We retrospectively reviewed the data of patients with acute AF-stroke (within 7 days of stroke onset) who were admitted to Gil Medical Center between April 2019 and April 2025. We included patients with known or newly diagnosed AF during hospitalization and with diffusion-weighted imaging-positive lesions. Additionally, patients underwent FMD assessment during hospitalization when they were neurologically stable.

We excluded patients who met any of the following criteria: (1) poor initial magnetic resonance imaging (MRI) quality; (2) incomplete clinical data; (3) a significant (> 50%) stenosis of the corresponding artery; (4) clinical syndrome typical of small-vessel disease (SVD) associated with a subcortical lesion with maximal diameter < 15 mm; (5) presence of other causes, such as dissection, vasculitis, cancer-related stroke; (6) undetermined causes such as two or more mechanisms (i.e., SVD or large artery atherosclerosis combined with AF-stroke) or negative; and (7) inability to reliably undergo high-resolution ultrasonography due to lack of cooperation. This study was approved by the local ethics committee of Gil Medical Center, South Korea (GAIRB number: 2023-119).

Clinical data and risk factors were obtained from the patients’ medical records and the stroke registry database. The use of previous medications, including antithrombotics and statins, were also assessed. Neurological deficits due to stroke were evaluated at admission using the National Institutes of Health Stroke Scale (NIHSS) score.

The time of stroke onset was estimated from the first abnormal finding in patients with clear stroke onset. If patients were eligible for IV alteplase treatment, they underwent endovascular thrombectomy (EVT) after receiving a standard dose of IV alteplase (0.9 mg/kg of body weight; 10% administered as a bolus followed by 1-h infusion of the remaining dose). All patients underwent EVT in accordance with national clinical practice guidelines and local protocols. Electrocardiograms were obtained from the emergency department and monitored during admission to the stroke units. Transthoracic echocardiography (TTE) was performed in all patients during admission. Left ventricular ejection fraction (LVEF) was measured using the biplane Simpson method. The left atrial volume index (LAVI) was calculated using the biplane area–length method and indexed to the body surface area. Newly documented AF cases were also evaluated.

ENI was defined as a ≥2-point decrease in NIHSS or ≥ 1-point reduction in motor NIHSS items within 24 h in non-thrombolysis patients (16). For thrombolysis patients, ENI was defined as an 8-point reduction or an NIHSS score of 0–1 at 24 h (17).

All patients underwent MRI using a 3.0 T Philips scanner (Philips Healthcare, Eindhoven, Netherlands). Stroke lesion patterns were categorized as multiple scattered, single confluent, and multiple with confluent lesions. Multiple scattered lesions were defined as multiple small ischemic lesions of < 15 mm in diameter. Multiple with confluent lesions were defined by the presence of lesions ≥ 15 mm along with additional lesions of any size in any arterial territory. Single confluent lesion was defined as a single lesion of ≥15 mm (18). The lesion location was categorized as anterior, posterior, or both.

During admission, endothelial function was assessed by measuring the FMD of the brachial artery on the non-paretic arm in response to hyperemia. This was performed using high-resolution B-mode ultrasound (Aplio 50 Toshiba SSA-700) with a 7.5 MHz linear array transducer. Measurements were performed once the patients were neurologically stable. The participants were asked to abstain from food and caffeine for 12 h before the ultrasound scan, and no medication was administered from midnight until after the procedure.

The brachial artery was located longitudinally at a standardized point approximately 2–3 cm above the antecubital fossa. The baseline diameter and blood flow velocity were recorded continuously for 10 s. A blood pressure cuff was placed on the forearm and inflated to at least 50 mmHg above the participant’s systolic pressure for 5 min to induce occlusion. Twenty seconds before the cuff deflation, the diameter and blood flow velocity were recorded. After 5 min, the cuff was deflated to allow for reactive hyperemia in the hand and forearm, which led to vasodilation of the brachial artery. The percentage of FMD was calculated as 100 × [(peak diameter/baseline diameter)/ baseline diameter] (19).

The baseline characteristics were compared between the ENI and non-ENI groups. Continuous variables were expressed as the mean ± SD or median (IQR); categorical variables as frequencies and percentages. Group comparisons were performed using t-tests, Mann–Whitney U tests, or chi-square tests, as appropriate. In addition, the patients’ baseline characteristics were compared according to the tertiles of% FMD. The chi-square test (Fisher’s exact test) was used for categorical variables and analysis of variance (ANOVA) was used for continuous variables. ANOVA and the Kruskal–Wallis test were used to determine significant differences. The predictors of ENI were identified using logistic regression analysis. Variables with p < 0.1 in the univariable analysis, along with clinically relevant covariates, were included in multivariate Model 1. Model 2 was constructed as a more parsimonious model by prioritizing clinically meaningful variables. In this model, lesion pattern and lesion location were excluded to reduce the risk of overfitting, whereas treatment modality (procedure) was retained due to its significant association with baseline NIHSS and its potential confounding effect on early neurological outcomes.

Furthermore, to address the potential floor effect associated with low baseline NIHSS scores, we performed a sensitivity analysis restricted to patients with a baseline NIHSS >2. Because baseline NIHSS demonstrated a significant interaction with treatment modality, we also constructed two separate multivariable models to account for this relationship.

Subgroup analyses evaluated the FMD–ENI associations stratified by age, sex, stroke features, vascular risk factors, and cardiac parameters. Within each subgroup, logistic regression models were fitted to estimate the odds ratio (ORs) of FMD for ENI, and interaction effects were assessed by comparing the ORs between the subgroups. All statistical analyses were conducted using the IBM SPSS version 21.0 software (SPSS, Chicago, IL, USA). Statistical significance was set at p < 0.05.

The baseline characteristics stratified by ENI status are presented in Table 1. Those with ENI had higher NIHSS (7 [4–13] vs. 2 [1–5], p < 0.001), more confluent (38.7% vs. 25.5%) and scattered with confluent pattern (29.3% vs. 18.1%, p = 0.007), and higher %FMD (6.5% ± 2.5% vs. 5.3% ± 2.2%, p = 0.001).

Furthermore, patients in the 3rd tertile (highest %FMD > 6.30) were more female (69.6% vs. 64.0% vs. 46.0%; p = 0.023), had less history of previous statin use (32.1% vs. 42.0% vs. 17.5%; p = 0.016), had higher initial NIHSS score (3 [1–6] vs. 4 [1.5–7] vs. 5 [2–12]; p = 0.020), and had a higher proportion of ENI (35.7% vs. 38.0% vs. 57.1%; p = 0.035) (Supplementary Table 1).

Univariate analysis showed that the history of smoking, higher initial NIHSS score, confluent lesion and scattered with confluent lesion pattern, and higher FMD% were significantly associated with ENI.

In multivariate Model 1, which included variables with p < 0.1 from the univariate analysis along with clinically relevant covariates, current smoking (adjusted OR [aOR] = 4.532, 95% confidential interval [CI] 1.620–12.679; p = 0.004), higher initial NIHSS score (aOR = 1.329, 95% CI 1.195–1.477; p < 0.001), and higher FMD (%) (aOR = 1.179, 95% CI 1.005–1.382; p = 0.043) remained independently associated with ENI. In the multivariate Model 2, female sex (aOR = 0.472, 95% CI 0.227–0.982; p = 0.045), current smoking (aOR = 4.199, 95% CI 1.747–10.092; p = 0.001), and higher FMD (%) (aOR = 1.247, 95% CI 1.076–1.446; p = 0.003) were identified as independent factors associated with ENI (Table 2).

A sensitivity analysis restricted to patients with a baseline NIHSS >2 showed that a higher %FMD remained significantly associated with ENI (Supplementary Tables 2, 3).

Subgroup analyses were performed to examine whether the treatment effects of FMD differed across clinically relevant patient subgroups. Statistically significant interactions were identified for hypertension (interaction OR: 1.41, 95% CI: 1.07–1.89, p = 0.016) and hyperlipidemia (interaction OR: 1.72, 95% CI: 1.08–3.33, p = 0.018). In contrast, no significant interaction effects were observed in the other subgroups (Figure 2).