GOALS: Using TimeML Annotations for an Information Extraction Approach to Support the Modeling of Clinical Guidelines

Clinical practice guidelines and protocols aim at raising the quality of healthcare. They are written in a narrative style and have to be translated into a computer-interpretable format to be usable in clinical software applications. In order to ease this challenging and laborious task for the modeler we developed a methodology called GOALS1. It is specified independently from the target computer-interpretable guideline language and uses a guideline’s text annotated with temporal concepts provided by TimeML as a starting point. It describes step-by-step how parts of the guideline’s model can be generated and finally assessed by means of an evaluation scheme. Information extraction techniques – machine learning algorithms and knowledge engineering methods – are applied to support the different steps in order to generate parts of the model automatically. A scenario-based application of GOALS shows the translation of temporally-related sentences of a clinical protocol into the corresponding semi-formal model.
Evaluation results are clear indicators for the GOALS methodology’s easing of the time-consuming modeling process.
Year of Publication
Thesis Type
PhD Dissertation
TU Wien
Supplementary Material