SEARCH

SEARCH BY CITATION

Abstract

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. Acknowledgements
  8. REFERENCES

For an eating disorder study over a period of 1 year, we measured total-body bone mineral using a Hologic QDR 1000W in a total of 157 subjects and observed anomalies that questioned the accuracy of such measurements. Using the recommended Enhanced software, a change in total bone mineral content (ΔBMC) correlated positively with a change in weight (ΔW; r = 0.66), but a loss of weight was associated with an increase in bone mineral areal density (BMD; r = 0.58), arising from a reduction in bone area (AREA). Both regressions were highly significant. The dominant factor in this relationship was a strong correlation between ΔAREA and ΔBMC, for all parts of the skeleton, r > 0.9, with a slope close to 1. This is implausible because bone area would not be expected to change. When Standard software was used, the slope of the ΔBMC/ΔW correlation was steeper, but the ΔBMD/ΔW regression became positive. An artefact of dual-energy X-ray absorptiometry processing was suspected, and phantom measurements were made. The phantom consisted of tissue-equivalent hardboard cut and stacked to form cylinders corresponding to the head, trunk, arms, and legs of a standard man. The skeleton was constructed from layers of aluminium sheet as an approximation of the average shape, BMD, BMC, and AREA in each region. When aluminium thickness was varied, BMD thresholds were found, approximately 0.4 g/cm2 for the legs and 0.2 g/cm2 for the arms. Above these, bone area rose fairly rapidly toward a plateau. At higher skeletal densities, the relationships between measured and true BMDs were close to linear, but slopes were less than unity, so that changes would be underestimated by 10–30%. Increases of thickness of the soft tissue of the phantom lowered AREA slightly. Uniform fat proportion increases led to decreases in BMC and AREA, but lard wrapped in an annulus around the limbs led to spurious increases in BMC and AREA of a similar magnitude to those observed in vivo, while BMD fell slightly, although there had been no true change of bone variables. Similar results were obtained with lard around the limbs of a volunteer. Reanalysis of phantom scans using Standard software confirmed the software differences noted in vivo. The phantom measurements offer an explanation of the anomaly in vivo and demonstrate that, under different circumstances, change in both BMC and BMD can be wrongly recorded. We believe that no valid conclusions can be drawn from measurements by the Holgic QDR 1000W of bone changes during weight change.


INTRODUCTION

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. Acknowledgements
  8. REFERENCES

THE REPRODUCIBILITY, or precision, of dual-energy X-ray absorptiometry (DXA) is well established and has led to its confident application to longitudinal studies of possible changes of bone mineral status with disease or treatment. The majority of applications concern spine or hip measurements, but there is an increasing use of total-body DXA, where coefficients of variation of repeated measurements of about 1% are obtainable.

The accuracy of DXA is much less well known, mainly due to the lack of a reference standard. The limited validity of the necessary assumptions means that high accuracy of whole-body bone mineral cannot be expected. This limitation has not been thought to be serious in longitudinal studies, where the measurement of changes is the chief concern. We have discovered evidence that even the measurement of changes may be seriously in error when subjects undergo significant weight change.

In a study of the relationship between bone mineral and weight change, patients with eating disorders and normal controls were measured twice, with an interval of 1 year. As expected, changes of total bone mineral mass (bone mineral content, BMC) correlated positively with weight change. However, changes of areal bone mineral density (BMD) appeared to correlate negatively with weight change, because bone area (AREA), which serves as the denominator in determining BMD, correlated positively with weight change and with BMC. Explanations for this unexpected finding might include: (1) a genuine change of AREA as bone mineral was added to or removed from the skeleton. This possibility is most unlikely and against anatomical, physiological, and radiological evidence. It would require bone width to change by over 10%, in some cases, in a relatively short time and in either direction. (2) There might be an artefactual dependence of AREA and/or BMC on soft tissue thickness or composition; this issue has been examined to some extent by a number of authors.1–7 (3) The measured AREA might, through an artefact of scanning or analysis, vary with the BMC or BMD independently of tissue thickness. Peel and Eastell have reported such an anomaly for spine scans.8 More detailed analysis of in vivo data and the performance of phantom measurements were undertaken to investigate the anomaly. Because DXA procedures are tailored to measurements of the human body, the phantoms were designed to be more anthropomorphic than those commonly used.

MATERIALS AND METHODS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. Acknowledgements
  8. REFERENCES

Subjects

Premenopausal women with a current or past eating disorder were recruited. The categories included current and recovered anorexia nervosa, current bulimia nervosa, and current and recovered subthreshold. Further details of the subjects, together with the clinical findings, will be published elsewhere. The results are included in this presentation only to illustrate the anomalies in DXA measurements. Age-matched control subjects with no eating disorder history were also recruited. The mean age of the subjects was 30.6 years, SD 8.7, range 17–52.

Total-body bone mineral, spine bone mineral, fat, and lean mass were measured by DXA, using a Hologic QDR 1000W (Hologic Inc., Waltham, MA, U.S.A.). Whole-body software Enhanced Whole Body, V5.51, and spine software V4.47P were used for analysis. The Enhanced Whole Body version was used because it provides a more regional assessment of tissue attenuation and is recommended by the manufacturer for greater accuracy. However, for comparison with other reports and because we have previously found that the different whole-body software leads to different results, the scans of 20 subjects were reanalyzed using the Whole Body version, which we shall refer to as Standard; this corresponds to the “Old” software referred to in our previous publications.1,2 One hundred and twenty-nine eating disorder subjects and 28 controls had two measurements 1 year apart.

For comparison, results from subjects measured as part of their clinical management were examined. There was no selection for clinical condition, weight change, age, or gender. Repeated measurements of a group of volunteers undertaken to establish reproducibility were also considered.

Phantoms or models

Preliminary studies demonstrated the desirability of varying phantom thickness, fat proportion, and bone density independently and of simulating the human body closely. The phantoms used were based on the Bush phantom, much used in the measurement of total body radioactivity.9 This consists of hollow circular and elliptical cylinders of polyethylene, approximately 3 mm thick, corresponding to the head, neck, chest, pelvis, arms, and legs of a standard man, which can be filled with water or other liquid. Anterorposterior thickness was 10 cm in the arms, 15 cm in the upper leg, 12 cm in the lower leg, and 20 cm in the trunk. The weight was 70 kg.

The phantom has the advantage that the X-ray attenuation characteristics can be varied by changing the liquid contents, but it introduces sharp discontinuities at intervals along the body due to the fat-like polyethylene end walls of the cylinders. Previously this had not been found to upset the Hologic analysis unduly, but had introduced problems with DXA instruments of some other manufacturers. Accordingly, a phantom of similar dimensions had been constructed from sheet hardboard. Consisting largely of cellulose, this is reasonably equivalent to soft tissue in its X-ray attenuation characteristics. The hardboard was cut and stacked in horizontal sheets to form cylinders of the same dimensions as those of the Bush phantom. The soft tissue attenuation characteristics were constant, but the thickness could be varied.

The skeleton was cut from sheet aluminium, mostly 1.68 mm thick. The dimensions were chosen to approximate those of a standard man and the full skeleton had BMC, BMD, and AREA of each component similar to those found in vivo. The head, pelvis, and ribs were complex and invariate in thickness. The “bones” of the arms, legs, and spine were rectangular and the number of sheets could be varied. The full skeleton contained four thicknesses of aluminium in the arms and spine and five in the legs, but measurements were carried out also at reduced effective BMDs. Bone width was 25 mm in the arms and 45 mm in the spine and legs. The bone equivalence of the aluminium was established by calibration against calcium hydroxyapatite, using the Hologic DXA apparatus in the spine scanning mode, so that results could be plotted against nominal BMD values. The total AREA was 2300 cm2 and the nominal maximum BMC 2120 g. A DXA scan in Fig. 1 illustrates the phantom.

thumbnail image

Figure FIG. 1.. DXA scan of phantom.

Download figure to PowerPoint

The Bush phantom was used to investigate the variation of measured BMD, BMC, and AREA with the fat content of the soft tissue while the thickness remained constant. Measurements in vivo established that the limbs, and particularly the legs, made an important contribution to the changes in bone mineral and fat content of the body, so one thigh section, 15 cm in diameter, was used as a variable component. Paraffin wax was cast into the bottom semicylinder and the remainder filled with water. The leanness was increased by dissolving different amounts of sodium chloride in the water, giving a range of percent fat (as measured by the QDR 1000W) from 18–58%. This mimics the situation in vivo since leanness is determined in X-ray attenuation terms largely by the mineral salt content. The thigh section, with its aluminium femur, was scanned in the upper and lower leg positions, with the remaining sections, with a full skeleton, in their normal positions. It was desirable to maintain a simulation close to reality because some of the bone measurements with the QDR 1000W are position dependent.

The hardboard phantom was used to investigate the variation of bone measurements with soft tissue thickness and with skeletal BMD. A limited change of fat proportion was also achieved by interleaving thin aluminium sheet with the hardboard of the lower leg portion, 12 cm in diameter, used in different positions, reducing the measured fat from 40 to 12%. Although aluminium has similar attenuation characteristics to bone, it is legitimate to use it also to modify the soft tissue attenuation, provided the relative thickness distribution matches that of the “fat” component. Any tissue can be simulated by a combination of two other substances.10

The closest simulation of the body changes found during weight change was obtained by scanning the hardboard phantom in its standard form, with complete skeleton, with and without packets of lard, contained in plastic bags and spread out to thicknesses of about 10 mm, wrapped around the limbs, adding 1 kg to each arm, and upper and lower leg. To investigate the effect of added fat with a different initial fat proportion, lard was wrapped around the leaner lower leg sections, which were placed in turn in the arm, upper and lower leg positions, with the trunk and head being unaltered. The phantoms were scanned in the same way as the subjects, but with the skeleton on the couch under the “soft tissue.” Sufficient repeated measurements were obtained for each combination of phantom, thickness, and composition to give adequate statistical significance of any differences.

In light of the results of phantom measurements, the effect of added lard was investigated in vivo. A fairly lean volunteer, with measured fat proportion of 12%, was scanned three times with and three times without 6 kg of lard wrapped around his arms and legs in the same manner as with the hardboard phantom. These scans and those of the phantom with wrapped lard were also reanalyzed using the Standard software.

RESULTS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. Acknowledgements
  8. REFERENCES

Subjects

The results that triggered our concern are shown in Fig. 2, where changes in total-body BMD and BMC using Enhanced Analysis are plotted against change in weight. The results were very similar whether they were plotted as absolute changes or percentages. There are highly significant linear regressions, positive for BMC, but negative for BMD. There were some substantial changes of weight and it may appear that the regressions are dominated, or even determined, by outliers. However, when all results with a weight change greater than one standard deviation (SD; 5 kg or 9%) away from the mean were removed, the revised regressions were still highly significant, with slope and intercept not significantly different from those with the full population. The regression equation indicates that a 10% change of weight was associated with a BMC change of approximately 2.5%. During the period of the study, 39% of the subjects lost and 40% gained more than 1% in weight.

thumbnail image

Figure FIG. 2.. (Upper) Correlation between changes in total BMC and weight. (Lower) Correlation between changes in total BMD and weight.

Download figure to PowerPoint

Further inspection revealed that the negative dependence of BMD on weight was due to an apparent positive correlation between changes in bone area and weight. The analysis provides results for regions of the body as well as the total, so regression equations for the correlations between changes in BMC (ΔBMC) and weight (ΔW) and ΔAREA and ΔW are presented in Table 1A. With the exception of the pelvis and the head, all the regressions were highly significant. The slopes differed between body regions, but the slope for ΔAREA was always similar to that for ΔBMC. As a consequence, BMD changed little with weight; there was a negative correlation for the total and for the legs, but no other significant correlations.

Table Table 1.. LINEAR REGRESSION EQUATIONS OF PERCENTAGE ΔBMC AND ΔAREA AGAINST PERCENTAGE ΔWEIGHT IN VIVO
Thumbnail image of

The most significant correlation was that between changes in AREA and BMC, Fig. 3. There were similar correlations for each of the regions. The linear regressions are given in Table 2A. The slope of the regression equation for the pelvis was significantly lower than that for the total body, but there were no significant differences between the other regression coefficients.

thumbnail image

Figure FIG. 3.. (Upper) Correlation between changes in total bone area and BMC. (Lower) Correlation between changes in bone area and BMC of lumbar spine.

Download figure to PowerPoint

Table Table 2.. LINEAR REGRESSION EQUATIONS OF PERCENTAGE ΔAREA AGAINST PERCENTAGE ΔBMC
Thumbnail image of

The foregoing results are for the whole group studied. Separate calculations were also made for groups comprising patients with continuing eating disorders, recovered patients, and normal controls. The corresponding regression equations for total-body bone mineral were not significantly different from each other or from those for the whole group.

It can be noted in Fig. 2 that there are intercepts for the regressions, so that at zero weight change there was a 1.1% increase in BMC, a 1.6% increase in BMD, and a reduction of 0.5% in AREA. No evidence of a change in the calibration of the scanner was evident in the quality control records. These are based on daily measurements of the Hologic spine phantom, so there was a possibility that they might not be representative of the stability of total-body scanning. Further evidence was obtained from repeated measurements of a semianthropomorphic total-body phantom11 that had been scanned for other purposes at intervals over a period of 4 years, including the duration of the study reported here. There were no significant changes of total body AREA or BMC. Over the 4-year period, there was a significant increase of BMD (p = 0.02), but during 1 year this amounted to only 0.4%.

The data from the subject measurements were examined for evidence of calibration change. A group of 45 subjects in whom weight change during 1 year was less than 2% was selected and the percentage change in total BMC was plotted against the initial scan date; these dates covered a period of 10 months. There was no significant correlation, supporting the existence of a stable calibration. Given the variance of the BMC and AREA changes (SEE of the regressions 2.8%, range about twice this), it is not too surprising that there were apparently mean changes of 1% and 0.5% in a year, particularly as they may be related to the anomaly under investigation.

In addition to the results based on the usual regions of interest (ROIs), some individual patient scans were examined to investigate the contributions made by areas of low BMD, because the phantom measurements suggested that these might have a disproportionate effect. Manually selected ROIs were placed over the hands and ribs. Typical measured BMDs in these areas were found to be 0.3–0.4 g/cm2. However, the initial contributions to the total BMC were less than 5%, so that any influences of a BMD threshold would have a negligible effect on the total-body bone changes.

The absorptiometer also measures the fat and lean components of the soft tissue and, in view of the influence these may have on the bone results, the results are presented in Table 3. It will be noted that the fat in the limbs comprises more than 60% of the total and contributes more than 50% of the changes. Changes in the fat component made up 67% of the weight change on average, but 80% in the legs.

Table Table 3.. CONTRIBUTION OF BODY REGIONS TO THE TOTAL BONE AND SOFT TISSUE MASSES AND TO CHANGES, EXPRESSED AS A PERCENTAGE
Thumbnail image of

The results of the reanalysis of scans using the Standard software are shown in Table 4. There were no significant differences between the mean values of ΔBMC, ΔAREA, or ΔBMD for the selected group and the whole population. The slope of the ΔBMC/ΔW plot was little different from the whole population for Enhanced analysis, but steeper for Standard. Because there was no significant difference between the ΔAREA/ΔW plots for the two protocols, the ΔBMD/ΔW relationship was different. The negative slope observed with Enhanced was reversed with Standard. The ΔAREA/ΔBMC plot was steeper for Enhanced. There were significant differences between the mean initial bone measurements, with Enhanced producing a higher BMC and BMD and a lower AREA, included in Table 4. However, it was found that the ratios of Enhanced/Standard results were dependent on the percentage fat of the body as determined by the absortiometer. The parameters of linear regressions of this correlation are also included in Table 4. The two protocols give the same result for all three variables at about 30% fat, but at lower fat proportions. Enhanced Analysis leads to values that are up to 20% higher.

Table Table 4.. LINEAR REGRESSION EQUATIONS OF TOTAL-BODY RESULTS FROM 20 SUBJECTS ANALYZED WITH ENHANCED WHOLE BODY SOFTWARE AND STANDARD SOFTWARE
Thumbnail image of

All subjects also had separate lumbar spine scans on the same visits. There was no dependence of ΔBMC or ΔAREA on ΔW. However, there was a strong correlation between ΔBMC and ΔAREA, although with a much lower slope than that found for the total body, Fig. 4. The linear regression, using percentages, was ΔAREA = 0.312 × ΔBMC + 0.25, r = 0.61, p < 10−10, SEE = 1.6.

thumbnail image

Figure FIG. 4.. Plots of AREA and BMC against uniform fat proportion in phantom limb with no variation of bone or soft tissue thickness. The regression equations are:

  • equation image

where ΔF is the change in fat%.

Download figure to PowerPoint

Software is available to make use of the attenuation data to measure the body thickness at any point in terms of the equivalence of epoxy resin plastic, used as one of the calibration standards. This facility was used to measure the thickness over the abdomen and midthigh of a number of subjects who had changed weight appreciably. Changes of thickness were then correlated with weight change. There was no significant difference between the slopes for the two regions, and a combined slope of 2.4 mm/kg was found.

To ascertain whether anomalies occurred in different populations, the database of total-body scans performed for clinical purposes was examined, and 62 who had had two scans at an interval of at least 1 year were identified. They were not part of a weight change study, but there had been some such changes, in both directions. The SD of ΔW was 4.4 kg, and 25 subjects had changed by more than 2 kg. When the whole group was considered, there was no correlation between total-body ΔBMD and ΔW, but both ΔBMC and ΔAREA rose with ΔW. The parameters of the regression equations are included in Table 1B. The intercepts were not statistically significant. The correlations ceased to be significant when the analysis was confined to the 37 subjects whose weight changed by less than 2 kg. However, for each of the categories, there was a highly significant correlation between ΔAREA and ΔBMC, as presented in Table 2B. The results for the 12 men included in the group were analyzed separately. There were no significant differences of any of the regressions from those of the whole population.

The records of repeated total-body scans undertaken to establish short-term reproducibility were examined. Two measurements had been made without an interval on 17 subjects. There were no systematic differences between them, but there was a significant correlation between the differences of AREA and BMC, Table 2C.

Phantoms

The effect of body thickness on the bone measurements was investigated with the hardboard phantom. When the thickness was reduced by the same proportion in each body compartment, the total-body BMC was unchanged, but the AREA increased significantly, by 2.2% for a 15% thickness reduction and 2.5% for a 25% thickness reduction. There was no further reduction when thickness was reduced to 50% of the original. The consequence of these changes was therefore an increase of BMD with increasing phantom thickness and weight. This is in the opposite direction to the findings in vivo, so the thickness effect does not provide an explanation for the anomaly; rather it dilutes it.

The effect of fat proportion on bone measurements when thickness and weight were almost constant was examined using the Bush phantom thigh section. In the analysis of the results, a fixed ROI was chosen to exclude the polyethylene end plates of the cylinder. It was found that the results were the same with the cylinder in the upper or lower leg position. Mean results from at least six measurements at each point are plotted in Fig. 4 against the fat proportion as recorded by the QDR 1000W. One of the points at 18% fat relates to the cylinder with water alone and the other to a combination of a semicylinder of wax and a solution of 200 g sodium chloride in 2.5 l of water. The close agreement of these points provides a justification of the means used to vary the soft tissue composition.

Both the AREA and BMC appear to change with the fat proportion. There was significant linear regression with percent fat, with similar negative slopes for both bone parameters. There was no significant change of BMD, a correct result because there had been no alteration of the aluminium skeleton. By contrast, the BMC measurements gave a false impression. The similarity of the regression coefficients for AREA and BMC leads to a correlation between ΔAREA and ΔBMC, just like that observed in vivo. However, AREA and BMC in the phantom decrease with increasing fat proportion, whereas they increase in vivo. Once again the results in vitro dilute, rather than explain, the measurements on subjects.

When 4 kg of lard was laid flat on the trunk and thighs of the hardboard phantom, there was a significant reduction of 2% in the total-body BMC but an insignificant reduction of AREA. When the lard thickness was doubled, using 8 kg, the BMC reduction increased to 3% and the ΔAREA became significant at 1%. For both added fat masses, the total-body BMD reduced significantly by 1%. These findings are similar to those observed when fat proportion or thickness were increased separately.

However, when lard was added as an annulus wrapped around the limbs, different results appeared. They are presented in Tables 5 and 6. Table 5 shows the results of measurements of the complete hardboard phantom with and without 1 kg of lard around each arm and 2 kg around each leg. Using the Enhanced protocol, results for the arms and legs were similar, with the addition of the lard leading to increases in both AREA and BMC of 5–7%. There was no significant change of BMD in the limbs, but there was in the total body. Interestingly, there was a highly significant change in the trunk AREA, although no lard was added here. The increase in the BMC of the trunk was not significant and the BMD decreased. The nominal AREA for each arm was 150 cm2, so it was overestimated in all the measurements. The arm BMC derived from the calibration against calcium hydroxyapatite was 139 g, so that was also overestimated. With a nominal arm BMD of 0.925 g/cm2, a correct value was obtained. The nominal leg AREA was 360 cm2. The measured values are below this, but a small proportion of the femur is excluded by the ROIs used. The nominal leg BMC of 416 g and BMD of 1.156 g/cm2 are exceeded.

Table Table 5.. RESULTS OF MEASUREMENTS OF THE COMPLETE HARDBOARD PHANTOM WITH AND WITHOUT THE ADDITION OF 6 KG LARD AROUND THE ARMS AND LEGS
Thumbnail image of
Table Table 6.. RESULTS OF BONE MEASUREMENTS OF HARDBOARD PHANTOM, WITH AND WITHOUT 1 KG LARD WRAPPED AROUND EACH LOWER LEG SECTION, PLACED IN THE ARM, LOWER AND UPPER LEG POSITIONS
Thumbnail image of

With the Standard software there was no significant effect of added lard on total AREA, BMC, or BMD, but BMC and BMD were reduced in the legs. All values of BMC recorded using the Standard protocol were lower than those measured using Enhanced, markedly so in the trunk.

The results of scanning with the lower leg section in the three limb positions are presented in Table 6. Fixed, manually chosen, ROIs were used in the analysis, within the ends of the bones, but wider than the soft tissue boundaries. There were four sets of conditions: with and without aluminium added to the hardboard to increase leanness and with and without added superficial lard, and they need to be considered in pairs. The differences between results from the arm and leg positions also need to be considered. There were no differences between upper and lower leg positions.

The position and soft tissue composition clearly have a big influence on the observed bone mineral parameters. The nominal AREA defined by the ROI and the width of the bone was 146 cm2 and values between 118 and 161 cm2 were observed. The BMC derived from the calibration against calcium hydroxyapatite was nominally 169 g, and the measured values ranged from 159 to 225 g. Section A of Table 6 shows that the degree of initial leanness had no significant effect on AREA or BMC, but altered the BMD by 1–2%. When lard had been added, the basic leanness did affect the bone measurements to a small extent, an increase of fat proportion leading to a 3% reduction of AREA and a 6–10% reduction in BMC. The points for the leg are included in Fig. 5 for comparison with the comparable results from the Bush phantom. The data are rearranged in Table 6B to show the effect of the addition of lard, which is seen to be important only for the arm position, increasing the AREA and BMC by 20–40%, with no change in BMD.

thumbnail image

Figure FIG. 5.. Variation of measured BMD with aluminium thickness expressed as nominal BMD in the arms, legs, and spine of the phantom.

Download figure to PowerPoint

There were some differences between the results from the same lower leg section scanned in the arm and leg positions (Table 6C). The positional effect was itself dependent on the soft tissue composition. With no added lard, the AREA and BMC in the leg position were 15–29% higher than in the arm position.

The effect of an annulus of lard around the limbs in vivo is illustrated in Table 7. Using the Enhanced protocol, the results are very similar to the measurements in vitro presented in Table 5. The linked increases in AREA and BMC are greater in the arms and slightly lower in the legs, but the total body changes are the same, and in each case there is a small but significant reduction of BMD. The use of Standard software gives very similar results in the limbs but introduces increases of AREA and BMC and reductions of BMD in the trunk. The increase in total AREA is greater with Standard software, so that the negative slope of BMD is increased.

Table Table 7.. MEAN RESULTS OF THREE MEASUREMENTS OF VOLUNTEER WITH AND WITHOUT 1 KG LARD WRAPPED AROUND EACH ARM AND 2 KG AROUND EACH LEG
Thumbnail image of

The variations of measured BMD with the thickness of aluminium in the skeleton, translated into nominal BMD from the calibration against calcium hydroxyapatite, are plotted in Fig. 5. The pattern varies with the position in the body. There are thresholds of about 0.4 g/cm2 for the legs and 0.2 g/cm2 for the arms, below which no bone is recorded. Exchange of the arms and legs of the phantom demonstrated that the differences were mainly due to the position in the body, but soft tissue thickness also had an effect. The spine registered some bone when none was present, as did the ribs.

The total-body results included in Fig. 5 were obtained by varying the aluminium thickness in the arms, legs, and spine by the same amount. The lowest total-body BMD plotted, 0.5 g/cm2, was achieved with BMDs of 0.23 g/cm2 in the arms, legs, and spine. The highest total-body BMD of 0.87 g/cm2 corresponds to 0.92 g/cm2 in the variable components. The apparent minimum in the total-body BMD is an anomaly due to AREA rising faster than BMC as aluminium thickness is increased at low BMDs, but it is not of practical significance because the proportion of the skeleton with BMDs below 0.5 g/cm2 is small.

At nominal BMDs above 0.5 g/cm2 for the arms, legs, and spine, the variations of AREA and BMC with aluminium thickness are such that the increases of measured BMD are close to linear. However, Fig. 5 does not reveal clearly how near the slopes are to unity. The measurements were further analyzed to determine the slopes at three BMD levels, and the results are presented as the measured increase in BMD for a true increase of 10% in Table 8. These results show that, at BMD levels similar to those found in humans, real changes of BMD are underestimated by 10–30%. Above a BMD of 0.6 g/cm2, the BMC was closely proportional to the nominal values, so a 10% change of leg, arm, or spine BMC resulted in a deviation from that expected by less than 2%.

Table Table 8.. MEASURED PERCENTAGE INCREASE IN BMD OF PHANTOM FOR A 10% TRUE INCREASE IN THE ARMS, LEGS, AND SPINE, AT THREE INITIAL BMD LEVELS
Thumbnail image of

There was a possibility that the use of aluminium rectangles to simulate the variable parts of the skeleton might place different demands on the bone recognition algorithms from those found in vivo. This point was investigated by substituting aluminium cylinders with the same outer diameter, but different wall thicknesses, in the arms. The resultant pattern of variation of AREA, BMC, and BMD with wall thickness was very similar to that found with rectangles. Because the bone mineral-equivalent volume density of aluminium is somewhat greater than that of the mineral dispersed in cortical bone itself, an alternative model was also used. Thin aluminium sheets of different widths were stacked to create an edge thickness gradient matching that of a thick-walled tube simulating a bone shaft. Again the results were very similar to those obtained with rectangles.

The results presented in Fig. 5 were obtained using the Enhanced Whole Body software. The scans were reanalyzed using Standard software. The results were little different. The BMD threshold was reduced for the legs to 0.2 g/cm2, but increased for the arms to 0.25 g/cm2. A threshold of 0.2 g/cm2 was introduced for the spine, and the spine BMD and BMC were underestimated at higher BMD levels. The slopes of the BMD plots at BMDs greater than 0.6 g/cm2 were similar to those with Enhanced software.

The relevance of the phantom measurements to the results in vivo can be assessed by considering the changes that would follow a weight change of 9 kg or a fat change of 6 kg. The results are summarized and illustrated in Table 9. The soft tissue thickness measurements showed that such a weight change would be associated with a thickness change of 2.2 cm, or 15% of the mean of 15 cm. The hardboard experiments with a constant fat proportion revealed no changes of BMC or AREA with thickness at this initial thickness. At greater thicknesses there was a reduction of AREA and an increase of BMD with increasing thickness. The changes were small and in the wrong direction to explain the observed anomaly.

Table Table 9.. MEASURED PERCENTAGE CHANGES OF TOTAL BMC, AREA, AND BMD FOR AN INCREASE IN BODY WEIGHT OF 9 KG OR FAT MASS OF 6 KG
Thumbnail image of

Changes of fat proportion without a change in thickness could also not provide an explanation. If all of a 9 kg increase in weight were attributed to the fat compartment it would result in an increase in fat proportion from the mean initial 24% to 40%. Figure 4 shows that such an increase would result in AREA reducing by 7.6% and BMC by 8.3%. These are again in the opposite direction to that observed in vivo, although they are compatible with there being no change in BMD.

A change of weight of 9 kg in vivo implies a change of fat of approximately 6 kg, the amount used in the lard annulus studies, the results of which are shown in Tables 5 and 7. The addition of 6 kg of lard led to increases of AREA of 5.5% and BMC of 3.4%, in both phantom and in vivo studies using Enhanced Analysis. In the clinical measurements, a corresponding 9 kg weight increase led to an AREA increase of 5.4% and a BMC increase of 5.4%, taking into account the intercepts, 4.3% and 5.9% increases if the slopes alone are considered. While the similarity of these results may be somewhat fortuitous, given the fact that we did not attempt to simulate fat distribution changes in the trunk, they do demonstrate that the results in vivo could be explained by the inappropriate effect of a realistic change of superficial fat in the limbs, without there being any real change in bone mass. The associated apparent change of BMD is in the opposite direction, but the magnitude of the error is less than that of ΔBMC and ΔAREA.

With the Standard Analysis, the annular lard had a similar effect on the total-body BMC and AREA of the volunteer to that with the Enhanced Analysis but no significant effect on the phantom. Weight change in vivo produced greater increases of BMC and AREA with the Standard Analysis.

DISCUSSION

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. Acknowledgements
  8. REFERENCES

The observation of an apparent increase of BMD with decreased weight, while BMC decreased, and vice versa, was certainly unexpected. It is explained by concomitant changes of bone area. Such changes are implausible. Cross-sectional studies show some evidence of an increase of bone width with age in men, but not in women.12,13 The changes are small and slow, at about 1%/year, quite insufficient to explain the bidirectional changes of up to 10% that we observed. Although inaccurate, the correlation between ΔAREA and ΔBMC is understandable. Manufacturers do not reveal the assumptions and algorithms that they use to identify bone-containing pixels, but they are likely to include a density thresholding process. As BMC rises, more pixels exceed a given threshold and add to the recognized bone area. Both our phantom and in vivo measurements show that such a correlation exists when there is, in fact, no real change of bone mass. In such cases, the use of BMD gives a result that is closer to the truth than does BMC. However, we cannot conclude that the use of BMD is always more accurate in measuring changes. The phantom measurements in which bone thickness was varied demonstrated some degree of underestimation of BMD, whereas BMC changes were correctly recorded. The demonstration that realistic superficial additions of lard to the limbs could account for the observed changes of AREA and BMC without there being any real bone changes does not prove that bone mass did not change with body weight. We made no attempt to simulate changes in trunk fat. These are likely to be much more complex in distribution, as short-term changes occur more rapidly in the highly heterogeneous visceral fat than in superficial layers. We have shown from CT scans that changes in abdominal fat have little mean effect on spine BMD measurements but increase errors in individual cases.14 We have no idea how total-body bone measurements are affected. Evidence of the complexity of the subject is provided by our finding that, even in the absence of any change of soft tissue in the phantom trunk, the bone measurements there were influenced by lard around the limbs.

Further complexities are illustrated by the results presented in Table 6. Spurious changes of AREA, BMC, and BMD in the limbs were dependent upon position, initial fat proportion, and added superficial lard. Variables not explored in these experiments included limb diameter and bone characteristics.

The anomalies presumably arise from the inadequacy of the fat distribution model used. In X-ray attenuation terms, there are three main body components, bone mineral, lean soft tissue, and fat soft tissue. Because only two photon energies are used, there are too few equations to obtain a complete solution and it is necessary to assume a fat distribution to estimate the proportion of fat overlying bone. This cannot be universally valid and it is apparent that even on average the models used by Hologic are inappropriate. The differences between the results from Enhanced and Standard Analysis illustrate this further. The Standard software calculates a single soft tissue baseline for the whole body, which introduces such anomalies as the BMC of the head appearing to change with the thickness of the trunk.1 The Enhanced software calculates a separate soft tissue background for each region of the body and is recommended especially when there is a large change of weight. The manufacturer does not reveal the details of the soft tissue correction, but our results show that the distribution of fat, as distinct from mere soft tissue thickness, is not adequately considered.

We had previously reported differences between results from Enhanced and Standard software and the dependence ot the ratio on fat proportion in measurements of normal subjects.1 The mean values of the variables were not significantly different, but that probably results from the lack of very lean subjects in the small population. Spector et al.15 also reported mean differences between Enhanced and Standard analysis using a Hologic QDR2000 in the pencil beam mode. Their ratios were smaller than ours, but they do not present the fat proportions in their subjects.

There have been reports of the effects of variations in soft tissue thickness without changes of fat distribution on bone measurements that produced results similar to those presented here. We previously used a smaller, semianthropomorphic phantom, without arms, to examine the variation of BMC and BMD with thickness and uniform fat proportion, using three different manufacturer's instruments.1 The Hologic QDR 1000W exhibited little variation of BMC or BMD with soft tissue thickness between 10 and 25 cm. Fat proportion also had little effect, although there was a statistically significant increase of BMC with fat proportion between 5 and 60% when analyzed with Standard software.

Using a Hologic QDR 1000W and a tank containing mixtures of oil and water, Jebb et al. found no effect on the BMD of an aluminium plate of thickness differences below a total phantom thickness of 25 cm and a fat proportion greater than 5%.3 There do not appear to have been any previous reports of the use of the more realistic addition of annular fat.

There have been two reports of total-body bone changes measured with a Hologic QDR 1000W during weight change. Jensen et al. studied 49 obese women and 2 men before and after 15 weeks of a low-calorie diet.4 They found a mean loss of 12.7% of weight, 27% of fat, and 5.9% of BMC. The slope of their %ΔBMC/%ΔW plot would therefore be 0.46, lower than the 0.70 from Standard software in our group of 20 subjects analyzed by both protocols, but higher than the 0.32 for Enhanced in this group and 0.27 in the whole population. Jensen et al. used software version 5.11, without indicating whether it was Enhanced. We believe that it was not because they report an unlikely proportional loss of head BMC equal to that from the rest of the body. We had found that using Standard software led to spurious changes of head BMC as soft tissue thickness over the trunk changed.1 Unfortunately, they did not present any BMD or AREA results, so we cannot judge whether their results demonstrate the anomaly that we found. Jensen et al.4 also placed up to 15 kg of lard flat on the couch beneath a control group of volunteers and found that it had no significant effect on the bone measurements, in line with our own experience.

Pritchard et al. also used a Hologic QDR 1000W to measure bone changes in a group of obese men.16 Eighteen of these dieted and lost a mean of 6.4 kg of body weight, or 7.3%, over a 12-month period. Their total-body BMC fell by 42 g, or 1.4%, and their BMD by 1.5%. The similarity of the proportional losses of BMC and BMD implies a lack of change of AREA and is contrary to our findings. Pritchard et al. used software version 5.47, but do not indicate whether the Enhanced option was employed. One difference between their study and ours was that they measured men whereas our weight-change study was confined to women. However, it is unlikely that the gender difference explains the disparity of our results because the changes in the small number of men in our general patient group agreed with those from the women. We can only surmise that unknown software differences explained the disparity.

In an investigation on the effects of kidney transplantation on bone mass, using a Hologic QDR 1000W with Enhanced software 5.57, Horber et al. showed that total-body BMC and BMD fell significantly, but AREA did not change.17 In the trunk, mean AREA, BMC, and BMD all fell, while in the arms and legs AREA rose by 2.5% and BMC rose by 1.3%, although the latter was not significant. No figures are given of any weight change. The correlations between changes in AREA and BMC in the regional observations provide some evidence of an artefact.

It would be desirable to know whether the observed anomaly is confined to Hologic absorptiometers. We have measured our hardboard phantom on a Lunar DPX and a Norland XR26 while varying the thickness of the aluminium skeleton.18 The relationships between measured BMD and true BMD were similar to those with the QDR1000W, including the minimum in the total-body plot, but there were no thresholds for the limbs. We have not been able to repeat the lard-wrapping experiment with these instruments.

Evidence from published total-body measurements on people is scant. Svendsen et al. used a Lunar DPX to measure bone mineral in 118 overweight women as they dieted for 12 weeks.19 The mean weight loss was 10 kg, but there was no significant change of total-body BMD. BMC was not reported. Compston et al. used a Lunar DPX to study 12 obese women before and after a 10-week period of severe dieting.20 The software used is not quoted. Body weight fell by 16.7%, BMC by 4.6% and BMD by 2.5%, implying some fall in AREA. Ramsdale and Bassey imposed a less severe dietary regime on 45 women and achieved a mean weight loss of 4.9%.21 A Lunar DPX-L, software 1.3, showed a BMC loss of 0.8% and BMD loss of 0.7%, so there is no evidence of our anomaly, although the changes were small. There have been very few reports of the use of Norland DXA equipment in longitudinal studies of total-body bone mineral and no evidence regarding anomalies is available.

We are not aware of any previous reports of the anomaly in whole-body DXA that we have described. A similar anomaly in DXA of the spine has been reported by Peel and Eastell.8 The results of our spine measurements agree with theirs, revealing a highly significant relationship between changes in AREA and BMC, although with a much lower slope than with whole-body measurements. Although less dramatic, the resulting underestimation of BMD changes is serious, in view of the much greater use of spine scanning.

It must be remembered that there are conditions when there is a genuine concordance between AREA and BMC, for example during growth. We limited our subject selection to those over 17 years to reduce this possibility. In addition, there were no height changes greater than 1 cm.

We cannot tell whether bone changes in the absence of weight change are recorded accurately. The universal correlation between ΔAREA and ΔBMC means that ΔBMD underestimates a change if ΔBMC is correct, but that ΔBMD would be more accurate if the changes were spurious, due to soft tissue changes or counting statistics. The phantom measurements demonstrated that both ΔBMC and ΔBMD could be wrongly recorded, but it would need a phantom approaching the complexity and variability of the human body to reach a more definitive conclusion. We can conclude, however, that measurements by the Hologic QDR 1000W of the effect of weight change on total body bone mineral have little validity. It is unknown whether similar anomalies are introduced by fat distribution changes with other manufacturers' DXA instruments, although we have some evidence of correlations between AREA and BMC. Changes in total-body BMC or BMD, especially when there is significant weight change, should be interpreted with caution, whichever DXA scanner is used.

It may be that modification of the fat distribution model could reduce the anomalies. The differences between the results using Enhanced and Standard software demonstrate the influence of this factor. However, the model can only be an approximation and there will always be uncertainties, due to the differences between the individual anatomy and the model.

The anomalies demonstrated here have the most dramatic effect on longitudinal studies, but it is obvious that the absolute accuracy of the bone measurements is suspect. This would affect comparisons with reference ranges and with results from other instruments. It would also limit attempts to use DXA measurements of bone mineral in multicompartment models designed to assess the fat proportion of the soft tissue22 and to correct underwater weighing measurements for the variable contribution of bone to lean tissue. We have previously shown that estimates of this contribution measured by DXA equipment from different manufacturers differ substantially.2,23 It now appears that errors are introduced by a single instrument due to differences of fat distribution.

Acknowledgements

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. Acknowledgements
  8. REFERENCES

We are grateful to Dr. R. Wrate for stimulating our interest in the anomaly, to Maisie Gard and Karen McPhail for organizing and conducting the clinical aspects of the study, and to Carol Millar for technical assistance. We should also like to thank the anonymous referees whose detailed criticism of an earlier version of the paper led to a reappraisal of our results, further experimentation, and enlightenment.

REFERENCES

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. Acknowledgements
  8. REFERENCES
  • 1
    Tothill P, Avenell A, Reid DM 1994 Precision and accuracy of measurements of whole-body bone mineral: Comparisons between Hologic, Lunar and Norland dual-energy X-ray absorptiometers Br J Radiol 67:12101217.
  • 2
    Tothill P, Avenell A, Love J, Reid DM 1994 Comparisons between Hologic, Lunar and Norland dual-energy X-ray absorptiometers and other techniques used for whole-body soft tissue measurements Eur J Clin Nutr 48:781794.
  • 3
    Jebb SA, Goldberg GR, Jennings G, Elia M 1995 Dual-energy X-ray absorptiometry measurements of body composition: Effects of depth and tissue thickness, including comparisons with direct analysis Clin Sci 88:319324.
  • 4
    Jensen BJ, Quaade F, Sorensen OH 1994 Bone loss accompanying voluntary weight loss in obese humans J Bone Miner Res 9:459463.
  • 5
    Mazess RB, Bisek J, Trempe J 1992 Effect of tissue thickness on fat content and bone using DEXA Bone 13:280.
  • 6
    Svendsen OL, Haarbo J, Hassager C, Christiansen C 1993 Accuracy of measurements of body composition by dual-energy x-ray absorptiometry in vivo Am J Clin Nutr 57:605608.
  • 7
    Laskey MA, Lyttle KD, Flaxman ME, Barber RW 1992 The influence of tissue depth and composition on the performance of the Lunar dual-energy X-ray absorptiometer whole-body scanning mode Eur J Clin Nutr 46:3945.
  • 8
    Peel NFA, Eastell R 1995 Comparison of rates of bone loss from the spine measured using two manufacturers' densitometers J Bone Miner Res 10:17961801.
  • 9
    Bush F 1946 Energy absorption in radium therapy Br J Radiol 19:1421.
  • 10
    Lehman LA, Alvarez RE, Macovski A, Brody WR, Pelc NJ, Riederer SJ, Hall AL 1981 Generalized image combinations in dual kVp digital radiography Med Phys 8:659667.
  • 11
    Nord RH 1991 Soft tissue composition phantom for DXA Osteoporos Int 1:203.
  • 12
    Ruff CB, Hayes WC 1988 Sex differences in age-related remodelling of the femur and tibia J Orthop Res 6:886896.
  • 13
    Mosekilde L, Mosekilde L 1990 Sex differences in age-related changes in vertebral body size, density and biomechanical competence in normal individuals Bone 11:6773.
  • 14
    Tothill P, Avenell A 1994 Errors in dual-energy X-ray absorptiometry of the lumbar spine owing to fat distribution and soft tissue thickness during weight change Br J Radiol 67:7175.
  • 15
    Spector E, LeBlanc A, Shackelford L 1995 Hologic QDR 2000 whole-body scans: A comparison of three combinations of scan modes and analysis software Osteoporos Int 5:440445.
  • 16
    Pritchard JE, Nowson CA, Wark JD 1996 Bone loss accompanying diet-induced or exercise-induced weight loss: A randomised controlled study Int J Obesity 20:513520.
  • 17
    Horber FF, Casez JP, Steiger U, Czerniak A, Montandon A, Jaeger P 1994 Changes in bone mass early after kidney transplantation J Bone Miner Res 9:19.
  • 18
    Tothill P, Cowen S, Hannan WJ, Wrate RM, Freeman CP, 1996 Anomalies in whole-body measurements using DXA. In: RingEFJ, ElvinsDM, BhallaAK, (eds.) Current Research in Osteoporosis and Bone Mineral Measurement IV. British Institute of Radiology, London, U.K., pp. 8586.
  • 19
    Svendsen OL, Hassager C, Christiansen 1993 Effect of an energy-restrictive diet, with or without exercise, on lean tissue mass, resting metabolic rate, cardiovascular risk factor and bone in overweight women Am J Med 95:131140.
  • 20
    Compston JE, Laskey MA, Croucher PT, Coxon A, Kreitzman S 1992 Effect of diet-induced weight loss on total body bone mass Clin Sci 82:429432.
  • 21
    Ramsdale SJ, Bassey EJ 1994 Change in bone mineral density associatied with dietary-induced loss of body mass in young women Clin Sci 87:343348.
  • 22
    Friedl KE, DeLuca JP, Marchitelli LJ, Vogel JA 1992 Reliability of body-fat estimations from a four-compartment model by using density, body water and bone mineral measurements Am J Clin Nutr 55:764770.
  • 23
    Tothill P, Han TS, Avenell A, McNeill G, Reid DM 1996 Comparisons between fat measurements by dual-energy X-ray absorptiometry, underwater weighing and magnetic resonance imaging in healthy women Eur J Clin Nutr 50:742752.