Cerebral Amyloid Angiopathy in the Anti-Amyloid Era

Why Stroke Neurologists Now Own This Problem

For most of its clinical history, cerebral amyloid angiopathy was something a neurologist inferred rather than diagnosed: an older patient with a lobar hemorrhage, a plausible story, and no better explanation. It was a stroke diagnosis, made after the event, with few consequences for management beyond avoiding anticoagulation.

That has changed twice over. First, the Boston criteria version 2.0 turned CAA into a diagnosis that can be made prospectively, from a standard MRI, in a patient who has never bled.¹ Second — and this is what makes the topic urgent for general neurologists rather than only for stroke specialists — the arrival of lecanemab and donanemab has made vascular amyloid the central safety variable in the treatment of Alzheimer's disease. Baseline microbleed count and cortical superficial siderosis now determine eligibility. Amyloid-related imaging abnormalities are the dose-limiting toxicity. And when a patient on one of these drugs arrives in an emergency department with acute focal deficits, the question of whether this is a stroke or ARIA-E has to be answered before thrombolysis is given — a decision with documented deaths behind it.^8,9

The clinical territory this creates does not belong cleanly to either the cognitive or the vascular service. It requires knowing what CAA looks like on MRI, what the inflammatory forms look like when they present as a rapidly progressive encephalopathy, what ARIA is and how it is graded, and how all of this alters ordinary decisions about aspirin, anticoagulation and thrombolysis. This review is organized in that order.

What CAA Is, and How Common

CAA is the deposition of amyloid-β in the walls of cortical and leptomeningeal small arteries and arterioles. The distinction from hypertensive arteriolosclerosis is anatomical and is the basis of every imaging criterion that follows¹⁰: CAA affects superficial vessels and produces lobar bleeding, whereas hypertensive arteriopathy affects deep perforators and produces basal ganglia, thalamic, pontine and cerebellar bleeding.

It is common. In a meta-analysis of 170 studies and more than 73,000 subjects, moderate-to-severe CAA was present in 47.5% of Alzheimer's disease brains and 23.0% of the general older population — and MRI captures under half of what pathology shows.¹¹ That single figure explains much of what follows: roughly half of the patients being considered for anti-amyloid therapy have meaningful vascular amyloid, and a substantial fraction of them will not have an imaging marker to declare it.

The vasculopathy is progressive and stereotyped. Amyloid accumulates in the tunica media and adventitia, smooth muscle cells are lost, and the vessel wall degenerates through fibrinoid necrosis, microaneurysm formation and, in the most advanced grades, the vasculopathic changes that precede rupture. This is why APOE ε2 and ε4 act at different steps: ε4 is associated with greater vascular amyloid deposition — ε4 carriers in the Rotterdam Scan Study had significantly more strictly lobar microbleeds¹² — whereas ε2 is associated with the vasculopathic changes and the tendency to rupture. It also explains why the disease is simultaneously hemorrhagic and ischemic — the same vessels that bleed also fail to perfuse, producing cortical microinfarcts numbering in the hundreds or thousands per brain at a mean diameter of about 200 µm, largely invisible below 7 T.

The clinical consequences fall into three groups. Hemorrhagic: lobar intracerebral hemorrhage, convexity subarachnoid hemorrhage, and transient focal neurological episodes. Cognitive: CAA is an independent contributor to dementia — in one cohort of CAA without ICH, 73% converted to dementia within five years.¹³ And inflammatory: the CAA-ri/ABRA spectrum, discussed below, together with its iatrogenic counterpart, ARIA.

Diagnosing CAA: The Boston Criteria Version 2.0

Version 2.0 was derived and validated across ten North American and European centers in 341 patients with MRI and histopathology, including an autopsy subgroup of 150.¹ The presentations that qualify for the clinical tiers are spontaneous ICH, transient focal neurological episodes, or cognitive impairment and dementia; convexity subarachnoid hemorrhage appears in the presentation clause of the pathology-supported tiers, and counts as a qualifying hemorrhagic lesion throughout. Figure 1 sets out the algorithm; Table 1 gives the criteria as published.

Qualifying presentations for the clinical tiers are spontaneous intracerebral hemorrhage, transient focal neurological episodes, or cognitive impairment and dementia; convexity subarachnoid hemorrhage appears in the presentation clause of the definite and pathology-supported tiers. Derivation and validation were across ten centers, n = 341 with an autopsy subgroup of n = 150; the reference standard was Vonsattel grade ≥ 2 at autopsy or ≥ 1 on biopsy or evacuated hematoma. Predictive values from this design are strongly prevalence-dependent and should not be transported to clinical practice.

What changed, and what it bought

Four things changed from version 1.5. The age threshold fell from 55 to 50 years. Cortical superficial siderosis and convexity subarachnoid hemorrhage were promoted to count as full hemorrhagic lesions, interchangeable with lobar ICH and microbleeds. Two non-hemorrhagic white matter markers were introduced — severe centrum semiovale perivascular spaces and the white matter hyperintensity multispot pattern. And the qualifying presentations were broadened to include transient focal neurological episodes and cognitive impairment.

In the whole sample, sensitivity for probable CAA rose from 70.8% to 79.8% while specificity fell from 88.8% to 84.7%; the area under the curve rose from 0.798 to 0.823 (p = 0.0005). In the autopsy-confirmed subgroup the trade was better still: sensitivity rose from 64.5% to 74.5% with specificity unchanged at 95.0%.¹ Probable and possible CAA combined reach 91.8% sensitivity but only 62.2% specificity — a reminder that possible CAA is a screening category, not a diagnosis.

The markers themselves

Table 2 gives the definitions and the sequence each is read on. Three points deserve emphasis in practice.

Microbleed detection is exquisitely sensitive to acquisition: pulse sequence, sequence parameters, spatial resolution, field strength and post-processing all alter the count,¹⁴ and formal rating scales such as the Microbleed Anatomical Rating Scale exist precisely because of it.¹⁵ The practical consequences are three.

Sequence choice can change the diagnosis. The criteria permit either T2∗-gradient-recalled echo or susceptibility-weighted imaging, at either 1.5 T or 3.0 T, without adjusting the lesion thresholds. Susceptibility-weighted imaging and higher field strength detect more microbleeds. A patient with one lobar microbleed on 1.5 T GRE — possible CAA — may have three on 3 T SWI, which is probable CAA, with no biological change whatever. When following a patient over time, and particularly when following a patient on an anti-amyloid antibody, keep the sequence constant.

Cortical superficial siderosis is the prognostic marker, not microbleed count. Focal siderosis involves three or fewer sulci; disseminated siderosis involves four or more.² In a prospective cohort of 240 CAA-related ICH survivors with meta-analytic pooling, any siderosis carried an adjusted hazard ratio of 2.4 for recurrent ICH and disseminated siderosis 4.4 (95% CI 2.0–9.9).¹⁶ A separate meta-analysis of six studies and 1,239 patients supplies the absolute figures: pooled annual recurrent ICH risk of approximately 3.9% with no siderosis, 9.1% with focal siderosis and 12.5% with disseminated siderosis.¹⁷ The two analyzes differ on whether focal siderosis is independently significant — it was not in the first, and was in the second — so the safest statement is that siderosis extent grades risk and that disseminated siderosis is unambiguously the high-risk phenotype. Lobar microbleed count, by contrast, has repeatedly failed to survive as an independent predictor of recurrence.

Cerebellar lesions are neutral. This is a change from version 1.5 that is widely missed. The v2.0 criteria panel states, in both the probable and possible rows, that a hemorrhagic lesion in the cerebellum is "not counted as either lobar or deep haemorrhagic lesion."¹ A patient with a cerebellar microbleed can therefore still meet probable CAA. Cerebellar hemorrhages can arise from either CAA or hypertensive arteriopathy, which is why version 2.0 treats a cerebellar lesion as neither lobar nor deep. Note the wording carefully: the carve-out is written for the cerebellum specifically, and the word infratentorial does not appear in the paper. Brainstem hemorrhagic lesions, however, are deep and therefore exclude the clinical probable and possible tiers — only the cerebellum is carved out as neutral.

A deep hemorrhagic lesion excludes the clinical Boston 2.0 probable and possible categories, but it does not exclude biological CAA. In one small surgical-pathology cohort of lobar ICH with available MRI, deep-seated microbleeds were present in 3 of 20 patients with histopathologically proven CAA.⁶⁹ In mixed-location hemorrhage and microbleed phenotypes, cortical superficial siderosis appears to separate subgroups: mixed disease without cSS is more often driven by hypertensive small-vessel disease, whereas mixed disease with cSS shows more CAA-type imaging markers and a higher recurrence risk.^70,71 The issue remains an area of active research.

The white matter features do not stack. Severe centrum semiovale perivascular spaces means more than 20 visible spaces in one hemisphere, corresponding to grade 3 or above on the Potter visual rating scale;¹⁸ the multispot pattern means more than 10 small round or ovoid subcortical hyperintensities, bilaterally, on FLAIR. Each counts as "one white matter feature," and the criteria only ever permit one to be used. Two together, with no hemorrhagic lesion, still give possible CAA.

GRE, gradient-recalled echo; SWI, susceptibility-weighted imaging; WMH, white matter hyperintensity; ICH, intracerebral hemorrhage. Only one white matter feature may be counted toward the criteria, so severe centrum semiovale perivascular spaces together with a multispot pattern, in the absence of any hemorrhagic lesion, still yields possible rather than probable CAA.

Presentations that should prompt the diagnosis

Lobar ICH in an older adult is the classic route, and the annual recurrence risk of roughly 7.4% is among the highest in vascular neurology.²

Transient focal neurological episodes are the presentation most often missed, and missing them is dangerous. They are recurrent, stereotyped and brief — usually under 30 minutes — and about half involve positive, aura-like spreading sensory or visual symptoms while the other half look like classical TIAs with negative deficits.¹⁹ The discriminating features are stereotyped recurrence and spread across contiguous cortical territory over minutes, rather than a deficit maximal at onset. The proposed mechanism is cortical spreading depolarization triggered by subarachnoid blood, and the imaging correlate is cortical superficial siderosis. The reason to get this right is stark: in pooled data, 24.5% of patients with these episodes sustained a symptomatic intracerebral hemorrhage within eight weeks.² Treating them as TIAs and starting an antiplatelet is an avoidable harm.

Convexity subarachnoid hemorrhage in an older adult, once aneurysm and trauma are excluded, is CAA until proven otherwise. Incidental lobar microbleeds found on an MRI ordered for another reason increasingly bring patients to attention, and in the anti-amyloid era they arrive with a specific question attached. Before accepting them at face value, exclude the mimics: cavernous malformations with their complete hemosiderin rim, hemorrhagic metastases with surrounding edema, calcification distinguishable on phase imaging, vessel flow voids traceable across slices, and radiation-induced microbleeds distributed along a former treatment field.²⁰

Where the criteria fail

This is the most important caveat in the review and it is systematically under-reported. Version 2.0 was derived in patients with hemorrhagic presentations. In an independent autopsy-validated cohort of 54 patients who were asymptomatic or had cognitive symptoms only, probable CAA by version 2.0 achieved a sensitivity of 28.6%, a specificity of 65.3% and an area under the curve of 0.47 — indistinguishable from chance — and was no better than version 1.5.²¹ The authors' conclusion was blunt: "Boston criteria v2.0 have low accuracy in patients who are asymptomatic or only have cognitive symptoms."

That population is precisely the one being screened before lecanemab or donanemab. The practical implication is not that the criteria should be abandoned but that a negative result in a cognitively impaired, non-hemorrhagic patient carries very little information. Absence of imaging markers does not exclude CAA, and given that MRI detects under half of pathologically confirmed disease,¹¹ it should not be presented to patients as if it did.

When MRI is not possible, and other diagnostic routes

The Boston criteria cannot be applied to CT. For patients presenting with lobar ICH who cannot undergo MRI, the Edinburgh criteria offer a validated CT-based alternative, built on three predictors — subarachnoid hemorrhage, finger-like projections of the hematoma, and APOE ε4 possession — with a c-statistic of 0.92.²² The simplified version, omitting genotyping, rules in CAA on finger-like projections plus subarachnoid hemorrhage with 87% specificity and rules it out on the absence of either with 81% sensitivity, and is what the 2025 international statement endorses for MRI-ineligible patients.²

Cerebrospinal fluid shows a characteristic CAA signature — low Aβ40 and Aβ42 with only modest tau elevation — which separates CAA from healthy controls well (Aβ42/40 area under the curve 0.90) but separates CAA from Alzheimer's disease poorly (0.69).²³ The discriminating axis against Alzheimer's disease is Aβ40 and tau, both lower in CAA, not Aβ42, which is equally low in both. Amyloid PET is sensitive — a negative scan makes probable CAA very unlikely — but only the occipital-to-global uptake ratio carries CAA-specific information, and the reported separation is small. Plasma p-tau217 is higher in the presence of CAA imaging markers (β = 0.29 for CAA presence; β = 0.12 per lobar microbleed count), with roughly half of that association mediated by brain amyloid burden on PET.²⁴ No plasma assay is validated for CAA as an indication; they are validated for amyloid pathology in the cognitive-impairment workup, which is a different question.

Two forms not to miss

Iatrogenic CAA should be considered in any patient under 55 presenting with lobar ICH, convexity subarachnoid hemorrhage, transient focal episodes or unexplained cognitive decline with strictly lobar microbleeds or siderosis. It follows exposure to cadaveric human CNS tissue — most often a cadaveric dura mater graft or childhood neurosurgery — and reflects seeded transmission of amyloid-β. Mean age at presentation is 37.7 years, mean latency from exposure 33.5 years, and mean age at exposure 3.3 years. The most striking diagnostic feature is that reported cases have almost all carried at least one APOE ε3 allele (about two-thirds ε3/ε3), without the ε2 or ε4 enrichment of sporadic disease.²⁵ The question to ask is simply whether the patient had brain surgery as a child.

Hereditary CAA should be considered in younger patients with a family history. Dutch-type (APP E693Q) presents with recurrent hemorrhage in the fifth and sixth decades; Iowa-type (APP D694N) presents as an autosomal-dominant dementia in the sixth or seventh decade, with widespread tangles and occipital cortical calcification.²⁶ Notably, in Dutch-type carriers the version 2.0 white matter features appear very early — the multispot pattern was seen at age 30 and severe centrum semiovale perivascular spaces at 29 — and version 2.0 raised the detection of possible CAA in presymptomatic carriers from 0% to 43%.²⁷ That is the strongest biological validation of the new markers, and it also exposes the limitation that the ≥50-year age threshold would exclude these patients entirely.

The Inflammatory Spectrum: CAA-ri and ABRA

A minority of patients with CAA develop an inflammatory reaction against the vascular amyloid. Historically two entities were described: CAA-related inflammation, with a perivascular, non-destructive lymphoplasmacytic infiltrate, and amyloid-β-related angiitis, with vessel-wall damage, marked T-cell infiltration and granulomas.²⁸ The distinction from primary angiitis of the CNS is that both show vascular amyloid, which primary angiitis does not.^29,28 The 2025 international statement has settled the nomenclature: "the term CAA-related inflammation (CAA-ri) is preferred for all cases, with the terms amyloid-beta related angiitis (ABRA) or CAA-related vasculitis reserved (as an additional descriptor) only where there is clear pathological evidence of an angio-destructive vasculitis."² In other words, one disease with a severity gradient, and ABRA as a histological descriptor at its severe end (Figure 2).

CAA-ri is rare — population-based incidence in Northern Ireland was approximately one per million per year³⁰ — but it is treatable, and it is the differential diagnosis that most often gets missed in a patient presenting with subacute cognitive decline and confusing white matter change.

Recognizing it

The syndrome is subacute encephalopathy in an older patient: headache, decline in consciousness, behavioral change, focal deficits and seizures, evolving over days to weeks. It is regularly mistaken for rapidly progressive dementia, posterior reversible encephalopathy syndrome, tumor, autoimmune encephalitis or stroke. Mean age is in the low seventies.

The imaging is the giveaway once you know to look: asymmetric, patchy or confluent white matter hyperintensity extending into the immediately subcortical white matter, often with mass effect, on a background of lobar microbleeds or cortical superficial siderosis. Leptomeningeal enhancement is common. The most useful recent numbers come from a comparison of 37 patients with CAA-ri against 158 with sporadic non-inflammatory CAA: parenchymal white matter hyperintensity in 89% versus 3%, sulcal hyperintensity in 78% versus 9%, gyral swelling in 86% versus 0.6%, and a median lobar microbleed count of 207 (IQR 33–811) versus 19 (IQR 7–58).³¹ A CAA patient with hundreds of microbleeds and asymmetric edema is a different clinical animal from one with a handful.

The distinction from primary angiitis of the CNS is age and imaging. In the largest comparison, CAA-ri patients had a mean age of 73 against 45 for biopsy-positive primary angiitis; at least one lobar microbleed was present in 94% versus 26%, cortical superficial siderosis in 27% versus 4%, and leptomeningeal enhancement in 70% versus 27%, while non-ischemic parenchymal gadolinium enhancement ran the other way at 16% versus 82%.

The first clinicoradiological criteria for CAA-ri were proposed by Chung and colleagues in 2011;³² the Auriel criteria, which modified them by requiring the white matter change to extend to the immediately subcortical white matter and by adding cortical superficial siderosis to the qualifying hemorrhagic lesions, remain as of mid-2026 the only formally validated set. They were derived in 17 pathologically confirmed cases against 37 non-inflammatory CAA controls, giving probable CAA-ri a sensitivity of 82% and a specificity of 97%.³³ Two caveats belong in any clinical use of them. The control group was CAA, not the realistic differential of tumor, PRES and encephalitis, so specificity against those mimics is untested. And possible CAA-ri has the same 82% sensitivity as probable while losing 29 percentage points of specificity, down to 68% — it buys nothing and should not by itself justify committing an older patient to months of immunosuppression.

Cerebrospinal fluid and biopsy

CSF helps mainly to exclude infection and malignancy. Raised protein is the rule — present in about 90% — but a lymphocytic pleocytosis is found in only around a quarter of cases, and CSF is entirely normal in a few per cent, so an unremarkable result does not exclude the diagnosis. The much-discussed CSF anti-Aβ autoantibody, reported to be elevated acutely and to normalize in remission,³⁴ has not been replicated in a subsequent series,³⁵ and the 2025 international statement records that the assay is "not validated for routine clinical diagnostic use."² It is not validated or routinely available for clinical diagnosis. Do not wait for it.

Brain biopsy is no longer routinely required when the clinicoradiological criteria are met. It remains indicated when the diagnosis is genuinely in doubt or when the patient fails to respond — and the criteria paper offers a usable interval, suggesting that "a reasonable follow-up approach would be to consider brain biopsy in empirically treated patients who fail to respond to corticosteroid therapy within 3 weeks."³³ That is an expert suggestion rather than a validated threshold; the 2025 international statement gives no interval, saying only that biopsy "should be considered after evaluating the risks and benefits" where there is diagnostic doubt or a lack of response to first-line immunotherapy.² It is not a benign procedure and its yield in this setting is instructive: in a series of 79 biopsies for suspected CNS vasculitis, primary angiitis was confirmed in 11% and an alternative diagnosis found in 30% — CAA being the commonest — with complications in 16% and serious complications in 4%.³⁶

Treatment

There has never been a randomized controlled trial of any therapy in CAA-ri or ABRA. The entire evidence base is one prospective multicenter cohort of 113 patients, retrospective series up to 104, a single treated-versus-untreated comparison of 48, and a pooled systematic review of 553.^6,37 Every regimen below is a reported practice pattern, not a validated protocol, and the treated-versus-untreated comparison is acknowledged by its own authors to be vulnerable to confounding by indication.

With that stated, the observational signal is strong and consistent. In the treated-versus-untreated comparison, clinical improvement occurred in 94% of treated versus 50% of untreated first episodes (odds ratio 16.0, 95% CI 2.72–94.1; p = 0.002), and radiographic improvement in 86% versus 29%.³⁷ In the prospective cohort, Kaplan–Meier clinical recovery reached 70.3% at three months, 80.2% at six and 84.1% at twelve.⁶ Pooled favorable outcome across 553 patients was 75%.³⁸

Three practical points follow from those data. First, clinical recovery outpaces radiological recovery — 70% versus 45% at three months — so do not escalate immunosuppression on the basis of residual FLAIR signal alone. Second, spontaneous improvement occurs in about half of untreated patients, which is why uncontrolled response rates overstate steroid efficacy. Third, and most actionable, never stop the intravenous pulse without an oral taper: abrupt discontinuation carried a hazard ratio of 4.68 (95% CI 1.57–13.93; p = 0.006) for recurrence.⁶

Relapse is the rule rather than the exception over time — approximately 16% at one year, 38% at two years and 38–40% at three — with a median time to recurrence of 5.2 months.^6,37 Every one of the fifteen relapsing patients in the prospective cohort had more than 10 microbleeds at baseline, which is a usable flag for closer follow-up. Seizures at presentation predicted poorer recovery (odds ratio 0.17, 95% CI 0.04–0.63).

Mortality in the largest prospective cohort was 13 of 113 over a median 12 months, and the leading cause of death was intracerebral hemorrhage (7 of 13), not the inflammation.⁶ That is the sentence to carry away: treating the inflammation does not treat the angiopathy. The underlying CAA does not remit, new ICH occurred in 7.1% within three months, and long-term management reverts to the CAA principles in the next section — with a blood-pressure target of ≤130/80 mm Hg.

As for whether ABRA should be treated more aggressively: in practice it often is, but the evidence is weak and partly contradictory. A pooled review of 139 published cases reported that ABRA more often required combination drug therapy,³⁹ and the largest ABRA-specific literature review assembled 94 cases with a mean age at diagnosis of 65 years;⁴⁰ but others find no outcome difference between glucocorticoids alone and combination therapy, and response to glucocorticoid monotherapy is already high. The 2025 statement provides a single unified algorithm and does not specify a different first-line regimen for ABRA. The defensible position is to treat all cases as CAA-ri and reserve escalation for limited response, poor tolerability or relapse.

ARIA: Amyloid-Related Imaging Abnormalities

ARIA was named in 2011 by an Alzheimer's Association Research Roundtable workgroup convened in response to signal changes emerging in trials of amyloid-lowering agents, principally bapineuzumab.⁴¹ A 2026 update from a reconvened Alzheimer's Association workgroup addresses ARIA in routine practice rather than in trials, and explicitly notes that its recommendations do not constitute a clinical practice guideline.⁴² The spectrum comprises ARIA-E — FLAIR hyperintensity representing vasogenic edema and sulcal effusion, usually transient — and ARIA-H — hypointensity on T2∗-GRE or SWI representing new microhemorrhages and superficial siderosis, generally permanent.

Why ARIA belongs in a review about CAA

The mainstream mechanistic account is that anti-amyloid antibodies mobilize parenchymal amyloid into perivascular drainage pathways and bind directly to vascular amyloid, driving inflammation and vascular disruption — with CAA as the substrate.^5,43 Steven Greenberg has put the conclusion plainly: ARIA-E is iatrogenic CAA-ri. The 2025 international statement reaches the same position, noting that the shared frequency of APOE ε4 homozygosity, the dependence on baseline CAA burden and the dose relationship together suggest "that ARIA-E might be an 'iatrogenic' form of CAA-ri."² Supporting the mechanism, antibodies that bind fibrillar amyloid most avidly produce the highest ARIA rates, while soluble-Aβ-targeting antibodies produce almost none — ARIA-E as low as 0.9% with solanezumab and 0.3% with crenezumab.⁴³

The strongest human evidence is a single autopsy. A 79-year-old APOE ε4 homozygote with four baseline microhemorrhages, entering the open-label extension of the lecanemab trial, developed headache after each of three infusions, then speech arrest and a generalized convulsion. Repeat MRI showed more than 30 microhemorrhages and multifocal swelling. She died five days after admission despite pulsed methylprednisolone. Autopsy showed severe CAA with perivascular lymphocytic infiltrate, reactive macrophages and fibrinoid degeneration of vessel walls — histologically indistinguishable from ABRA.⁴ This case simultaneously illustrates every recognized risk factor: ε4 homozygosity, baseline microbleeds, the early treatment window, and concurrent anticoagulation.

How common, and in whom

With lecanemab in the pivotal CLARITY-AD trial,⁴⁴ any ARIA occurred in 21%, ARIA-E in 13%, ARIA-H in 17%, symptomatic ARIA in 3%, and intracerebral hemorrhage larger than 1 cm in 0.7%. With donanemab in the pivotal TRAILBLAZER-ALZ 2 trial,⁴⁵ the corresponding figures were 36%, 24%, 31% and 6%. These are the original pivotal-dosing figures; a modified titration schedule subsequently reduced the ARIA-E rate and is reflected in the 2025 label. Rates rise steeply with APOE ε4 dose: any ARIA on lecanemab was 13% in non-carriers, 19% in heterozygotes and 45% in ε4 homozygotes; on donanemab, 25%, 36% and 55%.^46,47

The best risk-stratification dataset is a pooled multivariable analysis from the donanemab program, giving adjusted odds ratios for ARIA-E of 4.57 for ε4 homozygosity, 2.03 for heterozygosity, 2.53 for a baseline microbleed count of 2–4, 2.18 for cortical superficial siderosis, and 1.73 for a mean arterial pressure of 107 mm Hg or above — with antihypertensive use independently protective at 0.58.⁹ Two things are worth pulling out. Baseline microbleeds raise risk at a count of 2–4, which is still within label eligibility, so being eligible is not the same as being low risk. And blood pressure is a modifiable risk factor for ARIA, which is under-appreciated.

Real-world rates from three cohorts are broadly concordant with trial data — approximately 19–25% any ARIA, 12% ARIA-E, and 1–4.5% symptomatic ARIA — and an early real-world academic health-system series reported no ARIA-related deaths.⁴⁸ One natural experiment is worth noting: a center that excluded APOE ε4 homozygotes per European practice reported any ARIA of 18.6% and symptomatic ARIA of about 1%.⁴⁹

Screening, surveillance, and the 2025 change

Baseline exclusions are consistent across appropriate-use documents: more than four microhemorrhages, any area of cortical superficial siderosis, a prior macrohemorrhage above 10 mm, baseline vasogenic edema, more than two lacunar infarcts or a stroke in a major vascular territory, severe white matter hyperintensity (Fazekas 3), and MRI evidence of ABRA or CAA-ri.⁵⁰ The donanemab recommendations specify more than four microbleeds, cortical superficial siderosis, and "a major vascular contribution to cognitive impairment" as exclusions, and require a pre-treatment MRI obtained no more than 12 months before starting.⁵¹ Anticoagulation is an exclusion in the lecanemab recommendations, and the development of a new anticoagulant indication is a criterion for permanent discontinuation.

The surveillance schedule changed materially in 2025. On 28 August 2025 the FDA issued a Drug Safety Communication adding an earlier MRI before the third lecanemab infusion, citing six fatal ARIA-E cases and an analysis of 101 serious ARIA-E events by inter-infusion interval: 2% between the second and third infusions, 22% between the third and fourth, 40% between the fourth and fifth, and 36% thereafter.³ The current schedule for lecanemab is a baseline scan and then MRI at approximately one, two, three and six months of treatment; for donanemab, before the second, third, fourth and seventh infusions.

One detail from the FDA analysis deserves more attention than it has received: all 24 serious cases occurring before the fourth infusion presented with symptoms that prompted an urgent scan — they were not detected by scheduled imaging. Symptom vigilance, and a mechanism for patients to reach the service quickly, matter at least as much as the imaging calendar.

Recognition and management

Roughly three-quarters of ARIA-E is asymptomatic and found on surveillance imaging. When it is symptomatic, the mild phenotype is headache, confusion, dizziness, visual change, nausea and gait disturbance. The severe phenotype is different in kind: in a systematic review of 36 published severe symptomatic cases, altered consciousness occurred in 75%, focal neurological deficits in 66.7%, seizures in 38.9% and headache in only 36.1%.⁵² That is an encephalopathy, and it is the phenotype that produces the stroke-mimic problem.

The consensus imaging protocol from the American Society of Neuroradiology specifies three sequences — T2∗-gradient-recalled echo for ARIA-H, FLAIR for ARIA-E, and diffusion-weighted imaging to exclude acute ischemia — at 4 mm slice thickness, preferably at 3 T and never below 1.5 T.⁵³ Figure 3 and Table 3 set out the severity grading and the dosing decision. Two technical points are easy to get wrong. ARIA-E severity is graded by the single greatest dimension of the FLAIR abnormality, with multifocality upgrading mild to moderate. And ARIA-H counts are new incident lesions relative to the previous scan, not cumulative lifetime burden — a frequent source of confusion, and another reason to keep the blood-sensitive sequence constant.

Thresholds are those given in the current prescribing information for both approved anti-amyloid monoclonal antibodies. ARIA-H counts refer to new incident lesions relative to the immediately preceding MRI, not cumulative burden. Appropriate-use recommendations advise permanent discontinuation in a wider set of circumstances than the labels require: severe radiographic or symptomatic ARIA, any macrohemorrhage, one area of superficial siderosis, more than 10 microhemorrhages since starting treatment, more than two ARIA episodes, or the development of a condition requiring anticoagulation.

On corticosteroids, the review has to be honest. High-dose intravenous methylprednisolone — 1 g daily for five days followed by an oral taper — is recommended for severe or symptomatic ARIA by appropriate-use consensus, and was given in roughly three-quarters of published severe cases. No controlled data support its efficacy. The recommendation is extrapolated from spontaneous CAA-ri. In the published severe series, outcomes with corticosteroid treatment in 72% were complete recovery in 15, persistent deficits in 10 and death in 10 — a case fatality of about 28% once ARIA is severe and symptomatic, though publication bias toward severe outcomes must be acknowledged.⁵² The FDA labels themselves recommend only dose suspension and re-imaging; they do not mention corticosteroids at all. One further trap: the corticosteroid doses that appear in the Alzheimer's Association lecanemab toolkit — dexamethasone 0.75 mg/day or methylprednisolone 80 mg twice daily — are for infusion-related reactions, not ARIA, and must not be used as an ARIA regimen.

ARIA and the acute stroke service

This is the highest-yield section for a neurology readership and the one with the sharpest evidence gap. Both boxed warnings now carry the same sentence, differing only in the drug named: "Because ARIA-E can cause focal neurologic deficits that can mimic an ischemic stroke, treating clinicians should consider whether such symptoms could be due to ARIA-E before giving thrombolytic therapy in a patient being treated with [LEQEMBI / KISUNLA]." The lecanemab label adds that "fatal cerebral hemorrhage has occurred in a patient taking an anti-amyloid monoclonal antibody in the setting of focal neurologic symptoms of ARIA and the use of a thrombolytic agent."

At least two deaths across the class are directly linked to thrombolysis given during ARIA-E — the index case on lecanemab, in which autopsy showed histiocytic vasculitis with phagocytosis of vascular amyloid,⁸ and one in the donanemab program.⁹ There is no registry, prospective series or denominator. Any statement about the magnitude of thrombolysis risk in this population would be invented; the honest framing is two published deaths, no denominator, and a formal advisory recommending caution.

That advisory — the 2025 American Heart Association science advisory on vascular neurology considerations for anti-amyloid immunotherapy — is the operative document, and its key statements are worth quoting.⁷ On thrombolysis: "Pending further data demonstrating the safety of thrombolysis in this setting, treating clinicians might reasonably consider a high level of caution or avoidance of thrombolytic use in patients receiving immunotherapy." On thrombectomy: "Mechanical thrombectomy without thrombolytics for patients with established clinical indications is likely safe and should be performed." On anticoagulation: "Anticoagulation likely increases the risk of severe ARIA in the form of symptomatic ICH." On antiplatelets, reassuringly: "Antiplatelet monotherapy appears to be reasonably well-tolerated with immunotherapy." And on preparedness: "The considerations discussed here and methods for identifying individuals being treated with amyloid immunotherapy should be disseminated across the acute stroke team."

Note what these are — considerations, not graded recommendations, carrying no class or level of evidence. The acute stroke guideline goes further, though not in its numbered recommendations. The 2026 AHA/ASA guideline addresses this population in its Table 8, which grades contraindications to thrombolysis by an expert-opinion risk gradient rather than by class of recommendation, and which lists amyloid-related imaging abnormalities among the absolute contraindications: "The risk of thrombolysis related ICH in patients on amyloid immunotherapy or with ARIA is unknown and IV thrombolysis should be avoided in such patients."⁵⁴ The same table lists prior intracerebral hemorrhage as a relative contraindication and singles this population out within it: "Patients with known amyloid angiopathy may be considered as having higher risk than patients with ICH due to modifiable conditions (e.g. HTN, coagulopathy)… Treatment should be determined on an individual basis."⁵⁴ Neither statement carries a class or level of evidence, and neither drug is named anywhere in the guideline.

A companion review of the stroke-care implications reaches the same conclusions, adding that anticoagulation was exclusionary in the aducanumab but not the lecanemab or donanemab trials, that antiplatelet therapy in the lecanemab trial was associated with lower rates of ARIA-E, microhemorrhage and siderosis than lecanemab alone, and that ARIA-H risk is highest in the first six months — the window in which antithrombotics are best avoided if there is any discretion.⁵⁵ The operational conclusion for a stroke service is a short list. Know which of your patients are on these drugs — an electronic flag or a patient-held card is the practical implementation. Get MRI with FLAIR and DWI when the presentation is compatible with ARIA-E; DWI is what separates it from acute ischemia. Apply a high threshold for intravenous thrombolysis. Prefer mechanical thrombectomy where an accessible occlusion exists. Continue antiplatelet monotherapy where indicated. Avoid anticoagulation. And treat the blood pressure.

Table 4 sets the four entities side by side. The distinction that matters clinically is not pathological but operational: CAA is a substrate whose risk you manage, ARIA is a drug effect whose first move is to stop the drug, and CAA-ri is an illness you treat with corticosteroids — while ABRA is a histological label for the severe end of CAA-ri rather than a separate disease requiring a different first-line regimen.

The clinically useful distinction is operational rather than pathological. CAA is a substrate whose risk is managed; ARIA is a drug effect whose first move is to stop the drug; CAA-ri is an illness treated with corticosteroids; and ABRA is a histological descriptor at the severe end of CAA-ri, treated the same way. All four share the same underlying vasculopathy, and a patient may move between them — most obviously the patient with CAA who is given an anti-amyloid antibody.

Managing the Angiopathy Itself

Once CAA is diagnosed, four decisions recur: antiplatelets, anticoagulation, acute reperfusion therapy, and blood pressure. Figure 4 summarizes them and Table 5 gives the supporting evidence.

Antiplatelet therapy

RESTART randomized 537 survivors of intracerebral hemorrhage who had been taking an antithrombotic to start or avoid antiplatelet therapy, and found an adjusted hazard ratio for recurrent ICH of 0.51 (95% CI 0.25–1.03).⁵⁶ The imaging substudy of 254 participants is the analysis that matters here,⁵⁷ and it found no effect modification by microbleed burden or lobar distribution — every stratum, including five or more microbleeds and a strictly lobar distribution, had a point estimate favoring antiplatelet therapy. Cortical superficial siderosis was examined and likewise showed no interaction, although the subgroup was small and exploratory. A more recent trial of antiplatelets after ICH was terminated early at 69 of a planned 500 patients and is uninformative.

The 2025 international statement translates this into two positions worth memorizing. For primary prevention in probable CAA — no prior ischemic stroke or established vascular disease — antiplatelet therapy is not recommended, and this is a strong recommendation. Where there is a genuine indication after CAA-related ICH, "antiplatelet monotherapy may be safe and may be considered."²

Anticoagulation in atrial fibrillation

This is the hardest decision in the field and 2025 sharpened it rather than resolving it. PRESTIGE-AF randomized 319 ICH survivors with atrial fibrillation to a direct oral anticoagulant or none.⁵⁸ Ischemic stroke fell dramatically — hazard ratio 0.05 (95% CI 0.01–0.36), one event versus twenty — while recurrent ICH rose roughly tenfold (hazard ratio 10.89, 95% CI 1.95–60.72). Net clinical benefit was neutral (hazard ratio 0.69, 95% CI 0.33–1.40). No subgroup analysis by lobar location or CAA phenotype was reported in the main paper, though a secondary analysis identified cortical superficial siderosis as a predictor of recurrent ICH with a hazard ratio of 7.7 (95% CI 1.4–42.2), while lobar location and a probable-CAA label were not predictive.

For patients with microbleeds who have never bled, the calculus is different and much more permissive. In CROMIS-2, symptomatic ICH on anticoagulation ran at 9.8 per 1000 patient-years with microbleeds versus 2.6 without (adjusted hazard ratio 3.67)⁵⁹ — roughly 1% per year in patients with any microbleeds, an order of magnitude below the recurrence rate after an actual CAA-related lobar hemorrhage. A large collaborative analysis of 7839 patients with atrial fibrillation found microbleed presence associated with both intracranial hemorrhage (adjusted hazard ratio 2.74) and ischemic stroke (1.29) — a marker of a bleeding and ischemic phenotype, not a purely hemorrhagic one.⁶⁰ Intracranial hemorrhage rates exceeded ischemic stroke rates at 2–4 microbleeds (25 versus 12 per 1000 patient-years) and at 11 or more (94 versus 48), but these figures come from patients on anticoagulant plus antiplatelet, not anticoagulant alone — a distinction routinely lost in citation, and one that means the data identify no clean microbleed threshold for anticoagulant monotherapy.

Two earlier randomized datasets frame the same tension. SoSTART, a pilot non-inferiority trial in 203 patients, found recurrent intracranial hemorrhage in 8% versus 4% with an adjusted hazard ratio of 2.42 (95% CI 0.72–8.09) and was inconclusive;⁶¹ a pooled individual-participant meta-analysis of four trials and 412 participants found a hazard ratio of 0.68 (95% CI 0.42–1.10) for the composite of any stroke or cardiovascular death, driven by a clear reduction in ischemic events (0.27, 95% CI 0.13–0.56).⁶² Two trials that might have settled the CAA question will not. ENRICH-AF explicitly excludes lobar intraparenchymal hemorrhage and isolated convexity subarachnoid hemorrhage, so it will generate no CAA-applicable evidence. ASPIRE includes lobar ICH with no location exclusion and remains in recruitment. Until it reports, the operative rule from the international statement is that disseminated cortical superficial siderosis, not microbleed count, is the dominant contraindication.

Thrombolysis and thrombectomy

The 2026 AHA/ASA guideline made one genuinely new statement that is directly useful: in patients eligible for thrombolysis within 4.5 hours whose microbleed burden is unknown, treatment should be given "without first obtaining MRI to exclude CMBs" (Class 1, Level B-NR) — the point being not to delay treatment in order to look. With 1–10 microbleeds already demonstrated, thrombolysis is reasonable (Class 2a); above 10, its usefulness is uncertain (Class 2b). The 2019 guideline classified thrombolysis in a patient with a history of intracranial hemorrhage as potentially harmful (Class 3: Harm, Level C-EO).⁶³ That recommendation has been dropped. In the 2026 document, prior intracerebral hemorrhage appears not among the graded recommendations at all but in Table 8, as a relative contraindication to be settled case by case — and the same entry states that patients with known amyloid angiopathy "may be considered as having higher risk than patients with ICH due to modifiable conditions."⁵⁴ The direction of travel is toward individualized decision-making rather than prohibition, with CAA explicitly flagged as the higher-risk subgroup.

Where there is a large-vessel occlusion, the international statement is explicit: "In patients with probable CAA and an acute ischemic stroke due to large vessel occlusion, endovascular thrombectomy without thrombolysis is the preferred treatment option."² That preference is supported by a meta-analysis of 2051 patients showing that five or more microbleeds raised radiographic hemorrhage after thrombectomy (any ICH odds ratio 2.58, parenchymal hematoma type 2 odds ratio 5.33) but not adjudicated symptomatic ICH.

Blood pressure — the most under-sold intervention

In 1145 ICH survivors followed for a median of three years, inadequate blood-pressure control during follow-up carried a hazard ratio of 3.53 (95% CI 1.65–7.54) for recurrent lobar hemorrhage.⁶⁴ That effect size is comparable to disseminated cortical superficial siderosis — and unlike siderosis, it is modifiable. The 2022 AHA/ASA intracerebral hemorrhage guideline reaches the same target for long-term management after any spontaneous ICH.⁶⁵ The 2025 international statement makes it a strong recommendation for CAA specifically: "blood pressure should be regularly monitored to maintain a long-term target of ≤130/80 mm Hg to reduce risk of ICH recurrence," with home monitoring encouraged to reduce variability.² The same lever appears in the ARIA literature, where antihypertensive use was independently protective (odds ratio 0.58).⁹ A neurologist who does only one thing after diagnosing CAA should do this one.⁶⁶

Four Scenarios

The evidence above resolves into a small number of recurring bedside problems. Each of these is stated with the numbers that are actually available, and with an explicit note where none exist.

A 78-year-old with atrial fibrillation and six lobar microbleeds, who has never bled

This is the commonest version of the dilemma and it is less fraught than it feels. The relevant denominator is not the 7.4% annual recurrence rate that follows an actual CAA-related lobar hemorrhage — that patient has never had one. It is CROMIS-2, where symptomatic intracranial hemorrhage on anticoagulation ran at 9.8 per 1000 patient-years in patients with microbleeds against 2.6 without: roughly 1% per year in patients with any microbleeds.⁵⁹ For most patients with a meaningful CHA₂DS₂-VASc score, the annual ischemic stroke risk without anticoagulation exceeds that. Three modifiers matter. Look specifically for cortical superficial siderosis, which is the marker the international statement treats as the dominant contraindication. Avoid combining an anticoagulant with an antiplatelet unless there is a separate compelling indication, because the published crossover to net harm at high microbleed burden comes from patients on both. And treat the blood pressure to 130/80. Left atrial appendage occlusion is raised in the literature as an alternative in this situation, though without comparative trial data in a CAA population.

A patient with probable CAA who needs a coronary stent and dual antiplatelet therapy

There is no risk estimate for dual antiplatelet therapy in CAA. RESTART studied antiplatelet monotherapy; the international statement addresses monotherapy; no trial has enrolled CAA patients onto dual therapy. Saying so plainly is better than manufacturing a number. What can be said is that the recommendation for monotherapy does not extend to dual therapy, that the duration of dual therapy is the variable most amenable to negotiation with cardiology, and that a shorter course with earlier de-escalation to monotherapy is the reasonable compromise. Disseminated cortical superficial siderosis should raise the threshold further. Document the discussion, because this is a decision made without evidence on either side.

A patient with a prior lobar hemorrhage who presents with a large-vessel occlusion

Here the guidance is unusually clear. The international statement holds that in probable CAA with a large-vessel occlusion, "endovascular thrombectomy without thrombolysis is the preferred treatment option." The supporting safety data are reassuring: in a meta-analysis of five studies and 2051 patients,⁶⁷ five or more microbleeds raised radiographic hemorrhage after thrombectomy — any intracranial hemorrhage odds ratio 2.58, parenchymal hematoma type 2 odds ratio 5.33 — but did not raise adjudicated symptomatic hemorrhage.⁶⁷ A prior intracranial hemorrhage was classified in the 2019 guideline as making intravenous thrombolysis potentially harmful; the 2026 guideline drops that recommendation and reclassifies prior ICH as a relative contraindication, while singling out known amyloid angiopathy as carrying higher risk than hemorrhage from modifiable causes.⁵⁴ Go straight to the catheter.

A patient on lecanemab who arrives with acute hemiparesis and dysphasia

Assume ARIA-E until imaging says otherwise. Both boxed warnings require it, deaths across the class have followed thrombolysis in this setting, and the severe ARIA phenotype is dominated by focal deficits and depressed consciousness rather than headache. The discriminating investigation is MRI with FLAIR and DWI — FLAIR shows the edema, DWI shows or excludes the infarct. The MRI is not a clearance for thrombolysis: the 2026 AHA/ASA table advises avoiding intravenous thrombolysis in any patient on amyloid immunotherapy or with ARIA whatever the FLAIR shows,⁵⁴ so its role here is to separate ARIA from infarction and to route a thrombectomy-eligible patient to EVT rather than to license lysis. If a thrombectomy-eligible occlusion is present, proceed mechanically; the AHA advisory states that thrombectomy without a thrombolytic "is likely safe and should be performed." If there is no accessible occlusion and the imaging shows ARIA-E rather than infarction, treat the ARIA: suspend the antibody, consider high-dose corticosteroids while acknowledging they are unsupported by controlled data, and manage seizures, which occur in nearly 40% of severe cases. The logistical prerequisite for all of this is knowing, at the front door, that the patient is on an anti-amyloid antibody at all.

AF, atrial fibrillation; HR, hazard ratio; OR, odds ratio. Recommendation strengths (strong / weak) are those of the 2025 International CAA Association / World Stroke Organization scientific statement; classes of recommendation and levels of evidence are from the 2026 AHA/ASA acute ischemic stroke guideline.

What We Do Not Know

A review of this territory is obliged to be explicit about its holes. There has never been a randomized trial of any therapy in CAA-ri or ABRA, and no agent-specific response rate exists for cyclophosphamide, mycophenolate, azathioprine, rituximab, intravenous immunoglobulin or plasma exchange — the 2025 statement declares the last three of "unknown" efficacy. No validated risk-prediction model or nomogram for ARIA exists, despite the field's obvious appetite for one. There is no controlled evidence that corticosteroids improve ARIA outcomes. There is no cohort, registry or series describing outcomes of thrombolysis in patients on anti-amyloid antibodies — only individual deaths. No published cohort describes anti-amyloid therapy in patients with a prior history of CAA-ri, because they are excluded. And the Boston criteria have not been shown to perform adequately in the non-hemorrhagic, cognitively symptomatic population in which they are now most often applied.

Two nosological points are also unsettled. Newer "CAA-ri spectrum" criteria were proposed in 2025,⁶⁸ but no external validation of them has been identified, and the Auriel criteria remain the only validated set.³³ And the specific antibody mechanism linking CAA-ri and ARIA is contested: the imaging, genetic and pathological convergence is strong, but the CSF anti-Aβ autoantibody finding has not replicated, and complement and myeloid data point toward an innate rather than a humoral mechanism.

Conclusion

Cerebral amyloid angiopathy has become a diagnosis that general neurologists make, not one that pathologists confirm. The Boston criteria version 2.0 make that possible from an ordinary MRI, provided the protocol includes a blood-sensitive sequence and the lesions are counted consistently. Their weakness — near-chance performance in non-hemorrhagic, cognitively symptomatic patients — happens to fall exactly where the anti-amyloid era now asks the most questions, and that limitation should be stated to patients rather than glossed.

The inflammatory forms are rare, treatable and frequently missed. Recognizing the asymmetric subcortical edema on a background of lobar microbleeds, applying the Auriel criteria, and starting corticosteroids early converts a devastating encephalopathy into a condition with roughly 70% clinical recovery at three months — provided the intravenous pulse is followed by a proper oral taper. But treating the inflammation does not treat the angiopathy: the commonest cause of death in these patients remains intracerebral hemorrhage.

ARIA may well be the same reaction with a treatment-induced trigger — a hypothesis the 2025 statement raises and the autopsy literature supports, but which is not yet established. Most ARIA is asymptomatic and benign; the severe phenotype is an encephalopathy with a case fatality around one in four in published series, and it can present as a stroke. The single most useful thing a stroke service can do is to know which of its patients are receiving these antibodies, and to have decided in advance that such a patient with acute focal deficits gets an MRI with DWI before a thrombolytic is considered, and a mechanical route where one exists.

Finally, amid a literature dominated by difficult antithrombotic trade-offs, the best-evidenced and most modifiable intervention in CAA remains blood-pressure control. A hazard ratio of 3.5 for recurrent lobar hemorrhage, and a target of 130/80, are worth more clinical attention than they usually receive.

References

1. Charidimou A, Boulouis G, Frosch MP, et al. The Boston criteria version 2.0 for cerebral amyloid angiopathy: a multicentre, retrospective, MRI-neuropathology diagnostic accuracy study. Lancet Neurol. 2022;21(8):714–725.
2. Cordonnier C, Klijn CJM, Smith EE, et al. Diagnosis and management of cerebral amyloid angiopathy: a scientific statement from the International CAA Association and the World Stroke Organization. Int J Stroke. 2025;20(8):949–967.
3. US Food and Drug Administration. Drug Safety Communication: FDA recommends additional MRI monitoring earlier in treatment with Leqembi (lecanemab-irmb). 28 August 2025.
4. Solopova E, Romero-Fernandez W, Harmsen H, et al. Fatal iatrogenic cerebral β-amyloid-related arteritis in a woman treated with lecanemab for Alzheimer's disease. Nat Commun. 2023;14:8220.
5. Greenberg SM, Bax F, van Veluw SJ. Amyloid-related imaging abnormalities: manifestations, metrics and mechanisms. Nat Rev Neurol. 2025;21:193–203.
6. Antolini L, DiFrancesco JC, Zedde M, et al. Spontaneous ARIA-like events in cerebral amyloid angiopathy-related inflammation: a multicenter prospective longitudinal cohort study. Neurology. 2021;97(18):e1809–e1822.
7. Greenberg SM, Aparicio HJ, Furie KL, et al.; American Heart Association Stroke Council. Vascular neurology considerations for antiamyloid immunotherapy: a science advisory from the American Heart Association. Stroke. 2025;56(1):e30–e38.
8. Reish NFA, Maltsev D, Olsen T, et al. Multiple cerebral hemorrhages in a patient receiving lecanemab and treated with t-PA for stroke. N Engl J Med. 2023. doi:10.1056/NEJMc2215148.
9. Zimmer JA, Ardayfio P, Wang H, et al. Amyloid-related imaging abnormalities with donanemab in early symptomatic Alzheimer disease: secondary analysis of the TRAILBLAZER-ALZ and TRAILBLAZER-ALZ 2 randomized clinical trials. JAMA Neurol. 2025;82(5):461–469.
10. Charidimou A, Boulouis G, Gurol ME, et al. Emerging concepts in sporadic cerebral amyloid angiopathy. Brain. 2017;140(7):1829–1850.
11. Jäkel L, De Kort AM, Klijn CJM, Schreuder FHBM, Verbeek MM. Prevalence of cerebral amyloid angiopathy: a systematic review and meta-analysis. Alzheimers Dement. 2022;18(1):10–28.
12. Vernooij MW, van der Lugt A, Ikram MA, et al. Prevalence and risk factors of cerebral microbleeds: the Rotterdam Scan Study. Neurology. 2008;70(14):1208–1214.
13. Xiong L, Boulouis G, Charidimou A, et al. Dementia incidence and predictors in cerebral amyloid angiopathy patients without intracerebral hemorrhage. J Cereb Blood Flow Metab. 2018;38(2):241–249.
14. Greenberg SM, Vernooij MW, Cordonnier C, et al. Cerebral microbleeds: a guide to detection and interpretation. Lancet Neurol. 2009;8(2):165–174.
15. Gregoire SM, Chaudhary UJ, Brown MM, et al. The Microbleed Anatomical Rating Scale (MARS): reliability of a tool to map brain microbleeds. Neurology. 2009;73(21):1759–1766.
16. Charidimou A, Boulouis G, Roongpiboonsopit D, et al. Cortical superficial siderosis and recurrent intracerebral hemorrhage risk in cerebral amyloid angiopathy: large prospective cohort and preliminary meta-analysis. Int J Stroke. 2019;14(7):723–733.
17. Charidimou A, Boulouis G, Roongpiboonsopit D, et al. Cortical superficial siderosis and bleeding risk in cerebral amyloid angiopathy: a meta-analysis. Neurology. 2019;93(24):e2192–e2202.
18. Potter GM, Chappell FM, Morris Z, Wardlaw JM. Cerebral perivascular spaces visible on magnetic resonance imaging: development of a qualitative rating scale and its observer reliability. Cerebrovasc Dis. 2015;39(3–4):224–231.
19. Charidimou A, Peeters A, Fox Z, et al. Spectrum of transient focal neurological episodes in cerebral amyloid angiopathy: multicentre magnetic resonance imaging cohort study and meta-analysis. Stroke. 2012;43(9):2324–2330.
20. Charidimou A, Werring DJ. Cerebral microbleeds: a guide to detection and clinical relevance in different disease settings. Neuroradiology. 2013;55:655–674.
21. Graff-Radford J, Botha H, Wiste HJ, et al. Boston criteria v2.0 for cerebral amyloid angiopathy without hemorrhage: an MRI-neuropathological validation study. Neurology. 2024.
22. Rodrigues MA, Samarasekera N, Lerpiniere C, et al. The Edinburgh CT and genetic diagnostic criteria for lobar intracerebral haemorrhage associated with cerebral amyloid angiopathy: model development and diagnostic test accuracy study. Lancet Neurol. 2018;17(3):232–240.
23. Charidimou A, Friedrich JO, Greenberg SM, Viswanathan A. Cerebrospinal fluid biomarkers in cerebral amyloid angiopathy: new data and quantitative meta-analysis. Front Aging Neurosci. 2022;14:783996.
24. Kang SH, Kim JP, Jang H, et al. Cerebral amyloid angiopathy and downstream Alzheimer disease plasma biomarkers. JAMA Netw Open. 2025;8(5):e258842.
25. Banerjee G, Samra K, Adams ME, et al. Iatrogenic cerebral amyloid angiopathy: an emerging clinical phenomenon. J Neurol Neurosurg Psychiatry. 2022;93(7):693–700.
26. Grabowski TJ, Cho HS, Vonsattel JPG, Rebeck GW, Greenberg SM. Novel amyloid precursor protein mutation in an Iowa family with dementia and severe cerebral amyloid angiopathy. Ann Neurol. 2001;49(6):697–705.
27. van der Zwet RGJ, Koemans EA, Voigt S, et al. Sensitivity of the Boston criteria version 2.0 in Dutch-type hereditary cerebral amyloid angiopathy. Int J Stroke. 2024.
28. Salvarani C, Hunder GG, Morris JM, et al. Inflammatory cerebral amyloid angiopathy, amyloid-β-related angiitis, and primary angiitis of the central nervous system. Stroke. 2015;46(9):e210–e213.
29. Salvarani C, Hunder GG, Morris JM, et al. Aβ-related angiitis: comparison with CAA without inflammation and primary CNS vasculitis. Neurology. 2013;81(18):1596–1603.
30. Epidemiologic, clinical, and radiologic study of cerebral amyloid angiopathy-related inflammation in Northern Ireland. Neurology. 2025. doi:10.1212/WNL.0000000000214005.
31. Panteleienko L, Banerjee G, Mallon D, et al. Clinical-radiological spectrum of cerebral amyloid angiopathy-related inflammation. Ann Neurol. 2026;99:148–158.
32. Chung KK, Anderson NE, Hutchinson D, Synek B, Barber PA. Cerebral amyloid angiopathy related inflammation: three case reports and a review. J Neurol Neurosurg Psychiatry. 2011;82(1):20–26.
33. Auriel E, Charidimou A, Gurol ME, et al. Validation of clinicoradiological criteria for the diagnosis of cerebral amyloid angiopathy-related inflammation. JAMA Neurol. 2016;73(2):197–202.
34. Piazza F, Greenberg SM, Savoiardo M, et al. Anti-amyloid β autoantibodies in cerebral amyloid angiopathy-related inflammation: implications for amyloid-modifying therapies. Ann Neurol. 2013;73(4):449–458.
35. van den Berg E, et al. No replicating evidence for anti-amyloid-β autoantibodies in cerebral amyloid angiopathy-related inflammation. Ann Clin Transl Neurol. 2024;11(10):2563–2571.
36. Torres J, Loomis C, Cucchiara B, Smith M, Messé S. Diagnostic yield and safety of brain biopsy for suspected primary central nervous system angiitis. Stroke. 2016;47(8):2127–2129.
37. Regenhardt RW, Thon JM, Das AS, et al. Association between immunosuppressive treatment and outcomes of cerebral amyloid angiopathy-related inflammation. JAMA Neurol. 2020;77(10):1261–1269.
38. Laterza V, et al. Cerebral amyloid angiopathy-related inflammation: a systematic review and meta-analysis. J Neurol. 2026;273(7):439.
39. Chu S, Xu F, Su Y, Chen H, Cheng X, Yu J. Cerebral amyloid angiopathy (CAA)-related inflammation: comparison of inflammatory CAA and amyloid-β-related angiitis. J Alzheimers Dis. 2016;51(2):525–532.
40. Danve A, Grafe M, Deodhar A. Amyloid beta-related angiitis — a case report and comprehensive review of literature of 94 cases. Semin Arthritis Rheum. 2014;44(1):86–92.
41. Sperling RA, Jack CR Jr, Black SE, et al. Amyloid-related imaging abnormalities in amyloid-modifying therapeutic trials: recommendations from the Alzheimer's Association Research Roundtable Workgroup. Alzheimers Dement. 2011;7(4):367–385.
42. van Etten ES, Mahinrad S, Grill JD, et al. Amyloid-related imaging abnormalities (ARIA) in anti-amyloid therapies for Alzheimer's disease: an update from the Alzheimer's Association ARIA workgroup. Alzheimers Dement. 2026;22(4):e71361.
43. Merkel J, Perneczky R, Jessen F, et al. Amyloid-related imaging abnormalities (ARIA) in Alzheimer's disease: from pathophysiology to individualized risk assessment. Alzheimers Res Ther. 2026;18:79.
44. van Dyck CH, Swanson CJ, Aisen P, et al. Lecanemab in early Alzheimer's disease. N Engl J Med. 2023;388(1):9–21.
45. Sims JR, Zimmer JA, Evans CD, et al. Donanemab in early symptomatic Alzheimer disease: the TRAILBLAZER-ALZ 2 randomized clinical trial. JAMA. 2023;330(6):512–527.
46. LEQEMBI (lecanemab-irmb) injection, for intravenous or subcutaneous use. US prescribing information.
47. KISUNLA (donanemab-azbt) injection, for intravenous use. US prescribing information, revised July 2025.
48. Rezaii N, et al. Real-world characterization of amyloid-related imaging abnormalities (ARIA) in lecanemab treatment at an academic health system. medRxiv. 2026. (Preprint.)
49. Bregman N, Nathan T, Shir D, et al. Lecanemab in clinical practice: real-world outcomes in early Alzheimer's disease. Alzheimers Res Ther. 2025;17:119.
50. Cummings J, Apostolova L, Rabinovici GD, et al. Lecanemab: appropriate use recommendations. J Prev Alzheimers Dis. 2023;10(3):362–377.
51. Rabinovici GD, Selkoe DJ, Schindler SE, et al. Donanemab: appropriate use recommendations. J Prev Alzheimers Dis. 2025;12(5):100150.
52. Villarejo Galende A, González-Sánchez M, Hilario A, et al. Severe symptomatic amyloid-related imaging abnormalities in Alzheimer's disease: two case reports and systematic review of reported cases. J Alzheimers Dis. 2026;109(3):1099–1116.
53. Cogswell PM, Andrews TJ, Barakos JA, et al.; ASNR Alzheimer, ARIA, and Dementia Study Group. Alzheimer disease anti-amyloid immunotherapies: imaging recommendations and practice considerations for monitoring of amyloid-related imaging abnormalities. AJNR Am J Neuroradiol. 2025;46(1):24–32.
54. Prabhakaran S, Gonzalez NR, Zachrison KS, et al. 2026 guideline for the early management of patients with acute ischemic stroke: a guideline from the American Heart Association/American Stroke Association. Stroke. 2026.
55. Bilodeau PA, Dickson JR, Kozberg MG. The impact of anti-amyloid immunotherapies on stroke care. J Clin Med. 2024;13(5):1245.
56. Al-Shahi Salman R, Dennis MS, Sandercock PAG, et al.; RESTART Collaboration. Effects of antiplatelet therapy after stroke due to intracerebral haemorrhage (RESTART): a randomised, open-label trial. Lancet. 2019;393(10191):2613–2623.
57. Al-Shahi Salman R, Minks DP, Mitra D, et al. Effects of antiplatelet therapy on stroke risk by brain imaging features of intracerebral haemorrhage and cerebral small vessel diseases: subgroup analyses of the RESTART randomised, open-label trial. Lancet Neurol. 2019;18(7):643–652.
58. Veltkamp R, Norrving B, Heuschmann PU, et al.; PRESTIGE-AF Consortium. Direct oral anticoagulants versus no anticoagulation for the prevention of stroke in survivors of intracerebral haemorrhage with atrial fibrillation (PRESTIGE-AF): a multicentre, open-label, randomised, phase 3 trial. Lancet. 2025;405(10482):927–936.
59. Wilson D, Ambler G, Shakeshaft C, et al.; CROMIS-2 collaborators. Cerebral microbleeds and intracranial haemorrhage risk in patients anticoagulated for atrial fibrillation after acute ischaemic stroke or transient ischaemic attack (CROMIS-2): a multicentre observational cohort study. Lancet Neurol. 2018;17(6):539–547.
60. Soo Y, Zietz A, Yiu B, et al.; Microbleeds International Collaborative Network. Impact of cerebral microbleeds in stroke patients with atrial fibrillation. Ann Neurol. 2023;94(1):61–74.
61. SoSTART Collaboration. Effects of oral anticoagulation for atrial fibrillation after spontaneous intracranial haemorrhage in the UK: a randomised, open-label, assessor-masked, parallel-group, pilot-phase, non-inferiority trial. Lancet Neurol. 2021;20(10):842–853.
62. Al-Shahi Salman R, Stephen J, Tierney JF, et al. Effects of oral anticoagulation in people with atrial fibrillation after spontaneous intracranial haemorrhage (COCROACH): prospective, individual participant data meta-analysis of randomised trials. Lancet Neurol. 2023;22(12):1140–1149.
63. Powers WJ, Rabinstein AA, Ackerson T, et al. Guidelines for the early management of patients with acute ischemic stroke: 2019 update to the 2018 guidelines for the early management of acute ischemic stroke. Stroke. 2019;50(12):e344–e418.
64. Biffi A, Anderson CD, Battey TWK, et al. Association between blood pressure control and risk of recurrent intracerebral hemorrhage. JAMA. 2015;314(9):904–912.
65. Greenberg SM, Ziai WC, Cordonnier C, et al. 2022 guideline for the management of patients with spontaneous intracerebral hemorrhage: a guideline from the American Heart Association/American Stroke Association. Stroke. 2022;53(7):e282–e361.
66. Kozberg MG, Perosa V, Gurol ME, van Veluw SJ. A practical approach to the management of cerebral amyloid angiopathy. Int J Stroke. 2021;16(4):356–369.
67. Guo Y, Ma X, Bai X, Wang T, Wang Y. Intracranial hemorrhage risk in patients with cerebral microbleeds after mechanical thrombectomy for acute ischemic stroke: a systematic review and meta-analysis. Neurol Sci. 2024.
68. Charidimou A. Cerebral amyloid angiopathy-related inflammation spectrum disorders: introduction of a novel concept and diagnostic criteria. Ann Neurol. 2025;97(3):470–474.
69. Huhndorf M, Röcken C, Flüh C, Weiler C, Kuhlenbäumer G, Tegeler N, et al. Frequency of deep-seated cerebral microbleeds in patients with lobar hemorrhages and histopathological evidence for cerebral amyloid angiopathy. Front Neurol. 2023;14:1146737.
70. Das AS, Gokcal E, Biffi A, Regenhardt RW, Pasi M, Abramovitz Fouks A, et al. Mechanistic implications of cortical superficial siderosis in patients with mixed location intracerebral hemorrhage and cerebral microbleeds. Neurology. 2023;101(6):e636–e644.
71. Tsai HH, Pasi M, Tsai LK, Chen YF, Lee BC, Tang SC, et al. Microangiopathy underlying mixed-location intracerebral hemorrhages/microbleeds: a PiB-PET study. Neurology. 2019;92(8):e774–e781.