Consecutive recruitment of sICH patients admitted to Quzhou People’s Hospital was conducted from April 2020 to December 2022. The inclusion criteria were as follows: (1) first-ever stroke; (2) admission within 24 h of symptom onset; (3) age 18 years or older; (4) non-surgical treatment of hematoma; (5) intraparenchymal bleeding.

The exclusion criteria were the following: (1) hematoma caused by congenital or acquired coagulation abnormalities, hemorrhagic transformation of cerebral infarction, moyamoya disease, intracranial aneurysm, arteriovenous malformation, or neoplasms; (2) a history of other previous neurological diseases, such as ischemic or hemorrhagic stroke, intracranial tumors, or severe head trauma; and (3) specific diseases or conditions, such as infection in recent months, pregnancies, malignancies, uremia, liver cirrhosis, or chronic heart or lung disease (Supplementary Figure S1).

Additionally, healthy individuals who underwent physical examinations at our hospital were consecutively recruited as controls from May 2021 to July 2022. All controls met the following enrollment criteria: (1) no underlying diseases, including but not limited to hypertension, diabetes, hyperlipidemia, and coronary heart disease, and (2) normal results in routine blood tests, such as blood glucose, platelet count, hemoglobin concentration, and white blood cell count, in the normal ranges (Supplementary Figure S1).

This observational study was continually divided into a cross-sectional sub-study and a prospective cohort sub-study. In the cross-sectional assessment for investigating the longitudinal change of serum neuritin levels following ICH, venous blood was drawn at enrollment in controls, at admission in all patients, and at days 1, 3, 5, 7, and 10 after ICH in a portion of patients who were willing for blood collections at several time points. In the prospective cohort investigation for determining the feasibility of serum neuritin as a prognostic biomarker in ICH, all patients had their measurements of serum neuritin levels at admission and were observed for END and poor prognosis at the 90-day mark after ICH. This study was performed following principles set forth in the Declaration of Helsinki. The protocol of the current study was approved by the Institutional Review Committee at the Quzhou People’s Hospital (Opinion No: 2022–136). Written informed consent forms were obtained from patients’ relatives and controls themselves.

We collected basic information about the patients, including demographics (age and sex), vascular risk factors (hypertension, diabetes, and hyperlipidemia), medication history (use of statins, antiplatelet agents, and anticoagulants), and medical history. Stroke severity was assessed at admission using the GCS. All patients underwent baseline computed tomography scans of the head, and the hematoma volume was measured using the ABC/2 formula (19). Localizations of hematoma in lobular, subtentorial, intraventricular, or subarachnoid space were recorded. Upon arrival at the emergency center, non-invasive techniques were applied to reading systolic and diastolic blood pressures. A decrease of ≥2 points in GCS score within 24 h after admission was deemed as END (9). Patients with a modified Rankin Scale (mRS) score of ≥3 at post-stroke 90 days were considered to have a poor prognosis (20).

Venous blood samples of controls were collected at their recruitment into this study. Peripheral blood specimens at admission were collected from all patients, and those at days 1, 3, 5, 7, and 10 after ICH were gained from some patients who consented to blood collections at multiple time points. Blood glucose, potassium, and C-reactive protein levels and white blood cell counts were measured by employing routine methods. A quick deposition of blood specimens into 5 mL gel-containing biochemistry tubes (Ningbo Siny Medical Technology Co., Ltd., China) was undertaken to measure serum neuritin levels. Once coagulation was observed, blood samples were spun at 2,000 g for 10 min. Subsequently, the supernatants were moved into Eppendorf tubes (Eppendorf Tubes^® BioBased, China) for prompt preservation at conditions of below 80°C for later analysis. Quantifications were conducted in batches of serum neuritin levels. In detail, a batch of serum samples, which were collected in a recent quarter, were thawed for measurements at the end of a quarter in order to ensure time from blood drawings to biomarker measurements did not exceed 3 months to prevent protein from breaking down. By applying a commercially available kit (Article Number: EH0944; Wuhan Fine Biotech Co., Ltd.), serum neuritin detection by enzyme-linked immunosorbent assay was in duplicate fulfilled in a blinded mode by the same proficient personnel in accordance with the manufacturer’s instructions. The sensitivity of the kit was 18.75 pg./mL, and the detection range was 31.25–2,000 pg./mL, with both intra-assay and inter-assay variabilities below 10%. The dual measurements were calculated to acquire the average values for the statistical assessments in the end. A high correlation existed between the double measurements (ρ = 0.998, p < 0.001; intraclass correlation coefficient = 0.998, p < 0.001). Using the Bland–Altman plot, a satisfactory consensus was ascertained between the two measurements (Supplementary Figure S2).

The SPSS 23.0 (SPSS Inc., Chicago, IL, United States), MedCalc 20 (MedCalc Software, Ltd., Ostend, Belgium), GraphPad Prism 7.01 (GraphPad Software Inc., San Diego, California, United States), and the software package R (version 3.5.1)¹ were operated for plotting figures or statistical analyses. Normality evaluation by the Kolmogorov–Smirnov test was finished for measurement data, and subsequently, they were reported as means with standard deviations (SDs) or medians with 25–75 percentiles as deemed suitable. Qualitative data were presented as counts (proportions). The data comparisons between the two groups were completed using the chi-square test, Fisher exact test, t-test, or the Mann–Whitney U test as applicable. The Kruskal–Wallis H test was used to compare the differences in data among multiple groups. The Spearman correlation coefficient was generated to analyze the bivariate correlations. Subsequently, significantly correlated variables on univariate analysis (p < 0.05) were consolidated in the multifactorial linear regression model to identify variables that were independently relevant to serum neuritin levels. In order to discern independent predictors of poor prognosis indicated by mRS scores, mRS was regarded as an ordinal variable and a categorical variable (mRS 0–2 versus 3–6). The binary logistic regression model and ordinal regression model were constructed to ascertain independent predictors of poor prognosis. The variables with p < 0.05 on the univariate analysis should be integrated into the multivariable models for further analysis. Associations were expressed as odds ratios (ORs) and 95% confidence intervals (CIs). Similarly, independent predictors of END were determined using identical methods. Discrimination efficiency was assessed using receiver operating characteristic (ROC) curve analysis, with reporting of the area under the curve (AUC) and the corresponding 95% CI values. The restricted cubic splines were plotted to show a linear relationship, the combined models were visualized via the nomograms, calibration curves were generated to assess the stability of the models, and decision curves were produced to evaluate the clinical validity of the models. Half of all patients were randomly extracted for an internal validation of models. Weighted scoring systems were formed to optimize the prediction of poor prognosis and END after ICH. A two-sided p-value less than 0.05 signified a statistical difference.

Initially, the consecutive recruitment of 269 patients with ICH was completed in compliance with the prespecified inclusion criteria in Supplementary Figure S1. In total, 67 patients were ineligible and removed from this study in accordance with the predefined exclusion requirements in Supplementary Figure S1. Finally, a collective of 202 ICH cases were retained for further analysis. The baseline characteristics of all 202 patients are listed in Supplementary Table S1. Moreover, among these 202 patients, mRS scores 0, 1, 2, 3, 4, 5, and 6 were found separately in 17, 21, 73, 32, 36, 17, and 6 patients (median, 2; 25th–75th percentiles, 2–4).

A total of 91 and 57 patients experienced adverse outcomes at 90 days and END post-ICH, respectively. Moreover, a total of 54 consented to blood collection at multiple time points. Supplementary Table S1 shows no significant differences in baseline characteristics between all patients and the 54 who agreed to blood collection at several time points post-ICH (all p > 0.05). In addition, 100 healthy controls were recruited. A total of 53 males and 47 females composed controls, with a mean age of 61.1 years (SD, 14.0 years), and controls encompassed 20 tobacco smokers and 19 alcohol consumers. Statistically, there were no significant differences in age, sex, smoking, and alcohol consumption between all patients and the controls (all p > 0.05).

Serum neuritin levels of ICH patients promptly increased at admission, maintained elevated levels on day 1, reached the highest levels on day 3, and then gradually decreased from day 5 to day 10; and serum neuritin levels were significantly higher during the initial 10 days after ICH in patients than in controls (p < 0.001; Supplementary Figure S3). Serum neuritin levels in ICH patients were graphically shown to be significantly negatively correlated with GCS scores (p < 0.001; Supplementary Figure S4A) and to be substantially positively related to hematoma volume (p < 0.001; Supplementary Figure S4B). According to GCS, the patients were divided into three groups with scores of 3–8, 9–12, and 13–15 separately. In accordance with hematoma volume, the patients were categorized into two groups with bleeding sizes of below 30 mL and above 30 mL, respectively. In the form of a graph, serum neuritin levels were markedly highest in patients with the scores 3–8, were medium in those with the scores 9–12, and were profoundly lowest in those with the scores 13–15 (p < 0.001; Supplementary Figure S4C). In addition, serum neuritin levels were dramatically higher in patients with hematoma volume > 30 mL than in those with hematoma volume < 30 mL (p < 0.001; Supplementary Figure S4D). Moreover, within the framework of the Spearman test, serum neuritin levels were highly correlated with age, GCS scores, hematoma volume, diabetes mellitus, blood glucose levels, and intraventricular hemorrhage (all p < 0.05; Table 1). Moreover, a multivariate linear regression model was developed with the incorporation of the above-mentioned six significantly correlated variables, leading to the finding that admission serum neuritin levels were independently correlated with GCS scores and hematoma volume (both p < 0.05; Table 2).

Serum neuritin levels were significantly increased in patients with poor prognoses, as opposed to those with good prognoses (p < 0.001; Supplementary Figure S5).

Using the as a continuous variable in a linear analysis, it showed a clear positive correlation with the serum neuritin levels of patients (p < 0.001; Supplementary Figure S6A). Furthermore, the ICH patients were classified into seven groups based on their mRS scores from 0 to 6. There was a marked trend by a graphical representation that serum neuritin levels were increased in the order of the scores from 0 to 6 (p < 0.001; Supplementary Figure S6B). In the context of restricted cubic spline assessment, there was a linear concentration-response relationship between serum neuritin and risk of 90-day poor outcome (P for non-linear >0.05; Supplementary Figure S7). Moreover, under the ROC curve, serum neuritin levels efficaciously predicted post-stroke 90-day poor outcomes among this group of ICH patients (AUC, 0.771; 95% CI, 0.707–0.827), and serum neuritin levels >238.0 pg./mL distinguished patients with development of poor 90-day outcomes with specificity and sensitivity values of 88.29 and 56.04% (Youden index J, 0.4433), respectively (Figure 1).

From the perspective of statistics, mRS was recognized as an ordinal variable. Serum neuritin levels, GCS scores, hematoma volume, diabetes mellitus, blood C-reactive protein levels, and intraventricular hemorrhage significantly differed among seven groups, which were divided at a base of mRS (all p < 0.05; Table 3). With the help of the multifactorial ordinal regression analysis, GCS scores, hematoma volume, and serum neuritin levels were independently associated with mRS scores at 90 days after ICH (all p < 0.01; Table 4).

As shown in Table 5, patients with a poor prognosis, relative to those presenting with a good prognosis, tended to have significantly higher age, hematoma volume, blood glucose levels, and serum neuritin levels, were likely to exhibit substantially lower GCS scores, and were prone to display markedly higher percentages of diabetes mellitus and intraventricular hemorrhage (all p < 0.05). By applying the binary logistic regression analysis, GCS scores, hematoma volume, and serum neuritin levels emerged as the three independent predictors of poor prognosis at 90 days after ICH (all p < 0.01; Table 6).

As displayed in Figure 2, the ability of serum neuritin levels in prognosticating poor prognosis was equivalent to those of GCS scores (AUC, 0.807; 95% CI, 0.745–0.859; p = 0.408) and hematoma volume (AUC, 0.801; 95% CI, 0.739–0.854; p = 0.451). The independent predictors of poor prognosis, i.e., GCS, hematoma volume, and serum neuritin, were combined to configure a model, which was delineated by adopting a nomogram for visual representation (Figure 3). The combined model represented higher prognostic and predictive efficiency than any of GCS, hematoma volume, and serum neuritin (all p < 0.01; Figure 2). Under the calibration curve, the prediction model ran acceptably stably (Figure 4). Using decision curve analysis, the prediction model had satisfactory clinical fitness (Figure 5). Furthermore, 101 patients were randomly extracted from the whole group of 202 patients. Supplementary Table S1 showed no significant distinctions of baseline features between the whole group and the extracted 101 patients (all p > 0.05). There were no substantial differences in terms of the predictive ability of both serum neuritin levels and the combination model in anticipating poor prognosis under the ROC curve between all 202 patients and the extracted 101 patients (both p > 0.05; Supplementary Figures S8, S9). Moreover, we defined GCS scores of 3–8 as 2, 9–12 as 1, and 13–15 as 0. Hematoma volume ≥ 30 mL was defined as 1, and hematoma volume < 30 mL was defined as 0. Serum neuritin levels ≥238.0 pg./mL were defined as 1, while levels less than 238.0 pg./mL were defined as 0. Percentages of patients with poor prognosis, which were 13.0% (10/77), 44.6% (29/65), 78.1% (25/32), 94.7% (18/19), and 100% (9/9) successively, were gradually increased in the order of the total scores from 0 to 4.

Serum neuritin levels in patients suffering from END were significantly higher than those in those without (p < 0.001; Supplementary Figure S10). By utilizing a restricted cubic spline, serum neuritin levels were linearly correlated with END probability (P for non-linear >0.05; Supplementary Figure S11). Moreover, under the ROC curve, serum neuritin levels efficiently predicted END among this group of ICH patients (AUC, 0.794; 95% CI, 0.732–0.848), and serum neuritin levels >242.5 pg./mL distinguished patients at the development of END with specificity and sensitivity values of 86.90 and 63.16% (Youden index J, 0.5005), respectively (Figure 6).

As listed in Table 7, significantly lower GCS scores, substantially higher hematoma volumes, blood glucose levels, and serum neuritin levels, and markedly higher proportions of intraventricular hemorrhage and diabetes mellitus were confirmed in patients with END than in those without (all p < 0.05). Under application of the multivariate logistic regression analysis, GCS scores, hematoma volume, and serum neuritin levels were three independent predictors of END after ICH (all p < 0.05; Table 8). Moreover, Figure 7 shows that its discriminatory ability was similar to those of GCS scores (AUC, 0.836; 95% CI, 0.778–0.884; p = 0.314) and hematoma volume (AUC, 0.818; 95% CI, 0.757–0.868; p = 0.543). The three predictors were consolidated to establish a model, and subsequently, a nomogram was drawn for a visual manifestation of the prediction model (Figure 8). Under calibration curve evaluation, the prediction model was marked by high stability (Figure 9). Using decision curve analysis, the prediction model was clinically valid (Figure 10).

The combined model demonstrated higher END predictive efficiency for END compared to its individual components (all p < 0.05; Figure 7). Furthermore, no significant differences were observed in the predictive ability of both serum neuritin levels and the combined model for anticipating END under the ROC curve between all 202 patients and the subset of 101 patients (both p > 0.05; Supplementary Figures S12, S13).

We categorized GCS as follows: scores of 3–8 were defined as 2, scores of 9–12 as 1, and scores of 13–15 as 0. Hematoma volume was categorized as ≥30 mL (defined as 1) < 30 mL (defined as 0). Serum neuritin levels were categorized as ≥242.5 pg./mL (defined as 1) and <242.5 pg./mL (defined as 0). The percentages of patients with END increased progressively, with total scores ranging from 0 to 4: 6.5% (5/77), 12.3% (8/65), 56.3% (18/32), 94.7% (18/19), and 88.9% (8/9), respectively.