Mismatch between midline shift and hematoma thickness as a prognostic factor of mortality in patients sustaining acute subdural hematoma.

BACKGROUND: Acute subdural hematoma (ASDH) is a traumatic lesion commonly found secondary to traumatic brain injury. Radiological findings on CT, such as hematoma thickness (HT) and structures midline shift (MLS), have an important prognostic role in this disease. The relationship between HT and MLS has been rarely studied in the literature. Thus, this study aimed to assess the prognostic accuracy of the difference between MLS and HT for acute outcomes in patients with ASDH in a low-income to middle-income country.
METHODS: This was a post-hoc analysis of a prospective cohort study conducted in a university-associated tertiary-level hospital in Brazil. The TRIPOD (Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis) statement guidelines were followed. The difference values between MLS and HT (Zumkeller index, ZI) were divided into three categories (<0.00, 0.01-3, and >3). Logistic regression analyses were performed to reveal the OR of categorized ZI in predicting primary outcome measures. A Cox regression was also performed and the results were presented through HR. The discriminative ability of three multivariate models including clinical and radiological variables (ZI, Rotterdam score, and Helsinki score) was demonstrated.
RESULTS: A total of 114 patients were included. Logistic regression demonstrated an OR value equal to 8.12 for the ZI >3 category (OR 8.12, 95% CI 1.16 to 40.01; p=0.01), which proved to be an independent predictor of mortality in the adjusted model for surgical intervention, age, and Glasgow Coma Scale (GCS) score. Cox regression analysis demonstrated that this category was associated with 14-day survival (HR 2.92, 95% CI 1.38 to 6.16; p=0.005). A multivariate analysis performed for three models including age and GCS with categorized ZI or Helsinki or Rotterdam score demonstrated area under the receiver operating characteristic curve values of 0.745, 0.767, and 0.808, respectively.
CONCLUSIONS: The present study highlights the potential usefulness of the difference between MLS and HT as a prognostic variable in patients with ASDH. LEVEL OF EVIDENCE: Level III, epidemiological study. © Author(s) (or their employer(s)) 2021. Re-use permitted under CC BY-NC. No commercial re-use. See rights and permissions. Published by BMJ.

Background

Acute subdural hematoma (ASDH) is a traumatic lesion commonly found secondary to traumatic brain injury (TBI). The acceleration–deceleration resulting from trauma results in the rupture of the bridged veins, which causes the formation of hematoma.1–3 Damage to brain tissue cannot be attributed only to ASDH, as this hematoma is commonly associated with cerebral edema, brain concussions, and diffuse axonal lesions, which can worsen patient prognosis.4 5 Despite advances in prehospital and hospital support, this entity still has high mortality rates.6–8 The mortality rate in the literature ranges from 40% to 70%.6 9 10

Several factors are associated with the prognosis of these patients, such as the Glasgow Coma Scale (GCS) score at admission, age, pupillary response, and associated traumatic injuries.11 12 Radiological findings on CT, such as hematoma thickness (HT) and structures midline shift (MLS), have an important prognostic role and can be a determining factor for surgical management.13–17

MLS is caused by the hematoma itself and concomitant brain swelling.18 The effect of brain swelling can be shown on the initial head CT scan by the relationship of the MLS and the HT. Disproportionately increased MLS compared with HT would raise concerns of brain tissue injury and edema, and hence a higher intracranial pressure (ICP) and a worse prognosis could be anticipated. HT, MLS, ICP, and other factors including the severity of brain atrophy are interrelated.18–20

The concept of brain swelling is not new. However, complex scoring systems that assess radiological findings as prognostic factors for TBI include MLS and hematoma data in isolation and not associated, such as the Rotterdam21 and Helsinki22 scores.

The relationship between HT and MLS has been little studied in the literature.18 19 Zumkeller et al19 were the first to evaluate the difference between them as prognostic factors for survival in a cohort of patients. Recently, Bartels et al,9 in a retrospective cohort of 59 patients, determined that a difference between MLS and HT greater than 3 mm was associated with fatality. However, this parameter was not evaluated in a large sample and in low-income or middle-income countries (LMICs), a context of scarcity of resources that can benefit from simplified predictive models, thus requiring additional studies for external validation of its real predictive value. Thus, this study aimed to assess the prognostic value of the difference between MLS and HT for acute outcomes in patients with ASDH in an LMIC.

Methods

Study design

This was a post-hoc analysis of a previously conducted prospective cohort study. The study adhered to the principles of the Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD), the TRIPOD statement.23 All participants provided written informed consent to participate, and none of them were identified in this study.

Patients and population

The study was conducted at the Clinics Hospital of São Paulo, a tertiary-level hospital located in the largest city of Brazil, serving a population of more than 12 million people. Data were collected from a prospective data bank registry of patients with TBI admitted to the emergency department between January 2012 and December 2015. Inclusion criteria were defined as age greater than or equal to 18 years and traumatic ASDH confirmed by CT. Exclusion criteria were spontaneous or subacute ASHD, bilateral ASDH, epidural hematoma and patients with polytrauma. In our institution, any patient with intracranial abnormalities is eligible to be transferred to intensive care unit, which is subject to availability of bed. Therapeutic planning followed the recommendations provided by the Advanced Trauma Life Support and the guidelines provided by the Brain Trauma Foundation. Patients with non-operative management were submitted to a strict control of ICP, so when surgical evacuation is necessary it is immediately performed.

Outcome and variables of interest

Clinical variables

Epidemiological data such as sex and age were assessed. Trauma mechanisms were analyzed and classified as road traffic injury, fall, violence, and others. The GCS postresuscitation care for hospital admission and pupillary response were also included.

Radiological variables

The diagnosis of ASDH was made based on the admission CT evaluation performed using a multidetector 64-channel CT scanner (Philips Medical Systems World Headquarters, Best, The Netherlands) by a neurosurgeon and confirmed by an experienced radiologist. The Rotterdam and Helsinki scores were calculated according to the original article that describes them.21 22 These scores were chosen because they are complex models that include several radiological variables in their evaluation. Previous studies have already demonstrated that higher Helsinki and Rotterdam scores are associated with worse clinical outcomes in patients with TBI.8 16 18 The Marshall CT class was not evaluated because it was not originally constructed for outcome prediction and was developed in a patient population managed with protocols of care from the early 1980s, where aggressive surgical management for high ICP was not a common approach.24

MLS was assessed in the axial plane at the level of the interventricular foramen. This was measured as the longest perpendicular distance between the most displaced point of the pellucid septum and an imaginary sagittal line that associates the inner occipital protuberance and the frontal ridge.19 25 HT was assessed as the longest distance between the inner table and the cortex.6 Figure 1 shows an example of HT and MLS in a patient with ASDH. The HT was subtracted from the MLS to obtain the value described by Zumkeller et al19—the Zumkeller index (ZI=MLS−HT).

Figure 1

The thickness of the acute subdural hematoma (red arrow) was measured on a CT scan as the largest distance between the cortex and the internal table: midline, orange line; midline shift, orange arrow. HT, hematoma thickness; MLS, midline shift.

To assess its relationship with the outcome, a categorical variable with three cut-off points was created, categorized as ZI, using the authors’ original work as a guide or key parameter.19 The first category included patients whose MLS was lower than HT. The second category included patients whose calculation was between 0.01 mm and 3 mm. The third category included patients who had a difference of MLS−HT greater than 3 mm. This categorization takes into account the principle that there is an underlying brain swelling when MLS is greater than HT.

Patients were followed up during their hospital stay. The predictive value of the ZI and other variables was assessed for the primary outcome of 14-day mortality, based on reports in the literature for assessing acute outcomes of TBI.26 27

Statistical analysis

Categorical variables were presented using relative and absolute frequencies for descriptive purposes. Normally distributed continuous data were presented as mean and SD, and otherwise by median and quartiles. Continuous data were tested for normality using the Shapiro-Wilk test.

Clinical and radiological variables were presented in each category of the ZI and were compared. Categorical data were compared using the χ2 test. Normally distributed continuous data were compared using a one-way analysis of variance (ANOVA), and the post-hoc Tukey test was used to identify which pair had a significant difference. Data without normal distribution were compared using the Kruskal-Wallis test, and if the result was significant pairwise comparisons were performed.

Logistic regression analyses were performed to reveal the OR of the categorized ZI in predicting primary outcome measures. Analyses considered unadjusted models and models adjusted for age, GCS, and surgical intervention. The estimated survival times for the three ZI categories were estimated using the Kaplan-Meier model and compared using the log-rank test. A Cox regression analysis was also performed and the results were presented as HR and 95% CI.

A multivariate model was used to assess the discriminatory ability of the ZI against the complex scores of Rotterdam and Helsinki. Three models were created, all of which included age and GCS added from the Helsinki or Rotterdam or ZI score. The discriminative ability of multivariable models was evaluated by calculating the area under the receiver operating characteristic (ROC) curve (AUC). Calibration was assessed using the Hosmer-Lemeshow test.

All tests were two-sided and a final p value <0.05 was considered statistically significant. Data were analyzed using the Statistical Package for Social Sciences (IBM SPSS Statistics for Windows V.25.0; IBM, Armonk, NY, USA), and ROC curve analysis was performed using MedCalc V.19.1.3 (MedCalc, NY, USA).

Results

Patient characteristics

Overall, 114 patients were included in the study; 84.2% (n=96) were men and 15.8% (n=18) were women. The mean age was 47.7 years (SD 19.3). Regarding the trauma mechanism, 43.90% (n=50) were due to motor vehicle crashes, and the same percentage was associated with falls. A total of 64.90% (n=74) were categorized as severe TBI, and the others were classified as having moderate or mild TBI. Half of the admitted patients scored less than 6 points (IQR, 3–10) on the GCS.

The mean value of the ZI was −4.99 mm (SD: 6.80). In the categorical ZI assessment, 78.90% (n=90) of patients had negative values, 11.4% (n=13) between 0.01 mm and 3 mm, and 9.6% (n=11) greater than 3 mm. Of those who had values greater than 3 mm, 81.81% (n=9) died.

Table 1 presents the main characteristics of the sample in each category of the ZI. The Kruskal-Wallis test demonstrated that there was an effect of ZI on the GCS, and pairwise comparison demonstrated that there was a difference between the first and third categories of the ZI (p=0.025). One-way ANOVA demonstrated the effect of ZI on MLS. The post-hoc Turkey test showed that, on average, MLS was different between the ZI categories of 1 and 2 and 1 and 3 (p<0.001 and p<0.001, respectively); however, there was no difference between categories 2 and 3 (p=0.507). Although there was a difference in MLS, no significant differences were found in HT. As for radiological scores, the Kruskal-Wallis test revealed an effect of the ZI under the Rotterdam and Helsinki scores. The pairwise comparison showed a statistically significant difference in the distribution of the values of the two scores between categories 1 and 3 of the ZI (p=0.004 and p=0.001, respectively). Finally, the χ2 test showed that there was an association between the ZI and the presence of intraventricular hemorrhage (IVH), intracerebral hematoma (ICH), and 14-day mortality and in-hospital mortality outcomes.

Table 1

Clinical and radiological characteristics according to ZI category

	ZI <0	ZI 0–3 mm	ZI >3	P value
Age*	49.8 (19.6)	36.7 (12.9)	49.7 (15.6)	0.061
GCS†	6 (3–10)	5 (3–11)	3 (3–4)	0.031
Pupil reactivity				0.257
Both pupils	62.2 (56)	76.9 (10)	36.4 (4)
One pupil	17.8 (16)	15.4 (2)	36.4 (4)
Neither pupil	11.1 (2)	7.7 (1)	27.3 (3)
TBI class				0.874
Mild	15.6 (14)	7.7 (1)	1 (9.1)
Moderate	14.4 (13)	15.4 (2)	1 (9.1)
Severe	64.4 (58)	53.8 (7)	81.1 (9)
MLS, mm*	6.7 (5.45)	14.1 (7.8)	16.7 (5.1)	<0.001
HT, mm*	14.1 (7.12)	13.1 (7.9)	10.9 (7.7)	0.368
Rotterdam score†	5 (3–6)	5 (4.5–6)	6 (6–6)	0.002
Helsinki score†	6 (4–9)	9 (5–10)	11 (9–14)	0.001
Other injuries
Fractures	42.2 (38)	46.2 (6)	63.6 (7)	0.401
tSAH	5.6 (5)	100 (13)	100 (11)	0.498
IVH	21.1 (19)	7.7 (1)	45.5 (5)	0.077
ICH	50 (45)	46.2 (6)	90.9 (10)	0.032
DC	81.1 (73)	92.3 (12)	90.9 (10)	0.466
Outcome
14-day mortality	36.7 (33)	30.8 (4)	81.1 (9)	0.012
In-hospital mortality	48.9 (44)	46.2 (6)	90.9 (10)	0.027

*Mean (SD).

†Median (IQR); all other variables: % (number).

DC, decompressive craniectomy; GCS, Glasgow Coma Scale; HT, hematoma thickness; ICH, intracerebral hematoma; IVH, intraventricular hemorrhage; MLS, midline shift; TBI, traumatic brain injury; tSAH, traumatic subarachnoid hemorrhage; ZI, Zumkeller index.

Prognostic value of MLS, HT, and ZI for 14-day mortality outcome

HT was significantly associated with clinical outcome (p=0.027). Logistic regression analysis showed that HT (OR 1.073, 95% CI 1.015 to 1.135; p=0.013) was a significant predictor of mortality. MLS was also associated with 14-day mortality outcome (p=0.001) and was a significant predictor of mortality (OR 1.105, 95% CI 1.038 to 1.176; p=0.002).

The categorized ZI was associated with 14-day mortality outcome (p=0.012). Logistic regression demonstrated that this was a predictor of mortality with the reference category ZI >3 (OR 8.12, 95% CI 1.1651 to 40.01; p=0.01) (table 2). This category proved to be an independent predictor of mortality in the adjusted model for surgical intervention, age, and GCS (online supplemental table 1).

Table 2

Univariate logistic and Cox regression values in predicting 14-day mortality outcome according to Zumkeller index categories

Zumkeller index	Logistic regression			Cox regression
	OR	95% CI	P value	HR	95% CI	P value
<0.00*	–	–	0.033	–	–	0.014
0.01–3	0.768	0.219 to 2.689	0.679	0.854	0.303 to 2.412	0.766
>3	7.773	1.583 to 38.155	0.012	2.924	1.388 to 6.160	0.005

*Zumkeller index values <0.00 were included as a reference category in the model.

Survival analysis

The Kaplan-Meier survival analysis demonstrated a statistically significant difference in the survival function at the different levels of the ZI categorized for the 14-day follow-up (log-rank p=0.008) (figure 2). Cox regression analysis demonstrated that the third category of ZI was associated with 14-day survival (HR 2.924, 95% CI 1.388 to 6.160; p=0.005), as shown in table 2. This category proved to be an independent predictor of 14-day survival in adjusted Cox regression for surgical intervention, age, and GCS (online supplemental table 2).

Figure 2

ROC curves for the multivariate model including age, Glasgow Coma Scale and categorical Zumkeller index (model 1), Helsinki CT score (model 2), Rotterdam CT score (model 3). AUC, area under the ROC curve; ROC, receiver operating characteristic.

Multivariable predictive models

Multivariate analysis was performed for all three models. Model 1 included age, GCS, and categorized ZI; model 2 included age, GCS, and Helsinki score; and model 3 included age, GCS, and Rotterdam score. All variables included were independent predictors of mortality. Models 1, 2, and 3 demonstrated an AUC equal to 0.747 (95% CI 0.654 to 0.826), 0.767 (95% CI 0.676 to 0.843), and 0.808 (95% CI 0.721 to 0.878), respectively, as shown in figure 3. The test by DeLong et al28 showed no significant differences between the AUC values (p>0.05). All models demonstrated adequate calibration using the Hosmer-Lemeshow test (p>0.05).

Figure 3

Survival function for each category of Zumkeller index for 14-day mortality outcome.

Discussion

The present study stands out for presenting the prognostic value of a variable that is little reviewed in the literature, “Zumkeller index,” which was associated with acute outcomes and which may suggest the presence of cerebral edema or other associated injuries in patients with ASDH. The data presented corroborate the assumption by Zumkeller et al19 that an MLS that exceeds the thickness of the hematoma suggests the presence of associated cerebral edema and/or other parenchymal lesions, which contributes to worst clinical outcome.

In a review of approximately 3000 autopsy cases after TBI, Tandon29 concluded that ASDHs are rarely isolated lesions; in their study, 82% of the cases were associated with parenchymal lesions, which again demonstrates the importance of MLS evaluation. In our study, intraparenchymal lesions were found in 53.5% of patients, and a higher frequency of this finding was observed (90.9%) in patients who were classified as having a ZI >3. Several studies have associated the presence of ICH and its progression with poor prognosis.30–32 The study by Cepeda et al33 demonstrated that when studying the progression of ICH in patients with TBI, ASDH is associated more with this progression. In addition, it highlights the role of higher MLS values in the progression of ICH. In cases where the ZI was greater than 0, the role of MLS was notorious. Therefore, patients with these findings and associated ICH have a greater risk of lesion progression. Cepeda et al33 also demonstrated an unfavorable clinical outcome in 74% of patients with ICH progression, compared with 26% of those who did not. In addition, the number of patients undergoing decompressive craniectomy was significantly higher in patients with ICH progression (75% vs. 25%).

In our study, patients with ZI >3 had an increased frequency of IVH (45.5%) compared with 7.7% of patients with ZI between 0.001 and 3 and 21.1% of patients with negative ZI. According to the literature, the rate of occurrence of IVH in patients with moderate-to-severe TBI ranges between 7.1% and 22%.34–36 The presumed possible mechanisms of IVH are an extension of intracerebral hemorrhage into the ventricular system or rupture of the subependymal veins, which can be deformed by the negative pressure at the time of injury.37 Like ICH, several studies also associate IVH with a poor prognosis.21 The study by Laleva et al36 demonstrated that the main factor related to the onset of IVH is the presence of ICH on admission images. In our cohort, we observed high frequencies of the two lesions in patients with a ZI >3.

It was also observed that 100% of patients with a positive ZI had a traumatic subarachnoid hemorrhage (SAH). Diffuse bleeding resulting from rupture of the subarachnoid vessels in TBI is a predictor well described in the literature.38–40 Similar to what happens in the rupture of aneurysms, traumatic SAH induces vasospasm and cerebral ischemia, which can trigger inflammatory and neurotoxic processes and contribute to brain swelling, which contributes to worsening the outcome of patients.41 42

However, in cases of isolated ASDH, the ZI may play a role as a prognostic factor of mortality because, without a parenchymal lesion, the MLS exceeding the HT may be due to brain swelling and impairment of cerebrovascular reactivity.

The relationship between MLS and HT is promising in terms of prognosis and may help to estimate ICP. Recently, Liao et al20 constructed a model for ICP estimation based on a half-sphere finite-element model using only HT and MLS in ASDH. The ICP values obtained by the normogram created by the authors showed a high correlation with those measured, demonstrating an R2 coefficient of 0.744. Using estimates like these and predictive models like ZI, neurosurgeons can do more with less. In patients with TBI with multiple traumas, estimating ICP and brain swelling also helps prioritize treatment of different regions of the body.

We highlight the high mortality rate of patients with an MLS >3 mm in relation to HT. In our series, there was 81.8% mortality within 14 days for these patients, an even higher rate than that found by Zumkeller et al.19 When the entire length of hospital stay was assessed, 10 of the 11 patients categorized as ZI >3 died. Bartels et al,9 when studying this difference in a small cohort of patients, found a mortality rate of 100% when ZI was greater than 3 mm, which indicates the accuracy of this indicator for use in clinical practice. The authors further suggested that in these patients, the trauma resulted in greater damage than that generated by the hematoma, in addition to having an influence on the anatomy and physiology of the brain, resulting in an acute onset of swelling. This assumption helps to justify the poorer clinical presentation at admission of patients with positive ZI, demonstrated by significantly lower GCS values in the second and third categories of the ZI, as well as the poor pupillary response of patients with ZI >3. It should also be noted that values greater than 3 mm, in addition to being associated with a worse outcome, also implied a reduction in the estimated survival time.

Additionally, our results highlight the potential role of ZI as a variable for modeling studies, since a comparison with more complex scores such as Rotterdam21 and Helsinki22 demonstrated acceptable AUC values. We hypothesize that the values close to the three multivariate models created to assess the prediction of acute ZI outcomes against these scores are because the brain injuries covered by the scores, such as IVH, ICH, and SAH, are related to the positive values of MLS–HT. It should be noted that the intention of our work is not to suggest the replacement of other existing tomography scores because they can be used in the context of other brain injuries and include variables that are associated with the prognosis of TBI; our objective was to present a complementary index to these scores which assesses something not directly included in them. The objective of our multivariate analysis, which compared the ZI with these scores, was to demonstrate that the ZI has a discriminative ability similar to these more complex models, demonstrating its clinical utility in the context of outcome prediction.

To date, only two studies have evaluated the ZI: Zumkeller et al,19 who described the index, and Bartels et al.9 Neither of these studies were conducted in LMICs, which have distinct epidemiological contexts from high-income countries. As we know, LMICs have the highest burden of neurotrauma; however, most of the scientific articles published in journals originate from high-income countries.43–45 Interestingly, in Brazil, many centers already use this “Zumkeller index,” not as a prognostic variable but to decide whether to perform a primary decompressive craniectomy in the setting of ASDH.46 As it is a post-hoc analysis of a prospective study, it is not possible to suggest the indication of decompressive craniectomies for patients with positive ZI. However, cerebral edema and/or associated injuries in this group of patients are notorious. We then suggest the proper management of intracranial hypertension and that future studies of primary decompressive craniectomy address this variable.

Finally, regarding the epidemiological data for ASDH, the literature demonstrates that high mortality rates are more often associated with advanced ages.6 47–49 Wilberger et al,50 in their study, demonstrated that the average age of non-survivors was 59 years, whereas that of survivors was 41 years, a finding similar to that seen in other studies.3 9 In our study, the average age of survivors was 44.4 years and that of non-survivors was 53.4 years. Ryan et al,51 in their study, described falls as the main injury mechanism (57%), followed by automobile crash (23%), which was very close to the study by Leitgeb et al,52 which also presented falls as the main mechanism (51.9%), followed by automobile crash (22.2%). In the present study, the percentages of ASDH resulting from automobile crash and falls were the same (43.9%). Several authors have already demonstrated epidemiological differences in different socioeconomic contexts.53 54 In LMICs, where traffic laws and enforcement are not as effective, automobile crash and trauma mechanisms related to aggression are predominant. Thus, we emphasize that the present study is the first to evaluate the MLS–HT relationship in an LMIC, which contributes to the process of external validation of its practical utility in different contexts.

Study limitations

This study has some limitations. Despite an adequate number of patients for the proposed analyses, this study was restricted to a single center, which may limit the generalization of the findings. In addition, it is a post-hoc analysis of a prospective study. Thus, some variables, such as surgical intervention, could not be controlled. Despite this, we present an adjusted regression model for performing decompressive craniectomy associated with other clinical variables. Besides, to minimize bias of deaths from causes other than TBI, the primary outcome of our study was 14-day mortality. Several authors have already described that this is a useful time to assess prognosis in patients suffering from TBI. In a shorter follow-up period, it is possible to minimize other factors that could contribute to mortality, such as in-hospital infections and late complications. We encourage other authors from various centers around the world to assess the difference between MLS and HT for patients with ASDH and to investigate ZI in prospective cohorts as an indication of decompressive craniectomy. We emphasize the limitation of not providing data on the long-term outcomes of the study population. In future studies, the Glasgow Outcome Scale should be included. However, the difficulty of long-term monitoring of patients with TBI is not restricted to our study, which has been previously reported in the literature, mainly in LMICs.55 56

Conclusion

The present study highlighted the potential usefulness of the difference between MLS and HT (the “Zumkeller index”) as a prognostic variable for patients with ASDH. A worse clinical presentation by GCS was demonstrated in patients with positive ZI, as well as a worse classification in the Rotterdam and Helsinki tomography scores. A higher frequency of other injuries associated with ASDH, such as SAH, IVH, and ICH, was found in patients with a ZI greater than 3. In addition, the present study demonstrated a lower survival rate in this group of patients. We advocate that such an index be used in future modeling studies or evaluate its potential in future studies of primary decompressive craniectomies.