We conducted this systematic review and meta-analysis in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (Page et al., 2021). A comprehensive literature search was performed across multiple electronic databases, including PubMed, EMBASE, Cochrane Library, Scopus, and CINAHL, up to January 31, 2025. The search strategy incorporated relevant Medical Subject Headings (MeSH) and keywords, such as “Apolipoprotein E” (APOE), “APOE genotype,” “APOE ε2” and “APOE ε4,” alleles and their corresponding genotypes (ε2/ε2, ε2/ε3, ε2/ε4, ε3/ε4, and ε4/ε4), along with terms related to HS, ICH, cerebral hemorrhage, brain hemorrhage, and intracranial hemorrhage.

Two independent researchers (MN and AR) conducted the search, focusing exclusively on human studies without imposing language or publication date restrictions. To ensure a comprehensive dataset, we manually screened the reference lists of included studies, review articles, and previous meta-analyses for additional eligible studies.

Based on the eligibility criteria, we screened the titles and abstracts of the literature from the available electronic databases. We gathered the following information from each included study: the first author's name, publication year, ethnicity, country, study design, study period, target population, mean age, sex, sample size (number of HS cases and healthy controls), type of hemorrhage, matching criteria, APOE genotyping method, allelic distribution of APOE genotypes in HS cases and healthy controls, and source of control. The ethnicities of the populations were classified as Asian or Caucasian. We requested any missing data from the corresponding authors of the articles. Any disagreements were resolved through consensus among all involved authors.

The genotyping studies included in the systematic review and meta-analysis were carefully reviewed for APOE genotypes and alleles to ensure the quality of their genotyping methodology and data capture. Most studies used the ε3 allele, and the ε3/ε3 genotype was used as the reference to assess associations between the ε2 and ε4 alleles and their genotypes and the ε3 allele. While reviewing the remaining studies that did not explicitly mention the use of a reference genotype, it was evident that the ε3 allele was used in the experiments, as the results all involved variations in the ε2 and ε4 alleles with risk association, and the ε3 allele and its genotype showed protective association or no association. Therefore, the authors of these studies were contacted to ascertain whether the ε3 allele and the ε3/ε3 genotype were used as genotyping references for the remaining alleles. The respective authors confirmed that the ε3 allele was used as a reference standard for all the analyses. Hence, our meta-analysis assessed the association of risk with the ε2 and ε4 alleles, and with the ε3 allele and the ε3/ε3 genotype as the reference.

The methodological quality of the included studies was independently assessed by two reviewers (MN and AR) using the Newcastle-Ottawa Scale (NOS) (Kansagara et al., 2017), which evaluates selection, comparability, and outcome/exposure. Studies were categorized as high (≥7 points), moderate (4–6 points), or low ( ≤ 3 points) quality. Discrepancies were resolved through discussion with a third reviewer (PK).

Publication bias was assessed using Begg's funnel plot, while Egger's linear regression test evaluated plot asymmetry (Egger et al., 1997; Begg and Mazumdar, 1994). Sensitivity analyses were conducted to assess selection bias and heterogeneity by sequentially removing individual studies.

To assess the overall genetic risk association of the current APOE-based study with hemorrhagic stroke, both genotypic and allele-based models were used for comparative analyses against the reference allele ε3 and the genotype ε3/ε3, respectively. More importantly, the risk assessment in the genotypic model used a codominant comparison of APOE genotypes relative to the reference genotype. All the included studies were assessed for the Hardy-Weinberg equilibrium based on the availability of data from the control groups. The remaining studies were excluded from the final analysis.

A meta-analysis was performed to evaluate the association between APOE alleles/genotypes and HS risk, calculating pooled odds ratios (ORs) with 95% CIs. Heterogeneity was assessed using the I² statistic, with < 50% indicating low and >50% suggesting moderate to high heterogeneity (Higgins et al., 2003). A fixed-effect model (Mantel and Haenszel, 1959) was applied for I² < 50%, while a random-effect model (DerSimonian, 1996) was used for I² > 50%. Subgroup analysis by ethnicity and meta-regression explored sources of heterogeneity. Statistical analyses were conducted in STATA 13.0, with p < 0.05 considered statistically significant. However, the results must be interpreted cautiously due to multiple testing, which can lead to a Type I error.

The initial database search identified 1,849 articles, of which 1,764 were excluded after screening titles and abstracts. The full texts of 91 studies were assessed, with 67 excluded due to reviews, editorials, prior meta-analyses, non-hemorrhagic stroke studies, absence of control groups, or incomplete data. Finally, 24 studies met the eligibility criteria and were included in the systematic review and meta-analysis. The PRISMA flowchart (Figure 1) illustrates the study selection process.

The included studies, published between 1996 and 2024, examined multiple ethnicities across various geographic regions. Of the 24 studies, eight focused on Asian populations (Chowdhury et al., 2001; Ganaie et al., 2020; Das et al., 2016; Zhang et al., 2012; Kokubo et al., 2000; Misra et al., 2013; Nakata et al., 1997; Jiang et al., 2024), 19 on Caucasian (Catto et al., 1996; Duzenli et al., 2004; Martini et al., 2012; Seifert et al., 2006; Basun et al., 1996; Biffi et al., 2010; Garcia et al., 1999; Greenberg et al., 1996; McCarron and Nicoll, 1998; Tasdemir et al., 2008; Woo et al., 2013, 2005, 2002; Sawyer et al., 2018; Hostettler et al., 2022), three on African (Biffi et al., 2010; Sawyer et al., 2018; Atadzhanov et al., 2013), three on Hispanic (Biffi et al., 2010; Sawyer et al., 2018), and one on a mixed population (Sawyer et al., 2018). Among the Asian studies, three were conducted in India (Ganaie et al., 2020; Das et al., 2016; Misra et al., 2013), two in Japan and China each (Kokubo et al., 2000; Nakata et al., 1997), and one in Bangladesh (Chowdhury et al., 2001). Out of the 19 Caucasian studies, eight were performed in the American (USA) (Martini et al., 2012; Biffi et al., 2010; Greenberg et al., 1996; Woo et al., 2013, 2005, 2002; Sawyer et al., 2018) population, three in the British (Catto et al., 1996; McCarron and Nicoll, 1998; Hostettler et al., 2022) population, two in the Turkish (Duzenli et al., 2004; Tasdemir et al., 2008) population, two in the Swedish (Basun et al., 1996; Biffi et al., 2010) population, two in the Austrian (Seifert et al., 2006; Biffi et al., 2010) population, and one each in the Polish (Biffi et al., 2010) and Portuguese (Garcia et al., 1999) populations. Amongst the three African studies, two were conducted in the American (USA) (Biffi et al., 2010; Sawyer et al., 2018) population, and one was in the Zambian (Atadzhanov et al., 2013) population. Additionally, two of the three Hispanic studies were performed in the Spanish (Biffi et al., 2010) population, and one was in the American (USA) (Sawyer et al., 2018) population. Study quality varied, with six rated medium and 18 high. Seventeen were population-based, seven hospital-based, and most followed a case-control design, with two retrospective cohorts and one cross-sectional study. Table 1 provides an overview of baseline characteristics and quality assessments.

The APOE ε2 allele was assessed based on the location of the hemorrhagic stroke, encompassing 5,287 cases and 24,288 control subjects from 16 included studies. The locations were categorized into lobar, deep, mixed, non-lobar intracerebral hemorrhage (ICH), and rapidly involuting (RICH) or non-involuting (NICH) congenital hemangiomas. Additionally, those who could not be classified based on location were grouped under the hemorrhagic stroke category. It was found that the ε2 allele of the APOE gene had a significant association with the risk of hemorrhagic stroke for the overall location of all intracerebral hemorrhages (OR: 1.23, 95% CI = 1.12–1.35, I² = 0%). Moreover, there was a significant association of the ε2 allele with hemorrhagic stroke risk for lobar ICH (OR: 1.36, 95% CI = 1.11–1.68, I² = 0%). However, no significant association was observed for the deep lobar location of the hemorrhage concerning the ε2 allele and hemorrhagic stroke risk (OR: 1.04, 95% CI = 0.83–1.31, I² = 0%). The mixed, non-lobar ICH, RICH, and NICH studies were not categorized separately because they consisted of only one study. Those with undefined locations of hemorrhagic stroke exhibited a significant association between the risk of intracerebral hemorrhage and the ε2 allele of APOE (OR: 1.20, 95% CI = 1.03–1.41, I² = 0%). The overall results are illustrated in Supplementary Figure S2-5 and Table 2.

The association of the ε4 allele with HS risk across different hemorrhage locations was also analyzed in 17 studies with the same dataset. A significant association was found across all locations (OR: 1.31, 95% CI = 1.14–1.51, I² = 59.0%), with a stronger link for lobar ICH (OR: 1.49, 95% CI = 1.26–1.78, I² = 0%). However, no significant association was observed for deep ICH (OR: 1.13, 95% CI = 0.94–1.36, I² = 0%), undefined hemorrhagic stroke (OR: 1.26, 95% CI = 0.97–1.64, I² = 0%), or ICH with unspecified location (OR: 1.27, 95% CI = 0.80–2.02, I² = 78.8%). Subgroup analysis for mixed, non-lobar ICH, RICH, and NICH was not possible due to limited studies. Results are depicted in Supplementary Figure S2-7 and Table 2.

Begg's funnel plot and Egger's test were used to assess publication bias. The symmetrical funnel plot for ε2/ε2 and ε4/ε4 genotypes suggested no bias (Figures 3a, b). Egger's test confirmed this (p = 0.494 for ε2/ε2 and 0.760 for ε4/ε4). Secondary outcomes, including ε2/ε3, ε2/ε4, ε3/ε4 genotypes and ε2, ε4 alleles, also showed no publication bias (Supplementary Figures S2-8–12). Egger's p-values for all analyses are summarized in Table 2.

Meta-regression analysis was performed to assess the effect size based on study quality. No significant deviations were observed in the association between the ε2/ε2 and ε4/ε4 genotypes of the APOE gene and HS risk (Figures 4a, b). Similarly, analyses for ε2/ε3, ε2/ε4, ε3/ε4 genotypes and ε4 allele showed no notable deviations. However, the ε2 allele exhibited some deviation (p = 0.048), suggesting cautious interpretation of its association with HS risk. Table 2 summarizes the meta-regression p-values, with additional details in Supplementary Figures S2-13–17.

Sensitivity analysis using the Leave-one-out method showed no significant changes in effect size for the ε2/ε2 and ε4/ε4 genotypes of the APOE gene and HS risk (Figures 5a, b). Similarly, no notable variations were observed for ε2/ε3, ε2/ε4, ε3/ε4 genotypes or ε2 and ε4 alleles (Supplementary Figures S2-18–22), confirming the robustness of the findings.

Subgroup analysis confirmed significant associations between ε2/ε2, ε4/ε4, and ε2/ε4 genotypes and HS risk in Caucasian populations. In Asian populations, only the ε2/ε4 genotype showed a significant risk association. Further analysis by hemorrhage location revealed that ε2 and ε4 alleles were strongly linked to lobar hemorrhages, whereas no association was found with deep hemorrhages.

Our findings align with (Nie et al. 2019), which identified ε2 and ε4 alleles as risk factors for intracerebral hemorrhage (ICH) in Caucasians, but with a significantly larger sample (24 vs. eight studies). Similarly, our results corroborate (Marini et al. 2019), who found a strong link between ε2 and ε4 alleles and lobar ICH in Caucasians, with our larger dataset reinforcing their conclusions. Additionally, (Pan et al. 2021) demonstrated poor functional outcomes in ICH patients with the ε4 allele, further supporting our findings.

Beyond HS, the APOE gene has been extensively studied in neurodegenerative diseases, including Alzheimer's disease, where the ε4 allele is a well-established risk factor across ethnicities, including Indian populations (Farrer et al., 1997; Agarwal and Tripathi, 2014). APOE polymorphisms are also linked to lipid metabolism, cardiovascular diseases, and stroke risk, with associations found between APOE alleles and cholesterol levels, atherosclerosis, and oxidative injury protection (Ozen et al., 2022).

Conflicting evidence exists regarding the mechanisms behind APOE polymorphisms and HS risk. While ε2 and ε4 alleles are implicated in abnormal lipid metabolism, blood-brain barrier disruption, and vasogenic edema, some studies suggest APOE presence may protect against oxidative neuronal damage. Our study, the most extensive meta-analysis to date on APOE polymorphisms and HS, strengthens the link between ε2 and ε4 alleles and HS risk (Biffi et al., 2010; Zhang et al., 2014; Sudlow et al., 2006). These findings provide a robust foundation for future observational studies on APOE genotypes and their role in HS, offering potential clinical implications for risk stratification and targeted interventions. These findings can provide a guideline for conducting future observational studies on APOE genotypes to ascertain the relationship with the risk of HS, thereby facilitating clinical transition.

Despite its comprehensive approach, this study has some limitations. Key comorbidities like hypertension, diabetes, and prior stroke or cardiac disease were not accounted for, potentially influencing outcomes. False discovery rates were not assessed, which may affect comparisons. Some heterogeneity was noted, with meta-regression indicating bias; however, this was addressed through the random-effects model and sensitivity analyses. Importantly, the study demonstrated minimal heterogeneity and no publication bias, reinforcing the reliability of the pooled effect sizes. The lack of risk factor, confounder, and covariate data in the included studies limits the generalizability of the study outcomes and, hence, the results must be interpreted with caution.