![Show Menu](styles/mobile-menu.png)
![Page Background](./../common/page-substrates/page0398.png)
INFORMS Nashville – 2016
396
WB02
101B-MCC
Data Mining in Healthcare 2
Sponsored: Data Mining
Sponsored Session
Chair: zihao Jiao, MD, United States,
zihaobit@163.com1 - Doctors Performance In Emergency Rooms
Amir Mousavi, George Washington University, 2700 Wisconsin
Ave, Unit 305, Washington, DC, 20007, United States,
ahmn00@gmail.comEmergency Rooms are known as a vital ward in a hospital. Improving the
efficiency in the ER has been a challenging question for researchers. Throughout
the academic literature, people have defined different efficiency indexes in order
to tackle this problem. This research aims to apply Data Envelopment Analysis
technique in order to identify the performance (i.e. productivity) variation among
doctors and use this result as a component for the optimization model in order to
improve the efficiency of the system. The final goal will be to show how patient
waiting time is affected by doctors’ productivity and how the proposed scheduling
model using this information can reduce patient wait times in the ER.
2 - A Deep Feature Selection Approach For Personalized Medicine
Milad Zafar Nezhad, PhD Student, Wayne State University,
Wayne State Univesity, Detroit, MI, 48202, United States,
fq3963@wayne.edu,Kai Yang
Personalized Medicine has been defined in different ways in the literature. A good
interpretation for Personalized Medicine is “the use of combined knowledge
(genetic or otherwise) about a person to predict disease susceptibility, disease
prognosis, or treatment response and thereby improve that person’s health”. In
this research, we propose a new deep feature selection method based on deep
learning. Our method used stacked auto-encoders for feature representation in
higher level abstraction. We applied our approach to a specific precision medicine
problem. The results show that our feature learning and representation approach
leads to better results in comparison with otherwise.
WB03
101C-MCC
Advances in Emergency Department Operations
Management/Research
Sponsored: Manufacturing & Service Oper Mgmt, Healthcare
Operations
Sponsored Session
Chair: Soroush Saghafian, Harvard Univeristy, 79 John F. Kennedy
Street, Mailbox 37, Cambridge, MA, 02138, United States,
soroush_saghafian@hks.harvard.edu1 - Managing Emergency Operations
Eva Lee, Georgia Tech,
eva.lee@isye.gatech.edu, William Wang
This is joint with Grady Health System, Children’s Healthcare of Atlanta and
Emory University School of Medicine. Most scheduling is done based on
availability of physician’s preference time. The patients are then offered the best
possible time that may fit his/her doctor’s schedule. This study will identify the
needs of patients and develop a predictive model to estimate the individual needs
(and thus LOS for the appointment). This information is then incorporated within
a scheduling optimization framework for dynamic optimization. This allows for
optimizing the scheduled service as well as unexpected emergency service.
2 - The Impact Of Health Information Exchanges On Emergency
Department Length Of Stay
Jan Vlachy, Georgia Institute of Technology, Atlanta, GA, United
States,
vlachy@gatech.edu,Turgay Ayer, Mehmet U.S. Ayvaci,
Zeynal Karaca
Health information exchanges (HIEs) are expected to improve information
coordination in emergency departments (EDs), but their impact on ED operations
remains poorly understood. We study the effect of HIE on ED length of stay (LOS)
based on about 5.8 million ED visits in Massachusetts. We find that HIE a)
reduces ED LOS overall by 11.1%, b) is even more effective in teaching hospitals,
c) is less effective in crowded EDs, and d) its effectiveness depends on severity and
complexity of the patients. Our findings have implications for the nationwide HIE
adoption.
3 - Assignment Policy To Improve Emergency Department Boarding
Time: A Safety And Quality Of Care Perspective
Derya Kilinc, Arizona State University,
dkilinc@asu.edu,Soroush Saghafian, Stephen J Traub
One important reason for ED crowding problem is prolonged boarding time of
admit patients. We study effective ways of reducing ED boarding times by
focusing on the trade off between keeping patients in ED and assigning them to a
secondary inpatient unit. We model the patient flow problem as a parallel
queueing system and show that the optimal policy is a state-dependent threshold
policy. Since the optimal policy is hard to implement, we use a simple and
effective policy which we term penalty adjusted Largest Expected Workload Cost.
Using simulation model, we show that implementing the proposed policy can
improve patient safety by reducing the boarding times while controlling the
overflow of patients.
4 - Robust Data-driven Emergency Department Management Via
Percentile Optimization: Multi-class Queueing Systems With
Model Ambiguity
Austin Bren, Arizona State University, Phoenix, AZ, United States,
asbren@asu.edu,Soroush Saghafian
To help hospital Emergency Departments address overcrowding issues and
increase patient safety, we implement a robust multi-class queueing model to
overcome inherent ambiguities arising in parameter specification. Our technique,
based on percentile optimization, is uniquely suited to incorporate both learning
and the degree of optimism expressed by the manager. We demonstrate the
benefits of using our framework for improving current patient flow policies using
real-world data collected from a leading U.S. hospital and utilizing highly
effective, easy-to-implement management strategies.
WB04
101D-MCC
Optimization Methods in Smart Grid
Sponsored: Energy, Natural Res & the Environment,
Energy I Electricity
Sponsored Session
Chair: Andrew Liu, Purdue University, 315 N. Grant Street,
West Lafayette, IN, 47907, United States,
andrewliu@purdue.edu1 - Online Opf With Quasi-Newton Algorithm
Yujie Tang, California Institute of Technology, Pasadena, CA,
United States,
ytang2@caltech.edu, Krishnamurthy Dvijotham,
Steven Low
Optimal power flow is a central problem in the operation of power systems. So far
the majority of the literature deals with offline algorithms for traditional power
system applications, but the proliferation of distributed energy resources and
smart appliances in power networks motivates real-time, decentralized and
scalable algorithms. In this talk we will introduce an online OPF algorithm based
on quasi-Newton methods that is real-time and can track the optimal operation
when the state of the network is changing.
2 - Power System State Estimation In The Presence Of Bad Data
Ramtin Madani, University of California, Berkeley, Berkeley, CA,
United States,
ramtin.madani@berkeley.edu, Javad Lavaei,
Ross Baldick
This talk introduces a method for finding the unknown operating point of a
power network based on a given set of potentially corrupted measurements
including nodal active powers, nodal reactive powers, nodal voltage magnitudes
and line flows. We propose a conic optimization problem in order to handle
nonconvexity and deal with bad data simultaneously. The proposed convex
program is guaranteed to recover the exact vector of complex voltages as long as
the number of bad measurements is small. Numerical experiments on a large-
scale European system are performed to demonstrate the efficacy of the proposed
framework.
3 - Parallelized Interior Point Method For Security Constrained
Optimal Power Flow Problem
Na Li, Assistant Professor, Harvard University, 33 Oxford St,, MD
147, Cambridge, MA, 02139, United States,
nali@seas.harvard.edu,
Ariana Minot, Yannick Meier
Solving the security constrained optimal power flow problem (SCOPF) is
challenging due to the large size of the power system and large number of
contingencies. However, in SCOPF, different contingencies are only coupled via
the power injection control variables, yielding a sparse system. We design a
domain decomposition technique based on this sparsity to parallelize the problem
across different contingencies. For each subproblem associated with a
contingency, we exploit the network structure through graph coloring techniques
to enhance parallelism. In summary, we design an effective method to utilize two
layers of parallelism: 1) across contingencies and 2) across buses in the network.
WB02