Informs Annual Meeting 2017

WB08

INFORMS Houston – 2017

2 - Estimating Risk of Morbity and Mortality After Emergency Surgery with Machine Learning Jack W. Dunn, Massachusetts Institute of Technology, 24 Hardwick St, Unit 1, Cambridge, MA, 02141, United States,

3 - Activity Estimation using Ambient Sensor Network Data Yan Wang, University of South Florida, 4202 E Fowler Ave, ENB118, Tampa, FL, 33620, United States, yanwang2@mail.usf.edu Data from ambient sensor networks deployed in residential environments is used in daily activity recognition, behavior pattern discovery, abnormal behavior detection and physical and mental health assessment. In this work, we describe a spatio-temporal status tracking algorithm based Pyroelectric Infrared (PIR) motion sensor data to estimate resident’s location, time outside of home and sleep behaviors. The significance of this information is discussed as it relates to monitoring chronic health conditions and enabling older adults to age-in-place safely and independently. 4 - Occupant Activity Profiles from Smart Home Sensor Event Streams Garrick Aden-Buie, University of South Florida, 4202 E Fowler Ave, ENB118, Tamp, FL, 33620, United States, gadenbuie@mail.usf.edu Faced with a growing elderly population, learning and characterizing activity profiles of smart home occupants will support senior health care management for older adults living in homes augmented by ambient intelligence solutions that combine ubiquitous computing and artificial intelligence. In this work, we explore a bag of n-grams approach for creating occupant-specific activity profiles from event streams collected from passive, embedded sensors in real-world sensor networks in the homes of several older adults. Chair: Joseph Kapena Agor, North Carolina State University, 141 Hunt Club Lane, Apt C, Raleigh, NC, 27606, United States, jkagor@ncsu.edu 1 - Learning from Operational Experience and Benchmark Failure Experience in Telecom Sector in India Shaili Singh, Research Scholar, Indian Institute of Management, IIM Rohtak, Rohtak, 124001, India, singhshaili631@gmail.com, Mahua Guha Organizations learn from their own experiences and from experiences of peer organizations. Focus of this study is on learning from success and failure experiences. We identify success as enhanced operating experience and failure as benchmark failure experience which is measured based on the key performance indices set by Telecom Regulatory Authority of India and analyse the influence of aspiration performance discrepancy on organizational learning. Our findings suggest that in case of operating experience for performance above aspiration, firms fail to learn but for benchmark failure experience, firms learn from themselves. For performance below aspiration, we found no significant results. 2 - Learning Manufacturing Networks Dimitrios Katselis, University of Illinois at Urbana-Champaign, Urbana, IL, 61801, United States, katselis@illinois.edu, Seo Taek Kong, Carolyn Beck, R. Srikant Manufacturing networks are described by task precedence relationships, which indicate the flow of work in the network. The precedence relationships are designed by systems engineers, but they are often implemented quite differently than the intended design. The goal is to infer the precedence relationships in the implemented network using data about the observed start and end times of various tasks in the network. We will present an algorithm for performing the network inference and obtain bounds on the sample complexity by exploiting a relationship with the coupon collector problem. 3 - Arterial Probabilistic Traffic Modeling with Vehicle Probe Data using Machine Learning WB08 322B Learning Contributed Session

jackdunn@mit.edu, Lauren E.Berk, Dimitris Bertsimas, Haytham Kaafarani, Thomas Peponis, George Velmahos

Emergency surgery comprises an ever-increasing proportion of hospitalizations, and carries a higher risk of morbidity and mortality compared to non-emergent or elective procedures. We apply novel machine learning methods to develop models that predict the risk of mortality and associated co-morbidities following emergency surgery. Our approach delivers best-in-class accuracy whilst remaining intuitive and interpretable. We demonstrate the power of these models for patient counselling prior to surgery, as well as for benchmarking and improving the quality of emergency surgery care. 3 - Predicting Mortality to Prevent Deaths on the Liver Transplant Waitlist Yuchen Wang, Massachusetts Institute of Technology, Cambridge, MA, United States, yuchenw@mit.edu, Dimitris Bertsimas, Jerry L. Kung, Nikolaos Trichakis, Parsia Vagefi Current organ allocation process uses the Model for End-Stage Liver Disease (MELD) score to rank disease severity. We propose a new data-driven analytics- based method to predict the mortality rate within three months for patients on the liver transplantation waitlist. Our model not only outperforms the MELD score in accurately predicting mortality, but also addresses several of its shortcomings with respect to its inability to generalize to cancer patients. 4 - Identifying Exceptional Responders in Randomized Trials via Mixed-integer Optimization Alexander Weinstein, Massachusetts Institute of Technology, Cambridge, MA, United States, amw22@mit.edu, Dimitris Bertsimas, Nikita Korolko In randomized clinical trials, there may be a benefit to identifying subgroups of the study population for which a treatment was exceptionally effective or ineffective. In this chapter, we present an efficient mixed-integer optimization formulation that can directly find an interpretable subset with maximum (or minimum) average treatment effect. Using both simulated and real data from randomized trials, we demonstrate the effectiveness and stability of the optimization approach in identifying subsets with exceptional response and verifying their statistical significance. 322A Remote Patient Monitoring and mHealth Applications Sponsored: Health Applications Sponsored Session Chair: Julie Lynn Hammett, Texas A&M University, College Station, TX, 77840, United States, jhammett@tamu.edu 1 - Quantifiable Fatigue Risk Assessment through Activity Recognition Karla Ivette Gonzalez Coronado, Texas A&M.University, 23908 Killerbee Lane, New Caney, TX, 77357, United States, kalleyranch08@tamu.edu The development of a modern fatigue risk assessment is necessary. An algorithm to classify activities and a person’s heart rate variability through the day to track fatigue throughout the day is a solution for this. To develop this, 30 participants took part in a 90-minute study to identify and objectively define attributes of accelerometer readings, heart rate, breathing rate, and skin temperature and their role in understanding common daily activities. The activities included sitting, walking, picking up boxes, going up/down stairs, pulling ropes, and working on a computer. An application in mHealth is possible through the integration of this algorithm into smartwatches and smartphones. 2 - Remote Patient Monitoring System Model Julie Lynn Hammett, Texas A&M.University, 301 Holleman Dr E, Apt 728, College Station, TX, 77840, United States, jhammett@tamu.edu Healthcare providers are facing an increasing number of patients requiring long- term care, introducing new challenges to providing fast and affordable care. We present ongoing research to create a framework for the design, development, and implementation of remote patient monitoring (RPM) for chronic care. Using System Dynamics modeling, we demonstrate the potential impact of a remote health system on resource utilization in the healthcare continuum. The model shows the change in burden on existing resources when the flow of patients through a primary care facility changes. WB07

Bahar Zarin, PhD Candidate, University of Maryland, College Park, College Park, MD, 20742, United States, bzarin@umd.edu, Ali Haghani

This study proposes a probabilistic modeling framework for the prediction of arterial travel time distribution using Probe Data. We modeled spatiotemporal correlations of the network, including the intersections, using machine learning. We also investigated the effects of other factors such as driver behavior on the travel time. The data used in this study are INRIX vehicle trajectory and Open Street Map.

475