Outcome Prediction System Based on Different Types of Data in Electronic Health Records for Septic Patients

Haruya ISHIZUKA  Tsukasa ISHIGAKI  Naoya KOBAYASHI  Daisuke KUDO  Atsuhiro NAKAGAWA  

D - Abstracts of IEICE TRANSACTIONS on Information and Systems (Japanese Edition)   Vol.J101-D   No.3   pp.481-493
Publication Date: 2018/03/01
Online ISSN: 1881-0225
Type of Manuscript: Special Section PAPER (Special Section on Student Research)
medical data analysis,  mortality prediction,  machine learning,  text mining,  topic model,  

Full Text(in Japanese): FreePDF(868.6KB)

In this paper, we propose an outcome prediction system for septic patients combining severity scores, vital signs and nursing notes obtained from electronic health records. The proposed system consists of two stages. In the first step, we extract three features; severity score, mode proportion from vital signs using switching AR model and topic proportion from nurse notes using latent Dirichlet allocation. In the second step, we predict patients' outcome using the support vector machine based on these three features. Through demonstration experiments, we showed that the proposed system can predict an outcome more accurately than conventional methods.