• Jiun-Chi Huang, Yi-Chun Tsai, Pei-Yu Wu, Yu-Hui Lien, Chih-Yi Chien, Chih-Feng Kuo, Jeng-Fung Hung, Szu-Chia Chen, and Chao-Hung Kuo.
• Department of Internal Medicine, Kaohsiung Municipal Siaogang Hospital, Kaohsiung Medical University, Kaohsiung, Taiwan; Division of Nephrology, Department of Internal Medicine, Kaohsiung Medical University Hospital, Kaohsiung Medical University, Kaohsiung, Taiwan; Faculty of Medicine, College of Medicine, Kaohsiung Medical University, Kaohsiung, Taiwan; Graduate Institute of Clinical Medicine, College of Medicine, Kaohsiung Medical University, Kaohsiung, Taiwan.
• Comput Methods Programs Biomed. 2020 Oct 1; 195: 105536.

BackgroundIntradialytic hypotension (IDH) is commonly occurred and links to higher mortality among patients undergoing hemodialysis (HD). Its early prediction and prevention will dramatically improve the quality of life. However, predicting the occurrence of IDH clinically is not simple. The aims of this study are to develop an intelligent system with capability of predicting blood pressure (BP) during HD, and to further compare different machine learning algorithms for next systolic BP (SBP) prediction.MethodsThis study presented comprehensive comparisons among linear regression model, least absolute shrinkage and selection operator (LASSO), tree-based ensemble machine learning models (random forest [RF] and extreme gradient boosting [XGBoost]), and support vector regression to predict the BP during HD treatment based on 200 and 48 maintenance HD patients containing a total of 7,180 and 2,065 BP records for the training and test dataset, respectively. Ensemble method also was computed to obtain better predictive performance. We compared the developed models based on R2, root mean square error (RMSE) and mean absolute error (MAE).ResultsWe found that RF (R2=0.95, RMSE=6.64, MAE=4.90) and XGBoost (R2=1.00, RMSE=1.83, MAE=1.29) had comparable predictive performance on the training dataset. However, RF (R2=0.49, RMSE=16.24, MAE=12.14) had more accurate than XGBoost (R2=0.41, RMSE=17.65, MAE=13.47) on testing dataset. Among these models, the ensemble method (R2=0.50, RMSE=16.01, MAE=11.97) had the best performance on testing dataset for next SBP prediction.ConclusionsWe compared five machine learning and an ensemble method for next SBP prediction. Among all studied algorithms, the RF and the ensemble method have the better predictive performance. The prediction models using ensemble method for intradialytic BP profiling may be able to assist the HD staff or physicians in individualized care and prompt intervention for patients' safety and improve care of HD patients.Copyright © 2020 Elsevier B.V. All rights reserved.

11 keywords...

hide…