TY - GEN
T1 - Prediction of Hospital Readmission for Heart Disease
T2 - 7th International Conference for Smart Health, ICSH 2019
AU - Da, Jingwei
AU - Yan, Danni
AU - Zhou, Sijia
AU - Liu, Yidi
AU - Li, Xin
AU - Shi, Yani
AU - Yan, Jiaqi
AU - Wang, Zhongmin
N1 - Publisher Copyright:
© 2019, Springer Nature Switzerland AG.
PY - 2019
Y1 - 2019
N2 - Hospital readmissions consume large amounts of medical resources and negatively impact the healthcare system. Predicting the readmission rate early one can alleviate the financial and medical consequences. Most related studies only select the patient’s structural features or text features for modeling analysis, which offer an incomplete picture of the patient. Based on structured data (including demographic data, clinical data, administrative data) and medical record text, this paper uses deep learning methods to construct an optimal model for hospital readmission prediction, tested on a dataset of heart disease patients’ 30-day readmission. The results show that when only structured data is used, the deep learning model is much better than the Naive Bayes model and slightly better than the Support Vector Machine model. Adding a text model to the deep learning model improves performance, increasing accuracy and F1-score by 2% and 6%, respectively. This indicates that textual information contributes greatly to hospital readmission predictions.
AB - Hospital readmissions consume large amounts of medical resources and negatively impact the healthcare system. Predicting the readmission rate early one can alleviate the financial and medical consequences. Most related studies only select the patient’s structural features or text features for modeling analysis, which offer an incomplete picture of the patient. Based on structured data (including demographic data, clinical data, administrative data) and medical record text, this paper uses deep learning methods to construct an optimal model for hospital readmission prediction, tested on a dataset of heart disease patients’ 30-day readmission. The results show that when only structured data is used, the deep learning model is much better than the Naive Bayes model and slightly better than the Support Vector Machine model. Adding a text model to the deep learning model improves performance, increasing accuracy and F1-score by 2% and 6%, respectively. This indicates that textual information contributes greatly to hospital readmission predictions.
KW - Deep learning
KW - Hospital readmission
KW - Predictive analytics
UR - https://www.scopus.com/pages/publications/85076748322
U2 - 10.1007/978-3-030-34482-5_2
DO - 10.1007/978-3-030-34482-5_2
M3 - Conference article published in proceeding or book
AN - SCOPUS:85076748322
SN - 9783030344818
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 16
EP - 26
BT - Smart Health - International Conference, ICSH 2019, Proceedings
A2 - Chen, Hsinchun
A2 - Zeng, Daniel
A2 - Yan, Xiangbin
A2 - Xing, Chunxiao
PB - Springer
Y2 - 1 July 2019 through 2 July 2019
ER -