AI-assisted fracture detection on emergency X-rays — a randomized trial that made time in the department, not accuracy, the primary outcome (NCT06754137)

1An emergency X-ray gets read twice

When an X-ray is taken in the emergency department, the frontline physician reads it on the spot, and the formal report from a radiologist arrives afterwards. What the record of this trial states plainly is the operational reality at the participating hospitals: that report generally does not arrive before the patient has left.

In other words, at the moment treatment is decided, the treating physician is relying on a single reading. Placing AI at that point is less a contest over whether it beats a specialist than a question of whether it can fill the window in which specialist judgment is unavailable.

2What it means to make time the primary outcome

Most medical-AI trials put diagnostic accuracy at the centre. This one chose the time from triage to completion of emergency department treatment, and moved accuracy to the secondary outcomes. In an emergency department, reading correctly and keeping patients moving are different kinds of value, and the second one feeds directly into how many people the hospital can take.

Measuring diagnostic confidence, use of additional imaging, and missed fractures alongside it is an attempt to see from several angles how AI changes what physicians do. The result figures are not part of this registry record.

3One of the few trials that reached an answer

Of the 750 medical-AI clinical trials this site holds as of 2026-08-28, 164 are recorded as completed — 22 percent of the total. Another 247 are recruiting and 215 have not begun recruiting, so 62 percent are still on their way to a result. This trial began in October 2025 and completed in April 2026, placing it in the minority that finished within seven months.

Clinical evaluation of medical AI has not accumulated results at anything like the rate at which trials are registered. Against that backdrop, a pragmatic randomized trial that reached completion carries weight as a reference point.

4What to place as the primary endpoint

Most medical AI trials place diagnostic accuracy as the primary endpoint. This one chose something else.

When an X-ray is taken in the emergency department, the front-line physician reads it there and the formal radiology report arrives later. The record states explicitly that at the participating hospitals that report ordinarily does not arrive before the patient leaves.

Measuring diagnostic confidence, use of additional imaging and missed fractures alongside is a design for seeing from several angles how AI changes physician behaviour.