A normal heartbeat starts in the right atrium, spreads across both upper chambers so they contract together, and passes through a single electrical gate to the ventricles below. In atrial fibrillation, that coordination fails. The atria are driven by continuous disorganised electrical activity and no longer contract as a unit — they quiver. The gate passes impulses through erratically, so the ventricles beat irregularly and often fast. On the wrist this is an irregularly irregular pulse, and it is the oldest sign of the condition.
Two consequences follow, and they are almost entirely separate problems.
The first is symptoms. Palpitations, breathlessness, fatigue, reduced exercise tolerance. Losing the atrial contribution to ventricular filling and running at an uncontrolled rate makes the heart a less efficient pump. Some people are severely limited. Many feel nothing at all.
The second is stroke, and it is the one that kills. Blood that is not being pushed cleanly out of a contracting atrium can stagnate — particularly in the left atrial appendage, a small blind-ended pouch off the left atrium whose structure and flow characteristics make it the dominant site of thrombus formation [1] [2] [3] [4]. A clot that forms there and embolises travels up the carotid arteries to the brain.
The size of that risk is the reason this review exists. In the Framingham cohort followed 34 years, atrial fibrillation carried a near fivefold excess of stroke — and, uniquely among the cardiac risk factors examined, its effect did not weaken with age. The attributable risk of stroke from hypertension, coronary disease and cardiac failure all fell with advancing age; for atrial fibrillation it rose from 1.5 percent at ages 50–59 to 23.5 percent at ages 80–89 [5]. An earlier Framingham analysis put the proportion of strokes associated with the arrhythmia at 14.7 percent overall — 68 of 462 first strokes — rising from 6.7 percent at ages 50–59 to 36.2 percent at ages 80–89 [6]. In the very old, this one arrhythmia is behind more than a third of strokes.
It is common and getting commoner. United States prevalence rose from 0.1 percent below age 55 to 9.0 percent at 80 and over, with an estimated 2.3 million affected adults projected to exceed 5.6 million by 2050 [7]; the number of European Union adults aged 55 and over with atrial fibrillation is projected to more than double between 2010 and 2060 [8]. Global burden, incidence, prevalence and associated mortality all rose between 1990 and 2010, with AF-associated mortality roughly doubling in both sexes over that period [9] [10]. Incidence and prevalence have been characterised in Framingham [11] [12], Rotterdam [13], Olmsted County [14], Renfrew/Paisley [15], older adults [16] and long before all of them [17].
Here is the number people misquote. The lifetime risk of developing atrial fibrillation is 1 in 4 for men and women aged 40 and over — and 1 in 6 even in the absence of prior heart failure or myocardial infarction [18]. That is the risk of getting the arrhythmia, not the fraction of strokes it causes. The two "one in four" figures are different quantities and are frequently confused.
And much of it is silent. Atrial fibrillation is often paroxysmal — coming and going — and often asymptomatic, which means the first presentation can be the stroke. That is what makes detection a subject in its own right, and it is covered below.
The organising fact of management, which this review returns to repeatedly: the decision to anticoagulate and the decision about rhythm are made separately, on different evidence, for different reasons. Restoring sinus rhythm does not earn a patient the right to stop anticoagulation. Getting that wrong causes strokes.
Centerpiece: turning a list of risk factors into an annual stroke risk
Everyone with atrial fibrillation has an increased stroke risk. Not everyone has enough of one to justify lifelong anticoagulation, because anticoagulation causes bleeding. The clinical problem is therefore quantitative, and it is solved with a score.
CHADS2 was the first widely used version. It assigns one point each for congestive heart failure, hypertension, age 75 or over, and diabetes, and two points for prior stroke or transient ischaemic attack — the doubling being why the acronym carries a subscript 2 [19]. Validated in 1,733 Medicare beneficiaries with nonrheumatic atrial fibrillation who were not prescribed warfarin, across 2,121 patient-years and 94 ischaemic strokes, it achieved a c statistic of 0.82 (0.80–0.84), better than the two schemes it was built from [19].
Its output is the left panel of the figure: stroke rate per 100 patient-years without antithrombotic therapy of 1.9 at score 0, 2.8 at 1, 4.0 at 2, 5.9 at 3, 8.5 at 4, 12.5 at 5 and 18.2 at 6 [19]. The paper also states the pattern as a single number — the rate increased by a factor of 1.5 (1.3–1.7) for each 1-point increase [19].
The first check is on that summary. Fitting a constant multiplier to the seven printed rates recovers ×1.46 per point, which rounds to the stated 1.5, and the log-linear fit explains 99.99 percent of the variance. Those seven numbers lie almost exactly on an exponential. A clinical score built by adding up integers turns out to behave like a clean multiplicative model of risk — which is not obvious, and is what makes the score interpretable rather than merely predictive.
CHA2DS2-VASc refined it by adding what CHADS2 left out [20]: vascular disease, age 65–74, and female sex, with age 75 or over promoted to two points. It was developed by reclassifying the Birmingham/NICE schema and tested in 1,084 patients from the Euro Heart Survey [20]. The Danish nationwide registries then quantified it in 47,576 patients with a CHADS2 score of 0–1 who were not on a vitamin K antagonist or heparin: stroke or thromboembolism per 100 person-years at one year of 0.84 (0.65–1.08) at score 0, 1.79 (1.53–2.09) at 1, 3.67 (3.34–4.03) at 2, 5.75 (5.33–6.21) at 3 and 8.18 (6.68–10.02) at 4 [21].
The check the arithmetic was never given. The Danish paper prints five rates and never fits a multiplier to them. It is ×1.77 per point — a gradient about 21 percent steeper than the ×1.5 CHADS2 states for itself. The finer score does not merely re-rank patients into more boxes; it separates them faster per box. And its fit to a constant multiplier is slightly worse than CHADS2's (R² 0.974 against 0.9999) in a specific and useful way: the steps from 0→1 and 1→2 are larger than the steps from 2→3 and 3→4. The curve bends most steeply at the bottom — exactly the range in which the treatment decision is actually being made.
And the threshold falls out of the curve. Extrapolating the fitted CHA2DS2-VASc relationship, annual stroke risk crosses 2 per 100 person-years at a score of 1.52 — between 1 and 2, which is where guidelines place the anticoagulation threshold for men. The threshold is not an arbitrary line on an integer scale. It is the score at which the untreated risk crosses a level that the treatment's benefit reliably exceeds its bleeding cost.
Why the refinement mattered, in one comparison. The Euro Heart Survey analysis found the newer scheme's overall discrimination only modestly better (c statistic 0.606 against CHADS2's). What it did far better was the thing that matters at the threshold: patients it classified as low risk had no thromboembolic events at all, while 1.4 percent of CHADS2 low-risk subjects had one [20]. It also cut the proportion parked in the uninformative intermediate stratum from 61.9 percent to 15.1 percent [20]. And in the Danish cohort, patients with a CHADS2 score of 0 — nominally all low risk — had one-year event rates ranging from 0.84 to 3.2 percent depending on their CHA2DS2-VASc score [21]. CHADS2 was not identifying a low-risk group; it was averaging over one. The value of CHA2DS2-VASc is in ruling patients out of treatment safely, not in ranking high-risk patients more precisely. Further validations followed in low-risk groups [22], in Asian populations [23] and against clinical outcomes generally [24] [25], along with the effect on anticoagulation recommendations [26].
The right panel is the honest limit. Twelve published stratification schemes were applied to test cohorts. Observed stroke rates for patients labelled low risk ranged from 0 to 2.3 per 100 patient-years, and for high risk from 2.5 to 7.9 — the two labels separated by 0.2 percentage points of actual risk at their boundary. The fraction of the same patients called low risk varied from 9 to 49 percent, and high risk from 11 to 77 percent [27]. Alternative schemes continue to be proposed and some outperform both on discrimination — the ATRIA score reached c-indices of 0.73 in derivation and 0.70 in external validation, with positive net reclassification against CHADS2 and CHA2DS2-VASc [28] — as do echocardiographic and biomarker-based refinements [29] [30] [31] [32] [33] [34] [35].
Three limits on the figure itself. The Danish cohort was restricted to patients with CHADS2 0–1, so the CHA2DS2-VASc curve only spans scores 0–4 and the highest scores are not represented; the extrapolated crossing point at 1.52 is a fit, not an observation; and the two curves come from different populations, countries and eras, so their vertical offset should not be read as one score being safer than the other.
The teaching point. The score converts an unordered list of clinical facts into a single estimated annual probability, and treatment is offered once that probability crosses a threshold set by what the treatment can achieve. That is risk-based decision-making in its cleanest clinical form — and the right panel is the reminder that the estimate is only as good as the scheme producing it.
Pillar 1: measurement and diagnosis
The ECG, which is the whole diagnosis
Atrial fibrillation is an electrocardiographic diagnosis and nothing else will do. The tracing shows an irregularly irregular ventricular rhythm with no discernible P waves — no organised atrial depolarisation to produce them. No symptom, pulse check or imaging finding substitutes; a recording of the rhythm is required, which is why the practical problem is not interpretation but capture.
Catching it when it comes and goes
Atrial fibrillation is classified by how it behaves over time — paroxysmal (terminating spontaneously), persistent (requiring intervention to terminate) and permanent (accepted as the ongoing rhythm) — and this classification, set out in the successive management guidelines [36] [37] [38], drives everything from monitoring strategy to the realistic goals of rhythm control. Its practical consequence is that a normal ECG in clinic excludes nothing.
Ambulatory monitoring extends the recording window, from 24-hour Holter through multi-day patches to implantable loop recorders, which have been used to quantify arrhythmia burden after ablation with a precision intermittent recording cannot match [39].
Screening detects a real and treatable population. In a cluster-randomised trial across 50 English primary care centres and 14,802 patients aged 65 or over, active screening raised the detection rate of new atrial fibrillation to 1.63 percent per year against 1.04 percent in control practices — and systematic invitation to electrocardiography and opportunistic pulse-taking performed almost identically (1.62 versus 1.64 percent) [40]. Stepwise screening of an entire 75–76-year-old Swedish municipality is more striking still: of 848 participants, 12-lead ECG found previously undiagnosed silent atrial fibrillation in 10 (1 percent), and among 403 people with two or more CHADS2 risk factors who completed two weeks of twice-daily handheld ECG recording, 30 (7.4 percent) were diagnosed with paroxysmal atrial fibrillation — with 9 percent of the whole screened population identified as candidates for anticoagulation. The same study found that 35 of 81 people with already-known atrial fibrillation (43 percent) were not on oral anticoagulation [41]. The undertreatment of known disease was as large a problem as the undiagnosed disease. Systematic reviews of screening yield [42] and primary-care approaches [43] reach compatible conclusions.
Consumer wearables changed the scale of this. In the Apple Heart Study, 419,297 participants used a smartwatch irregular-pulse notification algorithm over a median 117 days. Only 2,161 (0.52 percent) were ever notified [44]. Among the 450 who returned analysable ECG patches, atrial fibrillation was present in 34 percent (97.5% CI 29–39), and the positive predictive value of a notification against simultaneous ECG was 0.84 for a subsequent irregular pulse notification and 0.71 for an irregular tachogram [44]. Smartphone photoplethysmography has been assessed for the same purpose [45], and the wider use of cardiac wearables reviewed [46]. Read carefully, those numbers cut both ways: the notification rate is low and its predictive value respectable, but two-thirds of notified participants who returned a patch did not have atrial fibrillation captured on it.
Device-detected atrial fibrillation, and a finding that complicates the whole story
Implanted pacemakers and defibrillators monitor the atrium continuously, which created a population nobody had been able to study: people with subclinical atrial fibrillation and no clinical arrhythmia. In ASSERT, 2,580 patients aged 65 or over with hypertension and no history of atrial fibrillation were monitored after device implantation [47]. Subclinical atrial tachyarrhythmias occurred in 261 (10.1 percent) by 3 months, and were associated with a hazard ratio of 2.49 (1.28–4.85) for ischaemic stroke or systemic embolism, with a population attributable risk of 13 percent [47]. Device-detected atrial fibrillation predicts stroke across large registries too [48], and the entity now has a formal European consensus definition [49].
But the temporal analysis of that same trial is the finding to sit with. Of the patients who had a stroke or systemic embolism, very few had subclinical atrial fibrillation in the month beforehand. Among the 14 with an episode detected more than 30 days before the event, the most recent episode had occurred a median of 339 days earlier. And 8 patients (16 percent) had subclinical atrial fibrillation detected only after their stroke, despite continuous monitoring for a median of 228 days before it [50].
That is hard to reconcile with the simple mechanical story of this review's opening. If clots formed in a fibrillating atrium and embolised promptly, episodes should cluster before events. They do not. Atrial fibrillation may be as much a marker of an abnormal atrium — an atrial cardiomyopathy with its own thrombotic tendency — as the proximate cause of each clot. The clinical implication is direct, and it is why ARTESiA tested apixaban specifically in subclinical device-detected atrial fibrillation [51] [52]. That trial's abstract is not available in OpenAlex — only its title — so this review does not quote its results, and readers should go to the trial report.
What else the workup is for
An atrial fibrillation diagnosis is the start of a search for drivers and consequences: hypertension, obstructive sleep apnoea, obesity, alcohol, thyroid disease, and structural heart disease on echocardiography. Left atrial size and structural remodelling both predict the arrhythmia and are produced by it [53] [54] [12]. Atrial fibrillation complicating acute myocardial infarction is its own well-described entity [55], as is atrial fibrillation after cardiac surgery [56] [57]. Bleeding risk is assessed alongside stroke risk, most commonly with HAS-BLED — hypertension, abnormal renal or liver function, stroke, bleeding history, labile INR, elderly over 65, drugs or alcohol — derived from 3,978 Euro Heart Survey patients with a c statistic of 0.72 overall [58] [59] [60]. Its purpose is frequently misunderstood: a high bleeding score identifies modifiable bleeding risks to correct, not a reason to withhold anticoagulation from someone whose stroke risk warrants it.
Pillar 2: management — two decisions, made separately
This is the section where the structure matters more than any individual fact.
Decision one: stroke prevention
Anticoagulation works, and the effect size is large. A meta-analysis of the randomised trials found that adjusted-dose warfarin reduces stroke by approximately 60 percent and antiplatelet agents by approximately 20 percent, with warfarin about 40 percent more efficacious than antiplatelet therapy (relative risk reduction 39 percent, CI 22–52, across 12 trials and 12,963 participants) — and absolute increases in major extracranial haemorrhage of 0.3 percent per year or less, smaller than the absolute stroke reductions [61]. The founding trials established this against placebo [62] [63] [64] [65], including in elderly community populations [66], and clopidogrel plus aspirin was tested as an alternative and found inferior to oral anticoagulation [67].
Warfarin's problem was never efficacy. It is a vitamin K antagonist, acting on VKORC1 — the enzyme whose variants cause both warfarin resistance [68] and much of the between-person dose variation [69], alongside CYP2C9 variants affecting dose requirement and bleeding risk [70], to the point that pharmacogenetic dosing algorithms were developed [71]. Add a long list of drug and food interactions [72], a narrow therapeutic window requiring regular INR testing, and bleeding that rises steeply outside it [73] [74], and the practical result was that a highly effective drug was widely underprescribed and often poorly controlled.
The direct oral anticoagulants solved the practical problem. Two mechanisms, both downstream of warfarin's: dabigatran directly inhibits thrombin, the active form of prothrombin [75], while rivaroxaban, apixaban and edoxaban inhibit coagulation factor X in its activated form. The four pivotal trials, each against dose-adjusted warfarin:
- Dabigatran (18,113 patients, median 2.0 years): stroke or systemic embolism 1.69 percent per year with warfarin against 1.11 percent with 150 mg twice daily (relative risk 0.66, 0.53–0.82, superior) and 1.53 percent with 110 mg (non-inferior). Major bleeding 3.36 percent per year with warfarin against 2.71 percent with the lower dose (P = 0.003) and 3.11 percent with the higher (P = 0.31) [76].
- Rivaroxaban (14,264 patients): 2.2 percent per year with warfarin against 1.7 percent in the per-protocol as-treated analysis (hazard ratio 0.79, non-inferior); in intention-to-treat, 2.4 against 2.1 percent (hazard ratio 0.88, non-inferior, P = 0.12 for superiority) [77].
- Apixaban (18,201 patients, median 1.8 years): 1.60 percent per year with warfarin against 1.27 percent (hazard ratio 0.79, P = 0.01 for superiority), with major bleeding 2.13 against 3.09 percent per year [78].
- Edoxaban (21,105 patients, median 2.8 years): 1.50 percent per year with warfarin, whose median time in therapeutic range was a creditable 68.4 percent, against 1.18 percent with high-dose edoxaban (hazard ratio 0.79, non-inferior) and 1.61 percent with low-dose [79].
A meta-analysis of these trials collected the class effect [80], and registry data confirmed the pattern outside trials [81]. The consistent finding across all four is less intracranial haemorrhage, which is what changes the risk-benefit calculation most for the patients at highest stroke risk. Practical guidance on their use is extensive [82], as is the management of bleeding on them [83], including specific reversal with idarucizumab for dabigatran [84] [85] and prothrombin complex concentrate [86]. Dosing in extremes of body weight has been studied [87] [88], and the boundary of their applicability was found the hard way: dabigatran was worse than warfarin in patients with mechanical heart valves [89], which is why "nonvalvular" appears in the indication.
For patients who genuinely cannot take an anticoagulant, apixaban was compared against aspirin in exactly that group and reduced stroke or systemic embolism by more than half (hazard ratio 0.45, 0.32–0.62) without significantly increasing major bleeding (1.4 against 1.2 percent per year) [90]. That trial largely removed aspirin monotherapy as a defensible option.
Left atrial appendage closure is the mechanical alternative — occlude the pouch where the clots form [91]. In PROTECT AF, 707 patients randomised 2:1 to a device or warfarin over a mean 3.8 years had a primary event rate of 2.3 against 3.8 per 100 patient-years (rate ratio 0.60), meeting both non-inferiority and superiority, with lower cardiovascular mortality (1.0 against 2.4 per 100 patient-years, hazard ratio 0.40) and all-cause mortality (3.2 against 4.8, hazard ratio 0.66) [92] [93]. Procedural safety improved with experience [94] [95], the approach has been used specifically in patients with contraindications to anticoagulation [96] [97], compared against direct oral anticoagulants [98] [99], and deployed across multicentre practice [100] [101] [102], with imaging protocols for sizing and planning [103] [104] [105].
And surgical occlusion, tested at scale in LAAOS III: among patients with atrial fibrillation undergoing cardiac surgery for another reason, with a mean CHA2DS2-VASc score of 4.2 and 76.8 percent still on oral anticoagulation at 3 years, appendage occlusion reduced stroke or systemic embolism from 7.0 to 4.8 percent (hazard ratio 0.67, 0.53–0.85, P = 0.001) over a mean 3.8 years, without excess perioperative bleeding, heart failure or death [106]. That result is important beyond the operating theatre: the benefit was on top of continued anticoagulation, which means the appendage contributes residual risk that drugs do not fully cover.
Decision two: rate or rhythm
Having settled anticoagulation, the separate question is what to do about the rhythm itself.
For two decades the answer was: it does not much matter, so keep it simple. AFFIRM randomised 4,060 patients with atrial fibrillation and high stroke or death risk to rhythm control or rate control. Five-year mortality was 23.8 percent with rhythm control against 21.3 percent with rate control (hazard ratio 1.15, 0.99–1.34, P = 0.08), with more hospitalisations in the rhythm-control arm [107]. RACE reached the same conclusion in 522 patients with recurrent persistent atrial fibrillation, where rate control was not inferior on a composite endpoint despite only 10 percent of the rate-control group being in sinus rhythm at 2.3 years against 39 percent of the rhythm-control group [108]. Further trials [109] [110], quality-of-life analyses [111], reviews [112] and secondary analyses of AFFIRM's mortality [113] and strokes [114] followed. A key AFFIRM on-treatment analysis found that the presence of sinus rhythm was associated with a lower risk of death, and antiarrhythmic drugs with increased mortality once sinus rhythm was adjusted for — the authors concluding that any beneficial antiarrhythmic effect of the available drugs was offset by their adverse effects. The same analysis found warfarin use improved survival [115], which is the two-decisions principle emerging from inside a rhythm trial.
Rate control also turned out not to need to be strict. RACE II randomised 614 patients with permanent atrial fibrillation to a lenient target (resting heart rate under 110) or a strict one (resting under 80 and under 110 on moderate exercise) [116]. Three-year cumulative incidence of the primary composite was 12.9 percent with lenient against 14.9 percent with strict control, meeting non-inferiority [116]. The easier target was no worse, and was far easier to reach.
Then EAST-AFNET 4 changed the picture, by changing when. It randomised 2,789 patients with early atrial fibrillation — median 36 days since diagnosis — and cardiovascular conditions to early rhythm control (antiarrhythmic drugs or ablation) or usual care in which rhythm control was reserved for symptoms [117]. Stopped early for efficacy after a median 5.1 years, the first primary outcome — cardiovascular death, stroke, or hospitalisation for worsening heart failure or acute coronary syndrome — occurred at 3.9 per 100 person-years with early rhythm control against 5.0 with usual care (hazard ratio 0.79, 96% CI 0.66–0.94, P = 0.005) [117]. Nights in hospital did not differ, and symptoms and left ventricular function at 2 years did not differ either — so the benefit was not symptomatic. Serious adverse events related to rhythm-control therapy occurred in 4.9 percent against 1.4 percent [117], which is the cost. The strategy also benefited patients with and without symptoms [118] and those with heart failure [119].
The reconciliation with AFFIRM is about timing and tools: EAST-AFNET treated patients within weeks of diagnosis, using safer drug options and ablation, rather than after years of established arrhythmia with the antiarrhythmics of the 1990s.
Catheter ablation rests on one 1998 observation. Mapping the spontaneous initiation of atrial fibrillation in 45 patients found 69 ectopic foci, of which 65 (94 percent) were in the pulmonary veins [120]. Electrically isolating those veins from the left atrium became the procedure, with defined electrophysiological endpoints [121] [122] [123], approaches for persistent disease [124] [125], the non-pulmonary-vein triggers that account for the rest [126], substrate imaging with fibrosis on delayed-enhancement MRI [127] [128], and worldwide surveys of methods, efficacy and safety [129] [130]. It has real complications, including the rare and often fatal atrio-oesophageal fistula [131], phrenic nerve injury [132] and stiff left atrial syndrome [133].
What ablation delivers is rhythm, and — in the trials — not clearly mortality. Against antiarrhythmic drugs it consistently wins on arrhythmia outcomes [134] [135] [136] [137], though as first-line therapy in paroxysmal atrial fibrillation MANTRA-PAF found no significant difference in cumulative arrhythmia burden over 2 years, with ablation better on the secondary measures (freedom from any atrial fibrillation 85 against 71 percent) and carrying procedural risk including one death from a procedure-related stroke [138]. CABANA randomised 2,204 patients to ablation or drug therapy: over a median 48.5 months the primary composite of death, disabling stroke, serious bleeding or cardiac arrest occurred in 8.0 against 9.2 percent (hazard ratio 0.86, 0.65–1.15, P = 0.30) — not significant — while death or cardiovascular hospitalisation (51.7 against 58.1 percent, P = 0.001) and atrial fibrillation recurrence (49.9 against 69.5 percent, P < 0.001) both favoured ablation [139]. The trial had 27.5 percent crossover from drugs to ablation [139], which is the standard caveat on its null primary result.
The exception is heart failure, where the effect is large. CASTLE-AF randomised 363 patients with symptomatic atrial fibrillation, NYHA class II–IV heart failure, ejection fraction 35 percent or less and an implanted defibrillator to ablation or medical therapy. Over a median 37.8 months the composite of death or hospitalisation for worsening heart failure occurred in 28.5 percent with ablation against 44.6 percent with medical therapy [140]. Earlier work had shown ablation feasible and beneficial in congestive heart failure [141], ablation beat amiodarone in persistent atrial fibrillation with heart failure [142], and left ventricular dysfunction can reverse after ablation [143] — the tachycardia-induced component of the cardiomyopathy resolving once the rhythm does. Set against this, AF-CHF had earlier found no mortality benefit from a drug-based rhythm-control strategy in atrial fibrillation with congestive heart failure (death from any cause 32 against 33 percent) [144]. Ablation and antiarrhythmic drugs are not interchangeable ways of achieving the same thing.
Antiarrhythmic drugs themselves remain in use for termination and maintenance — amiodarone for persistent atrial fibrillation [145] and acute rate control in the critically ill [146], ibutilide for flutter [147] — with the perennial constraint that the agents that suppress atrial fibrillation best also carry the most proarrhythmic and organ toxicity. Digoxin's safety in this population has been examined and no increased mortality found in one large analysis [148]. Upstream therapy targeting the substrate rather than the rhythm has been tried, for example with angiotensin receptor blockade to maintain sinus rhythm [149].
Treating the drivers, which is neither decision and belongs to both
Obesity. In Framingham, 5,282 participants followed a mean 13.7 years produced 526 incident cases, with body mass index associated with new-onset atrial fibrillation in part through left atrial dimension [150]; obesity also predicts progression from paroxysmal to permanent atrial fibrillation [151] [152]. And it is reversible: in 150 overweight or obese patients with symptomatic atrial fibrillation randomised to structured weight management against general lifestyle advice, with intensive cardiometabolic risk-factor management in both arms, the intervention reduced symptom burden and severity, atrial fibrillation episodes and cumulative duration on 7-day Holter, and left atrial area [153].
Alcohol is a risk factor for incident atrial fibrillation [154], and abstinence is a treatment: 140 patients drinking 10 or more standard drinks weekly with paroxysmal or persistent atrial fibrillation were randomised to abstain or continue. The abstinence group cut intake from 16.8 to 2.1 drinks weekly (a 87.5 percent reduction) and had greater freedom from recurrence and lower atrial fibrillation burden over 6 months [155].
Obstructive sleep apnoea. Recurrence of atrial fibrillation 12 months after cardioversion was 82 percent in untreated or inadequately treated sleep apnoea, against 42 percent in the CPAP-treated group and 53 percent in controls [156]. Treated apnoea improves outcomes after ablation too [157] [158], and the mechanism has been probed experimentally [159].
And hypertension, diabetes and coronary disease are the shared cardiometabolic substrate; guidelines fold their control into atrial fibrillation management [36].
Pillar 3: progress
Early rhythm control, and ablation earlier in the pathway. EAST-AFNET 4 is the shift [117] [118], reinforced in heart failure by CASTLE-AF [140] and the heart-failure analysis of early rhythm control [119]. The open question is which patients gain enough to accept a 4.9-versus-1.4 percent rate of therapy-related serious adverse events [117].
Screening at population scale. Consumer wearables make identification cheap and continuous where it was episodic and expensive [44] [45] [46], on top of the established yield from organised primary-care and community screening [40] [41] [42]. The unresolved question is what to do with what they find: device-detected subclinical atrial fibrillation raises stroke risk [47] [48] [49], but the episodes do not reliably precede the events [50], so the burden threshold at which anticoagulation becomes worthwhile is genuinely unknown. Atrial fibrillation also associates with cognitive decline and dementia, partly through subclinical cerebral infarcts [160] [161] — an outcome that screening advocates increasingly cite.
Risk-factor and lifestyle management as therapy rather than as advice, following the weight-reduction [153], alcohol-abstinence [155] and sleep-apnoea [156] [158] results. This is the least glamorous and possibly the most transferable finding in the field: three randomised or controlled interventions that change atrial fibrillation burden without touching the heart directly.
A gap this substrate cannot cover. The task specification lists factor XI inhibitors — abelacimab, asundexian, milvexian — as a progress area. This substrate contains no work on them at all: not a trial, not a review, not a mechanism paper. The rationale is easy to state and is not cited here because nothing in the substrate supports it. Readers should go elsewhere for that literature; this review will not assert results it cannot ground.
Two further directions the substrate does support: improved risk prediction, where the ATRIA score already outperformed CHADS2 and CHA2DS2-VASc on discrimination and was designed as a foundation for adding biomarkers [28] [32]; and the atrial substrate itself as a treatment target, with fibrosis on MRI predicting ablation outcome [127], extracellular matrix remodelling maintaining the arrhythmia [54], structural correlates identified in human atria [53], abnormal calcium handling from the sarcoplasmic reticulum implicated in the cellular mechanism [162], and the rapid-atrial-pacing model showing that atrial fibrillation remodels the atrium to sustain itself [163]. The cardiac ion channels through which several antiarrhythmics act and in which atrial-fibrillation-associated variants sit — SCN5A and KCNQ1 — are the molecular end of that story.
Dig deeper in lmmol
Stroke is the reason atrial fibrillation is treated aggressively, and the two reviews should be read together: this arrhythmia accounted for 14.7 percent of first strokes overall in Framingham and 36.2 percent at ages 80–89 [6], its stroke effect uniquely fails to weaken with age [5], and cardioembolic stroke is preventable by anticoagulation in a way that most stroke is not [61]. Heart failure has a two-way relationship with atrial fibrillation — each causes and worsens the other, rhythm control by drugs failed to help in AF-CHF [144], and ablation produced the largest effect seen anywhere in this literature in CASTLE-AF [140], partly by reversing tachycardia-induced ventricular dysfunction [143]. Coronary artery disease shares the risk-score logic exactly — a list of clinical factors converted into an annual probability that decides treatment — and intersects clinically when a patient on anticoagulation also needs antiplatelet therapy after stenting [164] [165]. Hypertension is the commonest driver, present in 70.8 percent of AFFIRM participants [107]. Obstructive sleep apnea and obesity are the two modifiable drivers with randomised or controlled evidence that treating them reduces the arrhythmia itself [156] [153]. Diabetes and glycemic control contributes a point to both risk scores [19] [20]. On the molecular side, the three anticoagulant classes map to three distinct targets: VKORC1 for warfarin [69], prothrombin for dabigatran [75], and coagulation factor X for the Xa inhibitors [77] [78] [79]; SCN5A and KCNQ1 are the cardiac ion channels behind the rhythm itself. The full collection is at health.