Informs Annual Meeting 2017

MA06

INFORMS Houston – 2017

MA06

4 - Supply Function Equilibria in Electricity Markets under Uncertainty Chaithanya Bandi, Northwestern University,

320C Health Data Analytics for Policy Making Sponsored: Health Applications Sponsored Session Chair: Ilbin Lee, PhD, Georgia Tech, 755 Ferst Drive, Atlanta, GA, 30332, United States, ilee79@gatech.edu 1 - Quantifying the Cost of Pediatric Depression in the Medicaid Population Erin Garcia, Auburn University, Auburn, AL, United States, eg3gatech@gmail.com Co-morbid depression often results in increased health care costs and worse outcomes in studies focused on particular conditions and groups of people. To determine the long term cost of pediatric depression in the Medicaid population, we identify the population of children with and without depression from the CMS data files. Each child with depression is matched to a child without depression based on a set of socioeconomic and health factors so that we can treat the data as paired samples. Healthcare utilization and cost data for each child in the final sample are computed from the CMS data, and pairwise analysis is conducted to identify differences between the populations with and without depression. 2 - Spatial Access to Psychotherapy Services for Publicly-insured Children Pravara M.Harati, Georgia Tech, Atlanta, GA, United States, pharati3@gatech.edu Access to specialty care is typically worse for publicly-insured children than for privately-insured. We estimate local level spatial access, including provider congestion and average travel distance, to child psychotherapy services by first using CMS Medicaid claims data to obtain the locations and annual capacities of mental health professionals who have provided psychotherapy to publicly- insured children. We then estimate the local level prevalence of mental health conditions using a simulation model based on demographic data. Finally we match the demand to supply in an optimization model to obtain the access measures, from which we can identify geographic areas needing improvement. 3 - Using Infusion Pump Data Analytics to Study Medication Safety Issues Kang-Yu Hsu, Purdue University, Gerald D. and Edna E. Mann Hall, 203 S Martin Jischke Drive, Lafayette, IN, 47907, United States, hsu66@purdue.edu, Poching DeLaurentis, Yuehwern Yih, Yuval Bitan, Dan Degnan An infusion pump is a medical device used to intravenously deliver fluids into a patient’s body in controlled amounts and rates. These infusion pumps have safety features incorporated in a software program known as Dose Error Reduce Systems (DERS) that uses customized Drug Limit Library (DLL). DLLs in infusion pumps may not be updated efficiently. The inconsistency between DLL in pumps and the most current DLL may put patient safety at risk such as dosage/rate miscalculation, false alerts and missed alerts. In this study, we quantify the impact of adopting out-of-date DLL through investigating infusion logs, and examine infusions which potentially jeopardized patient safety with outdated DLL usage. 4 - The Longitudinal Trajectories of Medication Adherence for Patients with Chronic Conditions Shan Xie, Purdue University, 315 N. Grant Street, W. Lafayette, IN, 47907, United States, xie34@purdue.edu, Alan Zillich, Yuehwern Yih Poor adherence to medication has been associated with increased healthcare costs and adverse health outcomes, thus improving medication adherence is important to realize the full benefit of medication therapies. The current claims-based quality reporting measures only provide an aggregate number to estimate patients’ adherence level over a long-term period, which are easy to implement but lack the ability to distinguish between different adherence patterns. Therefore, we used group based trajectory models to stratify patients into groups with distinct longitudinal adherence trajectories. We hope to provide useful insight to help design targeted and tailored interventions.

2001 Sheridan Rd, 566, Evanston, IL, 60208, United States, c-bandi@kellogg.northwestern.edu, Ermin Wei, Yuanzhang Xiao We study a two-stage electricity market with renewables. Each energy producer in the market has a portfolio of both renewable and conventional energy generators. In the real-time market, each producer tries to fulfill its DA committed energy with (zero-cost) renewables. We study the robust supply function equilibrium (SFE) where each producer has incomplete information about the other producers’ marginal costs and the distribution of renewable energy, and performs worst-case optimization. We fully characterize the unique robust SFE, and study the impact of the feed-in tariff on the equilibrium outcome. 320B Analytics & Optimization in Chronic Disease Screening and Management Sponsored: Health Applications Sponsored Session Chair: Turgay Ayer, Georgia Institute of Technology, Atlanta, GA, 30332, United States, ayer@isye.gatech.edu Co-Chair: Caglar Caglayan, Georgia Institute of Technology, 765 Ferst Drive, Atlanta, GA, 30332, United States, ccaglayan6@gatech.edu 1 - Analytics for Personalized Healthcare using Mobile Devices Anil Aswani, UC Berkeley, 8 10th St, Apt 1704, San Francisco, CA, 94103, United States, aaswani@berkeley.edu, Elena Flowers, Yoshimi Fukuoka, Ken Goldberg, William Haskell, Philip Kaminsky, Yonatan Mintz, Eric Vittinghoff, Mo Zhou Mobile technologies hold much promise for management of chronic diseases. However, algorithm design requires careful consideration of the data-transmission and computational capabilities of mobile technologies. This talk describes the development of a personalized physical activity intervention. First, we describe the results of a clustering analysis of objectively measured physical activity patterns in women. Insights from this analysis are used in development of a ROGUE bandit algorithm for the design of personalized physical activity interventions. A key innovation of our ROGUE bandit model is the ability to capture habituation that occurs with repeated exposure to a single treatment. 2 - A Comparative Analysis of Data Mining Approaches to Characterize Colorectal Cancer Screening Behavior Trajectories Maria Esther Mayorga, North Carolina State University, 400 Daniels Hall, Dept. of Industrial & Systems Engineering, Raleigh, NC, 27695, United States, memayorg@ncsu.edu, Rachel M. Townsley Patient behavior patterns can heavily impact outcomes and policy efficacy in the context of cancer screening, but are difficult to replicate in health policy simulation models. Using censored claims data we investigate the validity of common screening behavior assumptions and their impact on relevant individual health outcomes in the context of colorectal cancer (CRC). We compare data driven strategies for modeling trajectories of screening behavior over time. 3 - Assessing Multi-Modality Screening Policies for Women at High Risk of Breast Cancer Caglar Caglayan, Georgia Institute of Technology, 765 Ferst Drive, Atlanta, GA, 30332, United States, ccaglayan6@gatech.edu, Turgay Ayer, Donatus U. Ekwueme Women with breast density, family history of breast or ovarian cancer, or BRCA1 and BRCA2-mutation-carriers are at higher risk of breast cancer. For such women, non-mammographic modalities such as ultrasound or MRI, adjunct to or instead of mammogram, can overcome the limitations of mammogram and hence, be beneficial. Currently there is no consensus on the optimal use of these modalities in high-risk women. We study multi-modality breast cancer screening problem for the high-risk population and identify practical and cost-effective optimal screening policies. Our results show that screening with ultrasound is cost-effective and optimal over a wide range of budget levels in young high-risk women 4 - Identifying Factors that Influence Outcome for Prostate Cancer Patients Eva Lee, Professor and Director, Georgia Tech, Ctr for Operations Research in Medicine, Atlanta, GA, 30332-0205, United States, eva.lee@gatech.edu This work is joint with Kaiser Permanente. We analyze 10 years of prostate cancer patients to investigate factors that influence the outcome. Unsupervised learning will be performed to uncover outcome patterns and to subgroup patients according to their outcome and health similarities. Machine learning is then performed to identify factors that are critical for achieving treatment outcome. Modeling and computational challenges will be discussed. MA05

133