Table 1.
Characteristics of the study populations.
Table 2.
Pearson partial correlation coefficients of anthropometric variables and weight for the full DXA data set (IG+CV = 341 patients).
Table 3.
Pearson partial correlation coefficients between anthropometric variables and DXA-derived body compartments for the full DXA data set (n = 341).
Figure 1.
Multiple linear regression models based upon DXA measurements obtained from the Index Group predict the corresponding body masses in the Cross Validation group with good accuracy.
A) Bland-Altman analysis of actual versus predicted Total Fat has a very small average bias (i.e., the average of the differences between the two measurements) of 0.281 kg and a standard deviation of 3.44 kg (heavy dashed lines indicate the limits of agreement (mean bias±1.96 its standard deviation): −6.47 to 7.03 kg). B) Predicted Trunk Fat has an average bias of 0.328 (CI: −4.223 to 4.89 kg). C) Predicted Fat Free Mass has a bias of −0.176 kg (CI: −6.49 to 6.14 kg). D) Total Mass (FM + FFM) has an average bias of 0.672 kg (CI: −0.965 to 2.31 kg).
Figure 2.
FM increases faster than FFM as a function of weight.
Relationships between predicted body composition and weight (abscissa) to DXA-determined values (ordinate) for the full DXA population (IG + CV) are linear and show that with increasing mass, TF and FFM increase at similar rates which are less than for the total FM. Predicted weight is the sum of FFM and FM (see text for discussion).
Figure 3.
The relationship between the model estimates of FFM and FM is linear.
A) Body composition based on modeling DXA-derived data of the full DXA population (IG + CV) using anthropometric variables exhibits a linear relationship between FM and FFM (red line). The anthropometric model of Weltmann [13], derived from hydrostatic weighing in a similar obese female population, is similar (black line). In contrast, data obtained by Forbes [23] for a female population using isotopic dilution methodology is non-linear (blue line). B) The FFM/FM ratio approaches unity for body mass greater than 100 kg.
Table 4.
Coefficients of the linear regression equations predicting DXA body composition derived from the Index Group DXA population (n = 241).
Table 5.
Body composition of the Cross Validation group is predicted by the regression equations derived from the Index Group.
Figure 4.
The receiver operating characteristics of a nominal logistical regression using DXA-derived body composition as independent variables to estimate the probability of DM2 or either DM2 or MS in the full DXA population (IG + CV) show that a large percentage of the diseased individuals can be identified without a high cost of a false positive diagnosis.
A) The steep slope of the initial portion of the curve to the inflection point (*) indicates that a high proportion of individuals with DM2 can be identifying using DXA-derived FM and TF. Area under the curve is 0.78 with a standard error of 0.05 (24 subjects with DM2 and 327 without; p<0.0001). The 95% confidence interval is 0.68 to 0.88. [Regression equation: risk of DM2 = 1/(1+e−z), where z = 1.94−0.53(TF)+0.28(FM).] B) Similarly, the ROC for the detection of either DM2 or MS also shows inflection points, consistent with sensitive detection of subjects with disease. The AUC is 0.71±0.03 (103 subjects with disease, 238 without; p<0.0001). The 95% confidence interval for the AUC is 0.64 to 0.77. [Regression equation: k = 2.32−0.24(TF)+0.09(FM)].
Figure 5.
The receiver operating characteristic curve corresponding to a nominal logistical regression using body composition estimated by anthropometric modeling for subjects within the Endocrine Practice group (n = 1153) is consistent with good detection of DM2 or MS.
Similar to the DXA population, the ROCs are characterized by initial steep slopes indicating a high detection rate of true positives, but with a low false positive rate. A) The AUC for DM2 detection alone is 0.80±0.02 (1040 subjects negative, 113 positive; p<0.0001). The 95% confidence interval of the AUC is 0.78 to 0.84. [Regression equation: k = 0.14+0.18(Predicted FM)−0.59(Predicted TF)+0.12(Predicted FFM)]. B) The AUC for detecting DM2 or MS is 0.81±0.01 (821 subjects negative, 332 positive; p<0.0001). The 95% confidence interval is 0.78 to 0.84. [Regression equation: k = 0.45+0.33(Predicted FM)−0.83(Predicted TF)+0.06(Predicted FFM)].