Derivation of a Bleed to Brain Ratio to Predict the Need for Surgery in Head Injury
Correspondence Address: Source of Support: None, Conflict of Interest: None DOI: 10.4103/0028-3886.299134
Source of Support: None, Conflict of Interest: None
Keywords: Cerebral oedema, contusions, CT scan, head injury, head trauma, intracranial pressure
The Monro--Kellie doctrine states that the cranial compartment is incompressible and the cranium and its constituents (blood, CSF, and brain tissue) create a state of volume equilibrium, such that any increase in the volume of one of the cranial constituents must be compensated by a decrease in volume of another. Bleeding into such a closed cavity causes increased intracranial pressure. A high intracranial pressure is the cause of most of the brain damage that occurs after head trauma.
Current guidelines, such as the BTF guidelines, for surgical evacuation of intracranial bleed following head injury mention the volume of a particular bleed to consider the evacuation of that bleed. For example, a 30 ml extradural hematoma requires evacuation. However, the guidelines do not take into account the intracranial volume. The normal intracranial volume ranges from 1200 to 1700 ml. The rise in intracranial pressure should occur faster in a cranial cavity that has a lesser total volume for a given amount of bleed since the product of pressure and volume is a constant.
The decision for surgery is more difficult in the cases where there are multiple, irregularly shaped hematomas. The situation is further complicated in the setting of edema formation which also contributes to increased pressure.
So, we hypothesized that a ratio of total bleed volume to the intracranial volume should be a more appropriate measure to decide on the need for surgery than an absolute measure of the volume of bleed at any one site.
Patients of the age range of 16–50 years who underwent CT scans in our hospital within ten days of sustaining head injury were included in this retrospective study. The scans of the patient who was admitted for head trauma from May 2013 to October 2014 were used for this study.
Grossly depressed or elevated fractures or penetrating brain injuries were excluded since an exact intracranial volume could not be calculated using the methods described below. Since age-dependent atrophy of the brain allows relatively more bleed volumes to be accommodated in the brain with comparatively lesser symptoms, we chose to exclude patients above the age of 50 years. Patients younger than 16 years are more likely to develop edema related swelling of the brain and were excluded too.
We included all kinds of bleed including acute subdural hematomas, extradural hematomas, contusions, intraventricular bleeds, and subarachnoid hemorrhage.
The data on whether the patient underwent surgery or not was obtained from the case records and operation registers.
Planimetric method of volumetry was done on the console of a two-slice CT scanner (Somatom Spirit, Siemens AG, Berlin) by bracketing the Hounsfield units (HU) within a range 50–80 units to selectively measure the volume of blood clots within the brain. The native software of the CT scanning machine (Somaris/5.5 Syringo CT 2007P Spirit VB 35B, Siemens AG, Berlin) was used. The bracket of 50–80 HU was iteratively arrived upon by measuring different ranges to find the optimal one that matches visually identifiable blood clots while minimizing the chance of measuring structures other than blood.
The total intracranial volume was assessed by using a HU bracket of 0–80 units. This bracketing excluded skull, calcification, and air but included all other structures like the brain, the CSF, and the blood clots. The edema volume was calculated with a bracketing of 15–20 HU that matched low-radiodensity areas that were visually identified as edema around lesions like contusions. In the strict mathematical sense, the calculated ratio was the ratio of the bleed to the total intracranial volume, which we refer to it in this study as the bleed-brain ratio (BBR). Similarly, we calculated the bleed-edema-brain (BEBR) ratio by adding the edematous brain tissue volume to the total bleed volume and then dividing it by the total intracranial volume.
Two methods of planimetry were used as described below
Method 1 (precise method): Here, the range of interest (ROI) was defined by manually drawing within and along the skull bone, from the level of the foramen magnum, in each slice of the CT scan so that exact intracranial volume and bleeds within it was estimated.
Method 2 (ellipse method): In this alternative and easier method, the ROI was defined by using a circle/ellipse that encompasses the whole of the head instead of drawing the ROI exactly within the skull borders.
The advantage of this latter technique was that it was less time consuming and much less labor intensive. However, it was less accurate since the volume of scalp tissue and other soft tissue outside the skull and the volume of any bleed in these places also got measured [Figure 1].
While the first method typically took around 20 min, the second method could be completed in less than 5 min.
Both planimetric methods ensured that interobserver or intraobserver differences are minimized as it does not require the person to delineate the hematoma per se but define only the anatomical volume of interest. All measurements were taken using the same window level settings and scan parameters since these factors can also affect volumetry. Planimetric method of volumetry is superior to other approximations like the AXBXC/2 method.
A motor score of M6 (obeying commands) at the time of discharge was considered a good outcome and the rest were considered poor outcomes. It is well known that the motor score is not only linearly related to survival but preserves almost all the predictive power of the GCS.
Sample size estimation
Brain trauma foundation (BTF) guidelines state that intracranial bleeds like an extradural hematoma need to be surgically evacuated if the volume is greater than 30 ml. The intracranial volume is around 1,500 ml in an adult. So, in this case, the BBR is 0.02. We expected a similar or higher ratio to be found in those patients who required surgery in our study. In patients who did not require surgery, we expected this ratio to be less by at least 15%. The difference of 15% was taken arbitrarily since no estimates of the difference in proportion between the patients with and without surgical intervention were available in the literature. A sample size of 100 in each group was estimated with an expected difference in the proportion of BBR as 15% at 5% level of significance and 80% power.
Data management was done using REDCap (Research Data Capture, Vanderbilt University) electronic database.
The comparison of the ratios between the groups was carried out by using independent Student's t-test.
ROC curve was used to assess the predictive power of the BBR and BEBR to determine the need for surgical intervention. A cutoff value for the BBR with clinically applicable sensitivity and specificity was estimated from the ROC curve.
The area under the curve was used to compare the ROC curves generated using different ratios from different methods.
All statistical analysis was carried out at 5% level of significance and a P value <0.05 was considered as significant. Statistical analysis was carried out using SPSS version 21.
No extra scans or interventions were carried out for the sake of the study. Institute Ethics Committee cleared the study.
A total of 212 adult patients were included in this retrospective study. The mean age of the patients was 35 years (SD: 9.7). Among these patients, 184 (86.8%) were males. Ninety-one patients (43%) had undergone surgery and 121 (57%) patients were managed conservatively.
Patients had been managed according to the decisions made by experienced neurosurgical treating teams. But since the team did not operate on certain patients because of perceived futility of the surgical procedure given the poor neurological condition, we did an analysis by regrouping the nonoperated patients based on the need for surgery by retrospectively applying the standard BTF guidelines. The total number of patients who had undergone surgery or would have obviously required surgery as per BTF guidelines was 117 (55%) and this grouping was used for analysis.
The intracranial volume as estimated by the precise planimetric method ranged from a value as low as 955 ml to a maximum of 1,525 ml (mean: 1,219 ml, SD: 110). The mean intracranial volume in males was 1,238 ml (SD: 99 ml) and it was 1,092 ml (SD: 98 ml) in females. It was significantly higher in males (P < 0.001).
When the rough but easier, elliptical planimetric method was used, the mean cranial volume was 1,527 ml (SD: 180 ml) (Range: 1,103 ml–2,087 ml). The agreement between these two techniques were found to be good (Intraclass correlation coefficient was 0.85).
The bleed volume estimated by the precise planimetric method ranged from 34 ml to a maximum of 261 ml (median of 81 ml; mean 88 ml, SD: 37). When the bleed volume was estimated with the rough elliptical planimetric method, it had a median of 186 ml (range: 68–393 ml, mean: 189 ml SD: 57). The agreement between these two techniques was also found to be good. (Intraclass correlation coefficient was 0.82).
The edema volume estimated by the precise planimetric method ranged from 7 ml to a maximum of 89 ml (mean of 22 ml, SD: 13 ml, median: 18 ml). The mean edema volume estimated by method 2 (ellipse) was 45 ml (range 20–109 ml, SD: 15; median 43 ml). The agreement between these two techniques for estimating edematous brain volume was also found to be good (Intraclass correlation coefficient was 0.91).
BBR was the ratio of the estimated total bleed volume to that of the total intracranial volume. With the above values, we could arrive at the BBRs and these are given in [Table 1].
The mean BBR precise was significantly different in the group that required surgery compared with the group that was managed conservatively. Similarly, the bleed volumes and the ratios calculated by any of the methods were statistically different in the two groups. There were no statistically significant differences in the edema volumes and intracranial volumes of operated and nonoperated patients.
The utility of this ratio to predict the need for surgery was estimated by using ROC curves. The area under the curve for the BBR precise was 0.836 compared with 0.8224 for total bleed volume alone. Therefore, the predictive power of BBR precise was higher compared with that of total bleed volume when both volumes were estimated using the precise planimetric method. The predictive power of edema volume alone was also assessed using ROC but the area under the curve was only 0.4 [Figure 2].
We could not predict the need for surgery using the edema volume alone. Since edema volume was not of any predictive value, BEBR precise also did not gain any improvement in area under the curve compared with BBR. It had the same area under the curve as BBR precise. (ROC curve of BEBR is not shown).
Now, to answer the question of whether BBR estimated by the precise method is significantly superior compared with the predictive value of BBR estimated by simpler ellipse method, the ROC curves of both the techniques were compared. The area under the curve for BBR ellipse for predicting the need for surgery was 0.764 and this was lower than BBR precise [Figure 3].
The cutoff value of the BBR for use in clinical management was assessed using the ROC method.
We decided that as an initial diagnostic test, BBR should have a high sensitivity: we should be able to identify almost all patients who require surgery. So we thought a sensitivity of 90% was acceptable in a clinical setting where the consequences of missing out a patient who requires surgery would be more than the risk of falsely identifying a patient as an operative candidate. Patients who are detected by this diagnostic test to require surgery can be re-evaluated by the expert team to make a final decision.
From the ROC curve, we could find that a BBR of 0.0535 had a sensitivity of 90.6% and a specificity of 52.6% to predict the need for surgery. The cutoff value for BBR ellipse estimated with the faster but less precise planimetric method was 0.0895 (sensitivity of 90.6% and specificity 28.1%).
From the ROC curves, we could also find that the cutoff value of the total bleed volume for predicting the need for surgery was 62.3 ml (if we chose the identical sensitivity of 90.6%), but it had a lower specificity of 45.3%.
An exploratory sub-analysis of the relation of outcome to the BBR
Although BBR precise was correlated with the motor score at the time of acquiring the CT scan, it was not found to be significantly correlated with the motor score at the time of discharge (r = −0.162, P = 0.12), nor was it significantly correlated with a good outcome at the time of discharge (r = −0.169, P = 0.139).
The mean BBR of those who had a good outcome (n = 56) was 0.083 (SD: 0.042) and those who had a bad outcome (n = 22) was 0.071 (SD: 0.03). The mean BBR was not found to be significantly different (P = 0.15) between the outcome groups. This analysis was conducted only on those patients who had complete clinical data to derive an outcome score and thus has limited value.
The rise in intracranial pressure must occur faster in a cranial cavity that has a lesser total volume since the product of pressure and volume is a constant. If the rise in intracranial pressure occurs faster for any given volume of a bleed within a smaller cranial cavity, then it is important and intuitive to factor in the total volume of the intracranial cavity. Measuring the bleed volume in isolation might underestimate the risk of deterioration due to raised intracranial pressure in those patients having a lower than usual intracranial volume. In a patient with a relatively larger intracranial cavity, a larger volume of blood may be tolerated. Here, applying recommendations based on a fixed volume of bleed may result in unnecessary surgery when the patient could have been managed conservatively. Such considerations could be of relevance in conditions like hemorrhagic stroke too.
Another reason why we did this study is that in every day clinical practice, it is common to find multiple coexisting hematomas of various irregular sizes and shapes in the same patient. Very often, a calculation of the volume of hematoma is not made but a decision for surgery is made on visual impression. Even if a calculation of the volume of a hematoma is made, such a measurement is usually made using a rough approximation of the volume of an ellipsoid [(AXBXC)/2 method] 2. The recently terminated randomized controlled trial (STICH trauma trial) on traumatic intraparenchymal hematoma, the volume calculation of 10 ml to decide on surgery versus conservative management was based on this equation. So, it is quite possible that each of the small hematomas is below the BTF threshold for surgery but in combination exerts sufficient mass effect for surgical intervention. The BBR addresses this problem because it takes the total bleed volume into consideration.
In this study, we measured the total volume of bleed in the brain irrespective of the size, shape, kind (acute subdural hematoma, parenchymal hematoma, sub-arachnoid bleed, etc.) or the number of individual bleeds. The total bleed volume was the numerator and a precise measurement of the total intracranial volume was used as the denominator to find the BBR.
The wide range of intracranial volumes that we found between patients and genders lend some indirect support to our hypothesis that the same volume of bleed could exert a more mass effect in a smaller skull and a recommendation based only on a single of a site of bleed volume may need improvement. However, whether it truly occurs requires verification by measuring ICP in patients. We did not measure ICP in any of our patients.
The bleed volumes were significantly different in the group who underwent surgery compared with those who were managed conservatively. This was completely expected since the clinical decision for operative intervention was primarily based on the visual impression of the size of the hematoma by an experienced surgeon and because we regrouped the nonoperated patients by applying the BTF guidelines.
By factoring in the bleed volumes and the total intracranial volume into a single measure, we calculated the BBR. A BBR (precise method) of 0.0535 had a sensitivity of 90.6% and a specificity of 52.6% to predict the need for surgery. The method could identify 90% of the patients who underwent surgery, but could only identify around 50% of the people who did not require surgery.
The area under the ROC curve of the BBR was more compared with the area of total bleed volume ROC. Hence, BBR appears superior to total bleed volume alone in predicting the need for surgery.
Interestingly, edema volume was not significantly different between the patient groups. Estimation of edema volume was tricky since the Hounsfield bracketing units of 15-20 automatically included some parts of the brain like the periventricular areas where there was relatively low HU [Figure 4]. Adding edema volume to the bleed volume to generate the BEBR did not improve the predictive power. However, it is possible that in a subgroup of head injured patients with brain contusions which is prone to develop delayed perilesional edema over a period of days, BEBR may be better than BBR. We did not have a sufficient number of such patients to do the analysis to explore this possibility.
The bleed volume or BBR did not correlate with outcome. The outcome of a head trauma patient is dependent on very many factors. For example, a patient with little or no bleed could have a very low coma score due to diffuse axonal injury rather than due to mass effect of bleed and edema causing raised intracranial tension.
Limitations of the study
Although the scans of consecutively operated patients were taken, the scans of nonoperated patients were essentially a nonconsecutive, convenient sample and this could have introduced selection bias. The current cutoff value of the ratio should be again assessed for reliability and refined by testing on a future series of consecutive patients since the ratio might hold true only for this small sample of operated and conserved patients. We plan to do a prospective evaluation of the ratio in a consecutive series of patients to validate the ratio further.
The clinical decision to evacuate a hematoma is based not just on the volume of the lesion but on a wide variety of factors such as age of the patient, the clinical condition of the patient, biochemical and hematological parameters, other CT scan findings such as midline shift and obliteration of cisterns, intracranial pressure readings, and the response to medical management like anti-edema measures. We do not suggest that the BBR or any such number should replace a holistic decision-making approach. The BBR, which is based on only volumes derived from a CT scan, cannot approach the accuracy of a clinician who has access to all these factors.
However, the possibility that a ratio that takes into account the mass effect of the total bleed volume and the capacity of the cranial cavity is better than a measurement of the volume of a single bleed is introduced in this study. To the best of our knowledge, there are no other studies that have taken this relationship into consideration.
There were missing data elements regarding the clinical condition for some of the patients during the retrospective review of paper-based medical records and this could have affected the reliability of findings related to the outcome.
Planimetric methods are the most precise way of estimating the volume of a lesion in CT scans.
We found that the BBR was superior to total bleed volume alone in classifying patients who required surgery. Although the ratio estimated by the tedious planimetric method was most accurate, an easier less precise method had reasonable accuracy.
From the ROC curves, we could find that a BBR of 0.0535 estimated using the precise planimetric technique had a sensitivity of 90.6% and a specificity of 52.6% to predict the need for surgery in this retrospective analysis of a regrouped series of operated patients (and those who required surgery but not operated) and conservatively managed head trauma patients.
The BBR could be particularly useful in those head trauma patients who have multiple, irregular sized, small bleeds where the surgeon is undecided on the need for surgery. However, this technique requires further validation in a prospective, consecutive series of patients before it can be adopted for clinical or research use.
The study was done as part of the short studentship program under the ambit of the Indian Council of Medical Research.
We would like to acknowledge and thank the statistical analysis done by Dr. Harichandrakumar K.T, Assistant Professor in the Medical Biometrics and Informatics (Biostatistics), JIPMER, Pondicherry.
Financial support and sponsorship
Conflicts of interest
There are no conflicts of interest.
[Figure 1], [Figure 2], [Figure 3], [Figure 4]