Statistics from Altmetric.com
Editor,—We were interested in the suggestion by Mølgaard and colleagues that whole body bone mineral content (BMC) measured by dual x ray absorptiometry (DXA) should be considered as a function of bone area.1 Using the same type of DXA equipment (Hologic QDR-1000W) we measured whole body BMC in 300 normal girls aged 11–18 years.2 Our results for whole body BMC, both as a function of bone area and age, were essentially identical to those of Mølgaard et al.1 To assess the usefulness of area normalisation in patients we have analysed the results from 71 studies on 51 anorexic females aged 14–18 years. The patients’ clinical state was subdivided into “moderate” (n = 44) and “severe” anorexia (n = 27) based on a body mass index of 15.1–18.9 kg/m2 and < 15.0 kg/m2respectively. Height was within the normal range for age and there was no significant difference in age or height between the two groups.
Z(Age) and Z(Area) were calculated using the predicted values and SDs for age matched and area matched normal subjects from Mølgaardet al.1 Mean (SD) Z(Age) was −1.01(0.84) and −1.45 (0.95) for the moderate and severe anorexic groups respectively. For both groups Z(Age) was significantly lower than zero (p < 0.0001) as expected and the value for the severe anorexics was lower than that for the moderate anorexics (p = 0.05). However, Z(Area) was 1.85 (1.39) and 4.41(1.60) for the moderate and severe anorexic groups respectively. For both groups Z(Area) was significantly greater than zero (p < 0.0001) and the value for the severe anorexics was significantly greater (p < 0.0001) than that for the moderate anorexics, which is a most implausible result. Reduced bone mineral density in anorexia nervosa has been reported for specific anatomical sites3 and a significant reduction in total BMC in anorexic patients has also been observed.4
Our results in patients with anorexia nervosa illustrate that normalising whole body BMC by the measured area is clearly inappropriate. Misleading conclusions could be made if this method is used to assess the bone mineral status of patients who may have a reduced bone content as a consequence of disease. We believe this anomaly is related to an artifact of DXA processing. A related anomaly has been investigated in longitudinal studies of adult women and with an anthropomorphic phantom.5 The phantom results showed that a reduction in whole body BMC was accompanied by concomitant changes in the measured bone area, even although the true skeletal area of the phantom was not altered. Both in vivo and in vitro, a change of fat distribution led to spurious changes of BMC and bone area. These anomalies varied with the software version used in the analysis. A similar anomaly has been reported for DXA spine scans.6This artifact may be due to the need to assign a bone density “threshold” when determining which pixels represent bone, with the result that the area measured by DXA depends on the BMC value and the soft tissue composition and is not an adequate estimate of the true projected skeletal area.
Our results suggest that the bone area may be significantly underestimated in children with diseases which are known to result in loss of bone mineral. We conclude that for Hologic DXA systems normalisation of whole body BMC by area introduces significant errors which may result in misinterpretation of the bone mineral status. The substantial reduction in the coefficient of variation reported by Mølgaard et al 1 when BMC is normalised by bone area rather than age must be due in part to the spurious link between BMC and area.
Drs Mølgaard, Prentice, Cole, et al comment:
Hannah et al raise an important issue about the interpretation of bone mineral measurements in girls with severe anorexia. They report very high BMC for bone area Z scores in a group of anorexic girls aged 14–18 years.
Our method of interpretation separates out the various components that influence BMC,1-1 and, had it been used by Hannan et al, they would have been alerted to the fact that the apparently bizarre results were associated with unusually low values for bone area. We have recently seen the same anomaly in Cambridge data. Among adolescent girls examined twice 15 months apart, one developed severe anorexia between the two measurements (S Stear, A Prentice, unpublished data). She lost 5.9 kg, went from a body mass index of 17.4 to 15.2 and lost 12% of her whole body BMC. During the same period her bone area for height Z score decreased from −1.31 to −2.36 while her BMC for bone area Z score increased from −1.21 to +4.45. The problem is likely to be due to poor bone edge detection in these severely anorexic girls as a result of their abnormal soft tissue composition and thickness.
We have examined several hundred children with different chronic diseases likely to affect BMC, including cystic fibrosis, cancer, rheumatic diseases, and osteogenesis imperfecta. In none of these groups have we seen results like those described for the girls with severe anorexia. Of special interest is a group of children with osteogenesis imperfecta (qualitative collagen defect). They are known to have a very low BMC and often have fractures. The children are short, and have mean BMC for age and bone area for age Z scores of −3.7 and −4.9 respectively (A Lund, C Mølgaard, unpublished data). Despite a very low BMC for bone area (Z score −2.7) their bone area for height is only slightly reduced (Z score −0.37), implying that a low BMC is not always associated with a low bone area for height.
Hannan et al refer to a “spurious” association between BMC and bone area. There is indeed a strong association between these variables, but it is not spurious, it is genuine. This is firstly because bigger bones have more minerals on average, and secondly because of the nature of bone edge detection in absorptiometry. Although the algorithms have not been divulged by the manufacturers, it is reasonable to assume that bone is detected when beam attenuation increases above a set threshold. For two bones of the same external dimensions, as the beam moves from soft tissue into bone, the bone with the higher mineral content will cause more rapid attenuation, will be detected earlier and its measured bone area will be greater. As a result, although bone area is closely related to actual bone dimensions, it is also influenced by BMC. The relationship between BMC and bone area is not proportional in most cases, and is not removed by calculating bone mineral density (=BMC/bone area).1-7Normalisation of BMC for bone area, as in our stepwise diagnostic method,1-1 and in regression models,1-7 allows for this association. This approach has the advantage of not constraining the data to fit an arbitrary, often incorrect, relationship (that is bone mineral density) and so minimises the potential for misinterpretation.
In summary, we agree with Hannan et al that bone mineral assessments of girls with severe anorexia can be anomalous. This is due to erroneously low measurements of bone area, probably caused by problems with bone edge detection. It would be unwise to assume that the corresponding BMC or bone mineral density values are valid or clinically meaningful in this situation. The identification of this problem underscores the strength of our interpretative approach,1-1 which does not rely on the calculation of bone mineral density but makes use of all the information available in a decision making tree and ultimately normalises BMC for bone area.