New risk score of the early period after spontaneous subarachnoid hemorrhage: For the prediction of delayed cerebral ischemia

Abstract Background and Purpose The aim of this study is to identify the early predictors for delayed cerebral ischemia (DCI) and develop a risk stratification score by focusing on the early change after aneurysmal subarachnoid hemorrhage (aSAH). Methods The study retrospectively reviewed aSAH patients between 2014 and 2015. Risk factors within 72 hours after aSAH were included into univariable and multivariable logistic regression analysis to screen the independent predictors for DCI and to design a risk stratification score. Results We analyzed 702 aSAH patients; four predictors were retained from the final multivariable analysis: World Federation of Neurosurgical Societies scale (WFNS; OR = 4.057, P < .001), modified Fisher Scale (mFS; OR = 2.623, P < .001), Subarachnoid Hemorrhage Early Brain Edema Score (SEBES; OR = 1.539, P = .036), and intraventricular hemorrhage (IVH; OR = 1.932, P = .002). According to the regression coefficient, we created a risk stratification score ranging from 0 to 7 (WFNS = 3, mFS = 2, SEBES = 1, and IVH = 1). The new score showed a significantly higher area under curve (0.785) compared with other scores (P < .001). Conclusion The early DCI score provides a practical method at the early 72 hours after aSAH to predict DCI.


| INTRODUC TI ON
Aneurysmal subarachnoid hemorrhage (aSAH) is a serious subtype of hemorrhagic stroke carrying high mortality and morbidity. [1][2][3] It is well known that delayed cerebral ischemia (DCI) plays an important role in the development of unfavorable outcomes after aSAH. 4 The incidence of DCI is influenced by several characteristics, including demographics (such as age, sex, personal history, and past medical history), 5 clinical status, and radiological changes at admission. 6,7 Previous studies have established several grading systems based on clinical or radiologic factors to predict incidence of DCI or outcome to guide treatment. 8 The Glasgow Coma Scale (GCS), Hunt-Hess (HH), and World Federation of Neurosurgical Societies (WFNS) are the widely used clinical grading scales, focusing on signs and symptoms to assess brain injury and prognosis; however, these grading systems do not take into account the volume and severity of bleeding. 8 With the development of computed tomography (CT), some radiographic scales have been further established. By quantifying thickness and location of subarachnoid blood on CT image, the Fisher Scale (FS) and the modified Fisher Scale (mFS) predict the incidence of cerebral vasospasm and DCI. 9 Recent Subarachnoid Hemorrhage Early Brain Edema Score (SEBES) is a new scoring system which reflects the degree of early brain injury (EBI) to predict occurrence of DCI.
However, further studies are needed to determine the accuracy and effectiveness of this score. 7,10 The weakness of these radiological scales is underestimating the importance of the patients' clinical signs. Thus, some combined grading systems were promoted, such as VASOGRADE (VG) and the HAIR scale, to predict the outcome of aSAH patients. 11,12 However, these grading systems do not consider the importance of EBI. Recent studies have demonstrated that the incidence of DCI was associated with the degree of severity of EBI after SAH. 10 Our objective is to create a new risk score consistent with clinical and radiologic factors, and place emphasis on brain changes in the early period after SAH (within 72 hours) to predict DCI. Considering the complexity and multifactor aspects of aSAH progress, a new risk score would be established by risk stratification which integrates risk factors at early 72 hours. An efficient scoring system may provide early guidance for DCI prevention after aSAH.

| Study population
Our study retrospectively reviewed 1119 consecutive SAH patients admitted to our institution from January 1, 2014, to December 31, 2015. SAH was diagnosed by initial CT scan or lumbar puncture at the time of admission. Negative angiograms and arteriovenous malformation ruptures were excluded. The exclusion criteria also included the following: SAH due to trauma or suspicious trauma; patients with previous history of brain injury (such as stroke and cerebral hemorrhage, which left chronic change on the CT); SAH accompanied by serious comorbidities before SAH onset (such as severe coagulation disorders, malignant tumor, uncontrollable heart disease, and hypertension, which would interfere with clinical judgment); patients whose initial CT scan was not available for review; and patients whose initial CT was performed more than 3 days after initial presentation of SAH (in order to ensure the consistency of evaluation time of clinical and radiological data). All aspects of this study received approval from the Institutional Review Board of the Second Affiliated Hospital of Zhejiang University.
Informed consent was either obtained by the patients, family members, or waived by the Institutional Review Board.

| Variables
Demographic information, clinical, and radiological data of aSAH patients at admission were collected as the main variables.
Demographic information included age (analyzed by continuous variable and categorical variable which was stratified into >40, >50, >60, and >70), sex, history of smoking, drinking, hypertension, diabetes, hyperlipidemia, and use of anticoagulants. Early brain change was quantified by clinical and radiological variables. Clinical variables included WFNS grade 13 and HH scale. 14 Poor clinical condition was defined as high WFNS (4)(5) and HH (4)(5). Radiological variables included intraventricular hemorrhage (IVH) and intra-parenchymal hematoma on the initial CT scan, the mFS scale, 9 and SEBES scale. 7 Large amount of bleeding was defined as high mFS (3)(4); severe cerebral edema was defined as high SEBES (3-4).

| Outcomes
Outcomes were defined by occurrence of DCI. DCI was defined as appearing clinical vasospasm or/and delayed cerebral infarction. (a) Clinical deterioration (GCS by ≥2 points, or development of new motor deficits, which excluding other etiologies) was considered as clinical vasospasm; (b) new infarct on brain CT that was not visible on the initial CT, excluding infarctions that appeared around the aneurysm within 48 hours after aneurysm surgery or endovascular treatment, was considered as delayed cerebral infarction. 15,16 Other complications, such as rebleeding (new or expanded hemorrhage on CT), hydrocephalus, and seizures, were also recorded. All radiological data were independently and retrospectively evaluated by two blinded senior neurologists from our institution. An independent third examiner was used when there was a divergence between the two neurologists.

| Statistical analysis
Data are presented as mean ± SD, number (percentage), odds ratio (OR), and 95% confidence interval (CI). All P-values were two tailed, and a P < .05 was considered statistically significant. All statisti- For univariate analyses, continuous variables were compared between DCI patients and non-DCI patients by using unpaired Student's t-tests. Categorical variables were compared using chisquare or Fisher's exact tests. All variables with P < .10 in the univariate analysis were included in the multivariate logistic regression model. Multivariate logistic regression model using backward selection was used to determine the independent predictors of DCI.
Collinearity diagnosis analysis was performed to exclude the strong collinearity relation between variables before multivariate logistic regression. Multivariate logistic regression with stepwise backward selection was used to determine the independent predictors of DCI. The results of the multivariate logistic regression analysis are reported as regression coefficient (B), odd ratio (OR) at a 95% CI, and P-values. Based on the predictors obtained from multivariable logistic regression, we designed a risk stratification score to predict the incidence of DCI. Each predictor was given related risk score according to the ratio of corresponding B to minimum B (Bx/Bmin) and rounding to the nearest integer, which was considered to be associated with little significant difference to the calibration and discrimination of the model. 17 Performance of new DCI model was evaluated by assessing the calibration and discrimination. The discriminative ability of the risk score was first tested by the area under the receiver operating characteristics curve (ROC), and compared to other grading systems including HH, WFNS, mFS, and SEBES for prediction of DCI. The area under ROC curve (AUC) larger than 0.750 was considered to have good predictive accuracy. 17 Delong test was used to compare AUC values. 18 Calibration was assessed by the Hosmer-Lemeshow test and calibration plot in cohorts, and P-values > .05 defined good calibration.
A separate validation cohort of aSAH patients from January 2016 to April 2016 was used for internal validation of the new model. We applied the same inclusion and exclusion criteria. The performance was also evaluated by discrimination (AUC) and calibration.

| Patient characteristics
In total, 702 aSAH patients were included in the cohort. The mean age was 56.0 ± 11.2, ranging from 24 to 89, and 264 (37.6%) were male. The incidence of DCI was 27.9% (196/702) in the entire cohort  Table 1.

| Model development
Patients who suffered DCI were prone to having higher HH ( we assigned related scores to each predictor ( Table 3). The new risk score ranged from 0 to 7. The new score was named as EDCI score, which can be used to early predict DCI.

| Model performance
The distribution of the EDCI score is shown in Figure 2A. The score increase was associated with an increase in the DCI rate (P < .001 for trend). The positive predictive values for each score are shown in The discriminative ability of the EDCI risk score was good in the ROC (AUC = 0.785, 95% CI = 0.752-0.815; Figure 2B). The AUC was significantly higher compared with clinical and radiological scores. The EDCI score has the highest AUC among these grading systems (AUC WFNS = 0.724, AUC HH = 0.706, AUC SEBES = 0.660, AUC mFS = 0.627). The calibration ability was also good in Hosmer-Lemeshow test (P > .05; Figure 2C).
A total of 108 patients were included in our internal validation analysis. The baseline characteristics are described in Supplemental   Table S3. The AUC was 0.773 (95% CI = 0.683 to 0.848) for predicting DCI. The outcome was systematically similar to our predictions (P > .05 in Hosmer-Lemeshow test) ( Figure S1).

| D ISCUSS I ON
In this study, we identified a risk stratification model based on the early variables acquired within 72 hours after aSAH to predict the development of DCI. We named this new risk score "EDCI" score, which is  Large evidence proved higher mFS was a significant risk factor for DCI. 12,19,20 Additionally, among the aSAH patients, a higher WFNS score at admission was considered a risk factor for DCI. 7,20,21 The "WFNS" factor is derived from GCS score and focuses on signs and symptoms of SAH patients to reflect the primary brain injury at admission. 8 It is widely applied in many combined scores, such as VG 12 and modified WFNS. 22 Improving the FS, the mFS places greater emphasis on ventricular blood. Claassen et al 23 affirmed that IVH in lateral ventricles is an independent risk factor for the development of DCI (OR = 4.1, 95% CI = 1.7-9.8). Using the presence of bilateral IVH, they divided thin SAH into grades 0-1 and 2, thick SAH into grades 3 and 4. 24 The limitation was that large unilateral IVH was regarded as no IVH by mFS, which may weaken the influence of unilateral IVH. 25 Furthermore, in our study, we divided grades 0-2 and 3-4 into the same grade for easy record, which potentially weakens the importance of IVH. Thus, we added the IVH into the risk factors, and there was no collinearity between mFS and IVH. IVH also showed its value from multivariate analysis, which was similar in the HAIR score. 11 F I G U R E 2 A, DCI rate based on EDCI score. Distribution of the EDCI score (dark blue bars) and corresponding observed DCI rate (orange points) for each score. B, ROC of EDCI score and other grading systems. EDCI score keeps a highest AUC (AUC = 0.785, 95% CI = 0.752-0.815) among these grading systems (AUC WFNS = 0.724, 95% CI = 0.689-0.757; AUC HH = 0.706, 95% CI = 0.671-0.739; AUC SEBES = 0.660, 95% CI = 0.624-0.695; AUC mFS = 0.627, 95% CI = 0.624-0.695). P value was < .001 compared to each score. C, Calibration plot for predicted versus observed DCI for the risk EDCI score. Calibration plot, P = .522. AUC, area under receiver operating characteristics curve; CI, confidence interval; DCI, delayed cerebral ischemia; HH, Hunt-Hess; mFS, modified Fisher Scale; ROC, receiver operating characteristics curve; SEBES, Subarachnoid Hemorrhage Early Brain Edema Score; WFNS, World Federation of Neurosurgical Societies It should be mentioned that age, an important risk factor, was excluded from our score. The study has tested the age by continuous variable and categorical variable in univariate analysis. However, consistent with a recent promoted DCI risk score, there was no statistical significance found in our study. 26 The current studies on the prognostic value of age are controversial. Some studies suggested that older age is associated with lower incidence of DCI, 24 and another study suggested aging patients were more likely to suffer DCI compared with younger patients. 19 Interestingly, also other studies found that aging has no difference in the incidence of DCI. 27,28 This divergence may derive from the differences in the division of age groups in each study. Older age is associated with larger subarachnoid clot volume, which is considered to lead to DCI. 29  TA B L E 4 Related DCI risk to each risk point variable. We also excluded patients with serious comorbidities who may have had cerebrovascular events before SAH onset, as this may potentially interfere with clinical judgment. However, this may also result this manuscript cannot be generalized to total population of SAH patients. Second, an unequal distribution of patients in each stratum may be the common limitation of risk score. 30,37 This may be caused by the different score assignment of each variable. Third, despite validation of the EDCI score performance, it was not yet complete. The DCI rate for stratum 5-7 appears to abruptly change in the derivation cohort of our study. All in all, future studies need to use a larger external validation cohort to reassess performance of EDCI score in DCI prediction.

| CON CLUS ION
In summary, the EDCI score is a robust tool for quickly predicting the early incidence of DCI after aSAH. It obtains the risk factors available with 72 hours after aSAH and reflects the early brain change after aSAH. The EDCI score presents very good discriminative and calibration properties. Neurosurgeons may use this risk score to guide DCI prevention for aSAH patients.

CO N FLI C T O F I NTE R E S T
The authors declare that they have no conflict of interest.