2015 Informs Annual Meeting

WA33

INFORMS Philadelphia – 2015

2 - Marketing-driven Content Management in Large Organizations Roman Vaculin, IBM, 1101 Kitchawan Rd, Yorktown Heights, NY, United States of America, vaculin@us.ibm.com, Yi-min Chee, Ashish Jagmohan, Elham Khabiri, Richard Segal, Noi Sukaviriya Content management in large organizations is a complex process involving range of approaches such as text analysis or classification with taxonomies. We focus on facilitating management of legacy and creation of new content to improve marketing content effectiveness. We discuss techniques for (1) content characterization based on semantic concept models, (2) mapping textual content to legacy taxonomies, and (3) taxonomy enhancement for data-driven taxonomy rationalization, extension and mapping. 3 - Product Recommendations with Discovered Hidden Topics in Online Reviews Julie Zhang, Assistant Professor, University of Massachusetts Lowell, One University Ave., Lowell, MA, 01854, United States of America, juheng_zhang@uml.edu We use LDA (latent Dirichlet Allocation) to discover hidden information embedded in online product reviews for consumers to make informed purchase decisions. The discovered topics are used to make product recommendations to offer consumers better utility. 4 - Towards Automatic Identification of B2B Marketing Prospects Elham Khabiri, IBM, 1101 Kitchawan RD, Yorktown Heights, NY, United States of America, ekhabiri@us.ibm.com, Roman Vaculin, Richard Hull, Matthew D. Riemer In B2B marketing, it is critical to identify marketing leads effectively. We investigate into what extent the B2B leads identification process can be automated using online and historical data, and if algorithmic leads mining can achieve better performance than leads identified by human experts. We present a solution that combines deep learning to identify look-a-like companies with respect to historical leads, and mining of online data for identification of signals indicating potential leads. Chair: Muge Capan, Value Institue, Christiana Care Health System, 4755 Ogletown-Stanton Road, John H. Ammon Medical Education Center, Newark, DE, 19718, United States of America, Muge.Capan@ChristianaCare.org 1 - Determining an Optimal Schedule for Pre-mixing Chemotherapy Drugs Donald Richardson, University of Michigan, Ann Arbor, MI, donalric@umich.edu, Amy Cohn In collaboration with the University of Michigan Comprehensive Cancer Center, we have developed a data-driven, optimization-based approach to improving the timeliness of drug preparation for chemotherapy infusion patients while reducing staff workload and improving resource utilization. We compare the results from both static and dynamic decision policies to determine the optimal schedule for pre-mixing chemotherapy drugs at the cancer center. 2 - Using Text Analytics to Identify Labor and Delivery Sentiment from the Internet and Social Media Karen Hicklin, PhD Student, North Carolina State University, 111 Lampe Drive, Campus Box 7906, Raleigh, NC, 27695, United States of America, kthickli@ncsu.edu, Julie Ivy, Fay Cobb Payton, Evan Myers, Meera Viswanathan, Michael Wallis, Vidyadhar Kulkarni Pregnant women often seek counsel online from doctors, midwives, experienced and other expectant mothers to understand if their experience lines up with that of others and to seek opinions, suggestions and recommendations. We use text analytics to identify key attributes and preferences for women from internet data to identify important attributes that influence patient perceptions in regards to the birth experience that can be used as input parameters to inform delivery mode decision models. 3 - Data-driven and Analytical Approaches to Falls Injury Prediction and Rescue Resources Allocation Tze Chiam, Associate Director, Research Informatics, Christiana Care Health Systems, 4755 Ogletown-Stanton Road, WA33 33-Room 410, Marriott Data-driven Healthcare Operations Sponsor: Health Applications Sponsored Session

identify patients at high risk for injury due to falls and the appropriate responses based on these criteria. A discrete-event simulation is used to evaluate rescue strategies that yield fastest response time while minimizing interruptions to the system. 4 - Nurse Scheduling Optimization in the Neonatal Intensive Care Unit Muge Capan, Value Institue, Christiana Care Health System, 4755 Ogletown-Stanton Road, John H. Ammon Medical Education Center, Newark, DE, 19718, United States of America, Muge.Capan@ChristianaCare.org, Eric V. Jackson, Robert Locke Nurse scheduling is the process of assigning nurses to work shifts. A suboptimal schedule can impact the staffing ratios, nurses’ well-being, job satisfaction, and quality of care. Nurse scheduling in the Neonatal Intensive Care Unit is particularly challenging due to the complexity of care environment and required nursing skillset. We present a multi-objective optimization approach to allocate nurses to shifts while considering institutional requirements, workload fairness and nurses’ health. WA34 34-Room 411, Marriott Real World Problems, Innovative Approaches and Implementations at Geisinger Health System Sponsor: Health Applications Sponsored Session Chair: Christopher Stromblad, Senior Modeler - Operations Research, Geisinger Health System, 100 N. Academy Ave, Danville, PA, 17822, United States of America, ctstromblad@geisinger.edu 1 - Implementing Predictive Models in Five Operating Suites Improved Surgical Case Duration Accuracy Ronald Dravenstott, Geisinger Center for Healthcare Systems Engineering, 100 N Academy Ave, Danville, PA, 17822, United States of America, rwdravenstott@geisinger.edu, Eric Reich, Cheng-bang Chen, Priyantha Devapriya The length of a surgical procedure is generally predicted using a moving average of previous procedures performed by a given surgeon, a method which is often inaccurate and complicates Operating Room scheduling. This research developed and implemented predictive models for high-volume surgeries to seamlessly improve surgical scheduling by leveraging patient and provider data. The models now in place overcame real-time data inconsistencies and have improved scheduling. 2 - Anticipating Provider Orders in Outpatient Clinics Yi-shan Sung, Research Assistant, Penn State, 425 Waupelani Dr., Apt. 509, State College, PA, 16801, United States of America, yqs5097@psu.edu, Ronald Dravenstott, Priyantha Devapriya, Soundar Kumara Understanding provider ordering patterns for patients can enhance resource utilization in hospitals. By using patient and provider retrospective data, this research aims to develop statistical and network-based techniques to predict orders for upcoming appointments. These predictions can lead to a recommendation system supporting providers to increase order effectiveness and ease order documentation. The models will be validated with outpatient data from pulmonary clinics. 3 - RTLS Data Applications in Healthcare Analytics and Modeling Seth Hostetler, Lead Process Engineer, Geisinger Health System, Often, when developing models in healthcare, the lack of accurate and complete data is cited as a limitation. A real-time location system is able provide additional information concerning patient, staff, and asset locations and flows. This presentation will present results from applications in three application areas (ambulatory, inpatient, and emergency department) where RTLS data has been utilized to provide additional insight via analytic methods and improved modeling capabilities. 4 - Outpatient Clinic Provider Scheduling under Uncertainty Deepak Agrawal, PhD Candidate, Pennsylvania State University, 801 Southgate Drive, A4, State College, PA, 16801, United States of America, dua143@psu.edu, Christopher Stromblad, Soundar Kumara, Mort Webster Current clinic provider scheduling systems are inefficient, because scheduling is done manually and months in advance, most of the demand, capacity, and other constraints and their uncertainties are not considered. There is no objective way to evaluate the quality of a schedule; until after the fact. Therefore, the original schedule becomes suboptimal at the time of realization. A stochastic program to provider scheduling is proposed considering demand and capacity uncertainties using real data. 100 N Academy Ave, Danville, PA, 17822-2550, United States of America, sthostetler@geisinger.edu

Newark, DE, 19718, United States of America, Tze.C.Chiam@ChristianaCare.org, Kristen Miller, Bailey Ingraham Lopresto

As an effort to improve patient safety, Christiana Care Health Systems embarked on work to evaluate current rescue strategies for patient fall events. This study investigates the use of age, bone density, coagulation, surgery and fall type to

384