Intracranial hemorrhage is the second most common type of stroke after ischemic stroke. It is estimated that intracranial hemorrhage occurs in approximately 10–30 per 100,000 people each year, with a mortality rate of up to 50% within one month of intracranial hemorrhage, and functional independence is achieved in less than 40% of patients [1, 2]. The incidence rate is increasing, especially in developing countries [3]. A total of 60–70% of cases of intracranial hemorrhage are attributed to hypertension, usually located in the caudate-putamen (basal nuclei), thalamus, cerebellum and pons [4, 5, 6], which presents more serious performance issues and poor prognosis. However, there is still no appropriate treatment to improve the prognosis of patients with intracerebral hemorrhage [2, 7]. Due to the sudden onset of cerebral hemorrhage and poor prognosis, it is very important for the neurologist and the patient’s family to accurately evaluate the prognosis of the patient, especially their ability to live independently. There are several clinical prediction models [8], and the most commonly used model is Hemphill’s ICH (intracranial hemorrhage) score [9]. However, these scores are not very conveniently or widely used. Nomograms are visual displays of regression equations, which has gained widespread attention in recent years [10, 11]. Here, we used several common indicators to create a nomogram clinical prediction model to assist clinicians in assessing the prognosis of intracranial hemorrhage.

This retrospective study was approved by the Ethics Committee of Tongji Hospital of the Huazhong University of Science and Technology (TJ-IRB20220118). Informed consents were obtained from patients.

In the period from 2015–2021, 2114 ICH patients presented to our hospital. 947 patients were excluded as follows: 39 patients had primary IVH; 120 had primary SAH; 427 had incomplete records; 98 had secondary ICH; 12 had drug-induced ICH; 24 had amyloid angiopathy; 9 had intracranial venous sinus thrombosis; 196 had cerebral infarction hemorrhage transformation; and 22 had neoplastic bleeding. A total of 1167 patients were ultimately enrolled and divided into a training cohort and a validation cohort according to the visiting branch.

3.4 Nomogram Prediction Model

A nomogram to predict good prognosis (mRS score at six months follow-up $\le$2), ICH score, GCS score, irregular shape rate, uneven density, IVH relation and surgery were used to construct the nomogram clinical prediction model (Fig. 1).

Fig. 1.

Nomogram plot. (A) Conventional nomogram plot. (B) Pattern nomogram plot. Total points indicate the total points and are the sum of the four indicators (hemorrhage location, IVH, D-dimer, hypertension control rate). Pr is the probability of obtaining a good prognosis from the total points.

3.5 Calibration Curve

The calibration curve shows the consistency between the probability of a good prognosis for the patient predicted by the model and the actual result. The calibration curve showed good calibration (Fig. 2A,B).

Fig. 2.

Calibration curve. (A) The calibration curve for the training curve. (B) The calibration curve for the verification curve.

3.6 DCA Curve

The DCA (decision curve analysis) curve of this model is shown in Fig. 3. The threshold probability was $\ge$7%, and the use of this model to identify patients with intracranial hemorrhage who would achieve good prognosis was better than the ‘treat-all-patients’ or ‘treat-none’ schemes (Fig. 3).

Fig. 3.

Decision curve analysis of the prediction model in the validation cohort. The dashed line represents the model, the thick solid line represents the assumption that all subjects are actively treated, and the thin solid line indicates that all patients are not treated. Decision curve analysis (DCA) shows that when the threshold probability $>$7%, using this model to identify patients with intracranial hemorrhage who may return to an mRS score $\le$2 will be better than using the treat-all-patients and treat-none schemes.

Despite the development of many advanced surgical approaches and standard medical treatments, it has been reported that only 12–39% of hemorrhage patients acquire the ability to live independently [14]. It is critical to accurately assess the prognosis of hemorrhage patients to assist clinicians in formulating the best poststroke care program. Here, we used several indicators to construct a convenient clinical prevention model. Our results showed that the ICH score, GCS score, irregular shape, uneven density, IVH relation and surgery were related to the outcomes of intracranial hemorrhage patients. Some scholars have reported that intracranial hemorrhage patient prognosis is related to age, GCS score, blood pressure, hematoma location and volume [15, 16], intraventricular hemorrhage, use of anticoagulation drugs, hematoma expansion [17] and some inflammatory factors [18], which is consistent with our results.

Hemphill’s ICH score is a widely used scoring system that incorporates admission GCS score, age, hematoma volume, IVH relation, and infratentorial/supratentorial location [19, 20]. A meta-analysis conducted by Mattishent et al. [19] showed that the Hemphill-ICH score had the most validation queues (9 studies involving 3819 patients), and the area under the curve (AUC) was 0.80. The GCS score assesses the consciousness of patients by eye-opening response, verbal response, and motor response. Shah’s results showed that the GCS score was independently associated with functional outcomes at three months after traumatic intracranial hemorrhage [16]. Wang’s [21] results showed that both the GCS score and ICH score independently predicted 30-day mortality in ICH patients. Similarly, our results showed that the ICH score and GCS score at admission were independent predictors of 6-month prognosis in patients with spontaneous ICH. The ICH score and GCS score objectively reflect the state of ICH patients, which is potentially related to hematoma volume and other indicators and is likely to predict prognosis.

The irregular shape of the intracranial hemorrhage indicates multiple sites of hemorrhage, while the uneven density indicates active hemorrhage [22]. Therefore, some scholars speculate that these two imaging features can predict the prognosis of ICH patients [23]. Barras’ results showed that ICH patients with irregular shapes had larger bleeding volumes and were more likely to experience hematoma enlargement than patients with regular shapes, and uneven density was an independent predictor of hematoma enlargement [22]. Delcourt’s results showed that irregular shape was an independent predictor of death and severe disability in ICH patients, but uneven density was not a significant predictor of prognosis [24]. Masotti’s results showed that ICH patients with irregularly shaped hematomas were more likely to require observation in the intensive care unit (ICU) ward [25]. Wang’s results showed that irregular shape was independently associated with 30-day mortality in ICH patients [21]. Combined with our study, irregular shape and uneven density are important indicators for predicting the prognosis of ICH patients. More attention should be given to ICH patients with the abovementioned imaging characteristics to obtain a good prognosis.

According to previous reports, intracranial hemorrhage rupture into the ventricle is a predictor of poor prognosis [26]. Our univariate analysis and multivariate analysis showed that the IVH relationship was an independent predictor of prognosis. A retrospective study by Nishikawa et al. [27] showed that older age, IVH volume, acute hydrocephalus, and poor initial level of consciousness were independent predictors of poor prognosis of spontaneous intracranial hemorrhage. Some scholars found that enlarged ventricle hemorrhage in patients with spontaneous intracranial hemorrhage was also associated with poor prognosis [28, 29, 30]. Li et al. [30] found that increased ventricular hemorrhage (newly bleeding ventricular hemorrhage or an increase $>$1 mL) was an independent risk factor for poor outcomes at the 90-day follow-up (mRS score 3–6).

In some emergency situations, surgical treatment is an emergency measure to save the lives of ICH patients. The surgical methods include craniotomy, minimally invasive surgery and decompression surgery. Whether surgery improves the outcome of ICH patients compared with conservative treatment has not been determined. The Surgical Trial in Intracerebral Haemorrhage (STICH) study found that early craniotomy hematoma removal did not improve the prognosis of ICH patients and may be beneficial for patients with hematoma locations $\le$1 cm from the brain surface [31]. STICH II found that early surgical clearance of lob hemorrhage did not lead to better clinical outcome and only a slight survival advantage compared with medical conservative treatment alone [32]. For cerebellar hemorrhage, patients with ventricular hemorrhage or brain stem compression have better surgical treatment results [33]. The minimally invasive surgery with thrombolysis in intracerebral haemorrhage evacuation (MISTIE) III demonstrated that minimally invasive hematoma removal reduced 365-day mortality but did not significantly improve neurological function. The degree of hematoma clearance was associated with a good prognosis (mRS score 0–3) [34]. A total of 171 patients underwent minimally invasive hematoma removal in this study (148 in the training cohort and 23 in the validation cohort), and the average hematoma volume in these patients was significantly higher than the average (35 mL in the training cohort and 24 mL in the validation cohort). Our results show that surgical treatment improved neurological function at 6 months in patients with intracerebral hemorrhage. This may be because the operation reduces the time for the complete removal of the hematoma and alleviates the direct injury by the hematoma and secondary injury caused, such as inflammation. Additionally, all the ICH patients treated surgically in the study center were carefully managed in the ICU. Close care may also be a factor in the good prognosis.

A predictive model of patients with hypertensive intracranial hemorrhage established by Ding et al. [35] found that a GCS score $\le$12 points and a hematoma volume $\ge$25 mL were independent risk factors affecting prognosis. The average volume of hematomas in the patient population of this study was approximately 10 mL (training cohort, 11.16 mL and validation cohort, 8.2 mL). A small hematoma volume may not be able to achieve the corresponding statistical power, so similar conclusions cannot be drawn. Many reports have confirmed that hematoma enlargement (absolute hematoma volume increase $\ge$12.5 mL or a proportional increase $\ge$33% compared to the baseline CT scan) affects the prognosis of patients with intracranial hemorrhage [36, 37]. As this study was a retrospective study, complete hematoma enlargement data could not be obtained, so whether hematoma enlargement could be used as an important factor in this clinical prediction model could not be verified. Hemorrhage location is a potent indicator to predict intracranial hemorrhage patient prognosis. Hu et al. [9] conducted a retrospective study to confirm that D-dimer influences hemorrhage patient outcomes and reported that infratentorial hemorrhage induced poor outcomes at the three-month follow-up (p = 0.023, OR = 28.937, 95% CI 1.602–522.77). According to the literature, diabetes mellitus [38], NLR [39], coagulation factors [40], inflammation factors [41], anticoagulant use [42, 43] and electrolyte levels [44] are also important factors affecting the prognosis of intracranial hemorrhage, but these factors were not included in our clinical prediction model. We speculated that the predictive value of some factors could not be accurately identified due to the strong collinearity of the overabundance of basic variables.

There are several nomogram models for predicting the prognosis of patients with intracranial hemorrhage. Han et al. [1] established a nomogram model to predict 30-day mortality in patients with spontaneous intracranial hemorrhage, incorporating the GCS score, hematoma location, hematoma volume, white blood cell count, and D-dimer indicators. Similarly, the GCS score, hematoma location, hematoma volume, and primary intraventricular hemorrhage were included to construct a nomogram for predicting death within 2 days in intracranial hemorrhage patients [45]. In this nomogram model established by Song et al. [46] to predict the functional status (good: mRS score 0–3, poor: mRS score 4–6) of spontaneous ICH patients at the 3-month follow-up, midline shift, noncontrast computed tomography (NCCT) time from sICH onset, GCS score, serum glucose levels, uric acid levels, and Radiomics Score (Rad-score) were included. Comparing these results with the results of the current study, the short-term prognosis (2-day mortality) of patients with intracranial hemorrhage was mainly related to the characteristics of intracranial hemorrhage (GCS score, hematoma location, hematoma volume and IVH). After gradual stabilization (30-day mortality, mRS score at 3 months, mRS score at 6 months), prognosis may be related to other factors (D-dimer level, serum glucose level, uric acid level, white blood cell count and long-term blood pressure control).

As mentioned earlier, we used the ICH score, GCS score, irregular shape, uneven density, IVH relation and surgery to construct this clinical predictive model. According to the C statistic (0.840), this prediction model has good discriminability. The calibration curve shows that the model has good calibration in the training cohort and similar results in the validation cohort. In future clinical work, the use of the abovementioned convenient and simple indicators can accurately assess the possibility of intracranial hemorrhage patients living independently six months later and provide important information for the formulation of rehabilitation programs.

There are several limitations in our study. First, this is a single-center study, and the problem of selection bias cannot be completely avoided. Second, this was a retrospective study, and errors were inevitable in data collection. Third, the sample size of our data was small, so it was difficult to avoid statistical errors in the process of data analysis. Therefore, in some cases, the application of the prediction model should be combined with clinical findings. In the future, the results of prospective studies with large samples may increase the reliability and generalization of the prediction model.

We established a nomogram model to predict the prognosis of patients with intracranial hemorrhage that included the indicators of ICH score, GCS score, irregular shape rate, uneven density, IVH relation and surgery. The model needs to be confirmed in more large clinical trials.

GCS, Glasgow Coma Scale; ICH, intracranial hemorrhage; IVH, intraventricular hemorrhage; LDL, low density lipoprotein; HDL, high-density lipoprotein; HB, hemoglobin; WBC, white blood cell; NBC, neutrophil blood cell; LBC, lymphocyte blood cell; NLR, neutrophil lymphocyte ratio; DVT, deep venous thrombosis; PE, pulmonary embolism; CT, computerized tomography; SAH, subarachnoid hemorrhage; AVM, arteriovenous malformation; APTT, activated partial thromboplastin time; PT, prothrombin time; INR, international normalized ratio; AST, aspartate aminotransferase; RDW, red blood cell distribution width; PDW, platelet distribution width; PLT, platelet; mRS, modified Rankin Scale; OR, odds ratio; AUC, area under the curve; ICU, intensive care unit; NCCT, noncontrast computed tomography.

Research program formulation and data collection—YL, XL, JW CP; paper writing—YL; language polishing, paper review and editing—CP, ZT.

This study was approved by the Ethics Committee of Tongji Hospital of the Huazhong University of Science and Technology (TJ-IRB20220118). Informed consents were obtained from patients.

We are very grateful to Gaigai Li for his help in improving the language of this article.

This work was supported by grant 81901219 and 82071330 from National Natural Science Foundation of China.

The authors declare no conflict of interest.