Night-shift fatigue lowered chest X‑ray accuracy, and AI did not make up the difference
A prospective reader study from Germany, published online in European Radiology on 22 October 2026, found that radiology residents interpreted chest radiographs less accurately after a night on call, without slowing down, and that the AI tool tested did not restore their pre-shift performance.
What's new
Ten radiology residents (average experience about three years) each read sets of 33 chest radiographs before and after a first-call, on-site night shift. Their answers on five common findings (cardiomegaly, pulmonary congestion, pleural effusion, opacities and atelectasis) were compared with a majority vote of expert radiologists. Eye tracking during the reads was used to classify each resident as fatigued or not, based on a fall in the efficiency of their rapid eye movements after the shift.
Six of the ten residents met the fatigue definition. In that group, agreement with the experts (Cohen's kappa) fell from 0.71 before the shift to 0.61 after it. The non-fatigued group also dropped, but less (0.72 to 0.65). Reading time per film did not change, so the loss of accuracy did not show up as slower work.
After the shift the residents also read a set with AI assistance. The AI did not significantly improve accuracy in either group. It shortened reading time by about 10 seconds per film in the non-fatigued readers only.
This is a single-centre study with ten readers and one AI product, so the results may not apply to other tools, other settings or more experienced reporters. It does not show that AI is useless, only that this tool did not compensate for fatigue in this setting.
What it means for learners
- Tiredness hides itself. Readers in this study were just as fast when fatigued, so feeling that you are "keeping up" is not evidence that you are reading well. A fixed search pattern, such as the one in how to read a chest X-ray, is most useful exactly when you are tired.
- The findings tested are everyday ones. Cardiomegaly, congestion, effusion and collapse are covered in heart failure and pulmonary oedema, pleural effusion and lobar collapse.
- Do not hand the safety net to software. An AI prompt is one more input, not a second reader you can rely on when you are at your least sharp. Escalate uncertain films and follow your local policy on overnight reporting and senior review.
Source
- Siepmann RM, Khoobi M, von der Stück MS, Barajas Ordonez F, et al. From dusk till dawn? Night shift-induced fatigue, diagnostic performance, and AI assistance in chest radiography. European Radiology. Published online 22 October 2026. doi:10.1007/s00330-026-12969-8 (prospective single-centre reader study, 10 readers).
Also in What's new: UKHSA updates X-ray dose figures · MHRA withdraws the NGPod pH device.
For education only. Not a substitute for a formal report, senior review or your local policy. Never use this page to make decisions about a real patient.