In this study, we prospectively included adult patients with TBI who were admitted to the neurointensive care unit at Oslo University Hospital between January 2020 and August 2022. The reference method of measuring ICP, to which the noninvasive determinants were compared, was invasive monitoring, measured either parenchymally or intraventricularly. Transorbital ultrasound images and videos of the optic nerve sheath complex were acquired by a single experienced ultrasound operator (DFN). ONSD was measured manually at the bedside by the operator, whereas DI was calculated by semiautomated postprocessing of ultrasound videos by an investigator (RB) blinded to the invasively measured ICP at the time of processing. Patients were examined by transorbital ultrasound and were thereby included in the study, as per availability of the ultrasound operator.

Included patients were treated according to our institutional TBI management protocol, concordant with Brain Trauma Foundation guidelines [6], and inclusion in the study did not interfere with standard management. As per this protocol, maintaining ICP at ≤ 22 mm Hg was a main aim of management. Nonsurgical ICP-lowering measures included elevation of the head of the bed to 30 degrees, deep sedation, hypertonic saline, normocapnia, and thermoregulation to normothermia. Surgical ICP-lowering measures included CSF drainage and craniotomy with mass lesion evacuation. Last-tier measures included decompressive craniectomy and thiopental sedation to burst suppression.

In accordance with the Helsinki Declaration [7], proxy consent was obtained for unconscious patients at the time of inclusion. If the patient regained ability to give informed consent, this was obtained from the patient at a later stage. The study was approved by the Regional Committee for Medical and Health Research Ethics South East Norway (2018/136).

Adult patients (≥ 18 years old) with head CT scan findings consistent with TBI who were admitted to the neurointensive care unit with invasive ICP monitoring were eligible for inclusion. Invasive ICP monitoring was required to be recorded via a correctly positioned and functioning parenchymal sensor or an external ventricular drain. In patients with intraparenchymal ICP sensors, ultrasound data sets were acquired within 7 days of implantation to avoid significant drift from zero in the invasively measured ICP readings [8]. Where invasive measurements were performed via an external ventricular drain, no limitation to the in situ time was adhered to. Patients with unilateral or bilateral injuries to the orbital region were excluded from the study.

Transorbital ultrasound images and videos were acquired using a commercially available scanner (Philips Epiq 5G; Philips Healthcare, Andover, MA) with a linear array probe (Philips L14-7io). Established safety margins of ophthalmic ultrasound imaging were adhered to, with a mechanical index of less than 0.23. Serial measurements in the same patient were obtained at opportunity.

ONSD was measured at the bedside by manually identifying the optic nerve sheath 3 mm posterior to the retina in one plane. Ultrasound videos (10 s) for DI calculation were acquired after ONSD measurement for each eye by holding the probe still with the optic nerve sheath in focus. Invasive ICP was measured using either a parenchymal microsensor (Codman microsensor; Johnson and Johnson, Raynham, MA, or Raumedic Neurovent-P ICP sensor; Raumedic AG, Münchberg, Germany) or a ventricular catheter and was recorded at the time of the ultrasound examination.

Additional parameters, including physiological parameters (e.g., pulse rate, blood pressure etc.), Richmond Agitation-Sedation Scale score, details on surgical management, and the presence of CSF drainage or leak, were also acquired at the time of the ultrasound examination.

DI was calculated according to the method previously described by our research group [4, 5]. It estimates the magnitude of motion in the lateral direction on both sides of the optic nerve sheath (D1 and D2) over the cardiac cycle (Fig. 1) and quantifies the deformation of the nerve sheath complex according to the formula DI = (D1 − D2) / (D1 + D2). In this study, manual initialization of two opposite points on the optic nerve sheath was used before automatic speckle tracking of the lateral motion of the points was performed. The depth of initialization was standardized at 3 mm from the sclera. To reduce variability from the manual initialization, DI was processed three times, and resultant values were averaged for each eye. Averaged values between the two eyes were used for final analysis.

For each ultrasound video, the image quality was independently evaluated by two investigators (DFN and RB). A label of “adequate” or “too poor” image quality was applied to (1) the quality of the optic nerve sheath visualization and (2) the presence of movement artifacts. Discrepancies in “too poor” labels were reviewed to reach consensus. Examinations in which at least one quality parameter was labeled “too poor” after consensus for at least one of the eyes were categorized as “red.” Remaining examinations were categorized as “green,” representing adequate image quality.

End points and planned subgroup analyses were specified per study protocol prior to commencement of data collection. Primary end points were (1) correlation between DI, ONSD, or the combination of the two and ICP and (2) the ability of DI, ONSD, or the combination of the two to distinguish dichotomized ICP. Planned subgroup analyses included subgrouping by image quality score and by open versus closed skull vault. The latter was based on a hypothesis that any opening of the enclosed skull vault to the outside atmospheric pressure (decompressive craniectomy, CSF fistula/leak) or external resistance (open external ventricular or lumbar drain) may impact the physiology underlying DI and relates to the principles of the Monro-Kellie doctrine [9].

Correlations between ONSD and ICP and between DI and ICP were explored using the Pearson correlation coefficient R. Based on previous work, a positive correlation between ONSD and ICP and a negative correlation between DI and ICP was hypothesized. Correlations were therefore tested with a one-tailed t-test, and results reported with a one-sided 95% confidence interval (CI). Simple linear regression was used to study ICP as a function of ONSD or DI, and multiple linear regression was used for the combination of ONSD and DI. The coefficient of determination (R2) for the combined model was compared to that of the model predicting ICP based on ONSD alone, and the difference between the models was evaluated using analysis of variance.

To compare the two noninvasive parameters’ ability to distinguish dichotomized ICP, a cutoff at ICP ≥ 15 mm Hg was chosen. Additionally, cases were also dichotomized at an ICP of ≥ 20 mm Hg. For both dichotomizations, median ONSD and DI were calculated for each group and compared using a one-tailed Mann–Whitney U-test. Receiver operating characteristic (ROC) analysis was used to calculate associated areas under the curve (AUCs) with 95% CIs. Logistic regression was used to study the combination of ONSD and DI for predicting dichotomized ICP, and analysis of variance was used to compare this model to a logistic regression using ONSD alone.

Preplanned subgroup analyses by image quality score and open vs. closed cranial vault were performed by examining correlations between ICP and DI using Pearson correlation coefficients with one-sided 95% CIs in resultant subgroups. To further characterize potential physiological reasons for correlations between ICP and DI, we also performed exploratory analysis evaluating correlation between ICP amplitude and DI. Statistical analyses were performed using Matlab (R2022a; Mathworks Inc, Natick, MA), and p < 0.05 was considered significant.