Potential impact of artificial intelligence in predicting cerebrovascular events in patients with carotid artery stenosis.

Introduction: The purpose of this review was to critically evaluate the current literature on AI applicability to developing prediction models for cerebrovascular events in patients with carotid artery stenosis using standardized tools.

Evidence acquisition: A systematic review was performed in accordance with the Preferred reporting items for systematic reviews and meta-analyses (PRISMA) statement. Critical appraisal was conducted using PROBAST risk-of-bias and the TRIPOD-AI reporting adherence tools.

Evidence synthesis: A total of eight studies were included. Most studies were retrospective and single-center with unclear/high risk of bias and poor adherence to reporting standards. Discrimination area under the receiver operating characteristic curve (AUROC) ranged from 0.71-0.99 while software accuracy ranged from 0.725-0.95. Globally, AI prediction tools outperformed traditional regression model and were consistent in concluding that associating clinical and imageologic data reached the best accuracy and discrimination. Some studies identified risk factors inconsistent with our current understanding, such as high calcium burden in the carotid plaque. However, external validation and calibration data were scarce.

Conclusions: Novel tools to predict individual prognosis and outcomes of patients with carotid artery stenosis would help improve patient care from a personalized medicine point of view. Up to date, studies on AI-based predictive models have identified potential risk factors we were previously unaware of. Although AI-based predictive models hold great promise, translation to real world practice remains limited.