Tab. 2.

Summary of the results for models fitted for subsamples based on hospitals.

Model	Variables	Hosp. excluded	Sample Size	AUC	AUC nested CV	AUC of left- out data	Ssize left-out
1	Hem, morph, gender, age, angioarch	mayo (missing data)	245	0.8359	0.7547	-	-
2	Hem, morph, angioarch	-	289	0.7920	0.7297	-	-
2a	Hem, morph, angioarch	mayo	250	0.7679	0.6936	0.5842	39
2b	Hem, morph, angioarch	sinai	217	0.7928	0.7361	0.7081	72
2c	Hem, morph, angioarch	inova	118	0.7933	-	0.7507	168

Hem = hemodynamic variables, morph = morphological variables, angioarch = angioarchitectures. The column "AUC nested CV" refers to the AUC of the model after subtracting the estimated optimism by nested cross-validation. The sample size (Ssize) of the left-out data used for evaluation of the models by the split-sample approach is given in the last column