Table 5.

Predictive performances of different models in training set and validation set

	Accuracy (95% CI)	Sensitivity (95% CI)	Specificity (95% CI)	AUC (95% CI)
Training set
ANN	0.866 (0.838–0.894)	0.850 (0.821–0.879)	0.410 (0.370–0.450)	0.873 (0.846–0.900)
Logistic regression	0.711 (0.674–0.748)	0.662 (0.624–0.700)	0.337 (0.299–0.375)	0.720 (0.684–0.756)
APACHEII	0.615 (0.576–0.654)	0.569 (0.529–0.609)	0.367 (0.328–0.406)	0.629 (0.607–0.651)
SOFA	0.574 (0.534–0.614)	0.619 (0.580–0.658)	0.413 (0.373–0.453)	0.619 (0.596–0.641)
P value	< 0.001	< 0.001	0.029	< 0.001
Validation set
ANN	0.735 (0.714–0.756)	0.624 (0.601–0.647)	0.772 (0.752–0.792)	0.811 (0.792–0.830)
Logistic regression	0.722 (0.701–0.743)	0.604 (0.581–0.627)	0.744 (0.723–0.765)	0.752 (0.731–0.773)
APACHEII	0.401 (0.378–0.424)	0.333 (0.311–0.355)	0.841 (0.824–0.858)	0.607 (0.584–0.630)
SOFA	0.609 (0.586–0.632)	0.416 (0.392–0.440)	0.788 (0.769–0.807)	0.628 (0.605–0.651)
P value	0.272	0.197	0.095	0.002

ANN artificial neural networks, SOFA sequential organ failure assessment, APACHE acute physiology and chronic health evaluation, AUC area under the ROC curve, CI confidential interval