The effect of familiarization on squats done with a robotic exoskeleton

1. Introduction

For new exercise hardware validity, which refers to the similarity of its data to a device deemed the “gold standard” by practitioners, must be established before novel equipment are accepted by the scientific community [23]. Perhaps this is most true for hardware to be used during space flight, as microgravity is a novel setting in which humans incur significant muscle mass and strength losses [5, 20]. Hardware to abate in-flight losses must impart sufficient resistance with minimal space, mass and power needs in order to comply with the space craft environment. They include a robotic exoskeleton (Institute of Human and Machine Cognition; Pensacola, FL, USA) that enable squats to be performed without requiring gravity to operate or is impacted by a user’s body mass. This is achieved by a pair of actuators that create a loading stimulus independent of gravity. A side view of the exoskeleton, and depiction of squats done on the device, appear in Fig. 1. Hardware validity is impacted by the degree and type of familiarization given prior to testing [10, 28, 32]. Familiarization aims to limit learning effects with repeated testing to reduce systematic errors to the eventual data collected [28].

OPEN IN VIEWER

To date however, only one published paper examined the validity of the exoskeleton’s squat data [27]. That paper used Z-scores to identify outliers, and t-tests, Cohen’s d and Pearson product moment correlation (PPMC) coefficients to compare data from separate workouts done with the exoskeleton and a barbell [27]. Results included a large range of values. Roughly 10% of the total data set were outliers. Average t-test, Cohen’s d and PPMC values were 1.6, 0.51 and 0.37 respectively. The lower validity was attributed to the lack of familiarization to squats done on the exoskeleton, as the device was only available to the investigators for a limited time [27]. It was concluded more research was needed before exoskeleton could serve as in-flight hardware.

Familiarization reduces the intra-subject, and ultimately inter-workout, variability seen in studies that include repeated testing [10]. Intra-subject variability may be considered the most important type of measurement in validity studies [10]. As test data is collected, prior familiarizarion should improve data stability and validity of the device under inquiry [14, 15]. Familiarization may abate learning effects prior to repeated testing, yet some think it is not necessary [10, 26, 39]. Differences may be related to methodological issues. Some claim if familiarization entails a fixed protocol whereby a set number of repetitions and/or trials are practiced, it does not adequately address intra-subject variability, as individual subjects learn study-specific tasks at different rates [10].

Robotic exoskeletons, like the one in Fig. 1, are under consideration for use during long-term space flights such as manned missions to Mars. Yet the lack of familiarization, which likely led to higher intra-subject variability and lower validity values in the prior study, reduces the likelihood of the Fig. 1 device’s use as in-flight hardware [27]. The objective of our study was therefore to compare physiological, performance and perceptual responses derived from squats done on the Fig. 1 exoskeleton to those from a barbell, whereby the latter serves as the “gold standard”. Similar inter-workout responses help establish the exoskeleton’s validity and may improve the likelihood of its use for long-term manned space flights. Unlike the prior study [27], our subjects performed two familiarization sessions on the exoskeleton prior to two workouts. We hypothesize that squats on the exoskeleton, with familiarization prior to testing, will evoke similar inter-workout responses as those produced when the same exercise is done with a barbell.

All subjects completed four visits and none were injured from their project participation. Z-scores show less than 1% of our data were outliers and MANOVA results included non-significant differences. Anthropometric data for our female, male and total sample appear in Table 1. The raw data for our free weight metabolic, performance, HR and RPE variables, appear in Table 2. Corresponding exoskeleton values appear in Table 3. Tables 2 and 3 data were normally distributed. Table 4 shows t-test, Cohens d and PPMC inter-workout values for metabolic, performance, HR and RPE dependent variables. Results generally show more inter-workout peak force, as compared to peak velocity, variability. In particular, lighter loads had more peak force variability. Table 4 t-test results show 17 dependent variables had values less than 1.0. Cohens d results show eight dependent variables had values less than 0.2 and 17 produced values less than 0.4, while only one exceeded 0.5. PPMC results varied greatly. Generally lower PPMC values were produced by our performance-based dependent variables. In contrast, generally better relationships occurred with our post-exercise HR, metabolic and perceptual-based dependent variables.

Tables 5 and 6 display the raw VL sEMG data outcomes from our free weight and exoskeleton workouts respectively. Tables 5 and 6 data are presented by load, repetition and contractile mode. Our sEMG data were normally distributed. Inter-workout sEMG t-test, Cohens d and PPMC results appear in Table 7 and are displayed by load, repetition and contractile mode. ECC, as compared to CON, actions produced more variability. Results include only three inter-workout t-test values that exceed 1.0. They all occurred with ECC actions done against 23 kg. Cohens d results show 14 and 32 sEMG dependent variables less than 0.2 and 0.4 respectively.

Due to the similarity of our Bland-Altman plots and the number of dependent variables examined, we chose one plot to represent our entire data set. Figure 3 shows a Bland-Altman plot of our pre-exercise [BLa - ] data. It has dashed lines equal to ± 1.96 standard deviations, sometimes referred to as the limits of agreement. Since Z-score analyses previously omitted outliers, this plot, like those for all our dependent variables, had no outliers. Across all dependent variables (Tables 4 and 7 combined), average inter-workout t-test and Cohens d values equaled 0.65 and 0.24 respectively. PPMC coefficient results generally denote “moderate” or “high” relationships for dependent variables collected from our two workouts. The average inter-workout PPMC coefficient value, across all the current study dependent variables (Tables 4 and 7 combined) was 0.52. In general, there was considerable inter-workout similarity for our dependent variables.

OPEN IN VIEWER

Challenges to develop resistive exercise hardware for long-term space flights persist, such that some question if manned missions to Mars are tenable and safe for humans [5, 21]. Since it exceeds projected exercise hardware space, mass and power limits for Mars missions, NASA’s Advanced Resistive Exercise Device (ARED) is impractical for such flights [5]. In contrast, with its ability to provide a gravity-independent loading stimulus for the squat exercise without ARED’s limitations, the robotic exoskeleton is potential hardware that may address this issue. However the validity of the exoskeleton has been questioned [27]. Current results affirm our hypothesis with multiple statistical analyses that each examined the inter-workout similarity of our dependent variables.

Paired t-tests assess absolute inter-workout differences [27]. To achieve acceptable levels of validity, values less than 1.0 are desirable [27]. Paired t-tests were used to examine the differences between actual and age-estimated maximal heart rate and VO 2 max values [3]. The results included a large range of values that led to significant differences, as age-estimated maximal HR values averaged five beats per minute more than actual values, and age-estimated VO 2 max values varied 11–19% from actual results [3]. Across all dependent variables examined, the prior exoskeleton study produced an average t-test value of 1.6 [27]. The t-test results reported in that study included two dependent variables that were significantly ( 0.05) different [27]. In contrast our paired t-tests, across all dependent variables examined, produced an average value of 0.65. No current t-test values were significant, which suggests greater validity than the prior study [27].

Cohen’s d measures relative inter-workout differences. Cohen’s d values of 0.2 and 0.5, referred to as “small” and “moderate” differences respectively, were derived from the behavioral and social sciences [31]. It was suggested, due to differences in the treaments used and the potential for change among variables, those values were inappropriate for strength training research [31]. Revised Cohen’s d values for strength training suggest that those less than 0.8 denote small effect sizes for the type of subjects examined in our study [31]. Our results include, across all dependent variables examined, an average value of 0.24; in contrast the prior exoskeleton study had an average value of 0.51 [27]. Regardless of which scale is used, our results suggest minimal inter-workout differences.

Bland-Altman analyses plot delta measurements for paired values as a function of their mean [17]. They also identify heteroscedasticity and outliers; the former occurs when observed differences become greater as mean values increase, which implies increases in variability are a function of absolute measurement magnitude [17]. Despite their utility, few exercise studies employed Bland-Altman plots. Gait performance was examined over repeated test sessions done by stroke patients and identified 6% of the data as outliers [17]. Prior exercise studies with Bland-Altman plots examined intra- and inter-workout responses to novel hardware [6, 8]. Results included much test-retest agreement and less than 1% of the data were outliers, all inter-workout values [6, 8]. Our Bland-Altman results are like those from prior studies that administered familiarization prior to testing and used similar types of subjects [6, 8].

PPMC coefficients are used to assess validity [1, 11, 33, 35, 38]. Our PPMC coefficients produced an average value, across all dependent variables examined, of 0.52. Prior exercise studies used PPMC coefficients to examine RPE values as a correlate to HR in order to assess the validity of the perceptual measurements [9, 29, 37]. A mean PPMC value of 0.58 was seen between HR and RPE for exercise done by African-Americans [22]. In young females the HR-RPE PPMC coefficient was measured at 0.66 [29]. An unusually high correlation ( r = 0.98–0.99) between the aforementioned variables was elicited by tethered swimming [37], yet a meta-analysis of the HR-RPE relationship saw a mean PPMC coefficient of 0.62 [9]. Finally, the prior exoskeleton study produced an average PPMC value of 0.37 [27]. Thus our study’s average PPMC coefficient value approximates or exceeds those from prior studies. Differences in PPMC values between the current and prior studies may include, but are not limited to, methodological issues; as prior studies compared two distinctly different variables obtained at the same time for an assessment of concurrent validity, while we compared the same dependent variables from different workouts in order to examine predictive validity [9, 29, 35, 37, 38].

Intra-subject variability is considered by some to be the most important measurement in validity studies [10]. Our results show more variability for peak force, versus peak velocity, values. This is likely because our metronome controlled the cadence at which subjects squatted and limited our peak velocity variability. In contrast there was no procedure that controlled force exertion. Yet current results, particularly in light of the prior exoskeleton study, reinforce the need for familiarization [10, 27].

The ability of familiarization to yield similar inter-workout responses is due to several factors that include the type and duration of familiarization, the length of inter-test time intervals and types of subjects used [13, 24, 30, 32]. In regards to the latter, athletes may need less familiarization than sedentaries [16, 19, 25]. In addition subjects’ age, which may relate to their physical activity levels, impacted familiarization requirements to measure knee extensor strength; with less familiarization needed for younger, versus older, women [30]. More intense exercise tests also warrant greater familiarization, as higher degrees of physical exertion elicit learning effects when measured over multiple sessions [4, 14, 39]. In support of the last statement, 40 men did eight test sessions, without prior familiarization, to measure peak squat strength [28]. Only 30% achieved their peak strength by the third session, and over 50% needed at least six sessions [28]. Without familiarization, it was implied at least 9–10 trials were needed to obtain accurate values. It was concluded that familiarization was needed before tests when large muscles were engaged, and the number of sessions before tests should increase above that which is usually administered [28].

Different types of familiarization exist. They all aim to limit the intra-subject variability in subsequent test data in order to elicit similar inter-test session responses. While most familiarization strategies entail a fixed number of sessions, repetitions and durations to the stimulus under inquiry, they may not be as helpful to subjects who learn study procedures more slowly. Thus a fixed familiarization protocol lessens the validity of data provided by slow learners, since their subsequent test data exhibits less stability over repeated measurements [10]. In contrast we administered an individualized familiarization protocol, which did not limit subject’s exposure to the exoskeleton. Instead they did as many practice repetitions as they wished until they felt comfortable exercising on the device. We felt individualized familiarization was necessary given the exoskeleton’s novelty. Our results suggest subsequent workouts produced more stable data and less intra-subject variability. A recent vertical jump study also used individualized familiarization prior to testing and concluded it reduced intra-subject variability and improved statistical power [10]. Thus our hypothesis was affirmed, and we had less data variablity than the prior exoskeleton study, not only because we familiarized subjects, but also due to the type of familiarization administered [27].

While our statistical analyses supported our hypothesis, current results should be interpreted cautiously since there are study limitations. Though the exoskeleton is being considered for use aboard long-term manned space flights, our squat protocol is likely not sufficient to abate the lower body muscle mass and strength losses incurred by astronauts [20, 34, 36]. In addition, the exoskeleton used in this study has a maximum loading capacity of 59 kgs. Loading capacities of 136–182 kg for the squat exercise are more appropriate to reduce the in-flight muscle mass and strength losses produced by long-term space missions [20, 34, 36]. However due to its novelty and considerable promise as in-flight hardware, continued research with the robotic exoskeleton is warranted.