AI-Powered Risk Assessment and Prediction Models for VTE: A Narrative Review

Background: A major avoidable cause of illness and death in healthcare settings worldwide is venous thromboembolism (VTE). While there are conventional risk assessment methods, the capacity of AI and ML techniques to evaluate complicated, multidimensional clinical data and detect non-linear patterns that conventional statistical methods could overlook makes them attractive alternatives to traditional VTE risk classification methods. Objective: This narrative review synthesises evidence on the performance, feasibility, implementation, and contextual relevance of AI-powered VTE risk assessment models. Methods: A comprehensive narrative review was undertaken on PubMed, Google Scholar, ScienceDirect, and Web of Science for 2010–2025 research on AI and VTE. For VTE prediction or diagnosis, included research created, validated, or deployed AI/ML models such supervised learning algorithms, ensemble approaches, and deep learning architectures. Studies using standard statistical techniques or without performance indicators were eliminated. The narrative synthesis centred on the model's performance, the experiences of clinical application, and the obstacles to acceptance. Results: Traditional risk assessments were routinely surpassed by AI models across a variety of clinical groups. Deep neural networks, gradient boosting, and random forests all showed very good VTE prediction accuracy. Despite this, there are still a lot of obstacles to overcome, such as a lack of external validation, uneven performance among populations, a large number of false positives, restrictions on data quality, and inadequate infrastructure. Conclusion: Models driven by AI show great potential for enhancing personalised risk assessment and early VTE identification. Factoring in the challenges and shortcomings is crucial for effective use.