The study utilized the imaging examination database of the Second Affiliated Hospital of Guangzhou University of Chinese Medicine. We retrospectively analyzed carotid ultrasound examination data from January 2020 to January 2025 to identify eligible patients. All patients were inpatients in the neurology department, admitted for symptoms related to cerebrovascular and cardiovascular diseases. The study population consisted of two groups: the IS group and the RIS group. The IS group included patients who presented to our hospital for the first time with a stroke and no history of stroke. When included in the study, these patients did not experience a recurrence during follow-up. The RIS group included patients who had previously experienced an IS before undergoing carotid ultrasound examination and subsequently had a RIS during the follow-up period. Inclusion criteria: ① first-time diagnosis of IS at our hospital, supported by clinical symptoms and cranial CT or MRI findings; ② detection of carotid plaques with good-quality ultrasound images, complete laboratory tests and clinical data. Exclusion criteria included: ① poor image quality or incomplete data from carotid ultrasound examination; ② prior carotid stenting; ③ hemorrhagic stroke, cerebral hemorrhage, or cerebral artery malformation confirmed by head CT or MRI. Finally, 286 patients were enrolled in the study: the IS group (129 patients) and the RIS group (157 patients) (Figure 1). The Second Affiliated Hospital of Guangzhou University of Chinese Medicine ethics review board has approved this retrospective study (Ethical Approval No.: ZE2025-109-01). Written informed consent from patients was not required.

Clinical information was collected from the electronic medical record system. Including demographic data [gender, age, height, weight, and body mass index (BMI)], medical history [diabetes, hypertension, coronary heart disease, atrial fibrillation, and transient ischemic attack (TIA)], lifestyle factors (smoking status and drinking status), and laboratory indicators: total cholesterol (TC), high-density lipoprotein cholesterol (HDL), low-density lipoprotein cholesterol (LDL), triglycerides (TG), glycated hemoglobin (HbA1c), and fasting blood glucose (FBG).

Body mass index is calculated by dividing weight in kilograms by height in meters squared. Regarding lifestyle factors, drinking status was defined as an average ≥45 g daily for women and ≥90 g for men over the past decade. During that period, smoking status was defined as smoking at least of one cigarette daily. The following were the diagnostic standards for medical history: diabetes was defined as FBG ≥ 7.0 mmol/L or 2-h postprandial glucose levels ≥11.1 mmol/L. Hypertension was diagnosed based on systolic blood pressure ≥ 140 mmHg, diastolic blood pressure ≥ 90 mmHg, or the use of antihypertensive medication. Coronary heart disease was confirmed by imaging examinations, electrocardiograms, and typical symptoms such as chest pain or dyspnea on exertion (coronary angiography showing stenosis ≥50%). Atrial fibrillation was identified by electrocardiogram features. TIA was defined as a transient neurological deficit lasting less than 24 h without infarct lesions on imaging. Abnormal laboratory indicators were defined as follows: TG ≥ 1.7 mmol/L; LDL ≥ 3.4 mmol/L; HDL < 1.0 mmol/L (for men) or <1.3 mmol/L (for women); TC ≥ 5.2 mmol/L; HbA1c ≥ 6.5%; and FBG ≥ 7.0 mmol/L.

Carotid ultrasound examinations were performed using the Epic 7 ultrasound diagnostic device (4–8 MHz) and the IU Elite ultrasound diagnostic device (3–9 MHz). Plaque image analysis was conducted only on carotid arteries with patent lumens. When both carotid arteries were patent, the thickest plaque was selected for detailed examination. The location, size, and shape of the plaque were meticulously recorded, and the formula for calculating the degree of stenosis is as follows: Stenosis (%) = [(Diameter of the normal distal internal carotid artery − narrowest internal carotid artery diameter in the stenotic segment) / Diameter of the normal distal internal carotid artery] × 100%, according to North American Symptomatic Carotid Endarterectomy Trial (NASCET) (Barnett, 1991), carotid stenosis degree was classified as mild (<30%), moderate (30%–69%), and severe (70%–99%). The maximum wall thickness (MWT) was defined as the vertical distance from the lumen to the adventitia interface to the top of the plaque. The FC was identified as an arcuate, linear echogenicity structure between the plaque and the lumen. A thin FC was indicated if there was local echo disruption or incomplete visualization.

To ensure the evaluative accuracy of the Carotid Plaque-RADS, patient identities were anonymized. Their carotid ultrasound images were randomly assigned to two ultrasound physicians with over 5 years of experience for evaluation. The evaluators were unware of each other’s assessment results and unable to know patients’ follow-up result. In cases of discrepancy, a third, more experienced senior physician was consulted to determine the final classification. Prior to image interpretation, the ultrasound physicians received standardized training, and regular assessments of scoring consistency were conducted. Carotid plaques are categorized according to the Carotid Plaque-RADS categories (Saba et al., 2024): Carotid Plaque-RADS 1 indicates a normal vascular wall; Carotid Plaque-RADS 2 indicates a plaque with MWT < 3 mm and with no complex plaque features; Carotid Plaque-RADS 3a indicates MWT ≥ 3 mm, with LRNC and a thick FC, excluding complex plaque features; Carotid Plaque-RADS 3b indicates MWT ≥ 3 mm, with LRNC and a thin but intact FC; Carotid Plaque-RADS 3c indicates plaque ulceration, where “ulceration” refers to a surface cavity, known as a “healed ulcer plaque.” Carotid Plaque-RADS 4 is not associated with MWT thickness, but is graded by the presence of one of the following: IPH (4a), FC rupture (4b), or luminal thrombus (4c). Note: Active ulceration with luminal thrombus is classified under Carotid Plaque-RADS 4c, while “healed ulcer plaque” (3c) refers to a residual cavity without active flow communication.

This study systematically followed up with 286 patients who underwent carotid ultrasound examinations. The follow-up period extended from the date of admission to January 2025. Follow-up was conducted via medical record review and telephone interviews. For patients lost to follow-up, efforts were made to retrieve data through alternative means, such as contacting family members or reviewing available medical records to minimize data loss. The primary endpoint was symptomatic RIS: new-onset neurological deficit lasting ≥24 h and confirmed by DWI-MRI within 7 days of symptom onset, excluding hemorrhage, and all-cause mortality was the secondary endpoint. Patients were followed for up to 60 months (median follow-up: 38 months, IQR: 24–52 months), with censoring occurred at the first RIS, death, or January 2025, whichever came first. IS was diagnosed using the TOAST (Trial of Org 10172 in Acute Stroke Treatment classification) criteria (Adams et al., 1993). RIS was defined as the occurrence of new focal neurological deficits lasting more than 24 h, confirmed by diffusion-weighted magnetic resonance imaging showing new acute cerebral infarction lesions in the territory supplied by the carotid artery on the same side as the previously identified stroke. To ensure consistent diagnosis, the same imaging protocols and clinical criteria were applied throughout the study.

The diagnostic process for stroke recurrence was standardized, with two neurologists independently assessing clinical symptoms and imaging features. Disagreements were resolved by the clinical research team, ensuring the reliability and reproducibility of the data. All neurologists involved in the study underwent training to ensure consistent application of the TOAST criteria and DWI imaging interpretation. In cases of multiple RIS, only the first event was included in the statistical analysis to avoid bias. The diagnostic process adhered to internationally accepted stroke classification and imaging standards to ensure the scientific validity and reproducibility of data collection.

Statistical analysis was performed using SPSS Statistics (version 25.0, IBM) and R (version 4.2.3). Continuous data were presented as mean ± standard deviation and assessed via t-test. Categorical data were expressed as percentages and assessed using Chi-square tests. Risk factors for stroke recurrence were determined by univariate and multivariate logistic regression analysis, adjusting for age, sex, smoking status, diabetes, and BMI as potential confounders. A nomogram prediction model was constructed, incorporating factors that were statistically significant in the multivariate analysis, to predict stroke recurrence. The clinical effectiveness of the nomogram prediction model was assessed via calibration curves and decision curve analysis. The model’s fit was evaluated using the Hosmer–Lemeshow test, and its predictive capability was measured by the receiver operating characteristic (ROC) curve, with the AUC calculated. A significance level of P < 0.05 was used throughout the analysis.

This study included 286 patients, comprising 194 males (68%) and 92 females (32%). Their ages ranged from 60 to 92 years, with a mean age of 73.36 years. Carotid plaques were detected in 256 patients (89%) within the entire study cohort. During a 60-month follow-up period ending in January 2025, 129 patients experienced an IS, while 157 experienced a RIS. Among the RIS group, 147 patients (93.6%) had carotid plaques detected, and 30 patients (20.4%) of them had a severe carotid stenosis. The distribution of Carotid Plaque-RADS categories was as follows: 10 patients (6.3%) in Carotid Plaque-RADS 1, 38 patients (24.2%) in Carotid Plaque-RADS 2, 58 patients (36.9%) in Carotid Plaque-RADS 3, and 51 patients (32.5%) in Carotid Plaque-RADS 4. In contrast, among the IS group, 109 patients (84.5%) had carotid plaques detected, and 8 patients (7.3%) of them had a carotid stenosis degree of ≥70%. The distribution of Carotid Plaque-RADS categories was as follows: 20 patients (15.5%) in Carotid Plaque-RADS 1, 65 patients (50.4%) in Carotid Plaque-RADS 2, 35 patients (27.1%) in Carotid Plaque-RADS 3, and 9 patients (7.0%) in Carotid Plaque-RADS 4. Further analysis revealed that the detection rate of carotid plaques and the degree of carotid stenosis were closely associated with the occurrence of stroke events, particularly in the RIS group, where the prevalence of plaques and severe stenosis was higher.

A total of 286 patients were randomly assigned in a ratio of 3:1 to create a training set of 200 patients and a validation set of 86 patients. The external validation cohort consisted of 86 patients retrospectively enrolled from the same single-center temporal validation cohort (2019–2024) with identical inclusion criteria, representing a temporally distinct but demographically similar population. Within the training set, patients were further divided into two groups: the IS group, comprising 92 cases, and the RIS group, comprising 108 cases. Patients were divided according to whether they had a stroke recurrence during the follow-up period. We compared the training and validation sets with respect to age, height, weight, BMI, TC, HDL, LDL, TG, HbA1c, FBG, gender, smoking status, drinking status, hypertension, diabetes, TIA, coronary heart disease, atrial fibrillation, fibrinogen, carotid stenosis degree, and Carotid Plaque-RADS categories. There was no significant difference between the two groups (P > 0.05), indicating that comparisons can be made (Table 1).

In the univariate analysis, the RIS group exhibited significantly higher levels of LDL compared to the IS group (P < 0.05). No statistically significant differences were found between the two groups in height, weight, BMI, TC, TG, HDL, HbA1c, FBG, or fibrinogen levels (P > 0.05; Table 2). Significant differences were observed between the groups for hypertension, atrial fibrillation, carotid stenosis degree, and Carotid Plaque-RADS categories (P < 0.05), while no significant differences were found for gender, smoking status, drinking status, TIA, diabetes, or coronary heart disease (P > 0.05; Table 2). Additionally, the Carotid Plaque-RADS categories showed significant differences between the RIS and IS groups (P < 0.05; Table 2).

Multivariate logistic regression analysis revealed that statistically significant variables in univariate analysis—LDL, hypertension, atrial fibrillation, severe carotid stenosis, and Carotid Plaque-RADS 2, 3, and 4—were all independent risk factors for RIS (P < 0.05; Table 3). Notably, while TIA was associated with recurrence in the univariate analysis (P < 0.05), its effect was no longer independently significant after adjusting for confounding variables in the multivariate logistic regression model (β = 1.02, P = 0.16). This suggests that the association between TIA and recurrence may be mediated by other risk factors or due to collinearity. Consequently, TIA was excluded from the final multivariate logistic regression model.

To facilitate the intuitive presentation of the predictive model, we developed a nomogram incorporating five risk factors (Figure 2). The total score is calculated by adding up the individual scores assigned to each risk factor. This score is used to estimate the risk of RIS. For example, a patient with IS who also has hypertension (42 points), severe carotid stenosis (12 points), Carotid Plaque-RADS categories of 4 (100 points), an LDL level of ≥6 mmol/L (78 points), and a history of atrial fibrillation (68 points) would accumulate a total score of 300 points. This score corresponds to a predicted RIS risk of over 90%, indicating an extremely high risk of recurrence.

The P-value of the Hosmer–Lemeshow test was 0.354, indicating a satisfactory fit. Internal and external for calibration curves validation are presented in Figures 3, 4, respectively, and demonstrate that the nomogram predictions for RIS are consistent with actual outcomes. The internal validation results demonstrated that the calibration curve nearly aligned with the ideal curve (Figure 3), suggesting that the model have strong calibration and discrimination abilities within the training set. Specifically, the combined model incorporating clinical features and Carotid Plaque-RADS achieved an AUC of 0.79 (95% CI: 0.76–0.81). In contrast the model using only clinical features had an AUC of 0.72 (95% CI: 0.69–0.75), and the model using only Carotid Plaque-RADS had an AUC of 0.71 (95% CI: 0.68–0.76) (Figure 5). External validation results showed the calibration curve aligned with the ideal curve (Figure 4), further confirming the model’s stability and applicability in an independent validation set. The combined model integrating clinical features and Carotid Plaque-RADS achieved an AUC of 0.76 (95% CI: 0.69–0.82). The models using only clinical features and only Carotid Plaque-RADS had AUCs of 0.70 (95% CI: 0.62–0.78) and 0.69 (95% CI: 0.60–0.79), respectively (Figure 6).

Consistency and stability analyses show the model demonstrate good predictive performance across diverse datasets, indicating substantial clinical application value. Integration of Carotid Plaque-RADS categories into clinical features significantly improves the model’s predictive accuracy (P < 0.05), enabling more precise identification of high-risk patients.

Figures 7, 8 show decision curve analysis (DCA) curves applied to the RIS risk nomogram prediction model for the training and testing datasets respectively. The DCA curves for both datasets indicate the model provides substantial standard clinical net benefit and retains stable clinical utility advantages within the threshold probability range of 0.2–0.8. These results suggest the model has practical utility for clinical decision-making in RIS risk management.