Previous studies have indicated that deep ICH and lobar ICH differ in terms of clinical and radiological characteristics, incidence of HE, and prognosis [11, 12]. We further stratified deep ICH and lobar ICH into two subgroups, effectively controlling for the confounding effect of location, and identified distinct risk factors for HE in each group. When constructing predictive tools for HE using these risk factors, differentiation based on the site of hemorrhage should be considered.

This study defined HE as “an absolute increase in hematoma volume > 6 mL or a relative increase in volume from baseline to follow-up NCCT > 33%” and found no significant difference in the incidence of HE between deep and lobar ICH, consistent with previous research conclusions [10, 11]. However, our findings contradict those of Kuohn et al. [12], which may be attributed to differences in the standardization of the time from onset to NCCT and the definition of HE. Our study aligns with the research conducted by Roh et al. [10, 11]. In the differential analysis of clinical and radiological characteristics between deep ICH and lobar ICH, we not only identified clinical feature discrepancies consistent with previous studies, such as age, blood pressure, history of coronary heart disease, baseline hematoma volume, midline shift, and intraventricular hemorrhage [10,11,12], but also found that patients with lobar ICH had higher van Swieten scale scores compared to those with deep ICH, indicating more severe leukoaraiosis in lobar ICH. However, in subsequent subgroup multivariate analyses, we did not find a significant association with HE, which is in agreement with previous findings [15]. This suggests that the van Swieten scale is not effective for predicting HE. Our research also revealed that subarachnoid hemorrhage, fluid-level signs, irregular signs, blend signs, and island signs were more common in lobar ICH. Because of the superficial location of lobar ICH, it is more likely to be accompanied by subarachnoid hemorrhage. Given the differences in etiology, brain tissue and nuclear structure, and age at onset between patients with lobar and those with deep ICH, the reasons for the disparities in imaging markers such as fluid-level signs, irregular signs, mixed signs, and island signs may be multifactorial and require further investigation to elucidate the underlying mechanisms.

In this study, after Bonferroni correction, fluid level, admission GCS score, and time from onset to NCCT were independently predictive of HE in patients with deep ICH. Previous research has found that fluid level can be used to predict HE, and a interrater agreement analysis for fluid level achieved a consistency coefficient of 0.89, indicating its stability for HE prediction [16, 17]. Research has suggested that fluid level reflects abnormal coagulation within the hemorrhage, leading to early precipitation of higher density proteins [18]. In our study, fluid levels were significantly more prevalent in patients with lobar ICH than in patients with deep ICH, yet the correlation between fluid levels and HE was stronger in patients with deep ICH. Whether this is related to functional coagulation differences in different brain regions remains to be further investigated. Additionally, it is worth noting that this study included patients with ICH who had received anticoagulant therapy, and all patients were treated according to the anticoagulant reversal protocols recommended by the Chinese Cerebrovascular Disease Clinical Management Guidelines, based on the type of oral anticoagulant and INR values. As shown in Supplementary Table 6–8, we explored the relationship between anticoagulant therapy and fluid levels in the overall ICH, deep ICH, and lobar ICH cohorts and found that patients with ICH who had received anticoagulant therapy were more likely to have fluid levels, although the results were not statistically significant. In this study, the OR for the presence of fluid levels was 4.77, indicating that deep ICH with fluid levels was associated with a 4.77-fold increased risk of HE compared to deep ICH without fluid levels.

In this study, we defined the time from onset to NCCT as within 24 h; however, the majority of patients (74%) had a time from onset to NCCT of less than 6 h. As depicted in Supplementary Table 9, to investigate the impact of time from onset to NCCT on the incidence of HE, we analyzed the incidence of HE in each cohort when time from onset to NCCT was less than 6 h and less than 24 h. We found that when time from onset to NCCT was less than 6 h, there was a slight increase in the incidence of HE, with no significant difference observed between the deep ICH and lobar ICH groups. Because HE predominantly occurs within the first 3–6 h after onset [2], extending the time window beyond 6 h makes it difficult to capture HE occurrences. Adjusting for the rate of bleeding had no effect on the regression results, suggesting that the rate of bleeding may simply be an associated phenomenon of a shorter time from onset to NCCT.

The admission GCS score reflects the level of consciousness in patients, with lower scores indicating more severe coma. As seen in Table 1 comparing baseline characteristics of ICH at different locations, the admission GCS scores were significantly lower in the deep ICH cohort (median 12.00; IQR 9.00–13.00) compared to the lobar ICH cohort (median 14.00; IQR 11.00–15.00). Deep ICH typically involves neuro-nuclei, with worse consciousness levels indicating more severe bleeding and localized brain injury, thus increasing the likelihood of HE. In the lobar ICH cohort, there were no differences in GCS scores between groups, suggesting that GCS score may be a specific predictor of HE for deep ICH. Research by Morotti et al. has shown that admission GCS score and time from onset to NCCT are independent predictors of severe HE [8].

In the lobar ICH cohort, after the Bonferroni correction, irregular shape of the hematoma and fibrinogen level independently predicted the occurrence of HE. Previous studies have suggested that an irregular hematoma shape may represent secondary HE, with continuous bleeding and increased hemorrhagic pressure forcing the hematoma to expand into surrounding brain tissue. Vascular rupture under pressure shear leads to hematoma spread, losing the regular elliptical structure [18]. Low fibrinogen levels in the blood hinder primary and secondary coagulation processes, increasing the risk of HE [19]. Normally, fibrinogen levels increase with age [20]. In this study, the lobar ICH group (median 71.00 years; IQR 64.00–84.00 years) was significantly older than the deep ICH group (median 64.00 years; IQR 53.00–71.00 years), but the fibrinogen (g/L) level in the lobar ICH group (median 2.62; IQR 2.29–3.31) was lower than that in the deep ICH group (median 2.65; IQR 2.27–3.17), which may be related to hemostatic treatment and the use of antiplatelet medications such as aspirin and clopidogrel.

In the study by Lattanzi et al., higher neutrophil counts, lower lymphocyte counts, and a higher NLR were associated with poorer 3-month outcomes [14]. Supplementary Table 5 demonstrates a differential analysis of neutrophil, lymphocyte, and NLR levels between patients with good and poor prognosis, which is consistent with previous research. However, because of inconsistencies in previous studies, the relationship between NLR and HE remains controversial. In our analysis of HE, compared to the HE group, the non-HE group exhibited higher neutrophil counts, lower lymphocyte counts, and a higher NLR. In contrast, studies by Alimohammadi et al. and Kim et al. found that the HE group had higher neutrophil counts, lower lymphocyte counts, and a higher NLR, with the NLR being an independent predictor of HE [6, 21]. Conversely, Fonseca et al. reported that the HE group had lower neutrophil counts, lower lymphocyte counts, and a lower NLR, which aligns with our findings [22]. A meta-analysis by Shi et al. suggested no significant association between the NLR and HE [23]. To clarify the reasons for the discrepancies in these studies, it is necessary to analyze neutrophils and lymphocytes separately. Although there are conflicting results in the studies examining the NLR and HE, it is consistent across various studies that both the HE group and the group with poor prognosis have lower lymphocyte levels. A study by Mao et al. found that adoptive transfer of regulatory T cells in a model of autologous ICH can mitigate inflammatory responses, improve brain barrier integrity, reduce cerebral edema, and decrease neuronal death [24]. This could explain why reduced lymphocyte counts are significantly associated with poor prognosis and HE. The main contradiction lies in the inconsistent relationships between increased neutrophil counts and HE across different studies. Most research focuses on the inflammatory response’s effects on the blood–brain barrier, vascular permeability, and neuronal cells. However, inflammation also has a more direct effect on promoting coagulation and thrombosis. A study by Steppich et al. showed that neutrophil-activated proteases mediate endothelial cell damage and procoagulant responses [25]. There is a complex interplay between inflammation and hemostasis, involving proinflammatory cytokines, chemokines, adhesion molecules, tissue factor expression, platelet and endothelial activation, and microparticles. Inflammation increases procoagulant factors and also suppresses natural anticoagulant pathways and fibrinolytic activity, leading to a thrombotic tendency [26]. On the other hand, thrombin-induced secretion of proinflammatory cytokines and growth factors can exacerbate the inflammatory response, creating a vicious cycle. Further research is needed to elucidate the mechanisms by which neutrophil-mediated inflammatory injury and procoagulant effects influence functional outcomes and HE in ICH.

This study also has some limitations. Firstly, it is a retrospective study, which inherently involves some selection bias that is difficult to avoid. Secondly, in this study, the window for the first CT scan was defined as within 24 h, whereas HE predominantly occurs in the early stages of cerebral hemorrhage (3–6 h). This may result in some cases (26%) not capturing the progression of HE because of delayed CT acquisition. However, in our supplementary analyses, the incidence of HE did not vary significantly. Additionally, the follow-up CT scan was set to a 72-h window. As depicted in Supplementary Table 10, we also conducted a supplementary analysis, which showed an increased incidence of HE when the window was reduced to 48 h. It is important to note that HE primarily occurs in the early stages, and controlling the window for the first CT scan ensures capturing the pre-progression state of the hematoma, which directly affects the observation of HE on subsequent follow-up CT scans. However, extending the follow-up CT window does not alter the established fact of HE. The increased incidence of HE observed with a shortened follow-up window may be attributed to two factors: first, the inclusion of patients with a lower probability of HE in the extended time window, which reduces the overall incidence of HE, and second, in clinical practice, if patients do not show progression or sudden symptoms after admission, the follow-up CT scan is typically performed between 48 and 72 h post admission, indicating that these patients are mostly in an improving condition without HE. In research, the exclusion of these patients may not reflect the true incidence of HE, and careful consideration is required to decide whether they should be included in the study population. Lastly, the analysis is based on data from a single medical center, and the results need to be further validated.