Mining Typical Order Sequences from EHR for Building Clinical Pathways

Abstract

Clinical pathway is one of the key tools for providing standardized treatment for patients. However, building a new pathway from scratch is a time-consuming task for medical staffs, as it involves optimization of the treatment plan while preserving operability in a hospital. In this paper, we present a method for mining typical treatment processes from electric health records (EHRs) for facilitating creation of new pathways by providing base blocks. Firstly, we constitute occurrence and transition frequency matrices of clinical orders using all cases. Next, we compute the typicalness index for each order sequence based on the occurrence and transition frequencies. After that we perform clustering of all cases according to the similarity defined on the typicalness index. Experimental results on two disease datasets demonstrate that the method is capable of producing clusters that reflect differences of treatment processes without a priori information about order types.