Table 2.

Performance metrics from the tenfold cross-validation of individual EEG feature-based models for delirium prediction in critically ill patients. EEG variables were selected and ranked according to their predictive value using the Information Gain scoring method. The reported metrics include AUC, accuracy, precision, sensitivity, and specificity for each EEG variable analyzed. Abbreviations: AUC, area under the curve; EEG, electroencephalography; PDR, posterior dominant rhythm; PDA, polymorphic delta activity; RPPs, rhythmic or periodic patterns; SEDs, sporadic epileptiform discharges; SVM, support vector machine

Variable	Model	AUC	Accuracy	Precision	Sensitivity	Specificity
Absence of reactivity	Random Forest	0.488	0.543	0.000	0.000	0.974
Reduction in alpha activity	SVM	0.483	0.557	0.000	0.000	1.000
Burst-suppression/burst-attenuation	SVM	0.463	0.557	0.000	0.000	1.000
PDA	Naïve Bayes	0.529	0.486	0.143	0.032	0.846
SEDs	Logistic Regression	0.525	0.557	0.000	0.000	1.000
Absence of PDR	Naïve Bayes	0.592	0.514	0.440	0.355	0.641
Low voltage (< 20 μV)	Naïve Bayes	0.533	0.586	0.667	0.129	0.949
Predominant background frequency of less than 4 Hz	Naïve Bayes	0.600	0.614	0.591	0.419	0.769
RPPs	Naïve Bayes	0.558	0.600	0.667	0.194	0.923
Predominant background frequency of 4 to 7 Hz	Naïve Bayes	0.667	0.643	0.565	0.839	0.487