Opportunities and Challenges in Association and Episode Discovery from Electronic Health Records

83 guidance; machine understanding of human behavior, emotional, and physiological states; and the need to respond appropriately to unintended stimuli. Numerous machine perception , cognition, and communication challenges also remain such as image-guided intervention, speech and language understanding, two-hand-like manipulative dexterity, and learning systems that adapt to an individual's long-term change of state. Solutions likely lie at the intersection of new and ongoing research in computer science, materials, psychology, and neuroscience. The societal pressure to mitigate the healthcare crisis presents an unprecedented opportunity for computing , information science, and engineering. Whereas the pursuit of understanding the pathogenesis of disease will be accelerated with new algorithms and increasingly powerful computation and data architec-tures, we look to other computation-enabled means to provide additional avenues to the pursuit of quality of life. Multidisciplinary approaches are required to engineer a privacy-maintaining information infrastructure with secure, real-time access to unprecedented amounts of heterogeneous health, medical, and treatment data. New generations of algorithms must be developed to utilize the resulting global resource of population-based evidence for assisted discovery, knowledge creation, and even individual point-of-care decisions. Ana-lytics based on modeling phenomena ranging from the physiology of humans to their social interactions are required to optimize therapies ranging from molecular medicine to be-havioral interventions. Such advances in human-centered computing in combination with standardization and commercialization of unobtrusive sensing and robotics will trigger a disruptive change in health-care and wellbeing by empowering individuals to more directly participate. Finally, partnerships among academic , industrial, and governmental bodies are required to enable these computer science innovations and realize their deployment in order to help transform healthcare. Will Barkis is a AAAS science and technology policy fellow at the American Association for the Advancement of Science. Contact him at wbarkis@gmail.com. As healthcare practices, both small and large, move from traditional paper-based patient charts to electronic health records (EHRs), new opportunities are emerging for secondary uses of data captured as part of routine care. Such opportunities include not only traditional research meth-odologies involving relatively small cohorts of selected patients, but also large-scale data mining analyses encompassing hundreds of thousands or even millions of patients at once. Performing these nontraditional analyses has required novel computational approaches, sometimes borrowing from techniques originally developed in other elds such as genomics and network theory. Additionally, to interpret such large volumes of data in a meaningful