Grant: Discovery Grants Program - Individual

The process of temporal abstraction (TA) takes either raw or pre-processed stream data as input and produces context sensitive and qualitative interval based representations. Data Mining provides a mechanism to perform knowledge discovery on ever increasing volumes of data. That is, to transform it from data to information and ultimately to knowledge. Research has begun to apply TA concepts to the pre- processing steps of data mining. A number of limitations of existing temporal abstraction systems in particular relate to (1) their focus on one particular clinical event/condition, rather than creating an environment to explore multiple clinical events/conditions; (2) an inability to support complex abstracts that for example represent abstractions of behaviour across multiple streams and (3) an inability to support abstractions within a multidimensional context. These limitations are impacted by limitations within the supporting data mining principles for pre-processing temporally abstracted time series data. There is enormous potential to data mine historical de-identified patient datasets to detect common trends and patterns, occurring prior to the onset of disease or illness, that have common contextual behaviours rather than the same exact numerical values. To enable this, the proposed research program will build on established collaborative research projects to establish new approaches to data mining that incorporate mechanisms to detect behaviours rather than actual values are required. These approaches will provide a highly flexible environment to support a range of clinical research projects. In addition, these new approaches need to function where the de-identified datasets are located in many different locations. The research foci proposed are innovative and emerging with significant potential, supporting many national, regional and UOIT research priorities and goals. The research program proposed will support the training of masters and PhD students thereby creating the next generation of Health Informatics high quality personnel. Healthcare represents a significant portion (> 30%) of the Canadian National Budget. Health Informatics has the potential to reduce mortality and morbidity rates, improve quality of life and reduce health care costs. Intensive care units (ICUs) and in particular, Neonatal ICUs generate vast amounts of physiological patient data. Recent clinical research has shown the potential for common patterns in seemingly unrelated physiological data to potentially support earlier diagnosis of disease and the onset of illness in neonatal babies.