Comparative Analysis of Body Composition Results Obtained by Air Displacement Plethysmography (ADP) and Bioelectrical Impedance Analysis (BIA) in Adults

Abstract

Body composition assessment provides clinical and scientific information about health, including the metabolic risks associated with low or high body fat. The aim of the study was to (i) compare the agreement of the results of the body fat percentage from the air displacement plethysmography (ADP)—BOD POD and bioelectrical impedance analysis (BIA) method—InBody 770; (ii) compare the agreement of the classification of central obesity risk in young adult, healthy females and males using data from manual measurement of waist circumference (WC) and BIA analysis. The Bland–Altman plots were used to determine the clinical agreement between BIA, ADP, and various anthropometric measurements (ADP and anthropometry were utilized as reference techniques to compare variables estimated from BIA). Finally, 203 participants enrolled in this study. We found low agreement (Bland–Altman index: 6.4%) in body fat content (%BF) and Fat Mass Index between results from ADP and BIA methods. The InBody 770 analyzer revealed an underestimation in %BF for the total group and for males. The low agreement was also observed when comparing WC measured manually versus estimated by BIA, as well as with waist-to-hip ratio (WHR). Moreover, demonstrated systematic bias and/or proportionate bias between BIA and ADP indicate that these devices cannot be used interchangeably. WC should be measured manually, especially in females.

1. Introduction

Body composition assessment provides clinical and scientific information about health, including the metabolic risks associated with low or high body fat [1].

Currently, there are many methods for analyzing body composition. Historically, hydrostatic weighing was considered the gold standard for measuring body composition, but several limitations have led to air displacement plethysmography (ADP) now being the preferred method for determining fat mass (FM) and fat-free mass (FFM). Alternative methods for assessing body composition include the most popular and widely available—bioelectrical impedance (BIA), and very advanced imaging techniques such as dual-energy X-ray absorptiometry (DEXA), computed tomography (CT), and magnetic resonance imaging (MRI), which are very expensive and less frequently used [1].

However, the methods used in practice should not only be precise but also easy to use, practical, and suitable for a wide range of people. Over the last 20 or more years, ADP and BIA have gained popularity due to their ease of use and practical advantages [2].

BIA is a two-compartment model that uses low-frequency electrical currents of single or multiple frequencies to estimate whole-body or segmental-body composition. This method is based on differences in electrical conductivity and resistance throughout the human body, which allows for the distinction between two types of tissue (FM and FFM). FFM has high electrical conductivity, while FM has low electrical conductivity (high impedance) [3]. There are significant differences between BIA measurements performed using different BIA devices, depending also on the single- or multi-frequency analyzer used [4]. Multi-frequency bioelectrical impedance analysis (MF-BIA) technology, as opposed to single-frequency BIA (SF-BIA) (frequency of 50 kHz), uses different electrical frequencies (from 1 to 1000 kHz), which enables the estimate of extracellular water and intracellular water. The impedance is related to the volume of body water and is used to estimate FFM. FM and FFM are calculated from mathematical formulas using resistance, reactance, and impedance measurements. Pre-programmed equations estimate body fat percentage (%BF) and provide estimates of FM and FFM during evaluation [1,5]. Numerous prediction equations for quantifying body composition using BIA have been developed using high-standard procedures. However, the accuracy of the estimates is compromised by the lack of population-specific prediction equations [6].

Similarly, ADP is also a two-compartment model where body composition analysis is performed using the BOD POD system (COSMED USA, Inc., Concord, CA, USA), which involves measuring body volume by air displacement. At the same time, the individual is seated in the chamber and calculating the body density, which is used to estimate FM (% and kg) and FFM (% and kg) (based on the Siri equation) [1,7]. The accuracy of the BOD POD in determining body composition has been considered high, but some studies have shown that it overestimates body fat percentage in lean individuals [8,9,10]. In addition, testing conditions such as clothing worn during measurement and excessive facial or body hair may affect the accuracy of results [1].

BIA is a common method used to measure body composition in the general population because it is easily accessible, simple, and inexpensive [1]. However, the consistency of the results compared to other established techniques is important. In previous studies, the validity of BIA compared with ADP presented inconsistent results [11,12,13]. Results of research by Biaggi et al. [11] have shown that sometimes there are clear differences between the results of body fat content measured using BIA and ADP. Some studies have shown that the average body fat measured by BIA was less than ADP, and the limits of agreement were wide [11,12].

Modern BIA devices can also automatically estimate trunk circumferences, including waist circumference (WC) and waist-to-hip ratio (WHR). However, there is limited research validating BIA measurements against manual measurement values [14]. Simple anthropometric indicators such as WC, WHR, and waist-to-height ratio (WHtR) have been used to diagnose central obesity. Existing evidence suggests that central obesity is more strongly associated with cardiometabolic risk factors and chronic disease risk than overall obesity. Consequently, central obesity indices may provide a more accurate estimate of adiposity and may be more closely associated with mortality risk compared to body mass index (BMI) [15]. Current recommendations suggest confirming obesity either through direct measurement of %BF, if available, or via at least one anthropometric criterion (e.g., WC, WHR, or WHtR) alongside BMI, using validated methods and cut-off points tailored to age, sex, and ethnicity [16]. Additionally, analyzing body composition raw data is enhanced by calculating fat mass index (FMI; as FM⁄height²) and the fat-free mass index (FFMI; as FFM⁄height²), which eliminate the differences in %BF related to body height [17]. However, these calculations require a comprehensive body composition assessment.

However, because body composition assessment methods are based on different principles, the results of studies are inconsistent, possibly due to the different BIA devices and study populations. Therefore, the objectives of the study were: (i) to compare the agreement of the results of the body fat percentage from the air displacement plethysmography (ADP)—BOD POD and bioelectrical impedance analysis (BIA) method—InBody 770; and (ii) to compare the agreement of the central obesity risk classification in young adult healthy females and males using data from manual WC measurement and BIA analysis. ADP and manual measurement of WC were utilized as reference techniques to compare variables estimated from BIA (i.e., body composition values, WC, and WHR).

2. Materials and Methods

2.1. Participants

Two hundred twenty-nine participants enrolled in this study. Recruitment commenced in April 2022 and concluded in June 2023. Participants were recruited through posts on social media and announcements at the university. Inclusion criteria were: healthy adults (excluded when reporting chronic diseases or taking any medications) who were able to comprehend and consent to the study; being aged between 18 and 40 years; having a BMI between 18.5 kg/m² and 29.99 kg/m²; and those reporting having a stable weight and were not actively attempting to lose or gain weight. Exclusion criteria included pregnant females and those with standard exclusions for BIA and/or ADP, such as those with implanted defibrillator devices or prostheses. The final data included 203 participants (76 males and 127 females); 26 participants were excluded because of inappropriate BMI.

All participants gave their full informed consent for inclusion before participating in the study and confirmed the lack of contraindications and proper preparation for the performance of the analyses. This study was conducted according to the guidelines of the Declaration of Helsinki and was approved by the Ethics Committee for Scientific Research Involving Humans at the Institute of Human Nutrition Sciences, Warsaw University of Life Sciences, Poland, on 20 December 2021 (Resolution No. 52/2021).

2.2. Anthropometric Measurements

Manual anthropometric measurements included height, WC, and hip circumference (HC). Height (H) was measured with a portable stadiometer with the head in the horizontal Frankfurt plane and recorded with a precision of 0.1 cm (SECA 220, Hamburg, Germany). Anatomical WC was measured with a stretch-resistant tape that provides constant 100 g tension (SECA 201, Hamburg, Germany) while standing, in a horizontal plane, midway between the inferior margin of the ribs and the superior border of the iliac crest, to the nearest 0.5 cm [18,19]. HC was measured at the widest part of the hip bones of the buttock’s region in centimeters to the nearest 0.5 cm. All measurements were repeated twice. To minimize inter-observer and inter-device variability, all measurements were performed under strictly standardized conditions (room temperature of 22 °C, air humidity of 45%) by well-trained researchers using the same device. If a difference of more than 1% existed between the first two measurements, a third measurement was taken, and the averages were calculated to exclude the value with the largest difference from the three measurements and calculate the mean using the two remaining measurements. The intra-tester technical errors of measurement (TEM) were 0.5%.

Anthropometric indices were calculated based on anthropometric measures. WHR was calculated as WC (cm) divided by HC (cm), WHtR was calculated as WC (cm)/H (cm), and BMI was calculated as body weight (kg) divided by H (m²) [20,21,22,23].

According to the criteria of the International Diabetes Federation (IDF), the occurrence of central obesity in adults in the European population is diagnosed in the case of WC: ≥94 cm for males and ≥80 cm for females [20]. WHR was categorized as ≥0.90 for males and ≥0.85 for females, while WHtR was considered 0.5 for both sexes [20,21]. Cut-off points for BMI are defined by the World Health Organization, with normal weight classified as a BMI between 18.5 and 24.99 kg/m² [23].

2.3. Bioelectrical Impedance Analysis (BIA)

In this study, a multi-frequency bioimpedance body composition analyzer, InBody 770 (Inbody Co., Ltd., Seoul, Republic of Korea) with Lookin’Body 120, version 3.0.0.11 software, was used. The BIA device measures at six frequencies, ranging from 1 kHz to 1000 kHz. The BIA examination was performed under standardized conditions, according to the manufacturer’s protocol, following the BIA measurement procedures (refrain from intense physical exercise for at least 12 h before the examination; no caffeine or alcohol consumption for 24 h before the examination; avoid certain medications, such as diuretics, laxatives, and electrolytes; fasting or 4 h after a meal; bladder emptying 30 min before the examination and remove all metal jewelry). Measurements were taken twice. Further analysis included the FM, %BF, WC, and WHR results.

Manual assessment was used as a reference method to assess the agreement between WC and WHR, as determined by the two methods.

2.4. Air Displacement Plethysmography (ADP)

BOD POD (COSMED USA, Inc., Concord, CA, USA) was used for ADP. System quality checks and scale calibrations were carried out on a daily basis in accordance with the manufacturer’s guidelines. TGV was predicted by the BOD POD system using OMNIA, version 2.3.0.6 software based on age, sex, and height.

Plethysmography determines body volume (Vraw) based on the pressure/volume relationship based on Boyle’s law (isothermal condition) and Poisson’s law (adiabatic condition). Body volume measurements in BOD POD take place primarily under adiabatic conditions. There is a certain volume of air maintained under isothermal conditions, as contained in the lungs, near the skin or hair, and in clothing [2,24]. The average amount of air in the lungs during normal tidal breathing, as well as the volume of thoracic gas (VTG), can be measured or predicted. Thoracic gas volume (TGV) can be quantified as an additional measure while in the chamber or estimated by the BOD POD. The effect of isothermal air near the skin surface is estimated by computing the area artifact (SAA). SAA is automatically calculated by the BOD POD software as follows:

SAA (L) = k (L/cm²) × BSA (cm²),

where k is a constant (derived empirically by the manufacturer; and BSA is the body surface area. BSA is calculated from the body mass and height using the DuBois formula. The SAA is typically 1.0 L for average-sized adults [2]. Then, the body volume is calculated using the formula:

V_B (L) = Vraw (L) − SAA (L) + 40% TGV (L),

where V_B is the body volume corrected for SAA and TGV. As part of the test procedure, the subject is also weighed; hence, the body density (D) can be calculated as M/V_B. Then, the %BF is estimated based on the Siri model or Brozek equation [24].

The ADP examination was performed under standardized conditions, according to the manufacturer’s protocol. Participants were instructed to wear form-fitting clothing, such as a swimsuit or single-layer compression shorts, and a single-layer sports bra for females. Hair was thoroughly compressed by a swim cap.

The measurement included two assessments of the participant’s raw body volume, but if these differed by more than 150 mL, the BOD POD software instructed to perform a third assessment and used the mean of the two closest values in subsequent computations. If no two measurements met the acceptance criteria, the entire trial was repeated. Repeated measurements in five healthy males and females had an intraclass correlation R of 0.98 and a standard deviation of 0.005 g/cm³ for body density. %BF and FFM were computed using the Siri equation using the BOD POD software. Further analysis included the FM and %BF.

Further analysis also included the FM results from ADP and BIA. On this basis, the FMI was also calculated as FM (kg) divided by H (m²) [17]. Moreover, to classify the participants according to their body fat content, the values proposed by Kelly et al. were used, presented in

Table 1. FMI classification according to Kelly et al. [25].

FMI Class [kg/m2]	Severe	Moderate	Mild	Normal	Excess Fat	Obese
		Fat Deficit				Class I	Class II	Class III
Males	<2	2 to <2.3	2.3 to <3	3–6	>6 to 9	>9 to 12	>12 to 15	>15
Females	<3.5	3.5 to <4	4 to <5	5–9	>9 to 13	>13 to 17	>17 to 21	>21

[25].

2.5. Statistical Analysis

The sample size for the study group was initially estimated to be 199 participants using G*Power, version 3.1.9.7 software. The calculations assumed a small effect size (0.2), two dependent groups, and the difference between two dependent means (matched pairs), a statistical power of 0.8, and an alpha significance level of 0.05. The final effect size of the results obtained in the study is 0.374, and the test’s statistical power is 0.8034.

The normality of the distribution of continuous variables and their differences was tested using the Shapiro–Wilk test. Descriptive characteristics are displayed as means ± the standard deviation (SD) for the total sample, both males and females. The mean values were compared with the Mann–Whitney test to describe the difference between females and males. If necessary, differences in the percentage distribution of the categorical variables were verified with the chi2 test with Yates’ correction. Differences in the mean result of WC, WHR, WHtR, %BF, and %FFM assessed or calculated from two methods were evaluated using the Wilcoxon matched pair test (unnormal distribution) or t-student for repeated measurements. The Pearson’s correlation coefficient (r) was also calculated. The r was evaluated according to the following thresholds: <0.1 (trivial), 0.1–0.3 (weak), 0.3–0.5 (moderate), 0.5–0.7 (strong), 0.7–0.9 (very strong) and 0.9–1.0 (almost perfect) [26].

As part of the accuracy evaluation, paired t-tests were performed to identify differences in BF%, WC, and WHR estimations between methods. Simple linear regression models were fitted, and the Coefficient of Determination (R²) and Standard Error of Estimates (SEE), Mean Absolute Percent Error (MAPE), and Root Mean Square Error (RMSE) were reported.

The agreement between the ADP and BIA for %BF, WC manual, and WC from BIA; WHR calculated, and WHR from BIA was evaluated by the Bland–Altman plots [27,28]. Bland–Altman plots showing systematic bias and the limits of agreement (LOA) for the two compared devices or measurements were created [28]. The Bland–Altman index (%) was calculated as the percentage of participants falling outside the LOA. LOA was calculated using ±2 SD from the mean value of the dependent variable. Agreement between methods was considered when the mean value of the dependent variable was not different from zero according to a paired t-test (p > 0.05). To ascertain whether this bias remained, regardless of the results (independent variable), the homogeneity of the dependent variable was assessed by simple linear regression analysis using the p-value of the beta (β) parameter. Thus, homogeneity was considered at a p-value > 0.05 of the β parameter [29]. Good clinical agreement was interpreted as the Bland–Altman index, defined a priori as the limits of maximum acceptable differences (expected limits of agreement), demonstrated when at least 95% of the differences were within ±2 SD, which corresponds to an index value of 5% [30,31]. From a clinical point of view and from the assumption of the Bland–Altman method, only the clinician who uses the test results can decide whether the mean deviation and LOA are acceptable or not [31]. The above value was set for the measurements performed. The agreement was visually assessed via Bland Altman plots.

To assess the differences and agreement between the methods, various statistical measures were applied. The Mean Absolute Deviation (MAD) was used as an indicator of the differences between measurements obtained by the two methods. The slope and intercept derived from the least squares regression analysis were utilized to evaluate the consistency of measurement trends between the methods, where an ideal agreement would be indicated by a slope of 1 and an intercept of 0. To assess the distribution of errors, the β (beta) value was calculated, representing systematic deviations between the methods.

The Lin’s Concordance Correlation Coefficient (CCC) was used to confirm the results of the agreement as follows: <0.8 is unacceptable, 0.81–0.89 is poor, 0.90–0.94 is moderate, 0.95–0.99 is substantial, and >0.99 is almost perfect agreement [32].

Moreover, ordinary least product regressions (OLP; Model II linear regression) were used to determine whether systematic and/or proportional bias existed between BIA and ADP. Systematic bias (consistent under/overestimations) was present when the 95% confidence interval of the intercept did not include 0. Proportional bias was present when the 95% confidence interval of the slope did not include 1.0. p-values below 0.05 were considered statistically significant in all analyses. The statistical analyses were performed with Statistica 13 (TIBCO Software Inc., Tulsa, OK, USA; StatSoft Polska, Cracow, Poland).

3. Results

3.1. Characteristics of the Participants

The descriptive characteristics of the study group are displayed in

Table 2. Characteristics of the participants.

	Total n = 203	Males n = 76	Females n = 127	p-Value
Age (years)	25.01 ± 4.57	26.21 ± 4.72	24.30 ± 4.34	<0.001
Anthropometrics
Height (m)	1.73 ± 0.09	1.81 ± 0.06	1.67 ± 0.06	<0.001
Weight (kg)	68.94 ± 12.56	80.44 ± 10.18	62.06 ± 7.98	<0.001
WCmanual (cm)	79.40 ± 9.13	86.13 ± 7.35	75.38 ± 7.61	<0.001
WCBIA (cm)	81.29 ± 8.07	85.25 ± 8.35	78.91 ± 6.91	<0.001
p-Value	<0.001	ns	<0.001
HCmanual (cm)	97.87 ± 6.07	99.32 ± 5.97	97.01 ± 5.98	0.008
Indices
BMI (kg/m2)	22.96 ± 2.56	24.37 ± 2.39	22.11 ± 2.28	<0.001
BMI normal weight (%)	168 (83)	50 (66)	118 (93)	<0.001
BMI overweight (%)	35 (17)	26 (34)	9 (7)
WHRmanual	0.81 ± 0.07	0.87 ± 0.06	0.78 ± 0.05	<0.001
WHRBIA	0.84 ± 0.05	0.85 ± 0.06	0.84 ± 0.04	ns
p-Value	<0.001	0.028	<0.001
WHtRmanual	0.46 ± 0.04	0.45 ± 0.04	0.47 ± 0.04	<0.001
WHtRWC-BIA	0.47 ± 0.04	0.47 ± 0.05	0.47 ± 0.04	ns
p-Value	<0.001	ns	<0.001
Body composition
%BFBIA (%)	21.68 ± 7.61	15.42 ± 5.80	25.43 ± 5.92	<0.001
%BFADP (%)	22.28 ± 8.00	16.66 ± 7.01	25.65 ± 6.54	<0.001
p-Value	0.015	0.002	ns
%FFMBIA (%)	78.32 ± 7.61	84.58 ± 5.80	74.57 ± 5.92	<0.001
%FFMADP (%)	77.72 ± 8.00	83.34 ± 7.01	74.35 ± 6.54	<0.001
p-Value	0.015	0.002	ns
VFA (cm3)	62.06 ± 27.40	52.30 ± 26.36	67.90 ± 26.42	<0.001

Mean and SD; WC, Waist circumference; HC, Hip circumference; BMI, Body mass index; WHR, Waist-to-hip ratio; WHtR, Waist-to-height ratio; %BF, Body fat percentage; BIA, Bioelectrical impedance analysis; ADP, Air displacement plethysmography; %FFM, Fat-free mass percentage; VFA, Visceral fat area; ns, not statistically significant.

. The average age of the group was 25 ± 5 years, and the average BMI was 22.96 ± 2.56 kg/m² with a range from 18.5 kg/m² to 29.8 kg/m². Of the total sample, 83% had a BMI in the normal weight range (n = 168), while only 66% of males had a normal body weight.

In anthropometric measurements, the mean difference between the manual measurement of WC in the group of females was approximately 3.53 cm lower than that calculated from BIA. In males, higher values, on average by 0.9 cm, were found for manual measurements. %BF estimated by both methods—ADP and BIA—was statistically significantly higher in females than in males, in contrast to other measurements and indices. In anthropometric measurements, the comparison of results for the entire group revealed statistically significant differences between the two methods for WC, WHR, and WHtR (p < 0.001 for all). Similarly, significant differences were observed in %BF (p = 0.015) and %FFM (p = 0.015). Among men, significant differences were observed in WHR (p = 0.028) and in %BF and %FFM (p = 0.002 for both). Among women, significant differences were found in WC, WHR, and WHtR measurements when considering data obtained from different methods (p < 0.001 for all).

3.2. Assessment of Accuracy and Agreement for %BF, WC, and WHR

Based on the regression analysis for %BF, the R² and error values (except MAPE) allow us to conclude that both methods accurately estimate %BF (

Table 3. Summary of accuracy %BF from ADP and BIA device; WC measured manually and by BIA; and calculated WHR and WHR estimated by BIA.

	Total n = 203				Males n = 76				Females n = 127
	R2	SEE	MAPE	RMSE	R2	SEE	MAPE	RMSE	R2	SEE	MAPE	RMSE
%BFADP	0.8	2.8	15	3.5	0.8	3.2	22	3.5	0.7	3.3	7.4	2.8
%BFBIA
WCmanual	0.6	5.4	6.1	6.2	0.6	5.3	5.0	5.5	0.5	5.8	6.7	6.5
WCBIA
WHRmanual	0.1	0.1	7.8	0.1	0.1	0.1	6.7	0.1	0.1	0.1	5.2	0.1
WHRBIA

R², Coefficient of determination; SEE, Standard error of estimate; MAPE, Mean absolute percent error; RMSE, Root mean square error; %BF, Body fat percentage; BIA, Bioelectrical impedance analysis; ADP, Air displacement plethysmography; WC, Waist circumference; WHR, Waist-to-hip ratio.

). A weaker correlation was observed for WC measured manually and by BIA, and error values (except MAPE) permit the inference that both methods were not accurate in estimating WC (Table 3). WHR calculated from WC and HC measured manually and estimated by BIA showed a very weak correlation, and there was no linear relationship between the values (Table 3).

%BF determined by the BIA method is statistically different from ADP for the total group and males. There was no significant bias for females, which is small and equal to 0.219 (

Table 4. Results of accuracy and agreement for %BF (ADP as reference), WC, and WHR (manual measurements as reference).

	MAD	Lower LOA	Upper LOA	Number of Individuals Beyond the LOA	Bland–Altman Index (%)	Intercept	Slope	Distribution of Errors (β Value)
%BF
Total	0.600	−6.222	7.422	190 out of 203	6.4	−0.5367	0.0517	0.113
Males	1.237	−5.295	7.769	71 out of 76	6.6	−1.9646	0.1996	0.372
Females	0.219	−6.688	7.125	120 out of 127	5.5	−2.5376	0.1079	0.183
%BF and BMI < 25	0.398	−6.289	7.086	157 out of 168	6.5	−1.2419	0.0751	0.159
%BF and BMI ≥ 25	1.569	−5.664	8.801	35 out of 35	0	2.1188	−0.0243	−0.062
WC
Total	−1.886	−13.400	9.627	190 out of 203	6.4	−13.0283	0.1387	0.191
Males	0.874	−9.911	11.660	73 out of 76	3.9	13.3724	0.1459	−0.195
Females	−3.537	−14.260	7.190	118 out of 127	7.1	−12.2034	0.1123	0.138
WHR
Total	−0.034	−0.176	0.108	191 out of 203	5.9	−0.3935	0.4346	0.291
Males	0.017	−0.117	0.152	73 out of 76	3.9	0.1032	−0.0997	−0.071
Females	−0.065	−0.173	0.044	119 out of 127	6.3	−0.199	0.1659	0.112

%BF, Body fat percentage; WC, Waist circumference; WHR, Waist-to-hip ratio; BMI, Body mass index; MAD, Mean absolute difference; LOA, Limits of agreement.

, Figure S1). The Bland–Altman plots revealed that InBody 770 underestimated of %BF in the total group, and in males, the bias is 0.600 and 1.237, respectively. In addition, the regression line indicates that bias for males and females was not distributed homogeneously (for men: β = 0.372, p ≤ 0.001; β = 0.183, p = 0.039). The mean difference in %BF for the ADP and BIA methods was 0.600 percentage points for the total group. LOA was from −6.222% to 7.422%, and the Bland–Altman index was 6.4%, which indicates low compliance. Results of the Bland–Altman index for females and males were 5.5% and 6.6%, respectively. Moreover, in the subgroup with normal body weight, the Bland–Altman index was 6.5%, indicating a low agreement between ADP and BIA in the assessment of %BF in participants with BMI values indicating normal body weight.

WC determined by the BIA method was statistically different from manual measurement for the total group and for females (for males, there are no statistical differences; the bias is small and equal to 0.874) (Table 4, Figure S2). The Bland–Altman plots revealed that InBody 770 overestimated WC for the total group, and for females, the bias is −1.886 and −3.537, respectively. In addition, the regression line indicates that bias for the total group was not distributed homogeneously (β = 0.192, p = 0.006) along the entire range of average WC values. Considering sex, the regression line indicates that bias for males and females was not distributed homogeneously (for men: β = 0.195, p = 0.091; β = 0.138, p = 0.121) along the entire range of average WC values. The mean WC determined by the BIA method in females was 78.9 cm and 85,3 cm in males, while by the manual method, it was 75.4 cm and 86.1 cm, respectively (Table 2). The mean difference in WC for the manual and BIA method was −1.886 cm. LOA was from −13.400 to 9.627 cm, and the Bland–Altman index was 6.4%. Insightful results of the Bland–Altman index for females and males were 7.1% and 3.9%, respectively.

In consequence, WHR estimated by the BIA method is statistically different from manual measurement for the total group, for males and females (Table 4, Figure S3). The Bland–Altman plots revealed that InBody 770 overestimated WHR for the total group, and for females, the bias is 0.034 and 0.065, respectively. The mean difference in WHR for the manual and BIA method was 0.034 cm. LOA was from −0.176 to 0.108 cm, and the Bland–Altman index was 5.9%. Results of the Bland–Altman index for females and males were 6.3% and 3.9%, respectively.

CCC analysis confirmed that the agreement for %BF and WC of the tested methods is poor for the total group of males and females (

Table 5. Results of CCC and OLP analysis.

Measure	CCC (L-CI, U-CI)	(Est. L-CI, Est. U-CI)	(Slope L-CI, Slope U-CI)
%BF
Total	0.89 (0.87, 0.92)	(1.20, 3.91)	(0.80, 0.92)
Males	0.85 (0.79, 0.89)	(1.63, 4.90)	(0.64, 0.82)
Females	0.84 (0.79, 0.88)	(3.56, 8.09)	(0.68, 0.85)
WC
Total	0.75 (0.70, 0.79)	(0.15, 16.4)	(0.78, 0.97)
Males	0.75 (0.66, 0.82)	(−4.12, 25.4)	(0.70, 1.04)
Females	0.64 (0.56, 0.71)	(2.10, 23.6)	(0.66, 0.93)
WHR
Total	0.24 (0.15, 0.33)	(0.58, 0.75)	(0.12, 0.32)
Males	0.33 (0.16, 0.49)	(0.33, 0.73)	(0.14, 0.60)
Females	0.15 (0.07, 0.22)	(0.52, 0.76)	(0.11, 0.41)

%BF, Body fat percentage (ADP (ref.) vs. BIA); WC, Waist circumference (manual measurement (ref.) vs. BIA); WHR, Waist-to-hip ratio (manual measurement (ref.) vs. BIA), CCC, Lin’s concordance correlation coefficient < 0.8 is unacceptable, 0.81–0.89 is poor, 0.90–0.94 is moderate, 0.95–0.99 is substantial, and >0.99 is almost perfect agreement; Est. L-CI, Est. U-CI, and ordinary least products regression (OLP) estimates (Est.) not including 0 indicate systematic bias; Slope L-CI, slope U-CI, and OLP slope not including 1 indicate proportionate bias (i.e., the error increases as values increase).

). Inbody 770 demonstrates systematic %BF (Estimates not including 0) and proportionate (Slope not including 1) bias according to OLP for the total group, males and females. For WC Inbody 770, demonstrate systematic (Estimates not including 0) and proportionate (Slope not including 1) bias according to OLP for the total group and females. According to CCC, agreement magnitudes with calculated WHR were poor for BIA estimation (Table 5). Inbody 770 demonstrates systematic and proportionate bias, according to OLP.

3.3. Assessment of Obesity Based on Anthropometric Measurements and Body Composition Results

Considering certain discrepancies in recognition of the agreement of the discussed devices/methods, it was checked how the obtained results could affect further diagnostics or classification of the nutritional status of the examined persons. The FMI was calculated based on the FM measurements from ADP and BIA. Although the difference between mean values by methods was not statistically significant (p > 0.05), as shown in

Table 6. The differences in the FMI value calculated from BIA and ADP results.

	Total n = 203	Males n = 76	Females n = 127
FMIADP	5.14 ± 2.16	4.08 ± 2.00	5.78 ± 2.01
FMIBIA	4.99 ± 1.98	3.81 ± 1.65	5.70 ± 1.83
p-Value	ns	ns	ns

FMI, Fat mass index; BIA, Bioelectrical impedance analysis; ADP, Air displacement plethysmography; ns, not statistically significant.

, the Bland–Altman results showed low agreement for FMI in both sexes (

Table 7. Results of accuracy and agreement for FMI.

FMI	MAD	Lower LOA	Upper LOA	Number of Individuals Beyond the LOA	Bland–Altman Index (%)	Intercept	Slope	Distribution of Errors (β Value)
Total	0.151	−1.559	1.861	190 out of 203	6.4	−0.3179	0.0925	0.215
Males	0.272	−1.404	1.947	69 out of 76	9.2	−0.5302	0.2032	0.424
Females	0.078	−1.642	1.799	120 out of 127	5.5	−0.5063	0.1019	0.217

FMI, Fat mass index; MAD, Mean absolute difference; LOA, Limits of agreement.

).

The mean difference in FMI, calculated using the results from BIA and ADP methods, was 0.151 (Table 7). LOA was from −1.559 to 1.861, and the Bland–Altman index was 6.4%. Results of the Bland–Altman index for females and males were 5.5% and 9.2%, respectively.

Most individuals were classified as having a normal FMI, with 50% and 57% based on ADP and BIA results, respectively. According to ADP, 39% had a fat deficit, and 11% had abnormal fat mass. In comparison, BIA classified 36% as having a fat deficit and 7% as having excess fat. The differences in fat deficit classification between ADP and BIA were statistically significant in the total group. Full results are shown in

Table 8. The differences in the classification of participants based on body fat content according to Kelly et al. criteria [25].

FMI Class [kg/m2]	Severe	Moderate	Mild Fat	Normal	Excess Fat		Obese
	Fat Deficit					Class I	Class II	Class III
	n (%)	n (%)	n (%)	n (%)	n (%)	n (%)	n (%)	n (%)
FMIADP
Total	16 (8)	20 (10)	43 (21)	102 (50)	20 (10)	2 (1)	0 (0)	0 (0)
Males	7 (9)	6 (8)	14 (18)	34 (45)	14 (18)	1 (1)	0 (0)	0 (0)
Females	9 (7)	14 (11)	29 (23)	68 (54)	6 (5)	1 (1)	0 (0)	0 (0)
FMIBIA
Total	14 (7)	22 (11)	37 (18)	115 (57)	15 (7)	0 (0)	0 (0)	0 (0)
Males	7 (9)	10 (13)	10 (13)	41 (54)	8 (11)	0 (0)	0 (0)	0 (0)
Females	7 (6)	12 (9)	27 (21)	74 (58)	7 (6)	0 (0)	0 (0)	0 (0)
p-Values
total	<0.001	<0.001	<0.001	ns	<0.001	ns	ns	ns
males	ns	ns	<0.001	ns	<0.001	ns	ns	ns
females	ns	<0.001	<0.001	ns	ns	ns	ns	ns

FMI, Fat mass index; BIA, bioelectrical impedance analysis; ADP, Air displacement plethysmography; ns, not statistically significant.

.

Although the mean values of WC, WHR, or WHtR did not indicate the presence of central obesity in the study group, individual classification according to the adopted criteria revealed as many as 23% and 30% of people with abnormal WC, determined based on manual measurements and BIA estimates, respectively. These differences were statistically significant for the total population, as well as for both females and males. Similar results were obtained for both WHR and WHtR, as shown in

Table 9. Percentages of participants (%) with central obesity (abnormal WC), abnormal WHR, and WHtR estimated by BIA and measured manually.

Central Obesity Risk	Total n = 203 (%)	Males n = 76 (%)	Females n = 127 (%)	p-Value
Abnormal WCmanual—	46 (23)	16 (21)	30 (24)	ns
Abnormal WCBIA—	62 (30)	15 (20)	47 (37)	0.010
p-Value	<0.001	<0.001	<0.001	-
Abnormal WHRmanual—	35 (17)	23 (30)	12 (9)	<0.001
Abnormal WHRBIA—	71 (35)	18 (24)	53 (42)	0.009
p-Value	<0.001	<0.001	<0.001	-
Abnormal WHtRmanual—	41 (20)	23 (30)	18 (14)	0.006
Abnormal WHtRBIA—	43 (21)	21 (28)	22 (17)	ns
p-Value	<0.001	<0.001	<0.001	-

WHR, Waist-to-hip ratio; WHtR, Waist-to-height ratio; BIA, Bioelectrical impedance analysis; ns, not statistically significant.

.

Moreover, knowing that for the estimation of WC by BIA, the content of body fat in the trunk and the area of visceral fat (VFA) is important, a correlation analysis was performed between VFA and WC measurements or estimations (

Table 10. Pearson correlation coefficient value (r) for VFA and WC measurements or estimations.

WC	VFA (BIA)
	Total	Males	Females
BIA	0.683	0.906	0.881
Manual	0.324	0.632	0.597

WC, Waist circumference; VFA, Visceral fat area.

). A strong correlation was found between WC from BIA and VFA calculated by BIA, (especially when divided by sex (r = 0.906 for males and 0.881 for females). For manual measurements, a positive but weaker correlation with VFA was found (r = 0.324).

4. Discussion

In our study, we compared results from ADP and anthropometry, which were utilized as reference techniques to compare variables estimated from BIA, and found low clinical agreement between them. The results of body fat measurements in young adults from ADP differ significantly from those of BIA (InBody 770). InBody 770 showed underestimation and did not reveal acceptable agreement with BOD POD measures of %BF, especially in men (%FM BIA results lower by 1.237%). This has further implications for the assessment of metabolic health when using FM or FFM data for further classification or calculations. A low clinical agreement, with the tendency for overestimation by InBody 770, was also observed for WC, as well as for WHR, especially in females (WC BIA results higher by 3.537). These discrepancies generate differences in the classification of participants with or without central obesity, which may result in an incorrect assessment of the risk of cardiometabolic diseases.

In line with previous findings, BIA %BF did not agree with ADP [11,12,13]. Opposite results were observed in the study by Hillier et al. [13], where ADP results were 3.1% lower than those obtained from BIA. It should be emphasized that different statistical methods and the interpretation of obtained data in assessing agreement, as employed by various authors, can lead to contradictory conclusions. Applying different criteria for the estimated measurement errors or the Bland–Altman index may result in conflicting findings regarding the interchangeability of methods used to estimate body composition. Therefore, in our study, we additionally employed the CCC and OLP regression to verify the validity of our conclusions. Adopting a Bland–Altman index threshold of 15% makes it easier to achieve agreement between the tested methods; however, for our values, remaining at a 5% threshold appears more appropriate. Differences in %BF exceeding 1% (in men) lead to incorrect classification of patients into groups with or without obesity. Similarly, for WC (or WHR), differences reaching 2 cm (in the total group) or 3.5 cm (in females) are significant enough to preclude considering the BIA method interchangeable with the manual method. These findings underline the importance of considering clinical concordance when evaluating the interchangeability of body composition assessment methods, as even minor discrepancies may have significant implications for clinical decision-making and patient classification.

These differences may also be due to the model of body composition measurement equipment used. Accuracy varies widely among BIA instruments [33]. Many variables can also distract the measurement, e.g., hydration status in the case of BIA or body position in the case of ADP [1,34]. Similar, though not identical, equipment (MF-BIA: InBody 720) was used by Sullivan et al. [35] to compare BIA and ADP body composition values in a homogeneous obese population (BMI > 30.0 kg/m²). They observed no differences in %BF between BIA and ADP. In addition, significant correlations were observed, accompanied by low SEE values. These findings are partially consistent with our results in the subgroup of overweight participants (BMI ≥ 25 kg/m²) compared to the normal-weight subgroup (Table 4).

These results may also be related to the BMI, the mean values of which are within the normal range for both sexes but do not necessarily reflect normal body composition. Our findings showed that less than 8% of the female group were classified as overweight based on BMI, while BIA results indicated central obesity in 37% of them, and only 6% had excess fat, according to FMI. Similarly inconsistent results were reported by Petřeková et al. [36], where 22.1% of females had a BMI > 24.9, yet WHR and %BF were abnormally high in 50.4% and 44.3% of them, respectively. This leads to different interpretations of the health risks associated with high body fat. Furthermore, the FMI, calculated based on fat mass, has been proposed as a more reliable marker than BMI or %BF for identifying metabolic syndrome [25]. Classification of excess body fat based on FMI did not align with BMI estimates in our sample. Additionally, discrepancies in classification between ADP and BIA across the group make accurate assessments challenging. However, the findings of the large study by Oliveira et al. [37] suggest that BMI performed similarly to body fat-based indicators in identifying cardiometabolic outcomes. These estimates require further research.

Currently, many different indices have been proposed for the diagnosis of central obesity, such as WC, WHR, and WHtR. Existing evidence suggests that central obesity is more strongly associated with cardiometabolic risk factors and chronic disease risk than overall obesity. As a result, indices of central obesity may be more accurate than BMI in estimating adiposity and could, therefore, be more closely linked to the risk of mortality [15].

WC is considered the best and most simple measure of visceral fat and may be the best indicator for predicting cardiovascular risk factors. Advances in BIA devices have made it possible to estimate WC using special algorithms. However, available studies and our results showed clear differences between BIA estimates and manual measurement [14,38,39,40]. The Bland–Altman analysis prepared by Tanaka et al. [14] on 597 elderly individuals (mean age 64.6 ± 10.1 years) showed that bias was negative in both females and males, suggesting underestimation using BIA compared with manual measurements. This contrasts our results, where BIA overestimated WC in the total and female groups, but this may be due to age discrepancies. Comparable findings were observed in the study by Qin et al. [41], where BIA-estimated WC values were significantly higher than manual measurements in a group of 1496 middle-aged individuals.

Similarly, the estimation of central obesity based on WHR was statistically different between manual measurements and BIA results in our study, leading to an overestimation of the results based on BIA in the total and female groups. WHR was also ‘measured’ by BIA in the study by Song et al. [38] on a cohort of 10,030 middle-aged adults. WHR calculated by BIA in this study was higher than measured manually, but these methods were not compared directly to each other. Kim et al. [42] noted that WHR measured by BIA was significantly higher compared to anthropometry. Abedi Yekta et al. [40] observed a weak correlation between the manual method and BIA-based estimates and recommended avoiding the BIA application for WHR calculation. Some studies also suggest that BIA calculations provide greater accuracy in estimating VFA when compared to abdominal CT scans, as opposed to manually measured WC [39,43]. Additionally, WHR measured by BIA may better represent abdominal visceral fat. However, further research is required to determine the exact cause of the discrepancies between manual and BIA-calculated body composition metrics.

The observed differences in agreement between the manual method and the BIA method for estimating WC or calculating WHR in males and females are undoubtedly due to differences in body composition and fat distribution. Females have proportionally more FM, and males have more FFM, which is biologically determined. This suggests that detailed imaging of adipose tissue distribution provides a more comprehensive assessment of metabolic risk associated with sex differences. [44,45].

Sulis et al. [46] highlighted that sex and obesity-related differences must be considered when evaluating body composition. Our findings also revealed discrepancies between body composition assessments using bioelectrical impedance analysis (BIA) and air displacement plethysmography (ADP), with systematic biases being more pronounced in males. Such discrepancies may be attributed to variations in the interaction between measurement methods and the unique tissue characteristics of different populations. Specifically, bone and soft tissue composition may influence the accuracy of body fat estimation differently across sexes, as certain methods may either overestimate or underestimate body fat percentage (%BF) depending on the characteristics of the individuals being measured. Consequently, it is essential to explore how complementary technologies can be integrated to enhance the precision of body composition assessments. Combining different measurement tools could minimize bias and offer more reliable clinical decision-making, particularly when addressing sex- and obesity-related differences. Moreover, the accumulation of excessive adipose tissue is associated with hormonal activity. Estrogen promotes the accumulation of subcutaneous fat, which is more common in women, while testosterone is linked to the accumulation of visceral fat in men [47].

On a methodological note, the observed strong positive correlation between VAT and WC estimated by BIA confirms the need to refine algorithms for these parameters.

The strength of our study includes the involvement of a relatively large sample size and the use of professional equipment and standardized measurement conditions. The limitations of the study can be related to the limited BMI range (18.5–29.99). To further improve the results of the study, a wider range of participants could be included with a higher BMI, above 29.99, to evaluate further how %BF results differ between measurement instruments. Secondly, there is a lack of additional body composition assessment techniques for comparison. For example, using DEXA would allow for a comparison with an established, highly accurate device. Similarly, including other BIA devices, given the variety available on the market, would have enhanced the study’s practical significance. It is important to emphasize that there are no direct references in the scientific literature to methods for estimating waist circumference by InBody devices. InBody does not measure waist circumference directly but probably estimates it based on personal data and body composition results using mathematical models and algorithms, but it is not available to users. Thirdly, we did not investigate the multifactorial causes of excessive body mass or central obesity. Fourth, the study population consisted of relatively young individuals under 40 years of age. Fifth, ADP measurement was performed with TGV prediction using the BOD POD software. Finally, body composition measured by BIA may be affected by hydration and fasting, and that measured by ADP by changes in body position during measurement. We made every effort to instruct participants on proper preparation for measurement.

5. Conclusions

Our findings showed poor agreement between ADP and BIA (InBody 770) in %BF for males and between manual measurements in WC and calculated from BIA. Moreover, the statistical analysis demonstrated systematic bias and/or proportionate bias between BIA and ADP, indicating that these devices cannot be used interchangeably for determining fat content in males and WC and WHR in females. Given the documented differences between methods in clinics and medical facilities where patient body composition assessment is part of treatment and where testing efficiency is key, it appears that the use of the InBody 770 may provide an alternative to ADP with attention paid to possible bias. Still, WC should be measured manually, especially in females. It is important to consider the individual characteristics of the subjects when assessing nutritional status, taking into account gender differences and various types of body fat distribution.