Risk factors of cerebral vasospasm associated with ruptured arterio-venous malformation: A systematic review and meta-analysis

Research question

PROSPERO (CRD420251037660) records this research. This study used the PECO framework [33].

- Populations: adults (≥18 years) diagnosed with ruptured AVM

- Exposure:

• Demographic conditions, including younger age, female gender, and moderate or severe Glasgow Coma Scale (GCS) on admission (<8)

• Comorbid conditions including hypertension, obesity, diabetes mellitus, chronic renal disease, hyperlipidemia

• Substance use, including smoking, stimulants, and alcohol

• Lab and vital parameters: leukocytosis, anemia, hyperglycemia, hyponatremia, hypokalemia, fever, increased blood pressure, hypovolemia, hypotension, electrocardiogram (ECG) abnormalities.

• Poor prognosis for hemorrhage severity, including Hunt and Hess grading (III, IV, V), Fisher/mFisher rating (mFisher 3-4), and infratentorial placement.

- Comparator: none

- Outcomes: the incidence of cerebral vasospasm

Design and search strategy

This systematic review and meta-analysis follows PRISMA. Spontaneous burst AVM risk factors for cerebral vasospasm were examined from conception until April 2025. PubMed, Science Direct, Cochrane Library, and SSRN were searched systematically. AVM, arteriovenous malformation, risk factor, vasospasm, and cerebral vasospasm were keywords.

Inclusion and exclusion criteria

Studies assessing vasospasm risk factors were considered. We only included patients aged >18 with spontaneous burst AVM. We exclude studies in case series or case report form, editorial, commentary, review, and those with no data. Only studies available in English were further analyzed.

Data extraction and statistical analysis

Rayyan.ai originally excluded publications that did not match qualifying requirements from titles and abstracts. The screening was done separately by LADS, IGRA, and KKDC. Discussing disagreements led to a consensus. Each research’s author, year, sample size, study design, patient demographics (age, sex), and cerebral vasospasm diagnostic technique were retrieved separately using a data extraction sheet.

Risk-of-bias assessment

For observational research, the ROBINS-E tool evaluated study quality [32]. Seven biases were assessed: confounding, participant selection, exposure categorization, deviations from planned exposures, missing data, outcome measurement, and result selection. Studies were categorized as low, some concerns, high, or extremely high for bias. The risk-of-bias evaluation was done separately by LADS, IGRA, and KKDC. If assessors disagreed, the majority assessment was given.

Data synthesis and statistical methods

Meta-analysis was done with Review Manager. To account for study variability, a REM was used. RRs and 95% CIs were determined for each treatment comparison. Heterogeneity was evaluated using the I² statistic, thresholds of 0-30%, 31-50%, and >50% indicating insignificant, moderate, and considerable heterogeneity. Network consistency was assessed using the Q-statistic and the design-by-treatment interaction model.

Study selection and characteristics

A total of 972 records were identified, with 21 duplicates removed, leaving 951 for screening. Of these, 493 were excluded, and 458 reports were sought for retrieval. 443 were not retrieved, resulting in 15 reports assessed for eligibility. After full-text review, 9 were excluded, leaving 14 studies included. Additionally, 396 records were identified through citation searching, 22 were assessed for eligibility after 374 were not retrieved, with 14 subsequently excluded. The PRISMA flowchart is shown in Fig. 1.

Fig. 1.

PRISMA chartflow of the study included selection.

All included were observational, including prospective cohort, retrospective cohort, retrospective case control, and cross-sectional. Out of 16,304 individuals, 7,095 were in the vasospasm group and 9,209 were not. Two multi-center studies were undertaken, although most were single-center. Baseline population and clinical characteristics are presented in Tables 1 and 2.

Table 1.

Baseline population characteristics

Table 2.

Clinical characteristics

Risk of bias assessment

The traffic light graphic (Fig. 2) shows how the ROBINS-E tool evaluated bias in the observational studies. Each domain has low bias (green), moderate concerns (yellow), or substantial bias (red).

Fig. 2.

Risk of bias assessment

In general, most studies included in the review demonstrated some concerns (yellow) across several domains. Specifically, all studies except one showed some concerns in domains related to confounding (D1), postinterventions (D4), missing (D5), and selection reported results (D7). Notably, Chhor et al., 2011 had a high risk in both the confounding domain (D1) and the selection of participants domain (D3), suggesting methodological shortcomings that may impact their results. In contrast, domains D2 (exposure measurement) and D6 (outcome measurement) has little bias (green) in most trials, indicating good exposure and outcome assessment.

Overall, while the majority of studies presented certain methodological concerns, no study (other than Chhor et al., 2011) has a high overall risk of bias. These findings suggest that although the evidence base includes studies with some limitations, most studies provide reasonably reliable estimates, albeit with some caution due to potential residual confounding and reporting biases.

Outcomes

After data extraction was carried out, 2 of the 24 exposures were only included in one study, making it impossible to draw conclusions using the meta-analysis method, including younger age and obesity comorbid conditions. In addition (22 of 24 exposures), conclusions were drawn using an analysis method with visualization using a forest plot as follows (Figs. 3-17). Given the limited number and heterogeneity of included studies, these analyses are considered exploratory. Definitions of vasospasm varied across studies, including angiographic, transcranial Doppler–based, and clinical criteria. Significance was obtained for 12 exposures, including chronic renal disease, hyperlipidemia, smoking, stimulant use, leukocytosis, hyperglycemia, hyponatremia, hypokalemia, fever, hypovolemia, acute hypotension, and ECG abnormalities.

Fig. 3.

Forest plot of chronic renal disease.

Fig. 4.

Forest plot of hyperlipidemia.

Fig. 5.

Forest plot of smoking.

Fig. 6.

Forest plot of stimulant use as risk factors.

Fig. 7.

Forest plot of leukocytosis.

Fig. 8.

Forest plot of hyperglycemia.

Fig. 9.

Forest plot of hyponatremia.

Fig. 10.

Forest plot of hypokalemia as a risk fator.

Fig. 11.

Forest plot of fever.

Fig. 12.

Forest plot of hypovolemia.

Fig. 13.

Forest plot of acute hypotension.

Fig. 14.

Forest plot of ECG abnormality.

Fig. 15.

Forest plot of Hunt and Hess grading III, IV, or V.

Fig. 16.

Forest plot of mFisher 3-4 grading as risk factors.

Fig. 17.

Forest of infratentorial bleeding location as risk factors.

Chronic renal disease (RR= 0.5 [95% CI 0.46-0.67] p-v <0.0001) and hyperlipidemia (RR=0.78 [95% CI 0.73-0.84] p-v <0.00001) as comorbid conditions were both found as protective factors. Then the use of substances, including smoking and stimulants, was found to tend to increase the incidence of CVS (RR=1.31 [95% CI 1.11-1.54] p-v=0.001), RR=1.31 [95% CI 1.11-1.51] p-v=0.0002) respectively. In the lab and vital parameters section, 8 of them showed results as risk factors for CVS, including leukocytosis (RR=1.58 [95% CI 1.35-1.84] p-v<0.0001), hyperglycemia (RR=1.17 [95% CI 1.07-1.29] p-v=0.0009), hyponatremia (RR=1.45 [95% CI 1.23-1.71] p-v<0.0001), hypokalemia (RR=1.28 [95% CI 1.20-1.36] p-v<0.00001), fever (RR=1.38 [95% CI 1.23-1.54] p-v<0.00001), hypovolemia (RR=1.39 [95% CI 1.15-1.67] p-v=0.006), hypotension (RR=1.4 [95% CI 1.26-1.55] p-v<0.00001), ECG abnormality (RR=1.74 [95% CI 1.34-2.26] p-v<0.0001).

Protective factors

Interestingly, chronic renal disease (CRD) and hyperlipidemia were associated with a lower incidence of CVS. This finding may be related to intrinsic vascular alterations. In CRD, chronic endothelial dysfunction may lead to reduced vascular responsiveness to acute vasoconstrictive stimuli [2,37], while uremia-related immunomodulation may attenuate inflammatory pathways involved in vasospasm [5,19,38]. Similarly, hyperlipidemia may be associated with structural vascular changes, such as increased arterial stiffness and adaptive endothelial responses, potentially reducing susceptibility to acute vasoconstriction [27]. Although concomitant use of vasoactive medications and statins may contribute to this association, medication use was not consistently reported across studies [7,14,23,34]. Therefore, any contribution of these therapies remains uncertain, and the observed associations should be interpreted with caution.

Substances use

The use of substances, specifically smoking and stimulants, was found to significantly increase the incidence of CVS. These agents activate the sympathetic nervous system and directly induce cerebral vasoconstriction [1,31]. Smoking introduces nicotine and carbon monoxide, which impair endothelial function and promote catecholamine-mediated vasospasm [4,21,39]. Similarly, stimulant drugs (e.g., cocaine, amphetamines) trigger acute vasospastic crises via intense adrenergic surges and direct smooth muscle stimulation [6,8,10,31].

Laboratory and vital parameters associated factors

Several laboratory and clinical indicators were linked to cerebral vasospasm. Leukocytosis may indicate a high systemic inflammatory state, which causes vasospasm via cytokine-mediated endothelial dysfunction and oxidative stress [15,20]. Similarly, hyperglycemia may worsen vascular injury by promoting endothelial cell apoptosis and impairing nitric oxide production, thereby enhancing vasoreactivity [28,41].

CVS risk increases with electrolyte abnormalities. Syndrome of inappropriate antidiuretic hormone (SIADH) frequently causes hyponatremia and hypokalemia in neurocritical care. These disturbances can lead to hypovolemia and neuronal excitability, both of which increase cerebrovascular tone and predispose to vasospasm [3,24,29].

Fever, whether infectious or neurogenic in origin, may further exacerbate the inflammatory cascade and compromise cerebral autoregulation, thereby increasing vasospastic risk [22]. Moreover, hypovolemia and acute hypotension were shown to significantly impair cerebral perfusion pressure, potentially triggering compensatory vasoconstriction in vulnerable arterial segments [25,26,30,35].

Lastly, electrocardiographic abnormalities, which may reflect autonomic dysregulation or neurocardiogenic injury following subarachnoid hemorrhage, serve as a clinical indicator of increased sympathetic drive another key contributor to cerebral vasospasm [13,14,16,17,40]. These results emphasize the need for monitoring metabolic and hemodynamic stability in ruptured AVM patients, as these parameters are not merely epiphenomena but may actively contribute to vasospastic pathophysiology.

In contrast to aneurysmal subarachnoid hemorrhage, vasospasm following AVM rupture may arise through different mechanisms, particularly in the absence of diffuse subarachnoid hemorrhage. In such cases, vasospasm may be driven by localized effects of intracerebral or intraventricular hemorrhage, including the accumulation of blood breakdown products, focal inflammatory responses, and direct microvascular irritation. These mechanisms may result in more focal, angiographic, or subclinical vasospasm rather than the diffuse patterns typically observed in aneurysmal subarachnoid hemorrhage [9,12].

Most included studies reported unadjusted associations without controlling for important confounders such as hemorrhage severity, AVM characteristics, and treatment factors, limiting causal inference. Therefore, the observed relationships should be interpreted as associations rather than causal effects. Some variables, such as leukocytosis, electrolyte imbalance, and hypotension, may reflect disease severity or downstream responses rather than independent risk factors.

In addition, angiographic and hemorrhage patterns, such as subarachnoid, intracerebral, and intraventricular hemorrhage, were not consistently analyzed across the included studies. Diffuse subarachnoid hemorrhage is a well-established risk factor for vasospasm and may act as an important confounder. The lack of stratification by hemorrhage pattern may therefore influence the observed associations and should be considered when interpreting the results.

Strengths and limitations

This study used a comprehensive search strategy with independent data extraction and risk of bias assessment, enhancing reliability. However, several limitations should be noted. The small number of included studies and their heterogeneity in design, populations, and diagnostic criteria may limit the strength of the analysis. Several exposures were derived from only one or two studies, reducing the reliability of pooled estimates. Residual confounding and selection bias are likely due to the observational nature of the included studies and the lack of adjustment for key variables, including medication use. Most analyses were based on unadjusted data, limiting causal inference, and some variables may reflect disease severity rather than independent risk factors. In addition, heterogeneous definitions of vasospasm and inconsistent reporting of hemorrhage patterns may introduce misclassification and confounding. The use of multiple forest plots may also overrepresent the apparent precision of the findings; therefore, results should be considered exploratory and interpreted with caution.