Hemorrhage Pattern and Amyloid Burden in Cerebral Amyloid Angiopathy.
May 6, 2026DOI: 10.1161/jaha.125.048858Journal ArticleAHA
Written by: Elif GokcalJ Alex BeckerMitchell J HornOfer RotschildAlvin S DasAvia Abramovitz FouksJonathan RosandAnand ViswanathanKeith A JohnsonM Edip GurolSteven M Greenberg
This study used data from an ongoing single‐center observational cohort of patients with CAA at Massachusetts General Hospital, with enrollment procedures described previously.9 Briefly, participants met the modified Boston criteria for probable or definite CAA.3 The inclusion criteria included age ≥55 years old and magnetic resonance imaging (MRI) demonstrating (1) ≥2 hemorrhages (ICH or CMB) restricted to lobar, cortical, and cortico‐subcortical regions (cerebellar hemorrhages allowed) or (2) single lobar, cortical, or cortico‐subcortical hemorrhage accompanied by any degree of cSS; (3) absence of any hemorrhagic lesions in deep brain regions; and (4) absence of any alternative causes for hemorrhagic lesions (excessive warfarin [international normalized ratio >3.0]; blood dyscrasia, antecedent head trauma or ischemic stroke, intracranial tumor, vascular malformation, or vasculitis). Individuals with mild cognitive impairment, dementia, and any other neurodegenerative conditions were excluded based on comprehensive clinical and neuroimaging assessments.10, 11 Detailed information, including demographics and risk factors, was prospectively recorded at the time of cohort entry.All participants underwent structural MRI at 3T and 11C Pittsburgh compound B (PiB) PET imaging, as described below. The study was approved by the local institutional review board, and all participants provided written informed consent. No photographs, videos, or other identifiable personal information were used. The study was reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology guidelines. The data that support the findings of this study are available from the corresponding author upon reasonable request.All participants underwent a research MRI scan at 3T field strength. The imaging protocol included a 3‐dimensional T1‐weighted multiecho magnetization‐prepared rapid gradient‐echo sequence, fluid‐attenuated inversion recovery, T2‐weighted, and 3‐dimensional susceptibility‐weighted imaging sequences.Imaging markers of CAA, including lobar ICH, lobar CMBs, cSS, and CSO‐EPVS, were identified using established criteria.12, 13 ICH was defined as a parenchymal hemorrhage >5 to 10 mm in the largest diameter that does not meet criteria for a CMB.14, 15 cSS was scored using a validated multifocality scale.16 In this scale, each hemisphere was rated as 0 (no cSS), 1 (cSS in 1 sulcus or up to 3 adjacent sulci), or 2 (cSS in ≥2 nonadjacent sulci or >3 adjacent sulci). Right and left scores were summed (range, 0–4), with 0 indicating no cSS, 1 mild/unifocal cSS, and ≥2 severe/multifocal cSS. CSO‐EPVS was rated on axial T2‐weighted images using a validated 4‐point visual scale: 0=no CSO‐EPVS; 1=1 to 10 CSO‐EPVS; 2=11 to 20 CSO‐EPVS; 3=21 to 40 CSO‐EPVS; and 4=>40 CSO‐EPVS. A score of 4 was categorized as high‐degree CSO‐EPVS.17 ICH, CMBs, and cSS were rated on susceptibility‐weighted imaging and CSO‐EPVS on T2‐weighted images, with each marker assessed independently in separate sessions.
The FreeSurfer software suite (surfer.nmr.mgh.harvard.edu) was used to calculate WMH volume and estimated total intracranial volume, using an automated fluid‐attenuated inversion recovery‐based method independent of MRI ratings. WMH volume was expressed as a percentage of estimated total intracranial volume to account for individual differences in head size. For participants with large hemispheric ICH, measurements from the contralateral hemisphere were used and doubled to estimate total volume.
To capture variation in hemorrhagic lesions, participants were categorized based on the relative severity of cSS and CMBs. cSS severity was assessed using the previously defined multifocality score (range, 0–4), and total lobar CMBs were counted on susceptibility‐weighted imaging. Patients were then classified into 4 patterns: cSS‐D: cSS score ≥2 and ≤10 CMBs; CMB‐D: >10 CMBs and cSS score <2; both dominant: cSS score ≥2 and >10 CMB; and nondominant: cSS score <2 and ≤10 CMBs. The classification schema is illustrated in Figure 1. Among 67 patients with a diagnosis of probable or definite CAA, 5 individuals who met the criteria for the both‐dominant pattern were excluded from further analysis due to low sample size, resulting in a final cohort of 62 participants. Although ICH represents a hemorrhagic manifestation of CAA, it was not incorporated into the dominance classification, given the limited subgroup sizes that would have substantially reduced statistical stability.18
Representative susceptibility‐weighted images of each hemorrhage pattern. A, N‐D pattern in a participant with lobar intracerebral hemorrhages (squares) and a few CMBs. B, CMB‐D pattern in a participant with >10 CMBs (arrows) and mild cSS. C, cSS‐D pattern in a participant with severe cSS (circles) and few CMBs.
All PET scans were acquired on a Siemens/computed tomography ECAT HR+ scanner (Knoxville, TN) in 3‐dimensional mode and processed as previously described.19 Briefly, each scan was coregistered to each subject's magnetization‐prepared rapid gradient‐echo MRI using Statistical Parametric Mapping 8 software. MRI‐defined cortical and subcortical regions of interest were transformed into native PET space, and tracer uptake was sampled from each hemisphere. Carbon‐11‐PiB (11C‐PiB) (8.5–15 mCi) was injected as a bolus, with dynamic imaging from 40 to 60 minutes postinjection. The Logan graphical method was used to compute regional distribution volume ratios (DVRs) for each region of interest, using the cerebellar cortex as the reference region. Global cortical PiB DVR was calculated as the mean DVR across cortical regions of interest. In patients with a large ICH, results from the non‐ICH hemisphere were used for the analyses. All PET analyses were performed blinded to MRI ratings and WMH volume measurements.
Analyses were conducted on the final cohort (n=62), excluding the both‐dominant group due to low sample size. All analyses were exploratory and biologically informed, consistent with the observational study design and sample size, and were intended to characterize associations rather than support causal or predictive inference.
Continuous variables are presented as mean±SD or median (interquartile range) depending on distributional characteristics, and categorical variables are presented as counts and percentages. Normality was assessed using visual inspection of histograms and Q–Q plots in combination with the Shapiro‐Wilk test. Homogeneity of variance was assessed using visual inspection.
Group comparisons among the 3 subgroups (CMB‐D, cSS‐D, and nondominant) were performed using 1‐way ANOVA when parametric assumptions were met or the Kruskal‐Wallis test when they were not, for continuous variables. Categorical variables were analyzed using χ2 or Fisher exact tests, as appropriate. When a significant overall group effect was identified, post hoc pairwise comparisons were conducted as follows: for ANOVA, Bonferroni‐adjusted pairwise comparisons were used; for Kruskal‐Wallis tests, pairwise comparisons were performed using Mann‐Whitney U tests; and for categorical variables, pairwise comparisons were conducted using χ2 or Fisher exact tests. Bonferroni correction was applied for 3 pairwise comparisons, and adjusted P values are reported.
To examine associations between imaging markers and hemorrhage pattern, separate multivariable logistic regression models were constructed for each imaging marker (global cortical PiB DVR, high‐degree CSO‐EPVS, or WMH volume) rather than including multiple imaging predictors in a single model. Dominant hemorrhage pattern was modeled as the outcome (CMB‐D or cSS‐D versus nondominant and CMB‐D versus cSS‐D). This approach was chosen to limit model complexity and reduce the risk of overfitting, given the modest sample size. All regression models were adjusted for age and sex, based on their clinical relevance and established role as demographic confounders in CAA. Effect estimates with 95% CIs are emphasized along with the P values.
Given univariate differences in ICH prevalence between the cSS‐D and CMB‐D groups, we further evaluated this association using age‐ and sex‐adjusted logistic regression. Separate regression models were also used to examine the associations between global PiB DVR and CSO‐EPVS (logistic regression) and between global PiB DVR and WMH volume (linear regression, with WMH volume as the dependent variable), each adjusted for age and sex.
Because ICH was not incorporated into the dominance classification and represents a clinically distinct hemorrhagic manifestation within the CAA spectrum, sensitivity analyses were performed with additional adjustment for ICH in regression models to assess robustness.14, 18
To improve the interpretability of regression coefficients, global cortical PiB DVR and WMH volume were scaled by multiplying by 10 to reflect the effect per 0.1‐unit increase. WMH volume was additionally log‐transformed before modeling to reduce skewness, mitigate the influence of extreme values, and improve the linearity of its association with outcomes. In analyses where WMH was modeled as a dependent variable (linear regression), this transformation also improved residual distributional properties. Model assumptions, including linearity, were assessed using visual inspection of residual plots and were considered reasonable after transformation.
All statistical analyses were conducted using IBM SPSS Statistics (version 28.0). A 2‐tailed P<0.05 was considered statistically significant.
RESULTS
The final study cohort included 62 patients with a diagnosis of CAA, categorized into 3 groups: CMB‐D (n=31), cSS‐D (n=17), and nondominant (n=14). Baseline demographics and imaging characteristics for each group are summarized in Table 1. As expected, the groups significantly differed in terms of lobar CMB counts and presence of cSS, whereas demographics did not differ significantly among groups. The overall frequency of ICH did not differ significantly among groups (P=0.053). However, patients with the cSS‐D pattern showed a numerically higher rate of ICH compared with those with the CMB‐D pattern, and this association was significant in a logistic regression model adjusted for age and sex (odds ratio [OR], 1.65 [95% CI, 1.1–23.0]; P=0.031).
Baseline Demographics and Imaging Characteristics of Patients With Nondominant, CMB‐D, and cSS‐D Patterns
Values are presented as mean±SD or median (IQR) for continuous variables and n (%) for categorical variables. Overall group differences were assessed using 1‐way ANOVA (age), Kruskal‐Wallis tests (lobar CMB counts and WMH), and χ2 or Fisher exact tests (categorical variables), as appropriate. When the overall test was significant, pairwise comparisons were performed (Mann‐Whitney U tests for lobar CMB counts; χ2 or Fisher exact tests for categorical variables) with Bonferroni adjustment for 3 comparisons. Reported pairwise P values are Bonferroni‐adjusted. CMB indicates cerebral microbleed; CMB‐D indicates cerebral microbleed‐dominant; CSO‐EPVS, centrum semiovale enlarged perivascular spaces; cSS, cortical superficial siderosis; cSS‐D, cortical superficial siderosis‐dominant; eTIV, estimated total intracranial volume; ICH, intracerebral hemorrhage; IQR, interquartile range; and WMH, white matter hyperintensity.
Global cortical PiB DVR significantly differed among groups (P=0.011, Figure 2). In post hoc comparisons, both the CMB‐D and cSS‐D groups had significantly higher PiB DVR relative to the nondominant group (P=0.033 and P=0.014, Bonferroni‐corrected; Figure 2). These associations remained significant in logistic regression models adjusted for age and sex (CMB‐D versus nondominant: OR, 1.8 [95% CI, 1.1–2.8]; P=0.009; cSS‐D versus nondominant: OR, 1.6 [95% CI, 1.1–2.5]; P=0.020; Table 2). No difference was observed between the CMB‐D and cSS‐D groups (P=1.000).
Mean±SD values are indicated within each group. Group differences were tested using 1‐way ANOVA (overall P=0.011), with Bonferroni‐corrected post hoc comparisons: CMB‐D vs N‐D P=0.033, cSS‐D vs N‐D P=0.014, and cSS‐D vs CMB‐D P=1.000 (not significant). CMB‐D indicates cerebral microbleed‐dominant; cSS‐D, cortical superficial siderosis‐dominant; N‐D, nondominant; and PiB‐DVR, Pittsburgh compound B‐distribution volume ratio.
Multivariable Logistic Regression Models Evaluating the Independent Association of PiB DVR, High‐Degree CSO‐EPVS, and WMH Volume With Hemorrhage Patterns (CMB‐D and cSS‐D Versus Nondominant), Each Model Adjusted for Age and Sex
CMB‐D indicates cerebral microbleed‐dominant; CSO‐EPVS, centrum semiovale enlarged perivascular spaces; cSS‐D, cortical superficial siderosis‐dominant; PiB DVR, Pittsburgh compound B‐distribution volume ratio; and WMH, white matter hyperintensity.
High‐degree CSO‐EPVS significantly differed among groups (P=0.009, Table 1). In post hoc comparisons, both the CMB‐D and cSS‐D groups had significantly higher frequencies of high‐degree CSO‐EPVS compared with the nondominant group (CMB‐D versus nondominant, P=0.018; cSS‐D versus nondominant, P=0.015, each Bonferroni‐corrected), whereas the 2 hemorrhagic groups did not differ from each other (P=1.000). In logistic regression analyses adjusted for age and sex, high‐degree CSO‐EPVS remained independently associated with both the CMB‐D (OR, 9.3 [95% CI, 1.6–52]; P=0.011) and cSS‐D (OR, 10 [95% CI, 1.6–62]; P=0.013) groups relative to nondominant (Table 2).
Although the overall group difference in WMH volume did not reach statistical significance (P=0.055), WMH burden was numerically higher in both hemorrhagic‐dominant groups compared with the nondominant group (Table 1). These results remained nonsignificant after adjustment for age and sex (CMB‐D versus nondominant, P=0.089; cSS‐D versus nondominant, P=0.058).
Patients with high‐degree CSO‐EPVS had significantly higher global PiB DVR than those without (1.46±0.22 versus 1.29±0.25, P=0.007), and this relationship remained independent in a logistic regression model adjusted for age and sex (OR, 1.35 [95% CI, 1.0–1.7]; P=0.016). In a separate linear regression model, global PiB DVR was also independently associated with WMH volume (β=0.25 [95% CI, 0.02–0.94]; P=0.049), again adjusted for age and sex.
All associations remained unchanged in sensitivity analyses that included ICH as an additional covariate (data not shown).
DISCUSSION
In this study, we investigated whether amyloid burden varies among patients with CAA with distinct hemorrhagic patterns and whether these patterns are linked to downstream structural consequences.
We demonstrated that patients with CAA with either CMB‐D or cSS‐D patterns had greater brain amyloid burden, more frequent high‐degree CSO‐EPVS, and numerically higher WMH volumes compared with those without dominant hemorrhagic lesions.
Supporting our main hypothesis, our findings showed that hemorrhage‐dominant patterns (either CMB‐D or cSS‐D) were associated with significantly higher global cortical amyloid load compared with the nondominant group. Previous studies in non‐CAA populations have reported associations between either lobar CMBs or cSS and increased amyloid PET uptake, suggesting that vascular amyloid may contribute to microbleed and siderosis formation.20, 21 Within CAA cohorts, 1 study demonstrated increased regional PiB retention in areas affected by lobar CMBs, although similar data for cSS remain limited.22 Our results extend these observations by demonstrating that both hemorrhagic markers, CMBs and cSS, are associated with elevated global amyloid burden. These findings support models that even within patients with CAA, greater amyloid load promotes brain hemorrhages, which can manifest through diverse lesion types such as microbleeds and siderosis.
We also observed that both hemorrhagic‐dominant groups exhibited a higher frequency of high‐degree CSO‐EPVS, a marker of impaired perivascular clearance, and consistently numerically greater WMH volume, reflecting global ischemic injury in CAA.23, 24 In multivariable models, high‐degree CSO‐EPVS remained independently associated with both hemorrhagic patterns, whereas WMH volume was consistently numerically higher in hemorrhagic‐dominant groups, although these differences did not reach statistical significance. Similar to previous observations, when examined as a continuous measure across the entire cohort, higher PiB uptake was independently associated with both CSO‐EPVS burden and WMH volume, further supporting the hypothesis that these structural markers reflect downstream consequences of vascular amyloid accumulation.9, 19, 25, 26 Taken together, patients with more hemorrhagic lesions also have broader, likely parallel processes of amyloid‐related brain injury, including impaired perivascular clearance and global ischemic injury.
Notably, the only major difference between the cSS‐D and CMB‐D subgroups was that cSS‐D had a substantially higher frequency of ICH, whereas the 2 groups did not differ in PiB uptake or other CAA markers. This singular distinction suggests that the mechanistic pathways leading to ICH may diverge between these patterns, with cSS‐D potentially representing a more ICH‐prone phenotype. This interpretation aligns with prior studies reporting a strong association between cSS and increased ICH risk in CAA.27, 28 We also found that the CMB‐D group had a lower frequency of ICH (51.6%) than the nondominant group (78.6%) despite demonstrating a higher amyloid burden, suggesting that factors beyond global amyloid load influence the likelihood of ICH in CAA. This variability in ICH prevalence across subgroups further supports the rationale for using distinct hemorrhagic markers, rather than ICH itself, to define biologically meaningful CAA phenotypes.
Although our pattern classification also allowed for a fourth group characterized by both severe cSS and high CMB burden (both dominant), only 5 participants (7.4% of the initial cohort) met criteria for this pattern. This group was excluded from primary analyses due to the small sample size. The rarity of this combination in our study aligns with prior reports showing that cSS typically occurs in individuals with relatively lower CMB counts, supporting the idea that extensive cSS and high CMB burden may reflect distinct pathological mechanisms within the broader CAA spectrum.8
One proposed mechanism underlying the distinction between cSS and CMBs involves differences in the vascular compartments affected by amyloid pathology. A prior neuropathological study reported that moderate‐to‐severe cSS was associated with more advanced CAA in leptomeningeal vessels, but lower amyloid severity in cortical vessels.4 In addition, there was no association between the presence of cSS and microbleeds, suggesting that cSS arises primarily from vascular amyloid deposition in leptomeningeal vessels, whereas lobar CMBs may reflect amyloid burden in cortical vessels. This divergence may be further explained by underlying genetic differences. Several studies have linked apolipoprotein E (APOE) ε2 to cSS and APOE ε4 to lobar CMBs, raising the possibility that these genotypes may contribute to distinct vascular injury mechanisms.5, 6, 7 However, these studies are often limited by small sample sizes or meta‐analyses with heterogeneous cohorts, and typically assess cSS and CMBs in isolation. Notably, none included amyloid PET imaging, limiting insights into the interaction between genetic risk, amyloid burden, and hemorrhagic expression.
This study has several limitations. The sample size was modest; classification into 3 patterns further reduced the number of participants per group. Given the modest sample size, the regression models were intentionally kept simple. Some effect estimates, especially for dichotomized variables, may appear large with wide CIs and require careful interpretation. Overall, the findings should be interpreted as exploratory and hypothesis‐generating rather than confirmatory. Because the cohort was drawn from a single tertiary center, generalizability to broader clinical populations might be limited. Still, the study leveraged a well‐characterized CAA cohort with comprehensive multimodal imaging. APOE genotype data were not available, preventing analysis of potential genetic contributions to the observed hemorrhagic phenotypes. Future work integrating genetic data will be important for refining mechanistic insight. Our hemorrhage pattern classification intentionally did not account for the presence/absence of ICH, which may have introduced some heterogeneity in the study cohort. To address this, we repeated the regression models by including ICH as an additional covariate in sensitivity analyses. The results were unchanged, supporting the robustness of our findings. The cross‐sectional design limits causal inference, and longitudinal studies are needed to determine the temporal relationship between amyloid, hemorrhagic markers, and structural injury, an effort currently underway. PiB‐PET might not differentiate between vascular and parenchymal amyloid deposition. A previous neuropathological study in 44 participants, 24 of whom had a pathologic diagnosis of Alzheimer disease, reported that parenchymal plaques, but not regional CAA, were associated with PiB uptake.29 Importantly, that cohort differed substantially from ours; the participants were not clinically symptomatic patients with CAA but predominantly individuals with underlying Alzheimer disease pathology and minimal hemorrhagic markers, making their findings less applicable to a symptomatic CAA population. Still, we aimed to minimize the risk of Alzheimer disease copathology by excluding individuals with mild cognitive impairment or dementia at study enrollment. Finally, despite classifying patients into biologically meaningful subgroups, residual heterogeneity likely exists within each category; however, the simplicity of this classification supports its potential usefulness in both research and clinical settings. Despite these limitations, the consistency of associations across multiple models and sensitivity analyses supports the robustness of our findings and their relevance to understanding CAA mechanisms.
In conclusion, our findings support a close association between vascular amyloid burden and hemorrhagic manifestations of CAA, including a high burden of CMBs and severe cSS. The strength of the PiB–dominance association, combined with the observation that patients with hemorrhage‐dominant patterns also exhibit other amyloid‐linked imaging features such as high‐degree CSO‐EPVS, suggests that these phenotypes may reflect overlapping vascular pathways related to amyloid accumulation. These pathways may include impaired perivascular clearance, vascular dysfunction, and vessel wall weakening. A better understanding of these mechanisms may help identify biologically distinct subgroups with implications for prognosis and therapeutic targeting.
Overall, our results support the candidate treatment approach of targeting amyloid production, with the goal of lowering vascular amyloid burden and thereby reducing the risk for hemorrhagic lesions. This approach is currently being tested in a clinical trial.30
Sources of Funding
This work was supported by the National Institutes of Health, including grants from the National Institutes of Neurological Disorders and Stroke and the National Institute on Aging. Support was provided through awards to Dr Greenberg (R01NS096730, R01AG26484) and Dr Gurol (R01NS114526, K23NS083711).
Disclosures
All authors report no relevant conflicts of interest.