A Predictive Analytics Toolbox for Medical Applications Michael L. Valenzuela and Jerzy W. Rozenblit Electrical and Computer Engineering Department University of Arizona mvalenz@ece.arizona.edu, jr@ece.arizona.edu Allan J. Hamilton College of Medicine University of Arizona allan@surgery.arizona.edu Keywords: Management, Decision Support, Healthcare, Vi- sualization Abstract Ever more frequently business enterprises are benefiting from the collection and analysis of large data. This paper reports on a work in progress of adapting intelligence, predictive analyt- ics tools for analysis of medical data. While large data has been used in medical research, only recently has this become a trend for hospital management. A suite of tools originally developed for military intelligence analysts are repurposed for hospital management. The original design concepts is re- viewed, its medical applications and challenges are described along with an illustrative example. 1. INTRODUCTION Over the last two decades, information technology (IT) systems are being increasingly adopted by hospitals. Even seemingly mundane IT systems such as computerized patient records were already installed in 1997 at Cabarrus Family Medicine in Concord, NC [1]. In 1996, the Johnson Medical Center in Johnson City, TN, decided it needed a data ware- house to spot trends and anomalies [1]. The next natural pro- gression of this pattern is to use a computer system driven by the data warehouse to provide medical decision support. Hospital management is in need of decision-support sys- tems (DSS). DSS have been used in many domains, ranging from stability and support operations [2, 3], to intelligence analysis [4–6], to the medical domain [7–11]. Such a sys- tem would help hospitals in many ways: cut costs; discover and prevent causes of medical mistakes; decide whether a patient should recover from home or have a longer hospital stay; identify risk factors; optimize bed assignments; analyze work flow; and in general uncover actionable information. Yet, most prior DSS in the medical domain have focused on recommending drug prescriptions, diagnosing the source of chest pain, treating infertility, and promptly administering im- munizations [8,11]. Yes, hospital management systems have lagged behind that of medical science. A multitude of work applies advanced data analysis tech- niques to biomedical problems. [12] use hidden Markov mod- els (HMM) to shed light on the folding pathways of a com- plex protein. [13] built a metamorphic virus detector based on HMM. [14] uses reinforcement learning and an artificial neu- ral network to improve the outcome of fractionated radiother- apy. Back in 2002, support vector machines improved cancer classification using gene selection from 86% accurate to 98% accurate [15]. However, the use of advanced tools for hospi- tal administration has lagged behind advances in biomedical research, possibly due to the different nature of hospital ad- ministration. The demands on biomedical research are similar, but not the same as that of hospital administration. Both share ethical concerns, but biomedical ethics concerns itself with “playing God” and the dangers of the substances/organisms. Hospitals have to worry more about privacy concerns and patient safety. The top ranking concerns for hospitals have been financial concerns for the last 10 years [16]. As such hospitals are con- cerned with avoiding bad debt, reducing operating costs, and preventing lawsuits. Whereas in the world of research the pri- mary demand is get valid, significant results, before anyone else. In this paper, we present a single comprehensive tool, Med-Think TM . Med-Think, previously developed as an in- telligence analyst’s toolbox [4–6], reduces the gap between hospital management tools and those tools used for biomed- ical research. It offers a plethora of data visualization, ex- ploration, querying, analysis, and management capabilities. Even though Med-Think is still under development, it offers hospital management a data driven management and decision support system. The rest of this article proceeds as follows. Section 2 re- views alternative systems and briefly discusses the history of Med-Think. This is followed by a description of how Med- Think’s models data in Section 3. Section 4 describes Med- Think’s capabilities and applications. We discuss immediate challenges to the adoption of the system in Section 5. Last, we conclude and discuss future directions for Med-Think in Section 6. 2. BACKGROUND Due to the rapidly emerging interest in advanced data driven hospital management, the breadth, variety, and num- ber of these tools are likely to explode in the coming years. However, the recent demand for advanced data driven hospi- tal management tools has been left unmet. This is partially due to the lack of available tools. As such, we only review a 180