Background
Accurate neuroprognostication is essential in patients who remain comatose after cardiac arrest to identify patients who may recover and avoid futile treatment in those who will never awake. Current guidelines recommend multimodal prognostication no earlier than 72 h post-arrest to allow time for recovery and clearance of sedative drugs [
1]. Further investigation into markers with early prognostic value in comatose survivors after cardiac arrest may enable earlier multimodal neuroprognostication.
Neurofilament light chain protein (NfL) is a novel biomarker of neuroaxonal injury. Elevated circulating NfL at 24–72 h is highly predictive of poor neurological outcome after out-of-hospital cardiac arrest (OHCA) [
2‐
4]. Additionally, low levels of NfL at 24–72 h have been reported to accurately identify patients with good outcome [
3,
5]. The prognostic value is low early after ICU admission and the earliest time point when NfL can provide reliable prognostic information is unknown [
3,
6]. Previous studies reporting the highest prognostic performance of NfL included only selected cohorts of OHCA [
2,
3]. Further studies are needed to validate the predictive value of NfL in heterogeneous cohorts of post-cardiac arrest patients [
7].
The current study aimed to investigate the prognostic performance of NfL to predict good and poor outcomes in the first two days after cardiac arrest in an unselected cohort of OHCA and in-hospital cardiac arrest (IHCA) patients. We hypothesised that the predictive performance would be acceptable at 12 h and improve at 48 h due to accumulating brain injury and biomarker kinetics. We also hypothesised that NfL would be a more accurate predictor of outcome after OHCA than IHCA due to differences in patient and clinical characteristics [
8,
9].
Methods
Study design and setting
This was a retrospective, multicentre observational study of patients admitted after cardiac arrest to three Intensive Care Units (ICU) in southern Sweden from 2014 to 2018, with no pre-defined study size. The patients were consecutively included in the SWECRIT biobank, aiming to study biomarkers in critically ill patients (ClinicalTrials.gov no. NCT04974775, retrospectively registered July 2021). The study protocol was approved by the Regional Ethical Review Board in Lund, Sweden (registration no. 2014–47 and 2022-02681-01). Written informed consent was obtained from patients who regained mental capacity. The Standards for Reporting Diagnostic accuracy studies (STARD) guidelines were followed [
10].
Study population
All patients 18 years or older admitted to intensive care after cardiac arrest were eligible for inclusion. International guidelines on post-resuscitation care were followed, including multimodal neuroprognostication no earlier than 72 h after cardiac arrest for patients who remained unconscious [
11]. Blood samples were collected on ICU admission, 12 and 48 h after cardiac arrest. Samples drawn within six hours of the specified time points were included for statistical analysis. Patients with admission samples only (0–6 h after cardiac arrest) were excluded from this study.
Biochemical analyses
All samples were centrifuged, aliquoted, and frozen to −80 °C before storage in the biobank at Region Skane, Sweden (BD-47, SC-1922). The measurements of NfL were performed at the Clinical Neurochemistry Laboratory at the University of Gothenburg in September 2021 by staff blinded to clinical outcomes. Plasma NfL levels were measured using a single-molecule array (Simoa) NfL immunoassay on an HD-X analyser according to instructions from the kit manufacturer (Quanterix, Billerica, MA). Serum neuron-specific enolase (NSE) was analysed by the local laboratory in Region Skane as part of management in clinical practice [
12], using a Cobas instrument with electrochemiluminescent immunoassays (Roche Diagnostics, Rotkreuz, Switzerland).
Data sources and outcome
Patient data were collected from the International Cardiac Arrest Registry (INTCAR), the patient administrative system for intensive care units (PASIVA), the Swedish population register, and medical records. Routine electroencephalogram (EEG) was performed when indicated at the treating physician’s discretion. Neurological outcome according to the Cerebral Performance Category (CPC) scale [
13] was assessed two to six months post-arrest by personnel (physician or nurse) blinded to NfL levels. We considered CPC 1–2 as good outcome and CPC 3–5 as poor outcome.
Statistical methods
Clinical characteristics are presented as medians and interquartile ranges (IQR) or as counts and percentages. Groups were compared using Mann–Whitney U test, Pearson’s Chi-squared test, or logistic regression. Diagnostic performance was assessed by the area under the receiver operating characteristic curve (AUROC), defined from logistic regression models with and without addition of alternative methods for prediction of outcome. A model with clinical data included age, time to ROSC, witnessed cardiac arrest, bystander CPR (OHCA), shockable rhythm, and administration of adrenaline. The DeLong method was used to compare AUROCs for paired data.
To account for missing data, a multiple imputations approach [
14] using the package mice for R was used with the CART method (imputation by classification and regression trees) over
N = 20 multiple imputations. Pooled AUROCs were calculated over the imputed datasets using Rubin’s Rules, as implemented in the psfmi package for R.
We assessed partial AUROC (pAUROC) to predict poor outcome at 95–100% specificity. We determined pAUROC to predict a good outcome at 80–95% sensitivity, aiming to be highly predictive and to identify most patients with a good outcome. At the set specificities (95 and 100%) and sensitivities (80 and 95%), we determined cut-offs, positive predictive values (PPV, prediction of poor outcome), and negative predictive values (NPV, prediction of good outcome). We used a bootstrap method implemented in the R package pROC to calculate 95% confidence intervals (CI) and compare paired pAUROCs.
We used logistic regression models separately for OHCA and IHCA to adjust for covariates identified with stepwise backward regression. The backward elimination was continued for as long as the Akaike information criterion (AIC) decreased (R package MASS). We also evaluated differences in the performance of NfL to predict outcome after OHCA and IHCA with an interaction model, testing the interaction between NfL and cardiac arrest setting (OHCA vs IHCA). We tested if the covariates identified in the previous step eliminated the interaction, first using all covariates and finally only the covariates with a significant effect in the interaction model.
To reduce the skewness of the NfL measurements, we used log10-transformed data. Significance was set at
p < 0.05. All analyses were performed using R, version 4.1.2 [
15].
Discussion
This multicentre cohort study investigated the ability of plasma NfL to predict long-term neurological outcome in a heterogeneous population admitted to intensive care after cardiac arrest. NfL at 12 and 48 h after OHCA reliably predicted good and poor neurological outcomes. The predictive ability after IHCA compared to OHCA was lower at 12 and 48 h.
The prognostic performance of NfL at 12 h after OHCA in this study was similar to the accuracy of NfL at 24 and 48 h in two previous studies [
2,
4]. One study included an unselected OHCA cohort [
4], and the other investigated a selected group of OHCA patients of a presumed cardiac cause [
2,
16]. The performance was slightly higher in the COMACARE trial [
3,
17], which included a more selected patient group with initial shockable rhythm. The accuracy of any test will depend on the population studied, and the predictive ability will likely improve in highly selected populations of OHCA. Despite the heterogeneous OHCA cohort in this study, the ability of NfL to predict the outcome at 12 and 48 h was excellent, and superior to NSE. This novel finding supports the generalisability of NfL as a useful prognostic tool across the entire spectrum of OHCA patients in the ICU.
In the present study, the prediction of poor outcome at high specificity was better at 48 h compared to 12 h after OHCA. The improved prediction with time likely reflects evolving post-anoxic brain injury but may also be explained by patients not included at 48 h due to prior haemodynamic collapse and circulatory deaths. Although lower than at 48 h, the predictive ability of NfL at 12 h shown here may still be of clinical value.
Low levels of NfL have been shown to predict a good neurological outcome at 24 h after OHCA [
3,
5]. In this study, NfL, as early as 12 h after OHCA, had a high ability to predict good outcome. Early prediction of neurological recovery may aid in decision-making, e.g. regarding invasive procedures, and could help avoid inappropriate limitations in care and provide relevant information for relatives. Another potential clinical use of a low 12 h NfL value would be to tailor neuroprotective strategies, e.g. by attempting an early wake up in patients with low NfL. Our results suggest that the prognostic value of EEG is improved by NfL already at 12 h, supporting the prognostic performance of NfL when measured at least 12 h after cardiac arrest. Future studies could explore whether early prediction of outcome may be further strengthened by combining NfL levels at 12 or 24 h with an early (< 24 h) return of a continuous background pattern on EEG [
18,
19].
In this study, the predictive performance of NfL on admission to intensive care was low, which is in line with previous studies [
3,
6,
20]. In patients with poor outcome, NfL levels increase substantially as post-hypoxic brain injury develops and may reach a steady state between 48 and 72 h [
2,
3,
21]. NfL is also known to increase with age, even under normal circumstances [
22], and elevated NfL levels can be caused by neurological comorbidity and traumatic brain injury [
23‐
27]. NfL increase caused by other conditions is typically low compared to NfL levels released due to hypoxic brain injury [
28‐
30], but in very early prognostication and when low NfL levels are used to predict a good outcome, other causes of NfL elevation may be important to consider.
To our knowledge, NfL has previously not been analysed in an IHCA cohort. The prognostic accuracy of NfL was lower in IHCA compared to OHCA at both 12 and 48 h. However, a significant interaction between the location of arrest and outcome prediction by NfL was only found at 12 h. These results suggest an overall trend for lower predictive ability after IHCA, particularly early after cardiac arrest. IHCA and OHCA patients differ in key parameters that are relevant for prognostication. For example, the time to ROSC was shorter and WLST due to poor prognosis less common after IHCA than OHCA, suggesting a less severe brain injury in IHCA. This is also reflected by the higher level of consciousness on ICU admission after IHCA. When focusing on patients with good outcome, we found that NfL was generally higher in IHCA than OHCA, which may be explained by their higher age and more comorbidities in this cohort [
8,
9,
31]. NfL, as a marker of neuroaxonal injury, is not likely to identify patients with poor outcomes due to haemodynamic collapse or other causes of non-neurological death, which may contribute to the lower predictive ability of NfL in IHCA patients. When adjusting our analyses for age, time to ROSC, cardiac cause, shockable rhythm and administration of adrenaline, the difference between IHCA and OHCA in predictive ability of NfL was no longer significant, suggesting that patient and cardiac arrest characteristics may explain the lower predictive value of NfL after IHCA.
Strengths and limitations
Strengths of the present study are the heterogeneous cohort, allowing for more generalisable results, the multicentre design, the large sample size of OHCA patients, and the batch analysis of NfL samples. A limitation of this and other NfL studies is that there is no certified standard for analysis, and both analytical and preanalytical factors may vary, complicating comparisons of NfL levels reported by different studies. Before NfL can be implemented in clinical practice, future studies must establish a standard for analysis and normal values for NfL. This study is also limited by its retrospective design. The data quality depended on the documentation in the patient's medical notes, data on the presumed cause of death was unavailable, and the time to follow-up varied. The sample size of IHCA was small, particularly at 48 h.
Another limitation is that a lack of withdrawal of consent after regaining consciousness in some patients will lead to an underestimation of the proportion of survivors. Differences between included and excluded patients suggest that missing patients were due to early deaths and transfers from the ICU. Prognostication is not performed in clinical practice in patients who die or wake up early, but we cannot exclude that missed inclusions have affected the generalisability of the study.
Open AccessThis article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit
http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (
http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.