Support for Authoring and Transformation of CGPs

Clinical guidelines and protocols (CGPs) are promising means to improve the quality of health care. Supporting their application by computer systems in connection with electronic patient records increases conformance and reduces workload. To achieve this, CGPs must be translated to a computer readable form. The considerable effort to do so can be reduced by means of Information Extraction. Efficiently performing this task is still an open research frontier. At the same time, there is already valuable experience from previous efforts in the field, which is currently not taken to the field of creating CGPs in their original form (free text and diagrams, in natural language). In previous projects, we developed expertise in the fields of guideline modelling, with and without the use of Information Extraction, and in guideline validation, verification, and execution. Combining it with available open- source software, we intend to develop a comprehensive solution for authoring CGPs in formal and informal (freetext) form by a distributed team of experts with complementary expertise. To arrive at this, we will take the following actions. Evaluation of existing partial solutions for the task of authoring highly structured, large documents in a distributed environment by contributors with heterogeneous background. Integration of a suitable selection of open-source solutions, complementing them with custom-made middleware where required by the particular needs of the intended field of application. Further development of Information Extraction methods to support the transformation of CGPs to formal models. Creation of demonstration use-cases and evaluation of the prototype with domain experts. Creation of educational material and organisation of a workshop and individual meetings with interested experts to raise the awareness in the community for the then available solutions. The result of this research will be models and methods that can take the state of art in the important field of authoring formal and informal CGPs a significant step towards practical application. By applying these models and methods not only the transformation into computer-interpretable guideline formats will be facilitated, but also creators of new freetext CGPs will benefit from an environment that assists collaborative aspects and distributed work.

An important aspect in the treatment of patients is the application of evidence-based medicine, as described in clinical practice guidelines. To integrate guideline-based treatment in clinical practice, there are efforts to incorporate guidelines into clinical information systems supporting the medical staff. To achieve this, guidelines must be represented in a computer-interpretable format, i.e., the documents must be translated into a format a computer understands. In our project we dealt with the translation of guideline documents (available in natural language) into such a computer-understandable format as well as the maintenance of the resulting representations during a guidelines lifecycle. To this end, we first developed an intermediate representation that allows also users with little technical background to structure the content of a guideline for subsequent modelling in a more formal representation. We also analysed the requirements for a tool to support authors with different background (e.g., medical or technical) in modelling a guideline. As the modelling of a guideline is labour-intensive and complex, we explored ways to automate parts of the modelling process by applying various Information Extraction techniques for the different kinds of information a guideline consists of and that are relevant for execution: The system uses rules to detect relevant information and to assemble it in a structured way. In addition, we developed methods based on lexical and linguistic features and semantic relations using heuristic-, rule-based, and machine learning approaches. We also analysed temporal aspects in the context of treatment processes in guidelines and showed how the temporal information in the text can be used to identify activities and to order them to generate a workflow of the treatment plan. Furthermore, we developed methods to transform a computer-interpretable guideline model from one notation in another. This is important in the context of a guidelines lifecycle, as in this way models or parts of models that, for instance, have been generated by other organisations, can be re-used and easily integrated to promote the application of guidelines in clinical information systems.