Machine learning sharply improves serious-adverse-event classification in a hospital study

Healthcare AIMachine learningDenmark

Published: 25 July 2026. A peer-reviewed Scientific Reports study used data from 2,325 patients and seven vital signs to compare four machine-learning analyses with a threshold-based National Early Warning Score approach. At the same false-positive rate, the reported 24-hour model achieved a true-positive rate of 0.96 versus 0.50; at matched sensitivity, false positives were 0.06 versus 0.84.

Why it matters: Reducing alert fatigue without missing deterioration is a high-value clinical operations problem. Impact: Results justify prospective, multi-site evaluation, but they do not establish clinical benefit, safety or generalisability outside the study population. The online version is an unedited early-access manuscript. Peer-reviewed paper