Predictive value of circulating interleukin-6 and heart-type fatty acid binding protein for three months clinical outcome in acute cerebral infarction: multiple blood markers profiling study

The early prediction of long-term clinical outcome following ischemic stroke is important because of the high morbidity and mortality associated with stroke. Intravenous tissue plasminogen activator (tPA) is the only effective treatment for ischemic stroke, but less than 15% of all patients are eligible for thrombolytic therapy due to its narrow therapeutic window [1]. In patients who were not indicated for thrombolytic therapy, early initiation of treatment to reduce secondary neuronal damage is required for a favorable outcome. Age and initial clinical severity, represented by the National Institutes of Health Stroke Scale (NIHSS) score, are established clinical predictors for long-term outcome [2]. Brain imaging is a powerful diagnostic tool to determine the extent of neuronal injury which is associated with long-term stroke outcome. However, brain imaging has several limitations with regard to feasibility and cost-effectiveness.

Blood biomarkers can be a complementary tool for diagnosis [3], predicting prognosis [4, 5] and therapeutic monitoring of novel treatments in ischemic stroke [6]. Blood markers of neuronal injury, astroglial injury, inflammatory mediators, and/or thrombotic/haemostatic proteins are candidate blood markers for early stroke recurrence or long-term clinical outcome in previous case-control studies [7‐11]. For the clinical application of biomarkers for stroke prognosis, candidate blood markers must satisfy the proposed criteria for an ideal blood marker [4, 12]; 1) the marker has a statistically independent association with outcome after adjusting for clinical covariates; 2) the marker improves the discriminating ability of the established clinical model; and 3) the marker can reclassify patients at low or high risk for poor outcome across clinically relevant thresholds of predicted probabilities of stroke outcome.

Although many blood markers have been investigated to date, no single blood marker has been proven to be clinically useful to predict stroke outcome [12]. The reason is that most of the previous studies measured a single blood marker in a case-control design with variable time points of sampling. Profiling studies of multiple markers, the mechanisms of which are differentially associated with stroke pathophysiology, have tried to overcome the limitation of a single biomarker for stroke diagnosis [13] or outcome [4, 5, 10]. Several panels of inflammatory and haemostatic markers, such as IL-6, N-terminal pro-brain natriuretic peptide, C-reactive protein (CRP), fibrinogen, and/or D-dimer, have been suggested to be associated with poor clinical outcome and death [4, 5, 10]. However, whether the addition of blood biomarkers improves the predictability of the clinical predictors for stroke outcome remains controversial [4, 5, 10]. Therefore, clinically applicable panels of blood markers should be evaluated in subsequent studies.

The aim of the present study is to measure the blood concentration of multiple blood markers in stroke patients and investigate the clinically available blood markers for prediction of the long-term clinical outcome.

The demographic and laboratory characteristics between the favorable and poor outcome groups are described in Table 1. The poor outcome group was older, had a higher WBC count and had a higher prevalence of Afib than the favorable outcome group. The poor outcome group had a higher median initial NIHSS score and larger median DWI volume than the favorable outcome group. The univariate logistic regression model indicates that age (OR: 1.58 (per 10 years), 95% CI: 1.17 to 2.12, P = 0.002), initial NIHSS score (OR: 4.20 (per 3 points), 95% CI: 2.70 to 6.52, P < 0.001), Afib (OR: 4.29, 95% CI: 2.17 to 8.48, P < 0.001) and WBC count (adjusted OR: 1.11 (per 10³/μl), 95% CI: 1.03 to 1.19, P = 0.005) were associated with poor prognosis. TOAST classification, especially small vessel disease subtype and small infarct volume were associated with favorable outcome (P < 0.01). In the multivariate analysis of clinical variables, only age and initial NIHSS showed a strong association with poor stroke outcome (all P < 0.01), with the disappearance of significance for Afib and WBC count (P > 0.05). Among nine deceased patients (3m-mRS = 6), six died from the stroke itself (three large hemispheric stroke, two basilar stroke, and one cerebral hemorrhage), and three likely died from non-stroke causes (two cardiopulmonary distress, one sepsis).

The median concentrations of individual blood markers are shown in Table 2. The median sampling time after stroke onset was 10 hours (IQR: 7 to 16). The poor outcome group had higher median blood concentrations of hFABP, S100B, IL-6, MMP-9, CRP, and D-dimer than the favorable outcome group. The blood concentrations of NSE, TNF-α and PAI-1 were not different between the two groups. As described above, most of the stroke patients displayed negative GFAP and VSNL-1 in the peripheral blood during this time period, indicating that these proteins are unlikely to be released from the brain to the bloodstream during this time window of stroke.

The logistic regression analysis of individual blood markers after adjusting for age and initial NIHSS score revealed that plasma _logIL-6 (adjusted OR: 1.75, 95% CI: 1.25 to 2.25, P = 0.001) and _loghFABP (adjusted OR: 3.23, 95% CI: 1.44 to 7.27, P = 0.0045) reached a statistical significance (Table 3). The statistical significance of _logIL-6 and _loghFABP remained after further adjustment for 72-hour infarct volume (natural log-transformed) and Afib (_logIL-6, OR: 1.74, 95% CI: 1.24 to 2.44, P = 0.003; _loghFABP, OR: 3.23, 95% CI: 1.41 to 7.40, P = 0.001), time of stroke onset (_logIL-6, OR: 1.74, 95% CI: 1.24 to 2.44, P = 0.001; _loghFABP, OR: 3.23, 95% CI: 1.41 to 7.40, P = 0.005) and previous statin use (_logIL-6, OR: 1.81, 95% CI: 1.27 to 2.57, P = 0.001; _loghFABP, OR: 3.36, 95% CI: 1.46 to 7.47, P = 0.004). Adding the _logIL-6 and _loghFABP to the baseline clinical model improved the -2 log likelihood ratio (102.57 versus 123.01, P = 0.01) and goodness of fit (Hosmer-Lemeshow's test, P = 0.506). No multi-collinearity was found among the tested variables. In the Spearmann's correlation analysis, infarct volume was significantly correlated with IL-6 (r = 0.38, P < 0.001) but not hFABP (r = 0.126, P = 0.10).

In ROC curve analysis, the baseline clinical model showed good discriminating ability (AUC: 0.910) for stroke outcome. The addition of any single marker did not significantly improve the AUC value of the baseline clinical model. Adding a combination of _logIL-6 and _loghFABP to the baseline clinical model (AUC: 0.939, P < 0.03) significantly improved the discriminating ability of the baseline clinical model (Table 4 and Figure 1). The addition of other individual markers to the combined model of _logIL-6, _loghFABP and clinical model did not result in further improvement of AUC in ROC curve analysis (P ≥ 0.05 by the DeLong method). We further calculated the in-sample NRI index and IDI of _logIL-6 and _loghFABP in each model (Table 4). Adding _logIL-6 to the baseline clinical model significantly improved the NRI index (0.09, P = 0.04) but adding _loghFABP to the baseline clinical model did not improve the NRI index. The addition of both _logIL-6 and _loghFABP further enhanced the NRI index of _logIL-6 (0.18, P = 0.005). The in-sample reclassification table in the model with _logIL-6 and _loghFABP is presented in Additional file 2. In 64 patients with a poor outcome, seven were reclassified into higher risk categories, and only two were reclassified into lower risk categories. In 111 patients with a favorable outcome, only six were reclassified into higher risk categories, and 17 were reclassified into lower risk categories.

In the present study, we measured multiple blood markers that are plausibly related to the pathophysiology of ischemic stroke to predict functional outcome. The strengths of the present study include the prospective collection of stroke cases, measurement of multiple blood markers in a single study, exact diagnosis of stroke using DWI images in all cases and statistical analysis after adjusting for multiple confounding variables. We found that circulating IL-6 and hFABP levels showed an independent association with stroke prognosis after adjusting for clinical covariates. Moreover, the combination of these two markers showed a significant improvement of discrimination and reclassification for high and low risk of clinical outcome in stroke patients. Other blood markers failed to show a significant clinical value for the prediction of stroke outcome.

Previous studies demonstrated that IL-6 is one of the predictors of poor outcome in stroke [16‐20]. Our finding of the strong association between IL-6 and poor stroke outcome is consistent with those of several studies using multiple blood markers in prospective stroke cohorts [4, 5]. A systemic inflammatory response develops after brain damage during the acute phase of ischemic stroke. IL-6 is one of the most important mediators of the acute phase response to inflammation. The precise mechanism of the association between elevated IL-6 levels and poor clinical outcome is unclear. IL-6 up-regulates the expression of adhesion molecules, such as intercellular adhesion molecule-1 (ICAM-1), selectins and integrins on endothelial cells, leukocytes and platelets, during cerebral ischemia, which leads to secondary neuronal damage [21]. IL-6 also increases the levels of fibrinogen and von Willebrand factor, resulting in hypercoagulant status [22]. IL-6 expression is increased in the brain following ischemia, and damaged neurons may contribute to increased IL-6 levels [23]. Although the source of circulating IL-6 during the acute phase of stroke is still unknown, the early increase in IL-6 and its correlation with poor outcome support the role of acute inflammation in the pathophysiology of stroke.

Using a blood IL-6, we did not find any improvement in the prognostic performance of the clinical model for stroke outcome despite the close pathophysiological relationship between IL-6 and cerebral ischemia. Our results are consistent with previous studies which showed that IL-6 lacks clinical usefulness in the discrimination and reclassification between favorable and poor outcome groups of stroke patients [4, 5]. In our data, the combined use of IL-6 and hFABP significantly improved the discriminating ability of the clinical model poor stroke outcome. The addition of IL-6 and hFABP significantly reclassified patients at low- and high-risk for poor outcome across the clinically relevant threshold. hFABP is a small cytoplasmic protein (15 kDa) that is involved in active fatty acid metabolism in the myocardium and neuronal cell bodies in the central nervous system [24]. Plasma hFABP level is known to be a sensitive biomarker of acute coronary ischemia [25, 26]. In ischemic stroke, plasma hFABP levels were found to be elevated in several case-control studies [27‐29]. Plasma hFABP levels six hours after stroke were associated with early neurological severity (NIHSS score at 10 days) and long-term outcome (mRS at three months) [28, 29]. The precise mechanism of the association between blood hFABP and stroke outcome is unclear. One plausible explanation is that hFABP is released from the brain following cerebral ischemia. Proteomic analysis showed that hFABP levels in cerebrospinal fluid were elevated in deceased stroke patients [27]. However, we did not find a significant correlation between infarct volume and hFABP, indicating the extracerebral origin of hFABP release following stroke. A second explanation is that the elevation of blood hFABP reflects acute cardiovascular dysfunction associated with neuronal injury [30], which is associated with poor outcome. A third explanation is that hFABP is elevated in other conditions, such as heart failure [31], sepsis [32], pulmonary embolism [33], and metabolic syndrome [34], which may be an underlying substrate for poor stroke outcome. Further studies are required to clarify the origin of hFABP release during the acute phase of stroke and its predictive value for stroke outcome.

Neuro-astroglial markers are candidate stroke biomarkers because they are hypothetically more specific to brain injury than inflammatory and haemostatic markers. Several case-control studies showed that blood levels of S100B, NSE and GFAP are elevated in acute stroke, and are associated with poor stroke outcome [7‐9]. We evaluated the blood profiling of several neuro-astroglial markers, including candidate neuronal markers (VSNL-1 and Ngb) that have not been tested in human stroke patients [35, 36]. None of the tested markers, with the exception of hFABP, reached statistical significance. The lack of an association between blood levels and stroke outcome in the present study is attributable to their slow release kinetics from damaged tissue to peripheral blood (that is, their peak plasma levels were reached at 48 to 72 hours after stroke onset) [8, 28].

The limitations of the present study should be addressed. First, we did not perform serial measurements of the blood markers. The inflammatory process and release of neuro-astroglial proteins are prominent for several days after a stroke. Therefore, the measurement of blood markers within 24 hours after stroke onset weakens the potential application. The aim of our study is to search for panel markers within 24 hours of stroke onset for prediction of stroke outcome, as prediction of stroke outcome is required as soon as possible considering the high mortality and morbidity of stroke. We also excluded patients with their onset within six hours because most cases during this time period need thrombolytic therapy, which may confound the clinical outcome of cerebral infarction, and because our experience has shown that most of the tested biomarkers (especially neuro- and astroglial proteins) are not elevated within six hours after symptom onset. Second, it is unknown whether the elevation of blood markers represents a consequence or a pre-morbid status of stroke, as inflammatory and haemostatic markers are elevated in non-cerebral conditions, such as systemic infection, cancer or metabolic diseases. Further studies that involve serial measurement of blood marker levels are required to resolve this issue. Third, the sample size of our study was small. Our results should be interpreted with caution, and external validation is required to be generalized in clinical practice. Fourth, we could not examine the candidate blood markers for prediction of stroke outcome in the previous studies, including brain natriuretic peptide (BNP or NT-proBNP) [4], midregional pro-atrial natriuretic peptide (MR-proANP) [37], lipoprotein-associated phospholipase A2 (LpPlA2) [38], or copeptin [39]. Further studies that expand the panel of blood markers screened are required.

We found that blood IL-6 and hFABP levels are associated with poor clinical outcome in patients with ischemic stroke. Single markers did not enhance the discriminating ability of the clinical predictors for stroke outcome. The addition of the IL-6 and hFABP levels improved the prognostic performance of the clinical predictors for stroke outcome. Further studies in a large-scale prospective cohort are required to determine whether these blood proteins are clinically useful in routine clinical practice for prediction of those at high-risk of a poor clinical outcome, who need more intensive therapeutic strategies.