Mining Clinical Pathway Candidates from Order History Based on the Clustering of Order Sequences

In this paper we present a method for extracting common orders that may constitute building blocks for new clinical pathways. We firstly constitute occurrence and transition frequency maps of orders based on the order history stored in a hospital information system. Then we assess the ´typical ness´ of each order sequence according to the generated maps. After that, we perform clustering of order sequences based on the similarity measured on the typical ness. Experimental results demonstrate that our method could generate clusters that reflect difference of clinical processes induced by the difference of operation dates, as well as simple sets of orders that have high values of typical ness.