One preventable health condition is excess body fat, which may result in overweight or obesity, and is related to multiple chronic health diseases, such as metabolic syndrome, type II diabetes mellitus and CVD(Reference Roger, Go and Lloyd-Jones1). Healthy %BF ranges have been reported to be 10–22 % in males and 20–32 % in females at various ages(Reference Lohman2), with older individuals at the upper half of these %BF ranges. Because it is known that individuals who have %BF beyond healthy ranges are at risk of cardiometabolic disease, accurate assessment of %BF by age group is paramount in identifying at-risk individuals.
Historically, percent body fat (%BF) was estimated with reasonable accuracy by employing densitometry methods(Reference Fields, Goran and McCrory3), such as hydrodensitometry(Reference Brodie, Moscrip and Hutcheon4) and plethysmography(Reference Fields, Goran and McCrory3). %BF methods more accurate than densitometry methods include the three-compartment (3-C) model dual-energy X-ray absorptiometry (DEXA)(Reference Heymsfield, Smith and Wong5) and four-compartment (4-C) and five-compartment (5-C) models. The 3-C, 4-C and 5-C models are recognised as criterion reference methods in assessing %BF(Reference Pietrobelli, Heymsfield and Wang6).
Limitations to employing a criterion reference such as DEXA, hydrodensitometry, plethysmography and 4-C and 5-C models in clinical and health-care type settings include (1) high cost; (2) time intensive (∼30–60 min per assessment); (3) large space requirements and (4) specialised training, which in some cases certification is required. In contrast, recent advances in bioelectrical impedance devices, such as the Inbody 770 (IB770) (Inbody USA, Cerritos, California), have improved accuracy in estimating %BF and have demonstrated similar %BF relative to commonly recognised criterion reference methods, such as DEXA and hydrodensitometry(Reference Boykin, Tinsley and Harrison7–Reference McLester, Nickerson and Kliszczewicz11). It has been reported that compared with DEXA, IB770 underestimated %BF by only 1·2 % for men and 2·0 % for women(Reference Boykin, Tinsley and Harrison7–Reference McLester, Nickerson and Kliszczewicz11). These relatively small differences in %BF between DEXA and IB770 implies that IB770 may have appropriate accuracy in assessing %BF in clinical settings and provides a rationale for using IB770 as an acceptable criterion reference method. Moreover, IB770 offers several advantages for use in clinical settings: (1) minimal space (similar in size to traditional weigh scale); (2) specialised training or certification not needed and (3) and time efficiency, requiring ∼60 sec to assess %BF. The primary disadvantage of the IB770 is cost (starting at ∼$20 000 USD), although it is typically cheaper than DEXA (∼$50 000 USD).
In addition to IB770, skinfold prediction equations (SF-PE) can also be employed in clinical setting to calculate %BF, and concurrent validity has been established for SF-PE compared with criterion reference standards such as DEXA and hydrodensitometry(Reference Jackson and Pollock12,Reference Jackson, Pollock and Ward13) , with %BF typically within ±3–5 % of DEXA and hydrodensitometry. Advantageously, skinfold calipers are inexpensive, quick (∼60 sec needed to take measurements) and reliable and requires very little space and minimal specialised training. Although many studies have developed SF-PE to estimate %BF for specific populations, race, sex, age, athletic status and BMI(Reference Jackson and Pollock12–Reference Williams, Going and Lohman22), only a limited number of studies have compared multiple SF-PE to each other relative to a given criterion reference for specific populations.
Chambers et al. (Reference Chambers, Parise and McCrory23) used Caucasian older men and women with a mean age of 76·0, and compared %BF between a selected criterion reference (DEXA) and five difference SF-PE(Reference Jackson and Pollock12,Reference Jackson, Pollock and Ward13,Reference Durnin and Womersley15,Reference Gause-Nilsson and Dey16,Reference Visser, van den Heuvel and Deurenberg21,Reference Kwok, Woo and Lau24) . These authors reported %BF from four of the five SF-PE were significantly different than %BF from DEXA, while %BF from only 1 SF-PE were not significantly different than %BF from DEXA. Similarly, Silveira et al. (Reference Silveira, Barbosa and Noll25) used older 60–91 Brazilian males and females and compared %BF between DEXA and 4–5 difference SF-PE(Reference Durnin and Womersley15,Reference Lean, Han and Deurenberg18,Reference Visser, van den Heuvel and Deurenberg21,Reference Svendsen, Haarbo and Heitmann26) . These authors also reported only 1 of the SF-PE were not significantly different than %BF from DEXA for both older males and females. While comparing multiple SF-PE relative to a given criterion reference is important in assessing the accuracy of SF-PE, the results from these two studies were only applicable to older males and females, and were limited in the number of SF-PE they employed. Therefore, the purpose of the current study was to estimate %BF from commonly used SF-PE in Caucasian young, middle-age and older male and female adults and compare %BF from SF-PE to IB770, which was selected as the criterion reference for the current study. The hypothesis is that there will be no one SF-PE with similar (non-significantly difference) %BF compared to IB770 for all six age/sex groups, which included young males (YM), young females (YF), middle-age males (MM), middle-age females (MF), older males (OM) and older females (OF), but that each age/sex group will have one or more SF-PE with similar %BF to IB770.
Methods
Participants
Two hundred and two healthy males (n 96) and females (n 106) had their %BF assessed from 12 (males) to 13 (females) SF-PE and a selected criterion reference IB770. Participants were recruited by bulletin board announcements, posters, flyers, brochures and e-mail distributions within the California State University, Sacramento community. Although young, middle-age and older adult age groups vary and are somewhat arbitrary and not standardised in the literature, in the current study the young age group was defined as between 20–39 years old, the middle-age group was defined as 40–59 years old, and the older age group was defined as 60–80 years old.
Because %BF estimations are both sparse and have been shown to be less accurate compared with non-obese individuals, inclusion criteria included a non-obese adult population with a BMI of normal weight (18·5 kg/m2 ≤ BMI < 25 kg/m2) or ‘overweight’ (25 kg/m2 ≤ BMI < 30 kg/m2). Moreover, because race also affects the accuracy in using SF-PE, inclusion criteria for SF-PE chosen based on SF-PE developed using a similar population, race, sex, age, athletic status and BMI as in the current study. Like the current study, all SF-PE employed were comprised of primarily an adult Caucasian population of males and females with a similar age range, mean age and BMI. Because the current study was limited to a general adult Caucasian population, exclusion criteria were competitive athletes, anyone ≤ 19 years old, as well as underweight (BMI < 18·5 kg/m2) and obese (BMI ≥ 30 kg/m2) individuals. A flow chart of participant selection is shown in Figure 1.
Flow chart of participant selection.
This study was conducted according to the guidelines laid down in the Declaration of Helsinki, and all procedures involving human subjects were approved by the Institutional Review Board at California State University, Sacramento (protocol number AG2374). Written informed consent was obtained from all subjects.
Procedures
To help ensure consistent conditions, all participants were instructed to abstain from exercise, food intake, caloric beverages, caffeine and tobacco for 8 hours prior to testing (no alcohol or excess caffeine for 24 h)(Reference Heymsfield, Smith and Wong5,Reference Jackson and Pollock12,Reference Jackson, Pollock and Ward13,Reference Siri, Brozek and Henschel27) , but remain euhydrated. Subsequently, participants reported to the Human Performance Laboratory for body composition testing.
For IB770, procedures recommended from the manufacturer were used. Each subject stood on the footplate of the IB770 while grasping handles. Two electrodes were in contact with each forefoot, each hindfoot, each palm and each thumb. Each participant remained motionless and upright during testing. The IB770 estimated body composition across five segments (right leg, left leg, right arm, left arm and trunk) using six frequencies (1, 5, 50, 250, 500 and 1000 kHz). Body composition was determined using proprietary prediction algorithms built into the IB770 firmware. Total testing time on the IB770 was approximately 1 min.
For the skinfold measurements(Reference Heymsfield, Smith and Wong5,Reference Jackson and Pollock12,Reference Jackson, Pollock and Ward13,Reference Siri, Brozek and Henschel27) , Lange skinfold calipers (Beta Technology Inc.) calibrated with ± 1 mm accuracy were employed by a single trained experienced tester (approximately 32 years’ experience) who measured all participants. Standardised procedures for all skinfold measurements were employed and have been previously described(Reference Heymsfield, Smith and Wong5,Reference Jackson and Pollock12,Reference Jackson, Pollock and Ward13,Reference Siri, Brozek and Henschel27) . The eight skinfold sites assessed for both males and females were chest, midaxillary, triceps, subscapular, abdomen, suprailiac, thigh and biceps. Each skinfold site was measured for two trials, and recorded values were averaged(Reference Heymsfield, Smith and Wong5,Reference Jackson and Pollock12,Reference Jackson, Pollock and Ward13,Reference Siri, Brozek and Henschel27) . Total testing time to take all skinfold measurements was approximately 1–1·5 min. Subsequently, 2-site, 3-site, 4-site and 7-site skinfold measurements were summed. Waist (cm) circumference for males and females was also assessed and measured at the level of umbilicus. Participant mean (sd) age, mass, height, BMI, waist circumference and skinfold measurement data for males and females are shown in Tables 1 and 2.
Anthropometric characteristics (mean (sd)/range) for young males (YM), middle-aged males (MM) and older males (OM)
Skinfold (SF) measurement sites: Ch = chest; Ab = abdominal; Th = thigh; Sup = suprailiac; Tri = triceps; Sub = subscapular; Mid = midaxillary; Bic = biceps.
Anthropometric characteristics (mean (sd)/range) for young females (YF), middle-aged females (MF) and older females (OF)
Table 2. Long description
The table presents anthropometric characteristics and skinfold measurements for three age groups of females: young females (YF, 20-39 years), middle-aged females (MF, 40-59 years), and older females (OF, 60-80 years). The table has 23 rows and 15 columns. Column headers include Age (years), Body mass (kg), Body height (cm), BMI (kg/m^2), Waist (cm), and various skinfold measurement sites (Ch, Ab, Sup, Th, Tri, Sub, Mid, Bic) in millimeters. Additionally, there are columns for the sum of 7 skinfold sites, and sums of 3 skinfold sites in different combinations, and the sum of 2 skinfold sites. Each age group has data for mean, standard deviation (sd), and range. Row-wise data includes: Row 1: Age (years), YF Mean 27.4, sd 4.5, Range 22-36; MF Mean 47.9, sd 7.2, Range 40-59; OF Mean 72.0, sd 6.2, Range 60-80. Row 2: Body mass (kg), YF Mean 63.4, sd 8.5, Range 47-87; MF Mean 65.5, sd 8.5, Range 53-88; OF Mean 61.8, sd 9.1, Range 43-82. Row 3: Body height (cm), YF Mean 166.3, sd 7.7, Range 148-184; MF Mean 168.0, sd 7.0, Range 154-184; OF Mean 158.9, sd 5.4, Range 148-167. Row 4: BMI (kg/m^2), YF Mean 22.9, sd 2.5, Range 19-29; MF Mean 23.2, sd 2.1, Range 20-28; OF Mean 24.5, sd 3.5, Range 18-29. Row 5: Waist (cm), YF Mean 79.8, sd 6.5, Range 65-96; MF Mean 84.0, sd 8.0, Range 67-91; OF Mean 88.2, sd 9.0, Range 66-101. Row 6: Ch (mm), YF Mean 8.1, sd 3.6, Range 3-23; MF Mean 8.0, sd 3.5, Range 4-16; OF Mean 12.1, sd 4.3, Range 4-23. Row 7: Ab (mm), YF Mean 21.4, sd 7.0, Range 11-39; MF Mean 22.9, sd 6.5, Range 12-34; OF Mean 22.3, sd 6.8, Range 10-32. Row 8: Sup (mm), YF Mean 17.5, sd 8.0, Range 7-39; MF Mean 17.4, sd 6.8, Range 7-30; OF Mean 16.2, sd 6.3, Range 7-30. Row 9: Th (mm), YF Mean 21.9, sd 6.5, Range 14-38; MF Mean 23.6, sd 5.4, Range 15-33; OF Mean 26.3, sd 6.1, Range 8-35. Row 10: Tri (mm), YF Mean 16.0, sd 5.7, Range 7-33; MF Mean 18.7, sd 5.0, Range 10-30; OF Mean 19.1, sd 4.3, Range 12-28. Row 11: Sub (mm), YF Mean 13.4, sd 6.1, Range 7-29; MF Mean 14.4, sd 6.6, Range 6-31; OF Mean 15.8, sd 5.1, Range 6-27. Row 12: Mid (mm), YF Mean 12.2, sd 5.4, Range 5-23; MF Mean 11.5, sd 5.0, Range 6-24; OF Mean 13.1, sd 4.6, Range 5-21. Row 13: Bic (mm), YF Mean 9.3, sd 3.3, Range 4-19; MF Mean 10.9, sd 2.9, Range 6-17; OF Mean 11.9, sd 2.7, Range 7-17. Row 14: Sum of 7 SF (mm), YF Mean 108.4, sd 37.5, Range 58-198; MF Mean 116.6, sd 30.8, Range 64-177; OF Mean 124.8, sd 25.8, Range 71-175. Row 15: Sum of 3 SF (mm) (Th, Tri, Sup), YF Mean 55.4, sd 19.2, Range 30-105; MF Mean 59.7, sd 14.5, Range 34-93; OF Mean 61.5, sd 12.3, Range 37-85. Row 16: Sum of 3 SF (mm) (Tri, Ab, Sup), YF Mean 54.9, sd 19.8, Range 25-102; MF Mean 59.0, sd 15.8, Range 31-93; OF Mean 57.5, sd 13.6, Range 28-85. Row 17: Sum of 4 SF (mm) (Bic, Sup, Tri, Sub), YF Mean 56.2, sd 21.5, Range 26-118; MF Mean 61.5, sd 18.6, Range 33-105; OF Mean 63.0, sd 13.3, Range 34-93. Row 18: Sum of 2 SF (mm) (Bic, Tri), YF Mean 25.3, sd 9.0, Range 12-50; MF Mean 29.7, sd 7.9, Range 16-47; OF Mean 31.0, sd 7.0, Range 19-44.
Skinfold (SF) measurement sites: Ch = chest; Ab = abdominal; Th = thigh; Sup = suprailiac; Tri = triceps; Sub = subscapular; Mid = midaxillary; Bic = biceps.
Skinfold prediction equations employed
To estimate %BF, only SF-PE equations developed primarily from Caucasian adult male and female populations and for similar age and BMI ranges as in the current study were employed. All SF-PE are shown in Table 3, including the authors who developed each SF-PE, specific equations used for age and sex variations, specific skinfold sites, the number of skinfold sites used (1-site, 2-site, 3-site, 4-site or 7-site), the age range and sex employed in developing SF-PE and modified Siri equations used for different age and sex groups to estimate %BF(Reference Heymsfield, Smith and Wong5,Reference Jackson and Pollock12,Reference Jackson, Pollock and Ward13,Reference Siri, Brozek and Henschel27) . Variables other than sum of skinfold sites that were used in the SF-PE in Table 3 include age, sex, mass, height, waist circumference and BMI.
Regression equations for percent body fat (%BP) from skinfolds (SF) for all males (M), all females (F), young males (YM), middle-age males (MM), older males (OM), young females (YF), middle-age females (MF) and older females (OF)
Modified Siri equations(Reference Visser, van den Heuvel and Deurenberg21,Reference Williams, Going and Lohman22,Reference Moon, Eckerson and Tobkin28,Reference Escamilla, Yamashiro and Asuncion29) were employed as follows: %BF = ((4·95/Db) – 4·50) × 100 for YM and MM, %BF = ((4·97/Db) – 4·52) × 100 for OM, %BF = ((4·96/Db) – 4·51) × 100 for YF and MF and %BF = ((5·02/Db) – 4·57) × 100 for OF. Skinfold (SF) measurement sites: Ch = chest; Ab = abdominal; Th = thigh; Sup = suprailiac; Tri = triceps; Sub = subscapular; Mid = midaxillary; Bic = biceps.
Statistical analysis
Repeated measures 1-way ANOVA (P < 0·01) with post hoc tests for pairwise comparisons were employed for each of the three age groups for both males and females. To evaluate the agreement, systemic bias and proportional bias between IB770 and SF-PE, the Bland–Altman analysis was employed. Confidence limits of agreement were set at 95 %.
Results
Percent body fat between age groups and skinfold equation methods for males and females are shown in Tables 4 and 5. In Tables 4 and 5, all pairwise non-significant differences are superscripted next to %BF numerical values. The pairwise comparisons of the %BF values from SF-PE that were of primary interest were those with similar (i.e. not significantly different) %BF to IB770, which was the selected criterion reference.
Percent body fat (%BF) between age groups and skinfold equation methods for young males (YM), middle-aged males (MM) and older males (OM)
Skinfold (SF) measurements: Ch = chest; Ab = abdominal; Th = thigh; Sup = suprailiac; Tri = triceps; Sub = subscapular; Mid = midaxillary; Bic = biceps.
Non-significant differences are superscripted next to %BF numerical values in table according to schematic shown below. P < 0·01 = significant difference.
IB770 = not significantly different than Inbody 770 (Criterion Reference).
JP7 = not significantly different than Jackson & Pollock (1978), ⅀7 SF.
JP3a = not significantly different than Jackson & Pollock (1978), ⅀3 SF (Ch, Ab, Th).
JP3b = not significantly different than Jackson & Pollock (1985), ⅀3SF (Ch, Tri, Sub).
DW = not significantly different than Durnin & Womersley (1974).
GN = not significantly different than Gause-Nilsson et al. (2005).
P = not significantly different than Peterson et al. (2003).
V4 = not significantly different than Visser et al. (1994), ⅀4 SF.
V1 = not significantly different than Visser et al. (1994), ⅀1 SF.
C = not significantly different than Cicone et al. (2021).
W = not significantly different than Williams et al. (2003).
L4 = not significantly different than Lean et al. (1996), ⅀4 SF.
L1 = not significantly different than Lean et al. (1996), ⅀1 SF.
RO = not significantly different than Rojano-Ortega et al. (2024).
Percent body fat (%BF) between age groups and skinfold equation methods for young females (YF), middle-aged females (MF) and older females (OF)
Skinfold (SF) measurements: Ch = chest; Ab = abdominal; Th = thigh; Sup = suprailiac; Tri = triceps; Sub = subscapular; Mid = midaxillary; Bic = biceps.
Non-significant differences are superscripted next to %BF numerical values in Table according to schematic shown below. P < 0·01 = significant difference.
IB770 = not significantly different than Inbody (Criterion Reference).
JP7 = not significantly different than Jackson & Pollock (1980), ⅀7 SF.
JP3a = not significantly different than Jackson & Pollock (1980), ⅀3 SF (Th, Tri, Sup).
JP3b = not significantly different than Jackson & Pollock (1985), ⅀3 SF (Ab, Tri, Sup).
DW = not significantly different than Durnin & Womersley (1974).
GN = not significantly different than Gause-Nilsson et al. (2005).
P = not significantly different than Peterson et al. (2003).
V4 = not significantly different than Visser et al. (1994), ⅀4 SF.
V1 = not significantly different than Visser et al. (1994), ⅀1 SF.
C = not significantly different than Cicone et al. (2021).
L4 = not significantly different than Lean et al. (1996), ⅀4 SF.
L1 = not significantly different than Lean et al. (1996), ⅀1 SF.
RO = not significantly different than Rojano-Ortega et al. (2024).
For males (Table 4), seven of the ten SF-PE were similar to IB770 for YM (C, DW, JP3a, JP3b, L1, L4 and W). Five of the ten SF-PE were similar to IB770 for MM (DW, JP3a, JP3b, JP7 and W). Six of the thirteen SF-PE were similar to IB770 for OM (DW, JP3a, JP3b, P, RO and W).
For females (Table 5), only one of the nine SF-PE was similar to IB770 for YF (JP3b). Three of the nine SF-PE were similar to IB770 for MF (JP3a, JP3b and JP7). Three of the twelve SF-PE were similar to IB770 for OF (DW, P and RO).
%BF among all SF-PE had both similarities and differences for both males and females (Tables 4 and 5). Of all pairwise comparisons among SF-PE for age and sex, there were more significant differences than non-significant differences (except YM) among all SF-PE, which includes: (1) for YM, 47 % were significant differences and 53 % were non-significant differences; (2) for MM, 55 % were significant differences and 45 % were non-significant differences; (3) for OM, 55 % were significant differences and 15 % were non-significant differences; (4) for YF, 82 % were significant differences and 18 % were non-significant differences; (5) for MF, 80 % were significant differences and 20 % were non-significant differences and (6) for OF, 83 % were significant differences and 17 % were non-significant differences.
Bland–Altman plots for each age/sex group in which %BF between IB770 and the corresponding SF-PE showed similarity (i.e. not significantly different, Tables 4 and 5) are shown in Figures 2–7. Each plot includes the mean bias in %BF (mean difference between IB770 and the corresponding SF-PE) and also shows the systematic error between the two measurement methods. Each plot also includes upper and lower limits of agreement (mean difference ± 1·96 sd of differences), which represent the random error range within which 95 % of the differences between two measurement methods are expected to fall. Mean bias %BF values for all plots were near zero, with an average (sd) across all plots of −0·27 (1·27)%BF, and a range between ±0·11 %BF to 2·5 %BF. Limits of agreement for all plots collectively had an average (sd) %BF of −8·3 (3·0)%BF (range between −3·2 %BF to −14·0 %BF) for the lower limit and 7·7 (2·4)%BF (range between 2·7 %BF to 11·1 %BF) for the upper limit.
%BF between IB770 and C, DW, JP3a, JP3b, L1, L4, W for YM.
Figure 2. Long description
Panel A: A scatter plot compares the difference in body fat percentage (C YM - IB770 YM) against the mean body fat percentage (IB770 YM and C YM). The x-axis represents the mean body fat percentage, ranging from 8 to 24 percent. The y-axis represents the difference in body fat percentage, ranging from -10 to 8 percent. The plot includes lines for ULoA (4.7), Bias (1.76), and LLoA (-8.3). Panel B: A scatter plot compares the difference in body fat percentage (DW YM - IB770 YM) against the mean body fat percentage (IB770 YM and DW YM). The x-axis represents the mean body fat percentage, ranging from 8 to 26 percent. The y-axis represents the difference in body fat percentage, ranging from -8 to 8 percent. The plot includes lines for ULoA (5.8), Bias (0.28), and LLoA (-5.2). Panel C: A scatter plot compares the difference in body fat percentage (JP3a YM - IB770 YM) against the mean body fat percentage (IB770 YM and JP3a YM). The x-axis represents the mean body fat percentage, ranging from 8 to 24 percent. The y-axis represents the difference in body fat percentage, ranging from -15 to 15 percent. The plot includes lines for ULoA (8.0), Bias (-1.28), and LLoA (-10.6). Panel D: A scatter plot compares the difference in body fat percentage (JP3b YM - IB770 YM) against the mean body fat percentage (IB770 YM and JP3b YM). The x-axis represents the mean body fat percentage, ranging from 6 to 24 percent. The y-axis represents the difference in body fat percentage, ranging from -15 to 15 percent. The plot includes lines for ULoA (8.5), Bias (-1.63), and LLoA (-11.8). Panel E: A scatter plot compares the difference in body fat percentage (L1 YM - IB770 YM) against the mean body fat percentage (IB770 YM and L1 YM). The x-axis represents the mean body fat percentage, ranging from 8 to 26 percent. The y-axis represents the difference in body fat percentage, ranging from -8 to 8 percent. The plot includes lines for ULoA (7.0), Bias (0.31), and LLoA (-6.4). Panel F: A scatter plot compares the difference in body fat percentage (L4 YM - IB770 YM) against the mean body fat percentage (IB770 YM and L4 YM). The x-axis represents the mean body fat percentage, ranging from 8 to 26 percent. The y-axis represents the difference in body fat percentage, ranging from -4 to 8 percent. The plot includes lines for ULoA (5.6), Bias (0.85), and LLoA (-3.9). Panel G: A scatter plot compares the difference in body fat percentage (W YM - IB770 YM) against the mean body fat percentage (IB770 YM and W YM). The x-axis represents the mean body fat percentage, ranging from 6 to 28 percent. The y-axis represents the difference in body fat percentage, ranging from -6 to 6 percent. The plot includes lines for ULoA (4.4), Bias (-0.15), and LLoA (-4.8).
%BF between IB770 and DW, JP3a, JP3b, JP7, W for MM.
%BF between IB770 and DW, JP3a, JP3b, P, RO, W for OM.
%BF between IB770 and JP3b for YF.
%BF between IB770 and JP3a, JP3b, JP7 for MF.
Figure 6. Long description
Panel A: A scatter plot compares the difference in percent body fat (PBF) between IB770 and JP3a for males. The x-axis represents the mean PBF (IB770 and JP3a), ranging from 10 to 40 percent. The y-axis represents the difference in PBF (JP3a - IB770), ranging from -10 to 10 percent. The plot includes upper and lower limits of agreement (ULoA and LLoA) and a bias line. Panel B: A scatter plot compares the difference in PBF between IB770 and JP3b for males. The x-axis represents the mean PBF (IB770 and JP3b), ranging from 10 to 40 percent. The y-axis represents the difference in PBF (JP3b - IB770), ranging from -8 to 10 percent. The plot includes ULoA and LLoA lines and a bias line. Panel C: A scatter plot compares the difference in PBF between IB770 and JP7 for males. The x-axis represents the mean PBF (IB770 and JP7), ranging from 10 to 40 percent. The y-axis represents the difference in PBF (JP7 - IB770), ranging from -8 to 8 percent. The plot includes ULoA and LLoA lines and a bias line.
%BF between IB770 and DW, P, RO for OF.
Because the mean bias %BF was generally close to 0 %BF (perfect agreement) for all Bland–Altman plots, %BF was similar between IB770 and the corresponding SF-PE. A positive mean bias indicated SF-PE underestimated %BF compared with IB770, while a negative mean bias indicated the SF-PE overestimated %BF compared with IB770. The upper and lower limits of agreement show the range of agreement in %BF between IB770 and SF-PE. Finally, proportional bias occurred when difference in %BF between IB770 and SF-PE changed as the magnitude of the mean %BF changed. Proportional bias was observed to a greater extent in the three female groups (Figures 5–7) compared with the three male groups (Figures 2–4).
For the seven comparisons for YM in Figure 2, given DW and W had the lowest mean bias (near zero), the lowest levels of agreement and generally without proportional bias, DW and W are recommended to use when estimating %BF in YM (L1, L4 also acceptable). For the five comparisons for MM in Figure 3, given W had the lowest mean bias (near zero), the lowest levels of agreement and without proportional bias, W is recommended to use when estimating %BF in MM (DW also acceptable). For the six comparisons for OM in Figure 4 shows, given DW had the lowest mean bias (near zero), the lowest levels of agreement and without proportional bias, DW is recommended to use when estimating %BF in OM (W also acceptable). For the one comparison for YF in Figure 5, JP3b had relatively low mean bias and limits of agreement, but did have some proportional bias in those with below average %BF, where %BF was overestimated. Although JP3b is recommended to use when estimating %BF in YF, it is most appropriate for YF with average or above average %BF and should be used cautiously in individuals with below average %BF because of proportional bias. For the three comparisons for MF in Figure 6, given JP3a has the lowest mean bias (near zero), JP3b and JP7 have the lowest levels of agreement and all three had similar proportional bias, JP3a, JP3b and JP7 are all equally recommended for use for MF. However, given all three had proportional bias, the SF-PE for MF are most appropriate in those with near average %BF, given %BF was overestimated with below average %BF and underestimated with above average %BF. For the three comparisons for OF in Figure 7, given P had the lowest mean bias (near zero), the lowest levels of agreement and relatively low proportional bias, P is recommended to use when estimating %BF in OF. However, P did have some proportional bias showing that %BF was underestimated with above average %BF.
Discussion
This is the only known study that has examined the accuracy of various SF-PE in estimating %BF in young, middle-aged and older male and female adult groups. Assessing %BF accurately is important given excessive %BF increases cardiometabolic disease risk. SF-PE either underestimated, overestimated or accurately estimated %BF compared with IB770. As hypothesised, %BF in one or more SF-PE showed statistical similarity to the selected criterion reference IB770 for all six age/sex groups. Based on both statistical assessment and the Bland–Altman plots, similarity, agreement and accuracy observed in %BF between IB770 and SF-PE were determined. From these assessments, the following SF-PE are recommended to estimate %BF in YM, MM, OM, YF, MF and OF (%BF between IB770 and SF-PE shown): (1) YM: IB770 = 16·2 %; DW = 16·3 %; W = 16·2 %; L1 = 16·6 %; L4 = 17·1 % ; (2) MM: IB770 = 19·7 %; W = 19·7 %; DW = 22·3 %; (3) OM: IB770 = 24·4 %; DW = 25·3 %; W = 22·4 %; (4) YF: IB770 = 24·9 %; JP3b = 23·8 %; (5) MF: IB770 = 25·2 %; JP3a = 25·0 %; JP3b = 26·1 %; JP7 = 24·0 % and (6) OF: IB770 = 36·6 %; P = 36·2 %.
The findings from the Bland–Altman plots in the current study are similar to findings from Bland–Altman plots between bioelectrical impedance and SF-PE in the literature(Reference Arlindo de Sousa, de Macedo and Coutinho de Azevedo30–Reference Yesil, Kose and Ozdemir33). These plots visually and quantitatively exhibited an excellent agreement in %BF estimations between IB770 and SF-PE.
Proportional bias in the current study was similar to the literature that reported %BF involving bioelectrical impedance v. SF-PE(Reference Arlindo de Sousa, de Macedo and Coutinho de Azevedo30–Reference Valente, Bicho and Duarte32). Both in the current study and Valente et al., (Reference Valente, Bicho and Duarte32) the proportional bias pattern observed was SF-PE overestimated %BF when mean %BP decreased and underestimated %BF when mean %BP increased (Figures 5–7).
Compared with the literature, limits of agreement and mean bias values were similar although slightly lower in the current study, which may imply slightly better agreement with higher accuracy between IB770 and SF-PE. For example, Valente et al. (Reference Valente, Bicho and Duarte32) reported that their limits of agreement values for their nine Bland–Altman plots were approximately –10 %BF to 10 %BF, compared with an average limits of agreement in the current study of –8·3 %BF to 7·7 %BF. In addition, Valente et al. (Reference Valente, Bicho and Duarte32) mean bias was approximately 0–2 %BF for five plots and approximately 5–20 %BF for the remaining four plots, compared with a mean bias in the current study between 0–1 %BF for fourteen plots and between 1·1–2·5 %BF for the remaining eleven plots. Kotnik et al. (Reference Kotnik, Robic and Golja31) reported a mean bias of ±1·9 %BF across twelve plots and limits of agreement values similar to the current study. Yesil et al. (Reference Yesil, Kose and Ozdemir33) showed limits of agreement values from approximately –10 %BF to 10 %BF and a mean bias close to –1·0 %BF. Arlindo de Sousa et al. (Reference Arlindo de Sousa, de Macedo and Coutinho de Azevedo30) reported limits of agreement values between –10·1 %BF to 8·1 %BF and a mean bias of –1·0 %BF.
Although DEXA is commonly used as a criterion reference for estimating body composition, it has been reported to overestimate %BF by 3–4 % when compared with a 5-C model(Reference Moon, Eckerson and Tobkin28). Several validation studies have demonstrated similar %BF between IB770 and DEXA, with %BF variations of only 1·2 % for men and 2·0 % for women(Reference Boykin, Tinsley and Harrison7–Reference McLester, Nickerson and Kliszczewicz11). These studies demonstrate that IB770 is a close representation of an individual’s actual %BF when compared with DEXA, providing justification for its inclusion in the current study.
Numerous studies with similar populations and BMI employed SF-PE to estimated %BF used both DEXA and Bioimpedance as selected criterion references to estimate %BF(Reference Rojano-Ortega, Moya-Amaya and Molina-Lopez20,Reference Chambers, Parise and McCrory23,Reference Kwok, Woo and Lau24,Reference Escamilla, Yamashiro and Asuncion29,Reference Gonzalez-Torres, Anaya-Esparza and Trigueros Del Valle34–Reference Silveira, Barbosa and Rodrigues38) . Collectively, the mean %BF from all these studies were 25·8 % from DEXA and 26·2 % from Bioimpedance for males, and 34·9 % from DEXA and 32·0 % from Bioimpedance for females. Moreover, compared with DEXA, IB770 requires less space, requires less time to assess %BF (only about 1 min for IB770 compared with 30–60 min for DEXA), does not require specialised training and cost less. Consequently, the IB770 may be more practical and well suited for use in clinical settings and appears to be a reasonable alternative to DEXA as a selected criterion reference for assessing %BF.
The SF-PE employed in the current study were based on equations developed in the literature using a similar population, race, sex, age, athletic status and BMI. Like the current study, all SF-PE used were comprised of primarily an adult Caucasian population of males and females with a similar BMI, age range and mean age, with athletes and children excluded.
The most commonly used SF-PE over the past 4–5 decades were developed by Durnin & Womersley (DW)(Reference Durnin and Womersley15), who used 4-site skinfold and age to assess %BF, and were developed by Jackson & Pollock(Reference Jackson and Pollock12,Reference Jackson and Pollock17) and Jackson et al. (Reference Jackson, Pollock and Ward13) (JP3a, JP3b and JP7), which used 7-site and 3-site skinfolds and age to assess %BF. In the current study, JP3a, JP3b, JP7 and DW were amongst the most accurate and similar to IB770.
Peterson et al. (P)(Reference Peterson, Czerwinski and Siervogel19) employed a 4-C model using 4-site skinfold, age, height and BMI to assess %BF. These authors reported that DW, JP7, JP3a and JP3b equations underestimated %BF compared with P equations. In the current study, this was demonstrated to be true only for older females when using JP3a, JP3b and JP7. P was similar to IB770 in older males and females and most accurate in older females.
When averaged across the three age groups, %BF from the current study were very similar to %BF from the literature that developed SF-PE. For example, %BF between the current study, JP7, JP3a and JP3b were all between 17–18 % for males and 24–25 % for females, and these were also similar to the %BF values reported by Escamilla et al. (Reference Escamilla, Yamashiro and Asuncion29), which were 16·0 % for males and 24·6 % for females.
Numerous studies(Reference Durnin and Womersley15,Reference Gause-Nilsson and Dey16,Reference Peterson, Czerwinski and Siervogel19,Reference Chambers, Parise and McCrory23,Reference Ravaglia, Forti and Maioli37–Reference Jayawardena, Waniganayake and Abhayaratna41) have estimated %BF using SF-PE from Durnin & Womersley (DW), reporting similar mean %BF compared with the current study (≈22–24 % for males and ≈33–35 % for females). Moreover, Peterson et al. (P)(Reference Peterson, Czerwinski and Siervogel19) compared %BF from their SF-PE to SF-PE from DW, and for males reported a %BF of 22·7 % using P and 20 % using DW, and for females reported a %BF of 32·6 % using P and 31 % using DW. The %BF values reported by P were very similar to the %BP using P in the current study, which were 23·5 % for males and 32·6 % for females.
Lean et al. (Reference Lean, Han and Deurenberg18) used 1-site (L1) and 4-site (L4) skinfolds, and waist circumference for L1, to assess %BF. These authors reported a waist circumference of 88·1 cm for males and 78·3 cm for females, which were similar to mean waist circumference in the current study. These authors also reported a single %BF value of 22·3 % for males and 33·7 % for females, similar to %BF in the current study.
Cicone et al. (C)(Reference Cicone, Nickerson and Choi14) employed a 5-C model using 3-site skinfold, sex, age, BMI and waist circumference to assess %BF. For YF, %BF of 22·0 % for C was similar to %BF of 24·9 % from IB770. However, the middle-age group demonstrated a larger gap in %BF between C and IB770 (20·6 % v. 25·2 %), and the older age group demonstrated even a larger gap in %BF between C and IB770 (18·7 % v. 36·6 %). This same trend occurred in males, with good agreement in %BF between C and IB770 in YM (14·6 % v. 16·2 %), a larger gap in %BF between C and IB770 for MM (15·9 % v. 19·7 %) and even a larger gap in %BF between C and IB770 in OM (17·6 % v. 24·4 %). These results demonstrate that SF-PE from C is only valid for use in YM and YF. The reason for this discrepancy is that although C(Reference Cicone, Nickerson and Choi14) used an age range between 18 and 69 for males and females, which makes their SF-PE appear suitable for all age and sex groups, the mean age of their participants was only 26·5 years for males and 25·2 years for females, which invalidated the use of SF-PE from C for middle-age and older males and females.
Williams et al. (W)(Reference Williams, Going and Lohman22) used 1-site and 4-site skinfolds, age and waist circumference to assess %BF. SF-PE from W was only applied to males in the current study because calf skinfolds were used for females, which was not measured in the current study. These authors reported %BF of 22·9 % for males, similar to %BF of 19·4 % in the current study.
Rojano-Ortega et al. (R)(Reference Rojano-Ortega, Moya-Amaya and Molina-Lopez20) used 3-site skinfold, mass, height, sex and age but did not report %BF. These authors only reported a %BP from DEXA of 30·5 % for females, which was similar to %BF of 34·1 % for females in the current study.
Visser et al. (Reference Visser, van den Heuvel and Deurenberg21) used 2-site (V2) and 4-site (V4) skinfolds and sex to assess %BF. These authors only reported %BF for V4, which were 31·2 % for older males and 43·6 % for older females, similar to the current study, which were 30·3 %BF (V2 and V4) for older males and 43·2 %BF (V2) and 42·5 %BF (V4) for older females.
Gause-Nilsson & Dey (GN)(Reference Gause-Nilsson and Dey16) used 4-site skinfold, sex, mass and height to assess %BF. These authors reported %BF of 26·9 % for older males and 33·7 % for older females, which were quite different than %BF of 17·0 % for older males and 20·9 % for older females in the current study. These differences are perplexing given the input variables for GN were similar to the current study. In fact, if the age, mass, height and ⅀4 SF values reported by GN were input into the GN equation listed in their paper and shown in Table 3, the %BF for both older males and females would be similar to the %BF for older males and females in the current study. Moreover, the %BF from GN in the current study for older males and females are abnormally lower than both IB770 and all SF-PE except C, which as previously noted is low because C was not appropriate to use for older males and females given the mean age of their participants was only 25–26 years.
There were limitations to the current study. First, %BF estimations using SF-PE are affected by several factors, such as the ability of the tester to accurately measure skinfold thickness, quantity of subcutaneous tissue, the type of skinfold calipers used and their calibration accuracy and the validity and accuracy of SF-PE. Regarding the ability of the skinfold tester to accurately measure skinfold thickness, the individual tester in the current study and several of the studies listed in Table 3 were well trained, skilled and experienced. Although some of the studies from Table 3 did not report the training or skill level of the skinfold testers, there is a basic assumption that experienced skinfold testers are employed in studies whose primary focus is on taking skinfold thickness measurements to estimate %BF. Regarding different skinfold calipers, Lange, Harpenden and Slim Guide skinfold calipers are three of the most common and accurate calipers employed in assessing %BF from skinfolds in clinical and research settings, and while all are relatively similar in their ability to accurately measure skinfold thickness, differences still may occur among them. There are no known scientific studies that has compared all three of these calipers to each other to assess how different or similar they can measure skinfold thickness. However, Durnin & Womersley(Reference Durnin and Womersley15) reported no significant differences between skinfold measurements using Lange and Harpenden skinfold calipers, and nearly all of the calipers employed using the SF-PE in the current study used either Lange or Harpenden skinfold calipers. Regarding obese and normal-weight individuals, Duren and colleagues(Reference Duren, Sherwood and Czerwinski42) reported decreased accuracy in using skinfolds and SF-PE in obese individuals, which is why obese individuals were excluded in the current study, and why SF-PE from the current study are not appropriate for obese individuals. Moreover, as previously noted, the accuracy in estimating %BF varies according to race (e.g. Caucasian, Asian and black), sex (male v. female), age (young, middle age and older), body fat quantity, BMI, athletic status and other factors. In the current study, these factors were controlled as much as possible, such as including only the Caucasian race for all participants, excluding obese and underweight individuals, not including athletes or children, using SF-PE that were specifically designed for males or females, and specifically for younger, middle-aged or older individuals. One additional limitation is that IB770 instead of DEXA was selected as the criterion reference. However, recent advances in bioelectrical impedance devices, such as IB770, have demonstrated similar %BF relative to other criterion reference methods, such as DEXA(Reference Boykin, Tinsley and Harrison7–Reference McLester, Nickerson and Kliszczewicz11). It has been reported that compared with DEXA, IB770 underestimated %BF by only 1·2 % for men and 2·0 % for women(Reference Boykin, Tinsley and Harrison7–Reference McLester, Nickerson and Kliszczewicz11). These relatively small differences in %BF between DEXA and IB770 implies that IB770 may exhibit appropriate accuracy in assessing %BF, especially in clinical settings where DEXA is often not available or appropriate for use for the reasons previously described.
Conclusion
Although there was no single SF-PE that was found to be similar (not significantly different) and in agreement with IB770 for all 6 age and sex groups, the current study showed excellent agreement in %BF between IB770 and select SF-PE based on a mean bias close to zero and a relatively narrow range for the limits of agreement. Based both on statistical assessment and the Bland–Altman plots, similarity, agreement and accuracy observed in %BF between IB770 and SF-PE were determined. From these assessments, the following SF-PE are recommended to estimate %BF in YM, MM, OM, YF, MF and OF: (1) YM: DW, W, L1 and L4; (2) MM: W and DW; (3) OM: DW and W; (4) YF: JP3b; (5) MF: JP3a, JP3b and JP7; (6) OF: P. Given that SF-PE either underestimated, overestimated or accurately estimated %BF compared with IB770, this emphasises the importance of using SF-PE that are appropriate for different age/sex groups and provide as accurate as possible estimates of %BF. Assessing %BF accurately is important given excessive %BF increases cardiometabolic disease risk. Caution should be taken when using SF-PE among the 6 age/sex groups given that some SF-PE demonstrated proportional bias and underestimated or overestimated %BF compared with %BF from IB770.
Acknowledgements
All the participants who volunteered for our study.
The authors of this manuscript affirm we have no financial affiliation (including research funding) or involvement with any commercial organisation that has a direct financial interest in any matter included in this manuscript.
The authors have nothing to declare and no conflicts of interest.
(1) Designed research, R. F. E., I. S. T., R. A., K. Y., D. M. and M. M.; (2) conducted research, R. F. E., I. S. T., R. A., K. Y., D. M. and M. M.; (3) provided essential materials, R. F. E.; (4) analysed data or performed statistical analysis, R. F. E., I. S. T., R. A., K. Y., D. M. and M. M.; (5) wrote paper, R. F. E; (6) had primary responsibility for final content, R. F. E., I. S. T., R. A., K. Y., D. M. and M. M. and (7) all authors have read and agreed to the published version of the manuscript, R. F. E., I. S. T., R. A., K. Y., D. M. and M. M.
This study was performed at California State University, Sacramento, CA USA.
The authors of this manuscript affirm we have no financial affiliation (including research funding) or involvement with any commercial organisation that has a direct financial interest in any matter included in this manuscript. The manuscript work described has not been published previously except in the form of a preprint, an abstract, a published lecture, academic thesis or registered report. The manuscript is not under consideration for publication elsewhere. The manuscript’s publication is approved by all authors and tacitly or explicitly by the responsible authorities where the work was carried out. If accepted, the article will not be published elsewhere in the same form, in English or in any other language, including electronically, without the written consent of the copyright-holder. All participants provided written informed consent in accordance with the Institutional Review Board at California State University, Sacramento (protocol number AG2374).
Open access
