Table 4.

Different MRI gradings of TAI in various TBI injury severities: Adjusted^a ordinal and logistic regression analyses and cross-validated area under the curve (AUC)

				Model fit logistic regression modelb			Cross-validated AUCc
		Model	Variables included in the model	Pseudo R2	AIC	BIC	Mean AUC	Bootstrap 95% CI
Severe TBI	GCS score ≤ 8 n = 176	A0	Core variables and CT	0.15	585	626	0.80	0.68; 0.84
		A1	Standard TAI grading	0.16	581	632	0.81	0.73; 0.87
		A2	Stockholm MRI grading system	0.18	571	621	0.80	0.71; 0.86
		A3	TAI-MRI grading based on GCS score	0.17	577	627	0.81	0.71; 0.85
		A4	Trondheim TAI-MRI grading	0.19	571	628	0.84	0.72; 0.86
		A5	Vol. TAI on FLAIR (TAI-FLAIR volume model)	0.20	555	602	0.86	0.76; 0.90
		A6	Vol. TAI on FLAIR+ Vol. Contusion on FLAIR (TBI-FLAIR volume model)	0.21	551	605	0.90	0.78; 0.91
Moderate TBI	GCS score 9–12 n = 74	B0	Age, GCS score, and CT	0.15	214	241	0.76	0.69; 0.89
		B1	Standard TAI grading	0.16	219	253	0.69	0.64; 0.86
		B2	Stockholm MRI grading system	0.15	219	254	0.76	0.65; 0.86
		B3	TAI-MRI grading based on GCS score	0.16	219	253	0.73	0.63; 0.86
		B4	Trondheim TAI-MRI grading	0.17	219	258	0.77	0.63; 0.86
		B5	Vol. TAI on FLAIR (TAI-FLAIR volume model)	0.17	215	247	0.78	0.64; 0.86
		B6	Vol. TAI on FLAIR + Vol. Contusion on FLAIR (TBI-FLAIR volume model)	0.19	213	250	0.76	0.61; 0.84
	GCS score 13 n = 55	C0	Age and CT	0.15	67	73	0.66	0.51; 0.83
		C1	Standard TAI grading	0.15	69	77	0.72	0.50; 0.83
		C2	Stockholm MRI grading system	0.15	69	77	0.71	0.50; 0.83
		C3	TAI-MRI grading based on GCS score	0.16	69	77	0.76	0.47; 0.81
		C4	Trondheim TAI-MRI grading	0.15	69	77	0.73	0.51; 0.83
		C5	Vol. TAI on FLAIR (TAI-FLAIR volume model)	0.16	71	81	0.75	0.52; 0.85
		C6	Vol. Contusion on FLAIR (Contusion-FLAIR volume model)	0.26	63	73	0.78	0.55; 0.90
Mild TBI	GCS 14–15 n = 147	D0	Age and sex	0.05	102	111	NA
		D4	TAI gradingd on MRI	0.09	100	112
		D5	Vol. TAI on FLAIR (TAI-FLAIR quantitative model)	0.09	102	117
		D6	Vol. Contusion on FLAIR (Contusion-FLAIR volume model)	0.25	85	100

TAI traumatic axonal injury, TBI traumatic brain injury, AUC area under the curve, GCS Glasgow Coma Scale, pseudo R² McFaddens pseudo R², AIC Akaike information criterion, BIC Bayesian information criterion, CI confidence interval, Vol. volume, FLAIR fluid-attenuated inversion recovery, NA not applicable

^a Severe TBI is adjusted for core variables (age, pupil abnormalities and GCS score) and the worst Marshall CT score. Moderate TBI with GCS scores 9–12 are adjusted for age, GCS score and worst Marshall CT score. Moderate TBI with a GCS score 13 is adjusted for age and Marshall CT score. Mild TBI are adjusted for age and sex

For details on the contents and specifications for the different gradings, we refer to the Fig. 4

^b Ordinal logistic regression analyses with inverted 6 months Glasgow Outcome Scale Extended (GOSE) score as response variable for severe TBI and moderate TBI with GCS score 9–12. Binary logistic regression predicting disability (GOSE score ≤ 6) at 3 months for mild TBI and at 6 months for moderate TBI with GCS score 13

^c 10-fold cross-validation AUC analyses based on the logistic regression models but predicting poor outcome (GOSE score ≤ 4) for severe TBI and predicting disability (GOSE score ≤ 6) for moderate TBI. All values are given as mean AUC with 95% CI from the bootstrap samples

The most favourable model in each injury severity is indicated with italicised rows. The most favourable value in each category for each injury severity is in bold

^d Since patients with mild TBI only had TAI in the hemispheres or corpus callosum, all the clinical grading systems performed identically in this group