Risk factors and a prediction model for the prognosis of intracerebral hemorrhage using cerebral microhemorrhage and clinical factors

Background This study aimed to identify the risk factors and construct a prediction model for the prognosis of intracerebral hemorrhage (ICH) at discharge, 3 months, and 12 months. Methods A total of 269 patients with ICH were retrospectively enrolled at our hospital between January 2014 and August 2016. The prognosis of ICH was assessed using the modified Rankin Scale (mRS); an mRS score > 2 was considered a poor outcome. The primary endpoint was the 3-month mRS, whereas the secondary endpoints included the mRS scores at discharge and 12 months, and mortality. Results The Glasgow Coma Scale (GCS), National Institutes of Health (NIH) stroke scale, International Normalized Ratio (INR), blood urea nitrogen (BUN), epencephalon hemorrhage, and primary hematoma volume were significantly associated with a poor mRS score at 3 months. The predictive value of the prediction model based on these factors for a poor mRS score was 87.8%. Furthermore, a poor mRS score at discharge was affected by the GCS, NIH stroke scale, and primary hematoma volume; the constructed model based on these factors had a predictive value of 87.6%. In addition, the GCS, NIH stroke scale, and surgery were significantly related to a poor mRS score at 12 months; the predictive value of the constructed model based on the aforementioned factors for a poor mRS score was 86.5%. Finally, primary hematoma volume is significantly associated with the risk of 12 months mortality. Conclusions The study identified risk factors and constructed a prediction model for poor mRS scores and mortality at discharge, 3 and 12 months in patients with ICH. The prediction models for mRS scores showed a relatively high predictive performance.


Introduction
Stroke accounted for 12.2 million incident case of stroke in 2019, and 101 million prevalent cases of stroke (1).In China, there was an estimated 17.8 million adults presented a stroke, and 2.3 million of cases dying as a result (2)(3)(4).Stroke can be classified as ischemic or hemorrhagic, with the latter including intracerebral hemorrhage (ICH) and   (7).The prognostic factors for ICH have already been identified, including older age, higher blood plasma glutamate, tumor necrosis factor alpha (TNF-α), initial ICH volume, Scandinavian Stroke Scale score, dialysis, diabetes, Glasgow Coma Scale (GCS) score, bilateral dilated pupils, higher international normalized ratio (INR), hematoma in the cerebellum or brainstem, hematoma volume, and the presence of intraventricular hematoma (8)(9)(10).However, these studies did not assess the role of cerebral microhemorrhages (CMs) in ICH prognosis.CMs have been observed in 5% of healthy elderly individuals; the risk factors for CMs include a history of stroke or dementia, hypertension, and diabetes (11).Studies have illustrated that the number and location of CMs are significantly associated with the progression of ICH (12,13); however, whether they affect the prognosis of ICH remains unclear.
Therefore, we conducted this study to identify the potential prognostic factors for ICH and constructed a predictive model for functional status based on CMs and clinical factors.

Methods . Study design and population
Patients with ICH admitted to our hospital between January 2014 and August 2016 were retrospectively collected.This study was approved by the institutional review board of Beijing  .

Data collection and variable definition
Patients information was collected from electronic medical records, including age, sex, time from onset to admission, smoking, alcohol, body mass index (BMI), hypertension, diabetes mellitus (DM), hyperlipidemia, history of ischemic stroke, history of hemorrhagic stroke, history of subarachnoid hemorrhage, antiplatelet drugs, anticoagulant drugs, antihypertensive drugs, lipid-lowering drugs, antidiabetic drugs, systolic blood pressure (SBP), diastolic blood pressure (DBP), GCS, National Institutes of Health (NIH) stroke scale, white blood cell (WBC), platelet, fasting glucose, INR, creatinine, blood urea nitrogen (BUN), total cholesterol (TC), triglyceride (TG), high density lipoprotein (HDL), low density lipoprotein (LDL), glytamic-pyruvic transaminase (ALT), glutamic oxalacetic transaminase (AST), alkaline phosphatase (ALP), surgical treatment, location, primary hematoma volume, deep CMs, lobar CMs, subtentorial CMs, and total CMs.Blood samples were obtained in all patients within 1 h after admission.The fasting lipid profiles were obtained on the second day of admission.The susceptibility-weighted imaging showed very low signal intensity lesions was defined as CMs.The signal of CMs was similar to the venous on the susceptibilityweighted imaging phase image, and opposite the calcification signal.The inter-and intra-observer reliability for detecting CMs with Cohen's kappa values of 0.96 and 0.98, respectively.Similarly, the Cohen's kappa values of inter-and intra-observer reliability for assessing the location of CMs were 0.96 and 0.98, respectively.The modified Rankin Scale (mRS) was used to assess the prognosis of ICH, and a poor outcome was defined as an mRS score > 2 (14).Functional outcomes (mRS) were assessed after discharge, at 3 months, and at 12 months.The primary endpoint was the 3-months mRS, whereas the secondary endpoints included the mRS scores at discharge and 12 months, and mortality.

. Statistical analysis
The baseline characteristics of the patients with ICH with good or poor functional outcomes were assigned as continuous and categorical variables, respectively.Continuous variables are presented as mean (standard deviation) or median (interquartile range) according to the data distribution, while categorical variables are shown as number and frequency.Differences between good and poor functional outcomes were analyzed using the independent t test, Kruskal-Wallis test, or chi-square test.Univariate logistic regression analysis was performed to identify potential risk factors, and the factors were subjected to the multivariate logistic regression analysis using α = 0.05 and β = 0.10.The deep CMs, lobar CMs, subtentorial CMs, and total CMs were mandatory inclusion model, and the prediction model for functional outcomes at discharge, at 3 months, and 12 months was assessed using the receiver operating characteristic (ROC) curve with the area under the curve (AUC).All tests were two-sided, and P < 0.05 was regarded as statistically significant.SPSS version 18 for Windows (SPSS, Chicago, IL, USA) was used to perform statistical analyses.

. Baseline characteristics
Of 269 included patients, the mean age was 56.05 years, and 71% of included patients were male.The median time from onset to admitted was 5.16 h.Thirteen patients (4.83%) received surgical treatment, while the remaining patients treated with conservative therapy.The baseline characteristics according to the 3-month functional outcomes are shown in Table 1.There were significant differences between mRS ≤ 2 and mRS > 2 for age (P =

Discussion
This study first identified the risk factors for functional outcomes and mortality at discharge, at 3 months, and at 12months in patients with ICH.A multifactorial predictive model for the risk of poor functional outcomes was constructed, which could be used to screen high-risk patients, and effective treatments could be applied to improve the prognosis of ICH patients.Our study recruited 269 patients with ICH, of which 94 patients reported a 3-month mRS > 2. The 3-month mRS score was affected by the GCS, NIH stroke scale, INR, BUN, epencephalon hemorrhage, and primary hematoma volume.Moreover, the GCS, NIH stroke scale, and primary hematoma volume could affect the mRS at discharge, while the mRS at 12 months could be affected by the GCS, NIH stroke scale, and surgical treatment.Furthermore, the risk of 12-months mortality could affected by primary hematoma volume.The prediction model for mRS scores at discharge, at 3 months, and at 12 months showed a relatively high predictive performance.
Several studies have constructed prediction models for ICH patients (15-19).Fukuda et al. identified 187 patients with aneurysmal subarachnoid hemorrhage and found that a constructed model containing D-dimer was associated with a better discrimination ability for poor outcomes (15).Wang et al. (16) applied an automated machine learning-based approach to construct a prognostic model for patients with ICH and pointed out that the random forest provides the best predictive performance.(18) found that the important features for functional outcomes included the GCS, Charlson Comorbidity Index, ICH score, ICH volume, pupillary status, brainstem location, age, anticoagulant/antiplatelet agents, intraventricular hemorrhage, and cerebellar location, and the discriminative ability was high.Wu et al. (19) identified 83 patients with ICH and found that the combined model applied radiomic scores obtained from intraparenchymal hemorrhage, intraventricular hemorrhage, and clinical characteristics with high accuracy in predicting poor functional outcomes.However, these studies did not systematically identify the risk factors for poor functional outcomes in patients with ICH, and the characteristics of CMs have not been addressed.Thus, the current study was performed to identify risk factors and construct a prediction model for functional outcomes at discharge, at 3 months, and at 12months in patients with ICH.
Our study found that the risk factors at various time points differed; similar risk factors included the GCS and NIH stroke scale.A potential reason for this could be that these two scales reflect the neurological status of patients with ICH (20,21).Additional factors potentially influencing functional outcomes included INR, BUN, epencephalon hemorrhage, surgery, and primary hematoma volume.Several reasons could explain these results: (1) elevated INR has already been demonstrated to be associated with a poor Our study reported the risk of 12-months mortality could affected by primary hematoma volume, which could explained by the primary hematoma volume are significantly related to the severity of disease and subsequent treatments.Moreover, considering the incidences of mortality at discharge and 3months were lower than expected, thus the power was not enough to detect potential associations.In addition, although our study found that functional outcomes were not affected by CMs, especially deep, lobar, subtentorial, and total CMs, the prediction model based on CMs and clinical factors for mRS > 2 at discharge, 3 and 12 months had a relatively high predictive performance.Considering that the risk factors for poor functional outcomes have already been identified, effective strategies should be applied to patients at a high risk of poor functional outcomes.
This study has some limitations.First, the analysis was based on retrospective data, and the results of our study could have been affected by selection and recall biases.Second, our study was restricted by single-center study with a small sample size, thus the conclusions of this study should be recommended cautiously.Third, the severity of ICH was not restricted as an inclusion criterion, and the prognosis of ICH could be affected by the presence of more severe or mild symptoms.Fourth, the background therapies for ICH differed among the included patients, which could have affected the prognosis of ICH.Fifth, biomarkers levels change over the course of follow-up was not addressed, which needed further explored.Sixth, the predictive model was not verified using an external cohort.
This study identified the risk factors for functional outcomes and mortality at discharge, at 3 months, and at 12 months in patients with ICH, and the CMs were addressed.Moreover, a prediction model was constructed based on the identified risk factors with relatively high predictive performance, which could be applied in clinical practice to identify high-risk patients.Further large-scale prospective studies are required to validate the constructed model.

FIGURE
FIGUREReceiver operating characteristic curve for the risk of a -month modified Rankin Scale (mRS) score > in patients with intracerebral hemorrhage (ICH), including the six-component risk factor model.

FIGURE
FIGUREReceiver operating characteristic curve for the risk of a -month modified Rankin Scale (mRS) score > in patients with intracerebral hemorrhage (ICH), including the three-component risk factor model.
Katsuki et al. (17) used data from 140 patients with hypertensive ICH and found that the prediction model constructed using a deep learning framework was superior to the model derived from the ICH score, ICH Grading Scale, and FUNC score.Trevisi et al.
TABLE The baseline characteristics of collected patients.TABLEThe risk factors for -months mRS in ICH patients.
TABLE The risk factors for mRS at discharge in ICH patients.
TABLE The risk factors for -months mRS in ICH patients.
TABL26)he risk factors for -months mortality in ICH patients.BUN is an important index to assess renal function, which could reflect the severity of disease status in patients with ICH (23); (3) hemorrhage location is significantly related to the prognosis of ICH because the invasion sites are associated with the severity of disease (24); and (4) patients treated with surgery are significantly related to hematoma volume, which could affect the prognosis of ICH(25,26).Therefore, early intervention strategies should be implemented for identified modifiable risk factors in order to improve patient prognosis.