Journal Logo

SPECIAL COMMUNICATIONS: Methodological Advances

A Field-based Three-Compartment Model Derived from Ultrasonography and Bioimpedance for Estimating Body Composition Changes

TINSLEY, GRANT M.; RODRIGUEZ, CHRISTIAN; WHITE, SARAH J.; WILLIAMS, ABEGALE D.; STRATTON, MATTHEW T.; HARTY, PATRICK S.; SMITH, ROBERT W.; DELLINGER, JACOB R.; JOHNSON, BAYLOR A.

Author Information
Medicine & Science in Sports & Exercise: March 2021 - Volume 53 - Issue 3 - p 658-667
doi: 10.1249/MSS.0000000000002491

Abstract

The ability to accurately estimate body composition proves important in evaluations of health and disease. For instance, body fat percentage (BF%) is used as a risk factor for several chronic conditions such as obesity and eating disorders (1). In addition, optimal body composition plays a significant role in athletic performance (2,3). Valid and reliable assessments of body composition are crucial in sport as they may aid in selecting a practical performance-enhancing program, determine the effectiveness of an exercise or dietary intervention, or be used to monitor the health status of an athlete (4). Several approaches are used to estimate body composition, with the most common being a two-compartment (2C) model that separates the body into fat mass (FM) and fat-free mass (FFM). However, notable limitations of these 2C models exist as they assume constant density and hydration of FFM (4). In reality, FFM properties can vary substantially, revealing limitations of 2C models when these deviations occur (5–7). To improve the 2C approach, a three-compartment (3C) model that separates body mass (BM) into FM, total body water (TBW), and residual mass (consisting primarily of protein, minerals, and glycogen) can be implemented (1,8). Laboratory-based 3C (3CLAB) models use air displacement plethysmography (ADP) or hydrostatic weighing to estimate body density (Db). Although TBW is ideally estimated via dilution techniques, the time and cost burdens often preclude these assessments. As such, various bioimpedance-derived TBW estimates are often used (3). Although 3C models can be further expanded to four-compartment (4C) models by the incorporation of bone mineral content estimates from dual-energy x-ray absorptiometry (DXA), research suggests minimal variation between the 3C and 4C models (8,9). Thus, it is considered acceptable to validate simpler field methods using a 3CLAB model, particularly in situations in which bone mineral content is considered relatively stable (1,6,10).

Despite the well-known strengths of laboratory-based multicompartment models, the practicality of using methods such as ADP, DXA, and UWW is limited in many sports, applied research, and medical settings. Thus, more portable field methods with acceptable validity are needed. Relative to their laboratory counterparts, portable bioimpedance analyzers, handheld ultrasonography (US), and skinfolds (SF) are more practical options for body composition assessment in field settings. Although these techniques hold some utility individually, they can also be combined to create a field-based 3C model (3CFIELD). This is accomplished by obtaining an SF or a US-derived Db value and a portable bioelectrical impedance analysis (BIA)-derived TBW estimate (1,7,8). Few investigations have examined the validity of 3CFIELD to evaluate body composition. In the earliest report, Forslund et al. (7) evaluated the body composition of 22 healthy males using Db estimated from four-site SF and TBW estimated from single-frequency BIA and reported this field-based model to be accurate, as indicated by its strong correlation (r = 0.95) to a laboratory-based 4C model. Similarly, Esco et al. (1) and Nickerson et al. (8) demonstrated that SF and BIA provide valid estimates of BV and TBW, respectively, for use in a 3CFIELD model. These cross-sectional examinations lend preliminary support for the proposition that the 3CFIELD model is a valid alternative to laboratory-based multicompartment models for practitioners. However, to date, longitudinal examinations of 3CFIELD models are nonexistent. This is a notable limitation of the existing literature as quantifying changes in body composition over time is a primary component of evaluations in health, disease, and athletics. In addition, no studies to date have included US-based Db values in a 3CFIELD model. Although SF assessment is commonly used in field settings, limitations include difficulties determining the subcutaneous tissue boundaries via palpation, meaningful differences between calipers, and variation in the elasticity of skin and adipose tissue (11). Potential advantages of US are the ability to clearly visualize the subcutaneous tissue boundaries and a smaller influence of connective tissue properties (12,13).

The purpose of this study was to assess the agreement of a 3CFIELD model compared with a criterion 3CLAB model for tracking body composition changes during a supervised training intervention. Changes in FFM and FM were evaluated because of the importance of monitoring absolute values of these individual variables during exercise training and nutrition interventions. A secondary aim was to determine whether the 3CFIELD model demonstrated superior performance as compared with the two individual assessment techniques it contained (i.e., US and BIA). It was hypothesized that 3CFIELD would exhibit acceptable errors as compared with 3CLAB and would also demonstrate superior performance as compared with US and BIA individually.

METHODS

Overview

Resistance-trained males completed 6 wk of supervised resistance training (RT), performed 3 d·wk−1, in conjunction with daily ingestion of a high-calorie protein/carbohydrate supplement. All participants were encouraged to gain ≥0.45 kg·wk−1 and were weighed before each RT session to promote compliance with this goal. Body composition assessments took place at baseline (i.e., time 0 [T0]) and after the 6-wk training intervention (i.e., time 6 [T6]). Body composition changes detected by the 3CFIELD model, portable US, and portable BIA were compared with the criterion 3CLAB model. The experimental protocol was granted ethical approval by Texas Tech University’s Institutional Review Board, and all participants provided written informed consent before participation.

Participants

Individuals were potentially eligible for participation if they were between the ages of 18 and 40 yr, male, generally healthy, weight stable (defined as no change in BM >2.3 kg in the past 3 months), resistance trained (defined as performing resistance exercise 2–5 d·wk−1 for ≥6 months before screening), and willing to abstain from consumption of any supplement beyond a standard multivitamin or those provided as part of the study. Participants were ineligible if they reported previous administration of anabolic–androgenic steroids, had consumed creatine-containing dietary supplements within the past month, were taking medications that could reasonably affect study outcomes, had facial hair longer than ~1.3 cm (due to ADP assessments [14]), or had an implanted electrical device. Only males were recruited for the present study due to data indicating a desire for BM gain in nonoverweight university males as compared with a desire for BM loss in normal weight university females (15). Based on these data, successful recruitment of females desiring to adhere to the aggressive weight gain protocol was deemed unlikely. Interested individuals meeting these criteria completed the consent process before further evaluation of muscular performance. To begin the intervention, each participant was required to demonstrate a bench press one-repetition maximum (1RM) ≥1.0 BM and leg press 1RM of ≥2.0 BM. The participant flow chart is displayed in Figure 1.

F1
FIGURE 1:
Participant flow chart.

Intervention

Dietary program

Each participant was instructed to continue his typical dietary intake while consuming an additional high-calorie protein/carbohydrate supplement provided by the research personnel. After each training session, participants were supplied with a half-serving (~647.5 kcal, 5.5 g fat, 123.5 g carbohydrate, 26 g protein) of Super Mass Gainer™ (Dymatize Enterprises, LLC, Dallas, TX) to promote BM gain. In addition, participants were provided with containers of the supplement and instructed to ingest a half-serving on rest days. Each participant was instructed to achieve a minimum weekly BM increase of ≥0.45 kg (1 lb). To monitor general compliance with targeted BM gain, participants were weighed 3 d·wk−1 (before every RT session) in the laboratory. If participants did not achieve the desired BM gain, based on the weekly average, research personnel provided guidance to promote increased energy intake. Although the goal of the dietary program was simply to encourage a sufficient energy surplus to elicit BM gain, and macronutrient intake was not specifically prescribed, participants tracked their dietary intake via a multiple-pass, validated, automated 24-h dietary assessment tool (ASA24; National Institutes of Health, Bethesda, MD). This method estimated nutritional intake during the intervention to be (mean ± SD) 51.2 ± 19.2 kcal·kg−1·d−1, 6.1 ± 2.2 g·kg−1·d−1 of carbohydrate, 1.8 ± 0.8 g·kg−1·d−1 of fat, and 2.3 ± 0.7 g·kg−1·d−1 of protein.

RT program

Participants completed a 6-wk supervised RT program concurrent with the aforementioned dietary program. The RT program was constructed primarily to elicit hypertrophic adaptations. Each participant completed three RT sessions per week on Mondays, Wednesdays, and Fridays. All sessions were directly supervised by members of the research team who were certified as personal trainers or strength and conditioning specialists (CSCS). Trainers provided verbal coaching and encouragement to each participant throughout his training session. The weekly RT program included a lower-body session on day 1, an upper-body session on day 2, and a full-body session on day 3 (see Table, Supplemental Digital Content 1, Resistance training program, https://links.lww.com/MSS/C105). This allowed each major muscle group to be trained two times per week. RT sessions were completed by using free weights (barbells and dumbbells) and specific machines (hip sled, leg extension, and leg curl). As the goal of the RT program was to promote maximal FFM accretion, the majority of exercises primarily targeted larger muscle groups by way of multijoint movements. Exercise intensity was adjusted week to week based on repetitions in reserve (RIR) (16). This method incorporates an autoregulatory system, accounting for each individual’s lifting capabilities. Weeks 1 and 4 were set at an intensity of 2 RIR, weeks 2 and 5 were set at an intensity of 1 RIR, and weeks 3 and 6 were set at an intensity of 0 RIR. An RIR of 2 represents a set where 2 reps could theoretically be performed postset, whereas an RIR of 0 is taken to transient muscular failure. The load was adjusted as necessary to ensure that movements were performed in the specified repetition range at the relative intensity prescribed. In addition, workout logs were completed throughout the study to track relevant program variables.

Laboratory Assessments

Pre- and postintervention, participants reported to the laboratory after abstaining from eating, drinking, exercising, and using caffeine or nicotine for at least 8 h. Each participant was then interviewed to establish compliance to preassessment guidelines. For each assessment, participants were instructed to wear light athletic clothing and remove all metal and/or accessories before testing. Each participant provided a urine sample for assessment of urine-specific gravity (USG) with a digital refractometer (PA201X-093; Misco, Solon, OH). Urine-specific gravity was 1.0223 ± 0.0043 at T0 and 1.0238 ± 0.0051 at T6 (P = 0.21 via paired t-test). Each participant’s height was measured using a mechanical stadiometer (Seca 769, Hamburg, Germany). After the initial laboratory procedures, each participant was evaluated using ADP, bioimpedance spectroscopy (BIS), portable US, and single-frequency BIA. All equipment was calibrated as recommended by the device manufacturers each day before use.

For ADP (Bod Pod; Cosmed USA, Concord, CA), participants wore compression shorts and a swim cap, and assessments were conducted per manufacturer recommendations. Estimated thoracic gas volumes were used. Estimates of Db and FM were obtained and used in subsequent calculations.

For BIS (SFB7; ImpediMed, Carlsbad, CA), each participant remained supine for ~5 min immediately before assessment using manufacturer-recommended hand-to-foot electrode arrangement. The sites for adhesive electrodes were cleaned with alcohol wipes before placement of electrodes. Duplicate measurements were taken, and values were averaged. Assessments were reviewed for quality assurance through the visual inspection of Cole plots and were analyzed using the manufacturer-provided software (BioImp version 2.0.1). The BIS analyzer used in the present study implements 256 measurement frequencies to model the TBW content of the body. BIS was designated as the reference TBW method for the present investigation because of its use of Cole modeling and mixture theories, rather than regression equations, for TBW estimation (17,18). In addition, previous investigations have validated BIS for TBW estimation in active males. Specifically, as compared with TBW estimated by deuterium dilution, Gonçalves et al. (19) reported a total error (TE) of 1.5 L, and Kerr et al. (20) reported a TE of 2.6 L.

Immediately after BIS, participants remained supine for assessments via single-frequency BIA (RJL Quantum V; RJL Systems®, Clinton Township, MI). This analyzer implements an eight-point, bilateral, hand-to-foot electrode configuration (21). Before electrode placement, the attachment sites were cleaned with alcohol pads. Adhesive electrodes were then positioned according to manufacturer specifications on the dorsal surfaces of both hands and both feet. Assessments were performed in duplicate and averaged. Results were processed with the manufacturer’s software using their provided equation set (RJL BC Segmental version 1.1.2).

B-mode US estimates of subcutaneous adipose tissue thickness were obtained using a linear-array probe (Lumify L12-4; Philips Healthcare, Amsterdam, Netherlands) connected to an electronic Android tablet. After participants stood upright for ~5 min, transmission gel was applied to the seven measurement locations specified by Jackson and Pollock (22). Pressure applied by the transducer was intentionally minimized to avoid tissue compression, and a single image was taken at each location. Standard depth and gain were set per manufacturer recommendations and kept consistent for all measurements at a given site. A single trained technician performed all assessments for each participant, and additional personnel with expertise in US manually reviewed each image to ensure the appropriate designations of tissue interfaces. The trained technician demonstrated an intraclass correlation coefficient of 0.997 and a root-mean-square coefficient of variation of 4.0% for subcutaneous adipose tissue thickness estimates obtained on separate days. Raw single-layer tissue thicknesses were provided by the US software interface (MuscleSound®, Englewood, CO). These values were doubled to reflect the values obtained by SF (i.e., a double layer of subcutaneous tissue) and were subsequently used to calculate the sum of SF (SS) in millimeters and estimate Db using the seven-site equation of Jackson and Pollock (22):

SS=Σchest+axilla+triceps+subscapula+abdomen+suprilium+front thigh
Db=1.1120.00043499×SS+0.00000055×SS20.00028826×age

The US estimates of body composition were based on the Siri 2C model equation (23):

BF%=1004.95Db4.5

FM and FFM were estimated based on the BF% estimate and BM obtained from the calibrated scale associated with the ADP device.

3C Model Calculations

In addition to body composition estimates obtained by portable US and BIA, two 3C models were produced from data derived from the aforementioned assessments. Both models used the 3C equation of Siri et al. (23):

BF%=1002.118Db0.78TBW1.354

Db is expressed in grams per cubic centimeter, and TBW is expressed as a proportion of body mass.

The criterion 3CLAB model used ADP Db and BIS TBW, whereas the 3CFIELD model used US Db and BIA TBW. For both 3C models, FM and FFM were estimated based on the BF% estimate and BM obtained from the calibrated scale associated with the ADP device. The body composition estimates from ADP and BIS were not examined in the present analysis because of the inclusion of data obtained from these methods in the criterion model.

Statistical Analysis

Repeated-measures ANOVA was used to analyze FFM and FM estimates for the four assessment methods (3CLAB, 3CFIELD, US, and BIA) across time (T0 and T6). To generate the data set for the ANOVA procedures, multiple imputation with 100 iterations was performed using the mice package (24) in R to estimate missing data points and to preserve the full sample size. However, one very lean individual (BM, 55.5 kg; BMI, 17.4 kg·m−2; 3CLAB BF%, 4.0%; BIS BF%, 0.8% at baseline) who yielded negative BF% values on BIA (−1.1% at baseline; −1.2% postintervention) was excluded from the analysis, for a final data set of n = 25 (mean ± SD; age, 21.6 ± 2.6 yr; BM, 73.1 ± 10.2 kg; height, 178.5 ± 6.8 cm; BMI, 23.0 ± 2.8 kg·m−2; 3CLAB BF%, 15.0% ± 4.1%; baseline leg press 1RM relative to BM, 3.3 ± 0.8; baseline bench press 1RM relative to BM, 1.3 ± 0.2). ANOVA was performed using the afex package (25), and normality of residuals was assessed by visual examination of quantile–quantile plots supplemented by the Shapiro–Wilk test. When normality was violated, the BestNormalize package (26) was used to achieve a normal distribution. In the event of sphericity violations, Greenhouse–Geisser corrections were used. The data used for analysis, i.e., raw or transformed depending on whether the variable necessitated transformation, were examined for outliers using the rstatix package (27). No extreme outliers were present, and the outliers identified by the package were deemed to be real values based on body composition norms. Follow-up for significant effects was performed using pairwise comparisons with Tukey adjustment via the emmeans package (28).

Because of the component of individual-level error used in several of the validity metrics examined, the data set for evaluation of validity metrics contained only individuals who completed the entire study (i.e., only collected rather than imputed data were used). Therefore, a final sample of 18 individuals was used for estimation of validity metrics (Fig. 1; mean ± SD; age, 22.0 ± 2.8 yr; BM, 75.3 ± 10.8 kg; height, 179.8 ± 7.1 cm; BMI, 23.3 ± 2.8 kg·m−2; 3CLAB BF%, 15.2% ± 4.4%; baseline leg press 1RM relative to BM, 3.3 ± 0.8; baseline bench press 1RM relative to BM, 1.3 ± 0.2). The constant error (CE) was calculated as the mean difference between the 3CLAB criterion and each alternate method (i.e., alternate estimate minus 3CLAB estimate). Equivalence testing was used to evaluate whether each method demonstrated equivalence with the 3CLAB model based on a ±1.2-kg equivalence region, and the 90% confidence limits for the two one-sided t-tests were calculated (29,30). The equivalence region was selected based on the mean FM change in the present investigation, similar to a previous investigation (31). Equivalence testing was performed using the TOSTER software package (30). The Pearson product–moment correlation coefficient (r) between the 3CLAB body composition estimate and each alternate method was calculated, along with the associated 95% confidence limits and the coefficient of determination (R2). Ordinary least squares (OLS) and Deming regression were performed to determine whether the intercept and the slope of regression lines differed from the line of identity (i.e., the perfect linear relationship between methods with an intercept of 0 and a slope of 1). In contrast to OLS regression, which is commonly implemented in methodological investigations, Deming regression accounts for error in the measurement of both x and y variables and thus may be more appropriate when errors are present for both criterion and comparison methods. The 95% confidence limits for the intercept and slope were obtained, and the SEE (i.e., the residual SE) was obtained from OLS regression. For both regression analyses, the 3CLAB model was designated as the criterion variable (y), and the alternate model was designated as the predictor variable (x). The methods of Bland and Altman (32) were used alongside linear regression to assess the degree of proportional bias. The 95% limits of agreement (LOA) were calculated. The TE was calculated as follows:

TE=ΣBCALTBC3C_LAB2/n

where BCALT is the body composition estimate for the alternate method in question. All data were analyzed using R (version 3.6.1) (33). The primary packages used for estimation and presentation of validity metrics include psych (34), TOSTER (30), deming (35), DescTools (36), MBESS (37), and ggplot2 (38).

RESULTS

ANOVA

For FFM, ANOVA indicated a method–time interaction (P = 0.038 with sphericity assumed, P = 0.055 with sphericity correction, ηp2 = 0.11), along with main effects for method (P < 0.0001, ηp2 = 0.69) and time (P = 0.0001, ηp2 = 0.47). Pairwise comparisons indicated that 3CLAB FFM estimates (mean ± SD; T0, 62.0 ± 8.0 kg; T6, 66.0 ± 8.3 kg) differed from 3CFIELD (T0, 60.6 ± 7.7 kg, P = 0.0006; T6, 64.5 ± 7.6 kg, P = 0.008) and BIA (T0, 64.9 ± 8.0 kg, P < 0.0001; T6, 68.8 ± 7.0 kg, P < 0.0001) at both time points but did not differ from US at either time point (T0, 61.9 ± 6.9 kg, P = 1.0; T6, 65.1 ± 6.7 kg, P = 0.69). However, increases in FFM from T0 to T6 were observed for all methods: 3CLAB (4.0 ± 4.5 kg, P = 0.001), 3CFIELD (3.9 ± 4.2 kg, P = 0.0006), BIA (3.9 ± 4.2 kg, P = 0.01), and US (3.2 ± 4.3 kg, P = 0.009) (Fig. 2A).

F2
FIGURE 2:
ANOVA plots. A, Pairwise comparisons to follow up a method–time interaction for FFM changes indicated that criterion (i.e., 3CLAB) FFM estimates at T0 and T6 differed from 3CFIELD and BIA estimates, but not US. However, all methods detected an increase in FFM across time. B, Pairwise comparisons to follow up a method main effect indicated that 3CLAB FM estimates differed from 3CFIELD and BIA, but not US. However, follow-up for a time main effect indicated that all methods detected an increase in FM across time.

For FM, ANOVA indicated main effects for method (P < 0.0001, ηp2 = 0.69) and time (P = 0.004, ηp2 = 0.30), although no interaction was observed (P = 0.35, ηp2 = 0.04). Pairwise comparisons to follow up the method main effect indicated that 3CLAB FM estimates (T0, 11.1 ± 3.9 kg; T6, 12.4 ± 3.9 kg) differed from 3CFIELD (P = 0.002; T0, 12.5 ± 3.7 kg; T6, 13.9 ± 4.0 kg) and BIA (P < 0.0001; T0, 8.3 ± 3.9 kg; T6, 9.7 ± 4.8 kg), but not US (P = 0.97; T0, 11.2 ± 4.8 kg; T6, 13.3 ± 5.7 kg). Follow-up for the time main effect indicated an increase in FM from T0 to T6 (P = 0.004; 3CLAB, 1.3 ± 2.2 kg; 3CFIELD, 1.4 ± 2.2 kg; BIA, 1.4 ± 2.9 kg; US, 2.1 ± 2.6 kg) (Fig. 2B).

Validity metrics

For ΔFFM and ΔFM, only 3CFIELD demonstrated equivalence with 3CLAB based on the ±1.2 kg equivalence intervals (Table 1). 3CFIELD also demonstrated the lowest TE (1.0 kg), with US and BIA producing comparable TE values of 1.6 to 1.8 kg. Similarly, the SEE was lower for 3CFIELD (ΔFFM, 1.0 kg; ΔFM, 0.9 kg) as compared with US (ΔFFM, 1.6 kg; ΔFM, 1.3 kg) and BIA (ΔFFM, 1.5 kg; ΔFM, 1.0 kg). CE values ranged from −0.2 to −0.7 kg for ΔFFM and from 0.2 to 0.7 kg for ΔFM. 3CLAB values were correlated with changes detected by all other methods, although the strength of the relationships varied among methods. Bland–Altman analysis indicated that no proportional bias was present between 3CLAB and any other assessment method, except BIA ΔFM (Figs. 3 and 4). LOA ranged from 2.1 to 3.3 kg.

TABLE 1 - Body composition changes.
Descriptive Data a 90% TOST Correlation
Variable Model Mean SD Min Max CE SD (for CE) Equivalence LL UL TE r LL UL R 2
ΔFFM (kg) 3CLAB 3.1 1.8 0.4 6.2
3CFIELD 2.9 1.8 0.0 5.4 −0.2 1.0 Y −0.6 0.2 1.0 0.83 0.60 0.94 0.69
US 2.4 1.7 −0.3 4.9 −0.7 1.7 N −1.4 0.0 1.8 0.55 0.11 0.81 0.30
BIA 2.5 1.7 −0.3 5.2 −0.6 1.6 N −1.3 0.0 1.6 0.61 0.19 0.84 0.37
ΔFM (kg) 3CLAB 0.8 1.4 −1.2 4.0
3CFIELD 1.0 1.7 −1.9 4.3 0.2 1.0 Y −0.2 0.6 1.0 0.79 0.50 0.92 0.62
US 1.5 1.8 −1.3 6.2 0.7 1.7 N 0.0 1.4 1.8 0.49 0.02 0.78 0.24
BIA 1.4 2.2 −1.6 8.7 0.6 1.6 N 0.0 1.3 1.6 0.72 0.38 0.89 0.51
aDisplayed data are from the n = 18 participants with complete data. Data from the imputed data set (n = 25) were used for ANOVA procedures and are displayed in article text and Figure 2.
LL, lower limit; N, no (not equivalent based on ±1.2 kg equivalence interval); R2, coefficient of determination; TOST, two one-sided t-tests; UL, upper limit; Y, yes (equivalent based on ±1.2 kg equivalence interval).

F3
FIGURE 3:
FFM changes. Panels A–C depict ordinary least squares (OLS regression lines (dashed) and Deming regression lines (solid) as compared with the line of identity (dotted). Deming regression accounts for error in the measurement of both x and y variables. A, 3CFIELD ΔFFM: SEE = 1.0 kg; OLS: y = 0.67 + 0.84x; Deming: y = 0.17 + 1.01x; neither OLS nor Deming lines significantly differed from line of identity. B, US ΔFFM: SEE = 1.6 kg; OLS: y = 1.70 + 0.59x; Deming: y = 0.35 + 1.15x; only OLS intercept significantly differed from line of identity. C, BIA ΔFFM: SEE = 1.5 kg; OLS: y = 1.52 + 0.64x; Deming: y = 0.36 + 1.11x; only the OLS intercept significantly differed from line of identity. Panels D–F depict Bland–Altman analysis, with the solid diagonal line representing the relationship between the difference in body composition estimates, calculated as the comparison method estimate minus the 3CLAB estimate, and the average of comparison and 3CLAB estimates. The shaded regions around the diagonal line indicate the 95% confidence intervals for linear regression lines, the horizontal dashed lines indicate the upper and lower LOA, and the horizontal solid line indicates the CE between methods. Linear regression equations and 95% LOA values are displayed. For the Bland–Altman analysis, the slope of the linear regression line did not differ from zero for 3CFIELD (A), US (B), or BIA (C), indicating no proportional bias for any method.
F4
FIGURE 4:
FM changes. Panels A–C depict OLS regression lines (dashed) and Deming regression lines (solid) as compared with the line of identity (dotted). Deming regression accounts for error in the measurement of both x and y variables. A, 3CFIELD ΔFM: SEE = 0.9 kg; OLS: y = 0.15 + 0.66x; Deming: y = 0.01 + 0.80x only the OLS slope significantly differed from the line of identity. B, US ΔFFM: SEE = 1.3 kg; OLS: y = 0.30 + 0.38x; Deming: y = −0.12 + 0.61x; only the OLS slope significantly differed from the line of identity. C, BIA ΔFFM: SEE = 1.0 kg; OLS: y = 0.15 + 0.45x; Deming: y = 0.03 + 0.54x; only the OLS slope significantly differed from the line of identity. Panels D–F depict Bland–Altman analysis, with the solid diagonal line representing the relationship between the difference in body composition estimates, calculated as the comparison method estimate minus the 3CLAB estimate, and the average of comparison and 3CLAB estimates. The shaded regions around the diagonal line indicate the 95% confidence intervals for linear regression lines, the horizontal dashed lines indicate the upper and lower LOA, and the horizontal solid line indicates the CE between methods. Linear regression equations and 95% LOA values are displayed. For the Bland–Altman analysis, the slope of the linear regression line did not differ from zero for 3CFIELD (A) or US (B), indicating no proportional bias. However, the slope differed from zero for BIA (C), indicating the presence of proportional bias.

DISCUSSION

The purpose of this study was to assess the agreement between a 3CFIELD model and a criterion 3CLAB model for tracking body composition changes in response to a supervised training intervention. A secondary aim was to determine whether combining the US and the BIA techniques within the 3CFIELD model improved performance as compared with either of the techniques individually. A major finding was that only the 3CFIELD model demonstrated equivalence with the 3CLAB model; in addition, this model exhibited the lowest TE, SEE, CE, 95% LOA and the strongest correlation of the examined methods. As such, our hypotheses that the 3CFIELD model would exhibit acceptable agreement with 3CLAB and that the model would represent an improvement over US and BIA individually are supported. The findings of the current study suggest that, in the context of increases in FFM and FM, a 3CFIELD model, including Db estimated from portable US and TBW estimated from portable BIA, demonstrates relatively minimal differences for tracking body composition changes over time as compared with a 3CLAB criterion using ADP Db and BIS TBW.

The present analysis advances previous findings regarding the validity of field-based multicompartment model body composition assessments. Forslund et al. (7) demonstrated in a sample of 22 males that field-based multicompartment models can improve cross-sectional body composition estimates as compared with simple 2C models by reducing the number of physiological assumptions required, particularly those related to stable FFM characteristics. These researchers reported that a 3CFIELD model using SF-derived Db and a single-frequency BIA-derived TBW were strongly correlated with a laboratory-based 4C model (r = 0.95), although several desirable components of validity analysis (Bland–Altman analysis, estimation of TE, etc.) were not included in the report. Notably, Forslund et al. (7) used the same BIA TBW estimate in the 3CFIELD model and the criterion 4C laboratory-based model; this overlap is likely responsible for a portion of the strong agreement between models. In the present investigation, although TBW estimates were also produced from bioimpedance technologies, the TBW estimates for the 3CLAB and 3CFIELD models were obtained using distinct analyzers to ensure no direct overlap of data between models. In a similar manner, Esco et al. (1) used BIS-derived TBW estimates in their 3CLAB model and BIA-derived TBW estimate for the 3CFIELD model. In fact, the specific analyzers used by Esco et al. are the same as used in the present investigation, although the BIA analyzer is one generation newer in the present study. Esco et al. (1) confirmed the superior cross-sectional validity of a 3CFIELD method over the traditional 2C techniques it contained (i.e., SF and BIA) when estimating BF% in healthy young males and females. Similar to the present analysis, the 3CFIELD model exhibited lower SEE (2.2%) and TE (2.2%), as well as a higher r value (r = 0.96), relative to SF (SEE = 3.1%, TE = 5.7%, r = 0.92) and BIA (SEE = 3.0%, TE = 4.3%, r = 0.92) individually, when compared with the criterion 3CLAB model. More recently, Nickerson et al. (8) examined the performance of a 3CFIELD model, including SF Db and BIA TBW, along with other 3C models using DXA-derived BV estimates. As compared with a laboratory-based 4C model, the 3CFIELD model produced acceptable validity and the lowest TE of all 3C models for each body composition variable (TE = 2.2 kg for FM and FFM; TE = 2.95% for BF%). Although these three studies provide evidence that a 3CFIELD model is a valid method to estimate body composition in a cross-sectional manner, none provided direct evidence regarding the ability of a 3CFIELD model to track body composition changes over time. Thus, a novel aspect of the present investigation is that it may be the first study to examine the longitudinal performance of a 3CFIELD model for tracking body composition.

The present study is also unique in that US-derived Db estimates were used rather than the SF-derived estimates in previous 3CFIELD research (1,7,8). SF is a common field method of body composition estimation, although limitations of this assessment have long been recognized. These include difficulties determining the boundaries of subcutaneous tissue via palpation, potentially noteworthy variation between assessors and calipers, variation in the elasticity of skin and adipose tissue, and potential for a nonconstant force over a nonconstant area of contact throughout the range of the caliper’s opening (11). Some concerns of SF assessments may be particularly problematic in specific populations, such as the elderly and individuals with obesity (12). Potential advantages of US are the ability to clearly visualize the subcutaneous adipose tissue boundaries, minimal tissue compression when performed properly, and a lesser influence of adipose tissue thickness and connective tissue properties as compared with SF (12,13). However, limitations of US technology also exist, including a higher cost than most field methods, the need for an experienced technician, and outstanding questions regarding procedural standardization (12). Regarding the comparative agreement of these techniques, an early investigation by Fanelli et al. (39) demonstrated that subcutaneous adipose tissue thicknesses estimated by SF and US are equally effective in predicting Db estimated by hydrostatic weighing in lean males. The developed Db prediction equations for each method exhibited similar correlation coefficients, but the best prediction equation for the US technique (r = 0.809, SEE = 0.0078 g·mL−1) was slightly superior to the best prediction equation for the SF technique (r = 0.779, SEE = 0.0083 g·mL−1). Another trial from the same laboratory, conducted in males and females with obesity, indicated that the optimal prediction equation for the US technique (r = 0.819, SEE = 0.0095 g·mL−1) was superior to that obtained via the SF technique (r = 0.690, SEE = 0.0125 g·mL−1) (40). These investigations also highlighted the ability of US to overcome the compression effect of the SF technique, which is known to diminish Db prediction accuracy (39,40). Collectively, these and other findings indicate the utility of US to estimate Db, along with an accuracy that is comparable to, or greater than, the SF technique (12).

The present investigation examined body composition changes in the context of a supervised, progressive RT program. The RT regimen included twice-weekly training of all major muscle groups through dedicated lower-body, upper-body, and full-body sessions, as well as incorporating progression based on RIR (see Table, Supplemental Digital Content 1, Resistance training program, https://links.lww.com/MSS/C105) (16). Although this specific RT program likely differed from the habitual training programs of participants, the previous RT experience of participants indicates the likelihood that initial neural adaptations to training had previously occurred, thereby promoting muscular adaptations in response to the administered RT program (41,42). It has been posited that true skeletal muscle hypertrophy may not be detectable until 5 to 6 wk of RT (43), similar to the duration of the present investigation. However, although the specific energy surplus required to promote optimal muscular hypertrophy is still unknown, it is recognized that a positive energy balance provides an anabolic stimulus and can increase FFM, particularly when combined with RT (44). As such, the present protocol of an energy surplus in combination with RT represents a scenario in which true body composition changes may be detectable in a relatively short time frame. Indeed, notable increases in both FFM and FM were observed. Although the RT stimulus certainly promoted FFM accretion, the additional gain of FM was likely due to the purposeful overfeeding and relatively rapid prescribed rate of BM gain. The most direct applications of these findings are for those monitoring active individuals or athletes in a similar context (i.e., an energy surplus in combination with an RT stimulus to promote BM and FFM accretion).

There are both strengths and limitations of the present investigation. The meticulous body composition assessment methodology and the procedural standardization likely improved the quality of collected data. One notable aspect of this standardization was that the same trained technician, with previously demonstrated reliability, performed all US assessments. Furthermore, additional personnel with expertise in US manually reviewed each image to ensure the appropriate designations of tissue interfaces. The RT program was fully supervised by certified personal trainers or strength and conditioning specialists, who helped ensure an adequate training stimulus was provided. In addition, frequent monitoring of BM, rather than reliance on self-reported dietary records, helped ensure a sufficient energy surplus was present to elicit weight gain. Several limitations should also be noted. The usage of BIS rather than a dilution technique is a potential limitation of the TBW assessments, although previous investigations have validated both BIA and BIS for TBW estimation in groups of healthy adults (19,20,45). The selection of BIS for use in the criterion model in the present investigation was based on its estimation of TBW via Cole modeling and mixture theories rather than regression equations used by BIA (6,17). Although limitations to using bioimpedance-based TBW estimates should be noted, the usage of dilution techniques for TBW estimation is not common in applied research or field settings. Importantly, using bioimpedance-based TBW estimates in a 3C model has been demonstrated as superior to the alternative of using a 2C model that assumes constant FFM hydration (20). An additional limitation of the present study is the use of only resistance-trained males. The decision to include only males was in part due to previous data demonstrating a greater desire for weight gain in males as compared with females (15), indicating the likelihood of recruitment difficulties for female participants due to the prescribed weight gain protocol. In addition, time and resource limitations were a consideration. The relatively short duration of the present study could also be considered a limitation, although sufficient body composition changes occurred during this period to allow for the present analysis. Although our study used rigorous preassessment standardization procedures, as with many studies, we relied on aspects of self-reported adherence with these procedures. Finally, the sample size for the present investigation was relatively small and was largely dictated by time and resource constraints.

In conclusion, the present study demonstrated the potential usefulness of a 3CFIELD model incorporating US Db and BIA TBW as compared with a 3CLAB model consisting of ADP Db and BIS TBW for tracking body composition changes over time. As such, the 3CFIELD model may be suitable for implementation in field settings or even in laboratory settings when equipment for traditional multicompartment models is unavailable. In addition, this investigation demonstrated that combining US and BIA techniques within the 3CFIELD model improved body composition estimates as compared with either individual technique. This further demonstrates the importance of implementing multicompartment models rather than simple 2C models for improved body composition monitoring. Future research should examine the ability of 3CFIELD models to evaluate body composition changes in other contexts (e.g., weight loss) as well as in other populations (i.e., females, untrained or sedentary individuals, individuals with obesity, and high-level athletes). Ultimately, this line of inquiry may allow for enhanced accuracy of body composition monitoring in settings with resource limitations.

No funding was received for the present investigation. The bioelectrical impedance analyzer was donated to our laboratory by the manufacturer (RJL Systems®), and the ultrasound transducer and tablet were loaned to our laboratory by the manufacturer of the associated analysis software (MuscleSound®). The dietary supplements used in the present study were donated by the manufacturer (Dymatize Enterprises, LLC). None of these entities influenced any aspect of the present study or manuscript.

The authors declare no conflicts of interest. The results of the present study do not constitute endorsement by the American College of Sports Medicine. The results of this study are presented clearly, honestly, and without fabrication, falsification, or inappropriate data manipulation.

REFERENCES

1. Esco MR, Nickerson BS, Fedewa MV, Moon JR, Snarr RL. A novel method of utilizing skinfolds and bioimpedance for determining body fat percentage via a field-based three-compartment model. Eur J Clin Nutr. 2018;72(10):1431–8.
2. Roelofs EJ, Smith-Ryan AE, Melvin MN, Wingfield HL, Trexler ET, Walker N. Muscle size, quality, and body composition: characteristics of division I cross-country runners. J Strength Cond Res. 2015;29(2):290–6.
3. Schubert MM, Seay RF, Spain KK, Clarke HE, Taylor JK. Reliability and validity of various laboratory methods of body composition assessment in young adults. Clin Physiol Funct Imaging. 2019;39(2):150–9.
4. Ackland TR, Lohman TG, Sundgot-Borgen J, et al. Current status of body composition assessment in sport. Sports Med. 2012;42(3):227–49.
5. Tinsley GM, Graybeal AJ, Moore ML, Nickerson BS. Fat-free mass characteristics of muscular physique athletes. Med Sci Sports Exerc. 2019;51(1):193–201.
6. Moon JR. Body composition in athletes and sports nutrition: an examination of the bioimpedance analysis technique. Eur J Clin Nutr. 2013;67(1 Suppl):S54–9.
7. Forslund AH, Johansson AG, Sjödin A, Bryding G, Ljunghall S, Hambraeus L. Evaluation of modified multicompartment models to calculate body composition in healthy males. Am J Clin Nutr. 1996;63(6):856–62.
8. Nickerson BS, Tinsley GM, Esco MR. Validity of field and laboratory three-compartment models in healthy adults. Med Sci Sports Exerc. 2019;51(5):1032–9.
9. Nickerson BS, Tinsley GM. Utilization of BIA-derived bone mineral estimates exerts minimal impact on body fat estimates via multicompartment models in physically active adults. J Clin Densitom. 2018;21(4):541–9.
10. Wang ZM, Deurenberg P, Guo SS, et al. Six-compartment body composition model: inter-method comparisons of total body fat measurement. Int J Obes Relat Metab Disord. 1998;22(4):329–37.
11. Booth RA, Goddard BA, Paton A. Measurement of fat thickness in man: a comparison of ultrasound, Harpenden calipers and electrical conductivity. Br J Nutr. 1966;20(4):719–25.
12. Wagner DR. Ultrasound as a tool to assess body fat. J Obes. 2013;2013:280713.
13. Hyde PN, Kendall KL, Fairman CM, Coker NA, Yarbrough ME, Rossi SJ. Use of B-mode ultrasound as a body fat estimate in collegiate football players. J Strength Cond Res. 2016;30(12):3525–30.
14. Higgins PB, Fields DA, Hunter GR, Gower BA. Effect of scalp and facial hair on air displacement plethysmography estimates of percentage of body fat. Obes Res. 2001;9(5):326–30.
15. Neighbors LA, Sobal J. Prevalence and magnitude of body weight and shape dissatisfaction among university students. Eat Behav. 2007;8(4):429–39.
16. Zourdos MC, Klemp A, Dolan C, et al. Novel resistance training–specific rating of perceived exertion scale measuring repetitions in reserve. J Strength Cond Res. 2016;30(1):267–75.
17. Cole KS. Permeability and Impermeability of cell membranes for ions. Cold Spring Harb Symp Quant Biol. 1940;8:110–22.
18. Hanai T. Electrical Properties of Emulsions in Emulsion Science. London: Academic Press; 1968.
19. Gonçalves EM, Matias CN, Santos DA, Sardinha LB, Silva AM. Assessment of total body water and its compartments in elite judo athletes: comparison of bioelectrical impedance spectroscopy with dilution techniques. J Sports Sci. 2015;33(6):634–40.
20. Kerr A, Slater G, Byrne N, Chaseling J. Validation of bioelectrical impedance spectroscopy to measure total body water in resistance-trained males. Int J Sport Nutr Exerc Metab. 2015;25(5):494–503.
21. Moore ML, Benavides ML, Dellinger JR, Adamson BT, Tinsley GM. Segmental body composition evaluation by bioelectrical impedance analysis and dual-energy X-ray absorptiometry: quantifying agreement between methods. Clin Nutr. 2019;S0261-5614(19):33178–4.
22. Jackson AS, Pollock ML. Generalized equations for predicting body density of men. Br J Nutr. 1978;40(3):497–504.
23. Siri WE. Body composition from fluid spaces and density: analysis of methods. Techniques for measuring body composition. 1961;61:223–44.
24. Buuren Sv, Groothuis-Oudshoorn K. Mice: multivariate imputation by chained equations in R. Journal of Statistical Software. 2011;45(3):1–67.
25. Singmann H, Bolker B, Westfall J, Aust F, Ben-Shachar MS. afex: Analysis of Factorial Experiments; 2020.
26. Peterson RA, Cavanaugh JE. Ordered quantile normalization: a semiparametric transformation built for the cross-validation era. J Appl Stat. 2019;1–16.
27. Kassambara A. rstatix: Pipe-Friendly Framework for Basic Statistical Tests. 2020.
28. Lenth R. emmeans: Estimated Marginal Means, aka Least-Squares Means. 2020.
29. Dixon PM, Saint-Maurice PF, Kim Y, Hibbing P, Bai Y, Welk GJ. A primer on the use of equivalence testing for evaluating measurement agreement. Med Sci Sports Exerc. 2018;50(4):837–45.
30. Lakens D. Equivalence tests: a practical primer for t tests, correlations, and meta-analyses. Soc Psychol Pers Sci. 2017;1:1–8.
31. Tinsley GM, Moore ML. Body fat gain and loss differentially influence validity of dual-energy x-ray absorptiometry and multifrequency bioelectrical impedance analysis during simultaneous fat-free mass accretion. Nutr Res. 2020;75:44–55.
32. Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet. 1986;1:307–10.
33. Team RC. R: A Language and Environment for Statistical Computing. Vienna (Austria): R Foundation for Statistical Computing; 2019.
34. Revelle W. psych: Procedures for Psychological, Psychometric, and Personality Research. Evanston (IL): Northwestern University; 2019.
35. Therneau T. deming: Deming, Theil-Sen, Passing-Bablock and Total Least Squares Regression; 2018.
36. al ASem. DescTools: Tools for Descriptive Statistics; 2019.
37. Kelley K. MBESS: The MBESS R Package; 2019.
38. Wickham H. ggplot2: Elegant Graphics for Data Analysis. New York: Springer-Verlag; 2016.
39. Fanelli MT, Kuczmarski RJ. Ultrasound as an approach to assessing body composition. Am J Clin Nutr. 1984;39(5):703–9.
40. Kuczmarski RJ, Fanelli MT, Koch GG. Ultrasonic assessment of body composition in obese adults: overcoming the limitations of the skinfold caliper. Am J Clin Nutr. 1987;45(4):717–24.
41. Sale DG. Neural adaptation to resistance training. Med Sci Sports Exerc. 1988;20(5 Suppl):S135–45.
42. Moritani T, deVries HA. Neural factors versus hypertrophy in the time course of muscle strength gain. Am J Phys Med. 1979;58(3):115–30.
43. McGlory C, Devries MC, Phillips SM. Skeletal muscle and resistance exercise training; the role of protein synthesis in recovery and remodeling. J Appl Physiol (1985). 2017;122(3):541–8.
44. Slater GJ, Dieter BP, Marsh DJ, Helms ER, Shaw G, Iraki J. Is an energy surplus required to maximize skeletal muscle hypertrophy associated with resistance training. Front Nutr. 2019;6:131.
45. Haas V, Schütz T, Engeli S, Schröder C, Westerterp K, Boschmann M. Comparing single-frequency bioelectrical impedance analysis against deuterium dilution to assess total body water. Eur J Clin Nutr. 2012;66(9):994–7.
Keywords:

BODY COMPOSITION ASSESSMENT; ULTRASOUND; BIOELECTRICAL IMPEDANCE ANALYSIS; BODY DENSITY; WEIGHT GAIN; SKINFOLDS

Supplemental Digital Content

Copyright © 2020 by the American College of Sports Medicine