Topographical Patterns of Intraventricular Hemorrhage as a Novel Prognostic Imaging Biomarker in Intracerebral Hemorrhage

2.1. Study Design and Population

The internal cohort was derived from a prospectively established standardized clinical database of patients with spontaneous ICH at the Stroke Center of Yueyang People's Hospital between January 2019 and October 2024. For external validation, an independent cohort from a prospectively maintained registry at the Stroke Center of the Fourth People's Hospital of Changsha between December 2024 and June 2025 was used. Both cohorts strictly adhered to the same eligibility criteria.

Inclusion criteria were as follows: (1) computed tomography (CT)‐confirmed ICH with ventricular extension or primary IVH; (2) hematoma volume < 40 mL; (3) 18–90 years of age; and (4) hospital admission and treatment initiation within 24 h of ICH onset. Exclusion criteria were as follows: (1) IVH secondary to traumatic brain injury, ruptured aneurysm or vascular malformation, spontaneous subarachnoid hemorrhage, tumor hemorrhage, or ischemic stroke; (2) brainstem hemorrhage; and (3) severe physical or psychiatric conditions causing significant functional impairment or life‐threatening complications.

All patients were diagnosed and treated according to the Chinese Guidelines for Diagnosis and Treatment of Intracerebral Hemorrhage (2019) [21]. Surgical hematoma evacuation was generally considered for supratentorial primary hemorrhage ≥ 30 mL or cerebellar hemorrhage ≥ 10 mL with reduced consciousness. External ventricular drain (EVD) placement was performed in patients with hydrocephalus, with bilateral EVD placement when CT showed complete obstruction of the third ventricle or bilateral occlusion of the foramina of Monro. All EVDs were placed within 24 h of repeat cranial CT. EVD removal was considered when the daily drainage volume was < 100 mL and no progressive ventricular enlargement or elevated intracranial pressure was observed 48 h after clamping.

2.2. Baseline Characteristics

Demographic, clinical, imaging, and treatment data were collected from patient medical records, including sex, age, medical history, level of consciousness on admission, body temperature, systolic and diastolic blood pressure, hemorrhage location, hematoma volume, treatment modality, IVH scores, and overall IVH topographical characteristics.

2.3. Radiological Assessment

Initial CT scans were performed within 24 h of IVH onset and prior to EVD placement. Parenchymal hematoma volume was calculated using the formula (1/2 × A × B × C), where A, B, and C represented the maximum measured width, length, and height, respectively. IVH severity was evaluated according to the oGS (range, 0–12) and modified Graeb score (mGS; range, 0–32). Hydrocephalus was diagnosed on the initial CT scan obtained at admission, using the following criteria: maximum frontal horn diameter divided by the maximum cranial diameter in that plane (Evans' index) > 0.3, maximum frontal horn diameter > 45 mm, third ventricular diameter > 6 mm, or fourth ventricular diameter > 12 mm [22, 23].

2.4. Outcome Evaluation

The study outcomes were 30‐day mortality and 6‐month functional outcome. Functional outcomes were assessed using the modified Rankin Scale (mRS), with favorable outcomes defined as mRS 0–3 and unfavorable outcomes as mRS 4–6.

2.5. Classification of IVH Topographical Features

In the development cohort, 11 distinct IVH topographical patterns were defined based on the presence or absence of hemorrhage in four ventricular compartments: the left lateral ventricle, right lateral ventricle, third ventricle, and fourth ventricle. The patterns are as follows: (1) unilateral lateral ventricle only (left or right); (2) bilateral lateral ventricles only; (3) unilateral lateral + third ventricle; (4) unilateral lateral + fourth ventricle; (5) unilateral lateral + third + fourth ventricles; (6) bilateral lateral + third ventricle; (7) bilateral lateral + fourth ventricle; (8) bilateral lateral + third + fourth ventricles; (9) isolated third ventricle; (10) isolated fourth ventricle; (11) isolated third and fourth ventricles. Two senior neurosurgeons (R.Z. and J.L.), blinded to clinical outcomes, independently reviewed all CT scans to classify IVH topographical patterns. Inter‐rater reliability was assessed using Cohen's kappa coefficient, with values > 0.80 (on a scale from 0 to 1) considered to indicate excellent agreement. Discrepancies were resolved by consensus with a third senior neurosurgeon (A.C.).

2.6. IVH Topographical Score Development

First, we constructed three multivariate prognostic models: (1) baseline clinical factors + oGS, (2) baseline clinical factors + mGS, and (3) baseline clinical factors + IVH topographical features. Model discrimination was compared using the area under the receiver operating characteristic (ROC) curve (AUC) to assess whether IVH topographical features could substitute for two previous IVH scores in the prediction models.

Based on the regression coefficients (β values) of significant predictors from the constructed IVH topographical features model, a novel IVH scoring system named the IVH topographical score (ITS) was developed by assigning weighted points to these predictors. ITS reliability was validated in the external cohort (n = 102).

The ITS was compared with four established ICH scoring systems: the original ICH score (oICH) [10], ICH‐GS [11], and two modified ICH scores (mICH‐A and mICH‐B) [12]. Predictive performance was evaluated using ROC analysis, wherein curves were plotted, AUCs were calculated, and the optimal diagnostic cutoff values for each scoring system were determined. Confusion matrices were constructed, and the accuracy, sensitivity, specificity, precision, and F1‐score were determined.

2.7. Statistical Analysis

All statistical analyses were performed using R software (version 2020; R Foundation for Statistical Computing; Vienna, Austria). Statistical significance was set at p < 0.05. Categorical variables are expressed as frequencies (percentages) and compared using the chi‐square test or Fisher's exact test. Continuous variables with a normal distribution are presented as mean ± standard deviation and compared using the independent sample t‐test. Non‐normally distributed continuous variables are presented as median (interquartile range) and compared using the Mann–Whitney U test. Potential confounding factors were adjusted for in multivariable logistic regression models to identify independent predictors of mortality. As this was a retrospective analysis of a prospectively maintained database, no formal sample size calculation was performed prior to the study. All eligible patients meeting the inclusion criteria during the study period were included. A post hoc power analysis indicated that with 385 patients and 91 events, the study had 80% power to detect an odds ratio of 1.8 for a predictor with a prevalence of 20% at an alpha level of 0.05.

3.1. Internal Cohort

Table 1 summarizes the univariate analysis and multivariate logistic regression analyses in the internal cohort. Among the 385 patients, 294 (76.4%) survived and 91 (23.6%) died within 30 days. The mean age was similar between the two groups (64.6 ± 12.1 years vs. 66.2 ± 12.0 years, respectively). The mortality group had significantly lower Glasgow Coma Scale (GCS) scores on admission (6.0 ± 2.6 vs. 11.0 ± 3.6, p < 0.001), higher oGS (7.6 ± 2.9 vs. 4.2 ± 2.4, p < 0.001), higher mGS (17.2 ± 7.9 vs. 8.2 ± 6.4, p < 0.001), larger hematoma volumes (20.8 ± 12.1 vs. 10.6 ± 8.9, p < 0.001), and a higher likelihood of hydrocephalus (63.7% vs. 9.8%, p < 0.001) than the survival group. No significant differences were observed between the two groups in terms of hematoma location. Although the mortality group showed higher rates of hematoma evacuation (25.3% vs. 18.7%, p = 0.175) and EVD placement (40.6% vs. 33.7%, p = 0.224), these differences were not statistically significant. The mortality group exhibited a significantly lower proportion of isolated lateral ventricular involvement (unilateral: 6.6% vs. 38.1%, p < 0.001; bilateral: 2.2% vs. 10.2%, p = 0.029) and a significantly higher proportion of panventricular involvement (62.6% vs. 24.1%, p < 0.001). For the classification of IVH topographical patterns, inter‐rater agreement was excellent, with a Cohen's kappa of 0.89 (95% CI: 0.84–0.94).

3.2. Predictors of 30‐Day Mortality

The overall 30‐day mortality rate in the internal cohort was 23.6%. Comparative analysis of the predictive models showed that the IVH topographical features significantly outperformed the oGS and mGS in predicting mortality (AUC: 0.941 vs. 0.807 and 0.806, respectively; all p < 0.05, DeLong test; Figure 2A). Multivariate logistic regression identified IVH topographical features (odds ratio [OR]: 3.16; 95% confidence interval [CI]: 0.96–4.85; p = 0.033), hydrocephalus (OR: 4.02; 95% CI: 1.85–8.72; p < 0.001), hematoma volume (OR: 1.07; 95% CI: 1.04–1.10; p < 0.001), and GCS score (OR: 0.77; 95% CI: 0.69–0.87; p < 0.001) as independent predictors of 30‐day mortality (Table 1).

3.3. Role of IVH Topographical Patterns

All observed topographical patterns of IVH occurrence and mortality rates are shown in Figure 1. Unilateral lateral ventricular hemorrhage was one of the most common patterns (n = 118; 30.6%) but was associated with a relatively low mortality rate (5.0%). In contrast, patterns involving the third and/or fourth ventricles were associated with substantially higher mortality rates, particularly bilateral lateral hemorrhage with third and fourth ventricular involvement (n = 128; 33.2%; mortality, 44.5%) and isolated third and fourth ventricular hemorrhage (n = 19; 4.9%; mortality, 36.8%). Isolated third or fourth ventricular hemorrhage had a mortality rate of 25.0%. Furthermore, bilateral lateral hemorrhage with third or fourth ventricular involvement was associated with a significantly higher risk of death than unilateral lateral hemorrhage with similar ventricular involvement. Detailed frequencies and mortality rates for all 11 patterns are provided in Figure 1.

3.4. ITS Development

The ITS was developed based on the independent predictors identified by the logistic regression analysis (Table 1). Each predictor was assigned a different point value, yielding a possible total score of 9 (Table 2). The 11 IVH topographical patterns were grouped according to their ORs and assigned points as follows: isolated lateral ventricular hemorrhage (including unilateral and bilateral), which had the lowest OR, was assigned 0 points; isolated third or fourth ventricular hemorrhage, as well as unilateral lateral hemorrhage with third or fourth ventricular involvement, all of which showed similar ORs, were assigned 1 point; combined third and fourth ventricular hemorrhage and bilateral lateral hemorrhage with third or fourth ventricular involvement, both of which had approximately double the OR of the previous group, were assigned 2 points; and panventricular hemorrhage (bilateral lateral hemorrhage with both third and fourth ventricular involvement) was assigned 3 points. GCS scores were categorized as 3–8 and 9–15 points and assigned 1 and 0 points, respectively. Hematoma volume was dichotomized as ≥ 20 mL or < 20 mL and assigned 1 or 0 points, respectively. Hydrocephalus was treated as a binary variable and assigned 4 or 0 points, respectively.

The relationship between increasing ITS and 30‐day mortality is shown in Figure 3A. Among patients with scores of 0, 1, 2, 3, 4, 5, 6, 7, 8, and 9, the corresponding 30‐day mortality rates were 0.9%, 7.0%, 8.3%, 19.6%, 39.3%, 40.0%, 50.2%, 55.5%, 66.7%, and 90.6%, respectively.

3.5. Score Comparison

In the internal cohort, the ITS showed the highest AUC (0.906) for predicting 30‐day mortality (all p < 0.05, DeLong test), followed by the mICH‐A (0.862), mICH‐B (0.851), ICH‐GS (0.801), and oICH (0.799) scores (Figure 2B). In the external cohort, the ITS also showed the highest AUC (0.932) (all p < 0.05, DeLong test), followed by the mICH‐A (0.871), mICH‐B (0.860), ICH‐GS (0.818), and oICH (0.814) scores (Figure 2C). The mICH‐A score achieved the highest AUC (0.893) for predicting 6‐month functional outcomes (Figure 2D), followed by the ITS (0.887), mICH‐B (0.883), ICH‐GS (0.823), and oICH (0.809) scores. Calibration and decision curve analyses are provided in the Figures S1 and S2, respectively.

At the optimal cut‐off value determined by maximizing Youden's index in the internal cohort, confusion matrices were constructed for each scoring system to evaluate their classification performance for 30‐day mortality. The ITS demonstrated the highest overall accuracy (0.844), sensitivity (0.604), and F1‐score (0.647), indicating superior performance in identifying patients at risk of death while maintaining a balanced trade‐off with precision (0.696) and specificity (0.918). The confusion matrices and detailed comparative metrics for all scores are provided in the Figure S3 and Table S1, respectively.

3.6. External Cohort

Among the 102 patients in the external cohort (mean age 65.2 ± 12.3 years; range 33–89), 28 died (30‐day mortality rate: 27.5%). At 6 months, 42 (41.1%) patients achieved good functional recovery (mRS ≤ 3). Hemorrhage locations included the thalamus (50%), basal ganglia (16.7%), cerebellum (9.8%), lobar regions (6.8%), and primary IVH (11.8%). Detailed baseline characteristics of the external cohort stratified by outcome are presented in Table 3.

3.7. Outcomes

At 30‐day post‐ICH, among patients with scores of 0, 1, 2, 3, 4, 5, 6, 7, 8, and 9, the corresponding 30‐day mortality rates were 3.6%, 6.7%, 0%, 10.0%, 50.0%, 61.5%, 65.1%, 71.0%, 90.9%, and 87.5%, respectively (Figure 3B). At 6 months post‐ICH, unfavorable outcomes were observed across all ITS categories. No patient with an ITS > 5 points achieved favorable functional recovery. Among patients with ITS scores of 0, 1, 2, 3, 4, and 5, the rates of unfavorable outcomes were 28.6%, 45.0%, 25.0%, 53.3%, 75%, and 90%, respectively (Figure 3C).

Based on the relationship between the ITS and 30‐day mortality, we developed a risk stratification to classify patients into three risk groups: low risk (ITS 0–2), intermediate risk (ITS 3–6), and high risk (ITS 7–9). In the internal cohort, mortality rates were 5.2% (low), 35.8% (intermediate), and 72.4% (high). In the external validation cohort, the corresponding rates were 4.5%, 53.2%, and 86.5% (Table S2). This three‐tier stratification provides a practical tool for bedside risk assessment.

3.8. Subgroup and Sensitivity Analyses

We performed prespecified subgroup analyses stratified by age, sex, hematoma volume, GCS score, treatment modality, and hydrocephalus. The ITS maintained good discriminative ability for 6‐month functional outcome across all subgroups in the external cohort (Table S3). Several sensitivity analyses were conducted to assess the robustness of our findings. When a favorable functional outcome was redefined as mRS ≤ 2 (instead of mRS ≤ 3), the ITS remained predictive for 6‐month outcome (AUC 0.817, 95% CI 0.736–0.898) (Figure S4). When 6‐month mortality was used as an alternative endpoint, results were similar (AUC 0.893, 95% CI 0.827–0.959) (Figure S5). We also explored whether the effect of the 11‐category topographical classification was modified by hydrocephalus (Table S4). The interaction term was not statistically significant (OR = 0.24, 95% CI: 0.007–8.56, p = 0.43), indicating no detectable synergistic effect in our dataset. However, it cannot be excluded that this null finding is due to limited statistical power. Therefore, these findings should be interpreted with caution.

To assess the robustness of our classification against the modified Graeb score, we performed Spearman correlation analysis and pure imaging model comparisons. The classification was strongly correlated with the mGS (ρ = 0.847, p < 0.001). In pure imaging models without clinical covariates, the 11‐category topographical classification predicted 30‐day mortality with an AUC of 0.793 (95% CI: 0.75–0.84), which was comparable to that of the mGS (AUC 0.806, 95% CI: 0.76–0.85; DeLong test, p = 0.411). Detailed results are provided in Figures S6 and S7, respectively.

4.1. Limitations

Our study has several limitations. First, although we included a relatively large single‐center cohort of patients with IVH, the sample size of the development cohort remained limited owing to strict inclusion and exclusion criteria. In particular, the number of patients with isolated third ventricular or fourth ventricular hemorrhage was relatively small, which may have led to insufficient statistical power. Therefore, the negative findings for these patterns should be interpreted with caution. Second, both the internal and external validation cohorts in this study included patients who underwent surgical interventions such as hematoma evacuation or EVD. Although uniform eligibility criteria were applied in both cohorts and surgical decisions themselves may reflect disease severity, treatment‐related bias could still have been introduced. Third, to minimize the confounding effect of early mortality primarily driven by massive parenchymal hemorrhage or cerebral herniation on the prognostic effect of IVH itself, this study set an upper limit of 40 mL for the parenchymal hematoma volume in the included patients. The same criterion was applied in the external validation cohort, which is consistent with previous research [20]. This threshold allowed us to focus more specifically on the independent contribution of IVH but limits generalizability to patients with larger hematomas. Therefore, further validation in unselected ICH populations and in patients with larger hematoma volumes (> 40 mL) is warranted. Fourth, although the ITS outperforms conventional IVH scores in predictive performance and can rapidly estimate overall IVH severity without time‐consuming calculations, we acknowledge that no formal head‐to‐head comparison of completion time or ease of use has been performed against existing scoring systems. Future implementation studies are needed to directly compare the clinical efficiency of the ITS with these tools. Finally, despite rigorous external validation, the ITS still requires further validation in larger multicenter cohorts and across different ethnic populations to assess its generalizability.