The prognostic value of optic nerve sheath diameter in patients with subarachnoid hemorrhage

This was a retrospective, multicenter, observational study of adult patients with SAH admitted to the neurosurgical intensive care unit at ChungBuk National University Hospital (CBNUH) and Samsung Medical Center (SMC) between January 2012 and June 2017. The study was approved by the institutional review boards of CBNUH (CBNUH 2018-08-014-001) and SMC (SMC 2018-07-154). The requirement for informed consent was waived due to the retrospective study design. We included patients with (1) SAH admitted to the neurosurgical intensive care unit during the study period, (2) brain computed tomography (CT) within 12 h from the onset of SAH, (3) follow-up brain CT within 24 h of treatment for ruptured aneurysm, and (4) follow-up brain CT within 48 h from the onset of SAH if they were not treated for aneurysm. Of these patients, we excluded patients (1) under age 18; (2) with malignancy and whose expected life span was less than 1 year; (3) with insufficient medical records; (4) with a history of head trauma, neurosurgery, cardiac arrest, or chronic neurological abnormality on admission; (5) transferred from other hospitals after more than 24 h of SAH onset; or (6) with orbital anomaly, orbital mass lesions, and ocular or retro-orbital injury. We analyzed a total of 223 patients diagnosed with SAH in this study (Fig. 1).

In this study, we defined poor-grade SAH as grades 4 to 5 according to Hunt and Hess (H-H) [1, 4, 5]. The primary outcome was neurological status 6 months later assessed with a Glasgow Outcome Scale (GOS, 1 to 5) [12]. Patients with GOS scores of 3, 4, and 5 indicated favorable neurological outcomes whereas GOS scores of 1 and 2 indicated poor neurological outcomes. We thoroughly reviewed medical records, and two independent neurologists measured the patients’ GOS scores. Initial brain CT angiography and follow-up CT were performed within a specified time window. All the CT studies were performed using 64-channel scanners (Brilliance 64, Philips Medical Systems, Best, the Netherlands, at CBNUH and Light Speed VCT, GE Healthcare, Milwaukee, WI, USA, at SMC) with a 5-mm-slice width. Brain CT images were reviewed by two independent neurologists (SL and JAR). Investigators who were blinded to clinical information evaluated each of the patients’ CT scans using commercial image-viewing software (Picture Archiving and Communication System; Maroview 5.3 Infinitt Co., Seoul, Republic of Korea, at CBNUH and Centricity RA1000 PACS Viewer, GE Healthcare at SMC). The ONSD and eyeball transverse diameter (ETD) were measured using the same initial CT and subsequent scans. The ONSD was measured at a distance of 3 mm behind the eyeball, immediately below the sclera in a perpendicular vector with reference to the linear axis of the nerve (Fig. 2a) [11‐13]. The images were changed to the “chest/abdomen” window (window width 300 and window level 10) and magnified fourfold on the particular image slice that demonstrated the largest diameter of the optic nerve sheath [12]. The ONSD was measured from one side of the optic nerve sheath to the other as a section through the center of the optic nerve [13]. The transverse diameter of the eyeball was chosen because the ONSD is usually measured in the transverse plane [14]. ETD was defined as the maximal transverse diameter of the eyeball measured from one side of the retina to the other (in-to-in, Fig. 2b) [14, 15]. The ONSD_average and the ETD_average measured for the patient’s left and right eyes were averaged to yield a mean value. The ONSD_index was defined as median ETD (22.7 mm) multiplied by the average value of bilateral ONSDs divided by the average value of bilateral ETDs (22.7 × ONSD_average/ETD_average). The intracranial pressure (ICP) at the time of follow-up CT was designated as the immediate ICP after the CT scan.

Clinical and laboratory data were collected by a trained study coordinator using a standardized case report form. All data are presented as means ± standard deviations (SD) for continuous variables and numbers (percentages) for categorical variables. We compared data using the Student’s t-test for continuous variables and chi-square test or Fisher’s exact test for categorical variables. A scatter plot was drawn to ascertain the relationship between simultaneously measured ONSD and ICP. We calculated Pearson’s correlation coefficient (r) to evaluate the correlation between ONSD and ICP. We assessed the predictive performance of ONSDs and ONSD indices using the areas under the curve (AUCs) of the receiver operating characteristic (ROC) curves for sensitivity vs. 1-specificity. We compared AUCs using the nonparametric approach published by DeLong et al. [16] for two correlated AUCs. We obtained optimal cutoffs for each ONSD and their modifications to predict poor neurological outcomes by ROC curve and Youden index [17, 18]. We used the bidirectional elimination technique in the stepwise selection method to select clinically and statistically meaningful predictor variables. The adequacy of the prediction model was also determined using the Osius and Rojek test of goodness-of-fit [19]. All tests were two-sided, and p < 0.05 was statistically significant. We analyzed the data using IBM SPSS version 20 (IBM, Armonk, NY, USA).

The mean age of patients was 58.7 ± 13.0 years, and 84 (37.7%) patients were males. Hypertension (39.5%) and smoking (26.9%) were the most common comorbidities among the SAH patients. Eighty-three (37.2%) patients were H-H grades 4 or 5, and 96 (43.0%) were WFNS grades 4 or 5. WFNS grade was higher in patients included under the poor neurological outcome group compared with the favorable neurological outcome group (4.4 ± 1.1 vs. 2.4 ± 1.5, p < 0.001). Fisher grade and modified Fisher grade were also higher in the poor neurological outcome group compared with favorable neurological outcome group (3.9 ± 0.3 vs. 3.2 ± 0.7, p < 0.001 and 3.8 ± 0.4 vs. 2.9 ± 1.0, p < 0.001, respectively). Baseline characteristics of SAH patients are presented in Table 1. Anterior communicating artery (29.6%) and middle cerebral artery (25.1%) were the most common locations of a ruptured aneurysm. However, no aneurysm or ruptured aneurysm was detected in 14 patients (6.3%). Ruptured aneurysm was treated within 72 h in most patients (88.8%). Endovascular coiling of the ruptured aneurysm was performed in 124 (55.6%) patients, and surgical clipping was performed in 75 (33.6%) patients; delayed cerebral ischemia was observed in 46 (20.6%) patients. Treatment characteristics of SAH patients are presented in Table 2. Two hundred two patients (90.6%) survived to discharge. Of these 202 survivors, 186 (83.4%) had favorable neurological outcomes (GOS of 3, 4, or 5, Fig. 1). Among the 83 (37.2%) patients with poor-grade SAH, 66 survived to discharge (79.5%), and 51 (61.4%) had favorable neurological outcomes.

In this study, initial and follow-up ONSDs in the poor neurological outcome group were significantly greater than in the favorable outcome group (Table 3). However, initial ETD and follow-up ETD did not differ significantly between the favorable and poor neurological outcome groups (P = 0.504 and P = 0.295, respectively). In addition, changes in ONSD (∆ONSD_average) between initial and follow-up CT were not associated with prognosis. Although invasive ICP monitoring was performed in 80 (35.9%) patients during hospitalization, the ICP was measured in only 21 (9.4%) patients at the time of follow-up CT. Based on simple correlation analysis, the ICP showed a linear correlation with follow-up ONSD_average and follow-up ONSD_index (r = 0.525, P = 0.036 and r = 0.500, P = 0.048, respectively) (Fig. 3). In ROC curve analysis for prediction of poor neurological outcomes (Fig. 4a), the AUCs of ONSD_index were greater than those of ONSD_average during the initial CT and follow-up CT. However, no significant differences existed between the predictive performance of ONSD_index and ONSD_average during the initial CT and follow-up CT (P = 0.764, P = 0.086, respectively). The performance of ONSD_index in follow-up CT was significantly better than that of ONSD_index and ONSD_average in initial CT (P = 0.033, P = 0.022, respectively). The C-statistic of follow-up ONSD_index was 0.812 (95% confidence interval (CI) 0.754 to 0.861). A cutoff > 6.22 had a sensitivity of 70.3% (95% CI 53.0 to 84.1%) and a specificity of 80.7% (95% CI 74.2 to 86.1%). The C-statistic of H-H grade was 0.853 (95% CI 0.799 to 0.897). A cutoff > 3 had a sensitivity of 86.5% (95% CI 71.2 to 95.5%) and a specificity of 72.6% (95% CI 65.6 to 78.9%). Although no significant differences existed between the AUCs of H-H grade, the follow-up ONSD_average, and the follow-up ONSD_index, the composite of H-H grade and the follow-up ONSD_index [AUC 0.896, Brier score 0.078, goodness-of-fit (Osius-Rojek) Z = − 0.5052, p = 0.613] showed the highest ability to correctly predict poor neurological outcomes rather than either marker alone (all p < 0.006, Fig. 4b). Based on our definitions of poor markers as H-H grade > 3 and follow-up ONSD_index > 6.22, 107 (48.0%) patients had one or more poor markers (sensitivity 91.9% and specificity 60.8%). Of these patients, 39 (17.5%) carried two poor markers (H-H grade > 3 [poor-grade SAH] and follow-up ONSD_index > 6.22, sensitivity 64.9%, and specificity 91.9%).