Beyond the Binary: Rethinking Stroke Prevention in Atrial Fibrillation With Cerebral Amyloid Angiopathy

Stroke prevention in individuals with atrial fibrillation (AF) is one of vascular neurology’s most consequential balancing acts. Although oral anticoagulation (OAC) can prevent ischemic stroke (IS) in people with AF, intracranial hemorrhage (ICH) is a major concern, with an overall 45% mortality rate. Traditionally, IS risk is assessed with the antiquated (but omnipresent) CHA₂DS₂-VASc, counterbalanced against bleeding risk, which is typically assessed with the outdated (but ubiquitous) HAS-BLED score. Therefore, treatment is typically framed in binary terms: anticoagulate while accepting the bleeding risk (ie, presumably when CHA₂DS₂-VASc > HASBLED) or withhold OAC and face IS risk (ie, presumably when CHA₂DS₂-VASc < HASBLED). But this is a flawed concept.

Over the past decade, as MRI sequences such as gradient echo (GRE) and susceptibility-weighted imaging (SWI) have become more routine with higher-Tesla magnets and sharper resolution and definition, visualization of cerebral microbleeds (CMBs) has improved.^1,2 Hence, cerebral amyloid angiopathy (CAA) is becoming more commonly diagnosed in older individuals with dementia.³ Outcomes are worse for people with CAA with comorbid AF than those without AF because the very medications that help prevent IS may precipitate catastrophic hemorrhages. 

CHA₂DS₂-VASc and HAS-BLED scores do not adequately account for CAA-related hemorrhagic risk.^4,5 Both scores were designed when warfarin was the predominant, standard-of-care anticoagulant treatment, which currently represents only ~30% of all OAC prescriptions in the United States, and its use continues to decline annually. Most medications used for anticoagulation (70%) are direct OACs (DOACs; ie, DARE [dabigatran, apixaban, rivaroxaban, edoxaban]). Although more expensive than warfarin, DOACs pose significantly less bleeding risk, have fewer medication and food interactions, and do not require routine monitoring of anticoagulation effect.

Multifactorial Risk Stratification
Both AF prevalence and dementia incidence increase with age. Therefore, neurology and cardiology offices need to be prepared for increasingly high-stakes decisions as the US population ages. In addition to documenting CHA₂DS₂-VASc and HAS-BLED scores, physicians now also review GRE and SWI sequences, check for neurocognitive and neurodegenerative features on examination, and ask about genetic predispositions and family history.

Modern stroke care remains dynamic, and advances in hand-held neuroimaging with device-based acute IS and ICH detection, biomarker discovery, and artificial intelligence (AI) offer new opportunities to organize the array of patient data.⁶ Hence, the field is slowly overcoming the binary “anticoagulate or not” paradigm. The use of AI in neurocardiology has been shown to augment logistical efficiencies, including imaging-informed risk stratification, left atrial appendage closure (LAAC) referrals, lost-to-follow-up tracking, and biomarker-driven care.⁶ AI-based prediction models may define the next decade of risk–benefit decision-making in stroke. 

For CAA, the location of CMBs matters: cortical or lobar CMB distribution is a stronger predictor of ICH than involvement of deeper brain regions.^1,2 CMB burden also demonstrates a stepwise relationship with risk, with ICH rates increasing from <2% in individuals with 1 CMB to 4-fold higher in those with 5 to 10 CMBs.^1,6 Cortical superficial siderosis augments the risk of ICH. Mortality rates follow a similar pattern, increasing from 0.9% with 1 CMB to 3.8% with ≥5 CMBs.⁷ Patterns such as these provide clinicians with stronger tools to individualize risk discussions and present clear opportunities for the integration of AI-based platforms and machine-learning models.

Neurocardiac Program Efficiencies
In 2019, we formalized a neurocardiac program (NCP) integrating our university-based vascular neurology and cardiology disciplines in an attempt to streamline care for people with stroke who also need cardiac care. This alliance occurred in an organic, patient-centered manner, with a focus on hyperefficient co-management. Through evidence-based decision pathways, journal clubs, co-clinics, and inpatient-to-outpatient referrals, the NCP accelerates care delivery. Hospital leadership approved the NCP because it projected cost savings while using existing operations, staff, and support teams without incurring new expenses. Patients benefit from the NCP because it provides expeditious, conscientious care and improves follow-up, with the simultaneous goal of decreasing IS and ICH risk.

A specific, creative example of NCP collaboration is illustrated by the incorporation of cognitive screening into AF clinics. In our small study, nearly 20% of individuals with AF age >65 years screened positive for early dementia, and those with previous brain MRI had substantial underlying CMB burden.⁸ Hence, we found that it is reasonable for cardiologists to order brain MRI scans to uncover occult CAA or to assess for CMB burden in people with AF and dementia. Embedding such approaches into routine AF care allows us to co-stratify and co-identify the highest-risk individuals with greater granularity and frame treatment discussions more precisely.

Hybrid Prevention Models
The binary paradigm of “full-dose anticoagulation or not” is archaic, and more adaptive approaches are needed. However, recent evidence suggests that prescribing reduced-dose DOACs may not confer the anticipated safety benefit, with some analyses showing higher rates of bleeding compared with standard, appropriately dosed regimens.⁹ A retrospective analysis of so-called “DOAC failures” in people with IS and AF previously prescribed DOACs revealed frequent underdosing, nonadherence or noncompliance, and competing non-AF etiologies (eg, carotid atherosclerosis, lacunar disease).¹⁰ Another emerging concept is “AF-burden–triggered” anticoagulation;^11-13 however, this strategy needs further study before implementation.

LAAC in the Amyloid Brain
Device-based therapies further expand the toolkit for moving beyond the “anticoagulate or not” binary. The left atrial appendage (LAA) serves as a source of thromboembolism in ≥90% of AF-attributable strokes. LAAC procedures provide a safe and effective nonpharmacologic alternative to OAC by sealing off the LAA,^14,15 followed by a dual antiplatelet regimen for 6 months. More than a decade of LAAC use has demonstrated reductions in embolic IS and bleeding risk without effects on cognitive function.^16-18 LAAC procedures have become an increasingly considered option for individuals at high bleeding risk overall, but prospective data addressing individuals with CAA are lacking and urgently needed. 

LAAC requires timely referral, which can be accomplished through a multidisciplinary NCP to streamline the necessary shared decision-making process and documentation. Such collaboration bridges the traditionally siloed communication between cardiology and neurology. With help from AI tools, the shared decision-making process can be tailored to patient-specific treatment plans in a secure, ethically guided fashion.⁶

Emerging Tools
Looking forward, newer tools such as amyloid PET, which is already used for Alzheimer disease diagnosis, may allow more direct visualization of vascular amyloid deposits. Although not yet routinely used for CAA, amyloid PET can provide a noninvasive map of amyloid burden to enable more precise ICH risk stratification.¹⁹ Biomarkers (ie, plasma amyloid-β, tau, neurofilament light chain) are also rapidly advancing as minimally invasive surrogates of vascular pathology.²⁰ Genetic markers (ie, apolipoprotein E ε2 and ε4) may be associated with CAA severity and ICH risk.¹¹ Incorporating these biomarkers and genotypes into risk models may complement MRI findings and further refine shared decision-making for optimal per-patient selection of OAC or LAAC. When AI tools are used to integrate these multimodal clinical, imaging, biomarker, and genetic data into predictive algorithms, stroke prevention strategies can be further individualized.

Conclusions
Current guidelines and the outdated (but ubiquitous) CHA₂DS₂-VASc and HAS-BLED scores do not sufficiently capture the complexity of balancing embolic and hemorrhagic risks in individuals with CAA and AF.^13,21 Hence, we suggest the following: 

1. Neuroimaging-based risk stratification: MRI should be ordered and GRE and SWI sequences reviewed to assess CMB burden and cortical superficial siderosis in individuals with dementia and AF.

2. NCP efficiencies for LAAC: Expedited referral pathways for nonpharmacologic IS prevention strategies are needed for high-risk individuals with AF (eg, after ICH, high ICH risk, CAA).

3. Novel concepts: Hybrid AF-OAC models and arrythmia-based “burden-triggered” strategies are appealing, but clinical research is needed.

4. Biomarkers, genetics, and AI: Emerging tools offer substantial promise for more precise risk prediction.

The intersection of AF and CAA represents a conundrum in modern medicine: 2 distinct conditions with a real-world overlapping risk, converging to create situations unsolved by traditional tools. By integrating NCP with imaging-informed risk stratification, LAAC, comorbidity-focused (eg, AF-CAA) research, and future-forward biomarkers identified by AI tools, clinicians can build a more nuanced framework. Creative, smaller-scale, grassroots efforts (eg, dementia screening in AF clinics) are feasible and clinically revealing, as we have shown.⁸ Expanding such approaches will be essential in delivering stroke prevention strategies that are tailored for individuals with AF and CAA.