Neurofilament light chain for prognostication after cardiac arrest-first steps towards validation

Abstract

Background

After cardiac arrest, many patients remain comatose, and a substantial proportion do not survive. Neuroprognostication is essential for identifying patients with potential for recovery, and those with severe, irreversible hypoxic-ischemic brain injury. Neurofilament light chain (NfL) is a blood-based marker of neuronal injury that is under evaluation for neuroprognostication. NfL have potential advantages over the currently only guideline recommended blood biomarker for neuroprognostication, neuron-specific enolase, including earlier applicability. However, there is no consensus on optimal NfL cut-off levels. A previous large investigation in OHCA patients, identified NfL thresholds with high specificity for poor outcome, and the purpose of the present investigation is to validate these cutoffs.

Methods

The Blood Pressure and Oxygenation Targets in Post Resuscitation Care (BOX) trial included OHCA patients who were comatose at admission. Patients with at least one plasma biobank sample available at 24–48 h were included in this investigation. NfL was quantified by ELISA. Cerebral performance category score was estimated at 1 year. Diagnostic precision of NfL for prediction of poor neurologic outcome (CPC > 2) was determined by area under the receiver operator curve (AUROC), and the performance of previously identified cut-offs for a specificity of 100% were investigated.

Results

A total of 638 patients had a NfL measurement at either 24 or 48 h. The AUROC for prediction of poor neurologic outcome was 0.95 and 0.95 at 24 and 48 h, respectively. At 24 h, a cut-off of 1232 pg/mL had a specificity of 98%, for prediction of poor neurologic outcome, and false-positive results for 7 patients (1.4%). At 48 h, a cut-off of 1539 pg/ml similarly had a specificity of 98%, and false-positive results for 7 patients (1.3%).

Conclusions

The results of this investigation confirm the prognostic value of NfL for identification of risk of poor neurologic outcome after cardiac arrest. Previously identified cut-offs of 1232 pg/mL at 24 h, and 1539 pg/mL at 48 h, performed excellent with a very high specificity. This indicates that application of NfL will allow for reliable neuroprognostication as early as 24 h after cardiac arrest.

Trial registration:

ClinicalTrials.gov NCT03141099, registered April 30 2017

Supplementary Information

The online version contains supplementary material available at 10.1186/s13054-025-05579-1.

Introduction

During cardiac arrest, the brain rapidly depletes its energy reserves, placing resuscitated patients at high risk of hypoxic-ischemic brain injury [1]. Consequently, a substantial proportion of resuscitated out-of-hospital cardiac arrest (OHCA) patients are comatose at hospital admission and many remain unconscious for several days [2]. At this stage neuroprognostication is initiated to objectively assess who has a chance of meaningful recovery and who has suffered severe irreversible hypoxic-ischemic brain injury. The predominant cause of death for these patients is severe brain injury, and death often occurs after a decision of withdrawal of life-sustaining therapies based on neuroprognostication [1]. Therefore such decisions must be founded on very robust and reliable information and any risk of falsely identifying a poor prognosis should be minimized [1]. For this purpose, a multimodal approach is employed, including neurologic examination, brain imaging, electroencephalography (EEG) and other neurophysiologic tests, as well as blood biomarkers of brain injury [1, 3]. Currently, neuron-specific enolase (NSE) is the only bloodborne biomarker of brain injury recommended by guidelines for neuroprognostication [3]. There are, however, potential confounders for the evaluation of brain injury based on NSE levels, including potential falsely elevated levels in case of hemolysis [4, 5]. Neurofilament light chain (NfL), is a novel biomarker with potentially better diagnostic performance and earlier applicability than NSE [6–8]. However there is a lack of confirmatory studies and consensus on specific cut-off levels to apply in a clinical setting.

In neurons, the protein NfL forms part of the cytoskeleton and has been found to increase in the plasma phase of circulating blood in response to neuronal damage [9]. Accordingly, the concentration of NfL has been found to provide information on the extent of neuronal injury in neurodegenerative diseases, traumatic brain injury, as well as cardiac arrest [6, 10–13]. After cardiac arrest concentrations of NfL gradually increases in the circulation in case of neuronal injury, and have been evaluated in plasma as well as serum with good overall diagnostic precision for both, however limited data exist on which cut-off and which sample type to apply for identification of patients at high risk for a poor neurologic outcome after cardiac arrest, which negates inclusion in clinical guidelines [3, 6, 10, 12–14]. So far the largest study with proposed cut-off levels was performed by Moseby-Knappe et al. and based on serum biobank samples from the “Targeted Temperature Management after Cardiac Arrest” (TTM) trial [6]. In that study NfL was found to have superior performance compared to NSE after OHCA.

The purpose of the present investigation is to validate the cut-offs previously suggested by Moseby-Knappe et al. in a large cohort of OHCA patients. For this purpose, plasma samples obtained in the “Blood Pressure and Oxygenation Targets in Post Resuscitation Care” (BOX) trial were utilized to validate NfL cut-offs previously identified at 24 and 48 h after cardiac arrest [6, 15].

Methods

The BOX trial

The BOX trial was a randomized controlled trial in which resuscitated cardiac arrest patients who remained in coma at hospital admission were randomized in parallel to three physiologic interventions: higher or lower blood pressure target, a liberal or restrictive oxygenation target, and a longer or shorter total duration of device-based fever prevention [16]. The trial was approved by the ethics committee of the Capital Region of Denmark prior to initiation (see Declarations section for further details), and patients were enrolled at two tertiary heart centres in Denmark from 2017 till 2021. Patients with an out-of-hospital cardiac arrest, adult age, a presumed cardiac cause of the arrest, and coma at admission were considered for inclusion. The trial is registered at ClinicalTrials.gov with the trial identification number NCT03141099, registered April 30 2017. Trial design, primary results as well as 1-year follow-up have been published [15–19]. All three interventions were neutral [15, 17–19].

General treatment principles and neuroprognostication

All patients were comatose at hospital admission and were treated initially in an intensive care unit according to the international treatment guidelines which were current at the time of conduct of the trial [20]. For patients who remained comatose after cessation of sedation, multi-modal neuroprognostication was performed, and decisions on withdrawal of life sustaining treatments (WLST) were withheld until at least 72 h after cardiac arrest [20].

Follow-up and outcome data

Medical history, characteristics of the arrest, as well as in-hospital care were recorded, and patients were followed for one year after cardiac arrest for functional neurologic outcome and all-cause mortality. Cerebral performance Category (CPC) at one year was assessed by medical chart review by dedicated research personnel (medical doctors and research nurses) experienced in assessing neurologic and functional performance based on this score. Higher CPC scores indicate worse outcomes, and a dichotomous outcome of a favourable or poor neurologic outcomes was calculated based on a CPC of 1–2 or 3–5, respectively. Vital status at 1-year was assessed by medical chart review which is linked to the Danish national person data registry.

Measuring technique for determination of NfL levels in plasma

NfL was quantified in ethylenediamine tetraacetic acid (EDTA) plasma samples, that upon sampling had been spun at 2000 g for 10 min, at then frozen and stored in a biobank at −80 °C until time of analyses after the completion of the trial. Quantification was performed by using the Meso Scale Discovery (MSD) R-PLEX^® electrochemiluminescence Human Neurofilament L assay (MSD, Rockville, USA) and read on a Meso QuickPlex SQ 120MM (MSD, Rockville, USA) at the Department of Clinical Biochemistry (Rigshospitalet, Copenhagen, DK). Plasma samples were diluted 2-fold with assay diluent, and assay diluent was used as a blank. A subset of samples was analysed as duplicates (10%) and the remaining samples as singlets. The obtained CV was ≤ 15%. The calibration curve was fitted with a 4-parameter logistic model with a 1/y² weighting and used for calculation of NfL concentrations. A calibrator standard CV ≤ 15% and recovery range 80–120% (mean recovery = 100.3%) was found acceptable. For a comprehensive description see supplementary methods in supplement.

Measuring technique for determination of NfL levels in serum

NfL was also measured for a subset of patients in serum samples, that upon sampling had been spun at 2000 g for 10 min, and then frozen and stored at −80 °C until time of analyses after completion of the trial. Quantification was similarly determined by ELISA with a R-PLEX^® Human Neurofilament L Antibody assay and were read on a MESO QuickPlex SQ120 (MSD, Rockville, USA) [21].

Statistics

Categorical variables are presented as numbers (%) and compared by Fisher’s exact or χ2 as appropriate. Continuous variables are presented as median with interquartile range (IQR) and compared by the Wilcoxon rank sum test. Median levels of NfL at 24 h were compared between the MAP and PaO2 target interventions by the Wilcoxon rank sum test. Similarly, the median levels were compared at 48 h for all interventions individually. Logistic regression analysis tested associations between NfL levels at 24 and 48 h and neurologic outcome and all-cause mortality, and at 48 h for unawake patients at 48 and 72 h. Diagnostic performance for poor outcome was tested with receiver-operating characteristic analysis by calculating the area under the receiver-operating curve (AUROC). Confidence intervals for the AUROC were estimated by the DeLongs method. Serum NfL cut-off values with specificities and sensitivities have been published previously by Moseby-Knappe et al., and here we externally validate those and report sensitivity, specificity, true-positive, false-negative, false-positive, and true-negative rates for all-cause mortality and poor neurological outcome at each suggested serum cut-off value [6]. To identify specific cut-offs for prediction of neurologic outcome and morality in the present cohort, receiver operating characteristic (ROC) analysis was conducted to identify optimal NfL cutoff values at 24 and 48 h, targeting specificities of 100%, 99%, and 98%, for predicting of CPC-3-5 at one year and 365-day mortality. For each cutoff, sensitivity and 95% confidence intervals were calculated using stratified bootstrap resampling with 2000 replicates. Cut-offs with a specificity of 100% in the present cohort for CPC 1–2 and survival at one year were also identified. All statistical analyses were done in R, version 4.5.0 (The R Foundation for Statistical Computing).

Results

A total of 789 patients were included in the BOX trial, of these 638 patients had at least one plasma NfL measurement at 24–48 h and were hence included in this investigation, constituting 80.9% of the BOX trial intention to treat cohort. The number of available plasma samples in patients alive at 24 and 48 h was 523 and 560, respectively. This corresponds to a sample being available for 82% and 88% of patients alive at 24 and 48 h. Patient characteristics of those with compared to those without a sample can be seen in Supplemental Table 1; Reasons for unavailable samples other than the patient had expired included, biobank late initiated at one centre (initiated after 50 patients included at site), and no or poor quality of sample. Key differences in patients with or without a sample, include a higher fraction of patients with a sample being categorized as awake at 36–72 h and alive at 1 year; these differences mainly arise from patients dying early not being available for sampling. For the cohort of patients with at least 1 sample at either 24 or 48 h the median age was 63 years (interquartile range (IQR): 53–72), and 522 (82%) were men, also see Table 1. At one-year follow-up, 432 patients (68%) were alive with a median CPC score of 1 (1–1).

Table 1.

Patient characteristics

Characteristics	N = 638
Age, years, median (IQR)	63 (53, 72)
Female sex, no. (%)	117 (18%)
Bystander CPR, no. (%)	554 (88%)
Time to ROSC, min, median (IQR)	18 (12, 25)
ST-elevation in first ECG, no. (%)	277 (44%)
Left ventricular ejection fraction at arrival, %, median (IQR)	35 (25, 45)
Lactate at arrival, mmol per liter, median (IQR)	4.8 (2.9, 7.8)
Immediate coronary angiography, no. (%)	576 (90%)
Of which received immediate percutaneous coronary intervention, no. (%)	262 (46%)
Awake at 36 h, no. (%)	173 (27%)
Awake at 48 h, no. (%)	311 (49%)
Awake at 72 h, no. (%)	404 (63%)
One-year mortality, no. (%)	206 (32%)
Cerebral performance category scale at one year, n (%)
1	376 (60%)
2	33 (5.3%)
3	9 (1.4%)
4	1 (0.2%)
5	206 (33%)

CPR, cardiopulmonary resuscitation; ROSC, return of spontaneous circulation. For characteristics of patients without a sample compared to those with at least one sample at 24–48 h please refer to Supplemental Table 1

The randomized interventions of a higher versus lower blood pressure, and a liberal versus restrictive oxygenation, did not show evidence of an effect on plasma NfL levels in the respective group comparisons for each of the interventions at 24–48 h, and nor did a longer or shorter duration of device-based fever prevention of 36 compared to 72 h show evidence of an effect on NfL levels at 48 h; see Fig. 1, and Supplemental Table 2. The intervention groups were thus hereafter pooled for further analyses.

Fig. 1.

Neurofilament light chain levels according to randomization in the BOX trial. Shown individually for the three parallel randomizations of a high or low MAP target, a liberal or restrictive oxygenation target, and a targeted duration of device-based fever prevention for 36 or 72 h

NfL levels according to neurological outcome and one-year mortality

Median plasma NfL levels were lower in patients with a good neurological outcome compared those with CPC 3–5 to at both 24 and 48 h; 87.2 (IQR: 55.6–169) pg/mL compared to 3616 (992–8679) pg/mL, p < 0.001, at 24 h, and 101 (57.5–199) pg/mL compared to 6185 (1885–14353) pg/mL, p < 0.001, at 48 h (Fig. 2A). Similarly, median plasma NfL levels were lower in one-year survivors compared to non-survivors at both 24 and 48 h; 88.0 (IQR: 55.9–169) pg/mL compared to 3947 (1450–8929) pg/mL, p < 0.001, at 24 h, and 104 (58.0–208) pg/mL compared to 6711 (2838–15544) pg/mL, p < 0.001, at 48 h (Fig. 2B).

Fig. 2.

Diagnostic pression of neurofilament light Chain. (A) Neurofilament light chain (NfL) for prediction of a poor neurologic outcome at one year (B) Neurofilament light chain (NfL) for prediction mortality at one year

Prognostic performance and validation of previously identified cutoff levels for prediction of poor outcome

Plasma NfL performed excellent in neurological outcome prediction at both 24 h and 48 h, with an area under the receiver operator curve (AUROC) of 0.95 (95% CI: 0.92–0.97) and 0.95 (0.92–0.97) respectively, see Fig. 2.

When restricting the analysis to patients that had not awakened by 48 h (n = 300) and 72 h (n = 209) after cardiac arrest, NfL measured at 48 h yielded AUROCs of 0.96 (0.94–0.98) and 0.96 (0.94–0.99), respectively, for a poor neurological outcome (CPC 3–5).

For examining the sensitivity and specificity of specific NfL cut-off levels, we applied the levels suggested for serum NfL by Moseby-Knappe et al. (2019) at 24 and 48 h, see Table 2. At 24 h, the suggested 100% specific cut-off at 1232 pg/ml yielded a specificity of 98%, a sensitivity of 73%, and false-positive results in seven patients (1.4%) for poor neurological outcome in the current cohort. At 48 h, the 100% specific cut-off of 1539 pg/ml provided a specificity of 98%, and a sensitivity of 77%, for poor neurologic outcome, with seven patients being false-positive (1.3%). In total 9 patients had a false-positive test at either 24 or 48 h for a poor neurologic outcome. The cut-offs had similar performance for prediction of mortality at 1 year, see Table 2, with a specificity of 98% at both 24 and 48 h, and a total of 9 patients with a false-positive test at either 24 or 48 h. For characteristics of patients with a false-positive test compared to those with a true-negative for the suggested 100% specific cut-off, see Supplemental Table 3.

Table 2.

Neurofilament light chain cut-off levels for prediction of poor neurologic outcome and mortality

Time, hours	Cut-off Level pg/mLa	Specificity in Moseby-Knappe et al.a	Specificity in current cohort (95% CI)	Sensitivity (95% CI)	True-Positive	False-Negative	False-Positive	True-Negative	Total
Poor neurologic outcome:
24	1232	1.00 (0.98–1.00)	0.98 (0.96–0.99)	0.73 (0.66–0.80)	125 (24.5%)	46 (9.0%)	7 (1.4%)	333 (65.2%)	511
	641	0.99 (0.96-1.00)	0.97 (0.94–0.98)	0.79 (0.72–0.85)	135 (26.4%)	36 (7.0%)	11 (2.2%)	329 (64.4%)	511
	478	0.98 (0.95-1.00)	0.94 (0.91–0.97)	0.84 (0.78–0.89)	144 (28.2%)	27 (5.3%)	19 (3.7%)	321 (62.8%)	511
48	1539	1.00 (0.98-1.00)	0.98 (0.96–0.99)	0.77 (0.70–0.82)	151 (27.5%)	46 (8.4%)	7 (1.3%)	345 (62.8%)	549
	1140	0.99 (0.95-1.00)	0.97 (0.94–0.98)	0.80 (0.74–0.86)	158 (28.8%)	39 (7.1%)	11 (2.0%)	341 (62.1%)	549
	943	0.98 (0.93-1.00)	0.97 (0.94–0.98)	0.81 (0.75–0.86)	160 (29.1%)	37 (6.7%)	12 (2.2%)	340 (61.9%)	549
Mortality:
24	1232	1.00 (0.98–1.00)	0.98 (0.96–0.99)	0.76 (0.68–0.82)	125 (23.9%)	40 (7.6%)	7 (1.3%)	351 (67.1%)	523
	641	0.99 (0.96-1.00)	0.96 (0.94–0.98)	0.81 (0.74–0.86)	133 (25.4%)	32 (6.1%)	13 (2.5%)	345 (66.0%)	523
	478	0.98 (0.95-1.00)	0.94 (0.91–0.96)	0.85 (0.78–0.90)	140 (26.8%)	25 (4.8%)	23 (4.4%)	335 (64.1%)	523
48	1539	1.00 (0.98-1.00)	0.98 (0.96–0.99)	0.81 (0.74–0.86)	151 (27.0%)	36 (6.4%)	7 (1.2%)	366 (65.4%)	560
	1140	0.99 (0.95-1.00)	0.97 (0.94–0.98)	0.83 (0.77–0.88)	156 (27.9%)	31 (5.5%)	13 (2.3%)	360 (64.3%)	560
	943	0.98 (0.93-1.00)	0.96 (0.94–0.98)	0.84 (0.78–0.89)	158 (28.2%)	29 (5.2%)	14 (2.5%)	359 (64.1%)	560

Italics highlights that these specificities refer to the original investigation by Moseby-Knappe et al. and not the present investigation

^aCut-off levels and specificity from Moseby-Knappe et al. shown for NfL levels corresponding to 98%, 99% and 100% specificity for prediction of poor neurologic outcome in the original investigation [6]. A poor neurologic outcome was defined as CPC 1–2 at 1 year after cardiac arrest for this analysis, and mortality was similarly determined at 1 year

Identification of cut-off levels for poor and good outcomes in the present cohort

In the present cohort the cut-off levels that identified a poor neurologic outcome as well as mortality with a specificity of 100% were 5811 pg/mL at 24 h and 5582 at 48 h (Supplementary Table 4). The corresponding sensitivities for poor neurologic outcome and mortality at 1 year were 41% and 42% at 24 h, and 52% and 55% at 48 h, respectively. Conversely, the cut-off levels with 100% specificity for patients with a favourable neurologic outcome as well as survival were 32.3 pg/mL at 24 h and 51.6 pg/mL at 48 h. The corresponding sensitivities for a favourable neurologic outcome and survival at 1 year were 7% (5–39) and 7% (5–46) at 24 h, and 20% (16–36) and 20% (16–35) at 48 h, respectively.

Comparison of plasma and serum NfL measurements

A subset of patients had additional serum samples (77 patients) available at 24 and 48 h. NfL measurements in plasma and serum showed excellent correlation (Pearson’s correlation of log-transformed NfL values: 0.98, p < 0.001), and very high AUROC at both 24 and 48 h in the subset of 77 patients, Supplemental Figs. 1 and 2. For a Bland-Altmann plot of EDTA plasma versus serum samples see supplemental Fig. 3. Also, see supplemental Fig. 4 for a plot omitting samples with values at the lower level of detection.

Discussion

In the present investigation, NfL measured at both 24 and 48 h after cardiac arrest had excellent precision in identifying patients at risk of a poor neurologic outcome as well as death. When applying the cut-offs previously identified in the TTM trial by Moseby-Knappe et al. for a specificity of 100%, we found a numerically higher sensitivity for identifying patients at risk of a poor neurologic outcome compared to the original study. However, in our cohort, these cut-offs yielded a false positive rate of 1.3% and 1.2% at 24 and 48 h respectively, with similar results observed for prediction of mortality [6]. These findings are confirmatory for the applicability of NfL and previously identified cut-offs when applied in another large OHCA cohort, despite differences in the assay used and the choice of sample type, which in our study was plasma compared to serum in the original study. Our findings of a false-positive rate of 1% with the previous identified cut-offs for a specificity of 100% for a poor neurologic outcome or mortality, is supportive for the application of this diagnostic test, but also underscores the need for standardization of the test, as well as highlight the importance of a multi-modal approach to neuroprognostication which have been found to improve specificity [3, 8, 22]. Of the 9 patients with a false positive test at either 24 or 48 h, i.e. patients surviving till 1 year despite an NfL level above the cut-off, 5 were awake at 48 h, and 6 at 72 h. The follow-up period for one year was chosen to allow for possible improvement of functional and cerebral status, and for poor outcomes to manifest, and apart from a slightly higher creatinine in survivors with a false-positive test compared to those with a true-positive, there were no apparent differences in patient characteristics. Kidney function has not previously been described as a possible source of variation for NfL after cardiac arrest. However, in other settings, impaired kidney function and elevated creatinine has been associated with increased NfL concentrations [23, 24]. Still, the significance of this is unknown in relation to neuroprognostication. Further, there was no difference in utilization of renal replacement therapy between false-positives and true negatives.

This investigation also allowed for comparison of the two different matrixes of EDTA plasma and serum for a subset of patients. NfL levels in plasma and serum both had very strong correlations at 24 and 48 h. These findings align with previous comparisons of plasma and serum when analysed in non-cardiac arrest patients, where NfL levels are generally much lower than in patients with severe hypoxic-ischemic injury, and our findings importantly confirm that the same correlation exists at higher levels [25, 26]. Importantly, the diagnostic performance was excellent for both with nearly identical AUROC, and there were good levels of agreement between the tests in the range of the examined cutoff level for poor outcomes. Differences in not only sample matrix, but also assay used for the present investigation compared to the study by Moseby-Knappe, could potentially have influenced the differences in identified cut-offs [6]. In the present investigation we used a MSD ELISA assay due to experience with the assay and platform, as opposed to a single molecule array (SiMoA) assay from Quanterix as used in the original investigation [6]. A direct comparison of these two immunoassays in patients with multiple sclerosis, where NfL concentrations are generally much lower than in patients with hypoxic-ischemic brain injury, found a very strong correlation for serum NfL, r = 0.95, with the MSD assay generally measuring higher concentrations than the Quanterix assay [27]. This finding may at least partially explain for our finding of higher NfL concentrations in the present cohort for cut-offs with similar specificities as the study by Moseby-Knappe et al. As has also generally been established, this highlights the need for intra-assay calibrations, and also suggests that cut-offs for identification of severe brain injury after cardiac arrest should be validated for each assay, and possibly also for matrices, individually prior to clinical application [8]. Further, the present study selectively included patients with OHCA of a presumed cardiac cause, and accordingly we cannot rule out that the diagnostic precision or cut-off levels would be different for patients with other causes or in-hospital location of cardiac arrest.

Multiple studies have been published on the performance of NfL in predicting poor outcomes after cardiac arrest, however for these investigations the source of plasma or serum was biobank samples that were analysed post-hoc in large batches, similar to the present investigation [5–7, 12, 14]. Accordingly, there have been no published studies in cardiac arrest patients with NfL being measured under circumstances similar to a clinical application of the test. There are several analytical strengths to measuring the samples in large batches post-hoc, including lesser variation due limited influence of assay- and equipment variability. However, the measuring techniques utilized in the present study with a handheld ELISA method for quantification could potential have resulted in a relatively larger variability compared to a fully automated measuring technique as would be employed for routine biochemistry in clinical use. Further, in a clinical setting, samples would likely be analysed without a freeze-thaw cycle, potentially influencing the test results [25].

The finding of excellent performance of NfL already after 24 h from cardiac arrest, as opposed to NSE that only gains sufficient diagnostic precision at 48 h, is in line with previous investigations, and indicates that application of NfL in a clinical setting will allow for initiation of reliable multi-modal neuroprognostication at an earlier stage than current practice [3, 6, 28].

Limitations to the generalizability of our findings include that NfL was measured in patients selectively resuscitated from an out-of-hospital cardiac arrest with a presumed cardiac aetiology, and therefore might not be extendable to patients with other aetiologies, further as previously highlighted samples were analysed post-hoc in large batches from biobank samples that had undergone a single freeze-thaw cycle, and cut-offs might perform differently when applied in a clinical setting. Therefore, studies that incorporate measurement of NfL in a clinical setting after cardiac arrest, ideally with results blinded for treating clinicians to avoid self-fulfilling prophecies, would be highly valuable for validating cut-offs for identification of poor as well as good outcomes.

Conclusion

The results of this investigation of NfL for identification of risk of poor neurologic outcome and mortality in a large cohort of out-of-hospital cardiac arrest patients are confirmatory for the prognostic role of NfL. Previously identified cut-offs of 1232 pg/mL at 24 h, and 1539 pg/mL at 48 h, performed excellent with a very high specificity. This indicates that application of NfL in a clinical setting will allow for reliable neuroprognostication as early as 24 h after cardiac arrest.

Author contributions

MM, RB, CC, RFS, HS, JM, JK, participated in conceptualization and methodology. CC, RFS, JK, CH provided resources. All authors participated in the investigation. MM and RB performed statistical analysis, and RB contributed with visualization of results. CC, RFS, HS, JM, JK and CH participated in project administration, supervision and validation. MM, RB, and CH participated in writing of the original draft. All authors participated in writing, review and editing.

Funding

Open access funding provided by Copenhagen University. The BOX trial was funded by an unrestricted grant from The Novo Nordisk Foundation (NNF17OC0028706), Denmark, and measurement of NfL was sponsored by an unrestricted grant from The Heart Center Research Council, Rigshospitalet, Denmark.

Data availability

Data request will be subject to approval from the relevant authorities and the BOX trial steering committee.

Declarations

Ethics approval and consent to participate

The BOX trial was approved by the regional ethics committee of the Capital Region of Denmark prior to initiation (Approval no. H-16033436). Inclusion of incapacitated patients was performed after consent from a trial guardian in the acute setting in accordance with Danish legislation. Relatives were asked for consent as soon as possible, and patients who regained consciousness were also asked for consent.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.