By continuing to browse this site you agree to us using cookies as described in About Cookies
Notice: Wiley Online Library will be unavailable on Saturday 7th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 08.00 EDT / 13.00 BST / 17:30 IST / 20.00 SGT and Sunday 8th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 06.00 EDT / 11.00 BST / 15:30 IST / 18.00 SGT for essential maintenance. Apologies for the inconvenience.
Neurovascular Research Laboratory, Vall d'Hebron Institute of Research (VHIR), Universitat Autònoma de Barcelona, Barcelona, Spain
Neurovascular Unit, Department of Neurology, Vall d'Hebron University Hospital, Barcelona, Spain
Address correspondence and reprint requests to Dr Joan Montaner, Neurovascular Research Laboratory, Vall d'Hebron Institute of Research (VHIR), Vall d'Hebron University Hospital, Pg. Vall d'Hebron 119-129, 08035, Barcelona, Spain. E-mail: firstname.lastname@example.org
Acute ischemic stroke is a complex disease with huge interindividual evolution variability that makes challenging the prediction of an adverse outcome. Our aim was to study the association of bloodstream signatures to early neurological outcome after stroke, by combining a subpooling of samples strategy with protein array discovery approach. Plasma samples from 36 acute stroke patients (< 4.5 h from onset) were equally pooled within outcome groups: worsening, stability, and improvement (n = 3 pools of four patients each, for each outcome group). These nine pools were screened using a 177 antibodies library, and 35 proteins were found altered regarding outcome classification (p < 0.1). Processes of inflammation, immune response, coagulation, and apoptosis were regulated by these proteins. Ten representative candidates, mainly cytokines and chemokines, were assayed for replication in individual baseline plasma samples from 80 new stroke patients: β-defensin2, MIP-3b, plasminogen activator inhibitor 1 active, β-cell-attracting chemokine 1, Exodus-2, interleukin-4 receptor (IL-4R), IL-12p40, leukemia inhibitor factor, MIP-1b, and tumor necrosis factor-related weak inducer of apoptosis. Multivariate logistic regression analysis showed β-defensin 2 (ORadj 4.87 [1.13–20.91] p = 0.033) and IL-4R (ORadj 3.52 [1.03–12.08] p = 0.045) as independent predictors of worsening at 24 h after adjustment by clinical variables. Both biomarkers improve the prediction by 19% as compared to clinical information, suggesting a potential role for risk stratification in acute thrombolyzed stroke patients.
Early neurological deterioration after stroke is not easily predictable. The use of blood biomarkers might help in decision-making processes regarding this complication. By combining a sub-pooling of samples strategy with protein array discovery approach, we have found two new biomarkers: beta-defensin-2 and interleukin-4 receptor. Both biomarkers improve the prediction of poor-outcome over clinical variables in the acute phase of stroke.
tumor necrosis factor-related weak inducer of apoptosis
Stroke is the second cause of death and one of the main causes of disability worldwide (Roger et al. 2012). Early neurological deterioration (END) is an important concern during acute stroke management and in some series it occurs up to 40% (Arenillas et al. 2002; Lin et al. 2012). Although there is not an international consensus about END, one consistently used definition is the increase of 4 or more points in the National Institutes of Health Stroke Scale (NIHSS) within the first 48–72 h (Alawneh et al. 2009). Several causes of complication such as hemorrhagic transformation, arterial reocclusions, or malignant edema may appear in the first hours after stroke contributing to END. Although nowadays there does not exist a specific therapeutic treatment to solve neurological deterioration, the admission of stroke patients into specialized stroke units have demonstrated to prevent END and thus to reduce the rate of poor outcome after ischemic stroke (Roquer et al. 2008). Therefore, the prediction of END is one of the challenges in stroke, as an accurate identification of patients more prone to worsen might help to optimize the admission to the scarce stroke units.
Some acute neuroimaging factors have been associated with END, such as hypodensity or hyperdense middle cerebral artery sign on computerized tomography or large diffusion weighted image in magnetic resonance (Alawneh et al. 2009). As neuroimaging techniques are not broadly accessible, the use of blood biomarkers could be a more feasible option. For that proposal, the exploration of molecules which could anticipate the development of END is becoming increasingly popular. Some candidates have been explored in that context, such as interleukin-6 (IL-6) (Vila et al. 2000) or b-type natriuretic peptide (Montaner et al. 2012), although their added value to clinical prognostic models is still unclear (Whiteley et al. 2009; Montaner et al. 2012).
The complexity of blood and the interindividual variability make it difficult to validate differential protein expression to distinguish among disease stages. To obtain a common signature of protein changes which are associated with a specific stage, pooled-blood samples are being employed in other diseases (Ernoult et al. 2010; Fragnoud et al. 2012) and are recommended for high-throughput proteomics (Barker et al. 2006). Moreover, multiple subpools, which are generated by random distribution of individual samples, can be performed to estimate variation within population (Karp and Lilley 2009).
To go in depth in the physiopathology of ischemic stroke we performed the first exploratory study of the plasma proteome by screening an antibodies library with a subpooled samples approach. Nowadays there is an assortment of different antibodies libraries in multiplexed arrays available in the market that allow the study of hundreds of proteins involved in different pathways while using few amount of patients' samples. We have used the SearchLight® library, which included 177 antibodies covering the exploration of several cellular processes in a multiplex ELISA-based manner.
We aimed to discover a common signature of protein expression changes for those patients who have an early poor outcome after stroke. Furthermore, after replication of our results in individual stroke samples, we assessed the added value of our biomarker candidates to clinical predictive models by means of comparative statistical metrics such as Integrated Discrimination Improvement (IDI) and Net Reclassification Improvement (NRI) Indexes.
Materials and methods
Patients and protocol
We recruited patients who were admitted to the emergency department of the Vall d'Hebron University Hospital (Barcelona, Spain) from 2004 to 2010 with an acute ischemic stroke within the first 4.5 h after symptoms onset. This large cohort of 395 patients was used to select candidates for the different phases of derivation and replication of this study. Stroke diagnosis was performed based on a standardized protocol of clinical and neuroradiological assessments as previously described by our group (Mendioroz et al. 2011). All patients received intravenous recombinant tissue-plasminogen activator (rt-PA) in a standard 0.9 mg/Kg dose (10% bolus, 90% continuous infusion during 1 h).
Stroke severity was assessed by using NIHSS (Brott and Bogousslavsky 2000). We defined neurological improvement as a decrease in NIHSS score by 4 or more points, neurological stability as changes in NIHSS score of 3 or less points, and neurological deterioration as death or an increase in NIHSS score by 4 or more points at 24 h or 48 h (Brott et al. 1992). Clinical data were blinded to biomarker measurement.
Before administration of any treatment, peripheral blood samples were drawn from each patient in EDTA collection tubes. Plasma was immediately separated by centrifugation at 1500 g for 15 min at 4°C and was stored at −80°C until use.
The local ethical committee approved the study and written consent was obtained from all patients or relatives in accordance with the Helsinki declaration.
A scheme of our technical approach is shown in Fig. 1. From our cohort, we randomly selected 36 ischemic stroke patients to perform pooled plasma samples of each outcome group, which were balanced according to age, gender, NIHSS score at admission, and etiology subtype. Individual plasma samples were ice thawed and equal volumes (0.5 mL) of four different samples from patients with similar clinical characteristics were mixed by agitation during 2 h at 4°C to obtain a pool. In total nine pools were prepared for analysis, including three pools for each outcome group:
Worsening: patients with neurological deterioration during in-hospital stay.
Stability: patients without changes in neurological state during in-hospital stay.
Improvement: patients with neurological improvement during in-hospital stay.
Screening by multiplexed SearchLight® antibodies array library
A library of 177 anti-human antibodies was screened with multiplexed sandwich ELISAs from SearchLight® platform (Aushon BioSystems, Billerica, MA, USA). We screened the complete library, with all the antibodies that the company offers in a multiplex system, which is based on chemiluminescent detection of molecules whose respective capture antibodies were combined in 96-well plates. The molecules which are included in the library represent different biological processes from gene ontology pathways (Table S1).
Representative biomarkers were selected by their illustration of the discovery phase findings regarding biological processes, the possibility of being combined in multiplexed ELISA, and/or their statistical significance. Ten candidates were combined in SearchLight® custom arrays, including β-defensin 2 (BD-2), plasminogen activator inhibitor 1 (PAI-1) active form, macrophage inflammatory protein (MIP) 3b, MIP-1b, β-cell-attracting chemokine 1 (BCA-1), exodus-2, interleukin (IL) 4 receptor (IL-4R) and IL-12p40, and leukemia inhibitor factor (LIF). Tumor necrosis factor-related weak inducer of apoptosis (TWEAK) was analyzed by a single ELISA commercial kit (eBioscience, San Diego, CA, USA).
These candidates were tested in individual plasma samples from 80 new stroke patients fulfilling similar inclusion criteria than the derivation cohort. To achieve enough statistical power in the comparison among outcome groups, patients who worsened were first randomly selected from our cohort and afterward groups of patients who remained stable or improved were balanced in accordance to clinical variables.
In both phases (discovery and replication), biomarker results were blinded to clinical data. Each sample was assayed twice and the mean value was used, removing those results with either the intraassay coefficient of variation (CV) or interassay CV higher than 30%.
SPSS statistical package 15.0 was used, unless contrary is stated. Normality for continuous biomarker levels or clinical variables was assessed by Shapiro–Wilk test for the discovery phase and Kolmogorov–Smirnov test for the replication study. Those normally distributed variables (p > 0.05) were analyzed by Student's t-test or anova and mean and SD values are given, whereas for variables with non-normal distribution Mann–Whitney U or Kruskal–Wallis test were used and median and interquartile range are reported.
In the univariate analysis, intergroup differences were assessed by Pearson chi-squared test for categorical variables. Cut-off points with the optimal accuracy (both sensitivity and specificity) to predict outcome were obtained from receiver operator characteristic (ROC) curves for each individual biomarker. To build predictive models, all clinical variables which were associated with outcome at p < 0.1 in the univariate analysis were included in a forward stepwise multivariate logistic regression analysis. For those independent variables, odds ratio (ORadj) and 95% confidence interval (CI) were adjusted by NIHSS at admission, age, and sex. Afterward biomarkers alone or in combination were added by Enter method to clinical models.
The areas under the ROC curve (AUC) from models that include biomarkers were compared with AUC from only clinical model by DeLong's method (DeLong et al. 1988) with MedCalc software (version 12.4; MedCalc Software, Ostend, Belgium). Using R software (Hmisc and PredictABEL packages), NRI and IDI Indexes were calculated to assess the added value of the biomarkers to the clinical predictive model (Pencina et al. 2008; Pickering and Endre 2012). In the case of NRI test, pre-specified clinically relevant thresholds of predicted risk (≤ 10% and > 90%) were used to calculate reclassification of patients into risk outcome groups (Whiteley et al. 2012).
In all cases, a p ≤ 0.05 was considered significant at a 95% confidence level.
Discovery phase: proteome screening
Nine pooled plasma samples from stroke patients who worsened, remained stable, or improved during in-hospital stay were screened in a 177 antibodies library (Table S1). A schematic view of the design of the study and clinical data which corresponds to the whole derivation cohort (n = 36) is shown (Fig. 1). No difference was found for age or for NIHSS at admission regarding outcome.
From 177 analyzed proteins, 12 proteins were non-detectable in our pooled plasma samples. In total, 35 proteins were found to be altered (p < 0.1) regarding outcome classification. Inflammation and immune response were the main modified pathways, with changes in the expression of chemokines, cytokines, and their receptors, as well as other cellular and systemic processes (Fig. 2). From these 35 proteins, 29 were associated with worsening (such as metalloproteinases, several components of the MIP family, etc.), being either elevated in those patients who worsened or decreased in those patients who improved, and only six proteins were associated with improvement [Apo A-1, TRAIL, C-peptide, IL-8, PAI-1 active form, and osteopontin (OPN)], conversely (Table 1).
Table 1. Biomarker levels in pooled plasma samples regarding in-hospital outcome classification
List of 35 proteins that were found associated (p < 0.1) with outcome in at least one analysis (worsening vs. stability/improvement or improvement vs. stability/worsening). Those normally distributed proteins were expressed as mean ± SD and those non-normally distributed proteins were described as median (IQR). Statistically significant differences are expressed as bold p-value; n.a.: non-associated (p > 0.1).
Apo A-1 (μg/mL)
201.7 ± 56.4
303.2 ± 37.2
299.6 ± 33.3
254.2 ± 74.2
44.8 ± 22.1
34.7 ± 15.0
23.3 ± 7.4
45.4 ± 15.6
1.9 ± 0.6
1.0 ± 0.4
1.2 ± 0.5
1.3 ± 0.8
2.0 ± 0.4
1.8 ± 0.6
2.3 ± 0.4
1.7 ± 0.5
188.6 ± 68.6
125.3 ± 34.9
116.1 ± 45.1
161.6 ± 55.3
74.1 ± 12.1
51.9 ± 8.8
43.7 ± 8.4
64.0 ± 11.2
18.6 ± 5.2
7.5 ± 2.1
6.6 ± 0.9
13.5 ± 6.7
2.3 ± 0.5
2.2 ± 0.3
2.0 ± 0.0
2.3 ± 0.4
677.9 ± 104.8
563.0 ± 180.2
442.6 ± 115.4
680.7 ± 118.9
11.6 ± 1.9
19.0 ± 5.2
19.7 ± 4.9
15.0 ± 5.7
6.4 ± 4.2
5.8 ± 3.9
2.8 ± 2.4
7.6 ± 3.3
8.6 ± 3.1
9.5 ± 5.2
5.6 ± 2.8
10.9 ± 4.1
205.2 ± 23.3
166.6 ± 28.1
165.5 ± 34.1
186.5 ± 31.1
35.7 ± 6.8
45.2 ± 10.1
52.6 ± 7.6
37.0 ± 5.4
PAI-1 active (ng/mL)
1.7 ± 1.7
5.0 ± 2.1
5.4 ± 2.3
3.2 ± 2.4
PAI-1 total (ng/mL)
Protein C (μg/mL)
2.8 ± 0.6
2.2 ± 0.2
2.4 ± 0.1
2.4 ± 0.6
588.9 ± 189.0
313.2 ± 76.6
313.8 ± 62.7
450.7 ± 203.7
25.9 ± 5.8
27.1 ± 12.4
17.6 ± 3.5
31.3 ± 9.5
1068.7 ± 106.4
881.8 ± 82.0
830.2 ± 49.9
1001.1 ± 112.1
Replication phase in 10 candidates
There were no differences in demographic characteristics between the derivation (n = 36) and the replication (n = 80) cohorts (Table S2).
Our 10 selected proteins were mainly cytokines and chemokines, which were associated with worsening: BD-2, MIP-3b, BCA-1, exodus-2, IL-4R, IL-12p40, LIF, MIP-1b, and TWEAK. Nevertheless, PAI-1 active form was associated with improvement in the discovery phase. Results from MIP-3b and PAI-1 active form were excluded from further analysis because of high CV interassay.
We have explored the influence of several clinical factors on the level of our candidate biomarkers, without substantial findings (Table S3). Regarding outcome, BD-2 and IL-4R levels were elevated in those patients who worsened within 24 h (p = 0.046 and p = 0.062, respectively) and 48 h after stroke symptoms onset (p = 0.041 and p = 0.031, respectively) (Fig. 3a–d). Interestingly, this association with worsening occurs only in those patients who received rt-PA earlier (within the first 3 h from onset) and had a less established infarct (p = 0.058 for BD-2 and p = 0.024 for IL-4R) than those patients who received rt-PA beyond 3 h (p = 0.399 for BD-2 and p = 0.909 for IL-4R) (Fig. 4a and b).
New biomarkers for acute stroke prognosis
In our replication cohort, 21.5% of patients worsened within 24 h and 31.8% of patients worsened at 48 h, with 40% of worsening as a result of hemorrhagic transformation (that was symptomatic in 75% of the cases). Patients who worsened were more likely to have diabetes mellitus, higher NIHSS at admission, and to have suffered a previous stroke (Table 2). A cut-off of 1.15 ng/mL for BD-2 (82% sensitivity, 48% specificity) and of 503.40 pg/mL for IL-4R (53% sensitivity, 72% specificity) discriminated between patients who worsened or not at 24 h. At 48 h identical cut-off points remained discriminative with 76% sensitivity, 49% specificity for BD-2, and 52% sensitivity, 73% specificity for IL-4R. Any association was found between BD-2 or IL-4R and hemorrhagic transformation as a specific cause of worsening (data not shown).
Table 2. Univariate analyses for worsening at 24 h and 48 h in the replication cohort
Worsening 24 h
Worsening 48 h
Yes (n = 17)
No (n = 62)
Yes (n = 21)
No (n = 45)
Statistical trend (p < 0.1).
mRS, modified Rankin Scale; TOAST, etiology stroke subtype classification; NIHSS, National Institutes of Health Stroke Scale; IQR, interquartile range. Statistically significant differences between groups are expressed as bold p-value.
After including all associated variables into the multivariate logistic regression analysis, only NIHSS at admission (ORadj 1.10 [95% CI 1.00–1.22], p = 0.050) and diabetes mellitus (ORadj 5.17 [1.47–18.13], p = 0.010) were clinical independent predictors of worsening within the first 24 h after symptoms onset. When plasma levels above the cut-off for both BD-2 (ORadj 4.87 [1.13–20.91], p = 0.033) or IL-4R (ORadj 3.52 [1.03–12.08], p = 0.045) were added separately or in combination (ORadj 3.81 [1.43–10.14], p = 0.008) with the clinical predictive model, the discriminative power of the predictive model increased from an AUC of 0.719 (95% CI 0.606–0.815) to an AUC of 0.829 (0.727–0.905) (p = 0.056) (Table 3). Moreover, further statistical analyses showed how the combination of both biomarkers increased significantly the discrimination between patients who worsened or who did not (IDI Index 0.095, p = 0.033). Regarding patient reclassification into higher risk categories, BD-2 alone reclassified better both the events and non-events (NRI Index 28.2%, p = 0.009) than the combination of both biomarkers (NRI Index 19%, p = 0.020) (Table 3). Similar results were obtained when worsening was assessed at 48 h from stroke symptoms onset, and when the combination of both BD-2 and IL-4R allows the reclassification of 27.5% of patients (p = 0.020) (Table S4).
Table 3. Comparison between predictive models with only clinical variables and models including biomarkers for worsening at 24 h
Model – Worsening 24 h
Clinical + BD-2
Clinical + IL-4R
Clinical + Combination BD-2 & IL-4R
All logistic regression models were adjusted by NIHSS at admission, age, and gender; ORadj (95% CI) and p-value were given. Biomarkers were added to clinical logistic regression model using cut-off point: BD-2 > 1.15 ng/mL and IL-4R > 503.40 pg/mL. NRI: Net Reclassification Improvement Index (risk categories used: ≤10%, 10–90%, and >90%); percentage of reclassification given for both events (i.e., patients who worsened at 24 h) and non-events and for the sum of both (with 95% CI). IDI, Integrated Discrimination Improvement Index; index given for both events and non-events and for the sum of both (with 95% CI). AUC, Area under the ROC curve; area with 95% CI given for each model. Clinical model always used as reference model to compare. Statistically significant results expressed as bold p-values.
In this study, we first described the plasma protein profile which is associated with early neurological outcome after ischemic stroke. Furthermore, a methodological improvement was attempted both technically (by introducing the pooling strategy) and statistically (by using comparative metrics). As a result two new outcome biomarkers (BD-2 and IL-4R) were discovered.
We consider the use of pooling strategies highly suitable for the discovery of new candidates to become biomarkers. Pooling reduces the biological variability, as it is assumed that the expression in the pooled sample averages the expression of the individual samples which were contained in the pool (Kendziorski et al. 2005). Another good point of pooling is the reduced costs, both in number of assays and number of biological samples. The use of a subpooling strategy to gain accuracy by including some variability within each group (Zhang et al. 2007) and the replication of the discovery findings in individual samples (Sham et al. 2002; Walker et al. 2010) contribute to a more desirable approach.
Following this design, we found 35 altered proteins which are involved in biological and cellular processes with a known role in ischemic stroke, such as inflammation, apoptosis, or the coagulation cascade (Mehta et al. 2007; Jickling and Sharp 2011). Some of these proteins have been previously associated with stroke outcome, confirming the validity of our strategy: fibrinogen (del Zoppo et al. 2009), D-dimer (Welsh et al. 2009), protein C (Mendioroz et al. 2009), resistin (Efstathiou et al. 2007), metalloproteinase 2 (MMP-2) (Montaner et al. 2001), MCP-1/CCL2 (Worthmann et al. 2010), and MIP-1a/CCL3 (Zaremba et al. 2006).
We have only detected one protein, OPN, which discloses an inversely associated relation in our study and in previous literature; while we found higher levels in those patients who improved, in a previous study from our group OPN was oppositely associated with long-term poor prognosis (Mendioroz et al. 2011). Moreover, some molecules that have been associated with short-term prognosis in our study, such as C-peptide (O'Neill et al. 1991), P-selectin (Bath et al. 1998), and IL-8 (Zeng et al. 2013), have been studied in other cohorts without any association regarding stroke prognosis. Although following different approaches, the possibility of false-positive results could not be overlooked. On the other hand, those proteins typically associated with poor prognosis, such IL-6 (Smith et al. 2004) and C-reactive protein (CRP) (Montaner et al. 2006), were not associated in our pooled cohort. This could be related to the dilution effect which was commented above, a plausible explanation as neither IL-6 nor CRP have shown a great association with outcome in individual studies (Whiteley et al. 2009).
Nonetheless, our discovery experiment provides a list of interesting candidates which have not been previously explored in the context of stroke prognosis. From them, we chose 10 candidates (BCA-1/CXCL13, BD-2, Exodus-2/CCL21, IL-12p40, IL-4R, LIF, MIP-1b, MIP-3b, PAI-1 active form, and TWEAK) to be tested in our replication cohort. BD-2 and IL-4R were found as independent predictors of neurological worsening in the acute phase of ischemic stroke, within 24 and 48 h after symptoms onset, mainly when the infarct has not been fully established.
Human beta-defensins play a role in immune-inflammatory responses, mainly acting as antimicrobial peptides and also as chemoattractants. BD-2 is mainly expressed in the respiratory tract epithelia, but it can also be expressed by monocytes and macrophages and, at brain level, by capillary endothelial cells and astrocytes. BD-2 expression is inducible by cytokines, such as tumor necrosis factor alpha (TNF-a) or IL-1b, and bacteria (Schröder and Harder 1999). Moreover, in vitro and in vivo models have shown an increase in expression and the release of BD-2 after hypoxic/ischemic stimuli (Liu et al. 2009; Nickel et al. 2012). In stroke patients, only the copy number variant of BD-2 gene (DEFB4) has been studied and reflects a higher protein plasma concentration in those patients with more gene copies (Tiszlavicz et al. 2012). Thus, the higher levels of BD-2 which were associated with worsening might be reflecting the inflammatory state that is produced after stroke.
IL-4 is a multifunctional cytokine which is required for the development of Th2 cells and thus with anti-inflammatory properties. To exert this effect, IL-4 binds to membrane-bound IL-4R; however, there exists a soluble form of IL-4R (mainly produced by MMPs-mediated proteolysis (Jung et al. 1999a)) which can block or prolong IL-4 effects depending on its concentration (Jung et al. 1999b). Together with the higher expression of MMPs detected in patients who worsened, the association of soluble IL-4R with poor outcome is also in accordance with the pro-inflammatory state that is suggested by BD-2.
We also wanted to prove that, aside from being independent predictors, both biomarkers add value to clinical information. Clinical variables typically associated with neurological deterioration are stroke severity at admission and age, as non-modifiable factors, together with risk factors such as diabetes mellitus or arterial hypertension. The clinical variables that were independently associated with early worsening in our cohort comprise NIHSS at admission and diabetes mellitus, after being adjusted by age and gender. The deleterious effect of diabetes mellitus on stroke outcome has been previously observed by several groups, as recently reviewed (Desilles et al. 2013). When this clinical model was considered, comparison of AUCs showed how BD-2 and IL-4R, alone or in combination, improved the measure of discrimination of clinical variables, which changes from acceptable to excellent discrimination between patients who worsened and those who did not. However, some authors suggested that the comparison of AUCs is not the best way to know the additional value of biomarkers (Pepe et al. 2004). Therefore, we have employed statistical tools which are being applied to know the capacities of biomarkers regarding discrimination (IDI) and patients' reclassification into risk categories (NRI). These two tests are based on the risk prediction models (i.e. logistical regression models) and the probability of an event for each patient (Pickering and Endre 2012). Both IDI and NRI can be calculated separately for events (i.e., worsening) and non-events (i.e., non-worsening) to better interpret how the biomarker is adding value, if it is by better recognizing the events or if it is by reducing the rate of false positives for non-events. Regarding our results for IDI test, we found an improvement in discrimination by the identification of real events for BD-2 and particularly for the combination of both BD-2 and IL-4R. This may be explained because BD-2 cut-off point had great sensitivity and IL-4R increased specificity, thus, the combination of both biomarkers gained statistical power. Regarding reclassification of the patients into extreme risk categories, BD-2 alone showed the best performance at 24 h, reclassifying correctly a 28.2% of patients, mainly by reducing false positives. At 48 h, the combination of both BD-2 and IL-4R achieved this non-negligible reclassification rate. If confirmed in prospective studies as prognostic biomarkers, the measurement of plasma levels of BD-2 and IL-4R in the acute phase of stroke might help in decision-making processes, such as in giving information to patients and relatives, optimization of inclusion in specialized stroke units, evaluation of treatment benefits, or inclusion into clinical trials.
Our study stands with several limitations. The results of our discovery phase have not been corrected in accordance to multiple testing; we consider this phase merely exploratory and the limited sample size does not have enough statistical power to sustain this kind of correction. Moreover, sample size of our replication cohort is relatively small and limits the assessment of biomarkers association with specific worsening causes that could influence outcome, as has been suggested in larger cohorts, although non-modifiable factors are the main force (Grube et al. 2013); thus, an independently larger study might conduct subanalysis by specific causes and to assess the value of these biomarkers in all types of stroke patients as the results of this study can only be generalized to stroke patients who underwent thrombolysis. In the future, BD-2 and IL-4R might be further explored in a prospective cohort that is being recruited in our hospital and that will collect a more complete information regarding outcome. Furthermore, different detection methods (simple ELISA, Luminex arrays, or others) should be explored to confirm our results. Finally, molecules which have been found altered in our discovery phase, but have not been included in the replication because of the impossibility of performing multiplex ELISA as well as molecules that are not included in the discovery array might be of interest, such as the recently described stroke outcome biomarker copeptin (De Marchis et al. 2013) or others.
Acknowledgments and conflict of interest disclosure
T.G-B is supported by a pre-doctoral fellowship (FI09/00017) from the Instituto de Salud Carlos III. V.L is supported by a pre-doctoral fellowship from Vall d'Hebron Institute of Research. Neurovascular Research Laboratory takes part in the Spanish stroke research network INVICTUS (RD12/0014/0005) and is supported on stroke biomarkers research by FIS 11/0176. The authors have no conflicts of interest to declare.
We would like to thank all study collaborators, residents, neurologists, and nurses of the Stroke Unit and Neurology Ward from the Vall d'Hebron Hospital and especially all the patients who participated in this study.
The authors have no conflict of interest to declare.