MEASUREMENT OF BODY COMPOSITION CHANGES WITH WEIGHT LOSS IN POSTMENOPAUSAL WOMEN: COMPARISON OF METHODS

Abstract

Background

The accurate measurement of body composition changes is important when evaluating the efficacy of medical nutrition therapy and weight management programs, yet is not well documented in older women.

Objective

We compared methods of estimating energy-restriction-induced body composition changes in postmenopausal women. Design: 27 women (59 ± 8 y; BMI 29.0 ± 2.9 kg/m²; mean ± SD) completed a 9-wk energy restriction period (5233 kJ/d, (1250 kcal/d)). Changes in % body fat (Δ%BF), fat mass (ΔFM), and fat-free mass (ΔFFM) were measured by hydrostatic weighing (HW), air-displacement plethysmography (ADP), dual-energy x-ray absorptiometry (DXA), and deuterium oxide dilution (D₂O). The Baumgartner et al. (Am J Clin Nutr 53:1345−1353, 1991) four-compartment (4C) model with body volume from HW was the criterion method. The 4C model with body volume from ADP was also compared. Regression equations were developed based on 4C_HW (dependent variable) utilizing results of change (POST-PRE) for each method.

Results

The women lost 6.8 ± 3.2 kg; 9% of baseline weight. Based on 4C_HW, the body composition changes were −2.4 ± 4.5 Δ%BF, −4.7 ± 3.3 kg ΔFM, and −2.6 ± 4.4 kg ΔFFM. No differences were detected by ANOVA for Δ%BF, ΔFM, and ΔFFM among 4C_HW, HW, ADP, DXA, D₂O, and 4C_ADP. Bland-Altman limits of agreement showed differences between methods that ranged from 14.5 to −14.1 Δ%BF, 7.8 to −8.1 kg ΔFM, and 7.5 to −8.4 kg ΔFFM for individuals. A bias was shown with 4C_ADP overestimating Δ%BF (1.4 %) and FM (0.6 kg) and underestimating ΔFFM (−1.2 kg) compared to 4C_HW. The regression model was acceptable for %BF (4C_ADP, 2C_HW, and 2C_D2O); FM and FFM (4C_ADP, 3C_DXA, 2C_HW, and 2C_D2O), but not for other estimates of %BF, FM, FFM.

Conclusions

These body composition assessment methods may be used interchangeably to quantify changes in % body fat, fat mass, and fat-free mass with weight loss in groups of postmenopausal women. 4C_ADP overestimates Δ%BF and underestimates ΔFFM. When utilizing one of these comparison methods (4C_ADP, 3C_DXA, 2C_HW, 2C_D2O) to quantify changes in fat mass and fat-free mass for an individual postmenopausal woman, regression equations may be used to relate the data to 4C_HW.

Introduction

The ability of researchers and health professionals to accurately estimate changes in body composition is critical in determining the effectiveness of medical nutritional therapy and changes that occur with weight management programs. Researchers, health professionals and the public may benefit from technological advances that have increased available options for estimating changes in % body fat (Δ%BF), fat mass (ΔFM), and fat-free mass (ΔFFM). Hydrostatic weighing (HW) has been a long standing criterion method of determining body density (1). Hydrostatic weighing estimates body composition based on a 2-compartment (2C) model in which body weight is equal to the sum of FM and FFM. However, the use of HW is often confined to a research setting. Both air-displacement plethysmography (ADP) (2), another method also based on a 2C model and dual-energy x-ray absorptiometry (DXA), a method based on a 3-compartment (3C) model (3), (FM plus FFM (sum of lean tissue mass and total body bone mineral)) are increasingly available in health and fitness facilities.

The 2C or 3C (DXA) models may not accurately estimate changes in body composition due to energy restriction in overweight postmenopausal women (4, 5). Both models (6) assume the densities of FM (0.9007 g/cm³), FFM (1.1000 g/cm³) (D_FFM), and water content of FFM (73.2%) remain constant with aging, obesity and changes in body mass (4). Since these assumptions of density are based on early analysis of male cadavers (4), they provide potentially significant sources of error when calculating body composition because age, race, gender, level of body fatness and changes in body weight can influence fat-free compartments (protein, water, mineral) (5, 7-15). Nevertheless, Visser et al. (16) determined that most of the variability in D_FFM was accounted for by total body water (TBW) with only minor contributions from age, race, and gender. Visser et al. concluded that the assumption of constant D_FFM is valid in black, elderly and obese subjects, but observed a tendency for a lower D_FFM in older, white women.

The influence of weight loss on the assumed D_FFM is less well studied and documented. Researchers have found an increase in TBW:FFM with weight loss (5, 7, 13). This could be due to changes in TBW (13) and bone mineral content (BMC), which may (13-15, 17) or may not (5, 18-20) decrease with weight reduction. Thus, variability in D_FFM may result in errors in estimating FFM, and therefore %BF and FM, which would limit use of a 2C or 3C model (6).

Given the apparent possible assumption violations for the 2C and 3C models (DXA), a 4-compartment (4C) model has been suggested as the best in vivo body composition model available (16, 21, 22). The 4C molecular model reduces the potential for assumption errors by utilizing independent estimates of four components (FM, BMC, TBW and protein mass). This 4C model was developed specific to the use of HW for measuring body volume (BV). The development of ADP for measurement of BV led researchers to investigate whether BV_ADP may be substituted for BV_HW in the 4C model using various populations. While both utilize densitometry via the Siri equation to obtain %BF, they differ in methods used to obtain body volume (residual lung volume, 4C_HW, vs. thoracic lung volumne, 4C_ADP). Recently, investigators found small, but significant differences in BV and density and a trend toward significant differences between 4C model estimates of %BF (23).

Researchers have yet to present data comparing changes in body composition assessment using 2C_HW, 2C_ADP, 3C_DXA, TBW using the deuterium oxide (D₂O) dilution technique (2C), and 4C_ADP against 4C_HW in overweight and obese postmenopausal women with energy restriction. Therefore, uncertainty exists about whether mean estimates of Δ%BF, ΔFM and ΔFFM among these methods would agree and provide interchangeable measurements of body composition. Reporting of regression analyses, R², SEE and Bland Altman analyses has been recommended (24) to aid with understanding of the relationship between methods. Therefore, the purpose of this study was to examine the interchangeability of 2C-, 3C- and 4C_ADP models of body composition using the 4C_HW model as the criterion method for estimates of Δ%BF, ΔFM and ΔFFM in postmenopausal, overweight and obese women following energy restriction.

Subjects and methods

Subjects

The subjects for this study were 27 overweight and mildly obese, postmenopausal women (59 ± 8 y; BMI 29.0 ± 2.9 kg/m², see Table 1). This group of women was a subset of women who completed a 9-wk energy restriction weight loss protocol (described below). The primary aim of the parent study was to assess the effects of the macronutrient distribution and source of dietary protein on changes in weight and body composition with energy restriction. The Purdue University Committee on the Use of Human Research Subjects approved the protocol, and informed consent was obtained from each subject prior to enrollment in the study. Phone screening was conducted to determine if a potential subject met basic eligibility criteria which included: nonsmoker, age ≥ 50 y, ≥ 2 y postmenopausal (self-reported based on last menstruation), and body mass index (BMI) ≥ 25 and ≤ 34 kg/m2 at the time of screening. Potential subjects attended a medical screening visit and completed diet history questionnaires. The medical screening consisted of a written medical history, measurements of height (shoes removed), weight (nude, with weighed robe) on electronic scale, blood pressure, resting electrocardiogram, and clinical blood and urine chemistries. Medical history and screening data for all subjects were reviewed by the study physician prior to enrollment in the study. Subjects needed to be willing to discontinue use of all nutritional supplements (except calcium, due to recruitment difficulties), nonprescription medications and alcohol three weeks prior to starting and throughout the study period. All subjects received a monetary stipend.

Table 1.

Subject characteristics before and after 9-wk energy restriction¹,²

	Baseline	Post	Change (Δ)
Age (y)	59 ± 8
Height (m)	1.64 ± 0.05
Weight (kg)	76.9 ± 10.2	70.1 ± 10.6	−6.8 ± 3.23
BMI (kg/m2)	29.0 ± 2.9	26.1 ± 3.5	−2.9 ± 3.13
BVHW (L)	76.6 ± 10.3	70.2 ± 11.3	−6.4 ± 3.63
BVADP (L)	77.2 ± 10.5	70.0 ± 11.1	−7.2 ± 3.23
Db (HW) (g/L)	1.011 ± 0.0124	1.016 ± 0.014	0.005 ± 0.0113
Db (ADP) (g/L)	1.003 ± 0.0104	1.009 ± 0.011	0.006 ± 0.0075
TBW (kg)	32.3 ± 3.4	30.6 ± 4.0	−1.7 ± 2.73
TBW/FFM4C HW (%)	72.9 ± 3.4	73.2 ± 6.4	0.3 ± 0.05
BMC (kg)	2.4 ± 0.3	2.4 ± 0.3	0.0 ± 0.0
BMC/FFM4C HW (%)	5.5 ± 0.7	5.8 ± 0.8	0.3 ± 0.65

¹Mean ± SD, BMI, body mass index; D_{b (HW)}, body density measured by hydrostatic weighing; D_{b (ADP)}, body density measured by air-displacement plethysmography; BV, body volume; TBW, total body water; FFM, fat-free mass; BMC, bone mineral content; BMD, bone mineral density;

²n=27;

³P < 0.05, paired t-test;

⁴P < 0.05, t-test;

⁵P < 0.01, paired t-test

Experimental design

This 11-wk study included a 2-wk baseline-testing period (no dietary control), followed immediately by a 9-wk period of energy-restriction-induced weight loss, with testing repeated during the 9th wk of intervention. Subjects were assigned to one of four groups; three energy restriction groups consumed a lacto-ovo vegetarian basal diet (4186 kJ/d (1000 kcal/d)) plus 1047 kJ/d (250 kcal/d) of beef (48% carbohydrate (CHO), 26% protein (PRO), 26% fat (FAT) of total energy intake), or chicken (48% CHO: 26% PRO: 26% FAT), or carbohydrate/fat snack foods (58% CHO: 16% PRO: 26% FAT), and a control group was instructed to follow their habitual diet and to maintain body weight. All subjects were asked to maintain their usual levels of physical activity and to not begin any type of exercise program.

Fifty-seven Caucasian women started the diet intervention period. During the intervention period one subject dropped out for personal reasons, another subject was dismissed due to her need to start new medication, and one subject dropped out due to an unwillingness to adhere to the diet. Fifty-four women completed the study. The data from the control group (n=11) were excluded from these analyses since this group did not change body weight. We were unable to obtain residual lung volume measurements on all of the subjects due to technical difficulties with the equipment and because the procedure was uncomfortable for some of the women (n=10). Data from six women were excluded because they were unable to provide sufficient urine volume required for D₂O analysis. After excluding these subjects, data from 27 women were used for analyses.

Body Composition Assessment Procedures

Subjects were instructed to drink about 1400 mL (six 8-oz glasses) of water the day before testing. They were instructed to drink about 240 mL of water the morning of testing prior to arriving at the laboratory. Each subject arrived at the laboratory in the early morning after an overnight 12-h fast. Each subject was not permitted to consume food or water during testing and wore a tight fitting bathing suit for HW and ADP testing, and a bathing cap for ADP testing. All jewelry was removed. Testing was conducted in the following order: DXA, ADP and then HW, and completed within a two-hour period to control for diurnal changes in body composition.

Dual-energy x-ray absorptiometry

Wearing street clothing free of metal and/or plastic, each subject remained motionless in the supine position for 5−10 minutes, and was scanned on standard or thick mode based on subject size with a total body scanner (GE Medical Systems/LUNAR Prodigy™, Madison, WI). Prodigy software (version 6.7) was used to calculate the ratio of soft tissue (fat and fat-free) at the two energies measured (25) based on a 3C model in which BMC and bone-free soft tissue attenuated energy from the x-ray in a tissue specific manner. FFM was calculated as the sum of lean tissue plus BMC (CV = 1.7%). %BF is equal to FM divided by body weight (x 100). The same technician performed all analyses. Analyzing a phantom spine provided by the company assessed quality assurance, and daily calibrations were performed before all scans using a calibration block provided by the manufacturer. The coefficient of variation for percent fat from repeated scans on a group of 20 men and women ranging from 9% to 50% fat in our laboratory was 1.9%.

Air-displacement plethysmography

Air-displacement plethysmography was utilized to measure raw BV as previously described (2) using the Bod Pod® Body Composition System (Life Measurement, Inc., Concord, CA). Triplicate trials were performed on each subject. Before each trial the machine was calibrated using a precision 50 liter cylinder provided by the manufacturer. Each subject's thoracic lung volume was measured during normal breathing using a tube connected to the ADP breathing circuit system as described elsewhere (2). The 2C Siri equation (6) was used to estimate %BF after D_b was obtained. FFM was calculated as body mass minus FM. The CV for between week %BF in our laboratory using 37 men and women ranging from 9% to 50% BF was 4.0%.

Hydrostatic weighing

Hydrostatic weighing was used to measure BV as previously described (26) (Exertech® Body Density Measurement System, Dresbach, MN). Prior to each session the weighing platform located inside the stainless steel water tank was calibrated with a 7.82 kg metal bar, as specified by the equipment manufacturer. An additional 4.5 kg weight belt was tared and used to assist the subjects with submersion. Subjects sat or kneeled completely submerged underwater on the weighing platform and were instructed to maximally exhale air. A total of 8−10 replicate trials were performed in the underwater weighing tank with each subject. Values from the three heaviest trials were averaged to obtain body mass in water. The 2C Siri equation (6) was used to estimate %BF after D_b was obtained. FFM was calculated as body mass minus FM. Residual lung volume (air remaining in the lungs after maximal exhalation) was measured outside the tank using an open circuit oxygen-dilution technique (Vmax 229 series, SensorMedics Corp. Yorba Linda, CA). The CV for between week %BF in our laboratory using seven postmenopausal women ranging from 31% to 53%BF was 3.7%.

Deuterium oxide dilution

Total body water was determined using D₂O as previously described (27, 28). Upon arrival at the laboratory a baseline urine sample was obtained and then each subject consumed orally a 20-g dose of D₂O. Urine samples were again collected at 2, 3 and 4 hours following the dose. Subjects reached plateau for isotope dilution by the 3^rd hour as demonstrated by no significant difference in D₂O content between the 3^rd and 4^th hour samples, p = 0.46. Urine volume was measured and an aliquot was stored at −20°C. The D₂O concentration of the urine samples was determined using an Infrared Spectrophotometer 360 Water System (Nicolet Instrument Corp., Waltham, Mass.) (28). Total body water is based on mass balance and was calculated using a dilution principle (with corrections made for losses of isotope in urine). Calculations were used to obtain total isotope dilution (27). Total body water and FFM were calculated with the following formulas:

TBW = Calculated total isotope dilution / 1.04 (correction for exchange of isotope with nonaqueous hydrogen during equilibration in vivo).

FFM (kg) = TBW (L) / 0.732 (hydration constant)

TBW estimates FM based on body mass minus FFM. %BF was calculated as (FM/body weight X 100). The CV for between week %BF in our laboratory using 23 women ranging from 24% to 51%BF was 4.6%. The 0.732 hydration constant was the value originally used by Scholerb (27), which was adopted from guinea pig data. Our confidence using this number is supported by the results from 18 studies that determined the hydration of fat-free mass by chemical analysis of whole nonruminant animals (including three studies in humans) (29). The mean (±SD) fractional hydration of fat-free mass among the 18 studies was 0.731 ± 0.013, and 0.731 among the three human studies.

Compartment model

The 4C model with BV from HW of Baumgartner et al. (9) served as the criterion model. Body volume from ADP was substituted for BV from HW and provided a second 4C model using the same equation. This 4C model requires measurement of BV, TBW, BMC and body weight (BW) to calculate FM:

FM = [(2.75 × BV) – (0.714 × TBW) + (1.148 × TBBM) – (2.05 × BW)]

To obtain total body bone mineral (TBBM) an adjustment for the ratio of osseous to nonosseous mineral and loss of mineral during ashing of bone (9) was calculated as follows:

TBBM (g) = BMC × 1.279 To obtain percent body fat and FFM:

%BF = FM / BW FFM = BW minus FM

Statistical Analyses

Data were analyzed using two-way ANOVA to assess the main effects of diet group (between-subject factor) and measurement method (within-subject factor). These analyses established that there were no differential body composition responses to the energy restriction among the three diet groups. A paired t-test was used to compare subject characteristics in Table 1 (except for age and height) at PRE and PRE vs. POST. The relationship between the criterion method (4C_HW) and the comparison method single differences (Δ), was established using regression analyses.

The Bland-Altman statistical test was used to examine the likelihood of a comparison method agreeing to a criterion method. This procedure examined bias (mean difference) between the two methods and if the difference was related to the magnitude of the change. The standard deviation of the delta difference (ΔΔ) (single difference (Δ) 4C_HW minus single difference (Δ) comparison method) was examined and limits of agreement obtained. The bias was examined using paired t-tests. Bland-Altman tests were conducted using ANALYSE-IT Software, LTD version 1.71 (ANALYSE-IT Software, LTD, Leeds, England). ANOVA and paired t-tests were conducted using SPSS statistical software version 12.0 (SPSS Inc., Chicago, IL). A P level of < 0.05 was used for all significance testing. All values are reported as mean ±SD unless otherwise noted.

Results

Body composition

Body weight and BMI decreased with energy restriction, P < 0.01 (Table 1). Changes in body weight measured by 3C_DXA were not significantly different from weight measured by the 2C_ADP scale (−6.9 ± 2.9 kg, vs. −6.8 ± 3.2 kg, r = 0.98, P < 0.0001). Body volume decreased and D_b increased with energy restriction. D_b measured by 2C_HW and 2C_ADP were significantly different at baseline, P < 0.1. Changes in BV and D_b measured by 2C_HW were not significantly different from changes measured by 2C_ADP, P = 0.23, P = 0.37, respectively. Total body water was lower after energy restriction, P < 0.01, but TBW: FFM was not different. BMC did not change due to energy restriction, however, BMC: FFM increased after energy restriction, P < 0.01.

Body composition methods comparison

Table 2 presents the single difference (POST-PRE) for each method. There were no differences in mean Δ%BF, ΔFM and ΔFFM among the 6 methods: 4C_HW, 4C_ADP, 3C_DXA, 2C_HW, 2C_ADP, and 2C_D20.

Table 2.

Results of body composition changes (single difference) due to energy restriction in postmenopausal women

Method/Model	PRE	POST	Single difference POST- PRE (Δ)	Confidence Interval (95%)
% BODY FAT
4CHW	41.5 ± 5.4	39.1 ± 6.5	−2.4 ± 4.5	6.4 to −11.1
4CADP	43.6 ± 4.9	39.9 ± 5.5	−3.7 ± 3.0	2.1 to −9.6
3CDXA	43.0 ± 4.8	40.1 ± 5.7	−2.9 ± 1.9	0.8 to −6.6
2CHW	39.3 ± 5.8	37.1 ± 6.8	−2.2 ± 5.3	8.3 to −12.6
2CADP	43.5 ± 5.4	40.8 ± 5.7	−2.7 ± 2.7	2.6 to −8.0
2CD2O	42.3 ± 5.5	40.0 ± 5.8	−2.3 ± 4.8	7.0 to −11.6
FAT MASS (kg)
4CHW	32.0 ± 7.5	27.3 ± 6.6	−4.7 ± 3.3	1.6 to −11.1
4CADP	33.8 ± 7.5	28.3 ± 7.4	−5.5 ± 2.7	0.2 to 10.8
3CDXA	33.4 ± 7.7	28.6 ± 8.1	−4.8 ± 2.3	−0.3 to −9.4
2CHW	30.6 ± 7.8	26.2 ± 7.5	−4.4 ± 4.8	5.0 to −13.8
2CADP	34.0 ± 7.8	29.0 ± 8.0	−5.0 ± 3.1	1.1 to −11.1
2CD2O	32.9 ± 7.6	28.3 ± 7.6	−4.6 ± 4.0	3.3 to −12.5
FAT-FREE MASS (kg)
4CHW	44.4 ± 4.5	42.0 ± 6.2	−2.6 ± 4.4	4.9 to −9.9
4CADP	43.0 ± 4.2	41.9 ± 5.2	−1.1 ± 2.8	4.4 to −6.6
3CDXA	43.5 ± 3.6	41.5 ± 3.6	−2.0 ± 1.0	0.1 to −4.0
2CHW	46.2 ± 4.8	44.3 ± 5.4	−1.9 ± 5.3	8.6 to −12.3
2CADP	42.9 ± 4.1	41.0 ± 4.4	−1.9 ± 1.0	0.1 to −3.9
2CD2O	44.1 ± 4.6	41.8 ± 5.5	−2.3 ± 3.7	5.0 to −9.6

Values are mean ± SD unless otherwise noted; significant differences were not found among the methods using ANOVA. 4C_HW = 4-compartment model using hydrostatic weighing for body volume, 4C_ADP = 4-compartment model using air-displacement plethysmography for body volume, 3C_DXA, 3-compartment model dual-energy x-ray absorptiometry, 2C_HW= 2-compartment model hydrostatic weighing, 2C_ADP= 2-compartment model air-displacement plethysmography, 2C_D2O = 2-compartment model deuterium oxide dilution technique for measurement of total body water

In general Bland-Altman analyses showed very large differences between 4C_HW and other methods for Δ%BF, ΔFM and ΔFFM.

Δ% body fat

The results of the delta difference (single difference 4C_HW minus single difference comparison method) for each pair of method's estimate of Δ%BF are presented in Table 3). The limits of agreement showed considerable variation and lack of agreement between 4C_HW and each of the methods (14.5 to −14.1 %BF). A significant bias detected between 4C_HW and 4C_ADP (Figure 1B) indicated 4C_ADP overestimated Δ%BF compared to 4C_HW (4C_ADP > 4C_HW, bias = 1.3 %; limits of agreement, 8.1 to −5.4%). The limits of agreement suggest that the two methods would give measurements that differ by ∼8% to −5%. It appears that 4C_HW detects less change in Δ%BF compared to 4C_ADP. An adjustment to 4C_ADP (the new method being compared to 4C_HW) can be made by subtracting the bias from the estimates of Δ%BF from 4C_ADP (30). Bland-Altman plots of Δ%BF for 2C_HW, and 2C_D2O compared to 4C_HW (Figures 1H and 1L, respectively) showed that there was no recognizable relationship between the mean and the delta differences. In other words, the magnitude of the delta difference did not appear to be affected by the amount of Δ%BF, whereas this trend was observed between 4C_HW and 3C_DXA (Figure 1F) and between 4C_HW and 2C_ADP, (Figure 1J).

Table 3.

Comparison of methods and models to estimate body composition changes (delta difference) due to energy restriction in postmenopausal women to the four compartment hydrostatic weighing criterion method

Method/Models	Delta Difference single difference criterion method - single difference comparison method (ΔΔ)	Limits of Agreement1 (95%)
% BODY FAT
4CHW - 4CADP	1.4 ± 3.42	8.1 to −5.2
4CHW - 3CDXA	0.6 ± 4.5	9.4 to −8.3
4CHW - 2CHW	−0.2 ± 7.3	14.5 to −14.1
4CHW - 2CADP	0.4 ± 5.4	10.9 to −10.1
4CHW - 2CD2O	0.0 ± 4.1	8.0 to −8.1
FAT MASS (kg)
4CHW - 4CADP	0.6 ± 2.72	5.9 to −4.6
4CHW - 3CDXA	0.1 ± 3.1	6.2 to −6.0
4CHW - 2CHW	−0.9 ± 3.7	6.3 to −8.1
4CHW - 2CADP	0.2 ± 3.8	7.8 to −7.3
4CHW - 2CD2O	−0.2 ± 2.9	5.6 to −5.9
FAT-FREE MASS (kg)
4CHW - 4CADP	−1.2 ± 2.52	3.7 to −6.0
4CHW - 3CDXA	−0.4 ± 3.3	6.2 to −6.9
4CHW - 2CHW	−0.5 ± 4.1	7.5 to −8.4
4CHW - 2CADP	−0.4 ± 3.8	7.1 to −7.9
4CHW - 2CD2O	0.0 ± 3.0	6.0 to −6.0

Values are mean ± SD unless otherwise noted; 4C_HW = 4-compartment model using hydrostatic weighing for body volume, 4C_ADP = 4-compartment model using air-displacement plethysmography for body volume, 3C_DXA, 3-compartment model dual-energy x-ray absorptiometry, 2C_HW= 2-compartment model hydrostatic weighing, 2C_ADP= 2-compartment model air-displacement plethysmography, 2C_D20 = 2-compartment model deuterium oxide dilution technique for measurement of total body water.

¹Limits of agreement represent the agreement between 4C_HW and the comparison method based on Bland-Altman (ΔΔ ± 2SD)

²Comparison significantly different based on paired t-test, P < 0.05.

Figure 1.

Relation between a) change in % body fat (Δ%BF) estimated by 4C_HW compared with 4C_ADP; b) delta difference (ΔΔ) against mean change (Δ) in %BF by 4C_HW compared with 4C_ADP;, c) Δ fat-free mass (FFM) estimated by 4C_HW compared with 4C_ADP; d) ΔΔFFM against ΔFFM by 4C_HW compared with 4C_ADP; e) Δ%BF estimated by 4C_HW compared with 3C_DXA; f) ΔΔ%BF against Δ%BF by 4C_HW compared with 3C_DXA; g) Δ%BF estimated by 4C_HW compared with 2C_HW; h) ΔΔ%BF against Δ%BF by 4C_HW compared with 2C_HW; i) Δ%BF estimated by 4C_HW compared with 2C_ADP; j) ΔΔ%BF against Δ%BF by 4C_HW compared with 2C_ADP; k) Δ%BF estimated by 4C_HW compared with 2C_D2O; l) ΔΔ%BF against Δ%BF by 4C_HW compared with 2C_D2O

Δfat mass

The results of the delta difference for each pair of method's estimate of ΔFM compared to the 4C_HW estimate are presented in Table 3. Compared to 4C_HW model there was considerable variation in agreement between all methods for ΔFM (7.8 to −8.1 kg FM). A significant systematic bias was detected between 4C_HW and 4C_ADP (bias = 0.6, limits of agreement, 5.9 to −4.6 kg FM), however not between 4C_HW and other methods (Bland-Altman plots not shown).

Δfat-free mass

The results of the delta difference for each pair of method's estimate of ΔFFM compared to the estimate by 4C_HW are presented in Table 3. Compared to 4C_HW there was considerable variation for agreement between all methods for ΔFFM (7.5 to −8.4 kg FFM). A significant bias detected between 4C_HW and 4C_ADP indicated 4C_ADP underestimated ΔFFM compared to 4C_HW (4C_ADP < 4C_HW, −1.2 kg; limits of agreement, 3.7 to −6.0 kg), Figure 1D. The level of ΔFFM affected agreement between the methods. The Bland-Altman plot shows that the larger the ΔFFM the less likely the methods will agree.

Regression coefficients after energy-restriction induced weight loss

In general, the amount of variance between 4C_HW and other methods ranged from 0% to 66% for %BF, 0 % to 49 % for FM and 0% to 45% for FFM. %BF estimated with 2C_HW accounted for 66% of the variance in %BF measured by 4C_HW as indicated by the R² value (R² = 0.66, figures 1G). Other relatively high correlations included FM 4C_ADP (R²= 0.51) figure not shown, and FFM 4C_ADP (R²= 0.57), figure 1C, followed by %BF 4C_ADP (R²= 0.43), figure 1A and FFM D₂O (R² = 0.49), figure not shown. However, correlation coefficients were lowest for %BF 3C_DXA and 2C_ADP, R² = 0.00 and 0.03, figures 1E and 1I, respectively) FM 2C_ADP and 3C_DXA, R² = 0.07 and 0.18, respectively, and FFM 2C_ADP and 3C_DXA, R² = 0.0.02 and 0.33, respectively. Table 4 shows additional regression coefficients for %BF, FM, and FFM for each method in relation to 4C_HW. Slopes for prediction of %BF by 4C_HW from the comparison methods resulted in slopes from −0.1 to 1.0; prediction of FM 0.3 to 0.8 and prediction of FFM 0.5 to 2.1.

Table 4.

Regression results for body composition changes (single difference) due to energy restriction in postmenopausal women

Variable		R2	Slope	Y-intercept	SEE
Dependent	Independent
Δ% BODY FAT
4CHW	4CADP1	0.43	1.0 ± 0.22, 3	1.3 ± 1.1	3.4
4CHW	3CDXA	0.03	0.4 ± 0.5	−1.1 ± 1.6	4.5
4CHW	2CHW1	0.66	0.7 ± 0.12	−0.9 ± 0.6	2.7
4CHW	2CADP	0.00	−0.1 ± 0.3	−2.6 ± 1.32	4.5
4CHW	2CD2O1	0.37	0.6 ± 0.22	−1.0 ± 0.82	3.6
ΔFAT MASS (kg)
4CHW	4CADP1	0.51	0.8 ± 0.22, 3	0.0 ± 1.0	2.4
4CHW	3CDXA1	0.18	0.6 ± 0.32, 3	−1.9 ± 1.4	3.0
4CHW	2CHW1	0.42	0.4 ± 0.12	−3.1 ± 0.62	2.5
4CHW	2CADP	0.07	0.3 ± 0.2	−3.4 ± 1.22	3.2
4CHW	2CD2O1	0.49	0.6 ± 0.12	−2.2 ± 0.72	2.4
ΔFAT-FREE MASS (kg)
4CHW	4CADP1	0.57	1.0 ± 0.22, 3	−1.2 ± 0.62	2.6
4CHW	3CDXA1	0.33	2.1 ± 0.62, 3	1.8 ± 1.9	3.2
4CHW	2CHW1	0.42	0.5 ± 0.12	−1.5 ± 0.62	3.0
4CHW	2CADP	0.02	0.5 ± 0.8	−1.3 ± 1.7	3.9
4CHW	2CD2O1	0.45	0.7 ± 0.22, 3	−0.7 ± 0.7	2.9

Values are mean ± SE;

¹regression model is acceptable, prediction of variable with 4C_HW from variable by comparison method,

²significantly different than 0,

³confidence interval includes 1, 4C_HW = 4-compartment model using hydrostatic weighing for body volume, 4C_ADP = 4-compartment model using air-displacement plethysmography for body volume, 3C_DXA, 3-compartment model dual-energy x-ray absorptiometry, 2C_HW= 2-compartment model hydrostatic weighing, 2C_ADP= 2-compartment model air-displacement plethysmography, 2C_D20 = 2-compartment model deuterium oxide dilution technique for measurement of total body water

Discussion

The data from this study indicate that on a group basis the methods compared against the 4C_HW model (4C_ADP, 3C_DXA, 2C_HW, 2C_ADP, and 2C_D2O) are interchangeable to quantify group changes in %BF, FM and FFM with diet-induced weight loss in overweight and obese postmenopausal women. However, for an individual, there were large differences as indicated by the wide limits of agreement for each variable suggesting that individual estimates of changes in body composition when taken alone should be interpreted with caution. Currently there are no standard acceptable limits of agreement within body composition literature. Acceptance of limits of agreement requires use of clinical judgement because it is not a hypothesis test (31). For this reason, we support use of regression equations to obtain more acceptable individual estimates of changes in body composition that will relate to the criterion method, which in this case was the 4C_HW method. Additional consideration should be given to small magnitude changes that occur as a result of normal variation in day-to-day activities, fluid balance and diet. Therefore, to detect true changes in body composition due to energy restriction it is necessary to achieve changes in %BF, FM and FFM beyond the variation of the body composition method. The method variability of the body composition techniques, ∼1% to 4.6% in the present study, must be considered when choosing a method to estimate changes in body composition.

Comparison of the methods in the current study was based on changes in body composition due to energy restriction. Most of the previous studies have examined changes in body composition that occurred due to gastric surgery (32) or exercise (33-35), plus diet-induced energy restriction (5, 36). None used the same methods for comparison. In addition to these studies examining changes in body composition, there have been numerous cross-sectional studies comparing various combinations of 2C-, 3C- and 4C models, but since weight loss was not a factor, comparison with these studies is not appropriate.

Frisard et al. (37) compared 3C_DXA (criterion method), 2C_ADP and bioelectrical impedance analysis (BIA) to assess changes in body composition after weight loss in 56 obese men and women. Similar weight loss occurred in this study (−5.6 ± 5.5 kg vs. −6.8 ± 3.2 kg in the current study). They found significant bias whereby Δ%BF and ΔFFM were overestimated by 2C_ADP and BIA at a lower %BF and underestimated at a higher Δ%BF and ΔFFM when compared to 3C_DXA. High variability was also observed with FFM and FM. Frisard et al. (37) showed that 3C_DXA might be more accurate with larger changes in weight loss in contrast to Jebb et al. (38), whereby 3C_DXA overestimated %BF at higher tissue thickness. Comparing results among studies using 3C_DXA (criterion method) is complicated because 3C_DXA results have been shown to vary among instrument manufacturers, test modes (pencil beam compared with fan beam), software versions and tissue thickness (39). In addition, degree of obesity has been associated with measurement error and analysis mode should be consistent among laboratories (40). Recently, Schoeller et al. (29) demonstrated that 3C_DXA estimates for lean and fat mass can significantly vary among differing 3C_DXA models (hardware and software) under different laboratory conditions and recommended a 5% correction factor be applied when using QDR 4500A fan-beam 3C_DXA instrument. This was obtained from cross-sectional data that included men and women ages 19−82 y, n = 1195. We did not use this model in the current study and therefore would not apply this correction factor. The DXA instrument used in this study, LUNAR Prodigy™ fan-beam has been found to be similar to DXA using pencil-beam geometry (41). However, this was not demonstrated when QDR2000 pencil-beam vs. QDR 4500A fan-beam were compared (42).

Evans et al. (36) compared 3C_DXA, skinfolds, bioelectrical impedance and BMI using 4C_HW as the criterion model in premenopausal obese women who lost weight due to diet or diet plus exercise. Similar results for Δ%BF (ΔFM and ΔFFM were not reported) in both groups were found among the methods, but the authors indicated there was the limitation of accurately detecting small Δ%BF among these methods. They observed a bias between 3C_DXA and 4C_HW whereby 3C_DXA overestimated the loss in Δ%BF (1.3 ± 2.1% BF). We did not observe a bias between these two methods, but did observe a trend for 3C_DXA to be influenced by the magnitude of Δ%BF (Figure 1F).

In the present study, a comparison of agreement between the methods for individual differences between Δ%BF estimates from 4C_HW and 4_ADP, 3C_DXA, 2C_HW, 2C_ADP and 2C_D2O ranged from −14.5% to 14.1% (see Figures and Table 3). Large differences (10% to 15%) have been observed in other cross-sectional method comparison studies between 2C_ADP and 3C_DXA (criterion method) that used both larger and smaller populations (43-45). The limits of agreement in this study were narrowest between 4C_HW and 4C_ADP, 2C_D2O, 3C_DXA, 2C_ADP and 2C_HW following in increasing ranges of limits of agreement for Δ%BF. The estimates between 4C_HW and 4C_ADP and between 4C_HW and 3C_DXA each identified a trend that was not identified with the other methods (2C_HW, 2C_D2O, and 2C_ADP) (Figures 1H, 1J, lL). Bland-Altman plots identified a bias between 4C_HW and 4C_ADP for Δ%BF (1.3%, Figure 1b) and ΔFFM (−1.2 kg, Figure 1D). For Δ%BF, 4C_ADP estimated a higher Δ%BF and a lower estimate of ΔFFM than 4C_HW. No other biases were observed for the other comparisons to 4C_HW. Bias between 4C_ADP and 4C_HW would be likely due to differences in D_b measurements detected at baseline between the two methods. It should be noted that change in D_b was not significantly different based on method means (Table 1). However, the Bland-Altman plot between 2C_HW and 2C_ADP for D_b (figure not shown) yielded limits of agreement of −0.02 to +0.02 g/L for an individual. This may help explain why Bland-Altman plots of 4C_HW and 4C_ADP reveal a bias since this is the only component that differed between 4C_ADP and 4C_HW. The bias we observed between 4C_ADP and 4C_HW was in a small population within one laboratory. We chose to follow the same testing order for all subjects (3C_DXA, 2C_ADP, 2C_HW, respectively) which were all completed within 1−2 hours of each other. It's possible that a bias could have been introduced and that randomizing the methods would have overcome bias. A larger population with similar subject characteristics and comparisons based on data reflecting changes in body composition needs to be similarly studied before a specific correction factor is recommended. Previous method comparison studies between 2C_HW and 2C_ADP show that bias has been inconsistent in the direction of the differences between methods (24). It's possible that the limited number of subjects in this study influenced the bias since studies with largest mean differences have generally occurred in studies with the fewest subjects (24).

The individual differences between ΔFM and ΔFFM estimates from 4C_HW and 4C_ADP, 3C_DXA, 2C_HW, 2C_ADP, and 2C_D20 ranged between 7.8 to −8.1 kg and 7.5 to −8.4 kg, respectively. The differences were smaller between 4C_HW and 4C_ADP, with 2C_D20, 3C_DXA, 2C_ADP and 2C_HW following in increasing ranges of estimates for both ΔFM and ΔFFM. Overall, the individual estimates were greater for 2C_HW, 2C_ADP, 3C_DXA and 2C_D20 than 4C_HW and 4C_ADP for ΔFM and ΔFFM (Table 3). Although 4C_HW and 4C_ADP differed between each other to a smaller degree, these differences were still large enough to not be interchangeable.

The estimation of changes in body composition is a challenge due to a number of factors including: changes in D_b due to FM loss, muscle loss, TBW and BMC. The loss of FFM due to reductions in TBW can alter D_b (46) as was observed in the present study, Table 1. Research by Fogelholm et al. (5) showed change in TBW/FFM and change in D_b when comparing body composition estimates by 3C_DXA, skinfolds, bioelectrical impedance, BMI, 4C_HW, and HW during energy restriction in premenopausal obese women. They observed that 3C_DXA and methods based on the 2C model underestimated loss of FM and attributed this to the increase in TBW/FFM that was observed (5). In contrast, we observed a decrease in TBW, but no change in TBW/FFM, Table 1. Fogelholm et al. (5) attributed the change in D_b to a change in the water content of FFM (TBW), but our data do not support this since our subjects maintained the ratio of water content to FFM. A lack of change in TBW/FFM suggests there was a proportional change in FFM. Further, since the 2C and 3C models assume a constant fraction (TBW/FFM, ∼72−73%) (6, 47), it may partly explain why there were no group differences among the methods. In the current study postmenopausal women were adequately hydrated both prior to and during the energy restriction according to the normal standards noted above.

There was an increase observed for BMC/FFM, with no change in BMC (Table 1) observed. BMC changes with aging and varies due to body mass (25). Our results and others (5) do not show a change in BMC with short-term weight loss. Therefore, taken together these findings for changes in TBW/FFM and BMC/FFM suggest that the change with D_b may be attributed to those that occurred with FFM and FM independent of BMC. This emphasizes the importance of using a 4C model when assessing significant changes in body composition due to weight reduction.

At present, 4C_HW is not a realistic method for obtaining estimates of changes in body composition beyond research. More often in a clinical setting only one method is used to detect changes in body composition. Results of our regression analyses (Table 4) demonstrate the relationship between other methods and 4C_HW. This enables users of the other body composition methods to predict a 4C_HW estimate of body composition using one of the comparison methods. Thus, all of the methods in the current study, with the exception of 3C_DXA for %BF and 2C_ADP for %BF, FM and FFM, may be used to obtain an estimate for changes in body composition for a postmenopausal woman.

In summary, the results of the present study indicate that 4C_ADP , 3C_DXA , 2C_HW, 2C_ADP , and 2C_D2O provide comparable estimates of body composition changes with 4C_HW on a group basis. However, the wide limits of agreement between 4C_HW (the criterion method) and the comparison methods indicate that these methods are not interchangeable on an individual basis. Thus, regression equations should be employed in order to achieve estimates of changes in body composition that are related to a 4C model. These regression equations are limited to overweight and obese Caucasian women in general good health and to energy-restriction induced weight loss whereby physical activity was maintained. The lack of standard regression equations for varying populations and subject characteristics makes it currently problematic to interpret the reliability of individual estimates of body composition for any given method. Furthermore, interpretation of results for individuals with a given method requires knowledge of the limitations of the method. Thus, larger studies are much needed to further validate our regression equations and to examine the reliability of these methods for assessing weight changes in older populations.

Abbreviations

Δ%BF

change in percent body fat

ΔFM

change in fat mass

ΔFFM

change in fat-free mass

HW

hydrostatic weighing

2C

2 compartment

ADP

air-displacement plethysmography

DXA

dual-energy x-ray absorptiometry

3C

3-compartment

TBW

total body water

BMC

bone mineral content

4C

4-compartment

BV

body volume

D₂O

deuterium oxide dilution

CHO

carbohydrate

PRO

protein

D_b

body density

BW

body weight

TBBM

total body bone mineral

BMI

body mass index