AI tools boost liver cancer treatment decisions and ICU mortality forecasting

Researchers at University Hospital Münster evaluated four GPT variants (GPT‑4, o1, o3, 5.4) for assessing hepatocellular carcinoma. All achieved over 85% accuracy in liver‑function scoring, with GPT‑o3 showing the lowest error rates (2.8% for Child‑Pugh and no errors for ALBI). The models matched or exceeded interdisciplinary committee recommendations in up to 14% of cases and generated full reports in about two minutes per patient, cutting physician data‑gathering time from roughly 568 seconds to under 120 seconds and reducing labor costs by 300‑1300 times.

An international study published in *Intensive & Critical Care Nursing* analysed more than 360 000 Brazilian ICU patients from the COVID‑19 pandemic. The best AI algorithm predicted ICU mortality with about 85% accuracy, identifying both clinical and systemic risk factors such as care pressure and territorial inequalities. The results could help hospitals anticipate bed needs, prioritize high‑risk patients, and allocate resources more efficiently.