THE ARISING · Longevity & Healthspan
Could an ordinary chest X-ray help flag fracture risk?
In people already referred for bone-density testing, an AI model ranked fracture risk better than FRAX with bone density. It has not been shown to prevent fractures.
The same image taken to check the chest might eventually offer a second clue, about the likelihood of a major bone fracture. The evidence so far is from people who also had a bone-density scan.
The evidence comes from a cohort study published in NEJM AI in 2026. Researchers led by Kim and Kong built a computer model using 53,109 adults at one tertiary hospital. They then tested it in 85,284 patients at two separate institutions. The outcome was a major osteoporotic fracture: a break of the kind most tied to weak bones.
Doctors usually judge fracture risk with bone mineral density, measured by a dual-energy X-ray absorptiometry scan, and with FRAX, the Fracture Risk Assessment Tool. According to the paper, both capture risk incompletely in the osteopenic range. Osteopenia is the borderline zone between healthy bone and osteoporosis, and the authors note that much of the fracture burden occurs there.
The new model is a convolutional neural network, a type of image-analysis AI, built to estimate risk over time. It looks at an ordinary chest radiograph. That makes it an opportunistic approach: it uses an image already taken, with no added scan or radiation visit.
In the external validation groups, participants averaged 58.6 years of age, 68.9% were female, and follow-up averaged 77 months. The model reached a C-index of 0.843 (95% confidence interval, 0.837 to 0.850). The C-index measures how well a tool ranks people, so that those who go on to fracture are scored as higher risk than those who do not. A score of 0.5 is chance and 1.0 is perfect.
FRAX with bone density scored 0.779 (95% confidence interval, 0.770 to 0.787). For the 10-year area under the receiver operating characteristic curve, another measure of how well a test separates people who fracture from those who do not, the model scored 0.855 against 0.785 for FRAX with bone density.
In the 35,358-person osteopenia subgroup, the model’s 10-year discrimination score was 0.779, versus 0.659 for FRAX with bone density. That is the group where a bone density result alone often leaves the next step unclear. They conclude that the model identified patients at high fracture risk with discrimination exceeding that of FRAX.
Every participant had already been referred for a bone density scan. The authors say that estimates of absolute risk and of how many cases the approach would find need prospective testing in an unselected group of people having routine chest radiographs.
That testing is the next thing to watch. The next step would be to see how many people are accurately flagged in ordinary X-ray practice, how often unnecessary follow-up happens, and whether acting on an alert prevents fractures. This study measured risk ranking, not those real-world consequences.