Table 5.

Test-set performance of the best model and baseline logistic regression model for each outcome, displayed as mean ± std.dev. Statistically significant differences between best- and baseline models in terms of ROC-AUC and Average Precision are marked as * (Mann–Whitney U test, alpha = 0.05).

Outcome	Model	ROC-AUC	p-value	Average Prec	p-value	Accuracy	Sensitivity	Specificity	PPV	NPV
mRS > 2	QDA	0.87 ± 0.03*	p < 0.0001	0.60 ± 0.13*	p < 0.0001	0.72 ± 0.06	0.83 ± 0.08	0.71 ± 0.07	0.24 ± 0.04	0.98 ± 0.01
	Baseline LR	0.77 ± 0.05		0.40 ± 0.08		0.79 ± 0.05	0.51 ± 0.14	0.82 ± 0.06	0.24 ± 0.07	0.94 ± 0.01
mRS-Diff. > 1	MLP	0.70 ± 0.02*	p < 0.0001	0.19 ± 0.05	p = 0.2561	0.52 ± 0.03	0.74 ± 0.07	0.50 ± 0.04	0.13 ± 0.00	0.95 ± 0.01
	Baseline LR	0.65 ± 0.06		0.19 ± 0.06		0.66 ± 0.07	0.50 ± 0.16	0.67 ± 0.08	0.14 ± 0.03	0.93 ± 0.02
perm. nND	QDA	0.71 ± 0.04*	p < 0.0001	0.26 ± 0.08*	p < 0.0001	0.60 ± 0.10	0.65 ± 0.24	0.60 ± 0.12	0.11 ± 0.02	0.96 ± 0.02
	Baseline LR	0.49 ± 0.09		0.08 ± 0.02		0.69 ± 0.07	0.19 ± 0.16	0.73 ± 0.08	0.05 ± 0.04	0.92 ± 0.01
trans. nND	SVM	0.73 ± 0.07*	p < 0.0001	0.15 ± 0.05*	p = 0.0116	0.90 ± 0.03	0.00 ± 0.02	0.97 ± 0.03	0.22 ± 0.41	0.93 ± 0.00
	Baseline LR	0.63 ± 0.11		0.19 ± 0.10		0.74 ± 0.05	0.41 ± 0.19	0.77 ± 0.08	0.12 ± 0.06	0.95 ± 0.02
GOS < 5	GAM	0.79 ± 0.08*	p < 0.0001	0.45 ± 0.09	p = 0.0879	0.73 ± 0.05	0.69 ± 0.12	0.73 ± 0.06	0.30 ± 0.05	0.93 ± 0.02
	Baseline LR	0.75 ± 0.04		0.43 ± 0.09		0.74 ± 0.05	0.57 ± 0.13	0.77 ± 0.06	0.30 ± 0.06	0.92 ± 0.02

The QDA and GAM models for mRS > 2, permanent nND and GOS < 5 perform best in terms of Average Precision, too. mRS = modified Rankin Scale, GOS =  Glasgow outcome scale, nND = new neurological deficit, LR = logistic regression, QDA = quadratic discriminant analysis, MLP = multilayer perceptron, SVM = support vector machine, GAM = generalized additive model, ROC-AUC = area under receiver operating characteristic curve, PPV = positive predictive value, NPV = negative predictive value.