Surgical Oncology

Meta-analysis: ML predicts postoperative complications and early recurrence with moderate accuracy

This systematic review pooled 31 studies of ML models for cancer postoperative complications and early recurrence. Pooled performance for complications showed sensitivity ~0.75, specificity ~0.78, and AUC ~0.83. Proposed best models achieved higher sensitivity and AUC but results varied by tumor type, model, sample size, year, and country. Authors found no evidence of publication bias but emphasize heterogeneity and need for larger, standardized studies. 1

RCT protocol: AI-assisted education versus standard SMS after head and neck cancer surgery

This single-center randomized trial will enroll 100 postoperative head and neck cancer patients aged ≥60 to compare AI-driven personalized education with standardized SMS. Intervention delivers 12 months of phased education using an AI assistant with multimedia and real-time Q&A. Primary outcomes include mental health, social support, and quality of life measured at five timepoints to 12 months. Authors note single-center design and self-reported outcomes may limit external validity. 2

Cross-cancer review: AI shows accuracy but limited translational readiness for patient benefit

This structured review assessed AI translational maturity across 18 malignancies using a five-tier readiness framework. Breast and prostate AI tools reached higher maturity with FDA-cleared examples, but randomized evidence of patient-level benefit is absent. Most studies were retrospective, single-institution, and used curated datasets that differ from real-world deployment. Authors highlight persistent gaps between diagnostic accuracy and demonstrated clinical utility, plus disparities across populations. 3