Vascular Endothelial Growth Factor as Predictive Biomarker for Stroke Severity and Outcome; An Evaluation of a New Clinical Module in Acute Ischemic Stroke
Correspondence Address: Source of Support: None, Conflict of Interest: None DOI: 10.4103/0028-3886.271241
Source of Support: None, Conflict of Interest: None
Keywords: Acute stroke, prognosis, vascular endothelial growth factorKey Message: This research explores the role of VEGF and stroke severity. The inference could let us know the biomarker's correlation and functional status.
The prevalence of stroke seems to have risen exponentially in the past few decades and is projected to reach epidemic proportions concurrent to the rise in macro-vascular diseases such as diabetes in developing countries like India. The low mean age of stroke suggests that the social consequences of stroke are devastating. The prediction of outcomes after ischemic stroke is important for researchers and clinicians. A number of clinical variables have been identified as potential predictors of clinical outcome such as age and severity of deficit, which are directly predictive of outcome. However, the multivariable relationship of these predictors is unclear, which has led to contradictory reporting in the literature., A combination of clinical and imaging variables has been shown as a predictor in a risk adjustment model in the evaluation of stroke outcomes in both prospective and retrospective studies. As a consequence of stroke, several neurotrophic growth factors have been shown to increase in the borders of the infarcted areas (penumbra) within the ischemic zones. Vascular endothelial growth factor (VEGF) is a dimeric glycoprotein mitogen on endothelial cells known to increase vascular permeability, as an inducer of monocytes in chemotaxis, an inhibitor of extracellular apoptosis. It is an angiogenic protein with neurotrophic effects. It has been found that the serum VEGF concentration reaches its peak after seven days of ischemia and has stayed until the 14th day. Some studies have reported upregulation of VEGF in acute phase, as well as a direct relationship between VEGF levels and functional recovery., The increase in serum VEGF in the acute stage is found to be proportional to an improved National Institute of Health Stroke Scale (NIHSS) score after 3 months. Due to these properties of this biomarker it can be used as prognostic tool to predict stroke outcome.
The best validated clinical prognostic models continue to be inaccurate enough to predict outcome in patients with stroke. The objective of this research was (a) to study whether VEGF is a predictive biomarker after acute ischemic stroke of severity and functionality (AIS); (b) to examine the multivariable relationship between clinical characteristics and outcomes at three months after ischemic stroke; (c) to demonstrate the utility of a newly developed model (clinico-biomarker module) which can be used to make clinically useful predictions and prognosticate functional recovery in AIS.
A randomized controlled study with a sample size of 250 patients suffering from AIS within two weeks of the index event was conducted in a tertiary care hospital. Patients who consented to the study were prospectively enrolled and followed up for a period of three months after the onset. On enrollment, the objectives, study design, risks, and benefits were explained in detail to each patient or surrogate family members and written informed consent was obtained from the participants. All procedures performed were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards. The study was performed with the approval of the local ethics committee. Patients with the following illness were excluded: concomitant major illness affecting longevity of life such as cancer, progressive neurological illnesses affecting mortality and morbidity such as dementia, or other neurodegenerative diseases. All recruited patients were randomized into two groups: a study group and a control group. The control group consisted of patients in whom only the clinical predictor variables which constitute the clinical module were assessed. As for the study group patients, serum growth factor VEGF assessment was performed and added as as another variable to clinical module assessment. Peripheral venous blood samples were collected from patients in the study group for serum VEGF estimation. All patients had the following assessments at baseline and at the end of a three month period. NIHSS, Modified Ranking Scale (mRS), and Barthel Index (BI). According to NIHSS, patients with NIHSS (0–6) were classified as mild, moderate (7–20), and severe (>20). Patients in both the study and control arms were further divided into three groups: mild (0–6), moderate (7–15), severe to moderately severe (16–20), and very severe (20–42) based on the stroke severity (NIHSS score). The case and control group's modules for prediction of stroke outcomes were compared. Thirty healthy controls were included to measure the normal serum VEGF.
A detailed literature review was performed to identify clinical predictors of stroke outcome, and independent variables for the model were identified to avoid increased risk of type I errors. Eleven candidate variables were included depending on the sample size of the study population and outcome measures. These variables were selected based on the strength of the relationship to stroke outcome in the published literature, availability in the RANTTAS data, and the clinician's expertise and judgment. Eleven clinical variables and one biomarker variable were chosen.
The variables are shown in Appendix 1. The NIHSS is a valid and reliable rank order scale that measures 11 categories of neurological deficit by neurological examination in stroke patients. The NIHSS scoring was done by a trained personnel and certified for its proper use. Age, gender, stroke subtype, history of previous stroke, dysarthria, stroke localization, wakeup strokes, and Glasgow Coma Scale (GCS) were dichotomized as present or absent by history. Time from onset to presentation and prestroke functional status had trichotomous scoring using NIHSS under four categories: (0–6 scored as 1, moderate 7–15) as 2, moderately severe (15–20 as 3, and severe >20) as 4. The additional serum VEGF was scored between 1 and 4 as: 0–200 = 1, 200–300 = 2, 300–400 = 3, and 400–500 = 4. In this module, a higher subscore reflected a more severe initial deficit, with absent marked as 1 and present as 2. Dichotomization provided outcomes that were easily understood by clinicians and patients. The cutoff value of 15 for the NIHSS in the poor outcome model was based on the clinical judgment of the authors as well as other analyses.
The BI is a well-validated scale that measures the ability to complete 10 basic activities of daily living,, depending on the level of functionality. We compared the performance of the combined model (clinical and biomarker variables) with the clinical model (clinical variables only). The weights of each of the predictive variables for each model were determined by multivariable logistic regression techniques. Model discrimination was assessed by areas under the receiver operator characteristic (ROC) curves, which were computed by the nonparametric method. An ROC curve area of 0.5 indicates no discrimination, and an ROC curve area of 1.0 indicates perfect discrimination. The area under the ROC curves for the nested models (combined versus clinical models) were compared with likelihood ratio statistics.
Basic characteristics and demographics
Totally, 274 patients were screened, out of which 250 patients were recruited. The mean age for all patients was 56.7 ± 6.7 years. For a baseline serum VEGF sample between the 0th and 7th day of admission, 125 patients were assessed. Hypertension was the most common risk factor in all patients followed by diabetes and hypercholesterolemia [Table 1]. Around one-fourth of the patients had a history of alcohol in study group (25%), whereas in control group 40% of patients were positive. The mean NIHSS of all the patients was 17.1 ± 3.4 years, mRS ~3.4 ± 0.9, and BI ~39.3 ± 9.1, indicating that the cohort represents severely affected patients who were more than 50% dependent on their care givers.
Serum vascular endothelial growth factor and neurological severity
The VEGF scores in all study group patients was 378.4 ± 98.9 pg/ml, indicating moderately high scores in all patients with a score of 3 on the CB module. Out of 250 patients, 35 patients were classified as mild, 66 patients were moderately affected, and 149 were severely disabled. The mean cores with box plots in all three groups are shown with very high serum scores in the severely affected group (median 401pg/ml, 3rd quartile ~422.5pg/ml), with moderately affected patients having a median of 302.5pg/ml (3rd quartile ~389 pg/ml) (P = 0.05, 95% confidence interval (CI) = 4.5–1.2) [Figure 1]. There was no mortality in either study arm at the end of the three month study period.
Out of 35 patients, 14 of the study group patients had a mean VEGF score of 306 ± 92.9 pg/ml. The mean clinical module score was 13.5/26 and mean CB score was 16.1/30 without any statistical significant difference (P = 0.78). The control group had a score of 12/26 without any statistical difference between the two groups (P = 0.53). Mean NIHSS, mean mRS, and BI showed no statistical difference at 3 months [Table 2], although all patients improved with a mean BI score of 97.3, indicating minimal dependence on care givers.
Out of 66 patients in the moderately affected group, the mean clinical module score was 17.2/26 with mean VEGF of ~ 302 pg/ml. The CB module score was 19.6/30 with a mean increase of 2.4 (P = 0.56). The reduction in NIHSS and mRS were statistically significant as compared to baseline (mean 6 points reduction in NIHSS; P = 0.03 and 1.2 points in mRS; P = 0.34). BI was statistically significant between baseline and 3 months in all patients (41.3 versus 70.3;P = 0.02). The study and control groups when compared at 3 months did not reveal statistical significant difference in NIHSS, mRS, and BI (P > 0.05) [Table 3].
Severely affected patients
The mean VEGF score was markedly increased in all severely affected patients with a mean of of 420 ± 79.6 pg/ml. The mean CB module score was 23.4/30, whereas mean module score was 19.6/26 (P = 0.02; 95% CI ~4.1–1.5). NIHSS, mRS, and BI showed statistically significant improvement at 3 months in all patients (P < 0.05). When the study group and control groups were analyzed at 3 months, it was observed that BI and mRS (P = 0.05, 95%CI: −1.4 to −0.21; P = 0.05, 95%CI: 7.5 to − 2.5) showed statistical significant improvement while NIHSS did not (P = 0.16, 95% CI: 1.4 to −0.21) [Table 4].
Multiple regression analysis
Multiple regression analysis for 125 study group patients revealed that the CB model was fit to predict prognosis and severity in AIS [R2 = 0.86, F (23.4,6); P = 0.001], with NIHSS subscore, prestroke status, and VEGF being very strong predictors (P = 0.001, P = 0.001, and P = 0.003, respectively) of all variables studied. When only the clinical module was tested on all 250 patients, it was found that NIHSS subscore, time to stroke onset and prestroke functional status were the most common predictors [R2 = 0.79; F (45,9);P = 0.005].
In a stepwise regression analysis to predict the severity of stroke, it was observed that NIHSS and GCS followed by VEGF subscores were the most commonly predicted (r2 = 0.98; >0.96; >0.92), whereas all other variables lay outside the model. The ROC curve was drawn between the CB module and the clinical module. It was observed that CB module had a sensitivity of 79% with 21% specificity (R = 0.78, P = 0.05) followed by the clinical module with sensitivity of 53% and specificity of 67% (P = 0.12) [Figure 2], indicating that the module may be a good predictor for severely affected patients. The combined model had an area under the ROC curve greater than either of the other model for 6 of the 11 outcomes.
Bootstrap validation of the combined model for NIHSS, VEGF score and prestroke functional status was performed, which assessed internal validation and how accurately the models could predict the outcome in a new similar sample of stroke patients. All predictors were significantly correlated with the CB module and the bootstrap validation revealed that the bias in all the vairables was minimal; VEGF score [β ≤0.001, P = 0.001, 95% CI ~ 0.18–0.008], NIHSS subscore [β = 3.10, P = 0.001, 95% CI ~ 10–7.3], prestroke functional status [β = Z0.63, P = 0.02, 95%CI ~ 1.2–0.3] [Table 5].
The findings of the present study confirm VEGF as a neurobiomarker upregulated after AIS within seven days of measurement. This proangiogenic biomarker was significantly raised in severely affected (420 ± 79.6 pg/ml) patients with baseline NIHSS (mean 22.4) indicating its role post-ischemia [Figure 2]. It has been widely reported that VEGF is abundantly produced from neurons and vascular cells, thereby playing an important role in mediating neuronal survival and proliferation. The expression of VEGF is transcriptionally upregulated by hypoxia-induced factors in response to hypoxia. VEGF and its receptor Flt-1 are upregulated in neurons and vascular cells in peri-infarct areas in cardioembolic stroke., The increase in serum VEGF levels in the severely affected group was proportional to an improved NIHSS score after 3 months, than as compared to mild and moderately affected patients. After adjustment for covariates, serum VEGF levels were still significantly correlated with the long-term prognosis of ischemic stroke in our study as evidenced by other reports also., Slevin et al. reported that the mean expression of VEGF peaked after 7 days and was maintained up to 14 days, with lowest in the serum of patients with small infarct, increasing in patients with medium infarct and being the greatest in patients with large infarct, which suggests that VEGF could be a marker indicating the size of the infarct. Our patients reported a high mean of 458.3 ± 102 pg/ml as compared to 234.3 ± 98.3 pg/ml in healthy controls, which was similar to the result found in another study (220 pg/ml).
It was noted that in the study group that severely affected patients whose VEGF was high had statistically significant improvement at three months (for BI: P = 0.05;95% CI; 7.5 to − 2.5) than controls, indicating that increased upregulation of VEGF leads to better functional improvement and performance. We also observed that 51/80 patients had a good outcome with NIHSS <15, and 29 patients had a poor outcome with NIHSS >15, which is in congruence with the other report in which 3/29 patients who expressed exceptionally high mean VEGF levels (>1000 pg/ml) presented with the highest improvements in SSS rating. The ROC curve analysis revealed that area under the CB module to predict good outcomes was greater: (A = 0.96; P = 0.001; 95% CI: 0.996–0.934) as compared to the clinical module only (A = 0.89; P = 0.01, 95% CI: 0.95–0.83)[Figure 2], suggesting that the CB module was statistically superior in performance to the model using clinical variables alone especially in severely affected patients. In the future, this combined model could be tested in a heterogeneous stroke population who participate in a randomized acute stroke trial. Patients with mild and moderately affected symptoms had a mean VEGF score of 304.5 pg/ml and all patients improved significantly at 3 months (study and control group). The CB module scores for these patients did not differ between study and control groups in mild and moderately affected and it was also observed that patients with good outcome did not show an increased VEGF, which could reach statistical significance correlation.
Finocchi et al. combined clinical and imaging variables and categorized infarct volume into three categories (no lesion, medium lesion, and large lesion) and inferred that the infarct volume does not improve the predictive ability of such a model as the final size of an infarct cannot be detected on CT for several days after the event. The current availability of new imaging techniques, such as diffusion-weighted MRI, that identify lesion volume in the acute setting, may improve our ability to predict stroke outcomes in the acute setting. Toni et al. also evaluated a model that combined clinical and imaging information in which variables were chosen by stepwise techniques, which can make the resulting model very sensitive to the characteristics of the development data though it is prone to over-fitting. Henon et al., combined clinical predictors with CT head information. In this study, the investigators evaluated a relatively large number of predictor variables despite a low death rate of 16% (the specified outcome variable), thereby potentially increasing the risk of an over-fitted model with poor external validity. The fact that infarct volume and stroke subtype data were not collected limits our ability to use the models in an AIS setting, especially with larger volumes. However, this analysis does provide evidence that laboratory and clinical information together can be predictive of patient outcome at three months and that they constitute a more powerful tool when used together. Beta coefficients in bootstrap validation model [Table 5] confirmed that the error in VEGF and NIHSS subscore was minimal followed by prestroke functional status.
There are several limitations of this study. First, this model has only been tested in our prospective data set and not retrospectively validated with any other independent data. AIS classification and infarct volumes were not evaluated and hence were not included in the clinical model. We were careful, however, to avoid overspecification of the model by prospectively identifying the independent variables, avoiding stepwise variable selection techniques, and retaining all candidate variables in the final model. The serum VEGF was not recorded serially in all patients and not at of the three month period, values which would have elucidated the efficacy of the model and prognosticate recovery with a better understanding. A recent study by Medvedkova et al. to investigate an interrelationship between VEFG-1 elevation and neurologic recovery in short-term period among patients with ischemic hemispheric stroke found that VEGF-1 was 348.55 pg/ml, a higher level in mRS ≥3 versus mRS ≤2 population subgroups (P < 0.05) was observed. Diabetes mellitus type 2 was found to be an independent predictor for favorable neurologic recovery in the follow-up period. Increased VEGF expression may provide more long-term beneficial effects as a result of continued angiogenesis over several months., Additional longitudinal studies are required with other multiple growth factors/cytokines that are modified during ischemic stroke (including platelet-derived growth factor and basic fibroblast growth factor) in association with acute stroke classification and similar initial disability levels. Radiological and laboratory variables should also be included in the assessment modules to predict recovery post-ischemia.,
In summary, our results suggest that age, neurological functioning, a history of prior stroke, prestroke functionality, and VEGF biomarker level are all important independent predictors acutely and at in the three-month short-term period. The CB module combines all of these measures into a continuous survival score, which better predicts severity than data from a single domain. This score should therefore be considered a “work in progress,” with further investigations needed. Providing accurate prognostic tools is particularly important both for researchers in the health policy arena and for clinicians endeavoring to care for their patients.,
Serum VEGF is increased immediately till one week after AIS. Patients with severe deficits have a higher VEGF biomarker levels VEGF as compared patients with mild strokes. The addition of this biomarker in the clinical module wil help in prognosticating recovery after AIS. The new CB module can be validated in in future large cohort studies.
Financial support and sponsorship
The study was funded by Department of Science and Technology (DST).
Conflicts of interest
There are no conflicts of interest.
[Figure 1], [Figure 2]
[Table 1], [Table 2], [Table 3], [Table 4], [Table 5]