Supporting Computer-interpretable Guidelines' Modeling by Automatically Classifying Clinical Actions

Modeling computer-interpretable clinical practice guidelines is a complex and tedious task that has been of interest for several attempts to automate parts of this process. When modeling guidelines one of the tasks is to specify common actions in everyday’s practical medicine (e.g., drug prescription, observation) in order to link them with clinical information systems (e.g., an orderentry system). In this paper we compare a rule-based and a machine-learning method to classify activities according to the Clinical Actions Palette used in the Hybrid-Asbru ontology. We use syntactic and semantic features, such as the Semantic Types of the UMLS to classify the activities. Furthermore, we extend our methods by using 2-step classification and combining machine learning and rule-based approaches. Results show that machine learning performs better than the rule-based method on the classification task. They also show that the 2-step classification method improves the categorization of activities.