LungIMPACT
2026
AI-based chest X-ray prioritization in the lung cancer diagnostic pathway: the LungIMPACT randomized controlled trial
In 93,326 chest X-rays from primary care in the lung-cancer pathway, immediate AI prioritisation of the radiology worklist did not reduce time to CT or to lung-cancer diagnosis compared with AI-assisted reporting without worklist prioritisation.
Result
Primary endpoint: Time to CT and time to lung-cancer diagnosis (co-primary)
Median time to CT 53 vs 53 days (ratio of geometric means 0.97, 95% CI 0.93–1.02; P=0.31); time to lung-cancer diagnosis 44 vs 46 days (0.98, 95% CI 0.83–1.16; P=0.84). Lung cancer diagnosed in 558 (0.6%); AI–radiology discordance in 30.3% of CXRs.
- Subtopic
- Prioritisation & triage
- Design
- Multicentre RCT (randomised by day), UK
- Relevance
- Important: one of several pillars.
- Source
- Woznitza N, et al. Nat Med. 2026;32(5):1737-1744.doi 10.1038/s41591-026-04253-5
Discussion & critique
The largest randomised test of AI worklist prioritisation in imaging, and negative: with AI already running in both arms, pushing flagged chest X-rays up the queue changed neither time to CT nor time to diagnosis, so the authors recommend against prioritisation in this setting. Randomisation was by day rather than by patient, the unit of analysis is the X-ray rather than the person, and the trial isolates the effect of prioritisation only; it cannot say what the AI itself contributed to the pathway.