[1H4-J-13] AI application: medicine

[1H4-J-13-04] Clustering Difference of the features of the by Weight Values in the Weighted PLSA using DPC data

〇Kazuya Yamashita¹, Chisae Murata¹, Yuichiro Sakamoto², Eiichi Sakurai^1,2, Yoich Motomura^1,2 (1. AI Research Center,AIST, 2. Saga University)

Keywords:Diagnosis Procedure Combination Data, Probabilistic modeling, Bayesian networks, Probabilistic Latent Semantic Analysis, Big data, sepsis

The accumulation of DPC data has progressed, and the importance of these medical big data is recognized. At present, , there are few reports such as application for selection and comparison of a new treatment strategy using DPC. We used DPC data and PLSA to clarify simultaneous clustering of patients requiring intensive care and clinical practice and to extract clusters time transition patterns of patients and to evaluate feasibility of doctor's treatment strategy decision support algorithm we have verified. Furthermore, by using weighted PLSA, clustering can be performed which clearly shows the difference of attention variables by giving weight to 'attention' medical treatment, objective variables such as mortality rate, number of hospital days and medical expenses It is confirmed that it is. In this research, we treat all DPC items including variables that we have not used so far as variables, and investigate the differences in clustering features by changing the weight of each variable.As a result, it was shown that it is possible to generate clusters according to the purpose by adjusting the weight of each variable as a parameter according to the variable to be noticed.