David Limon, Varsha Satish, N. R. Srinivasa Raghavan, Miranda X. Morris, Mark T. Muir, Aashish Rajesh
Machine learning (ML), a branch of artificial intelligence, is rapidly transforming surgical complication and outcome prediction. Unlike traditional statistical approaches, ML can learn complex, nonlinear relationships across multiple variables, enabling more accurate and adaptable prognostication. Emerging ML-based tools have demonstrated strong performance across diverse surgical specialties, often surpassing conventional risk models. However, challenges remain, including opaque "black box" outputs, diminished performance during external validation, difficulty modeling rare events, and dependence on tabular data. These limitations can be mitigated but demand thoughtful design and rigorous validation. Importantly, ML introduces distinct methodological considerations unfamiliar to many surgeons. Successful clinical integration requires robust external validation and transparent sharing of trained models to ensure reproducibility and generalizability across diverse cohorts. By enhancing the precision of risk prediction, ML holds the potential to guide patient selection, optimize perioperative care, and strengthen shared decision-making between patients and surgeons.