Highlight
– The Multicenter Arteriovenous Malformation Research Study (MARS) reports an overall first intracranial hemorrhage (ICH) rate of 1.40 per 100 person-years (95% CI 1.20–1.64) among 2989 participants with unruptured brain AVMs.
– Independent predictors of first ICH: increasing age (notably age ≥60), presence of associated arterial aneurysms (HR 1.66), and cerebellar or supratentorial deep AVM location (HR 1.87).
– These data suggest a lower annual hemorrhage risk than the commonly cited 2–4% and should inform individualized management and counselling.
Background
Brain arteriovenous malformations (AVMs) are uncommon but clinically important vascular lesions that create direct shunts between cerebral arteries and veins without a normal capillary bed. The natural history of unruptured AVMs—particularly the risk of first intracranial hemorrhage (ICH)—is the central driver of decisions about conservative management versus microsurgical resection, stereotactic radiosurgery or endovascular treatment. Historically, widely used estimates for annual hemorrhage risk from unruptured AVMs have ranged from about 2% to 4% per year. Those estimates shaped clinical practice patterns and informed discussions about the risks and benefits of treatment.
However, existing estimates derive from heterogeneous sources, including small series, referral bias, or mixed cohorts of ruptured and unruptured lesions, producing imprecision. The MARS study (Multicenter Arteriovenous Malformation Research Study) was designed to provide more precise, contemporary estimates and to identify independent predictors of first ICH in a large international cohort of unruptured AVMs.
Study design
MARS pooled data from nine cohorts (two population-based and seven referral-based) with each cohort contributing at least 100 participants diagnosed with an unruptured brain AVM. The study period covered 2017–2023 and combined retrospective and prospective data collection. After exclusions, 3030 participants were included, of whom 2989 had unruptured AVMs at baseline and form the primary analytic cohort.
Key data elements included demographics, clinical presentation (seizures, focal deficits), angiographic features (nidus size, venous drainage pattern, associated arterial aneurysms, lesion location), and dates of diagnosis, any ICH events, treatment, death, and censoring visits. The primary outcome was time to first ICH after diagnosis of an unruptured AVM. The analytic approach used Cox proportional hazards models with baseline hazards allowed to vary by cohort (stratified Cox), multiple imputation for missing covariate data, and censoring at first AVM treatment, death, or last follow-up visit. The study accrued 11,339 person-years of observation for the unruptured cohort.
Key findings
Overall hemorrhage incidence
MARS observed 159 first ICH events among the 2989 participants with unruptured AVMs, yielding an incidence rate of 1.40 per 100 person-years (95% CI, 1.20–1.64). This point estimate is substantially lower than the commonly cited 2–4% annual rates used in many clinical discussions and reflects improved precision owing to the large sample and person-time accumulated.
Absolute risk approximation
Assuming a constant hazard (an approximation), a 1.40% annual rate translates to an approximate cumulative risk of 6.8% at 5 years and 13% at 10 years. These are approximate and assume stable hazard over time; individual absolute risk will vary by age, lesion features, and competing risks.
Independent risk factors for first ICH
Multivariable Cox regression identified three significant, independent predictors of first ICH:
– Increasing age at diagnosis: Relative to children (<20 years), older age categories showed higher hazards, with the strongest contrast for patients aged 60 years versus those younger than 20 (HR, 2.01; 95% CI, 1.14–3.57; P = .008). The middle age categories showed smaller and statistically uncertain differences.
– Presence of associated arterial aneurysms: AVMs with associated arterial aneurysms had higher hemorrhage risk (HR, 1.66; 95% CI, 1.06–2.59; P = .03). This finding underscores the concept that intranidal or feeding arterial aneurysms may contribute to rupture risk.
– Lesion location: Cerebellar or supratentorial deep AVM locations were associated with higher risk than cortical/other supratentorial sites (HR, 1.87; 95% CI, 1.16–3.00; P = .01).
Non-significant or lesser effects
Commonly discussed angiographic variables such as maximal nidus diameter, the presence of exclusive deep venous drainage (observed in 10% of the cohort), and seizures at presentation (45% presented with seizures) did not emerge as dominant independent predictors in the final multivariable models reported. Detailed effect sizes and confidence intervals for these variables were reported in the full manuscript and depend on modeling choices and imputed datasets.
Subgroup and sensitivity considerations
MARS included a mix of population-based and referral cohorts; the analytic strategy permitted baseline hazard differences across cohorts to mitigate heterogeneity. The investigators censored at treatment to estimate the natural history among untreated person-time, which is reasonable but can bias estimates if treatment is preferentially given to higher-risk lesions (informative censoring). Sensitivity analyses addressing such biases were discussed in the original report; clinicians should interpret estimates with this context in mind.
Expert commentary and context
How does MARS compare with earlier evidence?
The ARUBA randomized trial (A Randomized Trial of Unruptured Brain Arteriovenous Malformations, published 2014) claimed a benefit of conservative over interventional management for unruptured AVMs in its relatively short follow-up, but it was criticized for heterogeneity of interventions, limited follow-up duration, and issues with generalizability. MARS does not replace randomized evidence but supplies a modern, large-scale natural-history estimate derived from diverse clinical settings and provides greater granularity on risk factors.
Clinical implications
1) Counseling and shared decision-making: MARS provides clinicians with more precise baseline hemorrhage estimates and identifies higher-risk subgroups (older patients, associated arterial aneurysms, deep/cerebellar location). For many patients, especially younger individuals with low-risk location and no aneurysm, the annual hemorrhage risk may be lower than traditionally presented. This could alter the perceived balance between intervention-associated morbidity and the natural-history risk.
2) Stratified risk assessment: The identified predictors can help tailor counseling—patients with cerebellar or deep supratentorial AVMs or associated aneurysms may warrant stronger consideration of definitive treatment or closer surveillance.
3) Treatment selection and timing: The presence of modifiable lesion-specific risk markers (for example, treatable feeding artery aneurysms) might influence the choice and urgency of endovascular or microsurgical approaches.
Methodological strengths and limitations
Strengths include the large sample size, international multisite pooling, standardized definitions, and analytic methods that accounted for cohort-level baseline hazard differences and missing data. Important limitations include potential referral bias (7 of 9 cohorts were referral-based), informative censoring at time of treatment (riskier lesions may be treated earlier and thus removed from natural-history observation), heterogeneity in imaging and follow-up schedules, and residual confounding despite multivariable adjustment.
Mechanistic plausibility
The association between associated arterial aneurysms and increased hemorrhage risk is biologically plausible: aneurysms on feeding arteries or intranidal aneurysms are structural weak points prone to rupture. Deep and cerebellar locations have distinct venous drainage and anatomic constraints that may increase hemorrhage propensity or worsen clinical consequences when bleeding occurs. The age effect may reflect cumulative vascular vulnerability or comorbid changes in vascular compliance with aging.
Implications for practice and research
For clinicians: Use MARS estimates to refine risk communication. Replace rigid, generic 2–4% annual risk statements with individualized discussions incorporating age, presence of feeding/intranidal aneurysms, and lesion location. When counseling younger patients with low-risk features, emphasize that the absolute near-term hemorrhage risk is likely lower than older figures suggest, and weigh this against procedure-specific morbidity and patient preferences.
For researchers: Key next steps include developing and validating predictive models that combine clinical and imaging features to produce individualized absolute-risk estimates and decision aids. Addressing informative censoring by using methods that account for treatment assignment (e.g., inverse-probability weighting) and generating longer-term prospective follow-up—particularly in population-based cohorts—will further clarify the long-term natural history. Randomized studies with longer follow-up and well-defined interventions remain important for comparing the net benefit of treatment vs conservative management in specified risk strata.
Limitations of the MARS analysis (practical considerations)
– Referral bias: Overrepresentation of more complex or symptomatic cases in referral cohorts can distort incidence estimates; although the study stratified by cohort baseline hazards, generalizability to all newly diagnosed unruptured AVMs may be imperfect.
– Informative censoring at treatment: Patients judged at higher risk were more likely to receive treatment and thus were censored, potentially lowering the observed hemorrhage rate among the remaining untreated person-time.
– Imaging and adjudication variability: Multicenter data inevitably include variability in imaging protocols and event adjudication despite standardized definitions.
– Residual confounding: As with any observational analysis, unmeasured confounders may persist.
These limitations do not negate the value of MARS but should frame interpretation and translation into practice.
Conclusion
MARS provides a large, contemporary, and more precise estimate of the natural-history risk of first intracranial hemorrhage from unruptured brain AVMs: approximately 1.40% per year. Age (particularly ≥60 years), associated arterial aneurysms, and cerebellar or supratentorial deep location independently increased risk. These findings support more nuanced, individualized counseling and shared decision-making. They also identify key subgroups for whom interventional strategies may be prioritized and indicate directions for prospective risk modeling and trials to refine management strategies further.
Funding and clinicaltrials.gov
Funding sources and trial registration details are reported in the original publication: Kim H, Nelson J, McCulloch CE, et al. Risk of Future Hemorrhage From Unruptured Brain Arteriovenous Malformations: The Multicenter Arteriovenous Malformation Research Study (MARS). JAMA Neurol. 2025. Readers should consult the JAMA Neurology article for specific grant numbers, institutional support, and any clinicaltrials.gov identifiers related to component cohorts.
References
1. Kim H, Nelson J, McCulloch CE, et al. Risk of Future Hemorrhage From Unruptured Brain Arteriovenous Malformations: The Multicenter Arteriovenous Malformation Research Study (MARS). JAMA Neurol. 2025 Dec 1;82(12):1274-1281. doi: 10.1001/jamaneurol.2025.3581.
2. ARUBA Investigators (A Randomized Trial of Unruptured Brain Arteriovenous Malformations). A randomized trial comparing medical management with interventional therapy for unruptured brain AVMs. Lancet. 2014. (ARUBA trial; readers should consult the Lancet ARUBA publication for trial methods and outcomes.)
3. Stapf C, Mast H, Sciacca RR, et al. Predictors of hemorrhage in patients with untreated brain arteriovenous malformation. Neurology. 2006;66(2):169-171. (Representative older cohort study discussing risk factors.)
4. Al-Shahi Salman R, et al. Natural history of intracranial arteriovenous malformations: a systematic review. (Readers should consult major reviews for historical estimates and methodological critiques.)
Author note
This article is a critical synthesis and interpretation of the MARS study findings, intended for clinicians and policy-makers involved in managing patients with brain AVMs. For primary data, detailed statistical methods, and cohort-specific results, consult the full JAMA Neurology MARS publication.
