Prognostic Value of Magnetic Resonance Imaging in Moderate and Severe Head Injury: A Prospective Study of Early MRI Findings and One-Year Outcome

Abstract

The clinical benefit of early magnetic resonance imaging (MRI) in severe and moderate head injury is unclear. We sought to explore the prognostic value of the depth of lesions depicted with early MRI, and also to describe the prevalence and impact of traumatic brainstem lesions. In a cohort of 159 consecutive patients with moderate to severe head injury (age 5–65 years and surviving the acute phase) admitted to a regional level 1 trauma center, 106 (67%) were examined with MRI within 4 weeks post-injury. Depth of lesions in MRI was categorized as: hemisphere level, central level, and brainstem injury (BSI). The outcome measure was Glasgow Outcome Scale Extended (GOSE) 12 months post-injury. Forty-six percent of patients with severe injuries and 14% of patients with moderate injuries had BSI. In severe head injury, central or brainstem lesions in MRI, together with higher Rotterdam CT score, pupillary dilation, and secondary adverse events were significantly associated with a worse outcome in age-adjusted analyses. Bilateral BSI was strongly associated with a poor outcome in severe injury, with positive and negative predictive values of 0.86 and 0.88, respectively. In moderate injury, only age was significantly associated with outcome in multivariable analyses. Limitations of the current study include lack of blinded outcome evaluations and insufficient statistical power to assess the added prognostic value of MRI when combined with clinical information. We conclude that in patients with severe head injury surviving the acute phase, depth of lesion on the MRI was associated with outcome, and in particular, bilateral brainstem injury was strongly associated with poor outcomes. In moderate head injury, surprisingly, there was no association between MRI findings and outcome when using the GOSE score as outcome measure.

Introduction

M agnetic resonance imaging (MRI) is a sensitive method for the depiction of traumatic lesions in the brain parenchyma (Parizel et al., 2005). Nevertheless, the clinical usefulness of MRI examination in traumatic brain injury (TBI) has not yet been clearly established (Weiss et al., 2007).

The association between early MRI findings and outcome has been addressed in several studies, and the total burden of lesions as well as the depth of lesions in the brain have been related to outcome (Chastain et al., 2009; Firsching et al., 2001; Kampfl et al., 1998; Lagares et al., 2009; Levin et al., 1997; Pierallini et al., 2000; Wedekind et al., 2002; Yanagawa et al., 2000). Some of these earlier MRI studies are, however, associated with some limitations. First, there is a lack of cohort studies in which MRI has been used in consecutive head injury patients. Studies of MRI in TBI have often been performed in selected series (Kampfl et al., 1998), and the prevalence and implications of the findings may not apply to patients in other settings. Some studies were also performed before the introduction of the most sensitive MRI sequences that are now common in clinical protocols (Firsching et al., 2001; Kampfl et al., 1998; Levin et al., 1997). Furthermore, most studies have been restricted to prediction of poor outcome (Chastain et al., 2009), and some were performed in series with high mortality (Firsching et al., 2001). In typical diagnostic settings, however, MRI is performed in stabilized patients with high chances of survival. Finally, few have studied the value of MRI and established prognostic factors, such as age, injury severity, pupillary reactivity, and computed tomographic (CT) characteristics in multivariable analyses (Murray et al., 2007).

Recently, we studied diffuse axonal injury (DAI) in a patient cohort of 106 patients with moderate and severe head injury, and found that DAI was only associated with a worse outcome when located in the brainstem (Skandsen et al., 2010). However, other types of traumatic lesions than DAI are seen on early MRI scans, and the aim of the present study was to explore the impact and depth distribution of intracranial traumatic lesions, regardless of their primary mechanism. In the same cohort of 106 patients with moderate and severe head injury we further explored the possible prognostic value of early MRI scans, and describe the prevalence and outcome of different types of brainstem injury (BSI).

Patients and Methods

Subjects

We evaluated for inclusion all patients aged 5–65 years, with moderate-to-severe head injury according to the Head Injury Severity Scale (HISS) criteria (Stein and Spettell, 1995), admitted between October 2004 and August 2008 to the neurosurgical department at St. Olavs University Hospital, Trondheim, Norway. Patients who were deemed clinically unsalvageable on admission and for whom no further treatment was performed, or patients who died of other injuries during the initial 24 h were excluded. Otherwise there were no exclusion criteria. The hospital is the only level 1 trauma center in a region of 660 000 inhabitants.

The MRI sample consisted of 106 individuals who were examined with MRI within 4 weeks (median 8 days) post-injury. This sample represents 67% of the 159 patients aged 5–65 years with moderate and severe head injury who were discharged alive from the hospital during the study period. Details about patients with and without MRI and the reasons for not being examined have been reported previously (Skandsen et al., 2010). Patients not examined with early MRI were older and more often experienced pre-injury conditions such as substance abuse or central nervous system disorders. No significant differences were found for injury-related variables or 12-month outcomes (median Glasgow Outcome Scale Extended [GOSE] score) between unexamined patients and patients examined with early MRI.

Magnetic resonance imaging

MR imaging was performed at the study hospital at 1.5 Tesla (Siemens Symphony or Siemens Avanto; Siemens Medical, Erlangen, Germany). Six patients were examined with a similar MR system in a neighboring hospital. The scan protocol consisted of sagittal TSE T2-weighted imaging, sagittal, transverse, and coronal T2 fluid-attenuated inversion recovery (FLAIR) imaging, transverse T2*-weighted gradient echo imaging (GRE), transverse SE T1-weighted imaging, and diffusion weighted imaging (DWI). MRI parameters have been reported previously (Skandsen et al., 2009).

Two neuroradiologists (K.A.K., rater A; and M.F., rater B) reported the imaging findings based on visual inspection. Disagreement was solved by consensus. Variables were predefined and designed based on previous findings in the literature (Firsching et al., 2001; Gentry, 2002; Levin et al., 1988; Mannion et al., 2007; Ommaya and Gennarelli, 1974): (1) deepest level of lesions depicted (Table 1 and Fig. 1), (2) brainstem injury as unilateral or bilateral, and (3) cortical contusions.

FIG. 1.

(A) Axial FLAIR MR image demonstrating signal intensity changes in the right frontal and left temporal lobes, interpreted as cortical contusions (hemisphere level). (B) Coronal FLAIR MR image demonstrating cortical contusions in the right parietal and temporal lobe, as well as signal intensity changes in the corpus callosum (central level). (C) Axial FLAIR MR image demonstrating signal intensity changes in the brainstem, unilaterally (arrow). (D) Axial FLAIR MR image demonstrating signal intensity changes brainstem, bilaterally (arrows; FLAIR, fluid-attenuated inversion recovery; MR, magnetic resonance).

Table 1.

MRI-Derived Depth of Lesion

Level	Definition
Hemispheres	Traumatic lesions^a confined to the cerebral cortex or lobar white matter
Central	Traumatic lesions in the corpus callosum, thalami, or basal ganglia
Unilateral brainstem	Traumatic lesions in the brainstem unilaterally^b
Bilateral brainstem	Traumatic lesions in the brainstem bilaterally^b

Signal loss compatible with micro-bleeds (in the GRE sequence), or increased signal intensity compatible with tissue edema (in the FLAIR sequences).

Traumatic lesions at a higher level may also be present.

FLAIR, fluid-attenuated inversion recovery MRI, magnetic resonance imaging.

For evaluation of reliability, 31 cases were drawn evenly and blindedly from the list of cases. These were scored by a third neuroradiologist (rater C, J.R.), who was blinded to clinical information and previous classification. Interrater agreement was calculated based on depth of lesion as described in Table 1.

CT imaging

The CT scans were reviewed by a radiologist (I.H.S.) and classified by the worst of the evaluated CT scans according to the Rotterdam CT classification. The score is based on the presence of CT characteristics found to be prognostic in TBI, such as traumatic subarachnoid hemorrhage and/or intraventricular hemorrhage, compression of basal cisterns, midline shift, and absence of epidural mass lesion (Maas et al., 2005).

Injury-related variables

Injury-related variables were mechanism of injury, Glasgow Coma Scale (GCS) score (Teasdale and Jennett, 1974; recorded as lowest observed score observed at or after admittance, or before intubation in case of pre-hospital intubation), and noted as moderate (GCS score of 9–13) or severe head injury (GCS score of ≤8), hypoxic (oxygen saturation <92%) or hypotensive events (systolic blood pressure <90 mm Hg), before or at admittance to hospital, and any pupillary dilation (unilateral or bilateral). GCS score may be falsely low in cases of intoxication or major non-head trauma (Stocchetti et al., 2004). Accordingly, patients whose level of consciousness had returned to normal during the first 24 h, and there were other factors that could explain the initially low GCS score, were classified as mild head injury and excluded. Likewise, if the patient GCS score rose to 9–13 during the first 24 h, and other factors likely could explain the initially low GCS score, the head injury was downgraded from initially severe to moderate. Global outcome was evaluated by direct assessment or telephone interview 12 months post-injury using the structured interview for GOSE (Jennett and Bond, 1975; Jennett et al., 1981; Wilson et al., 1998). In order to reduce the potential error associated with the telephone setting, relatives or caregivers also provided information in two-thirds of the cases, and the assessor made the judgment of the best source of information to use. One of the patients was lost to 12-month follow-up, and 6-month outcome was therefore used instead. All other participants could be located and evaluated for outcome, except one who died from a new trauma a short time after the head injury.

Statistical analysis

Patient demographics and injury characteristics are presented as percentages, mean with standard deviations (SD), or median with interquartile range (IQR, 25–75%). Pearson's chi-squared test was used for comparison of proportions. The CT score, regarded as ordinal data, was treated as interval data rather than discrete levels in the regression analyses.

Relationship between possible prognostic factors and outcome in the upper range of the GOSE was explored using binary as well as proportional odds ordinal multiple logistic regression analyses. The latter was used when there were more than two outcome categories of interest. The dependent variable was GOSE score dichotomized into “good recovery” or “less-than-good recovery” (GOSE score ≤6) for binary multiple logistic regression analyses. For ordinal multiple logistic regression analyses, GOSE was reclassified into a variable of three categories: good recovery (GOSE score 7 and 8), moderate disability (GOSE score 5 and 6), or poor outcome (GOSE score 1–4).

In the binary logistic regression analysis, the number of covariates was limited to a maximum of one-fifth of the number of events in the smallest group (Vittinghoff and McCulloch, 2007).

For analysis of interrater reliability, we used the linear weighted Kappa statistic to determine consistency among raters. We adopted Altman's guidelines in which κ-values below .20 are considered poor, .21–.40 fair, .41–.60 moderate, .61–.80 good, and above .81 very good (Altman, 1999).

Available case analysis was used for the analyses of variables with missing data, which never exceeded five cases. Ninety-five percent confidence intervals (CI) are given where relevant. A two-sided p value <0.05 was regarded significant. Statistical analyses were performed using SPSS version 16.0 (SPSS Inc.,, Chicago, IL) and StatXact 8 (Cytel Inc., Cambridge, MA).

Ethics

The Regional Committee for Medical Research Ethics and the Norwegian Social Science Data Services (NSD) approved the study. For surviving patients, written consent was obtained from the patient, or their next of kin for individuals incapacitated or below the age of 16.

Results

Patients characteristics

Table 2 presents the demographic and injury-related characteristics and outcomes of the patients. Distribution of GOSE scores have been reported previously (Skandsen et al., 2010).

Table 2.

Patient Characteristics and Outcome

		No. of cases (%) ^a
Variable	Value	All (n=106)	Moderate (n=57)	Severe (n=49)
Age (years)	Mean (SD)	28 (15.9)	30 (16.2)	26 (15.4)
	<35 years	75 (71)	39 (68)	36 (74)
Sex	Male	80 (75)	42 (74)	38 (78)
Preinjury problems^b	Yes	10 (9)	6 (11)	4 (8)
Mechanism of injury	Traffic accident	56 (53)	25 (44)	31 (63)
	Fall	37 (35)	20 (35)	17 (35)
	Other	13 (12)	12 (21)	1 (2)
GCS score	(median, IQR^c)	9 [6–12]	12 [10–13]	6 [4–7]
Hypoxia	Yes	14 (13)	4 (7)	10 (20)
Hypotension	Yes	12 (11)	5 (9)	7 (14)
Pupillary dilation	Yes	12 (11)	1 (2)	11 (22)
Rotterdam CT score (worst)	1	4 (4)	2 (4)	2 (4)
	2	38 (36)	27 (47)	11 (22)
	3	39 (37)	23 (40)	16 (33)
	4	10 (9)	3 (5)	7 (14)
	5	11 (10)	2 (4)	9 (18)
	6	4 (4)	0	4 (8)
GOSE score^d	(median, IQR)	7 [5–8]	8 [6.5–8]	6 [3.5–8]

Unless otherwise stated.

Preinjury problems were present if substance abuse, or neurological or psychiatric conditions affected daily functioning at the time of injury.

Interquartile range (IQR).

GOSE score for 105 patients.

GCS, Glasgow Coma Scale; CT, computed tomography; GOSE Glasgow Outcome Scale Extended.

Interrater reliability

The interrater reliability of the classification of depth of lesion in 31 cases was in the very good range, with a weighted linear Kappa of 0.90 (95% CI 0.78,1.00; Table 3).

Table 3.

Interrater Agreement in Evaluating Depth of Lesion

	Rater A and B ^a
Rater C ^b	Hemispheres	Central	Brainstem unilateral	Brainstem bilateral	Total rater C
Hemispheres	12	2	0	0	14
Central	0	6	0	0	6
Brainstem unilateral	1	0	4	0	5
Brainstem bilateral	0	0	0	6	6
Total rater A and B	13	8	4	6	31

Rater A and B: Two neuroradiologists who had access to clinical information, and who solved disagreement by consensus.

Rater C: A third neuroradiologist who evaluated the scans blinded to clinical information.

Depth of lesion evaluated with MRI

Brainstem lesions were seen in 30 patients; in 22 cases (45%) of severe head injury, and 8 cases (14%) of moderate head injury (p<0.001). Fifteen patients had unilateral brainstem injury, all with concomitant lesions supratentorially described as DAI. Fifteen patients had bilateral brainstem injury. Of these, 14 cases were diagnosed as DAI, while 1 patient had bilateral ischemic brainstem injury secondary to compression.

One hundred and three (97%) of the patients had parenchymal lesions on the MRI scan. Of the remaining patients, two had only supratentorial epidural hematomas, and one had a normal MRI. The three patients without parenchymal lesions were assigned to the hemisphere category, which then comprised 47 patients. In 29 patients, corpus callosum, thalamus, or basal ganglia was the deepest level of injury (central level), and 27 of these had lesions in the corpus callosum. Nineteen patients had lesions in the thalami and/or basal ganglia. Cerebellar lesions were found in 12 patients. All patients with cerebellar lesions also had central or brainstem lesions and were classified accordingly. Table 4 presents the depth of lesion in moderate and severe patients separately.

Table 4.

Outcome in Relation to Depth of Lesion and Type of Brainstem Injury

	No. of cases (%)
	Moderate head injury				Severe head injury
		GOSE ^a score	GOSE score	GOSE score		GOSE score	GOSE score	GOSE score
Depth of lesion	n	7–8	5–6	1–4	n	7–8	5–6	1–4
Total	57	42 (74)	14 (25)	1 (2)	48	18 (38)	14 (29)	16 (33)
Hemispheres	34	25 (74)	9 (27)	0	12	8 (67)	4 (32)	0
Central	15	12 (80)	3 (20)	0	14	5 (36)	7 (50)	2 (14)
Brainstem, unilateral	7	5 (71)	1 (14)	1 (14)	8	4 (50)	2 (25)	2 (25)
Brainstem, bilateral^b	1	0	1	0	14	1 (7)	1 (7)	12 (86)

GOSE score for 105 patients.

13 cases of DAI, one case was secondary.

GOSE; Glasgow Outcome Scale Extended; DAI; diffuse axonal injury.

Prediction of outcome in moderate and severe patients

The data indicated that the predictor variable “depth of lesion” might have a different effect on outcome in the moderate and severe group (Table 4). In ordinal logistic regression, an interaction term (depth of lesion * HISS category; that is, moderate or severe head injury) was significantly associated with outcome (p=0.041). Thus we performed separate analyses for the moderate and severe patients.

In the moderate head injury patients, we used binary logistic regression to predict a less-than-good recovery, since all outcomes observed, except one, were in the favorable range (98%). Higher CT score and higher age were associated with a less-than-good recovery (GOSE score 1–6) in the univariable analyses. In multivariable analyses, only age was independently associated with outcome (Table 5). Ordinal logistic regression, with the GOSE score recoded into upper good recovery (GOSE score 8), lower good recovery or upper moderate disability (GOSE score 6 and 7), and lower moderate disability or poor outcome (GOSE ≤5), yielded practically the same results (not reported here).

Table 5.

Prediction of a Less-Than-Good Recovery in Patients with Moderate Head Injury: Binary Logistic Regression

Variable	Value	n	Good recovery ^{a b}	OR_crude	OR_adj	p Value (OR_adj)
Total		57	42 (74)
Age continuous				1.10 (1.05–1.16)
Pupils	Normal	55	40 (73)
	Dilation^c	1	1	n.a.	n.a.
Secondary events^d	Absent	51	37 (73)	(Reference)	(Reference)
	Present	6	5 (83)	0.53 (0.06–4.9)	0.08 (0.004–1.3)	0.076
GCS score	12–15	33	25(76)	(Reference)	(Reference)
	9–11	23	16 (70)	1.4 (0.41–4.5)	2.0 (0.43–9.3)	0.4
Rotterdam CT score		57		2.5 (1.1–5.6)	1.6 (0.57–4.6)	0.37
Depth of lesion	Hemispheres	33	24 (73)	(Reference)	(Reference)
	Central	15	12 (80)	0.69 (0.16–3.0)	1.4 (0.22–9.6)	0.7
	BSI, unilateral	7	5 (71)	1.1 (0.18–6.8)	1.1 (0.10–11.9)	0.9
	BSI, bilateral	1		n.a.	n.a.

Number of cases (%).

GOSE score 7–8.

Unilaterally or bilaterally.

Hypoxia or hypotension.

OR, odds ratio; adj, adjusted for age; n.a., not available; BSI, brainstem injury; GOSE, Glasgow Outcome Scale Extended.

In severe head injury, associations between outcome and potential prognostic factors were explored by use of ordinal univariable and multivariable logistic regression analyses (Table 6). Age was consistently associated with outcome, and subsequently we performed the other analyses controlling for age. Due to the small sample size, and the categorical character of the MRI variable, multivariable analyses with more variables than age was not warranted. A significant association with a worse outcome was shown for depths of lesion of injury below the hemispheres, and also for the Rotterdam CT score. When central lesions were compared to unilateral brainstem injuries, by altering the reference category, no difference between these two levels was found. The clinical variables “pupillary dilation” and “secondary adverse events” were also significantly associated with a worse outcome.

Table 6.

Association between Prognostic Factors and Outcome in Patients with Severe Injury: Ordinal Logistic Regression

Variable	Value	n	Good recovery n (%)	Moderate disability n (%)	Death or severe disability n (%)	Proportional OR unadjusted	Proportional OR adjusted for age	p Value OR_adjusted
Total		48	18 (38)	14 (29)	16 (33)
Age	Continuous	48				1.05 (1.01–1.09)
Pupils	Normal	37	17 (46)	12 (32)	8 (22)	Reference
	Dilation^a	11	1 (9)	2 (18)	8 (73)	9.4 (2.1–42.0)	11.7 (2.3–58.9)	0.003
Secondary events^b	Absent	33	16 (49)	11 (33)	6 (18)	Reference
	Present	14	2 (14)	3 (21)	9 (64)	7.2 (2.0–26.7)	7.1 (1.8–27.3)	0.004
GCS score	6–8	26	10 (39)	11 (42)	5 (19)	Reference
	3–5	22	8 (36)	3 (14)	11 (50)	2.1 (0.72–6.0)	2.1 (0.71–6.5)	0.17
Rotterdam CT score^c						2.1 (1.3–3.3)	1.8 (1.1–2.9)	0.013
Depth of lesion	Hemispheres	12	8 (67)	4 (32)	0	Reference
	Central	14	5 (38)	7 (50)	2 (14)	3.4 (0.72–16.6)	6.5 (2.2–38.7)	0.040
	BSI unilateral	8	4 (50)	2 (25)	2 (25)	2.8 (0.45–16.8)	7.9 (1.0–62.4)	0.050
	BSI bilateral	14	1 (7)	1 (7)	12 (86)	85.4 (10.7–678.6)	181.6 (16.5–2000)	<0.001

Good recovery: GOSE score 7–8; moderate disability: GOSE score 5–6; death or severe disability: GOSE score 1–4.

Unilaterally or bilaterally.

Hypoxia or hypotension.

Rotterdam CT classification.

OR, odds ratio; BSI, brainstem injury; GOSE, Glasgow Outcome Scale Extended.

Poor outcome and bilateral brainstem injury in severe patients

The distribution of outcome in patients with different levels of injury (Table 4) indicated that bilateral brainstem injury was in particular associated with a poor outcome. Thus we dichotomized the “depth of lesion” variable into having a bilateral brainstem injury or not, and computed positive and negative predictive values of this variable in a contingency table. Of patients with bilateral BSI, 86% experienced a poor outcome (positive predictive value=0.86), in contrast to only 12% of patients without such lesions (negative predictive value=0.88). Thus in this sample, we could compute a sensitivity of 0.75 and a specificity of 0.94 in predicting a poor outcome when bilateral BSI was present.

Poor outcome (GOSE score 1–4) was only seen in the severe patients, and thus this analysis was performed in this subgroup only.

Discussion

We performed early MRI examination in the majority of patients in a neurosurgical cohort of survivors of moderate to severe head injury. Brainstem injury was found in nearly half of the patients with severe injury, but also in some patients with moderate injuries, although almost exclusively unilateral lesions in the latter. We observed a striking difference in outcome between unilateral and bilateral brainstem injury. Presence or absence of a bilateral brainstem injury had a high positive as well as negative predictive value in a prediction of poor outcome. In moderate head injury, low age was a strong predictor of a good recovery, while no other factors could discriminate between a good recovery and disability. In severe injury, there was evidence, albeit weak, that the presence of lesions deeper than the hemispheres, but also higher age, CT score, pupillary dilation, and adverse events of hypoxia or hypotension were associated with a worse outcome.

Brainstem injury

Brainstem injury was found in 46% of the patients with severe injury, but also in patients with moderate injury. Firsching and colleagues performed MRI within 8 days in 102 consecutive patients with a minimum 24 h of coma, and found brainstem lesions in 57% (Firsching et al., 2001). Total mortality was 33% in their study, and the sample included many cases of bilateral pontine injury with 100% mortality. In our study, only two patients eventually died from the head injury, and no patients were in the vegetative state at follow-up. If only survivors from the study of Firsching and associates had been analyzed, seemingly the occurrence of brainstem injury in that study and ours would have been fairly similar. Mannion and co-workers detected brainstem injury in 13 of 46 patients with severe TBI, but this was not a cohort study (Mannion et al., 2007). Lagares and associates found brainstem injury in one-third of their sample of moderate and severe TBI patients, but they did not present data separately for the patients with severe TBI (Lagares et al., 2009). Taken together, we consider that our data together with results from previous studies indicate that nearly half of the patients with severe TBI who survive the acute phase have a brainstem injury.

The unilateral brainstem lesions were always seen in concert with DAI in our study. Thus we did not confirm the findings of Shibata and colleagues, who described a type of superficial BSI in patients without evidence of diffuse supratentorial injury (Shibata et al., 2000). Conversely, our findings are in accord with the autopsy study of Mitchell and Adams, who found damage elsewhere in the brain in all patients with primary brainstem lesions (Mitchell and Adams, 1973).

Patients with moderate head injury

We found only age to be associated with outcome in multivariable analysis in the moderate patients. Thus the value of MRI in moderate head injury was mainly that of documenting the brain injury; as parenchymal traumatic lesions were found in 95%. Even in the 8 patients with brainstem lesions, outcome was more or less the same as that for moderate injuries as a whole, although this finding must be interpreted with caution due to the limited number of patients. There are several possible explanations for this lack of association between MRI and outcome. First, MRI variables that quantify a volume of tissue destruction might have explained a significant proportion of the variance in outcome. There are, however, methodological issues to overcome also with that approach. The signal intensities depicted with FLAIR attenuate with time since injury as tissue edema is absorbed, and this would have been a source of error in a cohort study, for which time from injury to scan cannot be kept constant for clinical reasons. Regarding the gradient echo sequence, a correlation between outcome and number or volume of these has not been consistently shown (Carpentier et al., 2006), and the loss of signal represents paramagnetic phenomena rather than true areas of tissue pathology. Other MRI methods, like diffusion tensor imaging or MRI spectroscopy, have also shown promising results regarding relationship to outcome in selected samples (Garnett et al., 2000; Sidaros et al., 2008). At present, these sequences are not usually included in the clinical MRI protocols, but clearly, there is a need for protocols in the future that evaluate the structure and function of the brain tissue more precisely.

Next, one could argue that in moderate injury, global outcome, as measured with the GOSE, is also influenced by personal or environmental factors. These, however, are typically not described or controlled for in head injury trials or cohorts. More sensitive outcomes measures at the impairment level, like neuropsychological functioning or a measure of symptom complaints, might have been more directly associated with neuroimaging.

Patients with severe head injury

In contrast to what was found in the moderate group, MRI was valuable for outcome prediction in severe head injury. In prediction of poor outcome, the presence or absence of bilateral BSI showed good discriminative ability. This is in line with the study of Weiss and colleagues, who also showed a very good discriminative ability in a model of a pattern of MRI findings and some clinical signs (Weiss et al., 2008). Due to a lack of power to include all relevant clinical variables, we were not able to compare models with and without MRI variables. Since CT depicts the brainstem poorly, these data demonstrate the usefulness of MRI in the acute phase. Consistent with our findings, Lagares and associates in a similar prospective cohort study with 100 moderate and severe head injury patients, found that the additive prognostic information of MRI was most prominent in the more severe patients (Lagares et al., 2009).

Also in contrast to the moderate subgroup, the four-category variable “depth of lesion” was associated with outcome in severe injury, and we found that patients with central lesions, as well as both types of brainstem lesions, had a worse outcome than those with lesions confined to the hemispheres. A similar result was also found by Levin and colleagues in a pediatric sample, but the lesions were classified somewhat differently, and thus their results may not be directly compared to ours (Levin et al., 1997). This finding could appear contradictory to what we found in our recent study of diffuse axonal injury (Skandsen et al., 2010), in which no difference in good recovery was found among patients with or without DAI. Furthermore, we did not find a difference in outcome between DAI stage 1 and 2. However, the fact that the depth of lesion groups are different from the groups of DAI stages, can explain some of the difference. Also, stratification of the sample into moderate and severe injury performed in the present study, revealed that depth of lesion may be important only in severe head injury. Finally, the approach with ordinal regression analysis, exploiting data from several outcome categories, may be a useful approach. Nevertheless, it is important to emphasize that many patients with lesions deeper than the hemispheres recovered well. Furthermore, we could not confirm the findings of Carpentier and associates, that there are several cases with poor outcome despite few findings on the MRI, and where brainstem lesions not detectable with conventional MRI might explain the negative outcome (Carpentier et al., 2006). Only two patients in our study had poor outcomes and no visible brainstem injury, and both had severe contusions with high CT scores and elevated intracranial pressure, in addition to central parenchymal lesions. Thus we consider that early MRI depicts most cases of clinically-relevant BSI.

Strengths and limitations

We examined a large proportion, 67%, of all patients 65 years or younger with MRI. The patients not examined were older and more often had experienced pre-injury problems. Thus we consider the MRI sample to be fairly representative of previously healthy, surviving patients, below 66 years of age, with GCS scores of 3–13. Furthermore, data regarding the patients with severe head injury are population-based, as all severe head injuries are regionalized to our hospital. We achieved a very high participation rate in the database and no loss to follow-up, which further increases the generalizability of the data.

In contrast to most studies of prognosis, our study included mainly patients surviving the injury. This must be kept in mind if comparing the rates of poor outcome to other studies. In daily clinical work, however, MRI is usually performed in stabilized patients who are likely to survive. Thus we think that these data are important, as they extend the knowledge of the consequences of brainstem injury learned from other studies carried out in severe head injury patients with much higher proportions of poor outcome (Carpentier et al., 2006; Firsching et al., 2001; Mannion et al., 2007).

The sample size of 106 patients was fairly large compared to most other MRI studies, but smaller than recommended for prognostic studies in TBI research (Mushkudiani et al., 2008). Moreover, it was unexpected that separate analyses in patients with moderate and severe injuries would be preferable, and this was associated with loss of statistical power. The regression analyses in the subgroups of severe patients must therefore be interpreted with some caution; the confidence intervals are wide, and the odds ratios therefore inaccurate.

The neuroradiologists involved in this study could not be blinded to clinical information, since they also reported findings to the clinicians treating the patients. This may have influenced reliability. In a reliability study of 31 cases, for which the third radiologist was blinded, however, a very good interrater reliability was found. Another limitation is that the outcome assessment was not performed blindedly by personnel without any previous knowledge of the patients.

Conclusion

In this prospective cohort study of patients surviving moderate and severe head injuries, we found that early MRI could increase accuracy of prediction of poor outcome. In fact, the presence or absence of a bilateral brainstem injury alone could predict a poor outcome with high positive and negative predictive value in severe head injury. Furthermore, we found that the presence of central lesions or unilateral brainstem lesions on early MRI was modestly associated with a worse outcome in severe, but not in moderate, head injury. Acknowledging the exploratory nature of the study, we consider that moderate and severe head injuries might not be confidently analyzed using the same model. The study also indicates that neuroimaging, as well as the traditional prognostic factors, are insufficient for explaining the heterogeneity seen in the upper range of outcomes in TBI patients.

Taken together we feel that the present study demonstrates the clinical usefulness of MRI in head injury, in documenting the TBI, and in the sensitivity of the depiction of lesions with important prognostic information in severe head injuries.

Footnotes

Acknowledgments

We wish to thank neuroradiologist Mari Folvik (M.F.) for evaluation of MRI scans, and neuroradiologist Jana Rydland (J.R.) for her contribution to the reliability evaluation; Beate Holmqvist Karlsen, Otto Aarhaug, and Brit Sørum for participating in management of the database and the GOSE interviews; Kent Gøran Moen for his work with the data files; the residents at the neurosurgical department for collection of the injury-related variables; the staff at the MRI unit for their cooperation with the MRI examinations; Toril Skandsen (T.S.) and Ole Solheim (O.S.) have received research grants from the Liaison Committee between the Central Norway Regional Health Authority (RHA) and the Norwegian University of Science and Technology (NTNU), T.S. during the whole study period, and O.S. during the period of manuscript preparation.

References

Altman

D.G.

1999. Practical Statistics for Medical Research. Chapman & Hall: London.

Carpentier

, Galanaud

, Puybasset

, Muller

J.C.

, Lescot

, Boch

A.L.

, Riedl

, Cornu

, Coriat

, Dormont

, van Efffenterre

2006. Early morphologic and spectroscopic magnetic resonance in severe traumatic brain injuries can detect “invisible brain stem damage” and predict “vegetative states.” J. Neurotrauma, 23:674–685.

Chastain

C.A.

, Oyoyo

, Zipperman

, Joo

, Ashwal

, Shutter

, Tong

2009. Predicting outcomes of traumatic brain injury by imaging modality and injury distribution. J. Neurotrauma, 26:1183–1196.

Firsching

, Woischneck

, Klein

, Reissberg

, Dohring

, Peters

2001. Classification of severe head injury based on magnetic resonance imaging. Acta Neurochir, 143:263–271.

Garnett

M.R.

, Blamire

A.M.

, Corkill

R.G.

, Cadoux-Hudson

T.A.

, Rajagopalan

, Styles

2000. Early proton magnetic resonance spectroscopy in normal-appearing brain correlates with outcome in patients following traumatic brain injury. Brain, 123:2046–2054.

Gentry

L.R.

2002. Head Trauma. Lippincott Williams & Wilkins: Philadelphia.

Jennett

, Bond

1975. Assessment of outcome after severe brain damage. Lancet, 1:480–484.

Jennett

, Snoek

, Bond

M.R.

, Brooks

1981. Disability after severe head injury: observations on the use of the Glasgow Outcome Scale. J. Neurol. Neurosurg. Psychiatry, 44:285–293.

Kampfl

, Schmutzhard

, Franz

, Pfausler

, Haring

H.P.

, Ulmer

, Felber

, Golaszewski

, Aichner

1998. Prediction of recovery from post-traumatic vegetative state with cerebral magnetic-resonance imaging. Lancet, 351:1763–1767.

10.

Lagares

, Ramos

, Perez-Nunez

, Ballenilla

, Alday

, Gomez

P.A.

, Kaen

, Lobato

R.D.

2009. The role of MR imaging in assessing prognosis after severe and moderate head injury. Acta Neurochir, 151:341–356.

11.

Levin

H. S.

, Mendelsohn

, Lilly

M.A.

, Yeakley

, Song

, Scheibel

R.S.

, Harward

, Fletcher

J.M.

, Kufera

J.A.

, Davidson

K.C.

, Bruce

1997. Magnetic resonance imaging in relation to functional outcome of pediatric closed head injury: a test of the Ommaya-Gennarelli model. Neurosurgery, 40:432–440; discussion 440–431.

12.

Levin

H.S.

, Williams

, Crofford

M.J.

, High

W.M.

Jr. , Eisenberg

H.M.

, Amparo

E.G.

, Guinto

F.C.

Jr. , Kalisky

, Handel

S.F.

, Goldman

A.M.

1988. Relationship of depth of brain lesions to consciousness and outcome after closed head injury. J. Neurosurg, 69:861–866.

13.

Maas

A.I.

, Hukkelhoven

C.W.

, Marshall

L.F.

, Steyerberg

E.W.

2005. Prediction of outcome in traumatic brain injury with computed tomographic characteristics: a comparison between the computed tomographic classification and combinations of computed tomographic predictors. Neurosurgery, 57:1173–1182; discussion 1173–1182.

14.

Mannion

R.J.

, Cross

, Bradley

, Coles

J.P.

, Chatfield

, Carpenter

, Pickard

J.D.

, Menon

D.K.

, Hutchinson

P.J.

2007. Mechanism-based MRI classification of traumatic brainstem injury and its relationship to outcome. J. Neurotrauma, 24:128–135.

15.

Mitchell

D.E.

, Adams

J.H.

1973. Primary focal impact damage to the brainstem in blunt head injuries. Does it exist? Lancet, 2:215–218.

16.

Murray

G.D.

, Butcher

, McHugh

G.S.

, Lu

, Mushkudiani

N.A.

, Maas

A.I.

, Marmarou

, Steyerberg

E.W.

2007. Multivariable prognostic analysis in traumatic brain injury: results from the IMPACT study. J. Neurotrauma, 24:329–337.

17.

Mushkudiani

N.A.

, Hukkelhoven

C.W.

, Hernandez

A.V.

, Murray

G.D.

, Choi

S.C.

, Maas

A.I.

, Steyerberg

E.W.

2008. A systematic review finds methodological improvements necessary for prognostic models in determining traumatic brain injury outcomes. J. Clin. Epidemiol, 61:331–343.

18.

Ommaya

A.K.

, Gennarelli

T.A.

1974. Cerebral concussion and traumatic unconsciousness. Correlation of experimental and clinical observations of blunt head injuries. Brain, 97:633–654.

19.

Parizel

P.M.

, Van Goethem

J.W.

, Ozsarlak

, Maes

, Phillips

C.D.

2005. New developments in the neuroradiological diagnosis of craniocerebral trauma. Eur. Radiol, 15:569–581.

20.

Pierallini

, Pantano

, Fantozzi

L.M.

, Bonamini

, Vichi

, Zylberman

, Pisarri

, Colonnese

, Bozzao

2000. Correlation between MRI findings and long-term outcome in patients with severe brain trauma. Neuroradiology, 42:860–867.

21.

Shibata

, Matsumura

, Meguro

, Narushima

2000. Differentiation of mechanism and prognosis of traumatic brain stem lesions detected by magnetic resonance imaging in the acute stage. Clin. Neurol. Neurosurg, 102:124–128.

22.

Sidaros

, Engberg

A.W.

, Sidaros

, Liptrot

M.G.

, Herning

, Petersen

, Paulson

O.B.

, Jernigan

T.L.

, Rostrup

2008. Diffusion tensor imaging during recovery from severe traumatic brain injury and relation to clinical outcome: a longitudinal study. Brain, 131:559–572.

23.

Skandsen

, Kvistad

K.A.

, Solheim

, Strand

I.H.

, Folvik

, Vik

2010. Prevalence and impact of diffuse axonal injury in patients with moderate and severe head injury: a cohort study of early magnetic resonance imaging findings and 1-year outcome. J. Neurosurg, 113:556–563.

24.

Stein

S.C.

, Spettell

1995. The Head Injury Severity Scale (HISS): a practical classification of closed-head injury. Brain Inj, 9:437–444.

25.

Stocchetti

, Pagan

, Calappi

, Canavesi

, Beretta

, Citerio

, Cormio

, Colombo

2004. Inaccurate early assessment of neurological severity in head injury. J. Neurotrauma, 21:1131–1140.

26.

Teasdale

, Jennett

1974. Assessment of coma and impaired consciousness. A practical scale. Lancet, 2:81–84.

27.

Vittinghoff

, McCulloch

C.E.

2007. Relaxing the rule of ten events per variable in logistic and Cox regression. Am. J. Epidemiol, 165:710–718.

28.

Wedekind

, Hesselmann

, Lippert-Gruner

, Ebel

2002. Trauma to the pontomesencephalic brainstem—a major clue to the prognosis of severe traumatic brain injury. Br. J. Neurosurg, 16:256–260.

29.

Weiss

, Galanaud

, Carpentier

, Naccache

, Puybasset

2007. Clinical review: Prognostic value of magnetic resonance imaging in acute brain injury and coma. Crit. Care, 11:230.

30.

Weiss

, Galanaud

, Carpentier

, Tezenas de Montcel

, Naccache

, Coriat

, Puybasset

2008. A combined clinical and MRI approach for outcome assessment of traumatic head injured comatose patients. J. Neurol, 255:217–223.

31.

Wilson

J.T.

, Pettigrew

L.E.

, Teasdale

G.M.

1998. Structured interviews for the Glasgow Outcome Scale and the extended Glasgow Outcome Scale: guidelines for their use. J. Neurotrauma, 15:573–585.

32.

Yanagawa

, Tsushima

, Tokumaru

, Un-no

, Sakamoto

, Okada

, Nawashiro

, Shima

2000. A quantitative analysis of head injury using T2*-weighted gradient-echo imaging. J. Trauma, 49:272–277.