This Arkangel AI whitepaper is a narrative evidence synthesis, not a peer-reviewed study or a systematic review, of what AI can and cannot do today in surgery and anesthesiology, with a focus on Latin America. It draws on 28 sources from a targeted PubMed search and official WHO, FDA and PAHO documents (cutoff June 15, 2026) and grades six domains, from intraoperative hypotension alerts and complication prediction to surgical video, skill assessment and operating-room scheduling, by evidence stage, from retrospective development to routine implementation.
The evidence closest to clinical use is intraoperative hypotension prediction: a meta-analysis of 8 randomized-trial reports (568 participants counted with overlap), 7 of them on the Hypotension Prediction Index, found no significant difference in hypotension burden once the largest trial was included (mean difference 0.22 mmHg, 95% CI −0.03 to 0.48; I² 93.8%). Regional studies, such as a Bogotá surgical-duration model with an RMSE of 26 minutes and the LASOS cohort of 22,126 patients in 17 countries with 14.6% complications, show research capacity but not clinical effectiveness. The authors conclude that AI can support bounded perioperative tasks under local validation and governance, not autonomous intraoperative decisions.
