Description
This study constructed three models: customized SAPS-II, CART and RF for LOS-ICU prediction. Results showed that RF had the best predictive performance among the three models based on the MIMIC-III dataset. Moreover, we confirmed the validity of using SMOTE to deal with data imbalance in prediction model development. This study is useful for further efforts to help physicians make appropriate clinical interventions and do optimal medical resources allocation in ICU.
Learning Objective: After participating in this session, the learner should be better able to:
Learn performance of different models built for prediction of length of stay in ICU and their potential impact in patient care.
Formulate an approach to adoption of machine learning technology for length of stay in ICU prediction in ICU clinical practice.
Authors:
Jingyi Wu, Peking University
Ke Lin, Peking University
Yu Lin, Peking University
Yonghua Hu, Peking University
Guilan Kong (Presenter)
Peking University
Presentation Materials: