Development and validation of a web-based dynamic nomogram to predict individualized risk of severe carotid artery stenosis based on digital subtract angiography.

Abstract

Objectives: Delays in diagnosing severe carotid artery stenosis (CAS) are prevalent, particularly in low-income regions with limited access to imaging examinations. CAS is a major contributor to the recurrence and poor prognosis of ischemic stroke (IS). This retrospective cohort study proposed a non-invasive dynamic prediction model to identify potential high-risk severe carotid artery stenosis in patients with ischemic stroke.

Methods: From July 2017 to March 2021, 739 patients with ischemic stroke were retrospectively recruited from the Department of Neurology at Liuzhou Traditional Chinese Medical Hospital. Risk factors for severe CAS were identified using the least absolute shrinkage and selection operator (LASSO) and multivariate logistic regression (MLR) methods. The model was constructed after evaluating multicollinearity. The model's discrimination was assessed using the C-statistic and area under the curve (AUC). Its clinical utility was evaluated through the decision curve analysis (DCA) and the clinical impact curve (CIC). Calibration was examined using a calibration plot. To provide individualized predictions, a web-based tool was developed to estimate the risk of severe CAS.

Results: Among the patients, 488 of 739 (66.0%) were diagnosed with severe CAS. Six variables were incorporated into the final model: history of stroke, serum sodium, hypersensitive C-reactive protein (hsCRP), C-reactive protein (CRP), basophil percentage, and mean corpuscular hemoglobin concentration (MCHC). Multicollinearity was ruled out through correlation plots, variance inflation factor (VIF) values, and tolerance values. The model demonstrated good discrimination, with a C-statistic/AUC of 0.70 in the test set. The DCA and CIC indicated that clinical decisions based on the model could benefit IS patients. The calibration plot showed strong concordance between predicted and observed probabilities. The web-based prediction model exhibited robust performance in estimating the risk of severe CAS.

Conclusion: This study identified six key risk factors for severe CAS in IS patients. In addition, we developed a web-based dynamic nomogram to predict the individual risk of severe CAS. This tool can potentially support tailored, risk-based, and time-sensitive treatment strategies.