Study design

We performed a prospective cohort study at Rijnstate hospital, The Netherlands. Consecutive comatose patients after cardiac arrest were included for daily ONSD measurement, in addition to standard care. Patients were included between December 2019 and October 2021. The medical ethics committee Arnhem–Nijmegen approved the study protocol and waived the need for informed consent prior to study inclusion (2019–5586). In case of patient survival up to 72 h, deferred consent was obtained from the patient and/or relatives. The study is registered (ClinicalTrials.gov identifier: NCT04084054).

Study population

Consecutive comatose patients after cardiac arrest (in-hospital and out-of-hospital) were included within 24 h after cardiac arrest. Inclusion criteria were: Glasgow Coma Scale (GCS) ≤ 8 at admission, age ≥ 18 years, and admission to the intensive care unit (ICU). Exclusion criteria were pregnancy, traumatic brain injury, relevant eye surgery in medical history, pre-existing dependency in daily living (cerebral performance category (CPC) 3–4), or any known progressive brain illness, such as a brain tumour or neurodegenerative disease.

Standard of care

Patients were monitored and treated according to local protocols that were in line with international guidelines for comatose patients after cardiac arrest [1]. Targeted temperature management at 36 °C was induced as soon as possible after arrival at the ICU and maintained for 24 h. After 24 h, passive rewarming was controlled and normothermia was actively maintained. Patients generally received a combination of propofol, midazolam, and morphine for sedation and analgesia.

Decisions on withdrawal of treatment

Withdrawal of life sustaining treatment (WLST) was considered at ≥ 72 h after cardiac arrest, during normothermia, and off sedation. Decisions on WLST were based on European guidelines [1, 20] at the discretion of the treating physicians. The ONSD was never included in decisions on WLST and treating physicians were blinded to ONSD measurements.

Neurological outcome

Neurological outcome was assessed at 3–6 months after cardiac arrest by a standardized telephone interview by one of two researchers (MV, HK), blinded to ONSD measurements, according to the cerebral performance categories (CPC). Neurological outcome was dichotomized as “good” (CPC 1–2: no to moderate disability) or “poor” (CPC 3–5: severe disability, vegetative state, or death).

Study endpoints

The primary study endpoint is increase in sensitivity for predicting poor and good outcome after adding ONSD measurements to established parameters. Secondary endpoints include feasibility, inter- and intra-observer reliability, and differences in ONSD between patients with good and poor outcome.

Data acquisition and analysisEEG

Continuous EEG recordings were started in all patients as soon as possible after arrival at the ICU, always within 24 h after cardiac arrest, and continued for at least 3 days or until a patient’s decease or awakening, as part of standard care. Twenty-one electrodes were placed on the scalp according to the international 10–20 system. EEG recordings were performed using a Nihon Kohden system (VCM Medical, The Netherlands) from the study start to March 2021 and a BrainRT system (OSG, Belgium) from April 2021 onwards. Two reviewers (MV, HK) independently classified anonymized EEG epochs at 6, 12, 24, 36, 48, and 72 h after cardiac arrest blinded to the timing of the epoch, a patient’s clinical status, medication, and outcome. In case of disagreement, consensus was obtained by the consultation of a third reviewer (JH). EEG patterns were classified as suppressed with or without superimposed synchronous activity, continuous, or other patterns [4].

SSEP

SSEP recordings were performed off-sedation using a Nicolet EDX system (Natus Medical Inc., USA) as part of standard care at the treating physician’s request, generally between 48 and 72 h in patients who remained comatose after restoration of normothermia. Bilaterally absence of N20 responses was considered predictive of poor neurological outcome.

Pupillary light reflexes (PLR)

PLR were tested daily by treating physicians and categorized as present or bilaterally absent. Bilaterally absent PLR > 72 h after cardiac arrest was considered predictive of poor neurological outcome.

ONSD

ONSD was measured daily by trained personnel in the first 3 days after cardiac arrest, or until decease or awakening. Three consecutive measurements per eye were performed each day using an Affiniti 70C ultrasound system (Philips, The Netherlands). A linear probe with a frequency range of 3–12 MHz was used. Sterile ultrasound gel was placed on the probe and a sterile probe cover was placed over it, preventing ultrasound gel from touching the eye. The probe was placed transversally on the superior lateral part of the upper eyelid, angled caudally and medially with the head of the patient 30° elevated. No pressure was put on the eye. The field was reduced to a depth of 4 cm. The ONSD was measured 3 mm behind the retina [21] at the transition from the hyperechoic retrobulbar fat to the hypoechoic line, in the presence of hyperechoic striped bands, or at the transition from the hyperechoic retrobulbar fat to the hypoechoic region of the optic nerve, in absence of striped bands. These marker placements both correspond to the outer edges of the dura mater [22] (Fig. 1). The mean of three binocular ONSD measurements per day was used for further analysis.

Fig. 1

Examples of ultrasound images of the ONSD for a patient with good neurological outcome (left, ONSD = 5.01 mm) and a patient with poor neurological outcome (right, ONSD = 8.10 mm). The eyeball and optic nerve including its sheath are delineated in blue. The red horizontal line indicates the ONSD

Statistical analyses

Data are presented as mean ± standard deviation for continuous normally distributed data or median with interquartile range [IQR] for non-normally distributed data. To compare patients with good and poor outcome on a group level, we used Chi-squared tests for ordinal, and unpaired t tests or Mann–Whitney U tests for continuous variables.

Inter-observer reliability was assessed based on the ultrasound images of 10% of the included patients (n = 10), who were selected at random. One reviewer (MV) re-measured the ONSD of these patients offline, blinded for the original measurement. Intra-observer reliability was assessed based on three consecutive measurements per eye per day. Inter-observer and intra-observer reliability were calculated using the intraclass correlation coefficient (ICC) based on a two-way mixed-effects model with absolute agreement [23].

To test the additional predictive value of ONSD measurements on top of established parameters, we created a logistic regression model and a mixed-effects logistic regression model. The logistic regression model included two categorical variables (EEG classified as suppressed after 24 h, continuous within 12 h, or other; SSEP classified as not absent (not tested or present) or absent). PLR was excluded from the analysis because of the low frequency (n = 1) of bilaterally absent PLR in our cohort. ONSD measurements were normalized to z-scores. A mixed-effects logistic regression model with random intercept was trained (70% of data) and validated (30% of data). Fixed effects were the EEG, SSEP, ONSD, and time (days 1, 2 or 3). Study ID was used as a random effect. Predictive values of the models were evaluated using receiver operating characteristics (ROC): area under the curve (AUC), sensitivity to predict poor outcome at 100% specificity, sensitivity to predict good outcome at 90% specificity, and positive and negative likelihood ratios for predicting poor and good outcome. Additional predictive value of ONSD measurements was assessed by an increase in sensitivity for predicting poor and good outcome and a decrease in the Akaike information criterion (AIC). We checked for multicollinearity between our predictors using the variance inflation factor (VIF). Multicollinearity was assumed if VIF ≥ 5. Additional predictive value of ONSD measurements was assessed in the full data set and a subset (after exclusion of patients with a non-neurological cause of death). Sample size calculations were based on 10 patients per outcome group for every predictor added to the model. Adding 5 predictors (ONSD day 1, day 2, day 3, EEG, and SSEP), indicated the need of 100 patients (with an expected distribution of good vs. poor outcome of 50/50).

P values < 0.05 were assumed statistically significant. All statistical analyses were performed using R version 4.0.0.