VESS19 - Artificial Intelligence-Derived CTA Plaque Morphology Improves Long-Term Risk Stratification and Reclassifies Patients With Asymptomatic Carotid Disease

Objectives: We previously developed a 30-day stroke/death risk prediction model, incorporating quantitative plaque morphology to improve operative decision-making in asymptomatic extracranial carotid disease patients which currently relies only on clinical factors and stenosis degree. The objective of this study is to assess the long-term prognostic performance of this plaque-informed model by evaluating the three-year follow-up outcomes.

Methods: We analyzed single-institution data from 117 asymptomatic carotid stenosis patients (2010–2021) with available 3-year follow-up. Quantitative plaque morphology features were extracted from baseline head and neck computed tomographic angiography (CTA) using an artificial intelligence (AI)–based, histopathologically validated plaque analysis platform and incorporated into a risk assessment model. Independent clinical factors included age, sex, hyperlipidemia, diabetes mellitus, hypertension, and baseline statin therapy. Time-to-event outcomes (stroke/death) were analyzed using a Bayesian Weibull survival model. Model inference was performed using Markov Chain Monte Carlo (MCMC) sampling with 12,000 iterations, of which 8,000 posterior draws were retained after burn-in, from which median estimates and credible intervals were derived.

Results: Within 3 years of follow-up, 19 of 117 (16.2%) experienced stroke or death. Each patient’s three-year risk was estimated from posterior simulations, and patients were ranked according to the median of their posterior risk distribution, and stratified into tertiles representing low-, intermediate-, and high-risk groups. Observed 3-year event rates increased stepwise across risk tertiles, with event rates of approximately 7%, 15%, and 30% in low-, intermediate-, and high-risk groups, corresponding to a 4.3-fold gradient between high- and low-risk tertiles (Figure 1). The posterior probability of risk separation was 100%. Incorporation of CTA-derived plaque morphology into the risk model resulted in reclassification of 23% of patients relative to a model based on clinical covariates + stenosis degree (Figure 2). Patients reclassified to higher-risk categories demonstrated higher observed event rates, while those reclassified to lower-risk categories experienced fewer adverse events during follow-up.

Conclusions: Incorporation of plaque morphology assessment may enhance decision-making in asymptomatic carotid disease by improving identification of high-risk patients most likely to benefit from revascularization. This prediction model can also identify patients with stable plaque, who may be safely managed with intensive medical therapy, supporting more individualized treatment selection and warranting prospective validation in larger cohorts.