Learning to predict in-hospital mortality risk in the intensive care unit with attention-based temporal convolution network

Table 4 The performances of different ML models for prediction of in-hospital mortality in the test dataset

Methods	Sens	Spec	F1 score	Brier score	AUCROC	AUC-PR
Non-time series methods
DT	22.7%	96.9%	0.28	0.088	0.804(0.789–0.817)	0.381
LR	35.0%	96.8%	0.43	0.081	0.838(0.824–0.850)	0.459
RF	25.1%	98.5%	0.36	0.077	0.865(0.853–0.877)	0.511
SVM	29.1%	97.9%	0.39	0.080	0.822(0.808–0.835)	0.477
SAPS-II¹					0.777	0.376
APS-III¹					0.750	0.357
OASIS¹					0.760	0.312
Time series methods
LSTM²	46.1%					0.451
Attention-based TCN	67.1%	82.6%	0.46	0.142	0.837(0.824–0.850)	0.454

Statistical quantifications were demonstrated with 95% CI, when applicable. ML machine learning, attention-based TCN attention-based Temporal Convolution Network, LR Logistic Regression, SVM Support Vector Machine, SAPS Simplified Acute Physiology Score, APS Acute Physiology Score, OASIS Oxford Acute Severity of Illness Score, ¹, data referring to Hrayr et al. Scientific Data.2017; ², data referring to Ruo-xi Yu, et al. IEEE J Biomed Health Inform.2019

ISSN: 1471-2253