2015 Informs Annual Meeting

MB31

INFORMS Philadelphia – 2015

MB31 31-Room 408, Marriott Data Mining and Predictive Analytics in Health Care

The rapid growth of functional genomic data makes it possible to build models for predicting one high-throughput genomic data type from another data type. This can be formulated as a challenging big data regression problem which involves fitting millions of high-dimensional regressions simultaneously. To cope with the high dimensionality and heavy computation, we developed BIRD algorithm that leverages the correlation structure in the data to make computation fast and predictions accurate. 3 - Semi-automated Human Genome Annotation using Chromatin Data Michael Hoffman, Scientist, Assistant Professor, Princess Margaret Cancer Centre/University of Toronto, Toronto Medical Discovery Tower 11-311, 101 College St, Toronto, ON, M5G 1L7, Canada, michael.hoffman@utoronto.ca Segway is an integrative method to identify patterns from multiple functional genomics experiments. It discovers joint patterns in multiple genomic datasets using a dynamic Bayesian network model, simultaneously segmenting the genome and identifying clusters of similar segments. We apply Segway to ENCODE ChIP-seq and DNase-seq data and identify patterns associated with transcription start sites, gene ends, enhancers, and repressed regions. 4 - Identifying Genetic Risk Factors for Complex Traits using Functional and Association Data Jo Knight, Centre for Addiction and Mental Health, 250 College Street, Toronto, Canada, jo.knight@camh.ca, Mike Barnes, Mike Weale, Sarah Gagliano Our aim is to identify the genetic risk variants that contribute to disease. Genome wide association studies have identified some but many remain unknown. We seek to combine the association data with functional characteristics of the genome. Machine learning is used to derive a score to indicate whether a genetic variant is likely to be causal based on large amounts of functional data. We combine the functional score and the association score together in a Bayesian framework. Health Care Operations Sponsor: Health Applications Sponsored Session Chair: Tolga Tezcan, Associate Professor, London Business School, Regent’s Park, London, UK, NW14SA, United Kingdom, ttezcan@london.edu Co-Chair: Nicos Savva, London Business School, Park Road, London, NW14SA, United Kingdom, nsavva@london.edu 1 - Why is Big Data Underutilized? The advent of big data has brought the opportunity to track customer needs and offer service proactively. Motivated by a healthcare application we develop a queueing and game theoretic model to show that enrolling in such a data- tracking scheme generates positive externalities in the form of reduced waiting times. Nevertheless, we show that self-interested consumers will under-utilize this opportunity, leading to a welfare loss. 2 - A Two-sided Mechanism to Coordinate the Influenza Vaccine Supply Chain Kenan Arifoglu, University College London, Gower Street, London, WC1E 6BT, United Kingdom, k.arifoglu@ucl.ac.uk Rational consumer behavior and uncertain yield lead to frequent supply/demand mismatches in the influenza (flu) vaccine supply chain. To eliminate the inefficiency in the flu vaccine supply chain, we propose a two-sided mechanism which implements tax/subsidy payments on the demand side and a transfer payment on the supply side and aligns consumers’ and vaccine manufacturer’s incentives with the social optimum. The two-sided mechanism improves social welfare significantly. 3 - Analysis of Triage Systems in Emergency Departments Ozlem Yildiz, Simon Business School , University of Rochester, CSH 4-333. Simon Business School, Rochester, NY, 14627, United States of America, oyildiz@london.edu, Tolga Tezcan, Michael Kamali We study triage method decisions in emergency departments and provide a policy for determining when to apply provider triage (PT) based on operational and financial considerations using a queueing framework. We obtain closed-form expressions for the range of arrival rates in which PT economically outperforms traditional nurse triage using a steady-state many-server fluid approximation. We show via simulation experiments that the proposed policy performs within 0.82% of the best solution. Kraig Delana, London Business School, Regent’s Park, London, NW1 4SA, United Kingdom, kdelana@london.edu, Nicos Savva, Tolga Tezcan MB33 33-Room 410, Marriott Joint Session HAS/MSOM-Healthcare:

Sponsor: Data Mining Sponsored Session

Chair: Lior Turgeman, Data Mining and Operations Research, Joseph M. Katz Graduate School of Business, Roberto Clemente Dr, Pittsburgh, PA 1526, Pittsburgh, United States of America, tur.lior@gmail.com 1 - Predicting Hospital Readmission using Patient Encounter Data Atish Sinha, Professor, University of Wisconsin-Milwaukee, Lubar School of Business, Milwaukee, WI, 53201-0742, sinha@uwm.edu, Amit Bhatnagar, Arun Sen Under the Affordable Care Act, readmission rate has become a critical issue for hospitals. We analyze patient encounter data, obtained from an HIE, during a two-year period for a chain of hospitals. Our model incorporates two sets of factors, consumer demographics and encounter data, to predict readmission likelihoods and durations. 2 - A Mixed-ensemble Predictive Model for Hospital Readmission Lior Turgeman, Data Mining and Operations Research, Joseph M. Katz Graduate School of Business, Roberto Clemente Dr, Pittsburgh, PA 1526, United States of America, tur.lior@gmail.com, Jerrold H. May, Johnson Moore, Youxu Cai Tjader We present a novel approach for predictive modeling, using a mixed-ensemble classifier. The approach integrates a C5.0 tree as the base ensemble classifier, and a support vector machine (SVM) as a secondary classifier. By implementing our method for predicting hospital readmission of CHF patients, we were able to overcome some of the limitations of both C5.0 and SVM, as well as to increase the classification accuracy for the minority class, particularly when strong predictors are not available. 3 - A Decision Analytic Approach to Modeling Heart Transplant Survival Asil Oztekin, Assistant Professor Of Operations & Information Systems, Participating Faculty Of Biomedical Engineering & Biotechnology Program, University of Massachusetts Lowell, One University Ave. Southwick 201D, Lowell, MA, 01850, United States of America, Asil_Oztekin@uml.edu, Ali Dag, Fadel Megahed Due to the scarcity of donor hearts for transplantation, an accurate prediction of transplantation success plays an important role in the matching procedure between donors and recipients. A decision analytic framework based on Bayesian Belief Network is deployed here to address this issue. The results indicate that this decision analytic methodology yields superior results than the ones in the transplantation literature.It is a generic model which can be implemented in other transplant cases. MB32 32-Room 409, Marriott Big Data Analytics in Genomics Cluster: Big Data Analytics in Computational Biology/Medicine Invited Session Chair: Michael Hoffman, Scientist, Assistant Professor, Princess Margaret Cancer Centre/University of Toronto, Toronto Medical Discovery Tower 11-311, 101 College St, Toronto, ON, M5G 1L7, Canada, michael.hoffman@utoronto.ca 1 - A Spectral Approach for the Integration of Functional Genomics Annotations for Genetic Variants Iuliana Ionita-laza, Assistant Professor, Columbia University, 722 West 168 St, New York, NY, 10032, United States of America, ii2135@columbia.edu Over the past few years, substantial effort has been put into the functional annotation of variation in human genome sequence. Such annotations can play a critical role in identifying putatively causal variants among the abundant natural variation that occurs at a locus of interest. The main challenges in using these various annotations include their large numbers, and their diversity. I will discuss an unsupervised approach to derive an integrative score of these diverse annotations. 2 - Big Data Regression and Prediction in Functional Genomics Weiqiang Zhou, Johns Hopkins University Bloomberg School of Public Health, 615 N Wolfe Street, Rm E3638, Baltimore, MD, 21205, United States of America, kenandzhou@hotmail.com, Hongkai Ji

182