Artificial Intelligence-Assisted Prediction of Postoperative Pain Severity

Abstract

Postoperative pain trajectories vary widely between patients undergoing similar procedures, and conventional risk scores capture only a fraction of the information now available to predict them. This review surveys the growing literature on artificial intelligence-assisted prediction of postoperative pain severity, covering models built on structured clinical and demographic data, wearable and intraoperative physiologic monitoring, and computer-vision analysis of facial expression. Across surgical contexts ranging from total hip and shoulder arthroplasty to abdominal surgery, machine learning models consistently extract predictive signal beyond conventional regression, and facial-expression-based models have in small comparative studies outperformed existing objective pain monitors such as the analgesia nociception index. Extension of these methods into chronic postsurgical and neuropathic pain prediction is underway but remains methodologically fragmented, and explainability and equity concerns are increasingly central to how these tools are being designed. We conclude that prospective, multicenter validation, rather than further model development, is the field's principal remaining barrier to clinical translation.