Information structures and workflows in health care informatics
Abstract: Patient data in health care have traditionally been used to support direct patient care. Although there is great potential in combining such data with genetic information from patients to improve diagnosis and therapy decisions (i.e. personalized medicine) and in secondary uses such as data mining, this is complex to realize due to technical, commercial and legal issues related with combining and refining patient data.Clinical decision support systems (CDSS) are great catalysts for enabling evidence-based medicine in clinical practice. Although patient data can be the base for CDSS logic, it is often scattered among heterogenous data sources (even in different health care centers). Data integration and subsequent data mining must consider codification of patient data with terminology systems in addition to legal and ethical aspects of using such data. Although computerization of the patient record systems has been underway for a long time, some data is still unstructured. Investigation regarding the feasibility of using electronic patient records (EPR) as data sources for data mining is therefore important.Association rules can be used as a base for CDSS development. Logic representation affect the usability of the systems and the possibility of providing explanations of the generated advice. Several properties of these rules are relatively easy to explain (such as support and confidence), which in itself can improve end-user confidence in advice from CDSS.Information from information sources other than the EPR can also be important for diagnosis and/or treatment decisions. Drug prescription is a process that is particularly dependent on reliable information regarding, among other things, drug-drug interactions which can have serious effects. CDSS and other information systems are not useful unless they are available at the time and location of patient care. This motivates using mobile devices for CDSS. Information structures of interactions affect representation in informatics systems. These structures can be represented using a category theory based implementation of rough sets (rough monads).Development of guidelines and CDSS can be based on existing guidelines with connections to external information systems that validate advice given the particular patient situation (for example, previously prescribed drugs may interact with recommended drugs by CDSS). Rules for CDSS can also be generated directly from patient data but this assumes that such data is structured and representative.Although there is great potential in CDSS to improve the quality and efficiency of health care, these systems must be properly integrated with existing processes in health care (workflows) and with other information systems. Health care workflows manage physical resources such as patients and doctors and can help to standardize care processes and support management decisions through workflow simulation. Such simulations allow information bottle-necks or insufficient resources (equipment, personnel) to be identified.As personalized medicine using genetic information of patients become economically feasible, computational requirements increase. In this sense, distributing computations through web services and system-oriented workflows can complement human-oriented workflows. Issues related to dynamic service discovery, semantic annotations of data, service inputs/outputs affect the feasibility of system-oriented workflow construction and sharing. Additionally, sharing of system-oriented workflows increase the possibilities of peer-review and workflow re-usage.
CLICK HERE TO DOWNLOAD THE WHOLE DISSERTATION. (in PDF format)