Individualized MRI-Based Stroke Prediction Score Using Plaque Vulnerability for Symptomatic Carotid Artery Disease Patients (IMPROVE)

BACKGROUND: In symptomatic carotid stenosis, treatment decisions are currently primarily based on stenosis degree. We developed a clinical prediction model (Individualized Magnetic Resonance Imaging–Based Stroke Prediction Score Using Plaque Vulnerability for Patients With Symptomatic Carotid Artery Disease [IMPROVE]) incorporating the strong predictor, intraplaque hemorrhage on magnetic resonance imaging, stenosis degree, and risk factors to identify patients with high stroke risk. METHODS: IMPROVE was developed on data from 5 cohorts of 760 patients with symptomatic carotid disease on optimal medical treatment. Inclusion criteria included a recent transient ischemic attack/stroke (<6 months), magnetic resonance imaging–based information on intraplaque hemorrhage, no atrial fibrillation, and no immediate revascularization. IMPROVE was based on Cox regression using 5 expert-selected predictors and converted to 3-year ipsilateral ischemic stroke risk after internal validation. IMPROVE-based stratification was compared with care-as-usual using illustrative cutoffs: high risk was defined in IMPROVE as ≥ median 3-year IMPROVE risk, whereas in care-as-usual, it was ≥50% carotid stenosis. RESULTS: Sixty-five ipsilateral ischemic strokes occurred during a median follow-up of 1.2 years (interquartile range, 0.5–4.1). The IMPROVE model includes 5 predictors (hazard ratio [95% CI]): degree of stenosis (<50%: reference, 50%–69%: 4.54 [2.46–8.38], 70%–99% stenosis: 7.42 [3.45–15.95]), presence of intraplaque hemorrhage (5.61 [2.92–10.77]), classification of last event (ocular: reference, cerebral: 3.72 [1.11–12.52]), male sex (1.26 [0.64–2.48]), and age (1.14 [0.84–1.55] per 10-year increase). Internal validation revealed good accuracy (C statistic, 0.82 [95% CI, 0.77–0.87]) and no evidence of miscalibration (calibration slope, 0.93). Sensitivity for the illustrative IMPROVE cutoff was 92.6% (90.7–94.5) versus 80.6% (77.8–83.4) for care-as-usual. Specificity was 54.2% (50.7–57.8) for IMPROVE versus 52.9% (49.3–56.4) for care-as-usual. Patients stratified by IMPROVE as high risk had a higher incidence of ipsilateral ischemic stroke (24.0%) compared with the care-as-usual classification (20.7%). Among patients classified as lower-risk by IMPROVE and care-as-usual, 2.1% and 5.3%, respectively, experienced an ipsilateral ischemic stroke during follow-up. CONCLUSIONS: Using the presence of intraplaque hemorrhage on magnetic resonance imaging and 4 conventional parameters, the IMPROVE model provides accurate individual stroke risk estimates, which may facilitate stratification for revascularization after external validation.