**"What Is a Step?" Differences in How a Step Is Detected among Three Popular Activity Monitors That Have Impacted Physical Activity Research**

### **Dinesh John 1,\*, Alvin Morton 2, Diego Arguello 1, Kate Lyden <sup>3</sup> and David Bassett <sup>2</sup>**


Received: 28 February 2018; Accepted: 12 April 2018; Published: 15 April 2018

**Abstract:** (1) Background: This study compared manually-counted treadmill walking steps from the hip-worn DigiwalkerSW200 and OmronHJ720ITC, and hip and wrist-worn ActiGraph GT3X+ and GT9X; determined brand-specific acceleration amplitude (g) and/or frequency (Hz) step-detection thresholds; and quantified key features of the acceleration signal during walking. (2) Methods: Twenty participants (Age: 26.7 ± 4.9 years) performed treadmill walking between 0.89-to-1.79 m/s (2–4 mph) while wearing a hip-worn DigiwalkerSW200, OmronHJ720ITC, GT3X+ and GT9X, and a wrist-worn GT3X+ and GT9X. A DigiwalkerSW200 and OmronHJ720ITC underwent shaker testing to determine device-specific frequency and amplitude step-detection thresholds. Simulated signal testing was used to determine thresholds for the ActiGraph step algorithm. Steps during human testing were compared using bias and confidence intervals. (3) Results: The OmronHJ720ITC was most accurate during treadmill walking. Hip and wrist-worn ActiGraph outputs were significantly different from the criterion. The DigiwalkerSW200 records steps for movements with a total acceleration of ≥1.21 g. The OmronHJ720ITC detects a step when movement has an acceleration ≥0.10 g with a dominant frequency of ≥1 Hz. The step-threshold for the ActiLife algorithm is variable based on signal frequency. Acceleration signals at the hip and wrist have distinctive patterns during treadmill walking. (4) Conclusions: Three common research-grade physical activity monitors employ different step-detection strategies, which causes variability in step output.

**Keywords:** Step-detection; ActiGraph; Pedometer; acceleration; physical activity

#### **1. Introduction**

Wearable activity monitors are commonly used to detect steps to quantify physical activity (PA) [1]. While this seemingly standard metric should enable a direct comparison of PA among studies using step-counters, validation studies report discrepancies in steps from different commercial activity monitors [2–5]. This may be attributable to variability in criteria used by the devices to translate human movement into steps.

Furthermore, there may be within-brand differences in step-output. For example, steps using ActiGraph's ActiLife software on data from newer devices (i.e., model GT1M and higher), yield step-outputs that are not comparable to the older 7164 [6,7] and location of wear (hip vs. wrist) impacts ActiGraph step-output [8]. A substantial proportion of current knowledge on associations between objectively measured PA and health results from the nationally representative 2003–04/2005-06 National Health and Nutrition Examination Survey (NHANES), which used the ActiGraph 7164 [9–11]. However, the latest cycle of the NHANES (2011–12/2013-14) used a wrist-worn GT3X+ [12]. Additionally, while steps computed using ActiLife is most common in PA research, real-time step output is also available to a user via the device display screen in the new GT9X. However, the latter is computed using a real-time step-counting algorithm that is different from that in ActiLife. The real-time algorithm on-board the GT9X uses tri-axial (as opposed to uni-axial acceleration data in ActiLife), and filters out steps that are accumulated in bouts lasting less than 2 and 10 s on the hip and wrist, respectively. Thus, in the context of objectively measuring PA using motion sensors, the answer to the question "what is a step?" may be contingent on device-specific *proprietary* strategies used to detect and translate movement acceleration into steps.

Thus, our study examined the abovementioned question by first investigating step-output comparability among three devices that have significantly impacted PA research. This was followed-up with a determination of device-specific parameters used to define and detect a step. The three devices were the mechanical Yamax Digiwalker SW200, the piezoelectric cantilever accelerometer-based Omron HJ720ITC, and MEMs capacitive sensor-based ActiGraph monitors. A reason for selecting these three devices was that unlike various consumer-based devices that are currently available, these three brands of devices have significantly influenced the creation of commonly used step goals (e.g., 10,000 steps/day) [13–15] and/or have been widely used in both population-level studies (e.g., NHANES) [16–20] and in clinical trials [15,21–23]. However, these devices also use unknown proprietary strategies to define a step. Differences in proprietary strategies that dictate how a step is detected, may result in the invalidation of a step as a universal metric of PA and compromise inter-study comparisons of this metric. To our knowledge, no study has systematically determined device-specific step-detection parameters among these three brands of activity monitors. Such an examination will provide an improved understanding of "why" movements recorded as a step by one device are not classified as such by another device.

Thus, Aim 1 of our study compared manually-counted steps during treadmill walking to those from the hip-worn Digiwalker SW200 and Omron HJ720ITC, and steps from hip and wrist-worn GT3X+ and GT9X monitors processed using ActiLife software. Aim 2 determined device-specific raw acceleration amplitude (g-value) and/or frequency (Hz) thresholds to record steps. Results of Aim 2 may explain likely differences in step outputs among devices in Aim 1. A secondary aim of our study was to quantify features of high-resolution acceleration signals detected at the hip and the wrist and identify/describe visually distinctive patterns in the signals during walking. Such findings may further explain variability in output from Aim 1 and inform the development of superior non-proprietary step-detection methods when using accelerometer-based step-counters.

#### **2. Materials and Methods**

Twenty participants (Age: 26.7 ± 4.9 years; BMI: 26.1 ± 3.5 kg/m2; 12 males and 8 females; all right-handed) with no gait abnormalities and who were experienced in using a treadmill volunteered for this study. The study was approved by the Northeastern University Institutional Review Board and all participants provided written informed consent prior to commencing study participation. Figure 1 is a conceptual outline of study aims and associated experimental protocols, which are detailed below (Figure 1).

**Figure 1.** Conceptual outline of study aims and associated experimental protocols (acc. = acceleration; amp. = amplitude; freq.= frequency).

#### *2.1. Devices*

The hip-worn Digiwalker SW200 is a coiled spring-suspended pendulum-based mechanical device, which counts steps when vertical acceleration during hip movement sufficiently displaces the pendulum to complete an electrical circuit (Figure 2). The Omron HJ720ITC is typically worn on the hip, has two piezoelectric sensors that are placed perpendicular to each other on its circuit board, and uses proprietary criteria to detect steps in real-time. The ActiGraph GT3X+ and GT9X can be worn on both the hip and wrist. Data from these capacitive accelerometer-based devices [24] can be post-processed to obtain steps using ActiLife software (ActiGraph LLC., Pensacola, FL, USA).

**Figure 2.** Participant with the ActiGraph GT3X+ (#6), GT9X (#3), the Omron HJ720ITC (#4), and the Digiwalker SW300 (#5). Figures (**B**–**D**) depict the sensing elements (#1) inside the three devices.

#### *2.2. Aim 1—Human Treadmill Testing*

Participants wore a single Digiwalker SW200, Omron HJ720ITC, GT3X+, and GT9X at the waist using a snug elastic belt (Figure 2). The GT3X+ and the GT9X were positioned in line with the anterior axillary line on the left and right hip. The Omron HJ720ITC and Digiwalker SW200 were placed medially adjacent to the ActiGraph monitors. A GT3X+ and GT9X were worn at the most distal location on the left and right wrist. Monitor locations (left and right) were counterbalanced among participants to account for any confounding due to placement. Device placement was based on typical procedures followed in PA research, i.e., as per the manufacturer's recommendation (Figure 2).

The experimental protocol consisted of treadmill walking at 11 speeds ranging between 0.89 and 1.79 m/s (2 to 4 mph) in increments of 0.089 m/s (0.2 mph). The protocol yielded steady-state step-counts for one minute at each speed. A researcher manually counted steps (criterion) during the protocol. A researcher verified each treadmill speed using a handheld tachometer (Shimpo DT-105A, Nidec-Shimpo America Corp., Itasca, IL., USA).

#### *2.3. Aim 2—Testing to Examine Step-Detection Thresholds for Acceleration Amplitude (g) and Frequency (Hz)*

Processing simulated signals of known frequency and amplitude using device-specific algorithms lends maximum control and accuracy in determining the minimum acceleration required to detect a step (i.e., step-counting threshold). However, the Omron HJ720ITC and the Digiwalker SW200 do not have associated software or software features that facilitate such processing. Hence, to determine step-detection thresholds, we conducted shaker-testing for the Omron HJ720ITC and Digiwalker SW200, and simulated sinusoidal signal testing for the ActiGraph algorithm.

#### 2.3.1. Shaker Testing (Omron HJ720ITC and Digiwalker SW200)

Our orbital shaker (Model 1231A89, Thomson Scientific, London, UK) maintains a monitor in a fixed spatial orientation, while subjecting the device to a circular motion in the horizontal plane at adjustable frequencies and radii ranging from 0.25 to 8.0 Hz (15 to 480 rpm) and 1.6 to 16 cm, respectively. Both devices were tested at 4 different radii: 2.1, 2.6, 4.5, and 12.0 cm. Since, acceleration thresholds required to commence step counting may be different for these devices, an initial trial was conducted to determine monitor-specific testing frequencies. Testing frequencies were increased at the rate of 0.016 Hz (1 rpm). Each monitor then underwent 10 trials that yielded a minute of steady-state shaker testing at each testing frequency. The Omron HJ720ITC was tested from 0.97 to 1.17 Hz (58 to 70 rpm) at each of the testing radii. The Digiwalker SW200 was tested from 3.73 to 3.83 Hz (224 to 230 rpm) at 2.1 cm, from 3.35 to 3.45 Hz (201 to 207 rpm) at 2.6 cm, from 2.53 to 2.63 Hz (152 to 158 rpm) at 4.5 cm, and from 1.50 to 1.63 Hz (90 to 98 rpm) at 12 cm.

Devices were fastened to the shaker plate such that there was no movement artifact due to placement and each brand of monitor was tested separately. The protocol was video-recorded using a stationary high-speed camera placed at a height of approximately 30-cm from the shaker plate. Time-stamped video recordings were immediately used to retrieve and record steps/min at each frequency from the display screens of the two devices.

#### 2.3.2. Simulated Sinusoidal Signal Testing for ActiGraph

Simulated signal testing has been used to test characteristics of ActiLife software [25–27]. Signals were simulated at a sampling rate of 80 Hz for frequencies ranging from 0.2 to 2.4 Hz, in increments of 0.2 Hz. For each frequency, signals were generated at amplitudes ranging between 0.01 to 0.2 g in increments of 0.01 g. Simulated signals were generated using MATLAB (R2016B, MathWorks, Natick, MA., USA) and were processed using ActiLife v6.13.3 to generate step-output [25,26].

#### *2.4. Secondary Aim—Quantifying Selected Signal Features and Analysis of Acceleration Waveforms (Raw/Filtered) during Treadmill Walking*

We quantified specific amplitude and frequency characteristics of the high-resolution acceleration signal from a GT3X+ worn at the hip and wrist. Walking is rhythmic and produces a relatively consistent waveform pattern with acceleration peaks during a step. Average peak acceleration amplitudes were derived from steady-state walking steps at each treadmill speed. Frequency characteristics included the dominant and 2nd dominant signal frequencies derived using Fourier Transform in MATLAB. We also report visually detectable time-domain patterns in the acceleration signal during walking for raw and filtered acceleration (0.25–2.5 Hz). This bandwidth was selected for this demonstration as previous work has shown that it is sufficient to capture movement related to human ambulation [28].

#### *2.5. Data Analyses*

Step outputs during human, shaker, and simulated signal testing were compiled into a Microsoft Excel database. All ActiGraph data (80 Hz sampling rate) were processed using ActiLife v6.13.3.

For the treadmill protocol (aim 1), mean steps and estimation bias (95% confidence intervals) were derived using IBM® SPSS® Statistics software (*v.* 23, Armonk, NY, USA). Bias was defined as steps from direct observation minus monitor output. Acceleration amplitude and/or frequency thresholds required by a device to detect steps were derived from shaker and sinusoidal signal testing (aim 2). Orbital shaker testing subjected the Digiwalker SW200 and the Omron HJ720ITC to sinusoidal acceleration in the *x* and *y*-axes of the device where each revolution of the orbital shaker generates a curve that contains the two extrema of the acceleration sine wave with a zero-g value crossing. Hence, one revolution is one step. Details on computing peak acceleration amplitude during orbital shaker testing can be found elsewhere [29].

#### **3. Results**

#### *3.1. Aim 1–Human Treadmill Testing*

Table 1 depicts mean criterion and estimated steps and estimation bias (95% confidence intervals) from hip and wrist-worn devices during the treadmill protocol. Statistically significant underestimations were found for the hip-worn Omron HJ720ITC at 1.61 m/s (3.6 mph) and 1.79 m/s (4.0 mph), the hip-worn Digiwalker SW200 at 0.89 m/s (2.0 mph), 0.98 m/s (2.2 mph), and 1.16 m/s (2.6 mph), and both hip and wrist outputs from the GT3X+ and GT9X at all speeds. ActiGraph monitors were least accurate in detecting steps (Table 1).

#### *3.2. Aim 2–Testing to Examine Step-Count Thresholds for Acceleration Amplitude (g) and Frequency (Hz)*

#### 3.2.1. Digiwalker SW200

Table 2 contains mean steps detected during the shaker testing protocol for the Digiwalker SW200 and the corresponding acceleration amplitude computed at the testing frequencies. Our findings suggest that a total acceleration amplitude of at least 1.21 g is required to sufficiently displace the spring-levered mechanism in the Digiwalker SW200 and count a step. We used the lowest g-force when a step is detected to represent step-detection threshold. This accounts for marginal decrements in the g-force produced (during few to several rotations) that is attributable to friction and other environmental factors that impact the rotation of the shaker plate.

#### 3.2.2. Omron HJ720ITC

The Omron HJ720ITC counts a step when the detected motion satisfies both an acceleration amplitude and frequency threshold. Movement with a minimum acceleration amplitude of 0.10 g is required to detect steps (Table 2). At a radius of 2.1 cm, step counting did not commence until a frequency of 1.10 Hz (66 rpm), which generates an acceleration of 0.10 g. At all other radii, i.e., 2.6, 4.5, and 12 cm, acceleration amplitude at 1 Hz (60 rpm) and above were greater or equal to 0.10 g (range: 0.10 to 0.65 g). Our findings suggest that the Omron HJ720ITC records steps for those peaks in the acceleration waveform that have a minimum amplitude threshold of ±0.10 g, but only when the waveform has a dominant frequency ≥ 1 Hz (Table 2).


**Table 1.** Mean (SD) and bias (95% confidence intervals) for steps from the hip-worn Omron, Digiwalker, GT3X+, and GT9X and for the wrist-worn GT3X+ andduring the treadmill protocol. Bias was computed as criterion minus estimated steps; grey cells indicate a significant over, or underestimation of device output. **Table 2.** Shaker testing frequency, corresponding computed acceleration (g), and mean ± SD steps detected by the Omron HJ720ITC and the Digiwalker SW200 during orbital shaker testing. Total acceleration of 1.21 g is required for the Digiwalker SW200 to detect steps. For the Omron, at a radius of 2.1 cm, steps are not detected till an acceleration of 0.10 g is achieved. Although acceleration is ≥ 0.10 g at radii of 2.6, 4.5, and 12 cm, steps are not detected till a frequency of 1 Hz (60 rpm) is reached.


#### *Sensors* **2018** , *18*, 1206

 GT9X

#### 3.2.3. ActiGraph's ActiLife Algorithm

Sinusoidal signal testing demonstrated that the step-detection response of the ActiLife algorithm is variable based on both acceleration amplitude and frequency. Table 3 is a grid that shows the relationship between raw acceleration amplitude and frequency for the ActiLife step counting algorithm. The algorithm was most sensitive between frequencies of 0.7 and 0.8 Hz and required an acceleration amplitude of at least 0.07 g to detect a step (Table 3).

**Table 3.** Variable combinations of acceleration frequency and amplitude thresholds necessary to commence step counting using ActiLife software. 'x' indicates step detection, while '0' indicates the absence of step detection.


*3.3. Secondary Aim-Quantifying Selected Signal Features and Visual Analysis of Acceleration Waveforms (Raw/Filtered) during Treadmill Walking*

Overall, peak acceleration and the dominant signal frequency at the hip and wrist increased with speed (Figure 3A,B). Supplementary Table S1 contains values for the 2nd dominant frequency of hip and wrist movement (Figure 3).

**Figure 3.** Mean ± SD of peak high-resolution (80 Hz) acceleration detected in the *y*-axis of the GT3X+ at the hip (**A**) and wrist (**B**) during walking at various speeds. Dotted lines in 3A indicate step-detection total acceleration thresholds for the hip-worn Digiwalker SW200 (1.21 g) (DW) and the Omron HJ720 ITC (0.10 g) (OM). Please note that although the piezoelectric Omron HJ720 ITC is not sensitive to the force of gravity, we added the gravity component to denote the threshold for the Omron because, the figure depicts total acceleration (gravity component + movement acceleration). Dotted line in 3B is the frequency threshold for the Omron HJ720 ITC (1 Hz). We do not depict thresholds for ActiLife as it is variable based on signal frequency due to bandpass filtering. Additionally, specifications of the filter are proprietary and hence it is not possible to scale the filtered signal thresholds identified in the study to an unfiltered acceleration (g) value in Figure 3A.

Signal visualization (*y*-axis and tri-axial vector magnitude) revealed a consistent pattern where the shape of the acceleration waveforms for consecutive steps looked different, but alternate step patterns were similar. Figure 4 demonstrates this pattern using the signal detected in the *y*-axis of a GT9X at the hip and wrist. To further explore this phenomenon, we conducted video analyses of five participants who wore a GT9X at the dominant hip and the non-dominant wrist while walking at 1.61 m/s (3.6 mph) on a treadmill. Time-stamped videos were synchronized with the *y*-axis waveform from the GT9X at each site. For the hip, we defined an ipsilateral step as that by the leg on the same side as the hip on which the device was worn. The contralateral step was that taken by the opposite leg. Ipsilateral steps taken in concert with a forward arm swing of the non-dominant wrist generated acceleration curves that were larger than contralateral steps and the backward arm swing, respectively. Additionally, compared to raw data, filtered data are smoother and resemble sine-waves. The latter may simplify the detection of movements associated with a step. Thus, it is likely that monitor manufacturers may employ some form of acceleration signal filtering when detecting steps (Figure 4).

**Figure 4.** *y*-axis acceleration (5 s; 80 Hz) from a hip and wrist GT9X on a single participant and their corresponding bandpass filtered signals (0.25-2.5 Hz) processed using a fourth-order Butterworth filter. Dotted boxes indicate the same steps in the raw and filtered signals. 'co'= contralateral step (step taken by the leg on the same side as the hip-worn device); 'ip'= ipsilateral step (step taken by the leg on the opposite side as the hip-worn device); 'f'= forward arm swing; 'b'= backward arm swing.

#### **4. Discussion**

The overarching goal of this study was to examine how inter-device differences in proprietary step-detection strategies impact the comparability of step outputs during human movement. For this, we used devices that have significantly contributed to existing knowledge on the impact of steps/day on health. We first present a discussion that integrates findings from human testing in our study (Aim 1) and previous work with findings from our investigation of step-detection strategies in the three brands of activity monitors (Aim 2). This is followed by a discussion on observations from the secondary aim.

*4.1. Aims 1 and 2—Human Treadmill Testing and Examining Step-Count Thresholds for Acceleration Amplitude (g) and Frequency (Hz)*

#### 4.1.1. Digiwalker SW200

The device in our study required a total raw g-force threshold of 1.21 g to count a step. We placed the device flat on the shaker table, and subjected it to orbital motion in the horizontal plane. Such a placement eliminates the gravity component (1 g) that is exerted on the spring-based pendulum when the device is worn on the body (vertical plane). Thus, the additional vertical acceleration (beyond gravity) required to count a step is 0.21 g. Our empirically determined threshold of 0.21 g for the Digiwalker SW200 is lower than the previously reported threshold of 0.35 g [30]. Supporting evidence for our estimate can be inferred from the raw acceleration detected by the ActiGraph at the hip (Figure 3A). At speeds between 0.89 m/s (2 mph) and 1.25 m/s (2.8 mph), step-detection accuracy increased from 79 to 92%. At these speeds, mean ± SD of peak acceleration (including the gravity component) ranged between 1.30 ± 0.21 to 1.45 ± 0.22 g. Given that (i) peak acceleration increases with increasing speed; (ii) there is considerable variability in peak acceleration at each speed (~0.21 g, Figure 3A); and (iii) contralateral steps generate lower acceleration than ipsilateral steps (Figure 4); increasing walking speed results in an incremental proportion of steps that satisfy the acceleration threshold to detect a step, thereby resulting in a corresponding increase in step-detection accuracy (Table 1).

#### 4.1.2. Omron HJ720ITC

The acceleration amplitude threshold of 0.10 g in the Omron HJ720ITC does not include the gravity component, which is attributable to the use of piezoelectric sensors to detect steps. Piezoelectric sensors do not detect static acceleration and produce an electric voltage that is proportional to movement only. We found that beyond the 0.10 g threshold, subsequent increases in the amplitude of acceleration did not alter the frequency threshold of 1 Hz (60 rpm). Thus, the hip-worn Omron HJ720ITC may cause larger step-detection errors during slow walking that may be attributable to a low step frequency rather than an insufficient acceleration amplitude because during walking, it is improbable that acceleration at the hip will not exceed a threshold of 0.10 g. Correspondingly, Jehn et al. reported that a step-frequency between 80–90 steps/min was necessary for the Omron HJ720ITC to accurately detect steps [31].

#### 4.1.3. ActiLife

The ActiLife step-counting algorithm is based on proprietary thresholding of the bandpass filtered uni-axial (*y*-axis) acceleration signal. Here, the filtered acceleration waveform needs to cross both positive and negative values of a fixed proprietary acceleration amplitude threshold to qualify as a step. Consecutive positive and negative crossings of the amplitude threshold are equidistantly separated by a zero-g crossing, which is used as an event that marks a valid step [32]. Bandpass filtering eliminates the gravity component, which yields a zero-g crossing and a low amplitude threshold for step-detection (see Figure 4).

*Variability in acceleration thresholds:* Although step-counting in ActiLife is based on a fixed acceleration amplitude threshold, we found that at different signal frequencies, the step-detection amplitude threshold is variable (Table 3). This is due to bandpass filtering of the raw acceleration signal that occurs prior to signal thresholding to detect steps (described above). ActiGraph's bandpass filter is maximally sensitive at 0.75 Hz. As frequency components in the signal move away from 0.75 Hz within the passband (0.25 to 2.5 Hz), the amplitude of the signal is reduced symmetrically [27]. At 0.29 and 1.66 Hz, signal amplitude is reduced to 70.7% of that at 0.75 Hz [27]. This filter response explains peak sensitivity to detect steps between 0.7 and 0.8 Hz and the need for increasing acceleration amplitude (g-value) with increasing frequency to detect steps (table 3). During commonly preferred walking speeds of approximately 1.34 to 1.56 m/s (3 to 3.5 mph) in normal adults [33,34], the dominant frequencies in the acceleration signal at both the hip and wrist in our study were close to or greater than 1.66 Hz [1.78 ± 0.28 Hz at 1.34 m/s (3 mph) at the hip; 1.63 ± 0.46 Hz at 1.52 m/s (3.4 mph) at the wrist (Figure 3B)].

*Effect of monitor location on step*-*counting:* We are unaware of the reasons causing differences in step output between the hip-worn GT3X+ and GT9X. However, we discuss causes for differences between hip and wrist-worn ActiGraph devices and between wrist-worn GT3X+ and GT9X devices.

The ActiLife step-counting algorithm was developed for hip-worn devices and uses acceleration from only the *y*-axis to detect steps. Acceleration detected at the wrist are smaller in magnitude than those at the hip during walking at the same speed (Figure 4). This will result in fewer instances where the acceleration waveform satisfies thresholding criteria, especially at slower walking speeds. However, during a 24-h period under free-living conditions, a wrist-worn GT3X+ resulted in higher step counts than a hip-worn device [8]. This paradoxical finding may be explained by the fact that many activities of daily living involve hand movements that result in the detection of several false-positive steps.

A significant source of error between a wrist-worn GT3X+ and GT9X is a discrepancy in the orientation of the tri-axial coordinate system due to a dissimilarity in "form-factor" (Figure 5A,B). This inter-device discrepancy in axis orientations will significantly impact step counting accuracy between the GT9X and other ActiGraph devices at the wrist (Figure 5).

**Figure 5.** Orientation of the axes in the GT3X+ and GT9X when worn on the hip (**A**) and wrist (**B**), respectively.

Differences in individual gait, location of wear, and walking speed may impact thresholding of the signal in ActiLife, and thereby step-counting accuracy. These factors may yield a filtered signal that may not (i) cross both negative and positive thresholds required to count a step, or (ii) produce a zero crossing.

#### *4.2. Secondary Aim—Quantifying Selected Signal Features and Visual Analysis of Acceleration Waveforms (Raw/Filtered) during Treadmill Walking*

Although stepping involves displacement of the body's center of mass, which results in hip movement, the magnitude of peak-to-peak acceleration between two consecutive steps varies for hip-worn devices. During the contralateral step, i.e., a step taken by the left leg when the device is worn on the right hip, and vice versa, the sensor captures attenuated accelerations (Figure 4). This is because the device is not on the hip that is causing the step. At slower walking speeds, smaller accelerations detected during the contralateral step may not satisfy the step-detection criteria of the Digiwalker SW200 and ActiGraph, which may yield fewer step-counts (Figure 4).

At the wrist, the step waveform during the forward swing of the arm, tends to produce a distinctly larger waveform as compared to that during the backward arm swing. However, during walking, arm movement may not always reflect whole body movement (e.g., when holding an umbrella). Thus, arm swing patterns during walking may be less consistent than those observed at the hip. Signal processing techniques that are not limited to simple thresholding, but which also identify frequency features that are inherent in the signal during walking, may be useful to detect steps when wrist movement is restricted during ambulation.

#### *4.3. Strengths and Limitations*

A strength of this study is the use of three proprietary devices that have significantly influenced PA research. These devices use different hardware and strategies to detect steps, a universal metric of PA. Our study used human, mechanical, and software-based testing to demonstrate variability in step-detection parameters among tested devices, and how variability translates to discrepancies in steps. Weaknesses include limiting human testing to treadmill walking that may not represent natural over-ground walking, and exclusion of popular consumer wearables.

#### **5. Conclusions**

The Digiwalker SW200, the Omron HJ720ITC and the ActiGraph devices have significantly influenced the understanding of the dose–response between steps/day and health outcomes. However, the parameters to define and detect a step in all three devices are different.

	- - Step-detection threshold for the ActiGraph's ActiLife algorithm is variable based on signal frequency due to signal filtering.
	- - Our findings suggest that current step-detection strategies in ActiLife may be invalid for step detection using devices worn at both the hip and the wrist.

#### *5.1. Implications of Variability in Proprietary Step-Detection Parameters*

Similar to our study, previous work found that steps from hip-worn ActiGraphs and Digiwalker perform poorly at slower walking speeds [<~1.12 m/s (2.5 mph)], while Omron devices detect most steps at speeds above 0.89 m/s (2 mph) [3,35,36]. Another study reported that the Omron HJ720ITC was unreliable at slow walking speeds <0.83 m/s (1.85 mph.). Thus, the proprietary step detection thresholds in these three devices may be inappropriate to detect steps during slow ambulation, particularly when used in populations where gait speed is markedly reduced (e.g., older adults, disabled populations). Studies comparing the hip-worn Omron HJ720 ITC, the Digiwalker SW200, and ActiGraph devices to a criterion ankle-worn step-counter have reported significant underestimations of steps during 24-h free living [3,37,38]. Additionally, existing evidence that the Omron HJ720ITC does not count steps accumulated in durations lasting less than 4 s [38] in combination with findings on the frequency threshold from our study suggest that the step-counting

algorithm in the Omron HJ720ITC was designed to capture purposeful, rhythmic bouts of walking. However, activities of daily living often involve short/intermittent stepping bouts that may last few to several seconds [39,40]. Thus, several of these steps may go undetected by the Omron HJ720ITC.

From a public health perspective, our findings may explain differences in steps/day estimates for the US population [20,41]. One study using a spring-levered pedometer reported that adults averaged approximately 5117 steps/day [41]. Conversely, based on NHANES 2005–2006 accelerometer data, American adults accumulate 9676 steps/day. This nearly 2-fold difference in steps/day from two US population-level studies may be attributable to differences in step-detection parameters in the two devices used in these studies. Given a rapid turnover of new models for each brand of popular consumer wearables, such devices are likely more suitable for end-users in the general population. However, the adoption of consumer wearables in PA and public health research is quickly increasing [42,43]. Most of these devices use proprietary methodologies to detect steps and hence, differences may exist in the types of movements that are captured as steps. For example, one study examining steps during numerous simulated activities of daily living (including ambulation) demonstrated significant differences in step outputs between the Apple watch and ground truth [44]. Another study using a wrist-worn Fitbit reported that the device underestimated free-living steps by up to 30% as compared to directly-observed ground truth [45]. Recent work found that proprietary methods in the Apple Watch and the FitBit Blaze yield significantly different steps during 19 different activities and during free-living conditions [46]. One factor may be that the Apple watch likely uses a summarized acceleration derived from each axis of a triaxial accelerometer as compared to the FitBit Blaze, which uses individual information from each of the three axis [46].

#### *5.2. Outlook for the Future*

Devices and methodologies that use proprietary and/or rudimentary acceleration threshold-based strategies may insufficiently capture steps during purposeful ambulation and activities with intermittent movements. Furthermore, to establish steps as a standard metric of PA, it may be necessary for the research community to have a consensus on movements (and related acceleration signal characteristics) required to represent a step. Such a consensus framework may then be used to expedite the development of *a non*-*proprietary, open*-*source* step-detection methodology that does not rely solely on simple acceleration thresholds, but considers multiple acceleration features to classify movements as a step. This may help to standardize the step as a metric of PA. Similar approaches may be necessary when considering new methods for clinical populations to ensure uniformity across devices used in a specific pathological condition that alters gait.

**Supplementary Materials:** The following are available online at www.mdpi.com/1424-8220/18/4/1206/s1, Table S1: Mean ± SD of the second dominant frequencies (Hz) detected in the acceleration signal during the walking protocol.

**Author Contributions:** All authors were involved in the conception of the study, interpretation of data and findings, and drafting/editing of the article. Alvin Morton and Diego Arguello were involved in collecting data for the study.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

#### *Article*

### **Older Adults with Weaker Muscle Strength Stand up from a Sitting Position with More Dynamic Trunk Use**

#### **Rob C. van Lummel 1,2,\*, Jordi Evers 1, Martijn Niessen 1, Peter J. Beek <sup>2</sup> and Jaap H. van Dieën <sup>2</sup>**


Received: 16 March 2018; Accepted: 12 April 2018; Published: 17 April 2018

**Abstract:** The ability to stand up from a sitting position is essential for older adults to live independently. Body-fixed inertial sensors may provide an approach for quantifying the sit-to-stand (STS) in clinical settings. The aim of this study was to determine whether measurements of STS movements using body-fixed sensors yield parameters that are informative regarding changes in STS performance in older adults with reduced muscle strength. In twenty-seven healthy older adults, handgrip strength was assessed as a proxy for overall muscle strength. Subjects were asked to stand up from a chair placed at three heights. Trunk movements were measured using an inertial sensor fixed to the back. Duration, angular range, and maximum angular velocity of STS phases, as well as the vertical velocity of the extension phase, were calculated. Backwards elimination using Generalized Estimating Equations was used to determine if handgrip strength predicted the STS durations and trunk kinematics. Weaker subjects (i.e., with lower handgrip strength) were slower during the STS and showed a larger flexion angular range and a larger extension angular range. In addition, weaker subjects showed a greater maximum angular velocity, which increased with lower seat heights. Measurements with a single inertial sensor did reveal that older adults with lower handgrip strength employed a different strategy to stand up from a sitting position, involving more dynamic use of the trunk. This effect was greatest when elevating body mass. Trunk kinematic parameters were more sensitive to reduced muscle strength than durations.

**Keywords:** physical function; physical performance test; chair stand; sit to stand transfer; wearables; inertial sensors; accelerometers; gyroscopes

#### **1. Introduction**

The ability to stand up from a sitting position is essential for older adults to live independently and maintain an adequate level of physical activity. Older adults with a better sit-to-stand (STS) performance compared to age-matched controls showed shorter sitting periods, longer standing periods, and a higher number of locomotion periods in daily life, which indicates a more active lifestyle [1]. In community-dwelling older adults, sedentary behavior caused in part by difficulty to stand up was associated with increased risk of sarcopenia [2] and mortality [3–5].

Previous research clarified the dynamics of the STS movement to better understand its dynamics. Schenkman et al. distinguished four STS phases, which include the flexion momentum phase, the momentum transfer phase, the vertical extension phase, and the stabilization phase [6]. Riley et al. found indications that the momentum transfer phase, which starts with the lift-off from the chair seat,

represents the most demanding phase [7]. The STS involves a transition from an intrinsically stable three-point support to a dynamically stable two-point support [7]. Difficulty in rising from a seated position may directly increase the risk of injury since STS transfers were found to be responsible for 41% of all falls in nursing homes [8].

Muscle strength represents one of the most important factors contributing to success in rising from a chair [9] and has been identified as a factor determining the lowest chair height from which the functionally impaired elderly can rise [10]. STS transitions require the development of substantial muscle power and consequently many older adults perform such transitions close to their maximal ability [10,11]. In addition, functionally limited elderly individuals with lower quadriceps muscle strength showed a lower dynamic stability when performing the STS at their preferred speed [12]. Several studies have reported different strategies for standing up [13–15]. In general, functionally impaired elderly stand up with greater flexion of the upper body [16], possibly to achieve better postural stability [13]. Age-related reduction in muscle strength may determine the choice of STS strategy [17]. Using a muscle-actuated optimal control model, Bobbert et al. [18] found that optimizing the STS strategy can reduce the mechanical demands on all muscles by 45% compared to a normal STS. In particular, an STS strategy with greater trunk flexion led to reduced demands on the knee extensors. This finding is in line with the effects that are observed when subjects are instructed to increase trunk flexion [14]. Overall, the existing body of evidence suggests that older adults, especially adults with muscle weakness, may show adaptations in the manner in which they perform the STS.

In clinical practice, STS performance is usually evaluated by measuring the time it takes to perform a series of STS movements. Since the early 1990s, body-fixed inertial sensors have been used to quantify STS movements [19–24]. Single sensor instrumentation can be easily applied in an unobtrusive manner with high measurement reliability in a geriatric setting [25]. Instrumented measurements of STS performance yield several additional parameters that may provide a basis for a quantitative assessment of STS performance in a clinical practice setting [26]. Repeated STS movements with automatically detected sub-durations were proven to have stronger associations with health status, functional status, and physical activity than manually recorded STS durations, which implies greater clinical relevance outcomes of instrumented analysis than manually recorded durations [27]. Furthermore, instrumented analysis allowed for the assessment of the test's dynamic phases, which are likely more informative than the static sitting and standing phases [27].

The aim of this study was to investigate whether measurements of STS movements using body-fixed sensors yield parameters that are informative regarding changes in the STS kinematics in older adults with reduced muscle strength by using handgrip strength (HGS) as a proxy for overall muscle strength. HGS is an easy to use method to measure muscle strength in a clinical practice setting and has frequently been associated with overall and lower extremity muscle strength [28–30]. It was hypothesized that weaker adults (i.e., with lower handgrip strength) would be inclined to use an STS strategy involving a more dynamic use of the trunk.

#### **2. Materials and Methods**

#### *2.1. Participants*

Twenty-seven healthy older adults, living in sheltered housing or in the community (12 females; mean age: 73.8 ± 7.9 years; mean weight: 77.1 ± 13.2 kg; mean height: 173.4 ± 7.4 cm) participated in this cross-sectional study. The protocol was approved by the ethics committee of the Department of Human Movement Sciences of the Vrije Universiteit Amsterdam (ECB 2014-3M) before it was conducted. Prior to testing, all participants provided written informed consent.

#### *2.2. Instrumentation and Data Acquisition*

Trunk movements during STS were measured using a small and light (87 × 45 × 14 mm, 74 g) inertial sensor measurement system (DynaPort Hybrid, McRoberts, The Hague, The Netherlands), which was fixed with an elastic belt around the waist and placed over the spine. The experimental set-up is depicted in Figure 1.

**Figure 1.** The experimenter stayed in close proximity to the patient. The protocol was implemented on a computer. With a remote control, the start and end of every STS was marked and stored with the raw data.

This sensor measured acceleration and angular velocity in three directions at a rate of 100 samples/s (see Figure 2). A single sensor was used because this is more practical for clinical use than multiple sensors. A position near the center of mass was chosen to obtain an adequate reflection of the whole body movement [31]. Moreover, at this position the sensor was unobtrusive, easy to fasten, and did not hamper the participant's movements.

**Figure 2.** The main phases of the STS movement. The top panel shows the time series of raw acceleration (green/dashed-mediolateral, red/dotted anteroposterior, and blue/solid vertical), the lower panel shows the angular velocity (green/dashed-pitch, red/dotted-roll, and blue/solid -yaw) signals, and the insets illustrate the main phases of the STS movement, which are separated by vertical lines in both graphs.

#### *2.3. Test Protocol*

Participants were asked to stand up at their preferred speed from a height adjustable chair because effects of muscle strength are likely to be greater when standing up from lower chairs. Practice was allowed if needed. After standing up, participants were required to stand still for 10 s. Each participant performed two STS movements from three different chair heights at 100%, 90%, and 80%. The 100% chair height was defined according to the operational definition used by Schenkman [6] with the chair adjusted in such a manner that the participant's thighs were horizontal with shanks and feet

situated in a preferred orientation to stand up. The three chair heights were determined prior to performing the trials. The STS protocol was implemented on a computer, which randomly assigned the order of the height conditions. Markers set with the remote control at the start and end of each trial were stored with the raw signals to enable automatic data analysis. The three height conditions were performed with the arms folded in front of the trunk. When participants were unable to perform a certain condition, it was skipped and the next condition was offered.

#### *2.4. Signal Analysis*

The measurement of three-dimensional accelerations and angular velocities of the trunk allowed for a detailed analysis of the STS movement's different phases (see Figure 2). The acceleration and the angular velocity in the sagittal plane were used to calculate the trunk pitch angle [32]. Drift and noise were removed from the trunk pitch angle using the discrete wavelet transformation [19]. The dips in the trunk angle were used to detect a change in the trunk rotation direction. The start of the STS was defined as the end of the plateau before the first dip in the trunk pitch angle. Similarly, the end of the sit-to-stand was defined as the start of the plateau after the first dip in the trunk pitch angle. 'True vertical acceleration' was estimated by removing the influence of the trunk pitch angle from the vertical acceleration signal. Subsequently, the vertical velocity was derived by integrating this signal.

The following events were defined using commercially available software (MoveTest®, McRoberts®, The Hague, The Netherlands): the start of trunk movement, the end of the trunk flexion phase, and the end of the trunk rising phase [33] (see Figure 2). After automated identification of these phases, the duration, angular range, and maximum angular velocity of these phases as well as the vertical velocity of the extension phase were calculated. The mean of the two repetitions was used for statistical analysis.

#### *2.5. Handgrip Strength (HGS)*

HGS of both hands was measured using a digital handgrip dynamometer (Takei A5401) following a standardized protocol. Participants were standing in an upright position with their arms alongside their body. The handle was adjusted to fit their hand size. They were asked to squeeze the dynamometer as forcefully as possible.

Maximal HGS was measured twice for both hands with brief pauses between each measurement. The mean of all four measurements was used to estimate HGS. To assess the maximal HGS, at least three attempts are needed if HGS is considered to be a continuous variable while two attempts are sufficient to assess age-associated loss of muscle strength not caused by neurological or muscular diseases (dynapenia) [34].

#### *2.6. Statistical Analysis*

Normally distributed characteristics were presented as the mean and standard deviation (SD). Skewed (non-Gaussian) distributed continuous variables were presented as the median and interquartile range (IQR).

#### 2.6.1. Checks for Normality

STS parameters were checked for normality using the Kolmogorov-Smirnov test. Out of a total of 24 parameters, 8 were not normally distributed. Total STS duration and the duration of the flexion phase did not meet the criteria for normality for all three seat heights. For these duration parameters, skewness and kurtosis ranged from 2.0 to 2.1 and from 3.0 to 5.9, respectively. Also non-normal distribution included the maximum angular velocity during the flexion phase at the 80% chair height (D(24) = 0.187, *p* < 0.05) and the duration of the extension phase at the 90% seat height (D(25) = 0.206, *p* < 0.05).

#### 2.6.2. Analysis

All outcome variables were analyzed using Generalized Estimating Equations (GEE) to examine whether they were associated with HGS. GEE was chosen instead of a conventional least-squares regression because it can cope with missing values and takes into account that observations within a data set are dependent. In our study, these were the different seat heights for all subjects. In addition, body length, weight, age, and gender were treated as potential confounders. The main effects were calculated for HGS and these potential confounders. Because the effect of handgrip force was of interest and an interaction with task difficulty was plausible, the interaction between handgrip strength and seat height was also included in the initial model. The first iteration of the GEE analysis included HGS, HGS\*seat height, and all potential confounders. A backward elimination method was used to eliminate non-significant predictors. Seat height was always included in the model since this was the within-subject variable. All analyses were conducted in SPSS v. 21.0.

#### **3. Results**

From the 27 participants, 24 completed the entire protocol. One participant was unable to stand up from the 80% chair height while two participants were unable to stand up from both the 90% and 80% chair heights. All participants (i.e., also the three participants that did not complete the entire protocol) were included in the analysis. Participant characteristics and descriptive statistics are presented in Table 1.


**Table 1.** Descriptive statistics of study population.

Note: Data represents the mean ± SD or median (Q1–Q3).

The results of the durations and kinematics of the three seat heights and the model effects of GEE are presented in Table 2.

#### *3.1. Sit-To-Stand*

There was no significant effect of seat height on STS duration (*p* = 0.097). HGS was the only significant covariate showing that weaker subjects had a longer STS duration (*p* = 0.015).

#### *3.2. Flexion Phase*

There was no significant effect of seat height on flexion duration (*p* = 0.071). Angular range increased with lower seat height (*p* = 0.003), weaker subjects used a greater angular range (*p* < 0.001), and taller subjects used a greater angular range (*p* = 0.019). Age was the only covariate showing a significant effect on maximum angular velocity (*p* = 0.004) since angular velocity decreased with age.


*Sensors* **2018** , *18*, 1235

**Table 2.** movement

 for the three seat heights. Model effects of Generalized

 Estimating

 Equations are displayed for the condition factor Seat Height and for the

Durations (s), angular range (<sup>φ</sup> in ◦), angular velocity (ωmax in ◦/s) and vertical velocity (vmax in m/s) during flexion and extension of the Sit-to-Stand

#### *3.3. Extension Phase*

The duration of the extension phase tended to be longer for lower seat heights (*p* = 0.062) and was significantly longer for lower HGS (*p* = 0.003). The angular range was larger for lower seat heights (*p* = 0.002) and for lower HGS (*p* < 0.001) in the presence of a significant interaction effect of seat height and HGS (*p* < 0.001). Maximum angular velocity was higher for lower seat height (*p* < 0.001) and for lower HGS (*p* < 0.001). In addition, there was a significant seat height\*HGS interaction (*p* = 0.002). Older subjects showed a lower angular velocity (*p* = 0.026). Vertical velocity was higher for lower seat heights (*p* < 0.001) and for higher HGS (*p* = 0.011). Furthermore, vertical velocity was higher for taller subjects (*p* = 0.048), lower for older ages (*p* = 0.025), and higher for males (*p* < 0.005).

The most important outcomes are visualized in the scatterplots depicted in Figure 3. The four panels show the relation between handgrip strength (horizontal axes) and the STS duration, the flexion angular range, the extension angular range, and the extension angular velocity, respectively. As is evident from panel A, stronger subjects (i.e., with greater handgrip strength) performed the STS faster. Panel B shows that stronger subjects had a smaller flexion angular range while panel C shows that the weaker subjects had a much larger extension angular range, which increased with lower seat heights. Panel D shows a comparable effect for the maximum angular velocity during extension.

**Figure 3.** Scatter plots display values of the four STS parameters and handgrip strength at three different seat heights. Panel (**A**) shows the total duration of the SiSt phase. Panel (**B**) shows the angular range during the STS flexion phase. Panel (**C**) shows the angular range during the STS extension phase. Panel (**D**) shows the maximal angular velocity during the STS extension phase.

#### **4. Discussion**

The results showed that older adults with lower handgrip strength stand up with greater flexion of the trunk than older adults with higher handgrip strength. The weaker subjects also showed greater trunk extension with a higher angular velocity after seat-off, indicating a more dynamic use of the trunk. The hypothesis that older adults with lower muscle strength stand up with more dynamic use of the trunk was, therefore, confirmed. This finding is consistent with the results of Hughes et al. showing that functionally impaired elderly rising from chairs below knee height significantly increased peak hip flexion velocity and the time to rise when compared to rising from a knee height chair. In addition, they significantly decreased their mean center of mass/base of support (COM/BOS) separation at lift-off [16] in line with the suggestion that this strategy may facilitate balance control [13].

The single module instrumentation can be easily attached over undergarments beneath outer clothes in a manner that is unobtrusive to the subject. The participant's awareness of being assessed is low because the instrumentation is not visible for the participant. Data collection is fast and, with the remote control, the test leader can stay in close contact with the participant. Moreover, the online connection of the remote control makes it possible for a single test leader to simultaneously collect data and watch over the participant since it is no longer necessary to read out the stopwatch and write down the times. The automated analysis of the data provides detailed insight into the quality of the movements.

Physical performance tests that include the STS such as the Timed Up and Go [35] and the repeated STS as a sub-test of the Short Physical Performance Battery [36] measure the duration of the total test and provide no information about the STS sub-durations and kinematics. The duration of the static periods (standing and sitting) during the 5× repeated STS represents, at least for some older adults, a great part of the total duration [27]. The total STS duration as measured with a stopwatch provides incomplete information about the dynamic phases of the STS, which are particularly relevant. The present findings demonstrate that kinematic parameters of standing up might be useful for clinical practice. Trunk range of motion, maximum angular velocity, and vertical velocity might be used to identify STS strategies in clinical practice and to evaluate interventions.

Chair height has a major influence on the ability to successfully stand up from a sitting position. A previous study reported that older adults found that, in general, the use of hands during STS transfer from different bed heights made the performance less challenging. All participants reported that the lowest position was the most challenging position to perform the task [37].

Using backwards elimination from all covariates, HGS was proven to be the strongest predictor of the durations as well as the kinematic outcomes. Older adults with lower handgrip strength, therefore, seemed to resort to a different STS strategy, which involved a more dynamic use of the trunk. The present and previous studies have shown that muscle strength is likely an important determinant of successfully standing up from a sitting position [10,12,38,39]. However, besides strength training, the training of STS transfer skills may well form a critical element in interventions aimed at preventing loss of mobility in older adults and patients [40]. Improvement of the technical performance of the STS might be more sustainable than strength training. Analysis of STS strategy using body fixed sensors in combination with easy-to-use strength measures will help to objectively measure changes over time and better understand how older adults could improve their STS ability.

Several determinants like chair seat height and lower limb strength have a major influence on the ability to successfully stand up from a sitting position. Increasing lower limb strength is a possible solution. Seat height may need to be more closely scrutinized in areas frequented by frail elderly people. Augmentation of seat height by small increments facilitates chair rise performance. In the present study, we manipulated seat height. Standing up from a lower seat height commonly occurs in daily life such as when standing up from a sofa. However, the seating area of a sofa is longer and it includes a backward slope and a compliant seat [41]. Therefore, standing up from a sofa might be even more difficult.

Adaptations of STS behavior have also been described in patients with other health conditions. Patients with knee osteoarthritis primarily shifted their COM closer to the base of support provided by the feet in an attempt to achieve stability at and after seat-off [42]. Additionally, individuals with Parkinson's disease have difficulty rising from a chair as a result of reduced muscle strength, particularly at the hip amongst other factors [43]. In patients with a stroke, longer total time as well as changes at the initial phase and at the end of hip and knee extension phase were observed. The maximal hip flexion was lower during the rising phase from a seated position on the affected side in the stroke group [44]. Patients with end stage renal disease undergoing hemodialysis have high morbidity and mortality due to multiple causes. The results indicated that patients were slower to get out of the chair, which was measured by trunk flexion angular accelerations, time to peak trunk flexion, and overall STW completion time following the dialysis therapy session [45]. The present results add to these findings by showing that healthy older adults with reduced muscle strength also show significant adaptations in STS behavior, which is characterized by a more dynamic use of the trunk.

A limitation of the present study involves not measuring lower extremity strength and instead using HGS as a proxy for global muscle strength even though lower extremity muscle power was proven no better than knee extension torque or handgrip strength in the early identification of poor mobility [46]. Nevertheless, the change to a more dynamic STS strategy should be tested in a study protocol including lower leg strength measurements. To date, only a few studies have demonstrated the added value of using the instrumented STS in interventions [47,48]. The cross-sectional design of the present study precluded an analysis of how the STS strategy changes over time as muscle strength changes. It would be useful to examine whether older adults do revert to another STS strategy after an intervention is aimed at increasing lower leg strength. Longitudinal studies are needed to shed light on this issue. Finally, in the present study, using the arms during the STS procedure was prohibited by the protocol. The reason for this limitation was due to a focus specifically on the role of muscle strength and we, therefore, wished to eliminate the use of arm support or movement to compensate for lower leg strength. Future research could focus on how arm use might affect the STS dynamics.

#### **5. Conclusions**

Older adults with lower handgrip strength employed a different strategy to stand up from a sitting position than older adults with higher handgrip strength, which was characterized by more trunk flexion and a more dynamic use of the trunk during the extension phase. Changing to another strategy seems to be an adaptation of older adults to their reduced muscle strength. Measures of trunk kinematics were more sensitive to muscle strength than durations of STS phases.

**Acknowledgments:** We are very grateful to the older adults who volunteered to join this study and we thank Gijs van Cappelle for his support in the data collection.

**Author Contributions:** R.C.v.L., J.E., and J.H.v.D. conceived and designed the experiments. J.E. and R.C.v.L. performed the experiments. J.E. and M.N. analyzed the data. J.H.v.D., J.E., and R.C.v.L. contributed reagents/ materials/analysis tools. R.C.v.L., J.E., J.H.v.D., and P.J.B. wrote the paper.

**Conflicts of Interest:** R.C.v.L. is the owner and J.E. and M.N. are employees of the company McRoberts B.V., which is the manufacturer of the DynaPort.

#### **References**


© 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

MDPI St. Alban-Anlage 66 4052 Basel Switzerland Tel. +41 61 683 77 34 Fax +41 61 302 89 18 www.mdpi.com

*Sensors* Editorial Office E-mail: sensors@mdpi.com www.mdpi.com/journal/sensors

MDPI St. Alban-Anlage 66 4052 Basel Switzerland

Tel: +41 61 683 77 34 Fax: +41 61 302 89 18