Granularity-based temporal data mining in hospital information system

This paper proposes granularity-based temporal data mining method which constructs clinical process conducted by nurses. The methods consist of three process. First, data on counting sum of executed orders are extracted from hospital informaton system with a given temporal granularity. Then, similarity-based methods, such as clustering and multidimensional scaling (MDS) are applied to the data and the labels for grouping are obtained. By using the labels, rule induction is applied, and classification power of each attribute is estimated. The attributes are sorted by an index of classification power, the original dataset is decomposed into subtables. Clustering, rule induction and table decomposition methods are applied to the subtables in a recursive way. The method was applied to datasets stored in hospital information system stored in 10 years. The results show that the reuse of stored data will give a powerful tool for construction of clinical process, which can be viewed as data-oriented management of nursing schedule.