Tutorial on Thompson Sampling Daniel Russo, Northwestern University, Chicago, IL, United States, dan.joseph.russo@gmail.com Thompson sampling is an algorithm for online decision problems where actions are taken sequentially in a manner that must balance between exploiting what is known to maximize immediate performance and investing to accumulate new information that may improve future performance. The algorithm addresses a broad range of problems in a computationally efficient manner and is therefore enjoying wide use. I will present selected examples and insights from a recent tutorial on Thompson sampling. The talk will focus on applications in revenue management, and on practical challenges like nonstationarity, prior specification, and the need for approximate posterior inference. SA15

7