Ninety-eight patients hospitalized in the Geneva University Hospitals from April 2020 to May 2021, were diagnosed with SARS-CoV-2 infection and presented signs of AE. SARS-CoV-2 infection was confirmed by a positive SARS-CoV-2 reverse transcription polymerase chain reaction assay from a nasopharyngeal swab at the time of hospitalization. Out of those patients, 36 patients met the criteria of severe AE and were evaluated for GC administration.

The decision to treat with high-dose GC was based on an algorithm conceived by a multidisciplinary expert panel and approved by the institutional review board of the Geneva University Hospitals. Specifically, the criteria to treat with GC included (i) a Richmond agitation sedation scale (Sessler et al. 2002) (RASS) < -3 or the combination of an RASS ≥ -3 and (1) a confusion assessment method (Inouye et al. 1990) (CAM) ≥ 3 or (2) mutism, (ii) the exclusion of brain lesion, status epilepticus, other common causes of delirium (hypoxic, metabolic, toxic, etc.), and encephalitis that could explain the neurological symptoms, (iii) the presence of gadolinium enhancement at the level of intracranial arterial walls (assessed by three board-certified neuroradiologists and validated when common agreements were reached) (Fig. 2E and F), and (iv) the absence of signs of spontaneous remission within 48 h of protocol inclusion.

After careful consideration, 12 patients, who met all four criteria, were treated with high-dose GC (GT group). Treatment protocol consisted of the intravenous administration of 500 mg of methylprednisolone per day for 5 days, followed by prednisone 1 mg/kg (max. 80 mg/day) with tapering schedule. The non-treated (NT) group included 24 patients who met all first 3 criteria but who did not receive GC because of spontaneous improvement within the 48 h after the first neurological evaluation (Fig. 1). Clinical data was retrospectively retrieved from patients’ charts.

Flowchart of inclusion/exclusion criteria for GC, for hospitalized COVID-19 patients presenting with AE

The study was approved by the institutional review board of Geneva University Hospitals (protocol #2020-01206–approved May 25, 2020) and has been performed in accordance with the Declaration of Helsinki.

All patients in the GT and NT groups underwent an MRI on a Philips Ingenia (Philips Medical Systems, Eindhoven, The Netherlands) 1.5 T scanner equipped with a head and neck coil at the Geneva University Hospitals, within on average 6 days from AE symptoms onset. Time lag between the neurological symptoms onset and MRI acquisition was comparable between groups (mean GT = 7.18 \(\pm\) 6.3 days, mean CG = 5.5 \(\pm\) 4.1 days, p = 0.349).

Sequences included a T2-weighted (repetition time (TR): 3600 ms, echo time (TE): 100 ms, slice thickness 4 mm), susceptibility-weighted images (SWI, TR: 520 ms, TE: 0 ms, slice thickness 2 mm), a fluid-attenuated inversion recovery imaging (FLAIR, TR: 9000 ms, TE: 120 ms, slice thickness 4 mm), and a dynamic 3D contrast-enhanced MRA (TE: 17 ms, TR: 400 ms, image thickness 1.5 mm) of the neck vessels was performed from the aortic arch to the circle of Willis. Pre-contrast and post-contrast fat-saturated T1-weighted black blood VISTA images in all patients were acquired in the axial and coronal planes.

The following imaging features were considered: the number of vessels presenting walls enhancement, microbleeds, white matter hyperintensities (WMH), lacunes, cortical superficial siderosis, and enlarged perivascular spaces (EPVS) (Fig. 2).

Imaging features explored: microbleeds on SWI (A), white matter hyperintensities (B), EPVS (C), and lacunes (D) on T2-weighted, and endotheliitis as suggested by enhancement of the vessel wall of the left vertebral artery on VISTA sequences after gadolinium injection (F vs pre-contrast E)

Contrast enhancement of the vessel walls was retained as a possible marker of inflammation when homogeneous and circumferential (greater than 50% of the circumference) (Uginet et al. 2022a). Inflammatory atheromatous plaques, as a potential cause of such intracranial vessel enhancement, were excluded by angio‐MR, angio‐CT, or echo‐Doppler. Enhancement detection and number of vessels involved were ascertained by two board-certified neuroradiologists and validated when common agreements were reached.

The severity of white matter hyperintensities was assessed using the Fazekas scale (Fazekas et al. 2002).

The microbleeds anatomical rating scale (MARS) (Gregoire et al. 2009) was used to quantify the number and location of microbleeds, which were subsequently grouped into lobar, deep, and infratentorial.

The burden of EPVS was assessed on axial T2-weighted images in the basal ganglia and centrum semi-ovale and stratified into: < 10 EPVS, 11–20 EPVS, and > 20 EPVS. Lacunes were defined as rounded or ovoid lesions, with a diameter between 3 and 20 mm, with CSF signal density on T2-weighted images.

Rating was performed independently by two investigators (AR and GB), under the supervision of the head of the Neuroradiology Unit (KOL). In the case of discrepant ratings, a consensus was accomplished subsequently, by a third person (GA). As microbleeds have been associated to critical illness, we compared microbleeds rate between patients admitted or not to ICU.

The CSVD, cerebral amyloid angiopathy (CAA), and hypertensive arteriopathy (HTNA) scores was computed according to a previous study (Lau et al. 2017). Specifically, for CSVD score, one point was given for presence of (i) lacunes, (ii) 1–4 microbleeds, (iii) moderate-to-severe BG EPVS (> 20), and (iv) moderate WMH (total periventricular + deep WMH score 3–4), while 2 points were given for the presence of (i) ≥ 5 CMBs, (ii) severe WMH (total periventricular + deep WMH score 5–6). For CAA score, one point was given for the presence of (i) 1–4 lobar microbleeds, (ii) moderate-to-severe semi-oval center EPVS (> 20), (iii) deep WMH ≥ 2 or periventricular WMH > 3, and (iv) focal cSS, and two points for (i) ≥ 5 lobar microbleeds and (ii) disseminated cSS. For HTNA score, one point was allocated for (i) lacunes, (ii) periventricular WMH > 3 or deep WMH 2–3, (iii) ≥ 1 deep microbleeds, and (iv) presence of moderate-to-severe basal ganglia EPVS (> 10 EPVS).

The non-parametric Mann–Whitney U test was used to compare clinical and imaging features between groups. Spearman correlation test was used to correlate clinical and imaging features in the whole group. Statistical analysis was performed using Matlab (R2023a, Natick, Massachusetts: The MathWorks Inc).