Gaining New Medical Insights through Temporal Data Abstraction and Clinical Protocols

The aim of the proposed project is the experimental application and further development of promising approaches to temporal data abstraction and support of protocol-based care in neonatal intensive care, in order to enrich the medical knowledge of the field and to obtain feedback on the usability and applicability. Currently, intensive care faces two main problems: First, the medical staff obtain hosts of data from monitoring devices but lack of means to analyze them to the degree desirable. Second, while the benefits from computer support for protocol-based care are generally excepted, it is very difficult to introduce their computer-supported execution in fields, like intensive care. In a previous FWF project, P12797-INF, we developed a workbench containing tools for temporal data abstraction, for protocol development and execution, and for their task-specific visualization. We will apply these tools and methods to develop solutions for the above problems in an interdisciplinary way, building on extensive existing cooperations with several institutions in the field. Both the domain of neonatology and computer science will benefit from the proposed project. First, the knowledge about the physiological processes in premature neonates will be enriched. Second, state-of-the-art methods for temporal data abstraction and process modeling will be applied in practical settings, providing valuable feedback and fostering their further development.

The present research project dealt with two research problems: (1) How can we transform clinical treatment plans into a computer-interpretable format (2) How can we visualize such computer-interpretable treatment plans and the patient data task- and user-oriented. To provide an overview of both, actual and possible treatment courses to the medical personnel and the patients based on clinical guidelines, these guidelines have to be presented in a computer-interpretable format. For this reason, the documents have to be translated into a format a computer "understands". In our project we deal with the translation of guideline documents (which are available in natural language) into such a computer-understandable format. Due to the complex, cumbersome, and time-consuming translation process, which also demands detailed medical knowledge, our aim is to automate parts of this process. This task is accomplished using Information Extraction techniques: A system uses rules to detect relevant information and to assemble it in a structured way. We have developed a methodology (LASSIE) that transforms a clinical guideline in multiple steps using Information Extraction. We have analyzed the structure of guidelines to detect patterns supporting the formalization process. Furthermore, we analyzed temporal aspects in the context of treatment processes in guidelines and how they can be modeled in the formal representations. Using a tool we have developed (DELT/A), the user is able to review each steps` outcome, make adaptations, or add necessary information. Using this method is a substantial facilitation for translating guidelines and it also has the potential to support the development of clinical guidelines. Furthermore, we have developed a method to extract the content of tables in a PDF file within their structural context and implemented it prototypically (pdf2table). Apart from transforming clinical guidelines into a computer-interpretable format, the opposite transformation of a computer-interpretable guideline to a human readable format, is of equal importance. Accordingly, we developed interactive methods for visualizing medical treatment plans and temporal patient data (CareVis) in close cooperation with physicians. The results lead to the conclusion that a systematic view on visualization methods for time and time-oriented information is missing in the current state of the art of research. We were able to fill this gap in the context of the project by conducting a comprehensive survey of the available visualization methods as well as their systematic categorization and characterization. This structured overview is particularly interesting to support users in choosing appropriate techniques for their application problem at hand. Besides, a modular and application independent time- and data model has been developed based on existing approaches. Eventually, a concept for a framework to support the visualization process of time and time-oriented information has been developed in form of a step-wise process (TimeViz). Besides the methodical and conceptual works, we developed and implemented a novel visualization technique for representing planning data and medical treatment plans (PlanningLines). A controlled experiment has shown that users of PlanningLines made less mistakes and are faster in completing tasks when dealing with temporal uncertainties compared to a standard method. Due to the fact that information visualization is such a young discipline, no unique wording has been established so far. In order to support the scientific community in the attempt to make the used terms more precise, an international information and communication platform based on wiki technology that can be used actively by each user has been conceptualized and hosted (InfoVis:Wiki).