AI Chest X-Ray Prioritization Did Not Shorten Lung Cancer Pathway

Key Takeaways

• Immediate AI prioritization was not associated with shorter time to CT or lung cancer diagnosis.
• Secondary pathway outcomes were also similar, including urgent referral timing, treatment timing, stage at diagnosis, and lung cancer incidence.
• Discordance between AI and radiology reports was common, and reporting turnaround was shorter without faster downstream milestones.

In the multicenter UK LungIMPACT trial, immediate AI prioritization of primary care chest radiographs did not shorten time to CT. Median time to CT was 53 days in both groups, with a ratio of geometric means of 0.97 (95% CI 0.93-1.02; P=0.31). Across five NHS trusts, AI was available in both arms, and only reporting prioritization differed. That change in reporting order did not translate into faster movement through the lung cancer pathway.

The prospective, multicenter randomized controlled trial was conducted across five NHS Trusts in England in adults undergoing primary care-requested chest radiographs. After exclusions, 93,326 chest radiographs were analyzed, with 45,987 assigned to prioritization-on sessions and 47,339 to prioritization-off sessions. Randomization occurred by day and site, and qXR version 4.0 was used as a clinical decision support tool at image acquisition in both study arms. Images returned to local PACS with AI mark-ups for reporting in both arms, and on intervention days suspected-abnormal cases triggered worklist alerts. Reporters were trained before the study, follow-up continued to June 2025, and the coprimary endpoints were time to CT and time to lung cancer diagnosis.

The second coprimary endpoint also showed no difference in time to lung cancer diagnosis between groups. Median time to diagnosis was 44 versus 46 days, with a ratio of geometric means of 0.98 (95% CI 0.83-1.16; P=0.84). Secondary outcomes were likewise similar, including urgent referral timing, treatment timing, stage at diagnosis, and lung cancer incidence. Among 13,347 identified CTs, 2,766 occurred within 14 days, and lung cancer was diagnosed in 558 people, or 0.6% of analyzed radiographs. Median time to urgent referral was 14 versus 15 days, median time to treatment was 76 versus 72.5 days, and no significant site or quarter differences were seen for the coprimary outcomes.

Discordance between AI and radiology reports occurred in 28,261 chest radiographs, or 30.3% of the cohort, and reviews were completed for 26,505 of those reports. Expert radiology review identified actionable findings in 6,750 cases, while AI-positive and false-negative patterns varied across opacity and nodule classifications. Time from chest radiograph acquisition to report fell from 47 hours to 34.1 hours during prioritization use. That reporting gain was not accompanied by faster CT, diagnosis, referral, or treatment milestones across the measured pathway.

The authors concluded that AI prioritization of UK primary care-requested chest radiographs had no significant impact on the lung cancer pathway. They emphasized that the trial tested prioritization rather than AI presence versus absence and evaluated only one commercial algorithm. They also noted that time-to-CT analyses included downstream scans that were not always linked to the index radiograph, while discordance review used pragmatic caps and prespecified rules. Some exploratory analyses were not prespecified, and the interpretation remained narrow to this English NHS primary care chest radiograph setting.