Healthcare AI Daily Briefing

Research & Evidence

Agentic AI cuts irAE annotation time and improves accuracy in crossover study.

Journal article (Nature cancer)

In a randomized crossover study of clinical trial staff, an agentic LLM system with human verification reduced annotation time by 2.71 minutes per note (P<0.001), increased complete-match accuracy (odds ratio 1.45; 95% CI 1.01-2.08; P=0.045), and raised inter-annotator agreement from Krippendorff's α 0.22-0.51 to 0.82-0.85. Retrospectively the system reached macro-averaged F1 of 0.92 for current-event detection and 0.66 for multi-class severity grading; prospective silent deployment over three months yielded binary F1 of 0.72-0.79 at roughly $0.02 per note. [1]

Medical Imaging

Multimodal AI combining chest CT and echocardiography improves PAH diagnosis over single modalities.

Journal article (Journal of visualized experiments : JoVE)

In a retrospective observational study of 100 patients with suspected pulmonary arterial hypertension, an attention-based multimodal AI integrating chest CT and transthoracic echocardiography achieved sensitivity 92.2%, specificity 88.9%, accuracy 91.0%, and AUC 0.930 against right heart catheterization reference. This outperformed CT alone (AUC 0.860) and echocardiography alone (AUC 0.840); McNemar tests favored the AI over CT (p=0.0078) and echocardiography (p=0.00049). The authors designate the tool as auxiliary and call for external prospective validation. [2]

Industry & Products

Wachter warns AI will scale healthcare's alignment problem.

In a new essay, UCSF's Dr. Bob Wachter argues that AI's alignment problem in healthcare is a preexisting condition that AI will scale, not a novel risk from rogue systems. He contends AI embedded in EHRs will amplify existing conflicts where health-system incentives diverge from patient interests, and urges transparency so tools show tradeoffs rather than defaulting to a single recommendation. [3]

Industry & Products

Oracle's Ellison says AI will enable comprehensive healthcare platform at Oracle AI World.

HIStalk reports that Oracle chairman and CTO Larry Ellison told Oracle AI World attendees that AI will allow the company to build a comprehensive healthcare platform connecting care, research, insurance, and public health. The announcement frames AI as the enabler for unifying those domains under a single Oracle platform. [4]