Among the promising applications of artificial intelligence (AI) across surgical care, improving patient safety in the operating room is one of the most exciting avenues being explored. Patients put their trust in surgical teams to deliver consistent and competent care that will see them through their procedure safely. In other high-reliability industries, there is an often-unspoken expectation that objective, standardized means of quality control exist when human lives are at risk, and that these mechanisms are in place to guarantee that vital processes do not fail in the face of technological and human error.¹ Perhaps unsurprisingly, health care has traditionally lagged in this space, given the variability in safety protocols, equipment utilization, and educational standards across institutions and health systems. Data support the notion that human error plays a huge role in the morbidity of patients undergoing surgery, and that most of these errors occur in the intraoperative setting.² Applications of AI to standardize the delivery of care touch on all facets of patient safety, from surgical education to intraoperative decision support, and these promise to disrupt current standards and practices.

The initial and perhaps most pervasive use of AI in surgical care is risk prediction. Although this can refer to the analysis of large datasets generally, a crucial application of AI in this space is predicting surgical complexity. While complicated surgery does necessarily lead to complications, intraoperative adverse events certainly can be a surrogate for early postoperative outcomes. Surgical phase recognition can be done autonomously and used to classify procedures by complexity.³ Image-based deep learning models trained with preoperative CT scans can predict surgical complexity in abdominal wall reconstruction, outperforming expert surgeon opinion by a significant margin and accurately predicting surgical site infections.⁴ Understanding the complexity of a surgical case allows surgical teams to better prepare for, anticipate, and mitigate intraoperative adverse events, which may lead to a reduction in operative morbidity in both routine and complex surgical cases.⁵ Risk stratification preoperatively can also help clinicians better council patients on their individual ideal surgical options and bring a new level of personalization to surgical care.

Another exciting opportunity for innovation is in real-time decision support for surgical teams. The ability of AI-based technology to quickly process a range of data allows for feedback to occur on a second-to-second basis. Recognition of patient anatomy, surgical phases, and technical errors logically has led to the development of decision aids for surgical teams, delivering timely and specific information that may reduce or prevent harm to patients.⁶ Many surgeons operate alone or with trainees only during challenging cases with complex anatomy or pathology, but computer-vision AI algorithms may provide critical guidance and serve as a “copilot” through real-time image processing from surgical video. Built on large datasets of surgical cases, these algorithms can support surgeons making evidence-based intraoperative decisions that are based on prediction of patient outcomes, thereby maximizing patient safety.⁷

Feedback has been shown to benefit surgeons of all levels of training and experience, and AI can be used to provide objective and specific feedback that leads to improved performance.⁸ Whether in the simulation lab or operating room, providing trainees with detailed feedback has always been a key tenet of surgical education. AI algorithms can critique many of the core aspects of technical performance, from economy of motion to instrument accuracy, providing a new way in which to give objective and standardized instruction to trainee surgeons early in their careers.⁹ However, it is important to recognize that feedback has been shown to benefit surgeons of any expertise level. As high-level athletes use coaches to continuously improve their skills, this is complemented through the analysis of objective performance metrics, and AI can play a central role in augmenting surgical coaching in a similar way.¹⁰ Similarly, AI can be used to automate assessments of surgical skill for educational purposes.¹¹ If effectively implemented in credentialing processes, AI can help set benchmarks for competency in surgical procedures, ensuring that graduating residents are able to perform at a level that satisfies minimum safety standards in their fields.

The accuracy and reliability of surgical AI models continue to grow exponentially, and this has the potential to create a paradigm shift in how we deliver surgical care. While exciting, it is important to remain cautious regarding the integration of this technology into practice without completely understanding its clinical, ethical, medicolegal implications. Transparency will be key to introducing surgical AI into the mainstream. Just as any pharmaceutical or surgical instrument must be rigorously tested prior to use in the operating room, AI-based infrastructure needs to be exhaustively validated to minimize any bias or imprecision that could lead to patient harm. Identifying causative factors and establishing accountability in surgical safety events can be very challenging, without the addition of this new layer of complexity in the surgical environment. Our comprehension of how AI models make predictions is incomplete, and this remains a rate-limiting step toward bringing this technology into operating rooms around the world.