Can You Glean Bone Density From A Plain X-Ray?

"Validated in a peer-reviewed journal" doing a lot of work here.

A “breakthrough device” from 2023 has now popped up on the PRNewswire with 510(k) clearance: OsteoSight.

The press release happily describes the scope of the problem, that there are potentially missed opportunities to identify bone mineral density from plan x-rays acquired for other purposes. Sounds great! Their technique is “validated in peer-reviewed research”, and linked from the press release.

And, here are the performance characteristics for that validation:

At their pre-specified cut-point, that’s a sensitivity of 0.63 and a specificity of 0.85.

Assuming this diagnostic performance is true – and that’s a bit of a stretch considering their small training sample and exclusions of anything that might confound the algorithm – where do you apply it? In the high-prevalence populations where the low specificity isn’t punishing, these are (read: elderly women) the sorts of folks who are already targeted for routine screening and early treatment. Then, if this algorithm is deployed to try and catch extra cases in a low-prevalence population, the vast majority will be false positives.

Trickier still, who pays for it? The ROI has to be decreased total health expenditures for a public insurance (see: Medicare) program – and jumping from the possible incidental low BMD detection, to a sufficient quantity of new true positives from screening, to an effective early intervention, to preventing a fracture from a fall is a big ask.