Putting suspicious X-rays in front of the queue doesn't save a day: the LungIMPACT trial

In five NHS facilities, automatic prioritisation of AI chest X-rays has not shortened either the time to CT scan or the time to lung cancer diagnosis.

The idea seemed solid: let an algorithm immediately identify abnormal chest X-rays requested by community medicine, and move them to the top of the radiologists' reading list to speed up the rest of the process. The LungIMPACT randomized trial, published in Nature Medicine on March 24, 2026, tested this hypothesis in five British NHS facilities, with adult patients examined between July 2023 and December 2024, with Qure.ai's qXR 4.0 tool.

The result was clear and negative. The median time to CT scan was 53 days in both arms of the trial. The median time to diagnosis was 44 days on one side compared to 46 days on the other, a non-significant difference (geometric mean ratio of 0.98; 95% confidence interval: 0.83-1.16; p = 0.84). Lung cancer was diagnosed in 558 patients, or 0.6% of the radiographs, with no significant difference in the time to urgent referral, the time to treatment or the stage at diagnosis.

The authors' conclusion goes beyond the case of this tool: reordering a worklist adds cost and complexity without clinical benefit as long as the bottleneck is elsewhere in the care chain. Deploying a sorting AI only makes sense when coupled with a real redesign of the pathway.