Data supporting the findings of this study are available from the corresponding author upon reasonable request.

This retrospective study used a prospective registry from a single tertiary center. Between January 2012 and December 2020, 6,489 consecutive patients with acute cerebral infarction or transient ischemic attack (TIA) within 7 days of stroke onset were admitted to the Severance Stroke Center and prospectively registered in the Specialized Multi-center Attributed Registry of sTroke (SMART) Registry. During admission, all patients were evaluated and managed using a standardized protocol that included computed tomography, magnetic resonance imaging, carotid Doppler, transcranial Doppler, 12-lead electrocardiography (ECG), and standard blood tests. Cardiac embolic sources were evaluated using transthoracic echocardiography, transesophageal echocardiography (TEE), and 24-h Holter monitoring. Additionally, a continuous ECG was performed during the stroke unit stay. This study was approved by the Institutional Review Board of Yonsei University College of Medicine (approval number: 4-2022-0518), and the requirement for informed consent was waived owing to the retrospective nature of the study.

To determine the presence of potential sources of cardiac embolism, including LA/LAA thrombi, an echocardiographic study was routinely performed during admission. At our stroke center, TEE was also a part of the routine examination unless it could not be performed because of either the patient’s condition or failure to obtain informed consent. The TEE was performed by an experienced cardiologist. The TEE machine used a multiplane 5 MHz transducer attached to the tip of a gastroscope. Local pharyngeal anesthesia with 10% topical lidocaine was administered before the start of the examination. Transgastric, midesophageal, and basal views were recorded in a motion picture format. The presence of an intracardiac mass inside the left atrium and the left atrium appendage was also assessed. The Valsalva maneuver was performed with intravenous agitated saline injection to assess the presence of a right-to-left shunt.

In addition, between July 2006 and January 2017, multi-detector coronary computed tomography (MDCT) was consecutively performed for patients meeting any one of the following criteria [9] : (1) presence of atherosclerosis in cerebral arteries; (2) presence of two or more risk factors for coronary artery disease, such as hypertension, diabetes mellitus, dyslipidemia, smoking, and central obesity; and (3) men aged over 45 years and women aged over 55 years. MDCT was not performed when (1) high pulse rates were not controlled with beta-blockers at the time scheduled for MDCT (pulse rate >65 beats per minute), (2) the patient could not tolerate the study, (3) renal function was impaired with an estimated globular filtration rate of <60 mL/min/1.73 m², or (4) informed consent could not be obtained. Previously, MDCT was reported to have a sufficiently high detection rate to identify various cardioembolic sources, including LA/LAA thrombi [10].

We collected data on vascular risk factors, such as hypertension, diabetes, dyslipidemia, coronary artery occlusive disease, congestive heart failure, and atrial fibrillation type (paroxysmal or persistent/permanent). Atrial fibrillation was defined as paroxysmal when arrhythmia spontaneously resolved within 7 days. Stroke severity was determined using the National Institutes of Health Stroke Scale (NIHSS), whereas functional status was assessed using the modified Rankin Scale (mRS). Using TEE with or without MDCT, the ejection fraction, early mitral inflow velocity to mitral annular early diastolic velocity (E/e’) ratio, and presence of aortic atheroma were determined for each patient. Aortic atheroma was defined as the presence of atherosclerosis in the aortic arch or ascending aorta. The presence of relevant arterial occlusion was assessed using magnetic resonance angiography and/or computed tomography angiography. Relevant arterial occlusion was defined as an occlusive arterial lesion supplying an acute-infarcted area. The reperfusion therapy included intravenous alteplase infusion and endovascular thrombectomy.

The main outcome measured in this study was poor outcome, defined as an mRS score of >3 at 90 days after the index stroke event [11]. Furthermore, we also investigated the association between LA/LAA thrombus and outcomes defined as an mRS score of >1 or >2. In addition, we collected data on the occurrence of ischemic stroke (including TIA), hemorrhagic stroke (including intracerebral hemorrhage, intraventricular hemorrhage, subarachnoid hemorrhage, and subdural hemorrhage), myocardial infarction, bleeding events, major adverse cardiovascular events (MACE; a composite of stroke [ischemic or hemorrhagic], myocardial infarction, and cardiovascular death), and all-cause mortality during the 90-day follow-up period. Stroke or TIA was determined by a stroke specialist based on the clinical and radiological findings. Myocardial infarction was defined as the diagnosis of acute myocardial infarction based on clinical symptoms, cardiac enzymes, ECG, and imaging studies, such as coronary angiography. All bleeding events were recorded, including gastrointestinal, respiratory, and muscular bleeding requiring transfusion or hospitalization. Stroke neurologists and research nurses regularly contacted the patients or their caregivers during follow-up via regular face-to-face visits or telephone interviews, with or without medical chart review, to investigate outcomes, including mortality. In addition, we investigated the NIHSS score at discharge, occurrence of early neurological deterioration (END) during hospitalization, and duration (days) of hospital stay. END was defined as worsening of the total NIHSS score by more than 2 or any worsening of the NIHSS subscale on motor function of the arms or legs during admission. When patients with END underwent brain magnetic resonance imaging, we determined the infarction growth or the presence of new ischemic lesions using brain diffusion-weighted imaging performed at admission and at the time of END.

Descriptive statistics for the variables included in this study were analyzed. Categorical values are reported as counts with proportions and continuous variables as medians with interquartile ranges (IQRs). A comparative statistical analysis was performed between patients with and without LA/LAA thrombi. The chi-square test was performed for categorical variables and the Mann–Whitney U test was performed for continuous variables. Multivariable logistic or ordinal regression analysis was performed to assess the impact of LA/LAA thrombi on outcomes after eliminating the effect of covariates. Cox proportional hazards regression analysis was used to compare the events that occurred during follow-up. Data for the E/e’ ratio (48 patients, 7.6%) and ejection fraction (10 patients, 1.6%) were missing. Missing data were imputed using predictive mean-matching.

Furthermore, to reduce the potential effects of selection bias and confounding factors, we estimated the propensity scores for all patients enrolled for performing inverse probability of treatment weighting in this retrospective cohort study. Propensity scores were calculated for each patient using a logistic regression model including predetermined variables such as age (>75 years), sex, history of hypertension, diabetes, coronary artery occlusive disease, dyslipidemia, congestive heart failure, premorbid disability (mRS >3), glomerular filtration rate (>30 mL/min/1.73 m²), ejection fraction (>40%), E/e’ ratio (>15), presence of aortic atheroma, persistent/permanent atrial fibrillation, initial NIHSS score (>10), treatment with intravenous alteplase injection, and endovascular thrombectomy. The nearest neighborhood method was used, with a ratio of 1:2. Inverse probability of treatment weighting was calculated by using an inverse propensity score [12]. Symmetrical trimming was performed with a delta of 0.067, which is a previously proposed optimal cutoff, to achieve the final weighted dataset [13]. Balance was assessed using the standardized mean difference between the two groups before and after weighting. A 10% cutoff was used to define a significant imbalance [14]. P-values were two-sided. Multivariable analysis was performed to obtain the odds ratio for the weighted dataset. All statistical analyses were performed using R 4.0.0 (R Foundation for Statistical Computing, Vienna, Austria) [15].

Of the 6,489 patients admitted during the study period, we excluded 5,139 (79.2%) without NVAF, 708 (10.9%) who did not undergo TEE or MDCT, and 11 (0.2%) who were not followed up. Finally, 631 patients (9.7%) were included in this study (Figure 1). The median age of the included patients was 78.0 (IQR 70.0 to 83.0) years, and 57.7% of the patients were males. Compared with the patients excluded from this study, those included were older and more likely to have a history of hypertension, coronary artery occlusive disease, congestive heart failure, aortic atheroma, and worse glomerular filtration rate, and had a higher initial NIHSS score (Supplementary Table 1).

There were 68 (10.7%) patients with LA/LAA thrombus (56 observed on TEE and 26 observed on MDCT). Patients with LA/LAA thrombus were more likely to have congestive heart failure and persistent/permanent atrial fibrillation (22.1% vs. 5.7% and 95.6% vs. 73.5%, standardized mean difference 42.7% and 10.8%, respectively) (Table 1) than those without LA/LAA thrombus. Upon discharge, patients with LA/LAA thrombus were significantly more likely to be prescribed with a vitamin K antagonist than those without thrombus (75.0% vs. 43.2%, P<0.001).

The distribution of 3-month mRS scores is shown in Figure 2. Patients were likely to have a poor outcome when an LA/LAA thrombus was observed (42.6% vs. 17.4%, P<0.001) (Table 2). In the univariable analysis, a poor outcome was associated with the presence of LA/LAA thrombus along with age, male sex, history of diabetes, coronary artery occlusive disease, worse glomerular filtration rate, premorbid disability, high initial NIHSS score, and E/e’ ratio (>15) (all P<0.05). Multivariable logistic regression analysis adjusted for these significant factors on univariable analysis showed that LA/LAA thrombus was an independent determinant of a poor outcome (odds ratio [OR] 3.73, 95% confidence interval [CI] 2.03–6.83, P<0.001) (Supplementary Table 2).

In addition, inverse probability of treatment weighting analysis resulted in effective sample sizes of 59.0 for the group with LA/LAA thrombus, and 346.5 for the group without LA/LAA thrombus (Table 1). All the characteristics of the two groups were wellbalanced, as the standardized mean differences were not significant for any of the variables (Table 1 and Figure 3). Inverse probability of treatment weighting analysis demonstrated that LA/LAA thrombus was also independently and significantly associated with poor outcome (OR 4.05, 95% CI 1.98–8.32, P<0.001) (Table 3). LA/LAA thrombus was also an independent determinant when poor outcome was defined as mRS >2 (OR 2.55, 95% CI 1.31–4.99, P=0.006) but not on poor outcome with mRS >1 (OR 1.53, 95% CI 0.81–2.88, P=0.194) (Table 3). Multivariable ordinal regression analysis showed that LA/LAA thrombus was a significant factor for a shift in mRS score at 90 days (OR 1.69, 95% CI 1.01–2.83, P=0.046) (Supplementary Table 3). However, Cox proportional hazards regression analysis showed that LA/LAA thrombus was not associated with the occurrence of ischemic stroke, bleeding events, MACE, or all-cause mortality, while the risk of hemorrhagic stroke and myocardial infarction was different in the presence of LA/LAA thrombus (Supplementary Table 4).

When we compared the occurrence of END, NIHSS score at discharge, and relevant arterial occlusion between patients with and without LA/LAA thrombus (Table 2), patients with LA/LAA thrombus were more likely to have relevant arterial occlusive lesions than those without LA/LAA thrombus (32.4% vs. 20.6%, P=0.040 in the original population and 36.3% vs. 22.4%, P=0.047 in the inverse probability of treatment weighting population). For patients with relevant artery occlusion, carotid artery occlusion was more common in those with LA/LAA thrombus (Supplementary Table 5). Patients with relevant arterial occlusive lesions were more likely to have higher median NIHSS scores at stroke presentation than those without these lesions (median [IQR]: 8 [3–16] vs. 3 [1–8], P<0.001 in the original population and 11 [3–16] vs. 3 [1–8], P<0.001 in the inverse probability of treatment weighting population). In addition, the median NIHSS score at discharge was higher in patients with LA/LAA thrombus in the original population (2 [0–6.5] vs. 1 [0–3], P=0.013), and the median hospital stay (days) was longer before and after adjustment in patients with LA/LAA thrombus (8.5 [7–12] vs. 7 [5–9], P<0.001 in the original population and 8 [6.5–10] vs. 7 [5–9.5], P<0.001 in the inverse probability of treatment weighting population) than in those without LA/LAA thrombus. Among 59 patients with END, 36 underwent repeated brain diffusion-weighted imaging, and 50.8% had infarction growth. However, in terms of END during hospitalization, infarction growth, or MACE within 3 months, there was no significant difference between the two groups. When we investigated whether there was any difference in the outcomes for patients with LA/LAA thrombus after discharge according to the type of oral anticoagulant (vitamin K antagonists vs. direct oral anticoagulants), there was no difference between the groups (Supplementary Table 6).