Development and validation of a predictive model for the prognosis in aneurysmal subarachnoid hemorrhage

Abstract Background This study was to conduct a predictive model for the prognosis of aneurysmal subarachnoid hemorrhage (aSAH) and validate the clinical data. Methods A total of 235 aSAH patients were enrolled in this study, dividing into the favorable or poor prognosis groups based on Modified Rankin Scale (mRS) at 3 months postoperatively. Multivariate analysis was assessed using binary Logistic regression and Fisher discriminant analysis. The receiver operating characteristic (ROC) curve was used to determine the cut‐off value. Results Our findings showed that the high Glasgow Coma Scale (GCS) score 24‐hour after surgery reduced the risk of poor prognosis, and the surgical clipping and elevated neutrophil‐lymphocyte ratio (NLR) increased the risk of poor prognosis. The discriminant function was V = 0.881 × GCS score − 0.523 × NLR − 0.422 × therapeutic approach, and V = −0.689 served as a cut‐off value. When V ≥ −0.689, the good prognosis was considered among these patients with aSAH. The correctness for predicting the prognostic outcomes by self‐validation was 85.11%. Conclusion This predictive model established by a discriminant analysis is a useful tool for predicting the prognostic outcomes of aSAH patients, which may help clinicians identify patients at high risk for poor prognosis and optimize treatment after surgery.

the clinical data indicated that the prognosis of aSAH was poor after surgery. Therefore, prediction of prognostic outcomes is of great value for treatment options and assessment in patients with aSAH.
Several studies have mentioned that biochemical indicators could serve as predictive factors of the prognosis in patients with aSAH, such as lipoprotein-associated phospholipase A2, high-sensitivity C-reactive protein. 12,13 However, the change of a single indicator may not provide strong and sufficient clinical evidences for clinicians to diagnose the diseases. The predictive models are a derivative tool in statistics which are useful to predict the prognostic outcomes on the basis of the pooled evaluations of physical, laboratory, and radiologic examinations. 14 High-quality predictive models can be responsible for guiding clinical decisions and patient counseling, insuring rational allocation of resources to decline the healthcare costs, and improving the designs and analyses of clinical trials. 14 The optimal care for aSAH patients is applied in clinic, but high mortality cannot be avoided, and the long-term quality of life among survivors is unsatisfactory postoperatively. The establishment of a productive model may be beneficial in the management of patients with aSAH.
In the current study, we conducted a predictive model for the prognosis of aSAH and validated the clinical data, which may be useful for clinicians to improve the prognostic outcomes of patients after surgery.

| Patients
A total of 235 aSAH patients admitted to Meizhou People's Hospital were screened in this study between December 2016 and December 2018. All cases were consecutively recruited in the retrospective investigation that were divided into the favorable (n = 201) or poor (n = 34) prognosis groups based on Modified Rankin Scale (mRS) at 3 months postoperatively. 15 In this study, patients at score 4-6 manifested poor prognosis, while patients at score 0-3 showed favorable

| Laboratory examination
The blood biochemical indexes were tested using the automatic bio-

| GCS score
Glasgow Coma Scale score mainly includes Motor response (M), Verbal response (V), and eye-opening (E). The degree of coma is evaluated based on the sum of M, V, and E scores. A score above 14 was considered as normal, a score below 7 was coma, and GCS score ≤3 was indicated brain death or poor prognosis.

| Statistical analysis
Statistical analysis was performed using SPSS 23.0 (SPSS, Inc, Chicago, IL). Measuring data were presented as the mean ± standard (x ± s) and analyzed by ANOVA. Counting data were presented as n (%) with χ 2 or Fisher tests. The single-factor analysis of the prognosis at 3 months postoperatively was carried out by t, Kruskal-Wallis or χ 2 tests. Multivariate analysis was assessed using binary Logistic regression and Fisher discriminant analysis. The receiver operating characteristic (ROC) curve was used to determine the cut-off value. P < .05 was considered statistically significant.

| The baseline data of patients with aSAH
The process of patient selection was shown in Figure 1. Totally, 235 patients with aSAH were included in this study containing 93 males (39.60%) and 142 females (60.40%), with the mean age of (58.10 ± 9.97) years and the mean BMI of (23.08 ± 2.20). There were no statistical differences in gender (χ 2 = 0.04, P = .  (Table 1).

| Multivariate Logistic regression analysis for the prognosis at 3 months postoperatively
In Table 3, the stepwise regression analysis was used to analyze the influence factors of the prognosis at 3 months postoperatively in patients with aSAH. The findings showed that the higher the GCS score 24-hour after surgery, the lower the risk of poor prognosis (OR = 0.34, 95% CI: 0.18-0.63, P = .001). The risk of poor prognosis in the surgical clipping was higher than that in the endovascular therapy (OR = 3.34, 95% CI: 1.02-10.99, P = .033). In addition, the higher the NLR, the higher the risk of poor prognosis (OR = 1.13, 95% CI: 1.03-1.24, P = .008).

| The predictive model for the prognosis at 3 months postoperatively
The GCS score 24-hour after surgery, surgical techniques, and NLR was as independent variables, the prognosis at 3 months postoperatively was as the grouping variable, and Fisher discriminant analysis was carried out. Then, the discriminative equation was as follows:

| D ISCUSS I ON
Aneurysmal subarachnoid hemorrhage is a serious disease that threatens human health. Previous studies reported that >30% of mortality F I G U R E 1 ROC curve of the prognosis in patients with aASH was related to the existence of aSAH, and merely about 30% of aSAH patients could return to independent living. 16 Approximately 10%-25% of acute aSAH patients die after bleeding or before arrival at the hospital. 17 Thus, it is of great value to pay attention to the prognostic outcomes of aSAH patients to improve their quality of life and survival. In this study, we established a predictive model to assess the prognosis of aSAH using a discriminant analysis, and to conduct an internal validation to identify the effectiveness of this model. Totally, 235 aSAH patients were screened and observed the recovery 3 months after surgery. Patients at score 4-6 manifested poor prognosis, while patients Abbreviations: CI, confidence interval; GCS, Glasgow Coma Scale; NLR, neutrophil-lymphocyte ratio; OR, odds ratio.

F I G U R E 2 ROC curve of the prognosis in patients with aASH
Studies have shown that the discriminant analysis has been clinically used to select significant indicators. [18][19][20] In this work, we first applied this analysis to establish a predictive model to screen the major factors of prognosis in patients with aSAH, which may be availably applied in clinic. Our findings from multivariate Logistic analysis showed that the high GCS score 24-hour after surgery could reduce the risk of poor prognosis, and the surgical clipping and elevated NLR could increase the risk of poor prognosis. Previous studies have demonstrated that NLR was associated with the outcomes and could predict the courses of different medical conditions, such as ischemic stroke, 21,22 cerebral hemorrhage, 23,24 and major cardiac events. 25,26 In our study, we found that NLR was associated with the prognostic outcomes of aSAH patients. The high level of NLR could increase the risk of poor prognosis at 3 months postoperatively. One mechanism by which neutrophils can contribute to unfavorable outcomes following aSAH is the synthesis and secretion of matrix metalloproteinase. These are enzymes able to degrade any component of the extracellular matrix and play role in brain-barrier damage and development of secondary brain injury. 27 In addition, the mechanism of NLRinduced bleeding may be contributed by multifactorial pathophysiology, such as inflammatory reaction, immune dysfunction, and the production of reactive oxygen species (ROS). 5,28-32 aSAH can induce the leukocytosis and increased neutrophils by stimulating the systemic cellular responses that can cause the brain injury and delayed cerebral ischemia. 33,34 Early studies showed that the growth and rupture of cerebral aneurysms were promoted via the leukocyte infiltration and inflammatory reaction in the wall of aneurysms. [35][36][37][38] The results of histopathology found that the thinned wall of aneurysms was related to leukocytosis. 39 Furthermore, Sheheryar et al mentioned that elevated NLR at admission could predict higher inpatient mortality among aSAH patients. 40 Giede-Jeppe et al 29 reported NLR as an independent factor for unfavorable functional outcome in aSAH. These were consistent with our findings, indicating the high level of NLR may increase the risk of poor prognosis in patients with aSAH.
In addition, we also discovered that the risk of poor prognosis in the surgical clipping was higher in comparison with the endovascular therapy. The surgical clipping is a gold standard treatment in recent decades, and Yasargil applied the microsurgical techniques to neurosurgery to improve the surgical approach to intracranial aneurysms. 41 Endovascular therapy was initially based on the use of inflatable balloons in the aneurysmal cavity in 1970s. 42

| CON CLUS ION
We established a predictive model to assess the prognosis of aSAH using a discriminant analysis, and to conduct an internal validation to identify the effectiveness of this model. Our results revealed that the correctness for predicting the favorable prognosis was 85.67%, as well as for predicting the poor prognosis was 76.47%. The accuracy obtained by discriminant analysis was 85.11%, indicating that the effectiveness of this predictive model was relatively reasonable.
These findings obtained from our study may help clinicians identify patients at high risk for poor prognosis and optimize treatment after surgery.