Limitations of Foot-Worn Sensors for Assessing Running Power

unning power as measured by foot-worn sensors is considered to be associated with the metabolic cost of running. In this study, we show that running economy needs to be taken into account when deriving metabolic cost from accelerometer data. We administered an experiment in which 32 experienced participants (age = 28 7 years, weekly running distance = 51 24 km) ran at a constant speed with modi ed spatiotemporal gait characteristics (stride length, ground contact time, use of arms). We recorded both their metabolic costs of transportation, as well as running power, as measured by a Stryd sensor. Purposely varying the running style impacts the running economy and leads to signi cant differences in the metabolic cost of running (p< 0.01). At the same time, the expected rise in running power does not follow this change, and there is a signi cant difference in the relation between metabolic cost and power (p< 0.001). These results stand in contrast to the previously reported link between metabolic and mechanical running characteristics estimated by foot-worn sensors. This casts doubt on the feasibility of measuring

At the same time, the expected rise in running power does not follow this change, and there is a signi cant difference in the relation between metabolic cost and power (p< 0.001). These results stand in contrast to the previously reported link between metabolic and mechanical running characteristics estimated by foot-worn sensors. This casts doubt on the feasibility of measuring running power in the eld, as well as using it as a training signal. Keywords: accelerometer; running power; running economy; Stryd; metabolic cost of transportation 1. Introduction In the last decade, foot-worn sensors to assess and meaningfully analyze running metrics (e.g., step frequency, stride length, ground contact time) have gained increased attention and popularity [1–5]. These sensors are meant to improve laboratory and in- eld testing and training by delivering key performance data. The data derived from foot- worn sensors comprise spatiotemporal running parameters, including running power [1,6]. These devices are designed to be independent of factors such as slope, wind resistance, or fatigue. There has been signi cant prior research related to the use of foot-worn ac- celerometers for different purposes: they have been used for and are known to correlate with ground speed [7], running economy [8], and running power [6,9]. In this research, we aim to clarify whether it is possible to reliably measure running power using foot-worn sensors and use these data to objectively quantify effort. Being able to objectively evaluate and compare training effort are considered valuable to improve training programming and progression [10]. For example, two runners may cover a certain distance in the same duration, yet they experience different levels of exhaustion or underlying energy costs of running. This is mainly due to a complex interplay among the central and peripheral properties of cardiocirculatory, ventilatory, metabolic, and psychological capacities [11]. In practice, energy demands are measured as the rate of oxygen consumption normalized to body weight, specified as relative oxygen consumption VO2[mL min 1 kg 1 ][ . A reliable assessment of such exchange data via spiroergometric systems is applicable to Sensors2021,21, 4952.

cardiocirculatory, ventilatory, metabolic, and psychological capacities [11]. In practice, energy demands are measured as the rate of oxygen consumption normalized to body weight, specified as relative oxygen consumption VO2[mL min 1 kg 1 ][ . A reliable assessment of such exchange data via spiroergometric systems is applicable to Sensors2021,21, 4952.

Sensors2021,21, 4952 2 of 12 in- eld, as well as laboratory conditions. However, these systems can be considered a bit cumbersome and, therefore, require trained professionals to assess and interpret the data. Thus, the direct assessment of the energy output of an athlete without a spiroer- gometric mask seems to be promising from the perspective of athletes and coaches. In cycling, a common measure to evaluate absolute effort is power, measured in Watts. Power (=work per time) is often determined at a single point in the human–bike interface: the pedal. This deduction appears reasonable as the energy transferred from an athlete onto the street passes through this transmission point. Evidence shows that there is a very strong association between metabolic (measured by oxygen consumed) and mechanical power at the pedal in cycling (r= 0.97) [14]. With this background, it seems analogously plausible to measure running power also in a similar setup: The foot serves as the interface where the athlete's effort is exerted. Force plates would then be the point of reference to measure ground reaction force (GRF) and derive power applied to the ground. However, this setup would not be feasible in the eld. Alternative solutions for GRF measurement using specialized insoles are still expensive and still need to be validated. Hence, numerous companies have introduced small devices that can be easily attached to the shoelaces. These compact and consumer-friendly devices contain accelerometers [15]. By using Newton's second law, the force at the foot can be approximated via F = m a. A direct transfer of measurement parameters from cycling to running power, however, is problematic. The motion on a bike is very constrained, whereas runners move with more degrees of freedom [16]. Runners additionally use the muscle–tendon unit to regain and, thereby, conserve energy with each step [17]. These factors can be trained and altered and, therefore, affect the ratio between metabolic energy consumption and mechanical energy output. There are a number of factors in uencing the relation between energy consumed and energy expended in propulsion [18,19], commonly referred to as running economy (RE) [16,20]. In this paper,

to regain and, thereby, conserve energy with each step [17]. These factors can be trained and altered and, therefore, affect the ratio between metabolic energy consumption and mechanical energy output. There are a number of factors in uencing the relation between energy consumed and energy expended in propulsion [18,19], commonly referred to as running economy (RE) [16,20]. In this paper, we try to show that economy-related aspects need to be taken into account when considering running power. The primary objective of this work, then, is to clarify the information meant to be gained from foot-worn sensors and to present the results of a simple experiment that manipulates athletes' RE. We investigated the effects on energy intake ( VO2) and power produced (PW) when altering running parameters, such as stride length and step frequency. By keeping the running speed constant while simultaneously altering the ef ciency of energy utilization, we created a gap between VO2andPWas measured by a foot-worn sensor. We propose that we can alter VO2signi cantly whilst the power outputPWdoes not increase accordingly. This nding suggests that we are, in fact, not ready to measure metabolic running power in daily training with existing ready-to-use consumer products. 2. Methods 2.1. Participants and Study Design A total of 32 moderate endurance-trained runners (males (n = 22), females (n = 10)) participated in this controlled crossover trial. The participants' characteristics (mean SD) in- cluded: age = 28 7 years, BMI = 21.6 1.6, running VO2max= 58.6 13.1 mL min 1 kg 1 , and weekly running distance = 51 24 km over the prior four weeks. All participants had at least one year of distance running experience and were injury-free for at least the preceding three months. Participants chose their own footwear to best resemble their individual training and racing conditions. This study was approved by the local ethics committee (Ethics Commission German Sport University Cologne, Ethical Proposal No. 017/2021) and was in compliance with the declaration of Helsinki [21]. All participants were informed about the study design, and they signed informed consent for participation. The entire test protocol was carried out

to best resemble their individual training and racing conditions. This study was approved by the local ethics committee (Ethics Commission German Sport University Cologne, Ethical Proposal No. 017/2021) and was in compliance with the declaration of Helsinki [21]. All participants were informed about the study design, and they signed informed consent for participation. The entire test protocol was carried out in a single laboratory visit and was divided into two parts. First, each subject performed a combined incremental and ramp exercise test to determine their velocity at the aerobic threshold (AeT) [22]. For this, an incremental maximal oxygen uptake ( VO2max) treadmill test was conducted until objective exhaustion levels were reached [13]. Thereafter, following a 30 min rest, the participants ran another

Sensors2021,21, 4952 3 of 12 25 min on the treadmill with varying predetermined spatiotemporal running parame- ters. Throughout all tests, the participants wore a heart rate transmitter chest strap and receiver/watch (Garmin, Olathe, KS, USA). Athletes refrained from intense exercise 48 h prior to the test. 2.2. Testing Procedures 2.2.1. Combined Incremental and Ramp Exercise Test In order to determine their VO2max, as well as their AeT, the participants performed a combined incremental and ramp testing protocol. The initial speed of the subjects was set based on prior running experience and estimated 10 km race time at 2, 2.5, or 3 m s 1 . The combined protocol then consisted of the following steps: four 3 min stages (0.5 m s 1 speed increase per stage, 30 s rest in between), immediately followed by 90 s at the same speed as the last stage, and then a ramp test (0.2 m s 1 speed increase every 30 s) until the subjects reached volitional exhaustion [13]. In addition, capillary blood samples were taken from the earlobe of the participants for lactate analysis (EBIOplus; EKF Diagnostic Sales, Magdeburg, Germany) during the 30 s rest periods between stages and immediately after the ramp test. VO2max data were collected, using a breath-by-breath spirometric system (Zan 600, Zan Messgeräte, Oberthulba, Germany). This spirometric system was calibrated prior to each test, following the manufacturer's recommendations. The highest consecutive oxygen uptake values within 30 s during the nal part were considered as VO2max. VO2maxand objective exhaustion were veri ed for each participant following the criteria by Midgley and colleagues [13]. All participants were verbally encouraged and motivated in the same way towards the end of VO2maxtesting, and they ful lled objective exhaustion criteria (i.e., at least 4 out of 6 criteria). During a 30 min rest period between the all-out exhaustion test and the second part of the study, the athletes' AeT was determined using theminimum lactate equivalent(Lmin) method [22]. In order to do this, we t an exponential function to the lactate measures and the velocities measured during the four stages. Thereafter, we determined Lminas the

out of 6 criteria). During a 30 min rest period between the all-out exhaustion test and the second part of the study, the athletes' AeT was determined using theminimum lactate equivalent(Lmin) method [22]. In order to do this, we t an exponential function to the lactate measures and the velocities measured during the four stages. Thereafter, we determined Lminas the minimum of the ratio between this function and the velocity, using custom-built Python functions. Running power during all tests was measured using a Stryd sensor (Stryd Summit Powermeter, rmware 2.1.16;, Stryd, Inc., Boulder, CO, USA). This sensor was attached to an athlete's shoe using a clip in the shoelaces and connected to a watch via Bluetooth. This device is lightweight (8.5 g), unobtrusive (4 cm length), and did not impact the athlete's running form [15]. It was independently validated to strongly correlate with VO2in a recent study by Cerezuela-Espejo et al. [6]. In their study, which included two foot-worn and three additional sensors, the Stryd sensor resulted in the strongest correlation between displayed power and the metabolic cost of running. The sensor was used in accordance with the manufacturer's instructions, was reset between participants, and set to the weight and height of each athlete. It was attached to the bottom laces of the left foot and stayed xed for a single subject between tests. Data from the Stryd sensor and the heart rate monitor were sampled at 1 Hz and syn- chronized during the recording by a Garmin watch. Spirometric data were also recorded at a 1 Hz frequency and aligned manually during analysis. Potential errors due to misalign- ment were mitigated through data pooling. 2.2.2. Modulation of Running Economy In the second part of the study, the subjects ran for 25 min at a constant pace at their AeT, as calculated in the previous test. Before the test, each subject was instructed on the interrelation among step frequency, step length, and ground contact time while keeping a steady pace. They were also familiarized with the metronome (Weird Metronome, David Johnston,, cf. [ 23]). After a brief warm-up period of

for 25 min at a constant pace at their AeT, as calculated in the previous test. Before the test, each subject was instructed on the interrelation among step frequency, step length, and ground contact time while keeping a steady pace. They were also familiarized with the metronome (Weird Metronome, David Johnston,, cf. [ 23]). After a brief warm-up period of 2 min, the next 3 min were used to determine the subject's self-selected preferred step frequency (SF) at that pace. Next, at the same pace, the participants were instructed to

Sensors2021,21, 4952 4 of 12 perform four different variations to their running style for 3 min each. Between variations, the participants ran in their own unrestricted running form for 1 min. The four conditions in this experiment were as follows (order randomized between participants): • SF+10%: an increase of the step frequency by 10%. This was prescribed using a digital metronome. Participants were also verbally encouraged and supported when the target frequency was not met; • SF 10%: a decrease in step frequency by 10%. Again, a metronome was used to help the participants keep this running form. As additional mental help, the participants were instructed to lengthen their stride, as if they were gliding; • GCT: reduction of ground contact time. The participants were instructed to reduce the time spent in contact with the ground by around 20 ms. They were told the GCT during their self-selected running and a target GCT for this variation. Participants were instructed regularly to either keep their step exactly as is or to try and decrease their GCT further. As a mental image, participants were encouraged to imagine the treadmill to be covered in hot coals; • Arms: The participants were instructed to run without arm swing and, thus, without counterbalance to their running motion. The arms were either held above the head or in the neck in order to avoid effective use as a counterweight to rotational movement. After completing the four conditions, the participants continued running in their own running style for another 5 min. We used the VO2towards the end of this cool-down, to approximate fatigue. Assuming a linear relationship between time on the treadmill and fatigue, we then used this level of oxygen consumption to remove drift from the data of the four conditions, thus avoiding data artifacts due to randomization. Additionally, we used these data to investigate a “fatigued” condition. Figure a single subject. Panel (a) on the top displays the raw VO2andPWvalues over time. The vertical lines and alternating background colors signify the different phases of the test, as noted on the very top of Panel (a).

data of the four conditions, thus avoiding data artifacts due to randomization. Additionally, we used these data to investigate a “fatigued” condition. Figure a single subject. Panel (a) on the top displays the raw VO2andPWvalues over time. The vertical lines and alternating background colors signify the different phases of the test, as noted on the very top of Panel (a). These data rows were converted into single values by pooling over the last minute of each phase, displayed here as the horizontal bars. Figure 1. Protocol for the running economy variation test, all performed at a xed speed. Example of a single subject. Raw values for all measurements are plotted over time (measured at 1 Hz). (a) Oxygen consumption and power output during the running conditions. (b) Spatiotemporal running parame- ters in reaction to instructions for the experimental conditions. VO2for self-selected SF, 4 conditions and cool-down (average for last minute of each phase, cf. horizontal lines,* = signi cant difference): 2.06, 2.36 *, 2.49 *, 2.08, 2.47 *, 2.07.P W: 160.62, 159.84, 170.88, 156.34, 155.62, 160.58. Best viewed in color.

Sensors2021,21, 4952 5 of 12 In Figureb, we show the changes in the spatiotemporal running parameters that the subject adopted in order to ful ll the given instructions. As expected, the values changed in accordance with each other. Notably, this subject implemented the condition GCT by increasing the cadence. Other elucidations of this condition included highly elongated strides or shorter step lengths at the runner's usual cadence. 2.3. Statistics We calculated all the statistics in Python-3.6.8 using the scipy.stats-1.2.1 library. Through- outthe evaluations, signi cance levels ofp< 0.05 were set. We performed a repeated measures analysis of variance (ANOVA) to investigate the changes in VO2in response to the varying instructions during the six conditions (baseline, SF+, SF , arms, GCT, fatigue). We further modeled the adaptations in VO2by the participants in an unambiguous way as follows. 2.3.1. Oxygen Consumption during Altered Running Conditions The raw data for VO2andPWwere recorded at 1 Hz. For each subject, we took the last 1 min of a condition to calculate the average VO2andPWfor further analysis. We determined signi cant changes in VO2by analyzing the noise during the recording process. For this, we normalized each data row of the baseline condition (Minutes 3–5 of running) to its mean and t a Student t-distribution to the resulting combined noise for all participants. This noise incorporated both errors stemming from the inaccuracies of the recording device, as well as uctuations in the breathing patterns or other physiological changes. The resulting model described the variation in VO2that we expected to see in every measurement. We could then compare the baseline data rows for each subject with the data for all the conditions. If more than 10% of the data points in a condition were within the 95% con dence interval (CI) of the baseline measurement, we considered this condition to not be a signi cantly different VO2. This perspicuous process corresponded exactly to a t-test for signi cance withp< 0.05 with the added bene t that we could directly mark the nonsigni cant data points (cf. Figure 3, hollow points). 2.3.2. Difference in Slope of Power to Oxygen

The study investigates the reliability of foot-worn sensors in measuring running power.