Nodule AI was associated with 15% faster chest CT reporting, but slower emergency reads

Design

Retrospective before-after study, Weibull shared frailty model, single centre

Population

39,323 chest CT examinations in 19,433 patients

Primary outcome

Radiology reporting time

Effect

Median 21.3 to 18.2 min (-14.6%); adjusted HR 1.17 (1.14-1.21); emergency +7.1%

This retrospective before-after study at a Dutch tertiary centre compared 19,190 chest CTs reported before and 20,133 after introducing commercial AI for pulmonary nodule assessment, using a survival model adjusted for reader, examination type, location and requesting specialty.

Reports were completed faster after AI (adjusted HR 1.17, 95% CI 1.14-1.21). Median reporting time fell from 21.3 to 18.2 minutes, a 14.6% reduction. The largest reductions were in ECG-gated CT (41.1%) and among thoracic radiologists (25.0%). Emergency examinations took 7.1% longer. Exploratory modelling put the saving at about half a full-time radiologist.

AI is sold on efficiency, but real-world reporting time data have been scarce. This shows a modest saving that depends heavily on who reads and what is being read.

The increase in emergency reads is a reminder that AI output adds something to check when the clinical question is not about nodules.

• Measure reporting time before and after deployment rather than assuming a saving.
• Consider whether nodule AI output is needed on emergency CTs with a different clinical question.
• Expect the largest gains among subspecialty readers and in nodule-heavy work.
• Do not count the modelled workforce saving until your own data confirm it.

Why it matters

Efficiency claims for radiology AI now have a real-world figure, and it is smaller and more uneven than marketing suggests.

Don't overread it

This was a before-after comparison at one centre, so other changes over 2021-2024 may account for part of the time saved.

The statistics, in plain English

A hazard ratio of 1.17 here means reports were 17% more likely to be finished at any given moment, which translates to the three-minute median reduction. The workforce projection is an extrapolation, not a measured outcome.

Read the rest in the app

You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.

Get Daily Doctor — free

Scan to keep reading on your phone. No account needed to start.