Elsevier

World Neurosurgery

Volume 96, December 2016, Pages 562-569.e1
World Neurosurgery

Original Article
Outcomes and Complications After Endovascular Treatment of Brain Arteriovenous Malformations: A Prognostication Attempt Using Artificial Intelligence

https://doi.org/10.1016/j.wneu.2016.09.086Get rights and content

Purpose

To identify factors influencing outcome in brain arteriovenous malformations (BAVM) treated with endovascular embolization. We also assessed the feasibility of using machine learning techniques to prognosticate and predict outcome and compared this to conventional statistical analyses.

Methods

A retrospective study of patients undergoing endovascular treatment of BAVM during a 22-year period in a national neuroscience center was performed. Clinical presentation, imaging, procedural details, complications, and outcome were recorded. The data was analyzed with artificial intelligence techniques to identify predictors of outcome and assess accuracy in predicting clinical outcome at final follow-up.

Results

One-hundred ninety-nine patients underwent treatment for BAVM with a mean follow-up duration of 63 months. The commonest clinical presentation was intracranial hemorrhage (56%). During the follow-up period, there were 51 further hemorrhagic events, comprising spontaneous hemorrhage (n = 27) and procedural related hemorrhage (n = 24). All spontaneous events occurred in previously embolized BAVMs remote from the procedure. Complications included ischemic stroke in 10%, symptomatic hemorrhage in 9.8%, and mortality rate of 4.7%. Standard regression analysis model had an accuracy of 43% in predicting final outcome (mortality), with the type of treatment complication identified as the most important predictor. The machine learning model showed superior accuracy of 97.5% in predicting outcome and identified the presence or absence of nidal fistulae as the most important factor.

Conclusions

BAVMs can be treated successfully by endovascular techniques or combined with surgery and radiosurgery with an acceptable risk profile. Machine learning techniques can predict final outcome with greater accuracy and may help individualize treatment based on key predicting factors.

Introduction

Brain arteriovenous malformations (BAVMs) have an incompletely understood natural history. With contemporary advances in endovascular interventions and radiation treatment, the bias in favor of treatment has hampered the emergence of large observational studies on the long-term outcome of unruptured BAVMs. Nevertheless, patients with untreated BAVMs do harbor a lifelong risk of intracranial hemorrhage (ICH), with an estimated risk of 2%–4% annually.1, 2, 3

There is strong evidence that patients with ruptured BAVMs have more favorable outcomes than ICH from other causes. The Scottish Intracranial Vascular Malformation Study found significantly better functional outcomes in the BAVM group compared with other causes of ICH, with the risk of death or dependence in spontaneous ICH >7 times greater than BAVM-related ICH at 1 year and 15 times greater at 2 years, even when adjusted for age.4 Although there have been criticisms regarding the methodology of these studies,5 similar favorable outcomes also have been reported by others.6

Some studies predict that most BAVMs will rupture at least once over the patient's lifetime1, 7; however, true lifetime mortality figures are unavailable for reasons discussed previously, and our limited understanding of the natural history of these lesions often lends itself towards treatment despite a paucity of clear evidence to support this over a conservative “watch-and-wait” approach. A Randomized Trial of Unruptured Brain Arteriovenous Malformations (ARUBA) sought to address whether treatment of unruptured lesions was superior to conservative management; however, patient enrolment was halted because of an unacceptably high number of adverse incidents in the intervention arm.8 Given this unexpected result, a similar trial is unlikely to be pursued in the near future.9, 10, 11

Several retrospective studies have demonstrated numerous angiographic predictors of hemorrhage in untreated BAVMs, including intranidal fistulae12 and flow-related aneurysms,13 deep venous drainage,12 small nidus size,14 high feeding mean arterial pressure,14 deep location,15 venous stenosis,16 a single draining vein,17 and slow filling of feeding arteries.18 Systemic hypertension also has been implicated as a risk for hemorrhage.15

Predictors of hemorrhagic risk in partially treated BAVMs, however, are less well defined. Paradoxically, some studies suggest that hemorrhagic risk increases with decreasing diameter, postulated to be the result of greater feeding arterial pressure and fewer draining veins in small lesions.3, 19, 20 In addition, there is strong evidence that patients presenting with hemorrhage have a greater risk of rebleeding21, 22 and that this risk is considered to be greatest in the first year, when rebleeding rates area as high as 33%.23 Therefore, strategies for treatment of ruptured BAVM typically are more aggressive, with a treatment goal of complete obliteration.24, 25

Taking these factors into consideration, the management of BAVMs can be perplexing and depends on multiple clinical and imaging variables. A number of classification systems offer guidance for the management of BAVM, with the most established grading system devised by Spetzler and Martin.26 This system, however, originally was proposed to classify BAVMs in terms of potential neurosurgical challenges and is not necessarily a predictor of outcome in untreated BAVMs.27 Therefore, additional clinical and imaging variables, such as mode of presentation or angioarchitectural features, are not included in this system.26, 27 Subsequently, variations of the Spetzler-Martin classification, such as the Lawton supplementary grading system, were introduced to improve preoperative risk prediction and patient selection28 by including additional factors such as age, presentation, and nidal compactness.27

Because the natural history of BAVM often is protracted and treatment methods continuously evolve over patients' lifetimes, hemorrhagic risk categorization and prognostication is a very dynamic process. Therefore, this study aims to identify factors influencing final outcome in patients undergoing endovascular treatment of BAVMs, either alone or in combination with other modalities.

The usual approach in this regard is to develop logistic regression models; however, machine learning algorithms have been proposed as an alternative approach, with the advantage of easily incorporating new information to improve prediction performance.29, 30, 31 Because of the complexity of determining the final outcome of lesions treated endovascularly, conventional statistical models are cumbersome or insufficient for this purpose.

There has been a recent emergence in the use of artificial intelligence techniques in prognostication of similar neurovascular conditions such as acute stroke31; therefore, we aimed to review whether machine learning techniques are capable of predicting outcome for endovascular treatment of BAVM or to assist prognostication. Please refer to the supplementary digital content for a brief introduction into machine learning.

Section snippets

Methods

We performed a retrospective study on a prospectively collected database. Our institution has waived the requirement for informed consent for these retrospective studies.

Results

During the study period, 199 patients with BAVM underwent 659 embolization procedures (mean of 3.1 procedures per patient), combined with 76 radiosurgical interventions and 61 open surgical resections, with full clinical and imaging details available for analysis. This cohort consisted of 98 male and 101 female patients, with a mean age of 35.5 (SD ± 16.7) years. Mean duration of follow-up was 63 months.

The most common presentation was ICH (56%), followed by seizure (19%), headache (14%), and

Discussion

Patients with untreated BAVM have an annual lifelong risk of ICH of less than 5%,1, 2, 3 with evidence that patients presenting with rupture having more favorable outcomes than ICH from other causes.4 It is also predicted that most BAVMs will rupture at least once during the lifetime of the patient.1, 7 Treatment strategies are complicated, considering that ARUBA was halted because of an unacceptably high number of adverse incidents during intervention in unruptured AVMs,8, 32 disappointing

Conclusions

We have demonstrated an acceptable prediction accuracy of outcome after endovascular treatment of BAVM using machine learning techniques. There is further potential to improve prediction accuracy by the incorporation of larger multicenter datasets. Such models may be of use not only for prognostication and in predicting the outcome under different circumstances but can also assist in clinical decision making, in particular identifying patients who may benefit from further treatment, including

References (39)

  • J.D. Nerva et al.

    Treatment Outcomes of Unruptured Arteriovenous Malformations With a Subgroup Analysis of ARUBA (A Randomized Trial of Unruptured Brain Arteriovenous Malformations)-Eligible Patients

    Neurosurgery

    (2015)
  • J.P. Mohr et al.

    Viewpoints on the ARUBA Trial

    AJNR Am J Neuroradiol

    (2015)
  • M.A. Weber

    Cerebral arteriovenous malformations: Results of the ARUBA study

    Radiologe

    (2014)
  • M.P. Marks et al.

    Hemorrhage in intracerebral arteriovenous malformations: angiographic determinants

    Radiology

    (1990)
  • R. Willinsky et al.

    Brain arteriovenous malformations: analysis of the angio-architecture in relationship to hemorrhage (based on 152 patients explored and/or treated at the hopital de Bicetre between 1981 and 1986)

    J Neuroradiol

    (1988)
  • A. Kader et al.

    The influence of hemodynamic and anatomic factors on hemorrhage from cerebral arteriovenous malformations

    Neurosurgery

    (1994)
  • F. Turjman et al.

    Correlation of the angioarchitectural features of cerebral arteriovenous malformations with clinical presentation of hemorrhage

    Neurosurgery

    (1995)
  • F. Nataf et al.

    Angioarchitecture associated with haemorrhage in cerebral arteriovenous malformations: a prognostic statistical model

    Neuroradiology

    (1997)
  • X. Shi et al.

    An analysis of the correlation between angiographic and clinical findings in cerebral arteriovenous malformations

    Chin Med Sci J

    (1993)
  • Cited by (48)

    • Complications in the Endovascular Treatment of Intracranial Arteriovenous Malformations

      2021, Complications in Endovascular Surgery: Peri-Procedural Prevention and Treatment
    • Machine Learning for Brain Stroke: A Review

      2020, Journal of Stroke and Cerebrovascular Diseases
      Citation Excerpt :

      For stroke prevention, [21] mentioned 100% accuracy for classification of ischemic stroke sub types. If we see the results in stroke treatment category, [6] got optimal result, accuracy 91.43% and in prognostication, [3] received 97.5% accuracy. Figure 2 signifies 3 layers, consists of stroke category, dataset and outer layer represents optimal model of particular study.

    • Artificial intelligence in neuro, head, and neck surgery

      2020, Artificial Intelligence in Precision Health: From Concept to Applications
    View all citing articles on Scopus

    Supplementary digital content available online.

    Conflict of interest statement: The authors declare that the article content was composed in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

    View full text