B-2-3 自己組織化マップを用いた在院日数予測と評価(自己組織化マップ(1))

抄録

Diagnosis Procedure Combination / Per-Diem Payment System (DPC/PDPS), which has been introduced since 2003, standardizes medical records. It is possible to collect analyzable clinical information nationwide as the uniform format (DPC formatted data) from medical centers and hospitals. In this paper, we describe a prediction method of a length of stay in the hospital from DPC data using Self-Organizing Maps (SOM). Firstly we show an encoding method of training vectors for SOM from DPC data, secondly we propose a visualizing method for different lengths of stay in the hospital on the SOM clusters and finally we discuss the evaluation of availability of the prediction method.