JAMA + AI Weekly Update April 25, 2026

Could AI transform drug safety surveillance in pregnancy, accelerating research timelines from a decade to one day? A recent JAMA+ AI Conversations episode with Viktor Ahlqvist, PhD, of the Karolinska Institute explores this question. Drawing on his JAMA study of acetaminophen use during pregnancy as an example, Dr Ahlqvist describes how AI-enabled methods can strengthen observational research.

In a conversation with JAMA and JAMA+ AI Associate Editor Yulin Hswen, ScD, MPH, Dr Ahlqvist underscored the need for automated drug safety monitoring in pregnancy, a population historically excluded from clinical trials. While current regulatory and clinical guidance depends on slow, labor-intensive observational studies, he outlined how AI-driven pipelines could help to automate analyses and reduce years of work to as little as 24 hours.

Listen now on Spotify | Apple Podcasts | YouTube | JAMA.com.

Editor’s Picks in this week’s JAMA+ AI:

• A cohort study at a tertiary children’s hospital applied machine learning to predict postoperative length of stay for elective pediatric surgeries. Model deployment increased the number of scheduled procedures, reduced variability in bed occupancy, and minimized underutilization, suggesting the potential value of this approach in optimizing scheduling and stabilizing hospital capacity. (JAMA Pediatrics)
• Most voice biomarker studies overlook hearing health, even though hearing loss directly affects the speech and voice features used by AI. This omission can introduce confounding and misclassification, limiting generalizability particularly for older adults. The authors call for integrating scalable hearing screening, such as mobile audiometry, into voice biomarker studies. (JAMA Otolaryngology–Head & Neck Surgery)
• A new Viewpoint suggests that AI in pediatric intensive care should prioritize the recovery phase from severe illness, using decision support to guide treatment de‑escalation rather than escalation. The authors note that prediction models for acute deterioration face many barriers to adoption, and instead highlight the need for AI‑driven tools to optimize weaning of support. (JAMA Pediatrics)