Combined pulse wave velocity and triglyceride–glucose index to discriminate large-artery atherosclerosis from small-vessel occlusion in acute ischaemic stroke: a single-centre retrospective study

Objective Early etiologic discrimination between large-artery atherosclerosis (LAA) and small-vessel occlusion (SVO) is clinically relevant when advanced imaging is unavailable or delayed. We examined whether combining carotid pulse wave velocity (PWV) with the triglyceride–glucose (TyG) index improves early in - hospital discrimination of LAA vs. SVO, and developed a predictive model integrating PWV, TyG, and clinical indicators. Methods We performed a single-centre retrospective study including consecutive acute ischaemic stroke patients from September 2021 to November 2023. Carotid PWV (Wv, m/s) was measured bilaterally and averaged; TyG was calculated from fasting triglycerides and glucose. Restricted cubic splines (RCS) tested non-linearity. Multivariable logistic regression was applied to (i) estimate the adjusted association of PWV and TyG with LAA versus SVO, and (ii) identify independent predictors via backward stepwise selection and establish predictive model. Discrimination was assessed by ROC AUC with DeLong tests; calibration used bootstrap-corrected curves; internal validation used 1,000-bootstrap optimism correction; decision-curve analysis appraised clinical utility. Results Of 892 admissions screened, 298 were analysed (SVO 179; LAA 119). LAA showed higher PWV (median 17.72 vs. 15.02 m/s; p < 0.001) and higher TyG (8.85 vs. 8.64; p = 0.005) than SVO. RCS supported overall associations of PWV and TyG with LAA versus SVO without evidence of non-linearity. Compared with single markers (AUC: PWV = 0.627; TyG = 0.596), a two-marker model (PWV + TyG) achieved AUC 0.654, while a parsimonious four-variable model (PWV, TyG, sex, eGFR) reached AUC 0.685 and significantly outperformed either single marker (paired DeLong vs. PWV: ΔAUC 0.058, p = 0.043; vs. TyG: ΔAUC 0.089, p = 0.011). Calibration was acceptable after bootstrap correction. Conclusion PWV and TyG provide complementary, early in - hospital signals for moderate discrimination of LAA vs. SVO, with incremental value when combined. The predictive model constructed by combining PWV, TyG, and clinical indicators also performed well. These bedside markers provide adjunctive, early in-hospital risk stratification for suspected LAA versus SVO.