Clinical characteristics associated with ischaemic stroke and transient ischaemic attack among non-anticoagulated patients with atrial fibrillation

Atrial fibrillation (AF) is a major cause of ischaemic stroke (IS) and transient ischaemic attack (TIA). Risk stratification tools such as CHA₂DS₂ VASc (cardiac failure, hypertension, age ≥75 years [doubled], diabetes, previous IS/TIA [doubled], vascular disease, age 65–74 years, sex [female]) are widely used to estimate stroke risk and guide decisions regarding oral anticoagulation.

Despite guideline-directed recommendations, a significant proportion of patients with AF remain untreated with anticoagulation in routine clinical practice. Data describing the clinical characteristics of non-anticoagulated patients with AF who experience IS/TIA, and how they differ from event-free non-anticoagulated patients, are limited, particularly in contemporary multi-ethnic populations such as Aotearoa New Zealand. In this context, the contribution of AF pattern (paroxysmal versus persistent/permanent) and sex to stroke risk among untreated patients remains incompletely characterised.^1,2

We therefore examined clinical characteristics associated with IS/TIA among non-anticoagulated patients with AF.

Methods

We conducted a case–control study restricted to adults with non-valvular AF who were not treated with oral anticoagulation. Incident IS/TIA cases were patients with known AF prior to 1 September 2020 who experienced their index event between 1 September 2020 and 31 August 2021. These patients were sampled from the Auckland Regional Community Stroke Study (ARCOS) V study and had no record of oral anticoagulant dispensing prior to the index event. Details of the ARCOS study have been published elsewhere.³ In brief, ARCOS is a population-based stroke registry capturing first-ever and recurrent stroke/TIA events across Auckland in patients aged ≥16 years over 1 calendar year. Given the population-based design of ARCOS, patients who experienced an IS/TIA event during follow-up and presented for medical attention would be expected to be captured within the registry. Stroke/TIA events are independently adjudicated by a diagnostic committee.

Controls were patients with AF hospitalised prior to 1 September 2020 who remained free of IS/TIA through 31 August 2021 and who had no record of oral anticoagulant dispensing including during the corresponding observation period up to 31 August 2021; controls were identified from the Ministry of Health National Minimum Dataset. Similarly, all cases included in this study had been hospitalised with AF prior to 1 September 2020. Patients in whom AF was first diagnosed at the time of the index IS/TIA event were excluded.

Clinical records were systematically reviewed to confirm AF diagnosis using electrocardiograms, echocardiograms and clinician documentation. The AF pattern (paroxysmal vs persistent/permanent) was determined from clinical records prior to 1 September 2020. AF pattern was determined using a hierarchical approach: 1) explicit clinician documentation where available, otherwise 2) classification based on serial electrocardiograms, echocardiography reports and clinical notes demonstrating either paroxysmal episodes with reversion to sinus rhythm or continuous AF without reversion. AF pattern could be determined for all patients using hospital records, and there were no cases with an indeterminate classification.

Data collected included age, sex and prioritised self-identified ethnicity (i.e., Māori, Pacific peoples, Europeans and Others) as per national standards.⁴ Comorbidities relevant to CHA₂DS₂ VASc, as well as body mass index (BMI) and AF pattern (paroxysmal vs persistent/permanent), were collected prior to 1 September 2020 and were ascertained from clinical records. Cardiac failure was defined as left ventricular ejection fraction ≤40% on echocardiography. We also collected data on antiplatelet use (aspirin and/or clopidogrel) in the observation period for controls or prior to the index IS/TIA for cases. For context, characteristics of anticoagulated vs non-anticoagulated patients are shown in Appendix Table 1.

Continuous variables are summarised using medians with interquartile ranges and categorical variables as counts with percentages. Group comparisons used the Wilcoxon Rank-Sum and χ² tests, respectively. Multivariable logistic regression was used to estimate adjusted odds ratios (aORs) and 95% confidence intervals (CIs). Analyses were restricted to estimation of associations; absolute risks were not calculated.

As sensitivity analyses, stabilised inverse probability weighting (IPW) was used to assess the robustness of associations for female sex and AF phenotype to measured confounding. For each exposure, propensity scores were estimated using logistic regression models including age, comorbidities relevant to CHA₂DS₂ VASc, AF pattern (where not the exposure of interest), ethnicity and BMI. Missing BMI values (n=28) were handled using a missing-indicator approach, with BMI imputed to the median and a binary indicator included in the propensity model.

Stabilised weights were constructed using the marginal probability of the exposure and applied in weighted logistic regression models with robust standard errors to estimate aORs for IS/TIA. To reduce the influence of extreme weights, stabilised weights were truncated at the 1st and 99th percentiles. These analyses were prespecified as sensitivity analyses and were not intended to estimate causal effects.

This study has been approved by the Health and Disability Ethics Committee (ref: 2023 AM 9094) and the Auckland University of Technology Ethics Committee (24/4), and granted an exemption from obtaining informed consent. Analyses were conducted using Stata/BE 17 (StataCorp, College Station, Texas). This study is reported in accordance with the STrengthening the Reporting of OBservational studies in Epidemiology (STROBE) guidelines.⁵

Results

Among the 593 untreated AF patients, 121 experienced IS or TIA during the study period and 472 remained event free. Compared with controls, cases were substantially older (median age 80 [interquartile range (IQR) 69–86] vs 65 [56–74] years) and more likely to be female (57.0% vs 45.6%). Cases had a lower median BMI (27.5 vs 30.8kg/m²), all p<0.05.

Cases also had a higher prevalence of cardiovascular comorbidity, including cardiac failure, hypertension, diabetes, prior IS/TIA and vascular disease. As expected, cases had a higher CHA₂DS₂ VASc score (median 4 [IQR 3–5] vs 2 [1–3], p<0.0001); distribution in Figure 1.

View Figure 1, Table 1–2.

Ethnic distribution did not differ between cases and controls. Persistent or permanent AF was markedly more common among cases. Antiplatelet use was similar in controls and cases (32.2% vs 38.8%, p=0.168). Baseline data are summarised in Table 1.

In the multivariable logistic regression analysis (Table 2), older age was independently associated with higher odds of IS/TIA (aOR 1.04 per year [95% CI 1.02–1.06], p=0.001). A history of prior IS/TIA remained a strong independent risk marker (aOR 2.89 [95% CI 1.44–5.76], p=0.003). Persistent or permanent AF showed the largest adjusted association with IS/TIA (aOR 5.32 [95% CI 3.20–8.82], p<0.001).

Female sex was also associated with increased odds of IS/TIA (aOR 1.78 [95% CI 1.08–2.94], p=0.023). BMI was associated with lower odds of IS/TIA (aOR 0.96 per kg/m² [95% CI 0.93–1.00], p=0.031).

Cardiac failure was associated with IS/TIA in univariable analysis but did not remain statistically significant after multivariable adjustment. Similarly, hypertension, diabetes and vascular disease were not independently associated with IS/TIA after adjustment. There was a trend towards higher odds of IS/TIA in Māori (p=0.076).

Sensitivity analysis

Female sex

In a sensitivity analysis using stabilised IPW to balance baseline characteristics between females and males, the distribution of stabilised weights was well behaved, with a mean close to 1, a mild right skew and no evidence of extreme values (Appendix Figure 1). Weights were therefore truncated at the 1st and 99th percentiles. After weighting, baseline characteristics including age, comorbidities, AF pattern, BMI and ethnicity were well balanced, with similar weighted means and proportions across covariates. In the weighted analysis, female sex remained associated with higher odds of IS/TIA (trimmed aOR 1.66 [95% CI 1.08–2.55], p=0.02), consistent with the unweighted estimate.

AF phenotype

In a parallel sensitivity analysis examining AF phenotype, stabilised IPW showed a modest right-skewed distribution with a small number of higher weights (Appendix Figure 2); similarly, weights were truncated at the 1st and 99th percentiles. After weighting, baseline characteristics were reasonably balanced between patients with persistent/permanent AF and those with paroxysmal AF.

Persistent/permanent AF remained strongly associated with higher odds of IS/TIA in the weighted analysis (trimmed aOR 4.41 [95% CI 2.72–7.12]), consistent with the primary multivariable analysis, although the magnitude of association was modestly attenuated compared with the unweighted adjusted estimate (aOR 5.30 [95% CI 3.20–8.80]).

Discussion

In this case–control study of non-anticoagulated patients with non-valvular AF, IS/TIA events were concentrated within identifiable high-risk sub-groups. Persistent or permanent AF showed the strongest association with IS/TIA, supporting AF phenotype as an important marker of risk among untreated patients. Prior IS/TIA remained a dominant risk marker, highlighting substantial missed opportunities for secondary prevention in patients who remain untreated. The inverse association between BMI and IS/TIA, described as the obesity paradox, has been reported elsewhere.⁶ Increasing age was independently associated with higher odds of IS/TIA, consistent with extensive prior evidence demonstrating age as a dominant risk factor in AF-related stroke.^7–9 In comparison, anticoagulated patients were older and had a higher comorbidity burden and CHA₂DS₂ VASc scores (Appendix Table 1), consistent with risk-based prescribing.

Female sex remained associated with higher adjusted odds of IS/TIA in this untreated cohort. In sensitivity analyses using IPW to address measured confounding, the associations observed in the primary analyses were largely preserved. The association between female sex and IS/TIA among untreated patients with AF remained after covariate balancing, suggesting that sex-related differences were not fully explained by age, comorbidity burden, AF phenotype or ethnicity.

Contemporary Australian/New Zealand guidelines recommend use of the sexless CHA₂DS₂ VA score, reflecting evidence from large European cohorts that female sex acts as an age-dependent risk modifier rather than a uniform independent risk factor.^1,10 CHA₂DS₂ VA scores did not differ between men and women in our cohort. Given known differences in stroke epidemiology and ethnic composition in Aotearoa New Zealand, further local validation of sex-neutral risk stratification approaches may be warranted.

For AF phenotype, IPW resulted in partial attenuation of the very large association observed in the multivariable analysis, indicating some structural confounding by age and comorbidity; however, persistent/permanent AF remained strongly associated with IS/TIA after prespecified weight truncation.

These sensitivity analyses assess robustness to measured confounding and should not be interpreted as causal estimates, particularly given the case–control design and restriction to untreated patients. Although residual confounding and selection bias remain possible, the consistency of findings across analytic approaches supports the conclusion that stroke risk among untreated patients with AF is not evenly distributed, but clusters within recognisable high-risk phenotypes.

Other vascular risk factors (cardiac failure, hypertension, diabetes and vascular disease) showed a strong univariable association; however, these associations attenuated after adjustment, suggesting they were partly explained by co-existing high-risk clinical features.

Although Māori ethnicity was associated with higher adjusted odds of IS/TIA compared with Europeans, the estimate was imprecise and did not reach statistical significance. This likely reflects limited power, given the modest number of Māori cases, as well as the restriction to non-anticoagulated patients, which excludes higher-risk patients. We previously reported population-level data confirming increased IS risk in Māori.¹¹

Strengths of this study include the collection of individual-level patient data and adjudicated IS/TIA outcomes by the ARCOS diagnostic committee. Statistical analyses were guided by existing evidence and prespecified hypotheses, and key findings were further examined using sensitivity analyses based on IPW to assess robustness to measured confounding.

Limitations include the case–control design with cumulative sampling, which precludes estimation of absolute risk or stroke score performance (e.g., CHA₂DS₂ VA) and limits causal inference. Although ARCOS employs comprehensive case ascertainment methods, a small number of IS/TIA events may not be captured, particularly minor events not presenting to medical attention or deaths not attributed to stroke. This could result in limited misclassification of controls, although the direction of bias would be towards the null. AF classification was based on hospital records without access to primary care data, which may have resulted in some non-differential misclassification. Restriction to untreated patients introduces selection bias and limits generalisability, as this group is unlikely to represent the broader AF population and may reflect clinician, patient or system-level factors influencing anticoagulation decisions. The retrospective design and available dataset precluded detailed assessment of the clinical, patient-level and clinician-level reasons for non-anticoagulation. Although IPW was used to address measured confounding, residual confounding cannot be excluded. The generalisability of this study may be limited by the selected cohort of non-anticoagulated patients and by regional and ethnic differences in stroke risk and management practices. Finally, the modest sample size may mean that non-significant findings represent type II error.

In summary, among persistently untreated patients with AF, IS/TIA risk was associated with persistent/permanent AF, prior cerebrovascular disease, female sex and older age. These findings indicate that stroke risk in untreated AF is concentrated within identifiable high-risk sub-groups, highlighting missed opportunities for prevention in routine clinical practice.

View Appendix.