Synergizing Systemic Inflammation and Multimodal Ultrasound: Methodological Insights Into Predicting Carotid Plaque Vulnerability

Abbreviations

AUC

area under the curve

CEUS

contrast‐enhanced ultrasound

CRP

C‐reactive protein

EPV

events per variable

IPN

intraplaque neovascularization

LASSO

least absolute shrinkage and selection operator

NLR

neutrophil/lymphocyte ratio

VIF

variance inflation factor

VP

vulnerable plaque

To the Editor:

We were highly interested in reading the recent article by Liu et al, titled “Development and Validation of a Multimodal Nomogram for Carotid Vulnerable Plaques Based on Ultrasound Features, Clinical Risk Factors, and Systemic Inflammatory Markers.” ¹ The authors successfully developed and validated a high‐precision nomogram model by integrating contrast‐enhanced ultrasound (CEUS)‐derived intraplaque neovascularization (IPN) grading, systemic inflammatory markers, and clinical risk factors. The model demonstrated an area under the curve (AUC) of 0.889 in the validation cohort, providing a powerful tool for individualized stroke prevention. While we commend this outstanding work, we would like to offer some points for further discussion.

First, we address the issue of variable redundancy and multicollinearity in the predictive model. In the model constructed by the authors, which includes 8 factors, both the neutrophil/lymphocyte ratio (NLR) and absolute neutrophil count were retained. From a statistical perspective, the NLR is a derived ratio variable based on neutrophil count, and these 2 variables often exhibit significant multicollinearity in regression models. Although least absolute shrinkage and selection operator (LASSO) regression helps with variable selection, including both the “overall” variable and its “components” in the final model may affect the stability of the variable coefficient estimates. We suggest that future studies assess multicollinearity further using the variance inflation factor (VIF). ²

Second, clinical confounding factors, particularly the potential interference of statin use, warrant attention. Statins not only have lipid‐lowering effects but also significantly stabilize plaques, effectively reducing IPN occurrence and lowering systemic CRP levels. In Liu et al's study, statin use in the vulnerable plaque (VP) and non‐VP groups was listed as baseline data but was not included in the final model. If some patients in the VP group had been on long‐term statin therapy, their imaging characteristics (such as IPN grading) may have become decoupled from their actual biological risk. Performing subgroup analyses based on different intensities of lipid‐lowering therapy would help clarify the robustness of the model in the context of real‐world clinical treatment. ³

Additionally, we address the sample size adequacy and the methodological limitations of validation. The total sample size in the study was 146, with only 46 patients in the validation cohort. According to Riley et al's recommendations, building a robust binary outcome prediction model typically requires a higher event per variable (EPV) to avoid the optimistic bias resulting from overfitting. ⁴ While the authors used Bootstrap for internal validation, the single‐center nature of the study introduces uncertainty regarding the model's performance across different equipment platforms. We look forward to seeing external validation results of this model in multi‐center, large‐scale populations in the future.

Finally, future research could consider introducing more refined feature quantification methods. The distribution of microcalcifications within plaques and the biomechanical properties of the fibrous cap have significant impacts on plaque vulnerability. Integrating emerging technologies such as radiomics ⁵ or elastography may capture subtle structural changes that are difficult to identify with the naked eye, potentially overcoming the subjectivity limitations of conventional 2‐dimensional imaging. ⁶

In conclusion, the study by Liu et al marks a significant step in the multidimensional assessment of carotid plaques. With more rigorous statistical corrections and large‐scale external validation, this nomogram could become an invaluable tool in the field of stroke clinical prevention.

Data Availability Statement

Data sharing not applicable to this article as no datasets were generated or analyzed during the current study.