Neurological Complications and Noninvasive Multimodal Neuromonitoring in Critically ill COVID-19 Patients


 Background: The incidence and clinical presentation of neurological manifestations of coronavirus disease 2019 (COVID-19) remain unclear. No data regarding the use of neuromonitoring tools in this group of patients are available. Methods: This is a retrospective study of prospectively collected data. The primary aim was to assess the incidence and type of neurological complications in critically ill COVID-19 patients and their effect on survival, as well as on hospital and intensive care unit (ICU) length-of-stay. The secondary aim was to describe cerebral hemodynamic changes detected by noninvasive neuromonitoring modalities such as transcranial doppler (TCD), optic nerve sheath diameter (ONSD), and pupillometry. Results: Ninety-four patients with COVID-19 receiving mechanical ventilation and admitted to an ICU from February 28 to June 30, 2020, were included in this study. Fifty-three patients underwent noninvasive neuromonitoring. Neurological complications were detected in 47/94 patients (50%), with delirium as the most common manifestation. Patients with neurological complications, compared to those without, had longer hospital (36.8±25.1 vs. 19.4±16.9 days, p <0.001) and ICU (31.5±22.6 vs. 11.5±10.1 days, p <0.001) stay. The duration of mechanical ventilation was independently associated with risk of developing neurological complications (OR 1.100, 95%CI 1.046-1.175, p=0.001). Patients with increased intracranial pressure (ICP) measured by ONSD (19% of the overall population) had longer ICU stays. Conclusions: In conclusion, neurological complications are common in critically ill patients with COVID-19 receiving invasive mechanical ventilation and are associated with prolonged ICU length-of-stay. Multimodal noninvasive neuromonitoring systems are useful tools for early detection of cerebrovascular changes in COVID-19. Registration number: 163/2020


Background
Coronavirus disease 2019 , caused by the severe acute respiratory syndrome-coronavirus-2 (SARS-CoV-2) (1-4), is primarily a disease of the respiratory system, leading to a variety of clinical manifestations including dry cough, fever, fatigue, and respiratory failure (4). However, recent data suggest that COVID-19 is not con ned to the airways, but is also responsible for a syndrome of multiorgan dysfunction (MODS-CoV-2), including possible neurological involvement (5,6).
Coronaviruses may pass to the central nervous system by different routes (7,8), including hematogenous spread from the systemic to the cerebral circulation and lymphocyte invasion or dissemination from the cribriform plate and olfactory bulb to the brain (9,10). These hypothesis seem to be consistent with the loss of smell and taste described as-rst atypical, then quite prevalent-presentations of COVID-19 (11).
However, the neurologic manifestations of COVID-19 are highly variable, and can occur prior to diagnosis or as a complication late in the course of infection (7,8).
Noninvasive neuromonitoring systems are widely used in Neurointensive care settings for patients with primary cerebral damage; more recently, they are also being employed in critically ill patients in general as useful tools to detect neurological complications (22,23). In particular, transcranial Doppler (TCD) ultrasonography, optic nerve sheath diameter (ONSD) measurement, and quantitative pupillometry are safe, useful methods that can be applied at the patient's bedside to assess cerebral hemodynamics, as well as to monitor cerebral perfusion pressure and intracranial pressure noninvasively (22,23). To date, no studies have investigated cerebral hemodynamics in patients with COVID-19.
The primary aim of our study was to describe the type and frequency of neurological complications in a cohort of critically ill patients with COVID-19 receiving invasive mechanical ventilation in an intensive care unit (ICU) and the effects of these complications on outcome. As a secondary aim, we sought to assess changes in cerebral hemodynamics, their effects on outcome, and their role as potential predictors of neurological complications in a subgroup of patients who underwent noninvasive neuromonitoring (ONSD, TCD, and pupillometry).

Study design
This is a single-center, retrospective, observational study of prospectively collected data. The study was carried out during the COVID-19 pandemic, from February 28 through June 30, 2020, at the ICU of the San Martino Policlinico Hospital (SMPH) IRCCS for Oncology and Neurosciences, Genoa, Italy. The SMPH is the main hospital serving both the metropolitan area of Genoa (approximate population 840,000) and the wider Liguria Region (approximate population 1,543,000). The usual ICU capacity is 52 adult beds, increased to 74 during the peak of the SARS-CoV-2 outbreak in Italy. The study protocol followed Good Clinical Practice principles in compliance with the Declaration of Helsinki. Approval was obtained from the Ethics Committee of Liguria, Italy (registry number 163/2020), which waived informed consent for participation because of the retrospective nature of the study.

Study population
Patients aged ≥18 years, con rmed positive for SARS-CoV-2 infection by reverse transcriptasepolymerase chain reaction (RT-PCR) of nasopharyngeal swab specimens at the moment of ICU admission, and who were critically ill requiring invasive mechanical ventilation were eligible for inclusion.
Patients who were not neurologically evaluable due to life-threatening respiratory failure resulting in use of sedatives were excluded, as were those who died before sedation could be weaned.

Data collection
Overall population The following data were collected from patients' electronical records at the time of ICU admission: age in years; gender; body mass index (BMI) in kg/m 2 ; sequential organ failure assessment (SOFA) score (24); a series of comorbidities, namely hypertension, diabetes mellitus, respiratory disease (de ned as asthma or chronic obstructive pulmonary disease), end-stage renal disease (de ned as estimated glomerular ltration rate <15 mL/min/1.73 m 2 ), moderate/severe liver disease (de ned as compensated/decompensated liver cirrhosis) (25), and cancer, The highest C-reactive protein (normal range 0-5 mg/L) and D-dimer (normal range 0-500 mcg/L), as well as the lowest partial pressure of oxygen (PaO 2 ) (normal range 72-104 mmHg), were collected from daily test results throughout each patient's ICU stay. At the time of ICU and hospital discharge, data on ICU length of stay (ICU-LOS) (days), overall hospital LOS (days), duration of mechanical ventilation (days), neurological complications (type and number), and mortality were collected.

Neuromonitoring cohort
The following data were collected from patients who underwent noninvasive neuromonitoring during the day of assessment and throughout their ICU stay: ventilatory parameters [type of ventilation, positive endexpiratory pressure (PEEP) in cmH 2 O, pressure control or pressure support in cmH 2  Noninvasive neuromonitoring systems Ultrasound measurements were performed by two experienced operators (de ned as having received more than 5 years of training and performing more than 70 examinations/year) (DB, CR) and three mentored trainees in Anesthesia and Intensive Care (KC, FI, MB). MAP, heart rate, mean cerebral artery (MCA) ow velocities (diastolic, mean, and systolic), and ONSD were recorded during ICU stay, according to the clinical context and need (availability of personal protective equipment and clinical rationale).

Transcranial doppler (TCD)
A low-frequency (2 MHz) microconvex transducer (Philips SparQ ® ) was used to investigate intracranial vessels. The temporal window was preferred for passage of the Doppler signal for MCA assessment. Systolic (sFV), diastolic (dFV), and mean ow velocity (mFV) in the MCA were collected. MAP was also measured. The pulsatility index (PI) was calculated as the mean value between the right and left MCA ow velocities using the following formula [13]: Noninvasive ICP (nICP TCD ) was calculated according to the formula: where cerebral perfusion pressure (CPPe) was calculated as follows (27): Intracranial pressure (ICP) values > 20 mmHg were considered indicative of intracranial hypertension (27).

Optic nerve sheath diameter (ONSD)
A linear probe (Philips SparQ ® ) was used for ONSD evaluation. The probe was placed on the closed upper eyelid, and ONSD was evaluated 3 mm behind the retinal papilla. Two measurements were obtained from each optic nerve, the rst in the transverse plane and the second in the sagittal plane (28). Noninvasive intracranial pressure measured by ONSD (nICP ONSD ) was derived from a mathematic formula described elsewhere in the literature (29,30). Again, ICP values > 20 mmHg were considered indicative of intracranial hypertension (27).

Automated pupillometry
Pupillary light reactivity was measured by a handheld quantitative automated pupillometer (Neurolight Algiscan ® , ID-MED, Marseille, France) in both eyes. This device measures quantitative variation in pupillary light reactivity by using an infrared camera to record video footage of changes in the pupillary surface. Pupillary light reactivity was assessed by a calibrated light stimulation (320 lux for 1 second) with a precision limit of 0.05 mm. Quantitative reactivity was expressed as the percentage of pupillary light response, and baseline pupil size was expressed in mm. The pupillary constriction velocity (mm/sec) was also reported (31)(32)(33). Abnormal pupillary reactivity was de ned as an abnormal pupillary light re ex as reported by the pupillometer (e.g., a weaker than normal or "sluggish" pupil response) (34).

Statistical analysis
The results are expressed as mean ± standard deviation, median, 1 st quartile (Q1), 3 rd quartile (Q3), interquartile range (IQR), and absolute and relative frequencies. No sample size calculation was performed due to the retrospective design of this study. The Shapiro-Wilk test was used to assess the normality of distribution of continuous variables. The Mann-Whitney U-test was used to compare continuous variables, while categorical variables were compared with Fisher's exact test. Patient survival was estimated by the Kaplan-Meier method; the log-rank test was used to compare survival curves.
Continuous and categorical variables were entered into univariate Cox proportional hazard regression models, which returned regression coe cients and hazard ratios (HRs) with 95% con dence intervals (CIs) as the main outputs. The Efron approximation was used for each Cox model. The proportional hazards assumption for each signi cant Cox regression model was evaluated using correlation coe cients between transformed survival times and scaled Schoenfeld residuals. Variables signi cant on univariate Cox regression which satis ed the proportional-hazards assumption were carried forward to the multivariate model. A forest plot and a rank-hazard plot were provided for multivariate regression. The rank-hazard plot is able to visualize the relationship between the relative hazard of variables entered in a multivariate Cox regression model [37]. Logistic regression was performed to assess the risk factors associated with neurological complications. The Hosmer-Lemeshow omnibus test was used for goodness-of-t evaluation of each signi cant logistic regression model. Variables signi cant on univariate logistic regression were entered in the multivariate model, with regression coe cients and odds ratios (ORs) with 95%CIs as the main outputs. A receiver operating characteristic (ROC) curve was calculated for the multivariate logistic regression model, as well as sensitivity and speci city. Statistical signi cance was accepted at a two-tailed P-value <0.05 for all tests. Statistical analyses were conducted in the R software environment (version 3.6.3; R Foundation for Statistical Computing, Vienna, Austria).

Results
During the study period, 116 patients with COVID-19 were admitted to the SMPH ICU. Twenty-two patients were excluded because they did not meet the inclusion criteria. Thus, 94 patients were included in the nal analysis, of whom 53 underwent noninvasive neuromonitoring.

Overall population
The characteristics of the 94 patients admitted to our ICU who ful lled the inclusion criteria-with and without neurological complications-are described in Table 1. Neurological complications were detected in 47/94 patients (50%). Nine patients presented more than one neurological complication. The most common complications are reported in Table 2. Occurrence of neurological complications did not result in increased ICU mortality (p = 0.450) (Figure 1), but was associated with longer hospital (36.77±25.14 vs. 19.43±16.86 days, p <0.001) and ICU (31.51±22.64 vs. 11.51±10.14; p <0.001) stay compared to absence of neurological complications (Table 1).
No reliable Cox regression model could be obtained by entering the variables collected for the overall population (data not shown). On univariate logistic regression, duration of mechanical ventilation and CRP values were associated with risk of developing neurological complications (Table 3). Multivariate logistic regression demonstrated that the duration of mechanical ventilation was independently associated with the risk of neurological complications (OR: 1.1; 95% CI: 1.046-1.175; p = 0.001) ( Table 3) Figure 3). Additional data from the comparison between patients with normal ICP and those with high ICP as calculated by TCD and ONSD are reported in the ESM - Table 7. The outcomes of the Cox regression models for the patients who underwent noninvasive neuromonitoring are reported in ESM - Table 8 Table 4.

Discussion
The main ndings of our study are: 1) neurological complications are common in COVID-19 patients and have no effect on mortality, but can be associated with increased hospital and ICU length of stay; 2) the duration of mechanical ventilation is independently associated with the development of neurological complications; and 3) increased ICP (estimated by ONSD) and pupillary abnormalities are common, and associated with longer ICU length-of-stay.
To our knowledge, this is the rst study describing cerebrovascular dynamics in mechanically ventilated COVID-19 patients, which could potentially help to elucidate the underlying pathophysiology of neurological complications in this patient population. Moreover, to date, no studies have taken into account the possible secondary effects of mechanical ventilation and in ammation on neurological outcome.
There are several theories concerning central and peripheral neurological changes following SARS-CoV-2 infection: viral neurotropism, including trans-synaptic spread, endothelial or lymphocyte invasion by SARS-CoV-2; a hyperin ammatory and hypercoagulative state; or even mechanical ventilation-associated impairment (35). In our cohort, neurological complications were detected in half of the patients admitted to our ICU with con rmed COVID-19 pneumonia who ful lled the inclusion criteria. The most frequent complication was delirium (36.70%), followed by coma, critical illness neuropathy, ischemic stroke, stupor, encephalopathy, seizures, cognitive de cit, and depression. The frequency of delirium is in line with current COVID-19 literature, in which it has ranged from 26.80-73.60% (36,37). Delirium was identi ed both in the acute and in the post-ICU phases during the severe acute respiratory syndrome (SARS) and Middle-East respiratory syndrome (MERS) epidemics, with a possible detrimental effect on length of stay (38). Sedatives, analgesics, pain, psychological stressors, hypoxia, metabolic and electrolyte imbalances, infection, hyperthermia, sepsis, mechanical ventilation, light, and the use of physical restraints are well-known contributors to delirium occurrence in the ICU (39,40). Delirium is known to be associated with longer ICU stay and mechanical ventilation days, as well as increased risk of death at 6 months, disability, and long-term cognitive dysfunction (40,41). Our results are in line with these ndings; patients who developed neurological complications (mainly delirium) did not show increased ICU mortality, but they did have prolonged hospital and ICU stays, often exceeding 2 weeks, with a major impact on health expenditures and resource utilization-especially in the resource-limited setting of a pandemic.
Mechanical ventilation days and in ammation (assessed by C-reactive protein) were associated with the occurrence of neurological complications at the univariate analysis. This suggests that the magnitude of the in ammatory response and the severity of respiratory impairment may strongly affect the occurrence of neurological complications in COVID-19 (35).  (42)). As described in the literature, the threshold of increased nICP ONSD is 5-6 mm (28); this suggests that increased ICP is a common nding in COVID-19 patients. In fact, increased ICP measured with both ONSD and TCD was very common, and a large portion of patients (38.71%) exhibited altered pupillary reactivity.
Several factors can potentially cause increased intracranial pressure in patients with respiratory failure and pneumonia, including increased PaCO 2 , which can cause cerebral vasodilatation (43,44), or the use of high PEEP and consequent increased intrathoracic pressure (45). Indeed, we found that PEEP was higher in those who showed higher nICP, whether assessed by ONSD or TCD (as we reported in the ESM).
Although the difference was not statistically signi cant, it suggests that mechanical ventilation can interfere widely with cerebral hemodynamics.
Although common, the occurrence of increased ICP had no effect on cumulative probability of survival; it did prolong ICU-LOS when measured by ONSD, but not by TCD. This con rms that, in COVID-19 patients, noninvasive ICP monitoring may be essential for early detection of patients who are at risk of longer ICU-LOS with subsequent complications and di cult recovery. The incongruity between the results of the two noninvasive methods might be explained by differences in pathophysiological sensitivity and speci city for ICP assessment between the two (27); however, recent ndings suggest that ONSD is much more accurate than TCD for measurement of intracranial pressure in non-COVID-19 patients (46).
Although we found no correlation between altered neuromonitoring ndings and the occurrence of neurological complications, we strongly recommend the use of these methods in critically ill patients with COVID-19. Noninvasive neuromonitoring tools are safe, quick, low-cost, easily available, and can provide relevant data at the patients' bedside.

Limitations
This study has several limitations which must be addressed. First, this was a retrospective study of prospectively collected data. Therefore, data were collected within the clinical context of the COVID-19 pandemic (limited availability of personal protective equipment, clinical reasons, and so on). Thus, neuromonitoring data are neither complete nor available for all patients. Moreover, TCD, ONSD, and pupillometer measurements were intermittent, and were obtained at different stages of the patients' ICU stays. Continuous, daily, standardized monitoring would have provided more accurate data on the behavior of cerebrovascular hemodynamics in this population. Furthermore, because of the critical demands of the pandemic, we were unable to obtain multiple neuromonitoring measurements to reduce intra-and inter-observer variability among the operators. However, our team consists of a group of specialized physicians with ample experience in the use of noninvasive monitoring. The relatively small sample size of our study-which depended on the number of COVID-19 patients admitted to our ICU and was thus beyond our control-limits the strength of our conclusions and results, but this depends. Finally, as this is not an interventional study, the sedation and analgesia protocols were not standardized, but rather were based on the clinical needs of the patients, which may have had an impact on FV, ONSD, and pupillometer-derived values.

Conclusions
Neurological complications-particularly delirium-are common in COVID-19 patients and are associated with longer hospital and ICU stay.

Con ict of interest
The authors declare that the research was conducted in the absence of any commercial or nancial relationships that could be construed as a potential con ict of interest.
Authors' contribution DB, CR conceived of the study, designed the study, acquired data, interpretation of the data, and drafted the manuscript. GS analysis of data, interpretation of the data, critical revision of the manuscript and

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. SupplementalMaterialCC.docx