Silent Intralesional Microhemorrhage as a Risk Factor for Brain Arteriovenous Malformation Rupture
Background and Purpose—We investigated whether brain arteriovenous malformation silent intralesional microhemorrhage, that is, asymptomatic bleeding in the nidal compartment, might serve as a marker for increased risk of symptomatic intracranial hemorrhage (ICH). We evaluated 2 markers to assess the occurrence of silent intralesional microhemorrhage: neuroradiological assessment of evidence of old hemorrhage—imaging evidence of bleeding before the outcome events–and hemosiderin positivity in hematoxylin and eosin-stained paraffin block sections.
Methods—We identified cases from our brain arteriovenous malformation database with recorded neuroradiological data or available surgical paraffin blocks. Using 2 end points, index ICH or new ICH after diagnosis (censored at treatment, loss to follow-up, or death), we performed logistic or Cox regression to assess evidence of old hemorrhage and hemosiderin positivity adjusting for age, sex, deep-only venous drainage, maximal brain arteriovenous malformation size, deep location, and associated arterial aneurysms.
Results—Evidence of old hemorrhage was present in 6.5% (n=975) of patients and highly predictive of index ICH (P<0.001; OR, 3.97; 95% CI, 2.1–7.5) adjusting for other risk factors. In a multivariable model (n=643), evidence of old hemorrhage was an independent predictor of new ICH (hazard ratio, 3.53; 95% CI, 1.35–9.23; P=0.010). Hemosiderin positivity was found in 36.2% (29.6% in unruptured; 47.8% in ruptured; P=0.04) and associated with index ICH in univariate (OR, 2.18; 95% CI, 1.03–4.61; P=0.042; n=127) and multivariable models (OR, 3.64; 95% CI, 1.11–12.00; P=0.034; n=79).
Conclusions—The prevalence of silent intralesional microhemorrhage is high and there is evidence for an association with both index and subsequent ICH. Further development of means to detect silent intralesional microhemorrhage during brain arteriovenous malformation evaluation may present an opportunity to improve risk stratification, especially for unruptured brain arteriovenous malformations.
Optimal management of brain arteriovenous malformation (bAVM) is critically dependent on accurate assessment of the competing risks of interventional treatment versus those of the natural history of the disease. The strongest and most widely accepted risk factor is clinical presentation with a symptomatic intracranial hemorrhage (ICH); other risk factors such as age at diagnosis, deep-only venous drainage, and deep location are weaker in comparison.1,2 Among patients who present with an unruptured lesion—roughly half of all detected cases—additional risk factors are needed to further discriminate risk of future symptomatic ICH.
Although the occurrence of clinically “silent” hemorrhages in patients with bAVM has been recognized for some time,3–5 use of this information for risk stratification has not been developed. Because clinically symptomatic ICH at presentation (index ICH) is such a strong risk factor for future ICH, these silent intralesional microhemorrhages (SIM) may represent the same biological phenomenon and also signal increased risk for clinical ICH. Importantly, they may be amenable to detection by MR sequences sensitive to iron as proposed for other disease states.6,7
Our interest in SIM began when we noted that our animal model of the bAVM phenotype was associated with microbleeding.8,9 In initial comparisons to human samples, we found a much higher incidence of hemosiderin and Prussian blue positivity than expected, even in unruptured cases. To further explore the phenomenon, we examined a variable in our bAVM study project database1: neuroradiological evidence of old hemorrhage (EOOH). The surprising strength of the EOOH effect that we noted prompted us to systemically review histopathologic material.
Specifically, we investigated whether 2 markers of SIM that occurred before diagnosis were associated with a higher risk of symptomatic ICH, that is, studying the association of ICH with (1) EOOH; and (2) the presence of hemosiderin in archived paraffin blocks. We addressed the question of whether SIM–clinically silent episodes of bleeding in the nidal compartment–might signal increased risk of clinically symptomatic ICH.
All patients included are part of the University of California, San Francisco bAVM Study Project registry. Those enrolled prospectively since 2000 gave informed consent; cases between 1992 and 2000 were reviewed retrospectively. This work was approved by the University of California–San Francisco Committee on Human Research.
A standardized neuroradiological data collection form following the Joint Writing Group guidelines10 was completed by an attending neurointerventional radiologist. Available imaging data generally included tomographic imaging from an outside facility, where the presumptive diagnosis was made, and 4-vessel angiograms performed at the University of California, San Francisco.
We included only cases for which information on EOOH was recorded defined as CT or MR (generally T1- or T2-weighted sequences) evidence of bleeding before diagnosis. MR evidence included signal loss consistent with hemosiderin as well as indirect evidence of old hemorrhage, that is, encephalomalacia adjacent to the lesion consistent with a prior hematoma or calcification; CT was only useful for assessing parenchymal calcification and encephalomalacia.
We selected a series of paraffin blocks collected from between 1992 and 2011 enriched for (1) unruptured cases; and (2) those at the extremes of age (≤15 years and ≥60 years) to examine age-dependency seen in other forms of microbleeds such as lobar ICH.11 Five-micron sections were stained with hematoxylin and eosin.
Blinded to patient history, 2 neuropathology reviewers (T.S. and T.T.) used a 4-point scale (none, small, moderate, large) to gauge the relative amount of hemosiderin using birefringent as criteria or brownish material seen within the vascular wall or stromal tissue between vascular elements. Staining visualized only on the luminal surface, within vessel lumens, or inside of a macrohemorrhage and adjacent fibrinous material was not counted. Macrophage infiltration was assessed on a 5-point scale (none, minimal, focal, marked, extensive). For statistical analysis, both scales were dichotomized into absence or presence.
An exploratory experiment was performed to see how well ex vivo MRI correlated to histological evidence of prior hemorrhage (see online-only Supplemental Figures I and II; http://stroke.ahajournals.org).
For neuroradiological data, patients with and without EOOH were compared using descriptive statistics, including t tests for continuous variables and χ2 tests for categorical variables. Although a range of risk factors has been described as being associated with both index and follow-up ICH, we predefined a limited set for this analysis, including age at diagnosis (decade), sex, deep-only venous drainage, maximal bAVM size (cm), deep location, and associated arterial aneurysms (either intranidal or extranidal flow-related arterial aneurysms).2
Analysis was carried out in 2 stages. The first stage examined the relationship between initial clinical presentation with symptomatic ICH (index ICH) and EOOH. Both univariate and multivariable logistic regression analysis was performed using index ICH as the outcome and EOOH as the primary predictor. We chose to examine a multivariable model with inclusion of all predefined ICH risk factor variables irrespective of univariate significance.
The second stage examined the relationship between EOOH and its effect on the time to subsequent ICH after diagnosis in the natural course before any treatment. We performed Cox proportional hazards analysis of time to first ICH censoring patients at treatment, death, or last follow-up. Kaplan-Meier survival curves and log-rank tests assessed hemorrhage-free survival for patients with and without EOOH. Both univariate and multivariable Cox proportional hazards models were used, similarly including all putative risk factors for ICH. Sensitivity analyses included exclusion of cases with a history of stroke.
Because we believe that EOOH and index ICH are in the same causal pathway, we did not include index ICH in the multivariable model. However, as a sensitivity analysis, we performed a mediational analysis. We constructed a separate multivariable model substituting index ICH for EOOH and then an additional model with both index ICH and EOOH and their interaction.
For the hemosiderin analysis of paraffin block sections, univariate and multivariable logistic regression analyses were performed similar to that previously described but using hemosiderin positivity as the primary predictor. Sensitivity analyses included removing cases that underwent preoperative embolization and/or had a history of prior stroke and an exploration of the effect of delay between surgical harvest and either diagnosis or last ICH event.
Statistical analyses used Intercooled Stata Version 12. A probability value of P<0.05 was taken as the level of significance.
Association of EOOH With Index ICH
Baseline characteristics (n=975) are shown in Table 1; EOOH was present in 6.5% of patients. Representative images of cases scored positive for EOOH are shown in Figure 1. The fraction of presentation with index ICH was greater in the EOOH group (76% versus 42%; P<0.001). Other characteristics did not differ (P>0.05) between EOOH groups, the exception being a history of remote stroke (7 of 15 histories consistent with hemorrhagic stroke). Multivariable logistic regression results are shown in Table 2 and univariate results in online-only Supplemental Table I. The multivariable model showed that EOOH was highly predictive of index ICH (OR, 3.97; 95% CI, 2.10–7.50; P<0.001). Age, deep-only venous drainage, bAVM size, deep location, and associated arterial aneurysm remained highly predictive of index ICH in a multivariable model.
Association of EOOH With Subsequent ICH After Diagnosis
For longitudinal data (n=699), 4.9% of patients experienced a postdiagnosis ICH in 1626 patient-years of follow-up; there were 34 ICH events: 8 events in 41 patients with EOOH (20%) and 26 events in 658 patients without EOOH (4%).
Cumulative hemorrhage-free survival and number at risk entering each 5-year interval are shown in Figure 2 displayed on a time scale that contains the majority of events. There was a trend for unadjusted EOOH to be associated with a higher risk of subsequent ICH (log-rank test P=0.068). Multivariable Cox regression is shown in Table 2. In the multivariable model, EOOH remained an independent predictor of ICH (hazard ratio, 3.53; 95% CI, 1.35–9.23; P=0.010) after adjusting for our prespecified ICH risk factors.
For the mediational analysis, we found that the effect and CIs of index ICH (hazard ratio, 2.82; 95% CI, 0.99–8.04; P=0.052) were similar to that of EOOH (hazard ratio, 3.53; 95% CI, 1.35–9.23; P=0.01) and, more importantly, in a combined model of index ICH and EOOH, both displayed a 35% to 56% reduction in hazard ratio and became nonsignificant with no (P=0.464) interaction (data not shown). These observations are consistent with the 2 factors being at least partially in the same causal pathway, that is, previous silent hemorrhage leading to clinical presentation with an ICH representing a similar process that increases risk for ICH subsequent to diagnosis. A multivariable model substituting index ICH for EOOH (Table 2, bottom panel) showed a slighter lower effect size for index ICH (data not shown).
Prevalence of Hemosiderin Positivity and Macrophage Infiltration
A total of 127 patients were included in these analyses. The baseline characteristics are similar to Table 1, the exceptions being the enriched fraction of unruptured cases (70.4%) and a trend to have, as expected, fewer centrally (deep) located lesions with a lower fraction of deep-only venous drainage (data not shown).
We explored the relationship of the scoring system as a categorical predictor variable to the index ICH outcome. Although there was a trend toward a relationship between the scored amount of hemosiderin and risk of index ICH, due to the small sample sizes in some cells, the scoring scale was dichotomized into absence and presence for further analysis. The same was done for macrophage infiltration.
The raw categorical variables for macrophage infiltration and hemosiderin were highly correlated (Spearman ρ=0.74; P<0.001). Using logistic regression on dichotomized variables, hemosiderin positivity strongly predicted macrophage infiltration (OR, 29; 95% CI, 11 – 82; P<0.001). Hemosiderin positivity was found in 36.2% (29.6% in unruptured; 47.8% in ruptured; P=0.04) and macrophage infiltration was found in 43.3% (37.0% in unruptured and 54.4% in ruptured; P=0.058).
Association of Hemosiderin Positivity With Index ICH
Hemosiderin positivity was associated with index ICH in univariate analysis (OR, 2.18; 95% CI, 1.03–4.61; P=0.042). In multivariable analysis (Table 3; n=79), hemosiderin positivity, age at diagnosis, deep-only venous drainage, and associated arterial aneurysms remained significant (P<0.05). Full univariate results are provided in online-only Supplemental Table I. Excluding cases undergoing presurgical embolization or those with a remote history of stroke did not affect the relationship of hemosiderin positivity to index ICH in either univariate or multivariable models (data not shown). There was no clear relationship between EOOH and hemosiderin positivity, but in the 99 cases with data for both variables, there were only 5 with EOOH, making estimates unreliable (data not shown).
To explore potential confounding by either index ICH or postdiagnostic, preoperative ICH and surgical harvest, we restricted the logistic regression to include only cases with an interval of <4 days between the last clinical hemorrhage and surgical harvest. The OR for hemosiderin positivity remained similar and significant (data not shown). We also examined a model that adjusted for the period of time between diagnosis and surgical harvest. The interval between diagnosis and harvest was not significant (P=0.561) but the other effect sizes remained similar to those shown in Table 3 (data not shown).
We did not undertake time-to-event analyses for new ICH because there were only 5 outcome events in this subset of patients, making the models unstable and the results unreliable.
Association of Hemosiderin Positivity With Ex Vivo MRI
Online-only Supplemental Figures I and II show good correspondence between MR susceptibility and histopathologic demonstration of iron in resected specimens.
We examined 2 markers of SIM in patients with bAVM: (1) EOOH as seen on baseline tomographic imaging; and (2) hemosiderin positivity in resected surgical specimens. We made the following observations: (1) EOOH was independently associated with both index ICH (hazard ratio, 3.97; P<0.001) and new ICH after diagnosis (hazard ratio, 3.52; P=0.006) adjusting for other common ICH risk factors; (2) hemosiderin positivity was highly prevalent in both unruptured (29.6%) and ruptured (47.8%) bAVM; (3) hemosiderin positivity was independently associated with index ICH (hazard ratio, 3.91; P=0.03); (4) hemosiderin positivity and macrophage infiltration were highly correlated (P<0.001); and (5) ex vivo imaging of resected bAVM tissue demonstrated signal loss in areas corresponding to hemosiderin deposition. These data form a solid basis for planning future studies necessary to validate the use of SIM as a risk stratification tool.
Both hemosiderin positivity and EOOH are markers for the underlying biological phenomenon of SIM (Figure 3). Tissue histopathology is a relatively sensitive and specific means of assessing silent bleeding but is only useful for demonstrating the phenomena and cannot be used practically as a biomarker. The other predictor variable that we used, EOOH, is relatively insensitive because standard clinical imaging was not optimized for detecting iron, especially small amounts. Therefore, neither of these indices is proposed for future development per se. Rather, they provide evidence that use of modern iron-sensitive imaging is likely to be a fruitful approach.
The presence of prior hemorrhage in bAVMs is by no means a new finding.3–5 Curiously, despite the numerous reports supporting the ontology of the phenomena and suggestions that it be considered in risk evaluation, neither the clinical nor research community has heretofore developed “imaging evidence of prior hemorrhage” as a means of risk stratification. This may be in part attributable to the low sensitivity of older MR techniques. However, there has been considerable progress in susceptibility-weighted imaging.12
It is not clear how SIM corresponds to the imaging construct, “brain microbleeds,” that have garnered considerable recent attention in other cerebrovascular diseases, including the use of microbleeds as a means of risk prediction for recurrent lobar hemorrhages.6,7 Whatever the relationship to brain microbleeds is, the intralesional microhemorrhage phenotype in bAVM appears to be highly prevalent, and with iron-sensitive MR imaging methodology at higher MR field strength, SIM might represent a useful biomarker for gauging the risk of ICH in patients harboring bAVMs. In addition, the tight coupling of macrophage infiltration with hemosiderin positivity might offer another potential biomarker strategy using newer methods for imaging macrophages in the brain13 given the known inflammatory phenotype of the bAVM nidus.14
There are limited studies correlating histological examination with MR evidence of microhemorrhage, but the available suggests good correspondence.12,15 Under optimized ex vivo conditions, we found a good correspondence between histological presence of iron and MR susceptibility in bAVM tissue (see online-only Supplemental Figures I and II). However, much more work is needed to systematically compare tissue sections with modern iron-sensitive MR imaging techniques.
Our tissue studies suggest an overall prevalence of hemosiderin positivity of approximately 40%. We speculate that iron-sensitive MR imaging will yield at least a similar prevalence. Our tissue sections are but a small sample of the entire bAVM volume. Therefore, despite being sensitive, histological examination may not be specific. Our EOOH variable is the opposite: a large brain volume is interrogated but the MR images that we reviewed to generate the EOOH indicator variable were relatively iron-insensitive, resulting in an EOOH prevalence of only approximately 7%. Interestingly, other reports using similar MR methods estimated prevalence of prior hemorrhage as high as 23%.3–5
Although we found that a significant effect of EOOH is associated with new ICH after diagnosis in a time-to-event analysis, in this set of patients, we did not find a significant effect of deep-only venous drainage, unlike our previously reported findings and those of others.1,2 Therefore, the results for the time-to-event analysis should be interpreted with caution.
Our study has important limitations. The data reported here are primarily retrospective in nature and the neuroradiological data collection form was not designed to provide details on EOOH. Except in 2 cases in which “encephalomalacia” was noted, the neuroradiological consultant did not indicate on what the presumption of old hemorrhage was based. Excluding these 2 cases did not affect our estimates. Although we presume that the majority of the positive responses to the question were due to MR evidence of prior bleeding, we cannot confirm the nature of the findings.
It is possible that our histopathologic examination did not discriminate between blood from a SIM and the blood that resulted from an index ICH. An additional and important consideration that applies to the tissue analysis but does not apply to the preoperative brain imaging studies is that a period of time elapses between bAVM diagnosis and surgical resection. There are several considerations. First, a recent macrohemorrhage might confound interpretation of hemosiderin positivity. The interval between arrival of extravascular blood and its processing into hemosiderin is not precisely known, but based on available information, we chose 4 days to use in our sensitivity analysis.16,17 Furthermore, it is not known how long hemosiderin remains in the tissue after it is formed and phagocytized by macrophages, although presumably it is a period of many years.
Second, new SIM might occur between diagnosis and surgical harvest, especially with preoperative embolization. Therefore, hemosiderin positivity may reflect an event that would not be in evidence if, for example, one wished to use iron-sensitive MR to detect SIM for a baseline examination. Assuming that bAVMs are present some period of years before diagnosis, the generation of SIMs should be distributed over the prediagnostic natural history, so inclusion of an additional small time interval should minimally influence results. Although we cannot definitively address these points in this retrospective study, our sensitivity analyses suggest that there was not major confounding by timing considerations.
Third, our assessment of EOOH on index imaging in patients with hemorrhage may have introduced bias in that raters examining imaging with hemorrhage present may be more likely to grade a scan as demonstrating EOOH. A prospective study that considers baseline premorbid imaging in unruptured patients can address this concern.
In conclusion, we provide evidence that detection of SIM may represent a biomarker for risk of ICH and would appear to be eminently suited for modern MRI to develop a novel risk stratification tool to further optimize the management of bAVM. In particular, there is a pressing need to improve risk stratification for unruptured bAVMs: if the nearly completed A Randomized Trial of Unruptured Brain AVMs (ARUBA) trial (NCT00389181) suggests that even short-term outcome favors noninterventional management, there will be increased demand for additional means to assess ICH risk in the unruptured patient.
Sources of Funding
Supported in part by National Institutes of Health grants R01NS034949 and R01NS027713 (W.L.Y.), P01NS044155 (W.L.Y., H.S.), and K23NS058357 (H.K.).
We thank members of the University of California, San Francisco bAVM project (http://avm.ucsf.edu/) for assistance with data collection and report generation.
Soonmee Cha, MD; Christopher F. Dowd, MD; Anne Fedoroff, RN, BSN; Elizabeth Gardner, BA; Van V. Halbach, MD; Randall T. Higashida, MD; Philippe Jolivalt, BS; Brad Dispensa, BS, Timothy Shepherd, MD, PhD; and Yuanli Zhao, MD, PhD.
The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.111.647263/-/DC1.
- Received December 7, 2011.
- Revision received January 5, 2012.
- Accepted January 18, 2012.
- © 2012 American Heart Association, Inc.
- Kim H,
- Sidney S,
- McCulloch CE,
- Poon KY,
- Singh V,
- Johnston SC,
- et al
- Stapf C,
- Mast H,
- Sciacca RR,
- Choi JH,
- Khaw AV,
- Connolly ES,
- et al
- Yousem DM,
- Flamm ES,
- Grossman RI
- Cordonnier C,
- Klijn CJ,
- van Beijnum J,
- Al-Shahi Salman R
- Chen W,
- Guo Y,
- Bollen AW,
- Su H,
- Young WL
- Atkinson RP,
- Awad IA,
- Batjer HH,
- Dowd CF,
- Furlan A,
- Giannotta SL,
- et al
- Poels MM,
- Ikram MA,
- van der Lugt A,
- Hofman A,
- Krestin GP,
- Breteler MM,
- et al
- Thiel A,
- Radlinska BA,
- Paquette C,
- Sidel M,
- Soucy JP,
- Schirrmacher R,
- et al