Nomogram for predicting left atrial appendage dense spontaneous echo contrast and/or thrombosis (dense spontaneous echo contrast/left atrial appendage thrombus): a comparison with CHADS₂ and CHA₂DS₂-VASc scores based on three-dimensional transesophageal echocardiography

Introduction

Non-valvular atrial fibrillation (NVAF) is one of the most common tachyarrhythmias in clinical practice. Its prevalence has increased fourfold over the past 50 years (1), with a global prevalence of approximately 2–4% in adults (2), rising significantly with age. Stroke is the leading cause of mortality and disability in patients with atrial fibrillation (AF), and about 20% of strokes are attributable to AF. The core mechanism is cardiogenic thrombosis formed in the left atrial appendage (LAA). Therefore, accurate identification of LAA thrombus (LAAT) and pre-thrombotic status [such as dense spontaneous echo contrast (SEC)] is crucial for stroke risk stratification (3).

At present, the CHADS₂ [congestive heart failure, hypertension, age ≥75 years, diabetes mellitus, prior stroke/transient ischemic attack (TIA) score] and CHA₂DS₂-VASc (congestive heart failure, hypertension, age ≥75 years, diabetes mellitus, prior stroke/TIA, vascular disease, age 65–74 years, sex category score) score is a mainstream tools for guiding anticoagulant therapy in patients with NVAF. However, they are mainly based on clinical factors such as age and underlying diseases, without incorporating the anatomical and functional characteristics of the LAA, resulting in limited predictive accuracy. As the primary site of thrombus formation in AF, the morphological features of the LAA, including its shape, size, and echogenicity, are closely related to thrombotic risk. Guidelines recommend transesophageal echocardiography (TEE) as the gold standard for detecting LAAT and SEC (4). Compared with conventional two-dimensional TEE (2D-TEE), three-dimensional TEE (3D-TEE) allows real-time, dynamic visualization of the 3D structure of the LAA and provides more accurate morphological parameters. This study aimed to establish an optimized prediction model using 3D-TEE combined with clinical indicators, to precisely evaluate the risk of dense SEC/LAAT in patients with NVAF, and to provide evidence for clinical prevention and treatment. We present this article in accordance with the TRIPOD reporting checklist (available at https://qims.amegroups.com/article/view/10.21037/qims-2026-0805/rc).

Methods

Study place and design

A total of 159 hospitalized patients with NVAF who underwent 3D-TEE examination at The Second Hospital of Lanzhou University from July 2024 to December 2025 were retrospectively enrolled.

This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The Ethics Committee of The Second Hospital of Lanzhou University granted ethical approval on 08 June 2023 (No. 2023A-435). The requirement for informed consent was waived due to the retrospective design and the use of deidentified data.

Definitions and criteria

Inclusion criteria: (I) diagnosis of NVAF confirmed by electrocardiography or ambulatory electrocardiography; (II) successful completion of 3D-TEE examination with image quality sufficient for analysis; (III) complete clinical data; (IV) age ≥18 years.

Exclusion criteria: (I) patients with valvular heart disease or congenital heart disease; (II) patients with severe hepatic or renal insufficiency or coagulation disorders; (III) poor ultrasound image quality that precluded parameter measurement.

According to 3D-TEE findings, patients were divided into the dense SEC/LAAT group (positive group, n=50) and the non-dense SEC/LAAT group (negative group, n=109).

All examinations were performed in strict accordance with the standardized protocols established by the joint guidelines of the American Society of Echocardiography and the European Association of Cardiovascular Imaging [2015]. The examinations were conducted using a Philips CVx ultrasound system equipped with an S5-1 transthoracic probe (frequency 1–5 MHz) and an X8-2t transesophageal probe. All procedures were performed by cardiac sonographers with at least 5 years of experience.

Transthoracic echocardiography (TTE)

The ascending aortic diameter (AO), left atrial anteroposterior diameter (LAD), left ventricular end-diastolic diameter (LVEDd), and right atrial transverse diameter (RAD) were measured. Left ventricular ejection fraction (LVEF) was calculated using the biplane Simpson’s method. Doppler measurements included mitral E-wave velocity, e’-sep, E/e’ ratio, mitral regurgitation (MR), tricuspid regurgitation (TR), and pulmonary artery systolic pressure (PASP). For patients with AF, the median value from 3–5 consecutive cardiac cycles was recorded.

3D-TEE

After overnight fasting, the transesophageal ultrasound probe was inserted into the esophagus. Following clear visualization of the LAA by 2D-TEE, the left atrium (LA) and LAA were carefully examined for the presence of thrombus (solid or hypoechoic masses) or dense SEC (smoky, abnormally dense echogenic flow signals) (5). The 3D-ZOOM mode was activated to acquire and store dynamic images over 5 cardiac cycles for analysis. Structural parameters including the maximum orifice diameter, minimum orifice diameter, perimeter, area, maximum depth, and volume of the LAA were measured.

The echogenicity and type of the LAA (chicken wing, cauliflower, windsock, cactus) (6) were observed and recorded, as well as LAA orifice morphology (circular shape, elliptical shape), number of lobes (single, two, three, four lobes), presence of prominent pectinate muscles, and apex orientation (anterior, posterior, superior, inferior).

All parameters were independently measured by two experienced echocardiographers, and the mean values were taken as the final results. All echocardiographers performing image acquisition and analysis were blinded to patients’ clinical data and final grouping results.

Dense SEC was graded according to established echocardiographic criteria: grade 3–4 smoky, dense echo signals occupying most of the LAA cavity were defined as dense SEC. For differential diagnosis, LAAT was identified as localized solid or hypoechoic masses with clear boundaries and independent mobility; prominent pectinate muscles presented as linear or nodular structures continuously connected to the LAA wall, while artifacts were excluded by adjusting imaging gain, depth and multi-plane dynamic observation.

Any discrepancies in image interpretation and parameter measurement between the two readers were resolved through group discussion and consensus agreement. Formal interobserver and intraobserver reproducibility analysis was not performed in this retrospective study; however, all assessments were completed by senior sonographers to minimize subjective bias.

Demographic characteristics (age, sex), type of AF (paroxysmal, persistent), treatment history (use of oral anticoagulants or not), comorbidities (hypertension, diabetes mellitus, coronary artery disease, heart failure, history of TIA), laboratory parameters [platelet, hemoglobin, D-dimer, international normalized ratio (INR), creatinine], and clinical scores [CHADS₂, CHA₂DS₂-VASc, HAS-BLED (hypertension, abnormal renal/hepatic function, stroke, bleeding history or predisposition, labile INR, elderly, drugs/alcohol concomitantly score), and European Heart Rhythm Association (EHRA) scores] were collected from the hospital electronic medical record system.

We collected general demographic data, medical history, laboratory indicators and echocardiographic parameters of all enrolled patients. For anticoagulation-related information, only the status of anticoagulant use was documented. Detailed data on anticoagulant agents, treatment course, adherence and pre-TEE drug interruption could not be obtained from the existing medical records.

Statistical analysis

Statistical analyses were performed using SPSS26.0 and R4.2.0 software. Continuous variables with normal distribution were expressed as mean ± standard deviation, and comparisons between groups were conducted using independent-samples t-test. For continuous variables without normal distribution, median (interquartile range) [M (P₂₅, P₇₅)] was used, and Wilcoxon rank-sum test was applied for intergroup comparisons. Categorical variables were presented as frequency (%), and the Chi-squared (χ²) test was used for group comparisons.

First, the variance inflation factor (VIF) was used to test for multicollinearity among variables (VIF <10 indicated no significant multicollinearity). Univariate logistic regression analysis was performed to screen potential predictors of dense SEC/LAAT formation (P<0.10), which were then included in multivariate logistic regression analysis to identify independent predictors and construct a predictive model (Model 1).

The receiver operating characteristic (ROC) curve and area under the curve (AUC) value were used to evaluate the discriminatory ability of the model. Calibration curves were used to assess model calibration, and decision curve analysis (DCA) and clinical impact curves were applied to evaluate the clinical utility of the model. A two-tailed P<0.05 was considered statistically significant. The procedures for model construction are detailed in Figure 1.

Figure 1 Flowchart of the study. CHA₂DS₂-VASc, congestive heart failure, hypertension, age ≥75 years, diabetes mellitus, prior stroke/TIA, vascular disease, age 65–74 years, sex category score. CHADS₂, congestive heart failure, hypertension, age ≥75 years, diabetes mellitus, prior stroke/TIA score. AUC, area under the receiver operating characteristic curve; DCA, decision curve analysis; LAA, left atrial appendage; LAAT, left atrial appendage thrombus; NVAF, non-valvular atrial fibrillation; SEC, spontaneous echo contrast; TEE, transesophageal echocardiography; TIA, transient ischemic attack.

Results

Patients characteristics

A total of 206 patients with NVAF who underwent TEE between July 2024 and December 2025 were initially enrolled. Among them, 47 patients were excluded due to unqualified conditions. Specifically, 7 patients with congenital heart disease, 5 with valvular heart disease, 9 with poor image quality, and 26 with incomplete clinical data. Finally, 159 eligible patients were included in the final analysis (Figure 1).

Of the 159 included patients, 50 (31.4%) were assigned to the dense SEC/LAAT positive group and 109 (68.6%) to the negative group. In the positive group, 6 patients (12.0%) had isolated LAAT, 5 patients (10.0%) presented with both dense SEC and LAAT, and the remaining 39 patients (78.0%) had isolated SEC. Representative 3D-TEE images of patients in the dense SEC/LAAT positive group and negative group (Figure 2A,2B), respectively.

Figure 2 Representative 3D-TEE images of patients in the non-dense SEC/LAAT group (A) and the dense SEC/LAAT group (B). 3D-TEE, three-dimensional transesophageal echocardiography; SEC/LAAT, spontaneous echo contrast/left atrial appendage thrombus.

Comparison of baseline characteristics between the positive and negative groups revealed no significant differences in age, CHA₂DS₂-VASc score, and CHADS₂ score between the two groups (P=0.095, 0.096, and 0.065, respectively). HAS-BLED score, LAD, RAD, E/e’ ratio, and PASP were significantly higher in the positive group than in the negative group (all P<0.05).

Among LAA-related continuous variables, maximum diameter, minimum diameter, area, volume, and perimeter were significantly larger in the positive group (all P<0.05). The prevalence of persistent AF, abnormal INR, D-dimer >0.550 mg/L, reduced ejection fraction (EF), TR, MR, and EHRA score ≥ IIb was significantly higher in the positive group (all P<0.05), and the incidence of congestive heart failure showed a borderline difference (P=0.072).

For LAA-related categorical variables, the proportion of poor echogenicity was significantly lower in the positive group (16.0% vs. 57.8%, P<0.001). There were significant differences in LAA orifice morphology and LAA type between the two groups (P=0.011 and P=0.028, respectively). Sex, hypertension, diabetes mellitus, coronary artery disease, TIA, PLT, hemoglobin (Hb), creatinine (Cr), antiarrhythmic drugs (AAD), LVEDd, anticoagulant use, LAA apex orientation, number of lobes, and prominent pectinate muscles showed no significant differences between the two groups (all P>0.05).

In this study, we only recorded whether patients received anticoagulant treatment. Detailed information including specific anticoagulant types, duration of medication, treatment adherence, and whether anticoagulants were interrupted before TEE was unavailable due to incomplete medical records.

Univariate and multivariate logistic regression analyses were performed to identify independent predictors of dense SEC/LAAT formation in NVAF patients. The results of the regression analyses are presented in Table 1. Univariate analysis identified multiple variables significantly associated with dense SEC/LAAT development (all P<0.05). Multivariate logistic regression analysis with forward stepwise selection finally identified four independent predictors significantly associated with dense SEC/LAAT formation: D-dimer >0.550 mg/L [odds ratio (OR) =7.805, 95% confidence interval (CI): 2.044–29.795, P=0.002], EHRA score ≥ IIb (OR =6.255, 95% CI: 1.458–26.833, P=0.013), LAA poor echogenicity (OR =23.037, 95% CI: 5.651–93.909, P<0.001), and LAA orifice morphology (OR =0.537, 95% CI: 0.296–0.975, P=0.041) (Table 1).

Construction, evaluation, and clinical utility analysis of a predictive model for dense SEC/LAAT formation

Construction and performance comparison of the predictive model: multivariate logistic regression analysis identified four independent predictors—D-dimer >0.550 mg/L, LAA orifice morphology, LAA poor echogenicity, and EHRA score ≥ IIb. The predictive performance of the developed model (Model 1) was compared with two traditional clinical risk scores: the CHADS₂ score (Model 2) and CHA₂DS₂-VASc score (Model 3).

The results showed that Model 1 achieved an AUC of 0.892 (95% CI: 0.839–0.945), which was significantly higher than that of Model 2 (AUC =0.629, 95% CI: 0.540–0.718) and Model 3 (AUC =0.606, 95% CI: 0.515–0.697) (P<0.05) (Figure 3). At the optimal cutoff value of 0.29, Model 1 exhibited a sensitivity of 0.82, specificity of 0.83, accuracy of 0.82, and negative predictive value of 0.91 (Table 2).

Figure 3 Receiver operating characteristic curves of predictions by the three models. Model 1, Self-developed Model. Model 2, CHADS₂ Score. Model 3, CHA₂DS₂-VASc Score. AUC, area under the receiver operating characteristic curve.

Calibration curve analysis demonstrated favorable consistency between the predicted probability and the actual observed frequency for Model 1, with the calibration curve closest to the ideal diagonal line. However, all models showed a slight tendency toward overestimation at high predicted probabilities (>0.6) (Figure 4).

Figure 4 Calibration curves of predictions by the three models. The dashed diagonal line indicates perfect consistency between the predicted probability and the observed probability (perfect calibration). The calibration curve of Model 1 is closest to the diagonal, indicating its optimal calibration. All models show a mild overestimation trend in the high-risk region (predicted probability >0.6). Model 1, Self-developed Model. Model 2, CHADS₂ Score. Model 3, CHA₂DS₂-VASc Score.

Given the lack of an independent external validation cohort, we performed 10-fold cross-validation for internal validation to assess the model’s generalizability and correct for overfitting optimism. The full dataset was randomly split into 10 mutually exclusive subsets with equal sample size; the model was trained on 9 subsets and tested on the remaining 1 subset, with this process repeated 10 times to ensure all samples were used for testing exactly once. The 10-fold cross-validation procedure was repeated 100 times to obtain stable estimates, and the optimism-corrected mean values of core performance metrics were calculated.

The calibration slope derived from cross-validation reflects the model’s shrinkage degree. An ideal calibration slope equals 1; a slope >1 indicates overfitting, where predicted risks are overly extreme (overestimated for high-risk patients and underestimated for low-risk patients) in external populations. The calibration intercept and maximum calibration error (E_max) were simultaneously calculated to comprehensively quantify the deviation between predicted probabilities and actual observed event rates.

The cross-validation results showed that the model maintained an acceptable discriminative ability for dense SEC/LAAT after correcting for overfitting, with a mean AUC of 0.868 and a mean Brier score of 0.127 (Table 3).

However, the calibration performance of the model in cross-validation revealed a notable overfitting tendency: the mean calibration slope was 4.175 (substantially deviated from the ideal value of 1), the mean calibration intercept was −0.285, and the mean E_max was 0.385. These results indicate that the model’s predicted probabilities tend to be extreme on unseen data, with significant deviation from the actual risk, and the model should not be directly applied to clinical decision-making without further recalibration or external validation. Box plots were used to visualize the distribution of key indicators during 10-fold cross-validation (Figure 5).

Figure 5 Box plots of core metrics derived from 10-fold cross-validation. Brier, Brier score; C (ROC), area under the receiver operating characteristic curve; E_avg, average calibration error; E_max, maximum calibration error; R², coefficient of determination.

Presentation and validation of the predictive model

Based on the four independent predictors, a nomogram was plotted to intuitively demonstrate the predictive model (Figure 6). According to the patient’s specific indicators, the user can obtain the corresponding score for each factor, calculate the total score by summation, and then project it onto the risk axis to obtain the individualized predicted risk probability of dense SEC/LAAT formation.

Figure 6 This nomogram integrates four independent predictive factors: D-dimer >0.550 mg/L, LAA orifice morphology, LAA poor echogenicity, and EHRA score ≥IIb. In use, the corresponding score (points) is determined on each variable axis according to the patient’s condition, and the total score (total points) is projected onto the risk axis at the bottom (risk of event) to read the individualized predicted risk probability. EHRA, European Heart Rhythm Association; LAA, left atrial appendage.

Model validation demonstrated that the predictive performance of the nomogram was consistent with that of the previous logistic regression model, good agreement was observed between the predicted probabilities and the actual observations in the development dataset, confirming the reliability of the model’s visualized form.

Comprehensive evaluation of model performance

The ROC curve and calibration curve of the established nomogram are displayed in Figures 7,8, respectively.

Figure 7 Receiver operating characteristic curve of the predictive model. The area under the curve (AUC, i.e., C-statistic) of this model is 0.89 (95% CI: 0.82–0.94), indicating that the model has excellent discriminative ability. The gray shaded area represents its 95% CI. AUC, area under the receiver operating characteristic curve; CI, confidence interval.

Figure 8 Decision curve analysis of the predictive model. This figure evaluates the net benefit of using the model to guide clinical decisions (Model) compared with default strategies at different threshold probabilities. The default strategies include “intervening in all patients” (all) and “not intervening in any patients” (none). When the threshold probability is in a wide range from approximately 3% to nearly 100%, the net benefit of applying this model is higher than that of the two default strategies, indicating its clinical utility. The gray dashed line (Model irrelevant) and dotted line (Model relevant) assist in interpretation.

Clinical utility analysis of the model

The apparent DCA performed on the full development dataset showed that within the clinically relevant threshold probability range of 10–50% (commonly used for thromboembolic risk intervention), the net benefit of using Model 1 for dense SEC/LAAT risk stratification was higher than the two extreme reference strategies of “treat all patients” and “treat none” (Figure 8).

However, it should be explicitly noted that this apparent DCA result was derived from the full training dataset, which may be affected by model overfitting and cannot fully represent the model’s real-world clinical net benefit.

The optimism-corrected DCA via 10-fold cross-validation showed that Model 1 still maintained a positive net benefit within the clinically relevant threshold range of 10–50%, which was consistent with the core finding of the apparent DCA. However, the magnitude of the net benefit was slightly reduced compared with the apparent DCA, and the threshold range with positive net benefit was narrower than that observed in the apparent analysis, further confirming the impact of overfitting on the model’s apparent clinical utility.

Clinical impact curve analysis was performed to further evaluate the model’s practical clinical application value. The curve intuitively displayed the relationship between the number of high-risk patients classified by Model 1 and the true positive cases with confirmed dense SEC/LAAT at different risk thresholds (Figure 9). This curve provides reference evidence for clinicians to choose reasonable intervention thresholds according to specific clinical scenarios, patient risk profiles, and medical resource allocation, helping to determine the optimal timing and target population for intervention.

Figure 9 Clinical impact curve of the predictive model. The red curve (number high risk) represents the number of patients classified as high-risk by the model at different risk thresholds. The blue curve (number high risk with event) represents the number of true positive patients who actually experienced dense SEC/LAAT among those classified as high-risk. This curve helps balance the benefits (detecting true positives) and costs (unnecessary intervention for false positives) of intervention at different thresholds. SEC/LAAT, spontaneous echo contrast/left atrial appendage thrombus.

It should be emphasized that the model is only intended to serve as a supplementary tool for dense SEC/LAAT risk stratification in clinical research settings, and should not be used to directly guide anticoagulation treatment decisions without prospective external validation and clinical recalibration.

Discussion

Previous studies have demonstrated that the risk of ischemic stroke in patients with NVAF is five times higher than that in patients with sinus rhythm, making it an important risk factor for stroke. The emboli of cardiogenic stroke are mainly derived from the LA and LAA, and more than 90% of thromboembolic events in NVAF patients originate from the LAA (7). Therefore, the early diagnosis of LAAT holds important clinical significance for patients and can effectively reduce the public health burden. The guidelines recommend TEE as the gold standard for evaluating LAA morphology and structure, and the application of 3D-TEE has further improved the accuracy of such assessments. The CHA₂DS₂-VASc score, currently widely used in clinical practice, is more comprehensive than the CHADS₂ score but both scores were primarily developed for thromboembolic stroke risk stratification, not specifically for predicting dense SEC/LAAT detected by 3D-TEE, and thus have inherent limitations in this specific imaging endpoint and may miss some high-risk patients for SEC/LAAT. Studies have reported that although the CHA₂DS₂-VASc score is a convenient and widely used tool for overall thromboembolic risk assessment, its predictive value for dense SEC/LAAT is limited, with a C-statistic of only 0.606 and relatively poor predictive efficacy for this specific endpoint (8). Based on LAA structural parameters quantified by 3D-TEE and clinical indicators, this study successfully constructed a nomogram prediction model for dense SEC/LAAT formation in NVAF patients, which achieved an AUC of 0.892 and an accuracy of 82%, providing a new tool for clinical dense SEC/LAAT risk assessment.

Studies have found an association between blood biomarkers and dense SEC/LAAT. A study by Habara indicated that D-dimer has diagnostic discriminatory power, with an OR of 97.6 (95% CI:17.3–595.8) for LAAT and a negative predictive value of 97% (9). Additional research has confirmed a correlation between LAAT formation and D-dimer levels in NVAF patients (10). In a study by Liu, D-dimer levels were significantly higher in the LAAT group than in the non-LAAT group (180.0 vs. 90.0 µg/L, P=0.003) (11). Consistent with previous findings, the present study identified D-dimer >0.550 mg/L as an important independent predictor of dense SEC/LAAT (OR =7.806, P=0.003). Elevated D-dimer levels indicate the activation of coagulation and hyperfibrinolysis in the body, indirectly reflecting the risk of thrombus formation. The combined application of D-dimer with LAA structural parameters can improve the accuracy of dense SEC/LAAT prediction.

The clarity of LAA echogenicity was determined based on the presence or absence of SEC in the LAA. In TEE examinations, LAA was considered to have good echogenicity if no obvious smoky echoes were observed after increasing the gain (12). A study with a mean follow-up of 17.5 months in NVAF patients found that the incidence of thromboembolic events was 12% in patients with LAA SEC, compared with only 3% in those without SEC, confirming that LAA SEC is an independent risk factor for thromboembolism in NVAF patients. In the present study, LAA poor echogenicity was identified as the independent predictor with the strongest association with dense SEC/LAAT (OR =23.037, P<0.001). Poor echogenicity reflects the degree of blood stasis and prethrombotic state in the LAA, which may be related to decreased LAA systolic function and slowed blood flow velocity, creating favorable conditions for thrombus formation. 3D-TEE enables more accurate assessment of LAA echogenicity, providing technical support for the high predictive efficacy of the constructed model.

Previous studies on LAA orifice morphology have reported conflicting findings: Lee et al. (13) showed that a larger LAA orifice diameter is associated with cardiogenic embolic stroke and TIA, while Huang et al. (14) and Khurram et al. (15) emphasized other factors such as a smaller orifice diameter and abundant trabeculations. The present study found no significant differences in LAA depth or orifice diameter. Notably, LAA orifice morphology was identified as a negative predictor of dense SEC/LAAT in this study (OR =0.537, P=0.041). LAA orifice morphology may be associated with hemodynamic characteristics, and an optimal morphological structure can reduce thrombus risk, with the core mechanism linked to the uniformity of shear stress distribution. Studies have confirmed that a regular and symmetrical LAA morphology can reduce blood flow vortex and shear stress disturbance (16); clinical research has also found that the incidence of thrombus is significantly lower in patients with a circular LAA orifice than in those with a complex orifice morphology, indirectly supporting the role of symmetrical structures in optimizing hemodynamics (17). This phenomenon is consistent with the hypothesis that a circular ostium equalizes the wall shear stress of blood flow entering the LAA, which is speculated to exert an antithrombotic effect by reducing local blood stasis and endothelial damage.

The EHRA score is the core indicator for assessing the symptom severity of patients with NVAF, which is classified into four grades. A number of clinical studies have confirmed that an elevated EHRA score is significantly associated with an increased risk of dense SEC/LAAT formation, and patients with high grades (Grade III–IV) in particular constitute a key high-risk population (18). The European guidelines for the management of AF recommend the incorporation of the EHRA score into the thrombus risk stratification system for combined use with the CHA₂DS₂-VASc score. For patients with a high EHRA score, priority should be given to TEE to assess the risk of LAAT, so as to guide individualized anticoagulant therapy (19). In this study, patients were stratified by whether their EHRA score was ≥ Class IIB. The results demonstrated that the EHRA score was an independent predictor of dense SEC/LAAT (OR =6.256, P=0.014), indicating that patients with more severe symptoms face a higher risk of dense SEC/LAAT formation. This may be associated with a higher AF burden and more significant impairment of LAA function, highlighting the need for heightened clinical attention to thrombus risk screening in patients with a high EHRA score.

Beyond LAA anatomical morphology evaluated by 3D-TEE, two-dimensional speckle tracking echocardiography (2D-STE) has emerged as a valuable non-invasive technique to assess LA mechanical function. LA reservoir strain, a core parameter of STE, quantitatively reflects LA myocardial deformation capacity. Accumulated evidence has proven that impaired LA reservoir strain is strongly correlated with LAA blood stasis, dense SEC and LAAT, and its predictive value is supplementary to traditional clinical scores (20-22). LA dysfunction often precedes morphological changes and is an early pathological manifestation of AF-related atrial cardiomyopathy. Combining 3D-TEE LAA anatomical parameters with 2D-STE LA functional indices can realize multimodal echocardiographic assessment, further optimize thrombotic risk stratification and individualized anticoagulation management for NVAF patients. Regrettably, LA strain and other STE functional parameters were not included in the current model, which will be improved in our subsequent research.

The innovation of this study lies in the incorporation of 3D-TEE quantified LAA structural parameters into the prediction model, which compensates for the limitations of traditional scoring systems. The nomogram model is intuitive and user-friendly, enabling clinicians to rapidly calculate the risk of dense SEC/LAAT formation in patients and providing a quantitative basis for anticoagulant therapy decision-making. In addition, collinearity analysis revealed strong collinearity between LAA area and perimeter; future studies may consider optimizing the model structure using methods such as principal component analysis.

Notably, several quantitative 3D-TEE LAA structural parameters (including maximum diameter, minimum diameter, area, volume, and perimeter) showed significant intergroup differences in univariate analysis but were not retained in the final predictive model. This discrepancy was mainly attributable to multicollinearity and attenuated independent predictive values during multivariate adjustment. These morphological measurements are highly intercorrelated with one another and closely correlated with LAA orifice morphology and poor echogenicity. After controlling for confounding variables in the stepwise multivariate logistic regression, the significant predictive effects of these quantitative geometric parameters disappeared, indicating that they were not independent risk factors for dense SEC/LAAT. Excluding these highly collinear variables also reduced model redundancy, avoided overfitting, and improved the stability and generalizability of the final model.

This study has several key limitations that must be explicitly acknowledged. First, this was a single-center retrospective study, with inherent selection, information, and residual confounding biases. We mitigated these biases via strict pre-specified inclusion/exclusion criteria, consecutive patient enrollment, and standardized image evaluation by two independent echocardiographers. Second, the small overall sample size (n=159), with an even smaller dense SEC/LAAT positive group (n=50), reduces statistical power and increases the risk of false negatives (type II error). Third, we used a composite endpoint of dense SEC/LAAT to improve statistical power, meaning the model reflects overall cardiogenic thromboembolism risk rather than specific predictors of isolated LAAT. Fourth, this study did not include an external validation cohort; while 10-fold cross-validation was performed to assess internal validity, the cross-validated calibration slope of 4.175 indicates notable overfitting. The model’s predicted probabilities should therefore be interpreted with caution, and it should not be used for clinical decision-making without further recalibration or external validation. Fifth, the study lacked detailed anticoagulation data. We could only confirm whether patients received oral anticoagulants, while specific anticoagulant agents, treatment duration, medication adherence and pre-TEE anticoagulation interruption status were not available due to incomplete medical records. Since anticoagulation therapy strongly affects the formation of dense SEC/LAAT, this deficiency may have introduced residual confounding. Sixth, there is a possibility of misclassification of the dense SEC/LAAT endpoint due to the subjective nature of 3D-TEE image evaluation, despite standardized assessment criteria. Seventh, formal interobserver and intraobserver reproducibility testing was not performed for 3D-TEE parameters, which may introduce minor subjective interpretation bias despite standardized assessment by experienced sonographers. Finally, all these limitations result in limited generalizability of the study findings and the constructed nomogram. Future multicenter prospective studies are needed to externally validate the model, and penalized regression methods or bootstrap optimism correction should be considered to improve model generalizability.

Conclusions

The logistic regression model developed in this study, which incorporates 3D-TEE-derived LAA structural features and clinical indicators, demonstrates acceptable discriminative ability for predicting dense SEC/LAAT in patients with NVAF, with a cross-validated AUC of 0.868 in internal validation. The model complements the limitations of traditional stroke risk scores in predicting this specific imaging endpoint, and may serve as a supplementary tool for early risk stratification of dense SEC/LAAT in clinical practice. However, notable evidence of overfitting was observed in the model’s calibration performance, with a cross-validated calibration slope of 4.175. Therefore, external validation in an independent patient cohort or model recalibration is strongly recommended before the model is applied to routine clinical decision-making. Future multicenter prospective studies are warranted to further validate the model’s generalizability and optimize its predictive performance.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://qims.amegroups.com/article/view/10.21037/qims-2026-0805/rc

Data Sharing Statement: Available at https://qims.amegroups.com/article/view/10.21037/qims-2026-0805/dss

Funding: None.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://qims.amegroups.com/article/view/10.21037/qims-2026-0805/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The Ethics Committee of The Second Hospital of Lanzhou University granted ethical approval on 08 June 2023 (No. 2023A-435). The requirement for informed consent was waived due to the retrospective design and the use of deidentified data.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Schnabel RB, Yin X, Gona P, Larson MG, Beiser AS, McManus DD, Newton-Cheh C, Lubitz SA, Magnani JW, Ellinor PT, Seshadri S, Wolf PA, Vasan RS, Benjamin EJ, Levy D. 50 year trends in atrial fibrillation prevalence, incidence, risk factors, and mortality in the Framingham Heart Study: a cohort study. Lancet 2015;386:154-62. [Crossref] [PubMed]
2. Holmes DR Jr, Korsholm K, Rodés-Cabau J, Saw J, Berti S, Alkhouli MA. Left atrial appendage occlusion. EuroIntervention 2023;18:e1038-65. [Crossref] [PubMed]
3. Sun S, Ma C, Li Y, Lin J, Qian X. Development and validation of a nomogram superior to CHADS(2) and CHA(2)DS(2)-VASc models for predicting left atrial appendage dense spontaneous echo contrast/left atrial appendage thrombus. J Thorac Dis 2024;16:2102-14. [Crossref] [PubMed]
4. Hindricks G, Potpara T, Dagres N, Arbelo E, Bax JJ, Blomström-Lundqvist C, et al. 2020 ESC Guidelines for the diagnosis and management of atrial fibrillation developed in collaboration with the European Association for Cardio-Thoracic Surgery (EACTS): The Task Force for the diagnosis and management of atrial fibrillation of the European Society of Cardiology (ESC) Developed with the special contribution of the European Heart Rhythm Association (EHRA) of the ESC. Eur Heart J 2021;42:373-498. Erratum in: Eur Heart J 2021;42:507 Erratum in: Eur Heart J 2021;42:546-7. Erratum in: Eur Heart J 2021;42:4194.
5. Spethmann S, Stüer K, Diaz I, Althoff T, Hewing B, Baumann G, Dreger H, Knebel F. Left atrial mechanics predict the success of pulmonary vein isolation in patients with atrial fibrillation. J Interv Card Electrophysiol 2014;40:53-62. [Crossref] [PubMed]
6. Wang X, Xu X, Wang W, Huang H, Liu F, Wan C, Yao Q, Li H, Zhang Z, Song Z. Risk factors associated with left atrial appendage thrombosis in patients with non-valvular atrial fibrillation by transesophageal echocardiography. Int J Cardiovasc Imaging 2023;39:1263-73. [Crossref] [PubMed]
7. Cresti A, García-Fernández MA, Sievert H, Mazzone P, Baratta P, Solari M, Geyer A, De Sensi F, Limbruno U. Prevalence of extra-appendage thrombosis in non-valvular atrial fibrillation and atrial flutter in patients undergoing cardioversion: a large transoesophageal echo study. EuroIntervention 2019;15:e225-30. [Crossref] [PubMed]
8. Lip GY, Nieuwlaat R, Pisters R, Lane DA, Crijns HJ. Refining clinical risk stratification for predicting stroke and thromboembolism in atrial fibrillation using a novel risk factor-based approach: the euro heart survey on atrial fibrillation. Chest 2010;137:263-72. [Crossref] [PubMed]
9. Habara S, Dote K, Kato M, Sasaki S, Goto K, Takemoto H, Hasegawa D, Matsuda O. Prediction of left atrial appendage thrombi in non-valvular atrial fibrillation. Eur Heart J 2007;28:2217-22. [Crossref] [PubMed]
10. Zhong J, Xing LM. Predictive value of echocardiography combined with CT angiography for left atrial appendage thrombosis in patients with non-valvular atrial fibrillation. Eur Rev Med Pharmacol Sci 2023;27:10213-20. [Crossref] [PubMed]
11. Liu Y, Zhu D, Xiao Y, Zhu Y, Zhou Q, Ren L, Chen L. Risk factors of left atrial appendage thrombus in patients with non-valvular atrial fibrillation. Open Med (Wars) 2021;16:361-6. [Crossref] [PubMed]
12. Leung DY, Black IW, Cranney GB, Hopkins AP, Walsh WF. Prognostic implications of left atrial spontaneous echo contrast in nonvalvular atrial fibrillation. J Am Coll Cardiol 1994;24:755-62. [Crossref] [PubMed]
13. Lee Y, Park HC, Lee Y, Kim SG. Comparison of Morphologic Features and Flow Velocity of the Left Atrial Appendage Among Patients With Atrial Fibrillation Alone, Transient Ischemic Attack, and Cardioembolic Stroke. Am J Cardiol 2017;119:1596-604. [Crossref] [PubMed]
14. Huang K, Wen H, Liu C. Association Between Ischemic Stroke and Left Atrial Appendage Morphology in Patients With Atrial Fibrillation and Low CHA2DS2-VASc Scores. Tex Heart Inst J 2022;49:e207544. [Crossref] [PubMed]
15. Khurram IM, Dewire J, Mager M, Maqbool F, Zimmerman SL, Zipunnikov V, Beinart R, Marine JE, Spragg DD, Berger RD, Ashikaga H, Nazarian S, Calkins H. Relationship between left atrial appendage morphology and stroke in patients with atrial fibrillation. Heart Rhythm 2013;10:1843-9. [Crossref] [PubMed]
16. Masci A, Barone L, Dedè L, Fedele M, Tomasi C, Quarteroni A, Corsi C. The Impact of Left Atrium Appendage Morphology on Stroke Risk Assessment in Atrial Fibrillation: A Computational Fluid Dynamics Study. Front Physiol 2018;9:1938. [Crossref] [PubMed]
17. Nguyen DNQ, Ho DT, Tran TNH. Impact of Left Atrial Appendage Morphology and Function on Thrombosis Risk in Acute Ischemic Stroke: Insights from Transesophageal Echocardiography. Med Sci (Basel) 2025;13:63. [Crossref] [PubMed]
18. Wegner FK, Radke R, Ellermann C, Wolfes J, Fischer AJ, Baumgartner H, Eckardt L, Diller GP, Orwat S. Incidence and predictors of left atrial appendage thrombus on transesophageal echocardiography before elective cardioversion. Sci Rep 2022;12:3671. [Crossref] [PubMed]
19. Van Gelder IC, Rienstra M, Bunting KV, Casado-Arroyo R, Caso V, Crijns HJGM, et al. 2024 ESC Guidelines for the management of atrial fibrillation developed in collaboration with the European Association for Cardio-Thoracic Surgery (EACTS). Eur Heart J 2024;45:3314-414. [Crossref] [PubMed]
20. Kupczynska K, Michalski BW, Miskowiec D, Kasprzak JD, Wejner-Mik P, Wdowiak-Okrojek K, Lipiec P. Association between left atrial function assessed by speckle-tracking echocardiography and the presence of left atrial appendage thrombus in patients with atrial fibrillation. Anatol J Cardiol 2017;18:15-22. [Crossref] [PubMed]
21. Sonaglioni A, Lombardo M, Nicolosi GL, Rigamonti E, Anzà C. Incremental diagnostic role of left atrial strain analysis in thrombotic risk assessment of nonvalvular atrial fibrillation patients planned for electrical cardioversion. Int J Cardiovasc Imaging 2021;37:1539-50. [Crossref] [PubMed]
22. Gorcsan J 3rd. Can Left Atrial Strain Forecast Future Fibrillation? JACC Cardiovasc Imaging 2021;14:145-7. [Crossref] [PubMed]