New models for predicting mortality and poor prognosis after supratentorial intracerebral hemorrhage

Background: The study aimed to determine the multiple related risk factors and their prognostic significance of intracerebral hemorrhage (ICH), and to develop models for predicting poor outcome and mortality. Methods: We retrospectively analyzed 141 consecutive patients with acute ICH presenting to the neurological department of Tongji Hospital from December 2016 to April 2018. Independent predictors of 6-month prognostic significance were identified by logistic regression, and discriminative performance was assessed by using the area under curve (AUC) of receiver-operating characteristic (ROC). Models of poor prognosis and mortality were developed. Results: Independent predictors of poor outcome (mRS≥3) were as follows: NIHSS score, D-Dimer, mixed hematoma density, irregular hematoma shape, ICH volume, cerebral cortical atrophy and midline shift. The sensitivity and specificity of poor prognosis model were 87.5% and 87.4%, respectively. Independent factors associated with 6-month mortality includedplatelet counts at admission, NIHSS score, eGFR, presence of intraventricular hemorrhage ( IVH ), third ventricle Sylvian fissure distance. The sensitivity and specificity of the model for mortality were 90.0% and 94.4%, respectively. Conclusions: Two models for prognostic significance of ICH were developed and could well predict the poor prognosis and mortality.


Background
Intracerebral hemorrhage (ICH) is a common neurological emergency, accounting for 20% of all strokes, and usually leads to severe disability or death. Its mortality rate at 1 month is about 40%, and about 54% in 1 year, only 12-39% of patients achieve permanent functional independence [1][2][3] . Therefore, early identification of patients at high risk of ICH is crucial for prognosis.
Previously, a number of prognostic models for mortality and functional outcome after ICH such as the ICH score, the ICH-GS score, the modified ICH score, the FUNC score and ICH index [4][5][6][7][8] have been established. The variables that they include are GCS score, ICH volume, IVH, ICH location, infratentorial ICH origin, age etc.
Recently, some other factors were found to be associated with prognosis including hematoma enlargement, the use of antithrombotic drugs, do-not-resuscitate (DNR) treatment, intense antihypertension [9][10][11] , and newfound various CT signs (island sign, black hole sign, blend sign, hematoma irregularity and heterogeneity etc.) [12][13][14][15] . Earlier studies on the prognosis of ICH have covered a relatively simple and limited number of factors, and these models do not include newly identified CT signs, the sensitivity and specificity of the models also need to be improved. In our study, we have taken into account those new findings and expected to develop new prognostic models for ICH. Our data were anonymous, the requirement for informed consent was therefore waived [16] .
The images of baseline CT scan were collected including: the first head CT time from onset, ICH volume, the location of ICH, various CT signs, the midline shift, hemorrhage breaking into ventricles or subarachnoid spaces. The location of cerebral hemorrhage is divided into: lobe, basal ganglia, other (corpus callosum or capsula externa) and multi-site ICH. The ICH volume is calculated by superimposing the area of each bleeding layer and multiplying the total area by the thickness of the section [19] . Island sign, black hole sign, blend sign, liquid level, satellite sign, hematoma irregularity and heterogeneity are identified referring to previous literatures [12][13][14][15][20][21][22] . Two experienced clinicians who were blinded to the clinical profiles of the patients independently assessed all CT signs and discrepancies were determined by an experienced professor. Blend sign, black hole sign, liquid level, hematoma heterogeneity are uniformly defined as mixed hematoma density; island sign, satellite sign, hematoma irregularity are defined as irregular hematoma shape. Midline shift (MLS) was determined by creating a line connecting the anterior and posterior insertions of the falx cerebri, and measuring a perpendicular distance from the line to the septum pellucidum at the level of the foramen of Monro [23,24] . The linear measurement and visual assessment of the contralateral side of ICH were used to determine brain atrophy [25] , linear measurement markers including frontal ratio and third ventricle Sylvian fissure distance. White matter lesion (WML) was graded using the scale described by van Swieten et al [26] . Lacuna is defined as a circular or oval, subcortical, liquid (like cerebrospinal fluid density) cavity between 3 and 15 mm in diameter, consistent with previous small acute deep cerebral infarction or hemorrhage in the area of perforating branch arteriole [25] . Different treatments were recorded: conservative treatment, hematoma minimally invasive aspiration, extraventricular drainage (EVD), decompressive craniectomy, and other (hematoma minimally invasive aspiration + EVD + bone flap decompression).
Functional outcome and mortality were evaluated by the modified Rankin scale at 6 months. The mRS were divided into groups with good prognosis (mRS ≤ 2) and poor prognosis (mRS ≥ 3); basic recovery group (mRs ≤ 1), disabled group (2 ≤ mRS ≤ 5) and death group (mRS = 6); death and non-death group in univariate analysis. The mRS-9Q method was mainly used to determine the patient mRS score [27] .

Statistical analyses
For univariate analyses, the continuous variables were described by the mean ± SD values and compared using student's t test and ANOVA analysis. The categorical variables were expressed by frequency and percentage, and chi-square test of fourfold table and row list table, the Bonferroni corrected chi-square test and Fisher   exact test were used for comparison between groups. Multivariate logistic regression analysis was used to investigate the independent factors associated with the prognosis of ICH, those variables that reached P < 0.2 (or close to 0.2) in univariate analysis were considered for multivariable analysis, and the stepwise regression was performed backward. Finally, we determined the independent prognostic factors for ICH and developed models for predicting poor outcome and mortality. Univariate and multivariate receiver operating characteristic curve (ROC) analyses were performed, and the Yoden index was used to set up the cut-off value of continuous variables. Statistical analysis was performed with SPSS (version 22.0), and P < 0.05 was considered statistically significant.

Results
Of the 141 patients with ICH, the average age was 54.

Univariate Analysis
Good prognosis group (mRS ≤ 2 ) and poor prognosis group The pulse rate, presence with systemic inflammatory response syndrome (SIRS), N/L ratio, NIHSS score, white blood cell count, irregular hematoma shape, ICH volume, midline shift, third ventricle Sylvian fissure distance, the location of ICH and conservative treatments were associated with poor prognosis (Table 1).
Basic recovery group (mRS 0-1), disability group (mRS 2-5) and death group (mRS 6) Compared with the basic recovery group and the disability group, the pulse rate, the proportion of end-stage renal disease (ESRD), IVH and irregular hematoma shape, the presence of the island sign, the ratio of N/L was higher in the death group, glomerular filtration and PLT counts were lower. The proportion of patients with SIRS and subarachnoid hemorrhage, white blood cell count were lowest in the basic recovery group ( Table I in  Death and non-death group The pulse rate, presence with ESRD, SIRS, IVH and irregular hematoma shape, NIHSS score, ICH volume, PLT count, glomerular filtration rate, the location of ICH and conservative treatments were associated with mortality ( Table 2).

Multivariate Analysis
The multivariate logistic regression analysis showed that NIHSS score, D-Dimer, mixed hematoma density, irregular hematoma shape, ICH volume, cortical brain atrophy and midline shift independently predicted poor prognosis. Factors associated with mortality of ICH include: platelet count at admission, NIHSS score, IVH, third ventricle Sylvian fissure distance. Inflammation signs (elevated white blood cell count and N/L ratio) may increase the risk of mortality (P = 0.05, 0.093), and hematoma minimally invasive aspiration may reduce the risk of mortality (P = 0.053) ( Table 3).  NIHSS score at admission was independent predictors of poor prognosis of ICH and associated with high risk of hematoma enlargement (HE) [28][29][30] . In the risk score for predicting the 1-year functional outcome after ICH (ICH-FOS) [31] , as the NIHSS score increased by 1 point, the risk of poor prognosis increased to 1.1 times. We also found that NIHSS score was independently associated with long-term poor outcome and increased mortality. The ROC curve and the Yoden index showed that the cutoff value of NIHSS score for predicting poor ICH prognosis was 9.5 points, with sensitivity and specificity of 87.5% and 75.9%, respectively.
Studies on the effect of d-dimer on prognosis of ICH demonstrated that elevated baseline D-Dimer levels in the blood were independent risk factors for mortality and poor prognosis [32][33][34] . We also found that elevated D-Dimer levels can affect functional prognosis at 6 months, and the cut-off value of D-Dimer was 0.565 mg/L.
Sensitivity and specificity were respectively 66.7% and 55.2%. The possible mechanism is that elevated D-Dimer may reflect disturbance of the coagulation and fibrinolysis pathways [33] . Chih-Wei Chen et al found that dynamically-altered D-Dimer in cerebrospinal fluid (CSF) help to predict hematoma enlargement and poor prognosis, d-dimer in CSF is derived from clot lysis caused by elevated fibrinolytic activity in CSF, and is a more accurate indicator of intraventricular hemolysis than that in blood [35] .
ICH volume is one of the best predictors of poor prognosis and the cut-off values of volume vary [5,6,36,37] . Studies have found that the incidence of 30-day mortality and cerebral herniation are as high as 90% when basal ganglia hematoma volume ≥ 30 ml, even if the hematoma is removed by surgery within 24 hours after onset [38] .
In our study, the ROC curve and the Yoden index showed that the threshold for hematoma volume for predicting poor prognosis was 23.725 ml, and the sensitivity and specificity were 64.6% and 89.7%, respectively.
Previous studies on CT signs of ICH can be roughly divided into two categories: hematoma density and hematoma shape. Li Qi et al [22] proposed that black hole sign can predict hematoma enlargement, and is related to the prognosis of dysfunction (OR,8.19;95%CI:2.44-27.49; p = 0.001) at 3 months [15] . Blend signs, liquid levels are also found to be associated with poor prognosis of ICH [20,39,40] .
Barras et al [13] [14] . In our study, mixed hematoma density includes black hole sign, blend sign, liquid level, hematoma heterogeneity, and irregular hematoma shape contains island sign, satellite sign, hematoma irregularity, both are independent influencing factors for the prognosis. Due to the low positive rate and small sample size, the results of mixed hematoma density are not completely consistent with the previous studies. It is necessary to expand the sample size for verification.
In previous study, the midline shift (MLS) was found to be associated with the early neurological deterioration [41] . Kiphuth et al. used transcranial duplex sonography (TDS) to measure the midline shift within 2 weeks after ICH, and reported when MLS exceeded 4.5-7.5 mm, the mortality of ICH increased significantly. When MLS exceeded 12 mm, all had poor conditions [42] . We calculated specific values and classified MLS into quartiles, the result showed that patients with MLS of 1.43-2.42 mm have a higher risk of poor prognosis compared with MLS of 0-1.43 mm. We failed to find a direction correlation between poor prognosis and greater MLS, which may be explained by the small sample size.
We adopted linear measurements and visual presentation template methods to evaluate brain atrophy [24] , and found that severe cortical brain atrophy was independently associated with the prognosis, cortical brain atrophy may be more sensitive than central brain atrophy. Patients with the third ventricle Sylvian fissure distance less than 38.93 mm had a higher risk of death. Previous studies have found that severe WML and brain atrophy are associated with 90-day mortality or major disability [25] . Other studies have also confirmed that brain atrophy is an independent predictor of poor prognosis (OR = 1.09; CI:1.05-1.1; P = < 0.001), due to the presence of brain damage in patients with brain atrophy, which can lead to degenerative diseases such as dementia or subcortical vascular encephalopathy [43] .
Contrary to expectation, we found no signifcant relationship between lacunes, WML and prognosis, owing to the smaller sample size, the younger age of the study subjects and the lower positive rate of lacuna and WML.
Other factors such as thrombocytopenia also can increase the mortality of ICH. As for the relationship between white blood cell count, N/L ratio, hematoma minimally invasive aspiration and prognosis of ICH (P values were 0.05, 0.093, 0.053, respectively), more clinical results are needed to support.
The strength of this study is to focus on the multiple prognostic factors associated with ICH, innovatively develops new poor prognosis and mortality models of ICH, and both models have high sensitivity and specificity, the variables included in the analysis are relatively comprehensive.
We recognized several limitations in this study. First, our study was a single-center retrospective study with small sample size. Prospective studies are necessary to validate established poor outcome and mortality models. Second, other prognostic factors such as various inflammatory markers (hypersensitive C-reactive protein, interleukin, procalcitonin etc.), BMI, the admitted and the fourth day of NT-proBNP had a certain rate of loss. Third, the standardization of the sample was not enough. Finally, patients with cerebral small vascular disease only underwent CT without MRI, which might reduce the predictability. We aim to conduct a large sample prospective study in the future to overcome these limitations.

Conclusions
Two models for prognostic significance of ICH was developed and could well predict the poor prognosis and mortality. Technology. As our data were anonymous, the requirement for informed consent was therefore waived.

Consent for publication
Not applicable

Competing interests
The authors declare that they have no competing interests.

Availability of data and materials
The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.  NIHSS score at admission was independent predictors of poor prognosis of ICH and associated with high risk of hematoma enlargement (HE) [28][29][30] . In the risk score for predicting the 1-

Funding
year functional outcome after ICH (ICH-FOS) [31] , as the NIHSS score increased by 1 point, the risk of poor prognosis increased to 1.1 times. We also found that NIHSS score was independently associated with long-term poor outcome and increased mortality. The ROC curve and the Yoden index showed that the cut-off value of NIHSS score for predicting poor ICH prognosis was 9.5 points, with sensitivity and specificity of 87.5% and 75.9%, respectively.
Studies on the effect of d-dimer on prognosis of ICH demonstrated that elevated baseline D-Dimer levels in the blood were independent risk factors for mortality and poor prognosis [32][33][34] . We also found that elevated D-Dimer levels can affect functional prognosis at 6 months, and the cut-off value of D-Dimer was 0.565 mg/L. Sensitivity and specificity were respectively 66.7% and 55.2%. The possible mechanism is that elevated D-Dimer may reflect disturbance of the coagulation and fibrinolysis pathways [33] . Chih-Wei Chen et al found that dynamically-altered D-Dimer in cerebrospinal fluid (CSF) help to predict hematoma enlargement and poor prognosis, d-dimer in CSF is derived from clot lysis caused by elevated fibrinolytic activity in CSF, and is a more accurate indicator of intraventricular hemolysis than that in blood [35] .
ICH volume is one of the best predictors of poor prognosis and the cut-off values of volume vary [5,6,36,37] . Studies have found that the incidence of 30-day mortality and cerebral herniation are as high as 90% when basal ganglia hematoma volume ≥ 30 ml, even if the hematoma is removed by surgery within 24 hours after onset [38] . In our study, the ROC curve and the Yoden index showed that the threshold for hematoma volume for predicting poor prognosis was 23.725 ml, and the sensitivity and specificity were 64.6% and 89.7%, respectively.
Previous studies on CT signs of ICH can be roughly divided into two categories: hematoma density and hematoma shape. Li Qi et al [22] proposed that black hole sign can predict hematoma enlargement, and is related to the prognosis of dysfunction (OR,8.19;95%CI:2.44-27.49; p = 0.001) at 3 months [15] . Blend signs, liquid levels are also found to be associated with poor prognosis of ICH [20,39,40] . Barras et al [13] [14] . In our study, mixed hematoma density includes black hole sign, blend sign, liquid level, hematoma heterogeneity, and irregular hematoma shape contains island sign, satellite sign, hematoma irregularity, both are independent influencing factors for the prognosis. Due to the low positive rate and small sample size, the results of mixed hematoma density are not completely consistent with the previous studies. It is necessary to expand the sample size for verification.
In previous study, the midline shift (MLS) was found to be associated with the early neurological deterioration [41] . Kiphuth [42] . We calculated specific values and classified MLS into quartiles, the result showed that patients with MLS of 1.43-2.42 mm have a higher risk of poor prognosis compared with MLS of 0-1.43 mm. We failed to find a direction correlation between poor prognosis and greater MLS, which may be explained by the small sample size.
We adopted linear measurements and visual presentation template methods to evaluate brain atrophy [24] , and found that severe cortical brain atrophy was independently associated with the prognosis, cortical brain atrophy may be more sensitive than central brain atrophy.
Patients with the third ventricle Sylvian fissure distance less than 38.93 mm had a higher risk of death. Previous studies have found that severe WML and brain atrophy are associated with 90-day mortality or major disability [25] . Other studies have also confirmed that brain atrophy is an independent predictor of poor prognosis (OR = 1.09; CI:1.05-1.1; P = < 0.001), due to the presence of brain damage in patients with brain atrophy, which can lead to degenerative diseases such as dementia or subcortical vascular encephalopathy [43] . Contrary to expectation, we found no signifcant relationship between lacunes, WML and prognosis, owing to the smaller sample size, the younger age of the study subjects and the lower positive rate of lacuna and WML.
Other factors such as thrombocytopenia also can increase the mortality of ICH. As for the relationship between white blood cell count, N/L ratio, hematoma minimally invasive aspiration and prognosis of ICH (P values were 0.05, 0.093, 0.053, respectively), more clinical results are needed to support.
The strength of this study is to focus on the multiple prognostic factors associated with ICH, innovatively develops new poor prognosis and mortality models of ICH, and both models have high sensitivity and specificity, the variables included in the analysis are relatively comprehensive.
We recognized several limitations in this study. First, our study was a single-center retrospective study with small sample size. Prospective studies are necessary to validate established poor outcome and mortality models. Second, other prognostic factors such as various inflammatory markers (hypersensitive C-reactive protein, interleukin, procalcitonin etc.), BMI, the admitted and the fourth day of NT-proBNP had a certain rate of loss. Third, the standardization of the sample was not enough. Finally, patients with cerebral small vascular disease only underwent CT without MRI, which might reduce the predictability. We aim to conduct a large sample prospective study in the future to overcome these limitations.

Conclusions
Two models for prognostic significance of ICH was developed and could well predict the poor prognosis and mortality.