The main finding of this study is that NfL on ICU admission is a strong independent predictor of mortality in critical COVID-19.

For a better understanding of the role of NfL, GFAP and tau on ICU admission in critical COVID-19, a correlation network was created, revealing that NfL correlates primarily with age, renal function, CCI, CFS, altered consciousness before ICU admission, serum procalcitonin (PCT), and indirectly with interleukin-6 (IL-6) and albumin, apart from being correlated with GFAP and tau. GFAP and tau, consequently, had similar correlation patterns.

It is well-established that NfL increases with age and that age is a strong predictor of intensive care outcomes in COVID-19—in fact, better than the ICU gold standard Simplified Acute Physiology Score 3 (SAPS-3) [2]. Due to the correlation analyses and previously described associations of NfL, the multivariable logistic regression analyses of mortality included age, creatinine, CCI, and altered consciousness before ICU admission. GFAP and tau were also predictors of one-year mortality in critical COVID-19, albeit not independent predictors.

In univariate analyses, HrQoL in survivors showed that levels of NfL and GFAP on ICU admission differed for good versus poor PCS at one year, while no differences were seen for good versus poor MCS. Levels of tau did not differ for PCS or MCS at one year. There is no clear explanation for these differences, but our results are in line with those of Needham et al. [25].

The findings that NfL, GFAP and tau on ICU admission, i.e. before lengthy intensive care, showed predictive capabilities of one-year mortality in critical COVID-19 may suggest (1) a direct effect of SARS-CoV-2 on the CNS in critical COVID-19, (2) that they measure (possibly sub-clinical) pre-existing degenerative CNS disease, or (3) that they reflect a CNS effect of the secondary organ failure (e.g. hypoxia or hypoperfusion). These three possible explanations may also be viewed in light of a proposed increased BBB permeability in critical COVID-19 [6, 7].

Since levels of NfL in the cerebrospinal fluid (CSF) are much higher than in blood, initial studies primarily focused on analysing the CSF in neurological disease. The degree to which the BBB and blood–CSF barrier permeability influence the blood NfL levels is unclear. Explorative work used the CSF and serum albumin ratio to estimate BBB permeability. However, this ratio is rather a marker of the blood–CSF barrier [35, 36]. Patients with the highest CSF-to-serum albumin ratio also had the highest CSF and blood NfL levels, suggesting that an altered permeability of the blood–CSF barrier contributes to serum NfL [37, 38]. This relationship, however, was not observed in all studies [37,38,39,40]. Of note, there is an increased disruption of the BBB with ageing which may contribute to the increased levels of blood NfL in age-related diseases [41]. The increase in blood NfL levels with age may be driven by an increasing burden of comorbidities rather than the ageing process, such as the disruption of the BBB. Blood NfL levels have also been reported to depend on renal function and sex [42, 43].

The NfL and GFAP levels in critical COVID-19 in this study are within previously suggested reference intervals in healthy individuals [15, 16, 20]. Compared to previous COVID-19 studies, however, biomarker levels in the present study, especially NfL, are similar [24, 25]. NfL and GFAP levels in our age-matched controls were also significantly lower than in critical COVID-19. The previously reported slow release of NfL compared to GFAP is clearly shown in Figs. 6 and 7 [24], which also impacts absolute levels. Blood samples were collected within a relatively short time frame in the ICU and NfL levels in critical COVID-19 have been reported to normalise over time, up to 6 months later [24]. Our samples were collected at the defined start of ICU care, corresponding to an average 11 days after the onset of symptoms [2]. The differing levels of NfL and GFAP thus seem to adequately reflect health versus critical disease.

The precise cause of elevated NfL values in non-survivors cannot be determined in the present study. We have, however, corrected for comorbidities using the Charlson comorbidity index, making a pre-existing degenerative disease a less likely cause.

We speculate that the most likely cause for NfL elevation on admission in critical COVID-19, particularly in non-survivors, reflects a secondary CNS effect due to organ failure (hypoxia or hypoperfusion and consequent CNS injury) and recommend further studies to investigate this hypothesis in critical disease. If our hypothesis is correct, NfL could be used to evaluate intensive care in general, e.g. different sedation or drug strategies, blood pressure or ventilatory targets, using a biomarker reflecting perhaps the most important endpoint of intensive care—preserving cognitive abilities and emotional well-being.

The most important limitation of our study is the small sample size, as it limits our search for a more precise cause for the higher biomarker levels of CNS injury in non-survivors.

This study presents a prospectively and carefully studied population of critically ill COVID-19 patients. Serial biomarkers of CNS injury were collected over time and batch analysed together with healthy and age-matched controls, allowing for correct comparisons. We also allowed for corrections for renal function, comorbidities, and altered consciousness prior to admission. The detailed and mainly face-to face follow-up, including GOSE and HrQoL as measured by SF-36v2®one year after critical disease, is also a strength.