Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

TA61

4 - A Deep Active Survival Analysis Approach for Precision Treatment Recommendations Kai Yang, Wayne State University, 4815 Fourth Street, Detroit, MI, 48201, United States, Milad Zafar Nezhad, Survival analysis has been developed and applied in the number of areas including manufacturing, finance, economics and healthcare. In healthcare domain, usually clinical data are high-dimensional, sparse and complex and sometimes there exists few amount of time-to-event (labeled) instances. Therefore building an accurate survival model from electronic health records is challenging. With this motivation, we address this issue and provide a new survival analysis framework using deep learning and active learning with a novel sampling strategy. In the experimental study, we apply our approach on SEER- Medicare data related to prostate cancer among African-Americans and white patients. 5 - A Tensor Factorization Approach to Predicting Clinical Outcomes Qingpeng Zhang, City University of Hong Kong, 83 Tat Chee Avenue, 6/F, Academic 1, Kowloon, 12180, Hong Kong, In this talk, I am going to introduce the tensor factorization based models for the prediction of various clinical outcomes at the individual level. In particular, we focus on the prediction of chronic diseases and re-admissions of the elderly. Experiments with EHR data in Hong Kong will also be presented. n TA60 West Bldg 102B Feryal Erhun Sponsored: Health Applications Sponsored Session Chair: Feryal Erhun, University of Cambridge, Cambridge, CB2 1AG, United Kingdom 1 - Data-driven Machine Learning Algorithm for High-priority Drug-drug Interactions Discovery Ning Liu, Penn State University, 233 Leonhard Building, University Park, PA, 16802, United States, Qais Hatim, Soundar Kumara Drug-drug interactions (DDIs) are the primary causes for adverse events and have become serious concerns for both drug discovery and pharmacovigilance. Unfortunately, the identification of DDIs is challenging and require domain knowledge and enormous efforts from experts. The FDA Adverse Event Reporting System routinely collects adverse event reports from patients and provides a rich data source to discover unknown DDIs. We propose a machine learning method to find patterns from known high-severity DDIs and predict pairs of drugs that may potentially interact and result in adverse events. Our data-driven approach incorporates domain knowledge and provides an option for detecting DDIs. 2 - Outreach and Mobile Clinic Strategies for Vaccine Distribution Yuwen Yang, PhD Candidate, University of Pittsburgh, 3700 O’Hara Street, Benedum Hall of Engineering, 1043A, Pittsburgh, PA, 15261, United States, Hoda Bidkhori, Jayant Rajgopal In many low and middle income countries with geographically dispersed populations, last-mile vaccine delivery can be a complex process. Remote locations in these countries often do not have direct access to clinics/hospitals, and residents often face significant difficulty in obtaining routine vaccinations. An approach known as outreach is typically utilized to raise immunization rates in these situations. A set of these remote population centers are chosen as mobile clinics, and clinicians and support personnel are sent from a depot to each team to vaccinate people in the immediate surrounding area. We formulate this process as a mixed integer programming problem and discuss related issues. 3 - The Determinants of Process Innovation in the Pharmaceutical Industry Ivan Lugovoi, HEC Paris, 1 Square Theodore Judlin, Paris, 75015, France Utilizing a unique dataset of the process - patent expert evaluations merged with the sales of 50 generic pharmaceutical products over a 10-year period across 4 countries, we observe that process innovation is associated with a significant increase in a firm’s market share. To explain the origins of this effect we explore how market determinants: market concentration and growth, product lifecycle, and price elasticity, and technology determinants: vertical integration, technological diversification, and opportunity affect three dimensions (novelty, scope, and locus) of process innovation 4 - Capacity Management in New Drug Development: A Contract Research Organization Perspective Sjors Jansen, PhD Candidate, Eindhoven University, P.O. Box 513, Paviljoen E13, Eindhoven, 5600MB, Netherlands In recent years, pharmaceutical companies outsourced the development of new drugs to Contract Research Organizations (CROs). CROs conduct the development of a new drug for a fixed fee. CROs have a large research capacity, which they aim to fill with profitable research projects. Each project consists of multiple phases. Due to the uncertainty in the outcome of project phases, the

required capacity in the future is uncertain. Existing literature considers the problem from the perspective of a pharmaceutical company. CROs have different incentives and make different trade offs. Therefore, we model the capacity problem from a CRO’s perspective to help a CRO to select projects from pharma companies.

n TA61 West Bldg 102C Joint Session HAS/Practice Curated: Operations Research in Hospital Operations Sponsored: Health Applications Sponsored Session Chair: Miao Bai, PhD, Mayo Clinic, Rochester, MN, 55904, United States 1 - Dynamic Coordination of Exams in the Radiology Practice Miao Bai, Mayo Clinic, Rochester, MN, 55904, United States, Mustafa Y. Sir, Kalyan Pasupathy Radiology exams vary significantly in duration and technology requirements. Coordination between various exams with stochastic duration and requiring different scanner technology is challenging but essential to the timely fulfillment of demand and the efficient utilization of expensive scanner resources. Radiology exam coordination is further complicated by the great number of unscheduled exam requests that need to be completed within a specific time frame. To address this problem, we developed a real-time radiology exam coordination system based on a dynamic programming framework. 2 - Generalizability Among Hospitals that Percentages of Surgical Cases During Weekends or Holidays Change Proportionally with Total Caseload Over Decade Franklin Dexter, Professor, University of Iowa, 200 Hawkins Dr, 6JCP, Iowa City, IA, 52242, United States, Richard Epstein Previous studies showed how to calculate weekend and holiday anesthesia staffing to assure no greater than a prespecified percentage of days with =1 surgical case waiting longer than at baseline. We used Iowa Hospital Association data, Jan 2007 - Jun 2017; N = 42 hospitals with =10 cases during weekends or holidays per period. We confirmed that over many (10) years hospitals’ proportions of cases performed on weekends and holidays remain stable. The implication is that weekend and holiday staffing can be reevaluated annually, the interval over which changes in total caseloads are evident. 3 - Mathematical Methods for Efficient Allocation and Utilization of Hospital Resources Tracey Hong, Stanford University, Stanford, CA, United States, Emily Grimm, Michael Fairley, Steven Frick, David Scheinker, David Scheinker The rising cost of healthcare is increasing pressure on hospitals to improve the value of care, but hospitals often lack access to modern quantitative methods to maximize the efficiency of resource allocation and utilization. Many mathematical models have been developed to better inform hospital decisions, but relatively few have been implemented. We describe examples of several models, designed and implemented, at Lucile Packard Children’s Hospital Stanford, including optimal surgeon scheduling with integer programming, patient volume forecasting, bed assignments via simulation, and nurse staffing with a regression model. We emphasize the components that made implementation possible. 4 - Analytics Systems to Improve the Value of Surgical Care David Scheinker, Stanford - Lucile Packard Children’s Hospital, Huang Engineering Center, 475 Via Ortega, Stanford, CA, 94305, United States Variability in surgical supplies, procedure durations, operating room utilization, patient arrival rates to the PACU, and post-operative inpatient lengths of stay are associated with delays in care, staff frustration, and higher medical costs. Improving these processes with manual reviews and revisions can be prohibitively time consuming. We give an overview of the design and implementation of partially automated systems that use EMR data to reduce the variability, errors, and costs associated with these processes. The methods used include dynamic Bayesian networks, machine learning, causal inference, integer programming, and discrete event simulation.

265