Optic Nerve Diameter on Non-Contrast Computed Tomography and Intracranial Hypertension in Patients With Acute Brain Injury: A Validation Study

Abstract

Intracranial hypertension is a feared complication of acute brain injury that can cause ischemic stroke, herniation, and death. Identifying those at risk is difficult, and the physical examination is often confounded. Given the widespread availability and use of computed tomography (CT) in patients with acute brain injury, prior work has attempted to use optic nerve diameter measurements to identify those at risk of intracranial hypertension. We aimed to validate the use of optic nerve diameter measurements on CT as a screening tool for intracranial hypertension in a large cohort of brain-injured patients. We performed a retrospective observational cohort study in a single tertiary referral Neuroscience Intensive Care Unit. We identified patients with documented intracranial pressure (ICP) measures as part of their routine clinical care who had non-contrast CT head scans collected within 24 h, and then measured the optic nerve diameters and explored the relationship and test characteristics of these measures to identify those at risk of intracranial hypertension. In a cohort of 314 patients, optic nerve diameter on CT was linearly but weakly associated with ICP. When used to identify those with intracranial hypertension (> 20 mm Hg), the area under the receiver operator curve (AUROC) was 0.68. Using a previously proposed threshold of 0.6 cm, the sensitivity was 81%, specificity 43%, positive likelihood ratio 1.4, and negative likelihood ratio 0.45. CT-derived optic nerve diameter using a threshold of 0.6 cm is sensitive but not specific for intracranial hypertension, and the overall correlation is weak.

Introduction

Intracranial hypertension is a feared complication of acute brain injury. If left untreated, raised intracranial pressure (ICP) can cause secondary injury including ischemic stroke, herniation, and death.¹ Lumbar punctures in such cases are often contraindicated, and therefore definitive diagnosis of intracranial hypertension requires placement of an external ventricular drain or alternative invasive device, exposing the patient to the risk of infection² and hemorrhage.³ On the other hand, sustained intracranial hypertension greater than 20 mm Hg has been associated with worse outcomes in traumatic brain injury (TBI),⁴ and guidelines advocate for invasive neuromonitoring in select cases, and early treatment on this basis.⁵

The triaging decision to escalate care and intervene is challenging and clinical examination is often confounded by toxic or metabolic encephalopathy, seizure or post-ictal encephalopathy, or polypharmacy. Non-contrast computed tomography (CT) is widely employed to aid diagnosis; however, many of the reported signs of raised ICP (basal cistern compression, midline shift, Marshall scores) have relatively modest test characteristics when trying to risk-stratify patients,⁶ and often require specialist interpretation from neuroradiologists.

Transorbital measurement of the optic nerve sheath diameter (OND) on point-of-care ultrasound has become an increasingly popular tool because of its favorable test characteristics for predicting raised ICP. Meta-analysis of 10 studies estimated a pooled area under the receiver operator curve (AUROC) of 0.94 with sensitivities and specificities mostly in the 80–95% range.⁶ The optimal threshold for concern varies between studies; however, the largest studies suggest thresholds in the region of 0.59 cm^7,8 to 0.64 cm.⁹ Its use though has been limited by the requirement of experienced operators. Magnetic resonance imaging (MRI) has been similarly employed, with optimal reported thresholds of 0.58 cm.⁹

Given the widespread availability of CT and the knowledge that OND on MRI and CT are highly correlated,¹⁰ we explored the relationship between OND as measured on CT, and ICP. Prior work was able to demonstrate a linear relationship in a select patient cohort with TBI undergoing portable bedside CT and simultaneous ICP recording, and showed that using a threshold of 0.6 cm, OND on CT had a sensitivity of 97% to identify patients with intracranial hypertension.¹¹

In what follows, we report a validation study of OND measurement using CT in a large heterogenous cohort of acutely brain-injured patients. We identified patients admitted to our Neuroscience Intensive Care Unit with invasive ICP monitoring devices, and then identified non-contrast CT head scans taken with 24 h of documented ICP measurements. We hypothesized that OND was linearly related to ICP in this heterogenous cohort of patients, and that using a threshold of 0.6 cm, as used in prior work using CT, ultrasound, and MRI, we could screen for those at high risk for intracranial hypertension.

Methods

We performed a retrospective observational cohort study in a single tertiary referral Neuroscience Intensive Care Unit. Procedures followed were in accordance with the Helsinki Declaration of 1975. The Institutional Review Board at Weill Cornell Medicine approved this study (Approval number: 1309014352, Date: 10/14/2022, study title: “Weill Cornell Stroke Registry”) and waived the requirement for informed consent.

Patient population

We used our Neurocritical Care Electronic Data Utilization System (NUCLEUS), which contains comprehensive data on all patients admitted to the New York-Presbyterian/Weill Cornell Medical Center Neuroscience Intensive Care Unit between 2008 and 2020, to identify patients who had documented ICP measurements as part of their routine clinical care. ICP data were either transduced from external ventricular drains (EVDs), or from parenchymal monitors (“bolts”), with pressures measured in millimeters of mercury (mm Hg). Routine clinical care for these patients requires hourly documentation of ICP in the electronic medical record, which is then automatically extracted into a Microsoft SQL database as part of NUCLEUS. Of this cohort, we then identified those patients who had non-contrast CT head scans taken within 24 h of a documented ICP. We included patients who were older than 18 years of age, with any acute brain injury, if they had invasive ICP monitoring, a matching CT scan, and at least one optic nerve that could be reliably measured on their scan. Patients were categorized by class of brain injury, namely: 1) stroke, 2) trauma, 3) neoplastic, or 4) other.

Computed tomography and OND measurement

CT scanning was performed as part of routine clinical care with any of our general-use CT scanners (Optima 660/Lightspeed Xtra/Lightspeed Pro-16/Discovery HD-750; General Electric, Fairfield, CT, USA) or dedicated portable CT scanner (CereTom; Samsung Healthcare, Seoul, South Korea). Three investigators (JK, neurologist; HO, radiology resident; HE, medical student) blinded to the patient's clinical or demographic data measured bilateral OND in a random 10% of the total cohort for training and calibration purposes and to determine inter-rater reliability. The remaining measurements were performed by a single investigator.

CT images were viewed on a standard radiology Picture Archiving and Communication System (PACS) unit (Philips, Amsterdam, the Netherlands). The images were windowed using a standard abdominal preset for optimal contrast between the optic nerve complex and the surrounding orbital adipose tissue (level/center: 60 Hounsfield units; width: 360 Hounsfield units). Each investigator was recommended to magnify the images to best identify and measure the optic nerve. For each optic nerve, an optimal axial slice was identified that transected the nerve at its maximum diameter. For most patients the slice thickness was 5 mm (32 slice); however, if the patient happened to have higher resolution imaging (2.5 mm or 1 mm slices) that better transected the nerve, higher resolution imaging was used. The OND was measured 3 mm behind the globe, perpendicular to the linear axis of the nerve (Fig. 1). The orbital segment of the optic nerves are encased within the three layers of the meninges, with cerebrospinal fluid (CSF) traversing a trabeculated subarachnoid space.¹² The surrounding meninges and subarachnoid space are called the optic nerve sheath, which itself is surrounded by retrobulbar fat. The nerve, its sheath, and likely some of the dural reflection contribute to the final measurement.

FIG. 1.

Example labeled OND measurement from an axial slice of a non-contrast CT head. The right globe is highlighted in yellow, and the right optic nerve is highlighted in teal. The linear axis of the nerve is drawn with a yellow dashed line. The black calipers measure 3 mm behind the posterior aspect of the lamina cribosa. The red calipers measure the OND, perpendicular to the linear axis of the nerve. CT, computed tomography; OND, optic nerve sheath diameter.

Statistical analysis

We used Stata (StataCorp, TX, USA) and RStudio (PBC, Boston, MA, USA) utilizing the “psych” package.¹³ The sample size was of convenience. The intraclass coefficient (ICC) was calculated to assess the between-rater agreement of OND measurements for a random 10% subset of the cohort using a one-way analysis of variance (ANOVA) fixed effects model.¹⁴

For each patient, the diameter of each optic nerve was measured by a single investigator, and the largest OND (OND_max) was extracted. The maximum was chosen instead of the mean or other summary statistic because we wanted to prioritize sensitivity over specificity, and ease of use at the point of care. For each OND_max value, all ICP measurements taken within 24 h of the scan were used, meaning each OND_max had multiple associated ICP measurements.

Three regression analyses were performed. The first tested for a linear relationship between OND_max (independent variable) and ICP (dependent variable), corrected for within-patient effects using cluster-robust standard errors.¹⁵ As a sensitivity analysis, a second regression was performed that additionally included brain injury class as an independent variable, to explore any interaction between OND_max and brain injury class in relation to ICP. Finally, a zero-inflated negative binomial regression model tested for a relationship between OND_max (independent variable) and the area over the curve (AOC), defined mathematically as the product of the time spent ICP >20 and the magnitude of ICP elevation beyond that threshold.

We then assessed the test characteristics of OND_max for identifying patients with intracranial hypertension, defined as ICP >20 mm Hg. The receiver operator curve (ROC) was estimated using a non-parametric method based on Lorenz-like curves,¹⁶ and AUROC was calculated for the total cohort, as well as for each brain injury class, and subgroups for sex and age (above and below mean). Observed-versus-expected plots were used to assess the calibration of our classifier.

Finally, to validate prior thresholds reported using CT, we selected the OND threshold of 0.6 cm and report the sensitivity and specificity for identifying patients with intracranial hypertension.

Secondary analyses

To assess whether the time between collection of imaging and ICP measurement impacted the utility of OND measurements, the primary analysis was repeated using only ICP measurements recorded within 6 h of neuroimaging, and only if those measurements were recorded after neuroimaging occurred.

To assess whether OND differed in patients with grossly disturbed CSF flow dynamics, we repeated our primary analysis in a sub-cohort of patients who did not have a diagnosis of hydrocephalus, or craniectomy.

Results

Baseline characteristics

A total of 314 patients had documented ICP measurements and an associated CT scan with at least one optic nerve within the field of view of the CT scan (Table 1). The cohort was broadly representative of patients typically admitted to our Neuroscience Intensive Care Unit, 46.5% female, with brain injuries from various etiologies including stroke (that includes subarachnoid and intracerebral hemorrhage), neoplastic, trauma, and other. “Other” encompassed etiologies including infection and congenital hydrocephalus. Among the 314 patients, there were 5698 ICP measurements, meaning each OND_max reading was associated with on average 18 separate ICP measurements. The mean (standard deviation) ICP measurement was 8.9 (6.3) mm Hg, and 3.2% of readings qualified as intracranial hypertension (> 20 mm Hg). On a patient-level, of the 314 patients, 57 had at least one measurement that qualified as intracranial hypertension. The class of injury impacted the prevalence of intracranial hypertension, with patients with TBI having significantly more qualifying events (p < 0.001). In a similar fashion, the class of injury also impacted the prevalence of raised OND measurements, defined as OND >0.6 cm (p = 0.02).

Table 1.

Baseline Characteristics of Patient Cohort

	Total cohort	Stroke	Neoplastic	Trauma	Other
N	314	141	41	26	106
Age	58.7 (16.1)	61.9 (14.5)	57.0 (16.6)	63.7 (17.9)	53.9 (16.2)
Female	146 (46.5)	71 (50.4)	28 (68.3)	6 (23.1)	41 (38.7)
OND_max, cm	0.62 (0.09)	0.62 (0.08)	0.61 (0.07)	0.63 (0.10)	0.61 (0.10)
ICP, mm Hg	8.9 (6.3)	8.8 (5.9)	7.5 (4.7)	10.9 (6.6)	9.1 (7.1)
ICP >20, %	3.2	2.4	0.8	8.3	3.6
OND >0.6, %	58.0	57.5	56.0	64.2	57.5

Continuous data are presented as mean (standard deviation), categorical data as N (%).

CT, computed tomography; ICP, intracranial pressure; OND, optic nerve sheath diameter; OND_max, the largest measured OND from a single CT scan.

Inter-rater reliability of OND

Using the most conservative approach, the ICC of the measures of three raters measuring a random selection of 59 optic nerves was 0.79 (0.72–0.84; p < 0.001), consistent with excellent agreement.¹⁷ A summary plot can be found in Figure 2A.

FIG. 2.

Scatter plot (A) summarizing the inter-rater agreement of OND measurement in a random 10% of the identified cohort, corresponding to 59 measured optic nerves. The gray line represents perfect inter-rater agreement. Scatter plot (B) summarizing the relationship between OND and ICP. The red line represents the line of best fit, with gray shaded 95% confidence interval. Calibration plot (C) with expected-vs.-observed probabilities of high ICP. OND, optic nerve sheath diameter; ICP, intracranial pressure.

The linear association between OND and ICP

Linear regression analysis revealed a significant association between OND and ICP, adjusted for clustering within patients (β = 11.9 [6.6–17.2], p < 0.001). However, the R-squared of this model was only 0.03. As a sensitivity analysis, a second regression analysis was performed including the class of brain injury, and the interaction between OND_max and brain injury class. This second model remained significant, with little impact on the linear relationship demonstrated prior (β = 10.3 [2.9–17.7], p = 0.006); the R-squared of this second model was only 0.04. A third model exploring the relationship between OND and the area above the curve did not reveal a significant association.

OND as a screening test for intracranial hypertension

The AUROC for the total cohort was 0.68 (standard error 0.02). The test characteristics varied slightly by injury class, performing slightly better in patients with TBI (0.75) and neoplasm (0.73) compared with stroke (0.58) or other pathologies (0.67). When we excluded those with hydrocephalus or a hemicraniectomy, the AUROC was 0.63 (standard error 0.03). Sub-group analyses revealed similar AUROC between male (0.67) and female (0.65) patients, as well as between older (0.67) and younger patients (0.67). A detailed report of the sensitivity and specificity at different thresholds can be found in Supplementary Table S1.

Using a threshold of 0.6 cm, OND had the following performance characteristics: sensitivity 81%, specificity 43%, positive predictive value (PPV) 4%, negative predictive value (NPV) 99%, positive likelihood ratio (PLR) 1.4, and negative likelihood ratio (NLR) 0.45.

When the analysis was repeated with ICP measurements recorded in the 6 h after neuroimaging, the AUROC was 0.64 (standard error 0.04). Using the same 0.6 cm threshold, OND had the following performance characteristics: sensitivity 76%, specificity 42%, PPV 6%, NPV 97%, PLR 1.3, and NLR 0.55.

Discussion

In this single-center retrospective study, we attempted to validate the use of OND derived from non-contrast CT of the brain to predict intracranial hypertension. Specifically, we sought to validate a previously derived threshold of 0.6 cm. We found that OND measurement required minimal training or resources and showed excellent agreement between raters of different clinical experience and expertise. Using data collected as part of routine clinical care, we demonstrated a weak linear association between OND and ICP in a heterogenous cohort of patients receiving neurocritical care for various brain injuries. The previously proposed OND threshold of 0.6 cm was relatively sensitive at identifying intracranial hypertension, with similar test characteristics to prior CT-based analyses,^11,18 despite a much larger and more heterogenous cohort. However, the specificity was poor. This study represents the largest CT-based study of its kind; Sekhon and colleagues initially reported a similar relationship in 57 patients with TBI,¹¹ and Martin and associates reported comparable results in 54 patients with TBI.¹⁹

Current guidelines recommend invasive ICP monitoring in patients with severe TBI (Glasgow Coma Scale [GCS] score <9) who have a CT head scan with evidence of hematomas/contusions, mass effect, herniation, or compressed basal cisterns. In patients with severe TBI with a normal CT scan, ICP monitoring is recommended if the patient has any two of the following: 1) age >40 years, 2) unilateral or bilateral motor posturing, or 3) systolic blood pressure <90 mm Hg.²⁰ In patients without TBI, consensus recommendations for consideration of invasive ICP monitoring include GCS score <9, cerebral edema, mass effect, hydrocephalus, intraventricular hemorrhage on neuroimaging, or neurological worsening.²¹

The biological mechanism for the relationship between OND and ICP has been extensively investigated in pre-clinical models,²⁰ in post-mortem human data,^21,22 and in human volunteers undergoing intrathecal infusion,²³ and it appears to be a result of shunting of CSF from the intracranial compartment to the intra-orbital compartment as a means of pressure equilibration. The precise anatomical landmarks that are encompassed in the measurement of the OND depend on the spatial resolution of the technology. Ultrasound boasts an ability to visualize the nerve, sheath, and dura, allowing precise measurements of the sheath itself, albeit with some disagreement regarding the signal properties of different anatomical substrates within the nerve-sheath complex.²⁴ Some of this complexity likely contributes to heterogeneity within the literature when it comes to an optimal measurement threshold.⁶ The limited resolution of CT makes measurements simpler but sacrifices some of the precision afforded by ultrasound.

We found that using a threshold of 0.6 cm, CT-based OND is a sensitive tool but is vulnerable to false-positive results. This threshold is consistent with most of the ultrasound literature, the largest studies of which quote thresholds between 0.59 cm²⁵ and 0.64 cm.⁸ This is the largest cohort to date where the relationship between CT-derived OND and invasively measured ICP has been assessed.

The discriminative ability of CT-measured OND in our study is clearly less robust than MRI measures showing an AUROC of 0.94,⁹ and ultrasound measures that on meta-analysis have shown an AUROC of 0.94.^6,25 Some of this might be due to patient selection; much of the ultrasound literature focuses on TBI, and when our analyses are limited to patients with TBI exclusively, the test performance does show modest improvement (AUROC of 0.68 vs. 0.75). However, some of this discrepancy is likely due to both the spatial resolution of CT with regards to the optic nerve-sheath complex, and our experimental design. Specifically, we included ICP measures within 24 h of a measured OND, and so here we assess the ability of the OND to identify patients with any ICP >20 within a 24-h period.

On the basis of our results, the utility of CT-derived OND appears limited. If it were to be used, it would be best used in a clinical scenario where there is equivocal pre-test probability. Using the negative likelihood ratio presented above and the heuristic proposed by McGee,²⁶ an OND_max of less than 0.6 cm will reduce the post-test probability of raised ICP by between 15% and 20%. So, in scenarios where the pre-test probability is high, the value of CT-based OND is limited; however, in equivocal situations, the presence of a negative test may meaningfully reduce the post-test probability.

In reference to current TBI guidelines⁵ and non-TBI consensus recommendations,¹⁸ on the basis of our work, we believe the added value of CT OND would be limited to scenarios where a patient has a GCS score of <9, without neuroimaging stigmata of mass effect, but with unexplained neurological worsening. In those scenarios, if the OND_max is <0.6 cm, the concern for intracranial hypertension potentially might be lowered such that invasive monitoring could be avoided. However, it should be kept in mind that the sensitivity of the test is only roughly 80% and that the clinical sequelae of missing intracranial hypertension may be catastrophic. If the pre-test probability is already raised, for example in patients with neuroimaging showing midline shift or ventricle/cisternal compression, the OND_max should not be used as a reason to avoid invasive neuromonitoring.

Limitations

Our results should be considered in light of their limitations. Firstly, although this is a validation study, it is a retrospective observational study that relied on accurate record keeping as part of routine clinical care. Although this might be more representative of a real-life clinical environment, it is vulnerable to selection bias and measurement error. The study was also performed at a highly specialized referral center for neurocritical care, which might limit generalizability to other intensive care units, wards, or emergency departments. The indication for CT collection was also not defined, but presumably it was driven by a clinical concern for deterioration or reassessment of stability, which might also impact the results.

Further, although the raters were encouraged to focus purely on the OND and not review the rest of the scan in detail, it is possible that measurement bias could be introduced by findings of more striking brain injury. The proximity between neuroimaging and ICP measurement is also important to note given the dynamic nature of ICP. The purpose of this study was to assess the use of OND as a screening tool for intracranial hypertension and not as a surrogate measure of minute-to-minute changes in ICP. As such, in our primary analysis, the interval between neuroimaging and ICP measurements was set to maximize the statistical power of a potentially useful clinical inference (24 h). When this interval was reduced to 6 h, and only to scenarios whereby neuroimaging preceded ICP measurement, the results were qualitatively similar. It is notable that our sub-group analysis seemed to show worse performance in diseases where it would be hoped that OND would be most clinically useful (i.e., in the non-TBI patients).

Finally, our study had a relatively low prevalence of intracranial hypertension, which is likely due to most of the patients having EVDs in situ. The low prevalence impacts certain test performance characteristics as discussed above. This, in combination with a lack of standardized measurement time-points or imaging acquisitions limits the utility of our data set. The presence of EVDs in these patients undergoing OND measurements in theory means that normal CSF flow dynamics have been disrupted. When we repeated our analysis in those without grossly abnormal CSF flow dynamics (i.e., those with either established hydrocephalus or a hemicraniectomy), the AUROC was relatively similar; however, the impact of the EVD itself could not be assessed in this study.

Conclusions

In this large single-center retrospective study, we attempted to validate the use of non-contrast CT of the brain to measure OND as a means of risk stratifying patients for intracranial hypertension. A previously proposed OND threshold of 0.6 cm was sensitive for raised ICP, but not specific. When the pre-test probability of intracranial hypertension is equivocal, the presence of a negative test reduces the post-test probability by roughly 15–20%. Whether this is enough to sufficiently avoid invasive testing in select clinical scenarios remains to be seen. Prospective studies are required to best assess the performance of CT-derived OND, and other non-invasive measures, as screening tools for intracranial hypertension.

Transparency, Rigor, and Reproducibility Summary

This study was not pre-registered, and the sample size was not determined by power calculation, rather by convenience sample. Imaging acquisition and analyses were performed by team members blinded to relevant characteristics of the participants. All equipment and software used to perform imaging, pre-processing, and analysis are widely available from commercial sources. The key inclusion criteria and outcome evaluations are established standards. Certain data elements are available upon request of the corresponding author. This article will be published under a Creative Commons Open Access license, and upon publication will be freely available online.

Authors' Contributions

The listed authors met criteria for authorship as set out by set by the International Committee of Medical Journal Editors. All authors approved the final manuscript.

Funding Information

There was no funding provided for this research.

Author Disclosure Statement

JK, HO, HE, JHC, MRS, and AG declare that they have no competing financial interests. AEM and SM have served as expert consultants on neurological disorders. SM is in receipt of National Institutes of Health (NIH) funding (K23NS105948). HK serves as a principal investigator for the NIH-funded ARCADIA trial (National Institute of Neurological Disorders and Stroke [NINDS] U01NS095869), which receives in-kind study drug from the BMS-Pfizer Alliance, and ancillary study support from Roche Diagnostics; on clinical trial steering/executive committees for Medtronic, Janssen, and Javelin Medical; and on end-point adjudication committees for AstraZeneca, Novo Nordisk, and Boehringer Ingelheim. HK has an ownership interest in TETMedical, Inc.

Supplementary Material

Supplementary Table S1