Machine learning (ML) algorithms could reshape how we predict colorectal cancer (CRC) and advanced polyps (ACPs) before a colonoscopy is even performed. A recent meta-analysis, published in Gastroenterology and summarized by Jessica Nye, PhD, reviewed 14 studies involving over half a million patients across Europe and Australia.
These AI-driven tools demonstrated strong potential—particularly in CRC detection—with an average AUROC of 0.883, sensitivity of 83%, and specificity of 80%. However, the performance dropped when ACPs were included, and even more so when evaluating ACPs alone. The models varied widely, using everything from logistic regression to random forests and neural networks, and incorporating imaging, electronic medical records, and even volatile organic compounds.
Despite encouraging signals, the authors cautioned that heterogeneity in study methods and outcome definitions limits clinical applicability for now.