Feeding Artery Pressure and Venous Drainage Pattern Are Primary Determinants of Hemorrhage From Cerebral Arteriovenous Malformations
Purpose—The purpose of this study was to define the influence of feeding mean arterial pressure (FMAP) in conjunction with other morphological or clinical risk factors in determining the probability of hemorrhagic presentation in patients with cerebral arteriovenous malformations (AVMs).
Methods—Clinical and angiographic data from 340 patients with cerebral AVMs from a prospective database were reviewed. Patients were identified in whom FMAP was measured during superselective angiography. Additional variables analyzed included AVM size, location, nidus border, presence of aneurysms, and arterial supply and venous drainage patterns. The presence of arterial aneurysms was also correlated with site of bleeding on imaging studies.
Results—By univariate analysis, exclusively deep venous drainage, periventricular venous drainage, posterior fossa location, and FMAP predicted hemorrhagic presentation. When we used stepwise multiple logistic regression analysis in the cohort that had FMAP measurements (n=129), only exclusively deep venous drainage (odds ratio [OR], 3.7; 95% confidence interval [CI], 1.4 to 9.8) and FMAP (OR, 1.4 per 10 mm Hg increase; 95% CI, 1.1 to 1.8) were independent predictors (P<0.01) of hemorrhagic presentation; size, location, and the presence of aneurysms were not independent predictors. There was also no association (P=0.23) between the presence of arterial aneurysms and subarachnoid hemorrhage.
Conclusions—High arterial input pressure (FMAP) and venous outflow restriction (exclusively deep venous drainage) were the most powerful risk predictors for hemorrhagic AVM presentation. Our findings suggest that high intranidal pressure is more important than factors such as size, location, and the presence of arterial aneurysms in the pathophysiology of AVM hemorrhage.
- cerebral arteriovenous malformations
- cerebral circulation
- cerebral hemorrhage
- cerebrovascular disorders
Arteriovenous malformations represent a small proportion of the total incidence of stroke but typically affect otherwise healthy young adults. The main reason to treat cerebral AVMs is the prophylaxis against spontaneous intracranial hemorrhage. Hemorrhage is the initial manifestation of the disease in 50% to 60% of patients,1 2 and each bleeding episode is estimated to be associated with a 10% mortality and a 30% to 50% morbidity. Future hemorrhage in the natural course of the disease appears to be related to initial presentation with hemorrhage.3 4 5 6 Therefore, identification of a set of morphological and physiological risk factors that can predict a hemorrhagic clinical presentation has been the subject of several previous reports.3 7 8 9 10 11 12 13 14 15 16 17 However, because not all authors have reported that initial hemorrhage predisposes to subsequent hemorrhage,18 19 the precise nature of AVM natural history remains controversial.6
In 1994, our group demonstrated that a small nidus size and an exclusively deep venous drainage were important risk factors associated with an initial clinical presentation with an AVM hemorrhage.20 In that series, a small number of cases (n=52) also underwent cerebral arterial pressure measurement. The input pressure to the nidus, as measured in the feeding arteries, appeared to be a powerful predictive risk factor as well, suggesting that factors that might be associated with high intranidal pressure, ie, higher transmural pressure, might predispose to spontaneous hemorrhage. Since then we have expanded our database. Our present objective was to focus on factors hypothesized to be associated with AVM hemorrhage that reflected intranidal transmural pressure gradients, especially cerebral arterial pressure and detailed measures of AVM angioarchitecture obtained during superselective angiography, which may reflect cerebral venous pressure.
The angiographic and clinical data of prospectively enrolled consecutive patients with cerebral AVMs from the Columbia-Presbyterian Medical Center AVM study project were reviewed retrospectively in a blinded fashion by two neuroradiologists (D.H.D., M.C.V.) with regard to the variables listed in Table 1⇓. The study sample was divided into two groups on the basis of initial clinical presentation of the disease: hemorrhagic presentation or nonhemorrhagic presentation. Hemorrhage was defined as a clinically symptomatic event (sudden-onset headache, and/or seizure, and/or focal deficit) with signs of fresh bleeding on CT or MRI of the brain by report or direct examination of the imaging studies. The assignment of hemorrhage group was made by one member of the stroke neurology team (H.M., H.-C.K., A.H., J.P.M.) at the time of entry into the study database after review of clinical and radiographic data.
Only vascular pressure and morphological assessments obtained just before the first embolization or surgery were used in this study to eliminate any possible confounding effects of partial treatment. We do not generally perform either embolization or surgery during the acute phase of AVM hemorrhage. In such instances, the effect of the clot might falsely elevate FMAP. All patients who had presented with significant hemorrhagic events included in the data set had imaging studies (MRI or CT) demonstrating absence of mass effect from any recent intraparenchymal hemorrhage.
We attempted to record the following characteristics, as summarized in bottom part of Table 2⇓.
Patients were classified into three groups20 21 according to the maximal diameter of the nidus as estimated in the pretreatment angiogram or, if available, initial contrast-enhanced CT or MRI: (1) small (<2.5 cm), (2) medium (2.5 to 5.0 cm), and (3) large (>5.0 cm). When possible, all three dimensions of the nidus were measured on the initial contrast-enhanced CT or MRI or on a repeat angiogram performed just before treatment using standards (metal disks) taped to the skull.
Using Spetzler-Martin criteria, we rated malformations; an AVM was considered eloquent if it was primarily located in the sensorimotor, language, and visual cortex, hypothalamus and thalamus, internal capsule, brain stem, cerebellar peduncles, or deep cerebellar nuclei.22 All other AVM locations were considered noneloquent.
Arterial Supply Feeding the AVM
By inspection of the pretreatment angiogram, an assignment of major arterial supplies was made: (1) anterior (eg, internal carotid, anterior cerebral, middle cerebral); (2) posterior (posterior cerebral, vertebrobasilar system); and (3) extracranial. These assignments were not mutually exclusive, ie, a given patient may have had more than one grouping assigned.
Recruitment of Pial-to-Pial Collaterals
Recruitment of pial-to-pial collaterals was noted when there was angiographic evidence of anastomoses between long circumferential arteries supplying two adjacent vascular territories or between pedicles within the same vascular territory (Figure 1⇓). These collaterals contributed to AVM supply and/or normal brain supply distal to the AVM.
Aneurysms, defined as luminal dilatations greater than twice the width of the parent vessel, can occur on arteries related or unrelated to the AVM, ie, within or outside of the path of shunt flow. Those associated with feeding arteries were subclassified into proximal and distal, depending on their location relative to the length of these vessels from their site of origin to the AVM nidus. Intranidal aneurysms represent discrete focal dilatations or sacculations within the conglomerate of tortuous dysplastic vessels near the site of arteriovenous shunting. They were considered to be present when seen on at least one view without vessel overlap (Figure 2⇓).
To avoid bias caused by a subjective interpretation of aneurysms, their presence was determined by an arbitrary rating scale. An aneurysm was considered “definite” when it was evident on two orthogonal views, “probable” when demonstrated on one view with no overlap of neighboring vessels, “equivocal” when seen on one view with vessel overlap, “probably not present” when there was a suggestion on one view with vessel overlap, and “definitely not present” when undetectable on any view. For the purpose of this analysis, the ratings of definite and probable were collapsed into the category of “aneurysm present”; the remainder were classified as “aneurysm not present.”
The border of an AVM was considered “sharp” when its nidus composition was compact and well circumscribed. A “diffuse” AVM was composed of a widespread nidus with infiltration into adjacent parenchyma.
Venous Drainage Pattern
The patterns of AVM venous drainage were classified into three groups: (1) superficial, (2) deep, or (3) superficial and deep. Deep venous drainage was further specified as periventricular if the main outflow of the AVM established connection with the medial and lateral groups of the subependymal veins (Figure 3⇓). Venous stenosis was defined as a greater than 50% reduction in the caliber of any draining vein outflow pathway in at least one angiographic view.
Arterial Pressure Measurements
Intravascular pressures were measured using methodology previously described. During superselective angiography, FMAP was measured just proximal to the nidus using an intracranial microcatheter, 1.8F at its distal tip (Chimiotherapie).23 24 During craniotomy, pressures were measured by direct puncture with a 26-gauge needle at the beginning of surgery (n=6); a detailed description of intraoperative pressure measurement techniques, including hydrostatic considerations for nonsupine patients, was provided by Young et al.25 All pressure measurements were referenced to right atrial pressure and were compared with the simultaneously recorded SMAP either in the extracranial internal carotid artery or vertebral artery (embolization patients) or radial artery (surgical patients). To normalize any variation in systemic blood pressure among different patients, we also calculated the ratio of cerebral-to-systemic pressure (FMAP/SMAP) and used these values in subsequent statistical analysis.
The main grouping factor was patient presentation as hemorrhage versus nonhemorrhage. Categorical data were analyzed by the χ2 method; continuous variables were analyzed using an unpaired t test. Multivariate analysis was performed to determine the independent contribution of all measured variables to the clinical presentation as spontaneous intracranial hemorrhage. The potential explanatory variables were tested sequentially using a stepwise logistic regression procedure. The significance of the addition of any factor to the statistical model was determined by a χ2 test. Other incidental tests are described in “Results.” Data were analyzed using either the Statview 4.5 software package (Abacus Concepts) or SAS software (SAS Institute). Results were considered significant at P≤0.05 and are reported as either a prevalence or a mean±SD. ORs and associated 95% CIs are given as incremental risk per unit of measurement for continuous data and remain dimensionless for categorical data.
Tables 2⇑ and 3⇓ show the results of the univariate analysis relating hemorrhagic presentation to physiological and angiographic characteristics obtained during the initial superselective angiography. There were 340 patients reviewed who had initial presentation documented and at least some details of venous anatomy documented. Not all patients had all of the analyzed risk factors recorded, which accounts for the unequal sample sizes shown. These are listed in descending order of the ORs.
A multivariate logistic model was subsequently constructed for hemorrhagic risk. In a first analysis, the stepwise variable selection procedure considered all significant univariate predictors for which data were available for most patients to confirm and verify our previous findings.20 Factors associated (all P<0.01) with hemorrhagic presentation for the group as a whole (n=319) were posterior fossa location (OR, 4.3; 95% CI, 1.7 to 11.3); exclusively deep venous drainage (OR, 3.5; 95% CI, 1.8 to 6.6); and small size (OR, 3.2; 95% CI, 1.6 to 6.4).
The second and main analysis used a subset to specifically consider the intracerebral pressure data and detailed angiographic characteristics. For this multivariate logistic model, we used a stepwise variable selection procedure to consider several sets of significant univariate predictors. We examined models including the following: (1) FMAP, exclusively deep venous drainage, periventricular venous drainage (n=67); (2) FMAP, exclusively deep venous drainage, number of draining veins reaching a sinus (n=63); and (3) FMAP, exclusively deep venous drainage (n=129).
The best model was a combination of FMAP and exclusively deep venous drainage; the other variables added nothing significant to the multivariate model.
Because venous drainage was categorized as a dichotomous variable (exclusively deep or mixed venous drainage), we constructed two “risk curves” that related FMAP to the risk of having presented with hemorrhage: one for exclusively deep venous drainage and one for mixed venous drainage (Figure 4⇓).
We used both absolute FMAP and FMAP expressed as a ratio to systemic arterial pressure in all our analyses. Because there was no significant difference between the two in terms of predictive power and because absolute FMAP is more intuitive than a ratio, FMAP is the only one presented (Table 4⇓). There was no relationship of FMAP to the time between diagnosis and treatment by linear regression. With respect to time between diagnosis and treatment, there was no difference (Mann-Whitney U test) between groups presenting with hemorrhage (mean, 36.7±75.9; median, 2.9; range, 0.1 to 296.0 months) and nonhemorrhage (mean, 25.6±60.4; median, 4.1; range, 0.2 to 311.5 months).
Relationship of Aneurysms to Presenting Hemorrhage
We explored the association between SAH presentation and (1) the presence of any intracranial aneurysm; (2) whether the aneurysm was related or unrelated to the path of shunt flow; and (3) the presence of an intranidal aneurysm. Our sample to date was too small to yield sufficient power to detect significant differences between subcategories, as shown in Table 5⇓. The data in Table 6⇓ were generated using logistic regression, with SAH presentation as the dependent variable and aneurysm presence or characteristics as the independent variable.
The novel findings of this study are that when a number of putative risk factors, including detailed angiographic characteristics, are considered, FMAP and the presence of exclusively deep venous drainage were the only independent predictors for estimating the relative risk of hemorrhage at initial AVM presentation; nidus size, nidus location, and the presence of associated aneurysms were not. We present a model allowing point estimates of individual risk and the associated 95% CI.
In this study we have focused our attention on the interplay of hemodynamic and morphological factors associated with spontaneous intracranial hemorrhage; a discussion of other clinical presentations in our data set have been presented elsewhere.2 26 Spontaneous intracranial hemorrhage from a cerebral AVM should result from a finite set of physiological and anatomic abnormalities in a given patient. As outlined in Table 7⇓, several studies have addressed the issue of risk factor identification. Factors suggested include a prior bleeding episode, a periventricular nidal location, a diffuse AVM morphology, an elevated FMAP, an exclusive deep venous drainage, the presence of a single draining vein, venous stenosis, venous reflux into a dural sinus or deep veins, and the ratio of number of afferent-to-efferent systems.3 7 12 13 15 16 17 20 Collateral pial-to-pial arterial and venous recruitment have also been identified as factors associated with a decreased risk of hemorrhage.13 17
Higher FMAP was associated with AVM hemorrhage. The notion that feeding arterial pressure is causally related to AVM hemorrhage is supported by the fact that smaller AVMs tend to have a higher feeding artery pressure and that smaller AVMs, when they bleed, tend to have a higher hemorrhage volume.15 The results of the present study support the notion that pressure is an important determinant of risk independent of AVM size. It is not clear how input (feeding artery) pressure to the AVM is related to true intranidal pressure. However, FMAP has also been shown to correlate positively with draining vein pressure without any apparent influence of angiographic venous stenosis (r=0.59, P<0.05).25 Therefore, feeding artery pressure is probably a good surrogate for the relative transmural pressure gradient in the vessels in the AVM nidus.
The present study also concurs with previous findings that venous characteristics predominate among risk factors for AVM hemorrhage.12 13 16 17 20 We constructed logistic regression models to investigate the various detailed angiographic descriptors of venous anatomy, such as the number of draining veins reaching a sinus and periventricular venous drainage. None of these models was better than the one that used exclusively deep venous drainage. This probably reflects the fact that these variables are highly correlated with exclusively deep venous drainage and that this risk factor is a better predictor when used in conjunction with FMAP. Exclusively deep venous drainage may have a profound effect on AVM hemodynamics, by promoting an increased pressure gradient across the vasculature of the nidus. It is also possible that turbulent flow and elevated pressure in the deep venous system promote enhanced platelet aggregation and thrombosis, to a degree sufficient to cause AVM hemorrhage.27
Because we have accumulated a much larger data set than in our previous report,20 we also performed a secondary analysis of a larger group of cases (n=319) to confirm that the previously described influence of size and venous drainage pattern (in the absence of cerebral pressure measurements) was still similar; we found that size and exclusively deep venous drainage were still independent predictors in a multivariate logistic regression model. In this secondary analysis, posterior fossa location was also significant. This latter observation points out the utility of focusing on cerebral pressure measurement as a risk marker because it is a physiological parameter that is potentially measurable in all patients. Risk attributable to factors such as posterior fossa location (in this sample it was only 11% of cases) may only be present in a limited proportion of the population.
Because aneurysms have been implicated as contributing to the pathogenesis of spontaneous AVM hemorrhage,10 12 16 28 a detailed review of imaging studies was undertaken to relate aneurysms to the mode of hemorrhage. Unlike other series,12 16 we were not able to establish a statistical relationship between the presence of associated arterial aneurysms (within or outside of the path of shunt flow or intranidal) and the occurrence of intracranial hemorrhage. Moreover, SAH, which occurred in 17 of 58 (21%) patients of that cohort in our study, did not correlate with aneurysm incidence. However, our patient sample with detailed angiographic evidence of aneurysms was not large enough to achieve sufficient statistical power (using a method for categorical analyses with unequal-sized samples,29 the sample size requirements for determining the association between SAH presentation and feeding artery aneurysm with an 80% power would be 130). It is possible that many selective digital subtraction views with a high rate of acquisition are required to demonstrate all distal arterial or intranidal aneurysms associated with cerebral AVMs. We found only a 21% to 25% incidence of aneurysms in our sample, most of which were related to arterial feeders or intranidal in location.
There are several issues that deserve mention. There may be a selection bias in our cohort, which is derived from a single-center referral patient population. Clearly, a prospective population-based study would be more desirable to obtain accurate information on AVM natural history with respect to the risk of hemorrhage. It is hoped that information provided in this report might serve as a basis for the rational design of such a population-based effort.
Of particular interest is the potential that we may have underrepresented small AVMs. We recently reviewed commonly collected clinical and morphological factors in a total of 1266 patients in our own database as well as those from the groups in Berlin, Paris, and Toronto, which are also large referral centers.30 For small size (largest dimension <3 cm), our group percentage was similar to all groups (23% to 32%) except the Toronto group (68%). However, in our 125 cases that had both size and FMAP documented, there were 25 small (<3 cm) lesions. Of these, 15 presented with hemorrhage and 10 did not. Therefore, the logistic regression that related FMAP and presentation should not be unduly biased. Note that the percentage of hemorrhagic presentation for patients with AVMs smaller than 3 cm was 67% (284 of 424) for the combined Berlin/Paris/New York/Toronto databases. Even though the Toronto data set had a higher total proportion of small AVMs compared with the other data sets, the percentage of patients with AVMs <3 cm who presented with a hemorrhage was 69% (126 of 183). Therefore, this proportion is similar across all groups to the proportion observed in our cohort of small lesions with FMAP measurements presented in this article. The FMAP in supratentorial long circumferential vessels (cortical branches of the anterior, middle, and posterior cerebral arteries) tended to be lower than anterior choroidal and posterior fossa feeding vessels, although the sample sizes were too small to test for meaningful differences. The magnitude of difference in FMAP between groups (hemorrhage versus nonhemorrhage) appeared to be similar across all types of vessels.
We have made the assumption that patients can be separated into distinct groups based on their tendency to hemorrhage. Differences in risk distributions among AVM patients can be shown using the example shown in Figure 4⇑. Two hypothetical patients might have quite different mean (point estimate) probabilities of hemorrhage risk, with no overlap in the 95% CI, indicating that these patients have significantly different risks.
For each level of the risk variables studied, one can find patients who have bled. That some patients will be found to have bled regardless of their risk factors relates to the fact that the described associations are probabilistic and not causal. The long-term goal of this type of work is to identify patients with low or high probabilities of hemorrhage to (eventually) tailor treatment to their risk levels.
The final issue is that data predictive of hemorrhage may not be valid for going in the other direction (that is, for predicting a low risk of hemorrhage in patients not harboring those features). This situation is no different from the case in diagnostic testing where a test has high sensitivity but low specificity. Such a diagnostic test would be useful only when positive. We emphasize, however, that it is possible to combine variables in a multivariate model in which factors with high sensitivity can be combined with factors with high specificity for maximum discriminating power.
To summarize, high FMAP and exclusively deep venous drainage were independently predictive of hemorrhagic AVM presentation. Arterial aneurysms, regardless of their location and relationship to the AVM nidus, AVM size, and AVM location did not independently predict hemorrhagic presentation when considered together in a model that included FMAP. The findings in our series suggest that a high intranidal pressure, as reflected by high arterial input (FMAP) pressure or venous outflow restriction (exclusively deep venous drainage), is a primary factor associated with hemorrhagic AVM presentation. Further studies will allow more insight into the pathological mechanisms of AVM rupture. Additional refinement of risk factors may also be useful for guiding treatment decisions. For example, risk factors may be used in the implementation of clinical trials to guide stratification or to devise statistical models for risk-based (or assured) allocation.31 32 33
Selected Abbreviations and Acronyms
|FMAP||=||feeding mean arterial pressure|
|SMAP||=||systemic mean arterial pressure|
This work was supported in part by NIH grants RO1-NS27713 and RO1-NS34949. The authors wish to thank Steven Marshall, BS, and Joyce Ouchi for assistance in preparation of the manuscript. The authors gratefully acknowledge the support and contributions of the other members of the Columbia University AVM Study Group.
- Received November 17, 1997.
- Revision received March 9, 1998.
- Accepted March 26, 1998.
- Copyright © 1998 by American Heart Association
Valavanis A. The role of angiography in the evaluation of cerebral vascular malformations. Neuroimaging Clin N Am. 1996;6:679–704.
Mast H, Mohr JP, Osipov A, Pile-Spellman J, Marshall RS, Lazar RM, Stein BM, Young WL. ‘Steal’ is an unestablished mechanism for the clinical presentation of cerebral arteriovenous malformations. Stroke. 1995;26:1215–1220.
Pollock BE, Flickinger JC, Lunsford LD, Bissonette DJ, Kondziolka D. Factors that predict the bleeding risk of cerebral arteriovenous malformations. Stroke. 1996;27:1–6.
Crawford PM, West CR, Chadwick DW, Shaw MDM. Arteriovenous malformations of the brain: natural history in unoperated patients. J Neurol Neurosurg Psychiatry. 1986;49:1–10.
Marks MP, Bracci P, Steinberg GK. Natural history of cerebral AVMs correlated with angiographic risk factors of hemorrhage. Paper presented at: American Society of Neuroradiology 31st Annual Meeting; May 16–20, 1993; Vancouver, BC, Canada. Paper 91, p 75.
Porter PJ, TerBrugge KB, Stefani MA, Kerr RG, Montanera W, Willinsky RA, Wallace MC. A prospective study assessing risk factors for hemorrhage from arteriovenous malformations. Intervent Neuroradiol. 1997;3(suppl 1):96. Abstract 107.
Fleischer LH, Young WL, Pile-Spellman J, terPenning B, Kader A, Stein BM, Mohr JP. Relationship of transcranial Doppler flow velocities and arteriovenous malformation feeding artery pressures. Stroke. 1993;24:1897–1902.
Fogarty-Mack P, Pile-Spellman J, Hacein-Bey L, Osipov A, DeMeritt J, Jackson EC, Young WL. The effect of arteriovenous malformations on the distribution of intracerebral arterial pressures. AJNR Am J Neuroradiol. 1996;17:1443–1449.
Osipov A, Koennecke H-C, Hartmann A, Young WL, Pile-Spellman J, Hacein-Bey L, Mohr JP, Mast H. Seizures in cerebral arteriovenous malformations: type, clinical course, and medical management. Intervent Neuroradiol. 1997;3:37–41.
Fleiss JL. Unequal Sample Sizes in Statistical Methods for Rates and Proportions. 2nd ed. New York, NY: John Wiley & Sons Inc; 1981:44–46.
Hofmeister C, Hartmann A, Meisel J, Mansmann U, Lasjaunias P, terBruegge K, Pile-Spellman J, Young WL, Mohr JP. Epidemiological, clinical, morphological characteristics of 1266 patients with cerebral arteriovenous malformation. Cerebrovasc Dis. In press. Abstract.