Reliability and validity of a new accelerometer-based device for detecting physical activities and energy expenditure

Study design

The study included two protocols: investigating the reliability (n = 18) and validity (n = 19) of the Fibion accelerometer. All 37 participants were young and healthy volunteers, who had normal weight (i.e., BMI < 25 kg/m²) and were recreationally physically active. Exclusion criteria included acute and chronic diseases, which would prevent participants from prolonged sitting and/or standing or would interfere with the basic metabolic rate. All participants were informed of the study procedures and provided written informed consent prior to the testing. The study was carried out in accordance with the Declaration of Helsinki and was approved by the Ethical Committee of Shanghai Jiao Tong University (Approval number: ML16027).

Reliability protocol

A total of 18 young adults (eight females, 10 males; age 24.0 ± 2.8 years; BMI 22.8 ± 2.3 kg/m²) participated in the repeatability study. The protocol is a designed 45 min repeated test of walking, standing, and sitting, respectively (3 × 15 min, repeated twice, a total of 90 min). The participants were required to simultaneously wear Fibion and ActiGraph proximally on the left thigh, between the knee and the hip (Fig. S2A). However, the exact wearing location (i.e., medial or lateral) of each device was randomized but equally distributed among all participants. The ActiGraph was chosen because it is widely used and its validity and reliability have previously been studied (Welk, Schaben & Morrow, 2004; Esliger & Tremblay, 2006; Vanhelst et al., 2012; Lee, Kim & Welk, 2014; Aadland & Ylvisåker, 2015). Throughout the data recording, the participants were instructed to sit or stand still without shuffling their feet or repositioning their body, under observation by a research assistant. Walking was performed at a self-selected constant velocity. The procedure was strictly confined to the duration of 15 min, while any deviations from the protocol (i.e., uncommanded change of postures) were recorded.

Validity protocol

The validity protocol was performed by 19 young adults (9 females, 10 males; age 28.2 ± 3.8 years; BMI 21.0 ± 2.2 kg/m²), by using a 12 h guided sequence of tasks in the laboratory that simulated realistic daily activities. The participants were required to wear two Fibion devices (Fibion strapped to the front Thigh (FT) and worn in the pocket (FP)), as well as a portable gas analyzer (Cosmed K4b²; Cosmed, Rome, Italy) (Fig. S2B).

All participants were required to abstain from caffeine, alcohol intake, and unaccustomed exercise for 24 h prior to the measurements. The protocol commenced after resting metabolic rate was assessed in a fasting state following best practice guidelines (Compher et al., 2006). One supervisor and two assistants directly followed the participants in 4 h shifts. Specifically, the validity of the Fibion accelerometers was assessed using two criteria:

Criterion 1: 12 h guided sequence of tasks with direct observation

Direct observation of the designed 12 h guided sequence of tasks served as the criterion for detecting different types and intensities of PA (Table S1). The direct observation has been previously proven as a valid method, compared to indirect calorimetry (Lyden et al., 2014). In the present study, the 12 h guided sequence of tasks was designed to simulate an “ideal active working day,” according to current recommendations (Buckley et al., 2015). Thus, it included both computer-based office works (i.e., sitting, standing, walking, and cycling, etc.) and leisure activities in the afternoon (i.e., actually watching TV sitting on a sofa and exercising). In particular, walking was performed either within the laboratory or the hallway (short and long walks) or on the indoor track. Cycling was carried out on an ergometer at a self-paced intensity, which allowed participants to change the cadence as they do in the real-life conditions (i.e., commuting to/from work).

For the data collection, each of the tasks was coded as 0–12 based on the measurement logs of the direct observations, as was done previously (Lyden et al., 2017). The codes were subsequently separated into five different types (sitting, long sitting, standing, walking, and cycling) and three intensities (LPA, MPA, and VPA) of PA (Table S2). On the basis of the metabolic equivalent (MET; Ainsworth et al., 2011b), the PA intensities were defined as follows: long sitting was defined as uninterrupted sitting periods over 30 min. LPA was defined as any activities with an EE below three METs, excluding sitting. MPA was defined as activities with an EE between three and six METs. VPA was defined as activities over six METs. The entire 12 h measurement was recorded by video, to verify the correct timing of each criterion.

Criterion 2: indirect calorimetry

Indirect calorimetry served as the criterion for EE estimation. Pulmonary gas exchange of the participants was continuously measured throughout the 12 h guided sequence of tasks by a portable breath-by-breath gas analyzer (Cosmed K4b², Rome, Italy). The K4b² has previously been shown to be a valid and reliable device for estimating oxygen consumption (McLaughlin et al., 2001; Schrack, Simonsick & Ferrucci, 2010). Each participant was fitted with a rubber facemask (Hans-Rudolph, Kansas City, MO, USA), while the gas sensor was attached to a harvest and carried on the chest throughout the protocol (Fig. S2B). Prior to data recording, the device was calibrated according to the manufacturer’s guidelines (approximately 10 min before the start of the measurement). The facemask was removed from the face only during designated short breaks, that is, for food intake.

Data processing

Fibion data collected from the reliability and validity measurements were uploaded to the manufacturer’s web-browser-based online service (www.fibion.com/upload), in order to obtain detailed reports on PA types and intensities as well as the corresponding EE. The raw comma-separated values (CSV) files with minute-by-minute data from Fibion were subsequently exported for further analysis. Similarly, the obtained ActiGraph data were uploaded to the device-specific software (ActiLife 6) and subsequently exported to Microsoft Excel for further analyses. The ActiGraph data for each activity type (provided in seconds) were then synchronized with manual recordings. To obtain comparable data of EE, the breath-by-breath values collected with the K4b² were averaged for the duration of each task and were expressed as average “kcals/min” of each PA type and a 12 h total EE (Tables S1 and S2). No data were removed from the reliability study. However, for the validity test, a total of 8.2 ± 2.3% of the data, both from Fibion and criteria, were excluded due to visible artifacts (i.e., originating from movement artifacts, errors in the assessment of EE by gas analysis, or meal breaks). Furthermore, validity data from two participants (ID 16, 17) of FT but not FP were removed entirely due to wrong positioning of the thigh strap, according to the retrospective check of the video recordings.

Statistical analyses

Reliability

Table 1 presents the mean ± SD (ME, only for PA duration) of the repeated performances of Fibion (FT) and ActiGraph. The agreement of FT was moderate to good (CV%: 6.57–9.13; ICCs: 0.687–0.806) in all EE measures of sitting, standing, and walking, similar to the ActiGraph. FT accurately detected the duration of sitting (ME: 5.1 ± 1.2%), and walking (ME: 3.8 ± 0.3%), but not for standing (ME: 14.9 ± 2.6%). However, low correlations (ICCs: 0.189–0.459) were observed between subsequent measurements with both devices, especially in low-intensity PAs (sitting).

Validity

Table 2 presents the mean and SD of different PA durations (min) and EE (kcal/min) of Fibion compared with criterion 1 and 2.

Criterion 1

No significant differences were observed between FT and criterion 1 (i.e., PA duration) in sitting, long sitting, LPA, MPA, and VPA (all p > 0.05). By contrast, all types of FP (excepting LPA) significantly differed from criterion 1 (p < 0.05). When comparing FT and FP to the criterion 1, Bland–Altman plots revealed an individualized underestimation (positive mean bias) for standing and LPA, and an overestimation (negative mean bias) for all other PAs (Figs. 1 and 2). The LOA of PA durations ranged from −113.8 to 96.0 min for FT and from −325.7 to 344.5 min for FP. Furthermore, proportional errors were observed for FT (sitting, long sitting, standing, cycling, and MPA; (coefficient of determination) R²: 0.481–0.970) and FP (sitting, long sitting, and standing; R²: 0.912–0.989) and (Figs. 1 and 2). The low correlations (ICCs, FT: 0.016–0.638; FP: −0.046 to 0.650) also indicated that the MEs were random.

Criterion 2

The EE of standing, LPA, MPA, VPA, as well as the overall 12 h EE did not differ between FT and criterion 2 (all p > 0.05; ICCs: 0.363–0.894). However, all other activities significantly differed between FP and criterion 2, except for sitting, standing, LPA, and 12 h EE (Table 2). Bland–Altman plots revealed an individualized underestimation for walking, cycling, and the 12 h EE, as well as an overestimation for sitting, standing, LPA, MPA, and VPA, when comparing FT to the criterion 2 (LOA: −3.93 to 4.71 kcal/min, Fig. 3). Moreover, an individualized underestimation for walking, cycling, LPA, and 12 h EE and an overestimation for sitting, standing, MPA, and VPA were observed, when comparing FP to the criterion 2 (LOA: −4.81 to 4.03 kcal/min, Fig. 4).