3D-DXA does not measure anything

In a previous blog post I explained that 3D-DXA can not measure the cortical thickness. However, the problem goes deeper. The way 3D-DXA works is that it does not measure anything at all. What it in fact does is take an educated guess of the bone geometry based on the geometry measured from a previously acquired set of hip CT scans. A mathematical technique called principal component analysis is used to describe the bone geometry with only a few numerical values by limiting the geometry to only the main variations. Any 3D bone model acquired by 3D-DXA is generated by these values, which describe only the main variations of the input population. Now, it should not surprise you that femur bones of women are much smaller than those of men. Also the bones of a Spanish population are considerably different than those from a Dutch population, which by the way are now officially the tallest people in the world. So what will happen when you try to reconstruct the 3D bone geometry from a DXA image of a Dutch guy using a model built from a predominantly elderly female population, as is the case with the commercially available 3D-DXA software? Well, nothing good I can tell you.

The way in which 3D-DXA finds the values that describe the bone geometry is through an optimization process. It searches through the numerical values that describe the model until the projection of the bone model is similar enough to the DXA image (or until it has taken too much time). So what you end up with is a bone model of an elderly woman that happens to be similar to the projection of the bone from the Dutch guy. That is of course if the projection of the model and the DXA image actually manage to match up. The last time Luis Del Rio Barquero showed me the software provided by Galgo Medical and DMS-Apellem it would not allow you to compare the DXA image with the final model projection. In fact, there is no way at all to assess whether the 3D reconstruction was successful or not. Even if the images do match up, the density of the bone might very well be systematically overestimated or underestimated, giving a false assessment of the bone quality and ultimately a false diagnosis. It should be absolutely clear, and I can not stress this enough, that this is not something that can be resolved using more research and development, this is a fundamental limitation of this technology. It was a nice PhD project but 3D-DXA has no place in real clinical practice or medical research. It therefore genuinely scared me how many poster presentation there were at the Annual Meeting American Society for Bone and Mineral Research last year, which do exactly that. They include a study on type 1 diabetes patients1, patients with early breast cancer under aromatase inhibitors2, long latent autoimmune diabetes3, celiac disease after 1-year on gluten-free diet4, and perhaps worst of all, a study on individuals with small bone size and low bone mineral density5. As Ben Goldacre said in his book “Bad Pharma”: “It feels as if some people, perhaps, view research as a game, where the idea is to get away with as much as you can”.

  1. Humbert L. ,Fonollà R., Keil M, Lehmann G, Sämann A, Neumann T. Analyzing the cortical and trabecular bone of type 1 diabetes patients using 3D-DXA – a longitudinal study. ASBMR 2016 Annual Meeting.
  2. X. Nogués, L. H. Humbert, M. Rodriguez-Sanz, S. Servitja, N. Garcia-Giralt, L. Garrigos, J. Rodriguez-Morera, L. Mellibovsky, M. Martinez-Garcia , R. Fonolla , Y. Martelli, J. Romera, A. Diez-Perez, I. Tusquets. Cortical and Trabecular Bone Analysis of Patients with Early Breast Cancer under Aromatase Inhibitors, B-ABLE Cohort, using 3D-DXA: a Longitudinal Study. WCO-IOF-ESCEO 2016.
  3. M. A. Guagnelli, R. A. Gomez-Diaz, N. Wacher, L. Humbert, Y. Martelli, G. Gonzalez-Castelan, P. Clark. Cortical and trabecular bone analysis with 3D-DXA in patients with Type 2 Diabetes, Long Latent Autoimmune Diabetes in adults (LADA) and healthy controls: a preliminary report. WCO-IOF-ESCEO 2016.
  4. María Belen Zanchetta, Ludovic Humbert, Martelli Yves, Vanesa Longobardi, Mariela Sesta, cesar bogado, Jose Zanchetta. 3D-DXA analysis of the changes in cortical and trabecular bone in patients with celiac disease after 1-year on gluten-free diet. ASBMR 2016 Annual Meeting.
  5. B. M. Camargos, B. C. Silva, L. Humbert, L. Del Río. Clinical Applicability of 3D-DXA in Individuals with Small Bone Size and Low Bone Mineral Density: a Case Report. WCO-IOF-ESCEO 2016.