Missed Fractures on X-Ray and Why They Happen

The short answer: The fractures missed most often on X-ray are the subtle ones in anatomy that hides them: the scaphoid and other carpal bones at the wrist, the femoral neck in older patients with osteopenic bone, the cervical spine, the small bones of the foot, radial head fractures visible only as a joint effusion, and rib fractures on films ordered for something else. They are missed for the same handful of reasons every time, and almost none of those reasons is a lack of knowledge. A study of malpractice claims against radiologists in the United States, published in Radiology, found failure to diagnose to be the most commonly alleged error, with extremity fractures the second most frequently missed finding after breast cancer.

Updated July 2026.

Every radiologist knows the list. Ask any group which fractures they worry about and you will hear the same five or six body parts, in roughly the same order, with a story attached to each one. That consistency is the interesting part. If the misses were random, they would be scattered across anatomy. They are not. They cluster tightly, which means they are a property of how these studies get read rather than of who is reading them, and that in turn means they respond to process changes rather than to exhortation.

Which fractures are most commonly missed on X-ray?

The pattern in the published error literature and in departmental discrepancy meetings is remarkably stable across settings. These are the repeat offenders.

Why are fractures missed on X-ray?

Four mechanisms account for most of it, and they compound.

The first is satisfaction of search. Once one abnormality is found, the probability of finding a second one on the same study drops sharply. A patient falls off a ladder, the wrist fracture is obvious, and the radial head fracture on the elbow film ordered at the same time gets a shorter look than it would have alone.

The second is search pattern. Radiographs are read against a clinical question, and attention follows that question. This is why rib fractures on a chest film ordered for shortness of breath and vertebral compression fractures at the edge of the field of view come up so often in discrepancy review. Nobody was looking there, and nothing about the study asked them to.

The third is volume and fatigue. Error rates in radiology are measurably worse at the end of a long shift than at the start, and plain films are the study most likely to be read at speed because each one individually is quick. Several hundred quick reads still add up to a full day of sustained attention, and sustained attention is the thing that degrades. This is the same arithmetic behind radiologist burnout, viewed from the quality side rather than the wellbeing side.

The fourth is the film itself. A technically limited study, a patient who could not be positioned, a portable film in a resuscitation bay. The genuinely occult fracture belongs in this group too, because on a day-one radiograph it may be invisible to any reader.

What is an occult fracture?

An occult fracture is a fracture that is present but not visible on the initial imaging study. It is not the same thing as a missed fracture, and conflating the two makes departmental review conversations go badly. A missed fracture was visible and was not reported. An occult fracture was not visible, and it appears later on repeat radiographs once resorption at the fracture line widens it, or immediately on MRI or CT. The scaphoid is the archetype: this is precisely why immobilize-and-reimage pathways exist for clinically suspected scaphoid injury with normal radiographs, and why no imaging tool, human or software, removes the need for that pathway.

Are missed fractures a common cause of malpractice claims?

Yes, and they sit near the top. The malpractice claims analysis published in Radiology found that failure to diagnose was the most commonly alleged error in suits against radiologists in the United States, and that within that category extremity fractures ranked second in frequency, behind breast cancer. Reviews of emergency department imaging error reach the same conclusion from a different direction, consistently identifying failure to identify fractures as one of the largest single categories of diagnostic error in that setting.

The reason is not that fractures are the hardest thing radiologists read. It is that they are extremely common, read quickly, and produce a clear before-and-after record when something goes wrong. A missed subtle finding on a complex MRI is hard to litigate. A missed hip fracture that displaced two days later is not.

Which fractures do emergency departments miss most?

The emergency setting shifts the list slightly, because who reads the film first changes. Where an emergency physician makes the initial interpretation and the formal radiology read follows hours later, the discrepancies concentrate in exactly the fractures that need experience to recognize: radial head effusions, subtle cervical spine alignment, pediatric buckle fractures, and the second fracture on a patient who already has an obvious one. Overnight coverage makes it worse, since the studies read at 3am are read by whoever is available under the least favorable conditions of the whole week. Groups that cannot fill the overnight seat internally usually turn to locum or nighthawk coverage, and even where the sourcing and candidate screening is automated, the lead time on a credentialed radiologist is measured in months rather than weeks.

Can AI catch fractures that radiologists miss?

Sometimes, and the honest version of the answer has three parts.

The evidence is genuinely strongest for extremity fractures at the wrist, ankle and shoulder, which happens to overlap well with the list above. It is weaker and considerably more variable for ribs and spine. And it depends on your equipment: a multisite evaluation published in the American Journal of Roentgenology in 2026 found that a commercial fracture algorithm performed differently depending on which radiography vendor produced the image, which matters a great deal if your sites run a mixed fleet and the vendor validated on one of them.

Regulatory bodies have moved cautiously in the same direction. In January 2025, the United Kingdom's National Institute for Health and Care Excellence issued a conditional recommendation allowing several fracture detection tools to be used in urgent care while further real-world evidence is generated, alongside a trained professional rather than instead of one. That is the correct posture, and it is the posture worth adopting internally regardless of where you practice.

What software actually changes is attention and ordering. A flag is a prompt to look again at a region on a film you were about to sign, and a flagged study can move up the reading list instead of waiting in arrival order. That is the whole of the mechanism behind AI fracture detection software, and it is also its limit. The dangerous failure mode is not the false positive, which costs a few seconds. It is the unflagged film that starts to feel cleared. Any group deploying this needs to say out loud, repeatedly, that a study without a flag has not been checked by anything.

If you are comparing tools in this category, the market leader by volume is Gleamer BoneView, now part of RadNet's DeepHealth after the March 2026 acquisition, and the honest feature comparison is on our Gleamer alternative page. A wider view of who does what across the category is in the best AI radiology software roundup.

What actually reduces the miss rate?

Four things, roughly in order of effect per unit of effort.

Fix the ordering, not the reading speed. Most of the time between a study arriving and a fracture being reported is spent waiting to be opened, not being read. Reordering the queue so trauma films do not sit behind routine pre-operative imaging costs nothing clinically and is the highest-yield change available to most groups. This is the argument in full on our radiology worklist software page, and the underlying numbers are in radiology report turnaround time.

Make the second look structural. Whether that second look comes from a colleague, a formal double-read on defined study types, or software, the point is that it does not depend on anyone remembering to do it. Voluntary second reads happen on the studies people are already worried about, which are not the studies being missed.

Keep the follow-up pathways. Clinically suspected scaphoid injury with normal radiographs gets immobilized and reimaged. That protocol predates every tool discussed here and survives all of them.

Read your own discrepancies. Most groups either do not systematically compare preliminary and final reads, or they do and nobody looks at the aggregate. The pattern in your own data will not match the generic list above, because it reflects your case mix, your staffing and your equipment.

How do you measure your own miss rate?

Three data sources are already in your systems. Addenda and report amendments show where a finding was added after signing. Preliminary-to-final discrepancies show where the overnight or emergency read differed from the formal one. And repeat imaging within a short window, particularly a CT or MRI following a normal radiograph of the same body part, marks the cases where somebody did not believe the film.

Pull three months of each, group by body part, and the list will be short and specific. It is a better basis for a decision about second-read tooling than any vendor's evidence pack, because it tells you which body parts are costing you and therefore what a tool would have to be good at to be worth buying. Bring that list to the demo and ask each vendor to run it, on your images, from your equipment.

If that shortlist points at trauma radiographs, our AI fracture detection page covers what a second pass does and does not do on plain films, and the questions worth asking before you sign anything are collected in radiology AI vendor evaluation questions.