Atrial fibrillation (AF) serves as a primary risk factor for cardioembolic stroke (Ozdemir et al., 2023), conferring a 3–5-fold increase in ischemic stroke risk (Maida et al., 2020). Furthermore, AF-associated AIS exhibits higher recurrence rates, greater disability burdens, and elevated mortality compared to other stroke subtypes (Choi et al., 2023). Notably, NVAF predominantly observed in aging populations, constitutes approximately 85% of global AF cases and NVAF-associated AIS accounts for over 50% of total stroke incidents (Lippi et al., 2021; Escudero-Martínez et al., 2023).

The pathophysiology underlying the increased stroke risk in AF is multifaceted. The irregular and chaotic atrial contraction characteristic of AF leads to blood stasis within the atrial chambers, particularly in the left atrial appendage (LAA), creating an environment conducive to thrombus formation (Bisson et al., 2018). This hemodynamic disturbance, coupled with endothelial dysfunction and a hypercoagulable state, constitutes Virchow’s triad for thrombogenesis in AF patients (Spartera et al., 2023). Additionally, AF is associated with systemic inflammation and oxidative stress, further exacerbating thrombotic risk through various molecular pathways including platelet activation, tissue factor expression, and disruption of the endothelial glycocalyx (Chousou et al., 2023).

While the CHA₂DS₂-VASc score remains the clinical benchmark for stratifying ischemic stroke risk in AF patients (Maheshwari et al., 2019), its predictive utility is constrained by a primary focus on clinical comorbidities, which have limited multi-dimensional predictive capacity. The score predominantly incorporates demographic factors and comorbid conditions-including congestive heart failure, hypertension, age, diabetes, prior stroke/TIA, vascular disease, and sex-without considering important physiological, anatomical, and hemodynamic parameters that may significantly influence stroke risk. Furthermore, the dichotomous nature of the scoring system fails to capture the gradation of risk associated with continuous variables such as left atrial size or biomarker levels.

Emerging evidence from recent studies demonstrates that ischemic stroke pathogenesis in NVAF patients involves multifactorial mechanisms spanning cardiac anatomical features, functional parameters, electrophysiological abnormalities, hemodynamic disturbances, and biomarker profiles (Wu et al., 2023). Left atrial enlargement and dysfunction have been significantly associated with thromboembolic events in AF patients, independent of traditional risk factors (Cho et al., 2021; Chung and Lee, 2022; Yu et al., 2025). Specific LAA morphologies and reduced LAA emptying velocities have demonstrated strong correlations with thrombus formation and subsequent embolic events (Tokunaga et al., 2020). Additionally, various biomarkers reflecting inflammation, coagulation activation, and myocardial injury have shown promise in enhancing stroke risk prediction models (Zhou et al., 2020). Among these, D-dimer—a marker of fibrin turnover and thrombus formation—has been validated as an independent predictor of ischemic stroke in patients with atrial fibrillation (Yuan et al., 2022; Shi et al., 2023). The white blood cell count, as a surrogate indicator of systemic inflammation, also correlates with atrial thrombogenesis by inducing endothelial dysfunction (Dolu et al., 2023; Li et al., 2022). Notably, clinical studies have demonstrated that the CHA₂DS₂-VASc/BS score, which incorporates B-type natriuretic peptide and cardiac troponin, yields improved predictive performance for stroke events in patients with non-valvular atrial fibrillation (Paulin et al., 2019). These findings further underscore the clinical value of integrating such multi-dimensional biomarkers into stroke risk prediction models.

Despite these advances in understanding the complex interplay of factors contributing to stroke risk in NVAF, current clinical risk stratification tools do not adequately integrate these multidimensional parameters. This gap in risk assessment potentially leads to suboptimal anticoagulation strategies in certain patient subgroups, highlighting the need for more comprehensive and precise risk prediction models.

This study aims to develop a refined predictive model for AIS in NVAF patients by systematic integration of clinical features, laboratory parameters, ambulatory electrocardiogram (AECG) recordings, echocardiographic findings, and Cardiac Multislice Spiral CT (Cardiac MSCT) metrics. Through this multifaceted approach, we seek to identify novel predictive factors and create a more accurate risk stratification tool to guide individualized preventive strategies and improve outcomes in this high-risk population.

Acute ischemic stroke (AIS) associated with atrial fibrillation (AF) is characterized by significantly higher rates of disability and mortality compared to other stroke subtypes (Elsheikh et al., 2024). This underscores the critical importance of early risk prediction and timely intervention to mitigate adverse outcomes in this high-risk population. The CHA₂DS₂-VASc score has been widely implemented in clinical practice for stratifying stroke risk in AF patients; however, its reliance on primarily categorical clinical variables presents substantial limitations. The score’s unidimensional approach fails to capture the complex pathophysiological mechanisms underlying thromboembolism in AF, resulting in a risk assessment tool that may lack precision for individualized clinical decision-making.

Contemporary research has elucidated that the pathophysiology of thrombus formation in AF extends beyond the traditional Virchow’s triad. While blood stasis from ineffective atrial contraction remains a fundamental mechanism, structural cardiac remodeling, endothelial dysfunction, systemic inflammation, and hypercoagulability all contribute significantly to thrombotic risk. This multifaceted pathophysiology necessitates a more comprehensive approach to risk assessment that integrates anatomical, functional, inflammatory, and hemodynamic parameters beyond what conventional clinical variables can provide.

In this study, we employed a systematic, data-driven approach to develop a more refined risk prediction tool for AIS in NVAF patients. Through sequential application of univariate logistic regression, LASSO regression for feature selection, and multivariate logistic regression, we identified eight independent predictors of AIS: age, admission systolic blood pressure (SBP), history of stroke, anticoagulant therapy status, left atrial diameter (LAD), left atrial appendage (LAA) filling defect, white blood cell count (WBC), and D-dimer levels. These parameters, which span clinical, anatomical, hemodynamic, and biochemical domains, were incorporated into a nomogram model that demonstrated excellent discrimination, calibration, and clinical utility upon internal validation.

Among the identified predictors, age and history of stroke are well-established risk factors that are also components of the CHA₂DS₂-VASc score. The heightened risk associated with prior stroke is consistently supported by robust evidence, including findings from the GARFIELD-AF registry reported by Hacke et al. (2020), which demonstrated a markedly increased risk of recurrent stroke in AF patients with prior stroke history (HR: 2.17, 95% CI: 1.80–2.63). Similarly, a nationwide Danish cohort study revealed that AF patients with previous stroke remained at substantial risk for recurrence despite secondary prevention strategies, with doubled risk among those who discontinued anticoagulation compared to those who maintained therapy (Hindsholm et al., 2024). The significant weight assigned to stroke history in the CHA₂DS₂-VASc score appropriately reflects its profound impact on risk stratification.

An intriguing finding from our analysis was the identification of admission SBP as a significant predictor of AIS in NVAF patients, while a history of hypertension did not demonstrate significant association. Although the acute-phase elevation of blood pressure during stroke is well-documented, our findings suggest that real-time blood pressure measurement may be more indicative of stroke risk than historical diagnosis of hypertension. This observation aligns with research by Ishii et al. (2017), emphasizing the critical importance of effective blood pressure management in AF patients. The relationship between blood pressure variability and stroke risk in this population warrants further investigation, as emerging evidence suggests that blood pressure fluctuations may independently contribute to adverse cardiovascular outcomes beyond mean pressure values.

The incorporation of LAA filling defects as detected by Cardiac MSCT into our prediction model represents an important advancement in risk stratification. Previous research has established that approximately 90% of cardioembolic events in NVAF patients originate from thrombi in the LAA (Zhang et al., 2024). While Cardiac MSCT offers a straightforward, time-efficient, and non-invasive method for detecting LAA thrombosis (Hajhosseiny et al., 2024), the potential for false-positive diagnoses must be acknowledged. Nicol et al. (2024) demonstrated that even a false-positive LAA thrombus finding remained independently associated with cardiogenic stroke (OR: 3.33, 95% CI: 1.42–7.81, P = 0.006). By incorporating LAA filling defects into our model, we capture important anatomical information that reflects thrombotic potential not adequately represented by clinical variables alone.

Left atrial enlargement emerged as another significant predictor in our model. Beyond being a consequence of AF, left atrial dilation serves as an independent predictor of thrombotic events (Tan et al., 2023). The left atrial diameter, readily assessed by echocardiography, has been shown to correlate with thrombotic risk in multiple studies. Research focusing on NVAF patients with low CHA₂DS₂-VASc scores found that increased left atrial diameter was independently associated with LAA thrombosis (OR: 1.088, 95% CI: 1.032–1.146, P < 0.05) (Kamili et al., 2024). Hirota et al. (2021) further demonstrated a dose-response relationship between left atrial dimension and ischemic stroke risk. Future refinements of prediction models might benefit from incorporating more nuanced assessments of left atrial structure and function, potentially including fibrosis evaluation through advanced imaging techniques.

The identification of WBC count and D-dimer levels as independent predictors highlights the crucial role of inflammatory and coagulation biomarkers in thrombotic risk assessment. Inflammation compromises vascular endothelial integrity and promotes hypercoagulability, creating favorable conditions for thrombus formation (Kelly et al., 2021). Dolu et al. (2023) reported elevated WBC count as an independent risk factor for left atrial thrombosis in NVAF (OR: 1.26, 95% CI: 1.05–1.51), and Li et al. (2022) demonstrated a positive correlation between leukocyte count and AIS risk. Similarly, D-dimer has been consistently associated with thromboembolic risk in AF patients (Yuan et al., 2022; Shi et al., 2023), maintaining its predictive value even in patients receiving anticoagulation therapy (Christersson et al., 2014; Siegbahn et al., 2016). The integration of these biomarkers into risk assessment represents a significant advancement toward personalized stroke prevention strategies in AF. Serial D-dimer measurements, in particular, could potentially enable dynamic risk assessment, offering advantages over static clinical risk scores.

Our study confirmed that absence of anticoagulant therapy is an independent risk factor for AIS in NVAF patients, consistent with established guidelines (Wang et al., 2024; Van Gelder et al., 2024). Concerningly, our data revealed that less than 25% of NVAF patients received appropriate anticoagulation, with many inappropriately receiving antiplatelet therapy instead. While anticoagulation decisions must be individualized, the widespread underutilization of guideline-directed therapy underscores the need for improved risk stratification and clinical decision support tools. Our nomogram model aims to address this gap by providing more precise risk assessment to guide appropriate preventive interventions, with its predictive framework built on a comprehensive importance ranking of 8 independent predictors derived from standardized regression coefficients of multivariate logistic regression and normalized LASSO feature importance scores; notably, LAA filling defect and D-dimer topped the ranking, followed by history of stroke, admission systolic blood pressure, age, LAD, WBC, with absence of anticoagulant therapy ranking the lowest.

The integration of multiple risk domains—clinical, anatomical, functional, and biochemical—into a single predictive nomogram represents a significant advancement in personalized stroke risk assessment for NVAF patients. The superior performance of our model compared to the CHA₂DS₂-VASc score has important clinical implications. While the CHA₂DS₂-VASc score remains valuable for its simplicity and widespread familiarity, our findings suggest that more comprehensive assessment tools incorporating hierarchically ranked risk factors (with left atrial appendage filling defect and D-dimer as the most impactful indicators) may better guide anticoagulation decisions, particularly in patients whose risk stratification remains ambiguous using traditional methods. Furthermore, the nomogram format provides an intuitive visual representation of risk that can facilitate shared decision-making between clinicians and patients.

Future research should focus on prospective validation of this model in diverse patient populations, integration of additional emerging biomarkers and imaging parameters, and development of dynamic risk assessment tools capable of capturing temporal changes in stroke risk. Machine learning approaches may further enhance predictive accuracy by identifying complex patterns and interactions among risk factors not readily apparent through conventional statistical methods. Additionally, exploration of genetic and epigenetic factors in AF-related stroke risk represents an exciting frontier for future risk stratification models.

This study has several limitations. First, as a single-center retrospective analysis, selection bias may influence our results. Second, the stability of the variable selection process (LASSO followed by logistic regression), while effective, may be sensitive to sample variations. Third, while the model demonstrated satisfactory performance in internal validation, it has not undergone external validation, limiting its generalizability to other clinical settings. Fourth, despite considering numerous potential factors, some indicators with possible predictive value, such as IL-6 (Singleton et al., 2021) and P-wave index (Maheshwari et al., 2019), were not included due to incomplete clinical data. Future large-scale, multicenter prospective studies are needed to refine and further validate the model, ultimately benefiting a broader patient population.

Age, admission systolic blood pressure, history of stroke, anticoagulant therapy status, left atrial diameter, left atrial appendage filling defect, white blood cell count, and D-dimer levels were identified as independent predictors of acute ischemic stroke in patients with non-valvular atrial fibrillation. The novel nomogram prediction model incorporating these multidimensional parameters demonstrated superior predictive performance compared to the conventional CHA₂DS₂-VASc score model. As a complementary tool, this approach allows for the personalization of anticoagulation strategies, which may lead to improved stroke prevention.