Device-detected Atrial Fibrillation: How Much is Too Much?

Why ‘How Much’ is the Wrong First Question

AF is the most common sustained cardiac arrhythmia and a major cause of stroke. Both European and US guidelines now recommend opportunistic screening for AF in individuals aged 65 years and older, particularly those with an elevated stroke risk.^1,2 As screening expands and populations using devices grow, services are increasingly encountering subclinical AF (SCAF), defined as asymptomatic AF detected by continuous monitoring, most commonly via cardiac implantable electronic devices (CIEDs) or implantable cardiac monitors. In device clinics these events are often first flagged as atrial high-rate episodes (AHREs): an algorithmic label based on rate/duration criteria that may represent true AF but can also reflect non-AF atrial tachyarrhythmias or artefact unless confirmed on stored electrograms.^1,3

AHRE/SCAF is detected in a substantial proportion of patients with devices and services increasingly manage arrhythmia burdens far below those studied in classic anticoagulation trials among patients with clinical AF.³ Device-detected AF is also often associated with an increased risk of progression to clinical AF and thromboembolic events.⁴ Guidelines therefore remain cautious and emphasise individualised decision-making rather than a universal anticoagulation trigger for short device-detected episodes.¹

This exposes a constraint: in device-detected AF, absolute stroke risk is often around ~1% per patient-year, limiting the maximum benefit any anticoagulant can deliver, while harm from bleeding remains non-trivial.^5–7 Randomised trials reiterate that trade-off. In NOAH-AFNET 6, edoxaban did not significantly reduce the primary efficacy outcome and increased major bleeding.⁶ In ARTESiA, apixaban reduced stroke/systemic embolism compared with aspirin but increased major bleeding, with low absolute event rates in both arms.⁷ A study-level meta-analysis combining NOAH-AFNET 6 and ARTESiA quantifies this balance: oral anticoagulation (OAC) reduced the relative risk of stroke by approximately 32%, but increased the relative risk of major bleeding by approximately 62%, with no significant effect on cardiovascular or total mortality.⁸

Taken together, this data shifts the question ‘how much is too much?’ onto different ground. The defensible answer cannot be based on time in minutes, but rather a framework that separates net-benefit selection from service delivery. As such, in device-detected AF, episode duration is a poor standalone trigger; management should be phenotype-led for anticoagulation and burden-led for workflow urgency. The pathway outlined below represents one way of organising the available evidence into a workable clinical structure, serving as a practical scaffold for clinical reasoning and should not be interpreted as guideline-level recommendations.

Diagnostic Confirmation is Non-negotiable

Before any risk model is applied, services must confront a basic truth about safety: AHREs do not inherently indicate AF until adjudicated. Device algorithms are designed for sensitivity and can overcall AF because of oversensing, noise, far-field signals or other atrial tachyarrhythmias.³ With the known bleeding risk associated with OACs, offering anticoagulation to someone based on a false-positive would be iatrogenic harm.

Importantly, an electrogram review may reveal that some high-rate episodes represent atrial flutter or focal atrial tachycardia rather than AF. This may obviate the need for long-term anticoagulation as these rhythms are more frequently amenable to catheter ablation.³

If the rhythm is unconfirmed, it should be classified as an unconfirmed AHRE and managed as a monitoring/progression signal, not as an anticoagulation trigger.³ This validity gate is well-suited to physiologist-led delivery and should precede any discussion about OAC. Artificial intelligence-based approaches to ECG and electrogram classification have demonstrated high accuracy in distinguishing AF from other atrial tachyarrhythmias and artefact, offering a route to faster and more consistent adjudication at scale.⁹

Observational Evidence: Burden Correlates with Risk, but Thresholding is Unstable

AF burden refers to the cumulative duration of AF expressed as a proportion of total monitored time. A related but distinct metric is the longest single episode duration, which captures peak episode length rather than aggregate exposure. Critically, the same numerical burden carries different weights depending on the duration of the observation window. For example, 1% over a fortnight of monitoring is not equivalent to 1% over a year, as shorter windows are more likely to capture clusters that may not accurately reflect long-term behaviour. Thus, it is important that when reporting burden, services should specify the monitoring period.

Observational cohorts consistently show a risk gradient: longer AHRE/SCAF is associated with higher thromboembolic risk.^4,5,10,11 The ASSERT trial had some key findings: device-detected SCAF episodes exceeding 6 minutes were associated with increased stroke/systemic embolism risk, albeit with low absolute event rates.⁵ A meta-analysis of patients without prior AF reinforces this gradient, with a notably elevated signal for episodes of 24 hours or longer.⁴

However, risk does not scale cleanly across the minutes-to-hours range. In a focused analysis of the ASSERT findings, although episodes >24 hours carried a clearer association with stroke/systemic embolism, intermediate durations often failed to cleanly define the presence or absence of SCAF.¹⁰ A systematic review/meta-analysis similarly showed wide variation in proposed duration thresholds across studies, consistent with cohort dependence, such as baseline comorbidity and competing risks, and methodological differences in detection and adjudication.¹¹

The correct inference is narrow but important: burden contains signal, but burden is not a stable, portable, single-number trigger to definitively decide the need for anticoagulation.

Randomised Evidence: Benefit Exists But Bleeding is Real

Randomised trials test what observational studies cannot: whether anticoagulation improves outcomes in device-detected AF. These have been the findings of the most relevant trials:

• NOAH-AFNET 6 (edoxaban versus placebo/no anticoagulation): no significant reduction in the primary efficacy composite and increased major bleeding.⁶
• ARTESiA (apixaban versus aspirin; enrolled SCAF episodes 6 minutes to <24 hours): reduced stroke/systemic embolism but increased major bleeding.⁷ ARTESiA’s upper enrolment boundary excluded episodes of 24 hours or longer, reflecting the ASSERT-derived view that such durations carry a sufficiently distinct risk profile and warrant separate management. The pathway here, therefore, primarily addresses the population that has SCAF of a duration below 24 hours, as this is where treatment decisions are most uncertain.

Because aspirin itself carries bleeding risk and only modest protection from stroke, the real-world balance versus no antithrombotic therapy in very low-risk patients may be even narrower than apixaban-aspirin comparisons imply.⁷ Across both trials, event rates were low and the margin was narrow. A study-level meta-analysis combining NOAH-AFNET 6 and ARTESiA confirmed reduced ischaemic stroke (RR ~0.68) at the cost of increased major bleeding (RR ~1.62), with no effect on cardiovascular death or all-cause mortality.⁸

The Post-trial Reality

The post-trial reality is that burden did not select benefit the way clinicians hoped it would and this is the key reason duration-threshold fails as a primary basis for decision-making.

In ARTESiA, baseline SCAF frequency and longest episode duration did not reliably associate with stroke/systemic embolism risk and did not clearly modify the treatment effect of apixaban versus aspirin.¹² Importantly, this applies within ARTESiA’s enrolled SCAF range and trial context, not as a claim that burden never matters.^7,12

Similarly, in NOAH-AFNET 6, a focused analysis of patients with long AHREs (including episodes ≥24 hours) did not demonstrate a treatment-effect pattern that rescues a duration-threshold rule.¹³ Even long episodes did not reliably create a wide net-benefit margin that would justify a universal duration trigger.

Therefore, the split should be explicit:

• Burden matters most for progression risk, surveillance intensity and triage of clinic workload.^3,10,11
• Phenotype (absolute risk) matters most when considering whether prescribing OAC is worth the cost of bleeding.^6,7,14

That split between decision and delivery is the organising principle of a workable modern pathway. It is worth noting that these thresholds refer to thromboembolic endpoints. Other consequences of device-detected AF, such as cognitive decline or heart failure progression, may present different burden cut-offs, but these outcomes are beyond the scope of this framework.

Phenotype: Vascular Disease and Absolute Risk as the Most Defensible Discriminator

If one phenotype lever is both clinically legible and linked to trial findings, it is vascular disease/absolute thromboembolic risk. A combined analysis of NOAH-AFNET 6 and ARTESiA suggests that patients with device-detected AF and vascular disease have higher event rates and may derive greater benefit from anticoagulation than those without vascular disease, who appear very low risk.¹⁴ This is not proof of a universal rule, but it is a defensible discriminator that is simple enough for service delivery. Machine learning models that integrate clinical, electrocardiographic and imaging features are beginning to outperform conventional scores for stroke risk prediction in AF populations and may, in time, refine phenotype classification beyond what the CHA₂DS₂-VASc score for AF stroke risk, and vascular disease status alone can offer.⁹ Until such tools are validated specifically for device-detected AF, a pragmatic three-tier classification can be applied:

• High phenotype: prior stroke/transient ischaemic attack (TIA) and/or established vascular disease, such as coronary/peripheral artery disease (CAD/PAD), where net benefit is more plausible, provided bleeding risk is actively modified.^1,14
• Intermediate phenotype: elevated CHA₂DS₂-VASc score without overt vascular disease, where decisions are often borderline.¹
• Low phenotype: low CHA₂DS₂-VASc score and no vascular disease, where routine anticoagulation is rarely a safe default.^1,14

Bleeding liability must be handled systematically. Guidelines emphasise that bleeding risk assessment should drive decisions over risk-factor modification and safe prescribing rather than serve as a reflex ‘withhold OAC’ tool.¹ Practically, this means documenting and addressing modifiable risks, namely blood pressure, renal function, interacting drugs such as non-steroidal anti-inflammatory drugs/antiplatelets, alcohol, anaemia and falls/frailty context, before finalising an OAC decision in borderline phenotypes.¹

Atrial Cardiomyopathy Markers: Useful Tiebreakers

Atrial cardiomyopathy is a plausible ‘missing variable’: structural and electrical atrial disease may create a prothrombotic atrial environment that persists beyond the timing of detectable AF episodes.¹⁵ However, in device-detected AF, the bar for using substrate markers to trigger lifelong OAC is higher than simple association. Most markers, such as left atrium (LA) enlargement, interatrial block and P-wave indices, are risk correlates, meaning they may enrich baseline stroke risk but have not been validated as selectors of the net benefit of anticoagulation. Thus, they have not been shown to reproducibly identify a subgroup in whom OAC’s absolute stroke reduction outweighs bleeding more than in others.^1,15,16 In the current randomised controlled trial (RCT) era, selection of OAC benefit is anchored to clinical phenotype rather than substrate surrogates.^6–8,14 Therefore, their defensible role is narrower: use substrate markers as risk enhancers and tiebreakers when the phenotype is intermediate and decisions are genuinely borderline and only when these substrate markers are already available or of low burden to obtain.^15,16 A pragmatic bundle could include echocardiographic LA enlargement (assessed using the LA volume index if available) and ECG markers such as advanced interatrial block or prolonged P-wave indices.^15,16 Critically, substrate evaluation should not delay safety steps, such as electrogram (EGM) confirmation, or the structured bleeding-risk optimisation that determines whether a marginal net benefit collapses.¹

The Deliverable: A Device-clinic Pathway

After confirmation of AF, the pathway is summarised in Figure 1 and can be delivered as three steps.

Step 1: Burden Class (Workflow Urgency)

Burden classes must be auditable service triage categories. To allow for a meaningful comparison between patients and over time, services should define a standard reporting window (for example, cumulative burden over the preceding 30, 60 or 90 days):

• Class A (low burden): rare, confirmed episodes, low cumulative burden, no clear progression trend.
• Class B (intermediate): recurrent confirmed episodes and/or rising burden over time but not sustained or rapidly escalating.
• Class C (high burden): sustained episodes, rapid escalation or repeated long episodes; pragmatically treat ‘approaching ≥24 hours’ as a clinician-review trigger because it marks progression and instability and is consistent with management arising from RCT findings (not because it is a universal stroke threshold).^7,10,13

Step 2: Phenotype Class (Net-benefit Selection for Oral Anti-coagulants)

High phenotype: prior stroke/TIA and/or established vascular disease (CAD/PAD); high absolute risk; modifiable bleeding risk.^1,14

Intermediate phenotype: elevated CHA₂DS₂-VASc without overt vascular disease.¹

Low phenotype: low CHA₂DS₂-VASc and no vascular disease.^1,14

Step 3: Action Grid (Phenotype × Burden)

Figure 1 summarises the proposed pathway: electrogram confirmation as the entry criterion, burden for clinic triage and phenotype for anticoagulation net-benefit.

Workflow Impact: Preventing Alert Fatigue by Separating Physiologist and Clinician Work

If phenotype-led anticoagulation and burden-led triage improve care, service delivery must be changed. At the physiologist level, this would involve EGM adjudication, burden classification, optimisation of detection settings and flagging phenotypes from the clinical record.³ At the clinician level, targeted review of intermediate and high phenotypes (Class B and C), plus red flags (prior stroke/TIA, suspected heart failure deterioration, bleeding complexity, rapid progression or difficult shared decision-making).^1,14

This reflects the reality of the device clinic, where consultants cannot review every AHRE alert. It also aligns with patient and clinician perspectives on ‘remote-monitoring-first’ pathways, where acceptability depends on clarity and agreement over escalation and communication.¹⁷ Outcome data from remote-heavy models suggest that, in engaged populations with clear escalation rules, reduced in-person evaluation can be safe.¹⁸

Conclusion

Duration-only thresholds fail in device-detected AF because they attempt to compress a multi-dimensional problem into a single number. Observational gradients exist, but evidence from clinical trials indicates that baseline SCAF frequency and longest episode duration do not reliably select a wide net-benefit subgroup for anticoagulation, and even long episodes do not reliably produce a duration-threshold rule.^7,8,12,13 Therefore, the pragmatic path forward is not to strive to find the duration in minutes that can be universally applied as a threshold, it is to implement a pathway for device-detected AF anchored to three principles:

• Electrogram confirmation of AF as the entry criterion.³
• Burden directs workflow urgency and surveillance intensity.^10,11
• Phenotype, especially vascular disease and absolute risk, drives anticoagulation decisions, with atrial substrate markers used only as pragmatic tiebreakers in genuinely borderline cases.^1,14–16

In device-detected AF, ‘too much’ is not a duration. It is confirmed AF occurring in a phenotype whose absolute stroke risk is high enough that preventing one stroke plausibly justifies the risk of bleeding from taking OACs.