Dual-Energy X-Ray Absorptiometry Prediction of Adipose Tissue Depots in Children and Adolescents

Abstract

Background

The measurement of adipose tissue depots in-vivo requires expensive imaging methods not accessible to most clinicians and researchers. The study aim was to derive mathematical models to predict total adipose tissue (TAT) and sub-depots from total body fat derived from a dual energy x-ray absorptiometry (DXA) scan.

Methods

Models were developed to predict magnetic resonance imaging derived TAT and sub-depots subcutaneous (SAT), visceral (VAT), and intermuscular (IMAT) from DXA total body fat using cross-sectional data (T0) and validated results using 1 (T1) and 2 (T2) year follow-up data. Subjects were 176 multi-ethnic healthy children ages 5 to 17 years at T0. 22 were measured at T1 and T2. TAT was compared to fat.

Results

At T0, TAT was greater than fat (12.5 ± 8.4 vs.12.0 ± 9.4 kg; p< 0.0001), with a quadratic relationship between TAT and fat which varied by sex. Predicted mean TAT’s were not different from measured TAT’s: T1: (9.84±4.45 kg vs. 9.50±4.37 kg; p=0.11) T2: (12.94±6.75 kg vs. 12.89±7.09 kg; p=0.76). The quadratic relationship was not influenced by race or age.

Conclusions

In general, the prediction equations for TAT and sub-depots were consistent with the measured values using T1 and T2 data.

Introduction

Childhood obesity continues to increase at an alarming rate having reached epidemic levels in many countries and it demands attention worldwide (1). Excess weight gain during childhood is one of the most serious global health problems. There is growing evidence that body fat or adipose tissue accumulation, especially visceral or intraabdominal adipose tissue accumulation, correlates with overweight and obesity, and is an indirect marker of metabolic diseases in adulthood. Both the total amount of adipose tissue and its distribution are risk factors for several metabolic disturbances. Excess accumulation of lipids in visceral adipose tissue is related to impaired insulin sensitivity (a pre-cursor to the development of type 2 diabetes) and unfavorable lipid metabolism in children (2–4). Intermuscular adipose tissue is negatively associated with insulin sensitivity in black adolescents (5).

Accordingly, appropriate and accurate techniques to determine body composition in infants and adolescents are needed for a greater comprehension of the mechanisms by which key conditions in infancy regulate and mediate adipose tissue distribution, the characterization of the severity of adiposity or malnutrition, and the evaluation of the efficacy of interventional strategies in the pediatric population. Nutritional status in children can be evaluated by means of simple anthropometric assessment methods. BMI is an index of excess weight relative to height, has a moderately high sensitivity for the general identification of the obese child, but a weak correlation with total body fat due to considerable changes in fat mass that occur during childhood (6). Body weight and BMI are insufficiently sensitive markers for the evaluation of fat distribution in children, as both are influenced by changes in body cell mass during growth (7). Waist circumference is used as a surrogate for abdominal fat mass in adults where the latter is associated with elevated risk for cardiovascular disease and abnormalities of glucose and lipids metabolism (8, 9). However, the more potent visceral adipose tissue may differ in quantity among individuals with similar waist circumference (10).

Compared to adults, children have greater variations in body composition (including fat and lean tissues) due to maturation, growth, puberty and gender, which decrease the accuracy for the estimation of total or abdominal fat using anthropometric assessment methods (11, 12). Reliable assessments for quantification of fat distribution during childhood include computerized tomography and magnetic resonance imaging (MRI). Unfortunately, MRI is expensive, requires sedation in young infants, and with computerized tomography a relatively high radiation exposure is required, that prohibits its use in pediatric studies (13). In several clinical and research trials, dual energy X-ray absorptiometry (DXA) has been used for the estimation of total and regional body composition in children with a relatively rapid scan time, minimal radiation exposure, and good reproducibility (14–16). Nevertheless, DXA does not allow for the separation of subcutaneous from visceral adipose tissue.

Therefore, the need for an accurate, safe, and a cost-effective approach for the assessment of adipose tissue distribution exists. During obesity treatment programs in children and adolescents, there is increasing interest in monitoring changes in body composition. Since DXA is the most commonly used body composition measurement method, the proposed adipose tissue distribution models could be useful in research and clinical setting where the need exists to estimate the size of the VAT or IMAT depots with respect to metabolic risk assessment in children and adolescents.

The aim of our investigation in this pediatric cohort was to derive models to predict MRI-derived total adipose tissue and sub-depots (subcutaneous, visceral and intramuscular adipose tissue) from DXA total body fat using cross-sectional data and to validate findings by means of longitudinal data.

Results

Subject characteristics

Percentile of body mass index (BMI) was calculated for each child during the investigation period.

Cross-sectional

The baseline (T0) study group included 176 healthy children and young adults ages 5 to 17 years. Participant’s characteristics are presented in Table 1. At baseline, Tanner classification was available on 77% (n=135) of participants; 47% of the children were at stage 1, 10% stage 2, 9% stage 3, 8% stage 4 and 26% stage 5, respectively.

Table 1.

Characteristic of the subjects by gender

	Follow-up points
	Baseline (T0)		1-year (T1)		2-year (T2)
	Girls (n=73)	Boys (n=103)	Girls (n=8)	Boys (n=14)	Girls (n=5)	Boys (n=17)
Age (years) (Range)	10.6 ± 3.3 (6–17)	11.2 ± 3.4 (5–17)	8.6 ± 1.2 (7–11)	9.4 ± 1.4 (7–11)	11.0 ± 1.0 (10–12)	10.6 ± 1.2 (9–12)
Weight (kg)	43.6 ± 18.7	47.6 ± 20.2	35.7 ± 11.4	34.2 ± 7.9	44.8 ± 14.2	43.0 ± 12.3
Height (cm)	127.8 ± 9.1	129.1 ± 11.1	137.9 ± 9.7	137.2 ± 8.9	147.9 ± 11.7	144.8 ± 10.1
BMI Perc.	64.5 ± 28.5	63.1 ± 28.4	56.4 ± 34.7	61.6 ± 23.5	72.9 ± 30.2	66.5 ± 29.7
Tanner stages (n,%)	1: 29, 46.0%	1: 35, 48.6%	1: 5, 62.5%	1: 11, 78.6%	1: 2, 50.0%	1: 10, 62.5%
	2: 6, 9.5%	2: 8, 11.1%	2: 2, 25.0%	2: 2, 14.3%	2: 1, 25.0%	2: 2, 12.5%
	3: 5, 7.9%	3: 8, 11.1%	3: 1, 12.5%	3: 1, 7.1%	3: 1, 25.0%	3: 4, 25.0%
	4: 5, 7.9%	4: 6, 8.3%	4: 0, 0%	4: 0, 0%	4: 0, 0%	4: 0, 0%
	5: 18, 28.6%	5: 15, 20.8%	5: 0, 0%	5: 0, 0%	5: 0, 0%	5: 0, 0%
Ethnicity (n)
Caucasian	9	17	1	2	0	2
Black	32	37	3	2	1	2
Hispanic	20	32	1	8	1	10
Asian	6	7	2	2	3	2
Others	6	10	1	0	0	0

BMI Perc.: body mass index percentile.

Longitudinal

Twenty-nine participants with T0 data were measured longitudinally; 15 with measures at T0, T2, and T2; 7 with measures at T0 and T1, and 7 with measures at T0 and T2. At T1, 73% of the 22 children were at Tanner stage 1, and at T2, 60% of the 20 children with Tanner stage data were at stage 1.

Separate analyses were conducted on the total cohort and on the subgroup where Tanner stage was available. Tanner stage as an independent variable in the models did not contribute significantly to the prediction equations for the subgroup; therefore, the prediction equations based on the total group are presented.

Adipose tissue and fat

Mean values for TAT, sub-depots (SAT, VAT, IMAT) and for Fat_DXA are presented in Table 2. At baseline, TAT was greater than Fat_DXA (12.5 ± 8.4 vs.12.0 ± 9.4 kg; p< 0.0001), with a quadratic relationship between TAT and fat (TAT=a0+a1*fat+a2*fat²; r²_PRESS=0.99, SEE_PRESS=1.01 kg), which varied by sex (Figure 1): equation coefficients (a0, a1, a2) for females are (2.191, 0.776, 0.003) and for males are (2.641, 0.792, 0.002) as presented in Table 3. Predicted mean TAT’s were not different from measured TAT’s: T1: (9.84 ± 4.45 kg vs. 9.50 ± 4.37 kg; p=0.11). Descriptive statistics for the difference, predicted minus measured, are mean (0.34 kg), standard deviation (0.97 kg), and interquartile range (1.28 kg). T2: (12.94 ± 6.75 kg vs. 12.89 ± 7.09 kg; p=0.76). Descriptive statistics for the difference, predicted minus measured, are mean (0.06 kg), standard deviation (0.83 kg), and interquartile range (0.93 kg). The quadratic relationship was not influenced by race or age.

Table 2.

Adipose tissue and sub-depots by MRI and fat by DXA

	Follow-up points
	Baseline (T0)		1-year (T1)		2-year (T2)
	Girls (n=73)	Boys (n=103)	Girls (n=8)	Boys (n=14)	Girls (n=5)	Boys (n=17)
FatDXA (kg)	13.7 ± 10.2	10.8 ± 8.6	10.1 ± 6.0	8.4 ± 5.0	17.0 ± 8.6	11.2 ± 7.4
TAT (kg)	13.8 ± 9.3	11.5 ± 7.5	10.3 ± 4.5	9.0 ± 4.4	16.7 ± 7.9	11.8 ± 6.7
SAT (kg)	12.8 ± 8.7	10.5 ± 6.7	9.7 ± 4.3	8.2 ± 3.9	15.5 ± 7.5	10.8 ± 6.0
VAT (kg)	0.46 ± 0.39	0.51 ± 0.42	0.35 ± 0.19	0.47 ± 0.34	0.64 ± 0.50	0.59 ± 0.56
IMAT (kg)	0.51 ± 0.45	0.48 ± 0.42	0.28 ± 0.16	0.30 ± 0.17	0.47 ± 0.22	0.37 ± 0.24

TAT: MRI total adipose tissue; Fat_DXA: total body fat from dual-energy X-ray absorptiometry (DXA); SAT: subcutaneous adipose tissue; VAT: visceral adipose tissue; IMAT: intermuscular adipose tissue. Results are given as mean ± SD.

Figure 1.

Relations (quadratic) between fat by dual-energy X-ray absorptiometry (DXA) and total dipose tissue (TAT) by magnetic resonance imaging in males (A) and females (B):

Males: TAT (kg) = 2.64+0.79*fat(kg)+0.002*fat(kg)² (R²_PRESS=0.99, p<0.0001; standard estimation error_PRESS = 1.01 kg, n=103).

Females: TAT (kg) = 2.19+0.78*fat(kg)+0.003*fat(kg)² (R²_PRESS=0.99, p<0.0001; standard estimation error_PRESS = 1.01 kg, n=73).

Table 3.

Prediction models for MRI derived adipose tissue depots

Models and variables	a0	a1	a2	a3	r2PRESS	SEEPRESS
TAT=a0+a1*fat+a2*fat2					0.99	1.01
• Females	2.19	0.78	0.003
• Males	2.64	0.79	0.002
SAT=a0+a1*fat+a2*fat2+a3*age					0.99	0.90
• Females	1.88	0.67	0.0043	0.05
• Males	1.98	0.72	0.0013	0.05
Log(VAT)=a0+a1*fat+a2*fat2					0.72	0.44
African Americans
• Females	−2.76	0.17	−0.0028
• Males	−2.52	0.17	−0.0028
Asian, Caucasian
• Females	−2.26	0.17	−0.0028
• Males	−2.02	0.17	−0.0028
Hispanic
• Females	−2.60	0.17	−0.0028
• Males	−2.36	0.17	−0.0028
Other
• Females	−2.26	0.17	−0.0028
• Males	−2.02	0.17	−0.0028
Log(IMAT)=a0+a1*fat+a2*fat2					0.75	0.38
• Females	−2.21	0.12	−0.0013
• Males	−2.05	0.12	−0.0013

SAT: subcutaneous adipose tissue; TAT: total adipose tissue; VAT: visceral adipose tissue; IMAT: intermuscular adipose tissue. SEE_PRESS: Standard error estimate; a0 is the constant term, a1 is the coefficient of fat (by dual energy-X ray absorptiometry), a2 is the coefficient of fat², and a3 is the coefficient of age in the prediction equations. Log(x) indicates the natural logarithm of x.

DXA prediction of adipose tissue sub-depots

The prediction equations for the sub-depots are presented in Table 3. Age was a significant predictor in the SAT quadratic model. Both VAT and IMAT equations were derived using the natural logarithm of the dependent variable. Race was a significant predictor of Log(VAT). Neither age nor race were significant predictors of Log(IMAT). The r²_PRESS for the prediction equations for sub-depots were SAT (0.99), VAT (0.72), IMAT (0.75).

SAT

A comparison of predicted SAT to measured SAT by sex revealed no significant differences at 1-year (p=0.2173) and at 2-year (p=0.7793) follow-up.

VAT

No significant differences were found for the comparison of predicted Log(VAT) and measured Log(VAT) by sex at 1-year (p=0.7604) or 2-year (p=0.2134) follow-up.

IMAT

At 1-year follow-up, there was a significant difference between predicted and measured Log(IMAT) for females (p=0.0170). Although there was no significant difference between predicted and measured Log(IMAT) for males, there was a statistical trend suggesting a difference (p=0.06026). In contrast, at two years follow-up there was a significant difference between predicted and measured Log(IMAT) for males (p=0.0184). There was a statistical trend suggesting a difference for females (p=0.0816).

Discussion

This study provides for the first time useful prediction equations to estimate total adipose tissue and its sub-depots in a pediatric population. These data indicate that DXA fat values can be applied to obtain estimates for MRI-derived adipose tissue mass in healthy children and adolescents during development.

Prediction of adipose tissue in 5 to 17 years old children and adolescents

A principal finding of this study is that predicted TAT does not differ from measured TAT with a quadratic relationship found between TAT and Fat_DXA. That is, TAT can be predicted from Fat_DXA and this relationship varies by sex. For a given amounts of Fat _DXA, males will have a higher predicted TAT than females. Our findings suggest that the association between TAT and Fat_DXA does not change with development, and no significant differences were detected by race.

At baseline and follow-up time points, TAT measures by MRI were on average 4% greater than fat values measured by DXA exhibiting generally higher values for girls than for boys by both techniques. These sex differences are in agreement with previous reports during childhood (17, 18). Comparisons of whole DXA fat and MRI-measured TAT in adults have been reported in the literature indicating that DXA and MRI adipose tissue values are highly correlated (19–21). In a study of adult HIV-infected patients and healthy controls, DXA-estimated peripheral fat was consistently larger than MRI-estimated adipose tissue in regional body fat depot. Moreover, with increasing levels of fat mass, the differences between DXA fat and MRI derived adipose tissue were greater (22). Postulated mechanisms include MRI’s inability to detect small adipose tissue depots (e.g., intramuscular or liver fat depots) below the resolution of MRI; therefore, these fat amounts are not included in MRI estimates (22). DXA has been found to underestimate fat mass in leaner children and overestimate fat mass in heavier subjects (23).

We show that fat mass measured by DXA is useful to predict, not exclusively TAT but also the sub-depots SAT, VAT, and IMAT, where VAT and IMAT were modeled using the natural logarithm. The prediction of adipose tissue sub-depots revealed that SAT varied by age and sex, and VAT varied by sex and race. Race influenced only VAT values with Africans Americans having less VAT than Hispanics, Caucasians, Asians and Others, when fat is held constant. The one exception was IMAT as the equation in this case did not predict values that agreed with the measured data at follow-up.

Prediction of adipose tissue (AT) with longitudinal evaluation

In a small sample of predominantly prepubertal children followed over two years, predicted mean TAT values were consistent with the measured TAT values at each time point. The results were not influenced by race or sex. Therefore, the predicted TAT for a given amount of total body fat was equivalent for males and females of differing races, however, males have more TAT than females for the same amount of total body fat. VAT and IMAT were not influenced by age. In general, the adipose tissue sub-depot prediction equations were valid using longitudinal data. However, we acknowledge that the sample size was small, with few children in some age groups and developmental stages.

Strengths and limitations

DXA offers some advantages over MRI as a means of estimating fat distribution including easier access to an instrument, an uncomplicated test technique for children as no cooperation is required, and a relatively low cost. A limitation of DXA is however, the radiation exposure to the volunteer, which despite being low (≤ 15 μSv per whole body scan for a child compared to 20 μSv for a chest X-ray), is unattractive for some parents (24). The technique is attractive and valuable for providing measurements of longitudinal changes in body composition, both at the total body and regional level (22, 25). The current study used the longitudinal measurements that were available on a subset of the sample rather than an independent sample in validation study which we acknowledge as a limitation.

Conclusion

Fat mass measured by DXA is useful to predict TAT, SAT and VAT. For the pediatric population it is possible to translate the fat measurement into a predicted TAT, SAT and VAT value that is expected to agree with measured TAT, SAT and VAT, but not for IMAT. For longitudinal studies in the pediatric population, DXA is an advantageous technique for the assessment of body compositional changes that occur during childhood.

Methods

Protocol

Subjects included in this analysis participated in one of two studies where the protocols involved the same whole-body MRI protocol and whole body DXA scan. Study 1 was a cross-sectional study (baseline data only, T0) and Study 2 employed a longitudinal design and participants were evaluated at baseline (T0), 1-year (T1) and 2-year (T2). The T0 data presented in this analysis represents combined data from both studies. The T1 and T2 data are from Study 2 only.

All medical and body composition evaluations were carried out on the same day with the subject clothed in a hospital gown and without shoes. Subjects reported for testing to the Body Composition Unit, St. Luke’s-Roosevelt Hospital Center. The studies were approved by the Institutional Review Board of St. Luke’s-Roosevelt Hospital (IRB# 00-069 and IRB # 03-072), and a written consent as well as a verbal assent were obtained from all participating children, young adults and their parents before participation.

Subjects

Subjects were a multi-ethnic group of 176 healthy children and young adults (n=137 from Study 1 and n=39 from Study 2). All participants were recruited through schools, local area newspapers advertisements, and flyers dispersed in different locations in the local community. Subjects were required to be ambulatory, free of medical conditions or metabolic disorders (e.g. acute or chronic diseases) and were not taking medications that could potentially affect the variables under investigation.

At each study point, body weight was measured using a scale (Avery Weight-Tronix digital scale, model DS-01, Pointe-Claire, Quebec, Canada) to the nearest 0.1 kg, and height using a wall mounted stadiometer (235 Heightronic Digital Stadiometer, Quick Medical and Measurement Concepts, Snoqualmie, WA) to nearest 0.1 cm. Pubertal status was evaluated by a pediatrician using the Tanner criteria at baseline and at follow-up points. Ethnicity was assessed by questionnaire reporting of the ethnic background of the child.

Dual-energy X-ray absorptiometry (DXA)

Total body fat content was estimated in all children with a whole-body scanner (Lunar Prodigy, GE Medical, Madison, WI; software versions 8.8 and 11.4) using computer generated and manually confirmed default lines on anterior view planogram. In our laboratory the coefficients of variation was 9.9% for fat mass measurements on the Lunar Prodigy (based on monthly values for water and methanol phantom assessments from January 2003 through December 2006). Reproducibility of DXA in children has been reported (26); however, because of concerns surrounding unnecessary radiation exposure in healthy children, scan reproducibility in children was not performed in our studies.

Magnetic resonance imaging (MRI)

Whole body MRI was performed to quantify total adipose tissue (TAT) and sub-depots subcutaneous (SAT), visceral adipose tissue (VAT), and intermuscular adipose tissue (IMAT). A 1.5-T scanner was used (6X Horizon: GE Milwaukee, WI) and images created with T-weighted spin-echo sequence and a 210ms repetition time and echo time of 17ms. Subjects lay motionless on the scanner platform with arms extended above their heads. Transverse images were acquired for the whole body with a between slice gap of 40mm in taller pediatric subjects and 25 or 35mm gaps for smaller pediatric subjects generating between 30 and 40 image slices per subject. The SLICEOMATIC (TomoVision Inc, Montreal, Canada) software was used to calculate cross sectional tissue areas according to previously established protocols (27). Adipose tissue volumes were converted to mass using the estimated density of both tissues 0.92 kg/L (28). The interclass correlation coefficients among three analysts (who each read the same scans twice, separately by a three month interval) for SAT, VAT, and IMAT in our laboratory are 0.99, 0.95, and 0.97, respectively.

Statistical analysis

Descriptive statistics, mean and standard deviation, were calculated for continuous variables. For categorical variables, number and percentage were calculated for each level of the variable. The paired t test was used to test the null hypothesis that the mean TAT was equal to the mean Fat_DXA. Multiple regression analysis was used to derive equations to predict TAT and each of the sub-depots SAT, VAT, and IMAT using the T0 data. Potential independent variables for the regression equations included Fat_DXA, age, race, sex, Tanner stage, and their interactions. The backward elimination method was used to select the subset of independent variables to retain in the prediction equations. The predicted residual sum of squares method (PRESS) was used to as an alternative to data splitting to validate the model (29). R²_PRESS and the standard error of estimate (SEE_PRESS) statistics are provided to give an assessment of the validity of the equations. Although T1 and T2 follow-up data do not represent an independent data set, they were used to check the prediction equations. Descriptive statistics, mean, standard deviation, and interquartile range, were calculated for the difference between the predicted and measured values. The hypothesis that the mean difference was equal to zero was tested using a paired t test. There were twenty-two children with follow-up data at T1 and T2. This sample size has a power of 80% to detect a difference of 0.6*(standard deviation of the difference) using a paired t test with a level of significance of 0.05 two-tailed. Separate analyses were performed for each variable. All statistical analyses were performed using the SAS version 9.2 software package (SAS Institute INC, Cary, NC) for personal computer. The level of significance for all statistical tests of hypotheses was 0.05.