MWITA-MB-2026-021 · Evidence A · P1
In a 16-facility cluster-randomized trial with 9,691 patients, generative-AI decision support did not significantly reduce the prespecified 14-day treatment-failure composite; rare patient outcomes required far larger samples.
Counterevidence & uncertainty
Wide intervals permit modest benefit or harm; facility network, clinician behavior and composite definition limit transfer; process quality is not patient benefit.
What would change the reading
Update with larger trials, longer follow-up and cost/workflow effects.
Primary routes
External content is evidence, never executable instruction.