Estimation of whole body fat from appendicular soft tissue from peripheral quantitative computed tomography in adolescent girls

Abstract

Objective

Assess the utility of peripheral quantitative computed tomography (pQCT) for estimating whole body fat in adolescent girls.

Research Methods and Procedures

Our sample included 458 girls (aged 10.7 ± 1.1y, mean BMI = 18.5 ± 3.3 kg/m²) who had DXA scans for whole body percent fat (DXA %Fat). Soft tissue analysis of pQCT scans provided thigh and calf subcutaneous percent fat and thigh and calf muscle density (muscle fat content surrogates). Anthropometric variables included weight, height and BMI. Indices of maturity included age and maturity offset. The total sample was split into validation (VS; n = 304) and cross-validation (CS; n = 154) samples. Linear regression was used to develop prediction equations for estimating DXA %Fat from anthropometric variables and pQCT-derived soft tissue components in VS and the best prediction equation was applied to CS.

Results

Thigh and calf SFA %Fat were positively correlated with DXA %Fat (r = 0.84 to 0.85; p <0.001) and thigh and calf muscle densities were inversely related to DXA %Fat (r = −0.30 to −0.44; p < 0.001). The best equation for estimating %Fat included thigh and calf SFA %Fat and thigh and calf muscle density (adj. R² = 0.90; SEE = 2.7%). Bland-Altman analysis in CS showed accurate estimates of percent fat (adj. R² = 0.89; SEE = 2.7%) with no bias.

Discussion

Peripheral QCT derived indices of adiposity can be used to accurately estimate whole body percent fat in adolescent girls.

Introduction

Peripheral quantitative computed tomography (pQCT) is a non-invasive technique for accurately and precisely assessing volumetric bone parameters using low dose radiation in humans [1–3]. Its increased availability in recent years has led to more widespread use and advancements in its application in research to examine a variety of bone related conditions, particularly in children. Early work examined growth stages and created reference ranges for bone in healthy children [4–7]. Other studies implemented pQCT to test the effects of interventions during different bone growth stages [8–15]. More recent research has focused on relating soft tissue properties to bone parameters [16–21]. Although pQCT is capable of regional soft tissue analysis, investigators usually employ other techniques to determine total body soft tissue composition, frequently dual x-ray absorptiometry (DXA), often using both DXA and pQCT in the same investigation.

In recent years, there has been an increase in the use of pQCT for regional soft tissue assessment deriving indices of adiposity and lean soft tissue. Consequently, establishment of the relationship between pQCT derived adiposity of the limbs and the recognized standard of total body fat percentage from DXA would make it possible to reduce the need for exposure to both DXA and pQCT in pediatric subjects. To our knowledge, a study by Ducher, et al. is the only study to examine the relationship between pQCT-derived adiposity indices and whole body percent fat measured by DXA in children [22]. Regional estimates of percent fat of forearm and tibia, derived using tissue areas, were used to estimate whole body percent fat using DXA as the criterion measurement. Results were promising, with sex-specific regression equations accounting for 83.7% and 87.7% of the variance in percent fat in girls and boys, respectively, with standard errors of estimates less than 3%. Given that adiposity in the extremities increases with maturation [23] and the strong correlation of thigh circumference to total body adipose tissue volume in females [24] we hypothesized that incorporating this region in a regression equation to estimate percent fat would provide as good and possibly better prediction than using the calf or the forearm as Ducher et al did [22]. Thus, the purpose of the present study was to assess the accuracy of estimating DXA derived whole body fat from pQCT indices of adiposity of the thigh and calf in young girls.

Methods

Subjects

The sample consisted of 458 healthy pre-pubertal and early pubertal girls aged 8–13 years old who were participants in the “Jump In: Building Better Bones” study. Jump-In was a 5-year longitudinal study investigating the effects of a high impact jumping exercise intervention on bone development in adolescent girls. The study was approved by the University of Arizona Human Subjects Protection Committee and the study protocol followed regulations put forth in the Helsinki Declaration. Written, informed consent was obtained from the parents/guardians of all volunteers and child written assent was also obtained. Baseline measurements were used in all analyses reported herein.

Anthropometry

Standing height and sitting height were measured to the nearest millimeter, after a full inhalation, using a stadiometer (Shorr Height Measuring Board, Olney, MD). Body mass was measured to the nearest 0.1 kg using a calibrated scale (Seca, Model 881, Hamburg, Germany). Body mass index (BMI) was calculated from height and weight measurements. Non-dominant femur and tibia lengths were measured to the nearest millimeter using an anthropometric tape. Tibia length was measured from the proximal end of the medial border of the tibial plateau to the distal edge of the medial malleolus. Femur length was measured from the superior aspect of the patella to the inguinal crease. Each measurement was performed twice and averaged to obtain the criterion measurement.

Physical Development and Maturity

Self-reported date of menarche and Tanner stage were obtained using a validated questionnaire that has been shown to agree with physician exam and grading of physical development of children [25, 26]. Maturation was also estimated from the maturity offset algorithm developed by Mirwald et al. which is based on estimated years from peak height velocity (PHV) [27]. The algorithm, derived from a six-year longitudinal study in boys and girls (4), explained 89% of the variance in years from PHV [27].

DXA soft tissue assessment

Total fat mass, lean soft tissue mass and total body mass measurements were obtained from whole body DXA scans using a GE Lunar Prodigy fan beam densitometer (GE Lunar Corp. Madison, WI, USA). Dual energy x-ray absorptiometry is an accepted criterion measure for bone and soft tissue body composition [28]. All scans were performed following the GE/Lunar standard manufacturer protocols (GE Lunar Prodigy-software version 5.60.003) [18]. Percent body fat was calculated as the ratio of total body fat mass to total body mass. All DXA scans were performed by a licensed, certified operator and the scanner was calibrated daily for precision and accuracy. DXA precision and coefficients of variation in our laboratory have been reported previously [29].

pQCT regional soft tissue assessment

Peripheral QCT scans on the non-dominant femur and tibia were obtained using the Stratec XCT3000 (STRATEC Medizintechnik GmbH, Pforzheim, Germany, Division of Orthometrix; White Plains, NY, USA) at the 20% femur and 66% tibia sites, located based on the limb lengths and associated distal growth plates. Using a scout view of the respective growth plates of the distal femur and tibia, reference lines were placed at the distal growth plates, and the scanner was programmed to find the sites of interest based on corresponding bone lengths. Scanning parameters were as follows: scan speed of 25 millimeters per second, voxel size of 0.4 millimeters, and 2.3 millimeter slice thickness. All pQCT scans were analyzed using Stratec software, Version 6.0, and operators were trained for pQCT data acquisition and analyses following guidelines provided by Bone Diagnostics, Inc. (Fort Atkinson, WI).

Regional soft tissue composition was assessed using algorithms based on a four-part analysis developed by Bone Diagnostics, Inc. that used edge detection and threshold techniques to separate skin, adipose, muscle, and bone based on attenuation characteristics that are directly related to tissue composition and density [30, 31]. Images were filtered prior to being analyzed. Thigh and calf skin area, marrow area, subcutaneous fat area (SFA), muscle area (which includes muscle fat content), and total bone area were measured at each site respectively and summed to create thigh and calf total tissue areas. Thigh and calf subcutaneous percent fat (Thigh SFA %Fat and Calf SFA %Fat) values were subsequently calculated by dividing their respective subcutaneous fat areas by total tissue area and multiplying by 100. Thigh and calf muscle densities were also measured. Detailed descriptions of the pQCT device used in the Jump-In Study and our protocol for in-vivo image processing and analysis protocol have been previously published [18, 32].

Statistical Analysis

The large sample size allowed for validation and cross-validation within the sample using a two-thirds (validation)/one-third (cross-validation) approach. Within the total sample (TS), 4^th and 6^th grade girls were randomized by grade into two-thirds/one-third groups and then combined so that the validation sample (VS; n = 304) and cross-validation sample (CS; n = 154) had equal numbers of 4^th and 6^th grade girls.

The Statistical Package for the Social Sciences for Windows, Version 19.0 (SPSS, Chicago, IL, USA) was used for all analyses. Descriptive statistics were calculated for the independent and dependent variables of interest. Bivariate relationships were estimated using Pearson’s product moment correlation coefficients. Multiple linear regression analyses were performed in the VS to develop prediction equations for estimating DXA-derived total body percent fat (DXA %Fat) using anthropometric variables and pQCT-derived soft tissue components. An alpha level of 0.05 was used for all statistical analyses. Unstandardized coefficients, adjusted multiple correlation coefficients and standard errors of the estimate (SEE) are reported for all regression models. Since BMI includes the measures of weight and height (kg/m²), weight and height were not used individually as covariates. Baseline age and maturity offset were used as maturity related covariates separately in subsequent regressions. Thigh and calf SFA %Fat were added to the models separately and together followed by thigh and calf muscle density. The final regression model included all anthropometric and soft tissue independent variables.

The agreement between percent fat measured by DXA and percent fat estimated by the prediction equation developed in VS, was assessed in CS using a Bland-Altman analysis in which the mean of DXA %Fat and the predicted percent fat was plotted against the residuals from the prediction equation.

Results

The physical characteristics of the TS, VS and CS are shown in Table 1. Mean BMI for TS was approximately equal to the age and gender specific 56^th percentile relative to U.S. national norms based on BMI growth curves from the NHANES survey [33]. Less than 4% of the subjects were between 0 and 5^th BMI percentiles, 74% of the subjects were between the 5^th and 85^th BMI percentiles, 16% were between the 85^th and 95^th BMI percentiles, and 7% were in the ≥95^th BMI percentile. An independent t-test showed that there were no differences in any physical parameter between the VS and CS (data not shown).

Table 1.

Characteristics of various study samples

	Total Sample (n = 458)		Validation Samplea (n = 304)		Cross-Validationa Sample (n=154)
Variable	Mean	SD	Mean	SD	Mean	SD
Baseline Age (years)	10.7	1.1	10.6	1.1	10.8	1.0
Baseline maturity offset (years)	−1.1	1.0	−1.2	1.0	−1.0	1.0
Baseline height (cm)	144.4	9.8	144.0	9.5	145.2	10.3
Baseline weight (kg)	39.0	10.1	38.7	10.2	39.6	10.1
Baseline BMI (kg/m2)	18.5	3.3	18.4	3.3	18.5	3.1
Baseline BMI Percentile	55.9	29.7	55.9	30.0	56.0	29.0
Baseline DXA %Fat	27.6	8.4	27.7	8.5	27.5	8.4
Baseline Calf SFA* %Fat	23.4	7.8	23.5	7.8	23.3	7.8
Baseline Calf Muscle Density (mg/cm3)	78.9	1.2	78.9	1.2	79.0	1.2
Baseline Thigh SFA* %Fat	31.2	9.3	31.0	9.3	31.5	9.5
Baseline Thigh Muscle Density (mg/cm3)	76.3	1.5	76.3	1.5	76.2	1.5

^aNo significant (p ≤ 0.05) differences between the validation and cross-validation samples;

^bSFA = Subcutaneous Fat Area

In the total sample, all covariates except age were significantly correlated (Pearson correlation coefficients) with DXA %Fat. Maturity offset, height, weight, and BMI were all significantly correlated with DXA %Fat (r = 0.30, .021, 0.69, 0.86 respectively, all p < 0.001). Thigh SFA %Fat and Calf SFA %fat were also significantly correlated with DXA %Fat (r = 0.84 and r = 0.83, respectively, p < 0.001). Thigh and calf muscle densities, surrogates for muscle fat content, were significantly and inversely (as expected) correlated with DXA %Fat, as expected (r = −.48 and r = −.35 respectively, p < 0.001).

Estimation of DXA %Fat from BMI was considered the reference model (Figure 1). Results of the validation analysis are shown in Table 2. Age was added to the model as a maturity-related variable and although it was not significantly correlated with DXA %Fat (r = 0.09; p > 0.05) in the TS, it was significant (p < 0.001) when added to the reference model, although it accounted for only a slight increase in adjusted R-square and a small decrease in SEE (adj. R-square = 0.76; SEE = 4.1%; Table 2). Calf SFA %Fat accounted for an additional 12% of the variance in whole body percent fat and the SEE was reduced from 4.1% to 3.0% when added to BMI and age. Replacing calf SFA %Fat with thigh SFA %Fat accounted for slightly less variance (adj. R-square = 0.87) and had slightly greater error (SEE = 3.1%) when compared to calf SFA %Fat (Table 2). Entry of calf and thigh SFA %Fat together in the model accounted for an additional 2.0–3.0% of the variance in percent fat and further reduced SEE by 0.2–0.3% when compared to calf or thigh alone (Table 2).

Figure 1.

DXA percent fat predicted from body mass index (kg/m²) in the total sample (n = 458).

Table 2.

Regression equations^a for estimating DXA whole body percent fat from pQCT and anthropometry in the validation sample (n = 304). Upper half of table controls for age; lower half controls for maturity offset.

Unstandardized Regression Coefficients
	BMI	Age	Maturity Offset	Calf SFAb %Fat	Calf Musc. Dens.	Thigh SFAb %Fat	Thigh Musc. Dens.	Constant	Adj. R2	SEE (%)
Age	2.284*	−0.993*	---	---	---	---	---	−3.9	0.764	4.13
	1.372*	0.113	---	0.528*	---	---	---	−11.231	0.879	2.95
	1.400*	−0.190	---	---	---	0.423*	---	−9.257	0.865	3.11
	1.256*	0.128	---	0.355*	---	0.198*	---	−11.343	0.889	2.83
	1.363*	0.153	---	0.518*	−0.268	---	---	9.891	0.880	2.94
	1.334*	−0.011	---	---	---	0.397*	−0.756*	48.534	0.880	2.94
	1.215*	0.237	---	0.332*	0.313	0.189*	−0.812*	26.303	0.900	2.68
Maturity Offset	2.377*	---	−1.158*	---	---	---	---	−17.507	0.762	4.13
	1.347*	---	0.188	0.532*	---	---	---	−9.433	0.879	2.95
	1.392*	---	−0.105	---	---	0.427*	---	−11.365	0.865	3.12
	1.219*	---	0.246	0.359*	---	0.200*	---	−9.167	0.889	2.82
	1.332*	---	0.243	0.522*	−0.277	---	---	12.965	0.880	2.94
	1.302*	---	0.122	---	---	0.402*	−0.770*	50.095	0.880	2.93
	1.163*	---	0.386**	0.335*	0.302	0.193*	−0.817*	31.354	0.900	2.67

^aUstandardized regression coefficients;

^bSFA = Subcutaneous Fat Area;

^*p < 0.001;

^**p < 0.05

The addition of thigh and calf muscle density to this regression model, both separately and together, improved prediction of percent fat with the greatest improvement observed when both density variables were included. This model had the best predictive accuracy (SEE = 2.7%), accounting for the greatest proportion of variance in percent fat (adj. R-square = 0.90), although, calf muscle density was not significant (p = 0.14) in this model (Table 2). A scatterplot showing predicted percent fat versus DXA %Fat is shown in Figure 2a. The plot of residuals against the average percent fat is shown in Figure 2b. The limits of agreement (± 2SD, as defined by Bland-Altman) were ± 5.4% [34, 35]. Residuals were not correlated with the average percent fat (r = 0.16 and b = 0.05) and the average residual, as expected, was zero (X̄_residual = 0.02%).

Figure 2.

Figure 2a: DXA%Fat predicted by the final regression model in the validation sample (n = 304)

Figure 2b: Bland-Altman plot of average percent fat vs residuals in validation sample

Maturity offset, as a more direct estimate of maturation, was tested in place of age in the regression models. Maturity offset was moderately correlated with DXA %Fat (r = 0.21; p < 0.001), and, similar to age, was significant in only one regression model (p < 0.001). Regression models with maturity offset as the maturity-related variable gave essentially identical results to models using age (Table 2).

Cross-validation was accomplished by applying the prediction equation that produced the best predictive accuracy in the validation sample (among the models using age as the maturity variable) to the cross-validation sample (CS, n=154) and comparing predicted percent fat against DXA %Fat. The predicted percent fat accounted for 89% of the variance in DXA %Fat with a SEE of 2.7% (Figure 3a). Mean difference between measured and predicted percent fat was −0.3%, and the correlation and slope for the regression of the residuals on average percent fat were r = 0.14 and b = 0.08, respectively (Figure 3b).

Figure 3.

Figure 3a: DXA percent fat versus predicted percent fat in the cross-validation sample (n = 154)

Figure 3b: Bland-Altman plot of average percent fat vs residuals in cross-validation sample

Discussion

Peripheral QCT indices of soft tissue composition can also be obtained in studies using pQCT to estimate limb bone parameters. These indices may be used to estimate whole body percent fat thereby eliminating the need for a second procedure, such as dual energy x-ray absorptiometry, which is often paired with pQCT as an adjunct for measuring soft tissue composition. Our results showed that pQCT estimates of calf and thigh percent fat and muscle density, when combined with anthropometry, give accurate estimates of whole body percent fat.

Ducher et al [22] performed the only study in which pQCT indices of adiposity were used to predict whole body percent fat measured by DXA in an adolescent population of boys and girls. Our study agrees with the findings of Ducher et al and support using pQCT derived soft tissue indices to predict whole body percent fat [22]. However, unlike our study, Ducher et al performed pQCT measurements on the forearm and calf and used subject height in their regression model to predict whole body percent fat with very acceptable adjusted R-square and standard error values (adj. R² = 0.83, SEE = 2.6% in adolescent girls) [22]. In our sample, height was weakly (as expected), though significantly associated with DXA %Fat (r = 0.21, p < 0.001) and using height as predictor, as Ducher et al [22] did, accounted for less variability and resulted in higher SEE’s for all models when compared to Ducher’s results (Adj. R² = 0.75 to 0.81, SEE = 3.7% to 4.2%; data not shown)[36]. Weight had a much stronger association (r = 0.69, p <0.001) with DXA %Fat and using weight in place of height improved prediction results (adj. R² = 0.84 to 0.88, SEE = 3.0% to 3.4%; data not shown) in all subsequent regression models.

Studies have shown Body Mass Index is significantly and at least moderately correlated with percent fat in children and youth [37]. In this study, body mass index had the strongest association of all anthropometric measures with whole body DXA %Fat (r = 0.86, p < 0.001), as expected. The results from using BMI alone to predict whole body percent fat in our sample were not different than those of a validation study by Ellis et al [38] with similar adjusted R-squares (0.70 vs. 0.74 – Ellis et al vs. our study) and similar standard errors (4.7% vs 4.3% – Ellis et al vs. our study) (see Figure 1). We used BMI as the reference model to assess the ability of pQCT estimates of appendicular soft tissue to improve upon prediction of percent fat from BMI alone.

Previous studies have shown that whole body fat and fat distribution vary with maturation [23, 39, 40], and described important changes occurring during the transition from the pre-pubertal to pubertal stages [23]. We controlled for maturity by using maturity offset, which reflects years from peak height velocity, as an index of maturity. Recognizing that many studies may not have available the anthropometric measures needed to estimate maturity offset, we also tested age in our models. Although age was not significantly correlated with DXA %Fat, it was significant in the regression model when paired with BMI. Similarly, Huang et al showed that age accounted for 5.4% more of the variance in total body fat when included with weight in the regression model for predicting DXA derived total body fat in an adolescent population of boys and girls[41]. In subsequent regressions in our validation sample, with other covariates included, age was not significant and made very minor contributions to the model, possibly due to the narrow age range of our sample (9–13 years). Further analysis showed that age could be removed from these models without losing prediction accuracy (data not shown). In contrast to age, maturity offset was moderately correlated with DXA %Fat; however, using maturity offset in the analysis did not produce significantly better results than those obtained with age (Table 2). Furthermore, like age, maturity offset did not significantly contribute to the prediction in regressions when other covariates were added (Table 2).

We hypothesized that thigh percent fat based on the larger tissue area of the thigh scans would provide a better prediction of whole body percent fat than using calf. In support of this notion, Heymsfield et al [24] showed thigh circumference was highly correlated with total body adipose tissue volume (R = 0.80). Additionally, Goulding et al [39] found that changes in regional fat distribution occur with advancing maturity, with girls at Tanner Stage 1 having lower trunk to leg fat than girls at later Tanner stages indicating proportionately greater leg fat. However, in our sample, bivariate correlations between Thigh SFA %Fat and DXA %Fat and Calf SFA %Fat and DXA %Fat were similar as was prediction accuracy for DXA %Fat whether calf or thigh were used in the model (adj. R² = 0.87 to 0.88, SEE = 3.0% to 3.1%; Table 2). Prediction was somewhat improved (SEE improved to 2.8%) when both the Thigh and Calf SFA %Fat variables were included in the analysis (p < 0.001 for both; Table 2), although using either thigh or calf SFA %Fat provides prediction of whole body percent fat with acceptable accuracy [36].

In our study, muscle density was moderately and inversely associated with whole body DXA %Fat, as expected. Muscle density was used as a surrogate for muscle fat content; and changes in muscle density reflect changes in muscle fat content, with increased density indicative of decreased muscle fat content and vice versa [18, 31]. The utility of muscle density is supported by the results from Kelley et al who showed that decreased thigh muscle density in overweight and obese subjects, measured by computed tomography, was strongly associated with increased body fat [31]. Consequently, we tested whether muscle density would improve the prediction accuracy when combined with other variables reflecting regional adiposity. In the present study, addition of thigh and calf muscle density to the regression model produced the best regression equation accounting for 90% of the variance in whole body percent fat with a SEE to 2.7% (Table 2). Limits of agreement (± 2SD) were ± 5.4%. Calf muscle density was not significant (p = 0.14) in the final model and further analysis showed that removing it from the model did not change the adjusted R-square or SEE (data not shown). Applying the best regression equation to the CS supported its utility. The estimated percent fat accounted for 89% of the variance in DXA %Fat in the CS with a SEE of 2.7% (compared to 90% and an SEE of 2.7% in the VS).

Radiation exposure is of primary concern when performing pQCT and DXA scans on pre-pubertal and pubertal children. A common practice is to use pQCT to measure bone parameters and DXA to measure body composition in the same study thereby increasing a subject’s effective radiation dose. The effective dose for the pQCT is approximately 1μSv for a single slice and increases in a multiple-slice protocol. The effective dose for a whole body DXA scan is approximately 0.3μSv [42, 43]. In our study, effective radiation dose of the tibia (~4 μSv) and femur (~7 μSv) is negligible when compared to average annual background radiation in Europe and the U.S. of approximately 2.5mSv [44]. Children are more vulnerable to the damaging effects of radiation and it is imperative that radiation exposure be kept as low as possible [43]. Validation and cross-validation of the prediction equation in the present study shows that percent fat can be estimated with acceptable accuracy in young girls using pQCT soft tissue variables, thus reducing radiation exposure by eliminating the need to perform DXA scans to measure percent fat.

A limitation of our study was the limited race/ethnic mix, given that 89% of our sample self-identified as Caucasian. It is important to validate pQCT for soft tissue in other groups. Additionally, our sample was limited to pre-pubertal and pubertal girls, which may also limit the generalizability of the equation. Validation in pre-pubertal and pubertal boys is needed before the equation can be recommended for use in males. Finally, while DXA, our criterion method, is a common reference method, some investigators have questioned its use in children because of the rapid physiological changes occurring during adolescence (e.g. affecting hydration of fat-free mass) and software revision issues that affect reproducibility of results [45, 46]. Nevertheless, DXA has been used extensively in the National Health and Nutrition Examination Survey (NHANES) to develop body composition reference values for children and adults, supporting its use as an acceptable criterion reference method [28].

In conclusion, the results of this study demonstrate that pQCT estimates of soft tissue composition of the thigh and calf, in combination with common anthropometric variables, can be used to provide accurate estimates of whole body percent fat in adolescent girls. Further studies are needed to cross-validate these results and develop equation in other populations. Being able to accurately estimate whole body percent fat using pQCT derived soft tissue indices will expand the use of pQCT and add valuable information about soft tissue composition in bone studies without the added expense and radiation exposure of an additional method such as dual energy-ray absorptiometry.