Blood neurofilament light chain as a predictive biomarker for functional outcome of acute ischemic stroke: a systematic review and meta-analysis

A systematic search was conducted on the Pubmed, Cochrane Library, and Embase databases from their inception to 21 October 2023, utilizing a combination of keywords, medical subject headings (MeSH) terms, and free words. The search strategy included variations of terms such as ‘ischemic stroke’, ‘ischemia stroke’, ‘cerebral infarction’, ‘brain infarction’, as well as ‘neurofilament light chain’ or ‘NfL’. Furthermore, reference lists of included studies were manually reviewed to identify any additional relevant studies. The search strategy is listed in Supplementary Table 1.

The study’s inclusion criteria consisted of the inclusion of patients diagnosed with AIS, including transient ischemic attack (TIA); the provision of risk ratios (RRs) or odds ratios (ORs) along with their corresponding 95% CIs to demonstrate the association between AIS and bNfL. Excluded from consideration were reviews, case reports, letters, conference abstracts, clinical short communications, brief reports and nonhuman animal model research.

Two independent reviewers (H-Y Xu and H-Y Lin) conducted a screening of titles and abstracts based on predetermined inclusion criteria. Studies that potentially met criteria underwent a full-text review for further evaluation. Any discrepancies in the inclusion or exclusion of studies during screening were resolved through discussion until a consensus was reached, with final discrepancies being resolved by Li Chen.

Two independent reviewers (H-Y Xu and H-Y Lin) extracted the following information from each study: first author, publication year, region, study design, sample size, median age, follow-up duration, sampling time, method used to detect baseline neurofilament light (bNfL), type of blood specimen, NfL levels, stroke-ascertainment criteria, etiological subtypes, recurrent ischemic events, mortality, definition and timing of outcome (all studies assessed the Modified Rankin Scale [mRS] at 90 days or later after symptom onset), correlation coefficients, and adjusted covariates. This information was then compiled into a predesigned table.

The quality assessment of the included studies was conducted utilizing the Newcastle–Ottawa Scale (NOS) [17], which comprises three components (selection, comparability and outcome). The NOS scores range from 0 to 9, with the risk of bias categorized into three levels: high (0–3), moderate (4–6) and low (7–9). A higher score indicates high quality. The assessment process was independently carried out by two authors (R-X Qin and L-D Shao).

The meta-analysis was conducted using the Comprehensive Meta-Analysis software, Stata 12.0. Effect sizes were computed by importing the adjusted OR from multivariate logistic regression in the original study, which were then used to calculate a pooled adjusted OR representing the predictive effect of bNfL on functional outcome in AIS. Fixed-effects models were utilized when I² was less than 50.0%, while random-effects models were employed in cases of higher heterogeneity (I² > 50.0%). Heterogeneity was assessed using the Q test and the I² statistic, which ranges from 0 to 100% and indicates the degree of heterogeneity across studies. An I² value greater than 50% signifies substantial heterogeneity. Subgroup analysis was conducted to evaluate the consistency based on clinical characteristics, sample size and study design. Publication bias was evaluated through funnel plots in conjunction with Eggers’ test and Begg’s test. Two-tailed statistical tests were employed with a significance level set at p < 0.05. Sensitivity analysis was utilized to assess the reliability of the findings.

Our initial search of electronic and manual databases yielded 690 records, of which 114 were duplicates. Following a review of titles and abstracts, 542 records were excluded, leaving 34 studies for full-text evaluation. Of these, 25 studies were excluded for various reasons, including being conference abstracts, comments, reviews, clinical short communication, brief report, lacking reported effect sizes or having uninteresting outcomes. Ultimately, nine studies involving 2302 participants were deemed for inclusion in the quantitative synthesis (meta-analysis) [9,10,13–16,18–20]. The flow chart for this study screening and selection is shown in Figure 1.

Table 1 displays the primary characteristics of the 9 studies incorporated in this systematic review and meta-analysis, encompassing publication dates spanning from 2015 to 2023. With the exception of three studies, all were prospective cohort studies. The sample sizes of the included studies varied from 41 to 615 participants. Of the nine studies, six originated from Europe and America, while three were from Asia. Furthermore, three studies reported recurrent ischemic events, four reported mortality rates, and three described infarct volumes. All included studies examined the association between bNfL levels and unfavorable functional outcomes in stroke patients. The NOS for all studies included in the analysis varied between 7 and 9. Upon thorough examination of the included studies, we identified some qualitative heterogeneity.

The studies by Traenka et al. [18] and De Marchis et al. [19] included TIA patients in addition to AIS patients, with the analysis of stroke functional outcomes also encompassing TIA patients. The average age of the participants in these studies exceeded 60 years old, a contrast to the studies by Traenka et al. [18] and Pedersen et al. [20]. Six studies utilized the single-molecule array method for detecting bNfL levels, a novel approach for measuring NfL levels [21]. However, the other two studies utilized the electrochemiluminescence immunoassay method for detecting the NfL levels [18,19]. Gendron et al. utilized the NF-Light digital immunoassay for this purpose [13]. Furthermore, the duration of blood sampling varied across the studies included in the analysis, with Traenka et al. collecting samples over a period of 6 (3–9.5) days, Tiedt et al. at 7 days, and Pedersen et al. at 4(3–6) days [9,18,20]. The study conducted by Gendron et al. [13] collected blood samples within two-time frames: 0–8 days and 9–20 days. Among the five studies included, blood samples were obtained within a maximum of 1 day, with De Marchis et al. [19] collecting blood at 160 (91–275) min, Uphaus et al. [10] within 24 h after admission, and Zhou et al. and Li et al. at admission [15,16]. Chen et al. [14] collected blood samples at 5.13 ± 4.47 h, 5.54 ± 4.66 h, and 27.21 ± 6.23 h. Additionally, it was noted that two studies used plasma specimens while the others used serum. All studies utilized CT or MRI to confirm stroke diagnosis, except for one study that did not provide details on stroke ascertainment [13]. Regarding the etiological subtypes of stroke, the majority of studies did not classify subtypes except for two [15,18]. Cardioembolism was found to be the most prevalent etiological subtype in three studies [13,14,19], with patients typically being younger compared with other subtypes. In the studies conducted by Tiedt et al. [9] and Uphaus et al. [10], undetermined etiology was the most common etiological subtype. However, Li et al. [16] showed that small vessel occlusion accounted for the highest proportion of the etiological subtypes.

The mRS was the predominant measure utilized to assess functional outcomes in the studies analyzed, with all NfL levels from blood samples included in the stroke outcome analysis. Furthermore, Tiedt et al. [9] defined recurrent ischemic lesions (RIL) at 6 months as an additional prognostic indicator. Li et al. [16] incorporated early neurological deterioration (END) as a functional outcome indicator in their study. Due to variations in statistical methodologies and prognostic indicators, the mRS was designated as the functional prognostic index for the meta-analysis, which aimed to quantitatively assess the predictive value of bNfL in determining functional outcomes in AIS. A mRS score of 0–2 indicates functional independence (good outcome). mRS 3–6 indicates at least moderate disability or death (poor functional outcome). A total of nine studies were included in the analysis, with some studies also examining infarct volume, recurrent ischemic events, and mortality as outcome measures. Among the included studies, three specifically analyzed infarct volume as a potential indicator of functional outcome. De Marchis et al. [19] utilized a semi-quantitative method to estimate infarct size in their study. Their findings indicated that there was no significant association between infarct size on admission MRI-DWI and sNfL levels, both in univariate analysis (p = 0.15) and after adjusting for age and NIHSS on admission (p = 0.56). In contrast, Tiedt et al. [9] employed software tools to measure infarct size and observed a positive correlation between infarct volumes and NfL levels on D7 (partial correlation coefficient r = 0.736, p = 1.5 × 10–15) and D3 (r = 0.370, p = 0.0003), although this relationship was not evident at earlier time points. Chen et al. [14] reported a final infarct volume of 13.9 (3.8–61.4) ml without detailing the method of measurement. Additionally, they observed no significant correlation between sNfL levels and infarct volume at different time points (T1, p = 0.92; T2, p = 0.99; T3, p = 0.1). Three studies included in the analysis investigated the predictive value of sNfL as a biomarker for recurrent ischemic events [9,10,18]. Traenka et al. [18] identified two stroke patients who experienced recurrent ischemic strokes at 5 and 45 days post-blood sampling, with corresponding sNfL levels of 124.0 and 427.7 pg/ml, respectively. Nineteen patients (20%) experienced RILs in the study conducted by Tiedt et al. [9]. The researchers determined that the presence of RILs served as an independent predictor of sNfL levels at 6 months post-stroke, even after adjusting for variables such as age, sex, hypertension, and infarct volume at baseline (p = 0.014). In a separate study by Uphaus et al. [10], 56 patients experienced recurrent stroke or death. Furthermore, the researchers found that sNfL levels were a valuable predictor of recurrent stroke or death (OR: 2.002; 95% CI: 1.213–3.302; p = 0.007), in addition to NIHSS, diabetes mellitus and age-related white matter change rating, as determined through multivariate regression analysis. Four of the studies included in the analysis reported mortality rates [13,14,16,20]. The detailed baseline characteristics of the studies included in the analysis are outlined in Table 1.

As all of the studies incorporated in the analysis were non-randomized controlled trials, the NOS scores were utilized for evaluation. In terms of selection criteria, five studies received four stars, while four studies received three stars. Regarding comparability, eight studies were awarded two stars, with one study receiving one star. For outcome assessment, five studies were assigned three stars, while four studies were allocated two stars. Overall, the included studies were deemed to be of high quality, as evidenced by the detailed assessment provided in Table 2.

Nine studies involving 2302 patients explored the relationship between bNfL levels and unfavorable functional outcomes in AIS. All ORs were calculated, revealing a high level of statistical heterogeneity (I² = 62.8%) among the studies. Consequently, a random effects model was chosen for the meta-analysis. The results, depicted in Figure 2, showed a pooled adjusted OR of 1.929 [95% CI: 1.459–2.550], indicating that patients with higher bNfL levels are at a greater risk of unfavorable functional outcome compared with those with lower levels. The assessment of publication bias in the funnel plot is presented in Figure 3, which does not indicate significant publication bias (p = 0.466, two-tailed). Similar results were obtained from Egger’s test got similar results (p = 0.229, two-tailed). Additionally, a sensitivity analysis was conducted, as shown in Figure 4, revealing that the findings are robust.

The presence of significant heterogeneity in the meta-analysis prompted an in-depth examination of its origins through an analysis of both clinical and methodological factors, such as sampling time, study region, participant age, blood specimen, sample testing method, inclusion of TIA and sample size, as detailed in Table 3. The findings revealed that nearly all subgroups exhibited potential sources of heterogeneity. The subgroup analysis revealed that studies with a sample size ≥200 (I² = 71.3%) and participants aged ≥60 years (I² = 71.3%) exhibited the highest heterogeneity, with I² values of 71.3% and 71.3%, respectively. Conversely, subgroups characterized by participants aged <60 years, plasma sample collection, and sampling time of ≥4 days showed no heterogeneity (I² = 0). Additionally, the association between bNfL levels and unfavorable outcomes remained consistent across all subgroups, except for sample testing methods and inclusion of TIA in the study design. The results of subgroup analyses are shown in Table 3 and the forest plot of subgroup analyses is depicted in Figure 5.

Our meta-analysis comprehensively and systematically reviewed the current available literature and found that the patients with higher bNfL levels is associated with higher risk of unfavorable functional outcome compared that with lower bNfL levels. bNfL carries value as a functional prognostic biomarker for AIS.

Two systematic reviews on the relationship between bNfL and stroke have been published [12,22]. While our meta-analysis aligns with the main findings of Liu et al., there are notable distinctions between the two studies. Specifically, Liu et al. [12] included data from five studies and 1346 patients, whereas our meta-analysis incorporated data from nine studies and 2302 patients. The increased statistical power resulting from the inclusion of nearly 1000 additional cases in our analysis positions it as the most up-to-date and comprehensive study on the subject, providing further support for previous findings in the literature. Furthermore, we conducted a comprehensive analysis of heterogeneity, examining both clinical and methodological factors through subgroup analysis, encompassing a total of seven aspects. In contrast, Liu et al. [12] only investigated heterogeneity based on blood sampling time. Additionally, a sensitivity analysis was performed to assess the stability of the results. Sanchez et al. [22] focused on temporal patterns of neurofilament light as a blood-based biomarker for stroke in their meta-analysis. The study encompassed a range of cerebrovascular conditions, such as AIS, and chronic cerebrovascular disease like cerebral small vessel disease and cerebral autosomal dominant arteriopathy with subcortical infarcts and leukoencephalopathy. AIS cases accounted for only half of the total sample. Additionally, this meta-analysis did not conducted the pooled OR for quantitatively assessing the predictive value of bNfL in AIS functional outcomes.

The NfL protein, an intermediate filament protein, is indicative of primary axonal injury and secondary neurodegeneration [23]. It is released into the cerebrospinal fluid and bloodstream following neuronal tissue damage [24], demonstrating high specificity to the nervous system. Previous research has suggested that abnormal concentration of NfL serve as a direct marker of the severity of neurological diseases [2]. Our meta-analysis, following a systematic review of the existing literature, presents the most up-to-date and comprehensive evidence supporting bNfL as a predictive biomarker of AIS outcome. Consistent with previous research, our study further confirmed the correlation between bNfL and outcomes of AIS. Ferrari et al. [8] identified NfL-D7 as a significant predictor of 3-month NIHSS and the most effective biomarker for AIS patients. Tiedt et al. [9] observed a continued elevation in sNfL levels for up to 6 months post-ischemic stroke. The increase in sNfL levels during the acute phase following ischemic injury can be attributed to the significant number of neuronal axonal injuries. The elevation of bNfL levels may be attributed to various potential mechanisms. Ischemic injury is considered a significant factor, while Waller’s degeneration is also suggested to play an important role. It is noted that large myelin axons, which are abundant in neurofilaments, are susceptible to Waller’s degeneration following a stroke [25]. Additionally, post-stroke immunity responses and inflammation are believed to be significant contributing factors. Shichita et al. [26] and Yan et al. [27] have demonstrated the involvement of immune and inflammatory processes in the pathophysiology of ischemic stroke injury in both animal models and human subjects. Furthermore, certain investigations have indicated that persistent blood-brain barrier breakdown probably prompt prolonged release of bNfL [28,29]. None of the included studies provided head-to-head comparisons of bNfL-augmented models versus standard clinical scores (e.g., NIHSS + age + infarct volume). Therefore, we cannot conclude that bNfL adds incremental predictive value beyond readily available covariates.

According to the subgroup analysis, the timing of blood sampling may be as a significant factor contributing to heterogeneity. Therefore, the selection of an appropriate blood sampling time is crucial for the accurate prediction of bNfL on stroke outcomes, as shown in [8,12,20,22,30,31]. However, our study indicates that measurement taken within a time frame of less than 4 days may not capture the relationship between bNfL levels and post-stroke functional outcomes. Further investigation into the temporal pattern of bNfL post-stroke is warranted to pinpoint the peak levels, which could have implications for utilizing NfL as a predictive biomarker for stroke outcomes. In the present subgroup analysis stratified by region, it was observed that patients with higher levels of bNfL exhibited a greater risk of poor functional outcome compared with those with lower levels of bNfL in studies conducted in Asia and Europe and America. These findings suggest that bNfL may serve as a predictive biomarker for AIS across various geographical regions. Furthermore, when stratified by study sample size, it was noted that studies with larger sample sizes tended to yield larger effect sizes than smaller studies. Additionally, stratification by age revealed that studies including individuals aged 60 years and older were associated with a higher risk of poor functional outcome. This observation appears to be associated with pronounced arteriosclerosis, heightened inflammatory reaction and severe ischemic conditions. Subgroup analysis focusing on blood specimen revealed significant heterogeneity in studies utilizing serum. Despite the widespread use of ultra-sensitive single-molecule array technology in most studies, potential variations in blood NfL levels between serum and plasma measurements should be carefully considered. What is noticeable is that no significant heterogeneity were found in the sample testing method and inclusion of TIA stratification. However, these results can not completely rule out the source of heterogeneity, after all, the sample size included in TIA is not large enough and studies using Electrochemiluminescence immunoassay are limited. It should be noted that these results should be interpreted with caution, because the results of the subgroup analysis were observational and larger studies are needed to evaluate the effect of bNfL as a predictor of stroke functional outcomes in the subgroups. The statistically significant differences may be due to smaller sample sizes and fewer studies included in these subgroups, rather than an absence of effect, and differences are unlikely to be clinically significant. At present, no interventional trial has used bNfL levels to triage patients to more intensive monitoring, prolonged ICU stay, or escalation of secondary prevention. Until such trials are completed, bNfL should be considered a prognostic enrichment biomarker for research cohorts rather than a trigger for specific therapeutic decisions.

To the best of our knowledge, this is the most recent and the comprehensive systematic review and meta-analysis on the correlation between bNfL and AIS outcomes to date. Several strengths of our study are evident. First, the results of our meta-analysis offer the most up-to-date and compelling evidence for informing updates to clinical practice. Second, we performed subgroup analyses to investigate potential sources of heterogeneity, considering both clinical and methodological factors across seven dimensions. The subgroup analysis should have improved the credibility of the study and provided a benchmark for future research. Thirdly, our meta-analysis comprised nine studies in which we sought to assess the predictive value of bNfL in determining functional outcomes following AIS. Moreover, our systematic review was not constrained by temporal or linguistic restrictions.

There are some limitations in our study that should be taken into account. First, there was notable heterogeneity among the included studies, likely stemming from variations in sampling time, study region, participant age, blood specimen and sample size. Additionally, there may be other unidentified sources of heterogeneity. Second, while some studies have reported the proportion of etiological subtypes, the precise relationship between bNfL and functional outcomes in stroke etiological subtypes remains unclear. It would be advantageous to conduct subgroup analysis according to stroke etiological subtypes. Third, it is important to note that while one study was observational, two were prospective observational studies, and the remaining studies were prospective cohort studies. Discrepancies in study designs may limit the ability to establish causality. Fourth, because all included studies dichotomised mRS at ≥3, we were unable to examine the full ordinal distribution. Future individual-patient-data meta-analyses should test whether bNfL predicts a shift across the entire mRS range. Finally, Follow-up interval differed across studies (7 days to 6 months). Because most cohorts reported only the 90-day mRS, we could not examine whether the prognostic value of bNfL attenuates or strengthens with longer follow-up. Individual-patient-data meta-analyses that incorporate repeated mRS measurements are warranted.