The effect of postural orientation on body composition and total body water estimates produced by smartwatch bioelectrical impedance analysis: an intra- and inter-device evaluation

A total of 126 participants were prospectively recruited for this cross-sectional evaluation. Participants were excluded if they had a significant amount of internal metal such as a metal plate or a complete joint replacement (to avoid interference with the injected currents from bioelectrical devices and artifact using DXA); had a pacemaker or any other electrical implant; were missing any limbs or part of a limb; or were pregnant; trying to become pregnant; or breast-feeding or lactating. Nine participants were excluded due to scheduling conflicts and thus, a diverse sample of 117 (42 Non-Hispanic White; 37 Non-Hispanic Black; 32 Asian; 6 Hispanic White) males and females (48 M; 69 F) between the ages of 18 and 39 (age: 21.4 ± 3.0 y; BMI: 25.3 ± 5.7 kg/m²) completed this study in its entirety.

Participants reported to the laboratory following an ≥ 8 h overnight fast from food, beverage, supplements and medication and after abstaining from exercise for ≥ 24 h. Upon arrival, participants were instructed to remove all external metal and accessories (jewelry, shoes, watches, etc.), and any excessive clothing, so that only tight fitting athletic clothing was worn during testing. Then, participants underwent measurements of height using a stadiometer, body mass using a calibrated digital scale (SECA 769, Hamburg, Germany), and body composition assessments using DXA (Lunar iDXA, General Electric, Boston, MA, USA), BIS (SFB7, Impedimed^®, Carlsbad, CA, USA), and SWBIA (Galaxy Watch 4^®, Samsung^®, Suwon, South Korea). To ensure the most equitable comparisons, the body mass collected from the calibrated digital scale was the body mass used for both SWBIA and DXA. As such, FM was calculated as: body mass x (BF%/100); and FFM was calculated as body mass minus FM.

The methods used for SWBIA [3], BIS [4], and DXA [7] have been described in detail elsewhere, and a summary of these procedures are presented in the following paragraphs with additional information unique to this study provided in greater detail. Moreover, this study is a secondary analysis that extends from a larger line of investigation (ClinicalTrials.gov; NCT05885672); though the dependent variables presented in this study are unique to the current investigation.

Body composition estimates from a single-frequency, hand- to-hand, tetrapolar SWBIA were produced using a common consumer grade 40 mm smartwatch paired to a mobile application (Samsung Health^®; Samsung^®, Suwon, South Korea). Importantly, this device has previously demonstrated acceptable test-retest reliability [3]. Using four individual electrodes positioned underneath the watch face (two electrodes in contact with the user’s left wrist) and within watch frame (two electrodes in contact with the user’s right middle and ring finger) this SWBIA injects alternating currents to measure the impedance of the user’s upper body FFM at a frequency of 50 kHz and subsequently generates estimates of whole-body BF%, FM, and TBW from proprietary algorithms embedded within the watch’s software [1,3]. Importantly, this SWBIA is equipped with an internal analog front-end chip and has signal electrodes positioned inside the reference electrodes to account for the greater impedance and current concentrations inherent to its small and narrow electrode configuration [1,3].

Prior to testing, each participants’ age, sex, height, and weight were entered into the affiliated mobile application which was subsequently synced with the SWBIA. Then, the smartwatch was fastened on the participant’s left wrist so that the bottom electrodes were in complete contact with the skin. After the watch was successfully secured, a series of three separate measurements were performed while participants were 1) seated, 2) standing, or 3) supine. For the seated position, participants were instructed to sit up straight in a standard chair with feet flat on the ground so that their legs were approximately 90°. For the supine position, participants were positioned on an exam table with their backs flat and hands to their side until the measurement required movement of the hands. For the standing position, participants were simply asked to stand upright. Participants remained in each position for ≥ 10 min prior to testing so that body fluid could be adequately redistributed. The smartwatch was removed and repositioned prior to each measurement.

After remaining in each position for ≥ 10 min, participants were instructed to pronate their left wrist, supinate their right wrist, and laterally raise both arms. Participants were then instructed to bring to their hands together as to make a “circle” or “hoop” with their arms, and to position the tips of their right middle and ring fingers so that they were in firm contact with the electrodes extending from the watch’s frame. Participants were also instructed to bend their left wrist so that contact between the right fingers and the surface of the left hand could be avoided. For the seated and standing conditions, participants raised their arms until parallel to the floor. For the supine position, participants raised their arms upward to a comfortable point where they could see the watch’s interface — representing the position most likely to be performed in practice. After the participant’s fingers were placed against the electrodes, participants remained still for 15 seconds while the measurement was completed.

TBW estimates were collected using a validated [8] multi-frequency, hand-to-hand, tetrapolar BIS (SFB7, Impedimed^®, Carlsbad, CA, USA). Prior to testing, participants were instructed to lay supine for ≥ 10 min while two adhesive electrodes were placed on their left posterior wrist and hand and left anterior ankle and foot. The hand and foot electrodes were placed approximately 5 cm distal to their limb-specific proximal electrodes [9] in accordance with the manufacturer guidelines. Then, a single measurement was performed to produce estimates of TBW at a frequency of 50 kHz; which was selected 1) to match the frequency of the SWBIA, allowing for the most equitable comparisons between SWBIA and BIS; and 2) because 50 kHz represents the standard frequency used in research and practice for both single- and multi-frequency bioelectrical impedance devices [4,10]. The sex, body density, body proportion, and hydration coefficients were the default recommendations established by the manufacturer as used in prior investigations [4,10]. BIS was calibrated daily prior to testing, and visual inspection of Cole plots were performed following each measurement. For reasons unbeknownst to the investigators nor the manufacturer, the BIS device used in this study experienced technical difficulties with the device interface that required the device to be serviced by the manufacturer. While this did not have an impact on any of the data reported in the present study, BIS was unavailable for testing during the visits of 10 individual participants, and a single participant had Cole plots outside of the acceptable bounds (unrelated to the technical difficulties). Thus, TBW data was missing/omitted for 11 participants. As such, 106 participants (62 F, 44 M) were used to assess the agreement between the SWBIA and BIS-derived TBW as opposed to the 117 used to evaluate the agreement between SWBIA and DXA-derived body composition estimates.

A Lunar iDXA with version 18 enCORE software was used to produce estimates of whole-body BF%, FM, and FFM. Participants were positioned on the DXA according to manufacturer’s recommended guidelines [7] and reflection scanning procedures were performed for larger participants [11].

Using a small-to-medium effect size of d = 0.30 for a two-tailed paired null hypothesis significance test, it was determined that 90 participants would yield 80.4% power at an α = 0.050. Differences in the descriptive characteristics by sex were assessed using independent samples t-tests. Repeated measures ANOVA with Bonferroni-Holm corrections for multiple comparisons were used to determine the differences in the SWBIA-derived body composition estimates by body position. The mean difference (MD) between SWBIA and the criterion DXA and BIS were calculated as the SWBIA estimate in question minus the estimates produced by either DXA (BF%, FM, FFM) or BIS (TBW) and evaluated using null hypothesis significance and equivalence tests. The equivalence regions used for equivalence testing were defined as ±1.0% for BF%; ±1.0 kg for FM and FFM; and ±1.0 L for TBW. Additionally, the agreement between SWBIA and the criterion DXA and BIS were evaluated using coefficients of determination (R2), Deming regression, Bland-Altman analyses, RMSE, and concordance correlation coefficients (CCC). Agreement between the predicted values and the line of identity produced using weighted Deming regression was determined if 95% confidence intervals (95%CI) for the intercept and slope contained the values 0 and 1, respectively. The 95% limits of agreement (LOA) and proportional biases were determined using Bland-Altman and linear regression techniques. All evaluations were conducted for the combined sample and by sex. Data were normally distributed based on the results of Shapiro-Wilk and visual inspection of Q-Q plots. Statistical significance was accepted at p < 0.050 and data were analyzed using R version 4.1.2 and Microsoft Excel version 16.

Informed consent has been obtained from all individuals included in this study.

The research related to human use has been complied with all relevant national regulations, institutional policies and in accordance with the tenets of the Helsinki Declaration, and has been approved by the authors’ institutional review board or equivalent committee.

Participant characteristics are presented in Table 1. Of the 117 participants, 66 were classified as normal weight (56.4%) and 51 were classified as having either overweight (n = 33; 28.2%) or obesity (n = 18; 15.4%). Additionally, 35.9% were White, 31.6% were Black, 27.4% were Asian, and 5.1% were Hispanic. All descriptive characteristics were significantly different between males and females (all p ≤ 0.008) with the exception of age (p = 0.391) and body mass index (BMI; p = 0.590).

The means for SWBIA body composition and TBW estimates by body position are presented in Table 2 and Table 3, respectively. BF% (F = 5.67; p = 0.004), FM (F = 6.35; p = 0.002), FFM (F = 6.35; p = 0.002), and TBW (F = 3.99; p = 0.020) were significantly different by body position in the combined sample. Post hoc analyses showed that all body composition estimates, with the exception of TBW (p = 0.051), were all significantly different between standing and supine positions (all p < 0.001). However, simple effects tests revealed data consistent with differences between standing and supine TBW (MD: 0.13 L, MD^95%CI: 0.02, 0.25; d: 0.21, d^95%CI: 0.03, 0.39; p = 0.026). TBW estimates were significantly different between the seated and supine positions (p = 0.015). There were no significant differences between the sitting and standing positions (all p ≥ 0.195).

For females, BF% (F: 6.79; p = 0.002), FM (F: 7.10; p = 0.001), and FFM (F: 7.11; p = 0.001) were significantly different across body positions, with post hoc tests revealing differences between standing and supine estimates (all p < 0.001). Despite a non-significant omnibus model for female TBW (F = 2.80; p = 0.064), simple effects tests revealed data consistent with differences between seated and supine TBW estimates (MD: 0.17 L, MD^95%CI: 0.06, 0.28; d: 0.38, d^95%CI: 0.13, 0.62; p = 0.003). There were no significant differences between the seated and standing positions for females (all p ≥ 0.153). No significant differences were observed across positions in the male only sample (all p ≥ 0.290).

For the combined sample and sex groups, paired null hypothesis significance tests revealed no differences between SWBIA and DXA regardless of body position (Table 2; all p ≥ 0.083). For the combined sample and females, both SWBIA FM and FFM demonstrated equivalence with DXA across all body positions; however, only standing BF% demonstrated equivalence with DXA for these groups. No estimate demonstrated equivalence with DXA for males. Results of the Bland-Altman and Deming regression analyses are illustrated in Figures 1–3 and Supplemental Figures 1-3, respectively. In the combined sample, Deming regression revealed no difference from the line of identity for supine FFM estimates only (Supplemental Figure 1), whereas differences were observed for all BF% or FM estimates.

Fig 1a-d.

Fig 2a-d.

Fig 3a-d.

RMSEs ranged from 4.37-4.40% for BF% and from 3.15-3.19 kg for FM and FFM; all of which were lowest for the supine position. R2 ranged from 0.82-0.93 and CCC’s ranged from 0.90-0.96; which were similar across body positions. Proportional biases were observed for all estimates in the combined sample (Figure 1a-d; all p ≤ 0.026). Large LOA were observed for BF% (±8.51-8.61%) and FM and FFM (±6.14-6.26 kg) with the lowest LOA observed while supine (Figure 1a-d).

For females, Deming regression revealed differences from the line of identity for BF% and FM, but not FFM independent of body position (Supplemental Figure 2). RMSEs ranged from 3.81-3.89% for BF% and from 2.54-2.57 kg for FM and FFM; all of which were lowest in the seated position. R² ranged from 0.82-0.97 and CCC ranged from 0.88-0.98; all of which were similar between body positions. Proportional biases were observed for BF% (all p < 0.001) and FM (all p ≤ 0.003), but not for FFM for females (all p ≥ 0.741) (Figure 2a-d). LOA were large for BF% (±7.43-7.63%) and FM and FFM (±4.96-5.07 kg) for females, with the lowest BF% and FM/FFM LOAs observed while seated and supine, respectively (Figure 2a-d).

For males, Deming regression revealed differences from the line of identity for FFM, but not BF% or FM, independent of body position (Supplemental Figure 3). RMSEs ranged from 4.98-5.13% for BF% and from 3.85-3.93 kg for FM and FFM, which were lowest in the supine position. R² ranged from 0.56-0.90 and CCC’s ranged from 0.72-0.94, which were similar across body positions. Proportional biases were observed for FFM across all body positions (all p ≤ 0.009) and for standing BF% in males (p = 0.045) (Figure 3a-d). LOA were large for BF% (±9.84-10.15%) and FM and FFM (±7.58-7.74 kg) for males with the lowest observed while supine (Figure 3a-d).

For the combined sample, paired null hypothesis significance tests revealed no significant differences in TBW between SWBIA and BIS (all p ≥ 0.491), and all SWBIA estimates demonstrated equivalence with BIS. However, when separated by sex, SWBIA TBW was significantly higher for males (all p ≤ 0.003) and significantly lower for females (all p ≤ 0.019) across all body positions relative to BIS, and TBW did not demonstrate equivalence with BIS for any position. Deming regression for TBW revealed no differences from the line of identity for all positions in the combined sample (Supplemental Figure 1) and while standing and supine and in males (Supplemental Figure 3); however, differences from the line of identity were observed across all positions in females (Supplemental Figure 2). RMSEs ranged from 2.79-2.80 L for the combined sample, 2.64-2.72 L for females, and 2.88-3.02 L for males; with lowest observed in the seated and supine positions for females and males, respectively.

R² ranged from 0.81-0.89 and CCC’s ranged from 0.88-0.94 across groups and were similar across body positions. No significant proportional biases were observed (all p ≥ 0.050) and LOA were large across all groups and body positions (±4.99-5.52 L); though LOA were lowest for males while supine.