Table 2.

Performance of the TBI prognostic model.

	All candidate variables			Excluding non-robust variables			Excluding non-robust & counterintuitive variables
Sample set	Training	Validation	Test	Training	Validation	Test	Training	Validation	Test
AUC (SD)	0.9372 (0.0236)	0.7822 (0.0126)	0.8094	0.9080 (0.0249)	0.7877 (0.0177)	0.8046	0.8912 (0.0252)	0.7836 (0.0189)	0.8085
Accuracy (SD)	0.8522 (0.0327)	0.7500 (0.0169)	0.7536	0.8165 (0.0317)	0.7484 (0.0250)	0.7440	0.8053 (0.0285)	0.7451 (0.0255)	0.7488
F1 score (SD)	0.8281 (0.0360)	0.7129 (0.0190)	0.7052	0.7855 (0.0344)	0.7104 (0.0299)	0.6864	0.7740 (0.0375)	0.7076 (0.0315)	0.7045
Sensitivity (SD)	0.8477 (0.0305)	0.7434 (0.0489)	0.7011	0.8008 (0.0319)	0.7394 (0.0527)	0.6667	0.8018 (0.0637)	0.7393 (0.0570)	0.7126
Specificity (SD)	0.8554 (0.0440)	0.7549 (0.0456)	0.7917	0.8279 (0.0450)	0.7549 (0.0439)	0.8000	0.8078 (0.0238)	0.7494 (0.0443)	0.7750
Precision (SD)	0.8106 (0.0529)	0.6880 (0.0330)	0.7093	0.7722 (0.0498)	0.6862 (0.0329)	0.7073	0.7500 (0.0252)	0.6813 (0.0329)	0.6966

Performance of the TBI prognostic model trained using all candidate variables, only robust variables, and robust and clinically validated variables. Standard deviation (SD) is calculated over 5 cross-validation folds.