Thromboembolic events and vascular dementia in patients with atrial fibrillation and low apparent stroke risk

Atrial fibrillation (AF) is the most common sustained arrhythmia, with rising global prevalence and substantial morbidity, mortality and healthcare burden. Stroke prevention in AF is guided by risk scores such as CHA2DS2-VASc, but these tools have modest discrimination and are based on historical cohorts. Current guidelines typically restrict oral anticoagulation to patients with specific risk factors and do not consider other thromboembolism-related outcomes such as vascular dementia. The research question addressed whether patients with AF who have a low perceived stroke risk (no prior stroke/TIA, CHA2DS2-VASc 0–1, and no anticoagulation) still experience increased rates of thromboembolic events, vascular dementia, and mortality compared with matched individuals without AF.

Prior studies have established that AF increases thromboembolic risk and is linked to poorer outcomes after stroke. Risk stratification tools (for example, CHA2DS2-VASc) show only modest predictive performance and may not reflect contemporary risk profiles or management. Associations between AF and cognitive decline or dementia have been reported, but confounding by vascular comorbidities and anticoagulant use has limited inference, and few studies have distinguished dementia subtypes. Evidence on silent cerebral infarctions and progressive cerebral lesions in AF supports a biological pathway for vascular dementia. Randomized evidence on anticoagulation to prevent cognitive decline in AF is scarce and largely in older populations, with ongoing trials now testing earlier anticoagulation in low-risk AF.

Design: Population-based, retrospective, matched, open cohort study using UK primary care electronic health records from the IQVIA Medical Research Database (IMRD), covering 1 January 2005 to 31 December 2020. Data were extracted using the DExtER tool. Ethics approvals were in place via the NHS South-East Multicentre Research Ethics Committee and IMRD Scientific Review Committee. Population: Adults aged 40–75 registered with eligible general practices for at least 1 year. Exposure: Recorded diagnosis of AF. Index date was AF diagnosis (for incident AF) or practice eligibility date (for prevalent AF). Matching: For each AF-exposed patient, up to four unexposed patients without AF were randomly selected matched on age (±1 year), sex and health authority region, and assigned the same index date. Exclusions: Prior stroke/TIA; any prior anticoagulant prescription (vitamin K antagonists or DOACs); CHA2DS2-VASc ≥2 (one point each for heart failure, hypertension, age 65–74, diabetes, vascular disease; two points for prior stroke/TIA/arterial thromboembolism; female sex counted as a point). Covariates: Age (continuous), sex, Townsend deprivation quintile, ethnicity, hypertension, and diabetes. Outcomes: Incident stroke, arterial thromboembolism, ischemic heart disease, pulmonary embolism, deep vein thrombosis; dementia (all-cause), vascular dementia, Alzheimer’s disease; and all-cause mortality. Follow-up: From index date to earliest of outcome, death/de-registration, practice ceasing data contribution, or 31 December 2020. Statistical analysis: Incidence rates per 1,000 person-years and Cox proportional hazards models estimated hazard ratios (HRs) with 95% CIs, adjusted for age, sex, deprivation, ethnicity, hypertension and diabetes. Proportional hazards assumptions were evaluated with Schoenfeld residuals. Subgroup analyses by age strata (<55, 55–69, ≥70 years) were performed. Sensitivity analyses included new AF only and censoring at time of anticoagulant initiation. Competing risk analyses (Fine–Gray) assessed death for stroke and vascular dementia, and ischemic stroke/ischemic heart disease for vascular dementia. Software: Stata. Sample: Of 5,199,994 eligible primary care patients, 290,525 had AF aged 40–75; after exclusions, 36,340 AF-exposed patients with low stroke risk and no anticoagulation were included and matched to 117,298 controls. Median follow-up was 4.4 years in AF and 5.0 years in controls (combined 831,005 person-years).

• Population: 36,340 AF-exposed patients with low perceived stroke risk (CHA2DS2-VASc 0–1, no prior stroke/TIA, no anticoagulation) matched to 117,298 controls by age, sex and region; median follow-up 4.4 vs 5.0 years; total 831,005 person-years.
• Stroke: 1,366/36,340 (3.8%) AF vs 1,796/117,298 (1.5%) controls; crude incidence 6.71 vs 2.92 per 1,000 person-years; adjusted HR 2.06 (95% CI 1.91–2.21; P<0.001).
• Arterial thromboembolism: 104/36,340 (0.3%) AF vs 123/117,298 (0.1%) controls; crude incidence 0.50 vs 0.20 per 1,000 person-years; adjusted HR 2.39 (1.83–3.11; P<0.001).
• Ischemic heart disease: 1,870/33,162 (5.6%) AF vs 3,030/113,222 (2.7%) controls; crude incidence 10.36 vs 5.15 per 1,000 person-years; adjusted HR 1.88 (1.77–1.99; P<0.001).
• Pulmonary embolism: 365/36,047 (1.0%) AF vs 808/117,063 (0.7%) controls; crude incidence 1.77 vs 1.31 per 1,000 person-years; adjusted HR 1.23 (1.08–1.39; P=0.001). No difference for deep vein thrombosis: 345/35,982 (1.0%) AF vs 981/116,790 (0.8%); crude incidence 1.68 vs 1.59 per 1,000 person-years; adjusted HR 0.97 (0.86–1.10; P=0.62).
• All-cause mortality: 3,246/36,340 (8.9%) AF vs 5,875/117,298 (5.0%); crude incidence 15.56 vs 9.44 per 1,000 person-years; adjusted HR 1.44 (1.38–1.50; P<0.001).
• Dementia: All-cause dementia 420/36,232 (1.2%) AF vs 788/117,056 (0.7%); crude incidence 2.03 vs 1.27 per 1,000 person-years; adjusted HR 1.17 (1.04–1.32; P=0.010). Vascular dementia 126/36,327 (0.4%) AF vs 162/117,273 (0.1%); crude incidence 0.60 vs 0.26 per 1,000 person-years; adjusted HR 1.68 (1.33–2.12; P<0.001). No association with Alzheimer’s disease: adjusted HR 0.85 (0.70–1.03; P=0.09).
• Subgroup/sensitivity: Results consistent across age strata and after restricting to new AF or censoring at anticoagulant initiation. Interaction P values for age were nonsignificant for vascular dementia and all-cause mortality; HRs for stroke, arterial thromboembolism, and ischemic heart disease decreased with age. Competing risk analyses did not materially change associations.

The study shows that, even among patients with AF deemed at low risk of stroke by conventional scoring (CHA2DS2-VASc 0–1) and not receiving anticoagulation, there are substantially higher risks of incident stroke, arterial thromboembolism, ischemic heart disease, vascular dementia, and all-cause mortality compared with matched controls. The increased risk of vascular dementia, but not Alzheimer’s disease, supports a thromboembolic and vascular mechanism linking AF to cognitive decline. Findings were robust across subgroups and sensitivity analyses, with no significant competing risk effects. These results challenge the sufficiency of current risk stratification and suggest that thromboembolic and cognitive risks manifest earlier in the AF disease course than typically appreciated. Clinically, a more dynamic, earlier approach to risk modification and integrated AF management may be warranted while awaiting randomized trial evidence on earlier anticoagulation.

Patients with AF at low apparent stroke risk experience significantly elevated risks of stroke, arterial thromboembolism, vascular dementia, and death compared with matched individuals without AF. The dissociation between vascular dementia (increased) and Alzheimer’s disease (not increased) underscores a vascular/thromboembolic pathway. These findings highlight the need to reassess early risk management strategies in AF. Ongoing randomized trials (DaRe2THINK and BRAIN-AF) will be critical to determine whether earlier initiation of DOACs in low-risk AF can reduce thromboembolic and cognitive outcomes. Future research should refine dynamic risk prediction, evaluate integrated management strategies, and explore mechanisms linking AF to vascular brain injury.

This retrospective observational study relies on routinely collected, coded primary care data, introducing potential misclassification, incomplete capture, and residual confounding. AF exposure included patients with past or resolved AF, and the presence or burden of ongoing AF could not be ascertained; if anything, this may bias associations toward the null. Risk factors were assessed at baseline and not updated over time. Dementia and vascular outcomes were identified via coding without adjudication, and incident rates for vascular dementia were low, limiting detailed analyses. Causality cannot be inferred from associations. The IMRD sample represents approximately 6% of the UK population with a slight skew toward younger/more affluent areas, which may affect generalizability. Bleeding outcomes were not assessed, and there was no linkage to external datasets. Potential prescription biases preclude causal inference regarding anticoagulant effects.