Use Treadmills with Caution: Walking Energy Expenditure and Metabolic Cost Are Elevated Compared to Overground Across Multiple Speeds in Healthy Young Adults

Abstract

Objectives: Treadmill walking is often employed for tightly controlled gait and energetics research, but growing evidence suggests that treadmill-based metabolic and biomechanical measurements may not directly reflect the ecologically valid mode of overground walking. While many previous studies focused on older adults, much less is known about how treadmill walking influences gait energetics and spatiotemporal parameters in young healthy adults across matched speeds. We investigated energy expenditure, metabolic cost of walking and spatiotemporal gait parameters in healthy young adults walking overground and on a treadmill at three speeds (slow—1.0, comfortable—1.3, fast—1.5 m/s). Our hypothesis was that at the comfortable speed, treadmill and overground energetics and gait parameters would be comparable. However, at slow and fast speeds, there would be a significant energetic penalty, accompanied by significant differences in spatiotemporal parameters. Methods: Twenty young participants (10 males and 10 females) completed a randomized cross-over walking protocol with a minimum of ten minutes treadmill familiarization at 1.3 m/s. Breath-by-breath oxygen consumption ($\dot{V}{O}_2$) and Respiratory Exchange Ratio were measured using a portable indirect calorimetry system and gait parameters were calculated from Inertial Measurement Units. Gross and net energy expenditures, costs of walking, cadence, average step and stride lengths, and walk ratio were calculated. A three-way mixed ANOVA was used for primary statistical analyses. Results: Treadmill walking was characterized by higher gross and net energy expenditures and metabolic costs (p < 0.001, η_p² = 0.6) across all speeds compared to overground. It was also characterized by faster cadence and shorter average step and stride lengths (p < 0.001, η_p² = 0.9). Additionally, there was an effect of sex (p = 0.01, η_p² = 0.3) on the gait parameters, with females exhibiting a faster cadence and shorter average step and stride lengths than males. Conclusions: Our findings show that treadmill walking imposes a medium-to-large metabolic penalty even in healthy young adults, with compensatory gait adaptations, possibly reflecting increased stabilization demands and altered neuromuscular control strategies. These results underscore the limits of generalizing treadmill derived gait data to overground walking and we caution against the uncritical use of treadmills, especially while trying to understand ecologically relevant human walking mechanics and energetics.

1. Introduction

Human walking is a fundamental aspect of locomotion that helps us to go from one location to another while minimizing metabolic costs, maintaining stability and optimizing task performance. Walking economy, or the Metabolic Cost of Walking (MCoW), is defined as the amount of energy used per unit of distance traveled and generally expressed after body-mass normalization. While overground walking is the ecologically valid way of walking, treadmills are extensively used in laboratories as a proxy for daily-life walking. This is because multiple speeds can be tightly controlled and continuous metabolic and biomechanical data can be easily measured. However, it remains ambiguous whether humans, in general, adopt an altered and less economical gait on treadmills and consequently have an elevated MCoW across speeds. Looking at the published literature in this field, we see that a consistent elevation in MCoW for treadmill walking compared to overground is reported in Older Adults (OAs). At the same time, there is a lack of published studies involving Young Adults (YAs) and the results are inconsistent, compared to those involving OAs. Therefore, it is imperative to compare MCoW and gait parameters between overground and treadmill walking, across a range of matched speeds in young males and females, to better understand the reasons for elevated MCoW in this population.

In general, MCoW is either expressed as the Gross Cost of Walking (GCoW), which reflects the total amount of metabolic energy consumed at the whole-body level, or in terms of Net Cost of Walking (NCoW), which is calculated by subtracting the Resting Metabolic Rate (RMR) from the Gross Energy Expenditure (GEE) ([1]). Many past studies ([2,3,4,5]; among others) have consistently shown that OAs increase antagonistic muscle co-contraction when walking on a treadmill and this is associated with an elevated MCoW. Ref. [6] reported higher oxygen consumption at matched speeds in OAs, whereas ref. [7] observed higher MCoW across multiple speeds in OAs, on treadmills compared to overground. Furthermore, ref. [8] have shown similar elevated treadmill MCoW at same speeds even in YAs, though the results in YAs are fewer in number and more equivocal ([9,10,11,12]). In previous work, ref. [13] found no NCoW difference at the same Preferred Walking Speed (PWS) between treadmill and overground in YAs, although OAs did show an elevation. Very recently, a comprehensive systematic review and meta-analysis by [14] showed that treadmill walking in healthy YAs and middle-aged adults significantly increases relative oxygen consumption and produces significant differences in kinematics and kinetics, with marked increases in cadence, while reducing stride and step lengths. Taken together, there are many differences between overground and treadmill walking energetics and so we cannot presume metabolic similarity between the two modes of walking.

From the published literature, we see that beyond metabolic differences, overground and treadmill walking are also biomechanically different, even when the speeds are matched. Treadmill walking is often characterized by higher cadence, shorter step and stride lengths, altered gait variability, and increased demands on balance and sensorimotor control ([15,16,17,18,19]). Several other studies also explored kinematics, stability constraints and short-term control of gait parameters on treadmills and found that they are different compared to overground ([20,21,22]). All these treadmill-specific gait features possibly arise due to differences in visual flow, and proprioceptive and vestibular control, compared to overground. All in all, both from a metabolic and biomechanical perspective, there are several differences between overground and treadmill walking and so this shows that the apparent equivalence that is assumed between daily-life overground and treadmill walking is not correct. As a result, it is important to know the true metabolic and biomechanical gait differences between overground and treadmill walking, which warrants further empirical research.

In addition to these commonly known differences, a much-understudied source of variability in treadmill walking studies arises from sex-related differences. Especially due to a shorter height and smaller legs, females typically walk with higher cadence and shorter step and stride lengths than males at matched speeds ([23,24,25]). Despite these differences, it remains unclear whether males and females respond similarly to the altered biomechanical and neuromuscular control constraints imposed by treadmill walking, compared to overground. Since it is well known that MCoW is more closely related to muscle force production, force-rate, and stabilization demands than to kinematics alone ([26,27]), similar (treadmill-specific) gait adaptations across sexes may nevertheless result in different metabolic consequences. Consequently, it becomes important to look at sex-related differences, since treadmill-specific constraints could interact with sex-related factors and failing to account for them would obscure actual gait differences between overground and treadmill walking.

Considering the existing literature and, importantly, to resolve the equivocal differences in YAs, we wanted to answer whether young, healthy adults incur a significant metabolic penalty on treadmills compared to overground across a range of speeds, and if so, can we explain it from some basic differences in spatiotemporal parameters? Additionally, we wanted to see, if we find significant anthropometrical differences between young males and females, can we also see significant differences in their MCoW and gait parameters? In this study, we had a few confirmatory hypotheses and explorative premises. To capture a range of walking speeds, we decided to have three walking speeds and defined them as: slow—1.0 m/s, comfortable—1.3 m/s and fast—1.5 m/s. Our confirmatory hypotheses were as follows:

H1: 

Gross and Net Energy Expenditures (GEE and NEE) will increase with increasing walking speeds;

H2: 

As seen in our previous study, [13], at the comfortable speed, which is close to the PWS of humans, there will be no significant difference in GEE, NEE, MCoW and spatiotemporal gait parameters between treadmill and overground; and

H3: 

At the slow and fast speeds, treadmill walking will show significant elevations in GEE, NEE and MCoW, accompanied by significant differences in spatiotemporal gait parameters. There will also be significant positive associations between the (walking surface-related) differences in the metabolic variables and the differences in the spatiotemporal parameters in this case.

Exploratively, we posited that if we see significant differences in basic anthropometrics between the young males and females, there would be sex-specific differences in spatiotemporal gait parameters, GEE, NEE and MCoW between treadmill and overground walking.

2. Materials and Methods

2.1. Participants and Ethics Declaration

A total of 10 healthy young Japanese males (mean age 23.3 ± 1.4 years) and 10 healthy young Japanese females (mean age 21.0 ± 1.9 years) were recruited for this study. The sample size was estimated and kept in the range of our previous studies [13,28] and the other likewise studies cited in [13]. Furthermore, the current study was part of a larger data collection campaign on energy expenditure modeling from wearable Inertial Measurement Units (IMUs), and we followed a recent study ([29]) to estimate our sample size and increased our sample size, compared to that study. Since there is a general lack of research in investigating the metabolic differences between treadmill and overground walking involving healthy YAs, and due to the equivocal results that we can see from the related published literature, we did not do an informed apriori power analysis. All participants were free from any specific medications, diabetes, neurological, musculoskeletal and cardiovascular diseases, past surgeries or lower-limb prostheses and had not experienced a fall in the past 6 months. The participants were able to perform regular daily chores without any assistance and were physically active. None of them were undergoing any special strength/endurance or gymnasium training programs. All participants were briefed about the experiments, their participation was completely voluntary and they could leave the experiment at any time. All of them gave written informed consent. Before starting the experiments, basic anthropometrics were measured: body-mass, height, waist circumference, dominant hand and leg, and Lower Limb Lengths (LLL) from the greater trochanter to malleolus and foot. Body Mass Index (BMI) and Waist-to-Height ratio were calculated from the measured variables. The ergonomics ethical review committee of the National Institute of Advanced Industrial Science and Technology (AIST, Tokyo, Japan) approved the experimental protocol (reference number: 2024-1431, approval number: 73002030-E-20240617-001) in accordance with the principles stated in the Declaration of Helsinki.

2.2. Pre-Experiment Factors

Before any measurements were taken, we advised the participants to have only a light breakfast or lunch (moderate intake of fats and carbohydrates and minimal intake of protein and fiber-rich foods) and to not consume any alcohol or caffeine-related products. They were specifically told not to smoke tobacco until the end of the experiments. We also ensured that at least 2 h of time was elapsed from their last food intake to any metabolic data measurement, to minimize the thermic effects of food. For the overground walking trials, the participants walked along a straight-line walkway, with 5 m straights, taking smooth turns at the end of the tracks, inside a laboratory (see Figure 1). For the treadmill walking trials, a floor-level treadmill (T101, Horizon Fitness Inc., Cottage Grove, WI, USA) was used, which was placed at approximately 20 cm from the ground (wheels to belt). Handrails were not provided during the treadmill walks and free arm swing was ensured. To minimize fatigue, we gave at least 4 min of seated rest between every walking trial and at least 10 min of seated rest while changing from one walking mode to another. We pseudo-randomized the walking speeds and half of the participants completed the overground walking trials first and the other half completed the treadmill walking trials first (see Figure 1).

2.3. Preparation

The participants were fitted with 4 IMUs (Movella DOT, Movella Inc., Henderson, NV, USA), on the right thigh, right shank, left thigh and left shank, and the data were collected at a sampling rate of 60 Hz. This sampling rate was chosen to enable real-time streaming of the IMU data to the dedicated mobile application, to be checked for any issues by one of the experimenters. A schematic diagram is shown in Figure 2. However, in this study we analyzed only the right thigh and right shank IMU data as the right side was the dominant side for all the participants. They were also fitted with a face mask (Hans Rudolph Inc., Shawnee, KS, USA) and a portable Mobile Aeromonitor AE-100i (Minato Medical Science Co., Ltd., Osaka, Japan) to measure rates of oxygen consumption and carbon dioxide liberation. All the devices were properly calibrated every day, before the start of each experiment, according to the manufacturer’s guidelines. All the participants were provided with Mizuno Wave Rider shoes (Mizuno Corporation, Osaka, Japan), according to their foot sizes, by the research team at AIST.

2.4. Data Collection

Though there were some participants who had previous treadmill running experience, none of them had prior treadmill walking experience. Nevertheless, we decided to familiarize each of our participants for at least 10 min on the treadmill at the Comfortable Walking Speed (CWS) of 1.3 m/s in the beginning. After that we gave them 5 min of seated rest and then, between the next 10 to 15 min, they were fitted with the IMUs, the face mask and the Mobile Aeromonitor. This ensured that there was at least a 15 min interval between the completion of their treadmill familiarization and any metabolic measurements, to ensure that all participants returned to their metabolic baseline. Before the start of any of the walking trials, we measured the seated RMR (see Figure 1) for 5 min (see [13,28]). After the resting measurement, all of them walked for at least 6 min at each of the walking speeds both overground and on the treadmill. For the overground trials, an experimenter timed the participants’ walking time on the track with a stopwatch during their walks on the 5 m overground track and ensured that within the first 2 min of the walks, all the participants were walking at the desired speed condition. If needed, verbal encouragements were given in native Japanese to ensure that the participants kept walking at the target speeds. In this way, we ensured that after they had achieved steady state and during the last 4 min of every walking trial, the participants were all walking at the desired speeds. The metabolic steady state was further visually confirmed by checking the wirelessly transmitted data from the Mobile Aeromonitor to its dedicated laptop, during the resting and the walking trials. At the end of the overground trials, the participants were verbally told to stop walking and then all the measurement devices were stopped. For the treadmill trials we increased the speed from 0 m/s to the desired speed within the first 1 min of the start of the walk and at the end, the treadmill speed was slowly reduced from the set speed to 0 m/s again within 1 min and then all the measurement devices were stopped.

2.5. Calculation of the Energy Expenditure and Metabolic Cost of Walking and Related Gait Parameters

For the resting trial, we rejected the first and last 30 s of data from the 5 min of constant sitting and calculated the RMR. The RMR (in J kg⁻¹ s⁻¹ or Watts kg⁻¹) was calculated using the resting Respiratory Exchange Ratio (RER) and the rate of resting oxygen consumption ($\dot{V}{O}_{2resting}$_, in l of O₂ kg⁻¹ s⁻¹) using the Lusk equation ([30]). For every walking trial, we did not analyze the first 3 min of data, as it was used to ensure a metabolic steady state, and the last 30 s of data, as it was used to stop all the measurement devices. From the steady state region of the walking trials, all the necessary metabolic variables were calculated after averaging. The Gross Energy Expenditure (GEE, in J kg⁻¹ s⁻¹ or Watts kg⁻¹) was calculated using the walking RER and the rate of walking oxygen consumption ($\dot{V}{O}_{2walking}$_, in l of O₂ kg⁻¹ s⁻¹) using the same Lusk equation ([30]). From the GEE, we subtracted the RMR (in J kg⁻¹ s⁻¹ or Watts kg⁻¹) to get the Net Energy Expenditure (NEE, in J kg⁻¹ s⁻¹ or Watts kg⁻¹). Then, we divided the GEE and NEE by the Walking Speed (WS, in m s⁻¹) to get the Gross Cost of Walking (GCoW) and the Net Cost of Walking (NCoW) in J kg⁻¹ m⁻¹. These were the primary variables of interest in this study.

Additionally, we calculated a few spatiotemporal parameters. The total number of gait cycles of every walking trial was calculated from the shank IMU data (from the vertically oriented z-axis, sagittal plane angular velocity from the gyroscope) using an open-source algorithm from [29], assuming steady-state bilateral symmetry. We used the algorithm for treadmill walking as in [29] and extended it for overground walking in this study. Although (in-lab) overground walking generally involves taking turns in addition to straight-line walks, it has been previously reported [31] that shank IMUs are also valid for detecting gait events during turns in overground walking, with high accuracy. Furthermore, as we knew the total walking times for each completed walking trial, we could directly calculate the cadence (steps s⁻¹). In doing so, all the recorded gait cycles were used for both overground and treadmill walking. This means that for overground walking, the gait cycles included in our analyses were from both the straight-line track (majority number of gait cycles) and the smooth turns at the ends (minority number of gait cycles) and they were not analyzed separately. After that, we used the cadence and the walking speed (in m s⁻¹) to calculate the average step length (m) and then doubled it to calculate the average stride length (m). We then calculated the walk ratio (WR) (in m s steps⁻¹) from the average step length and cadence. Finally, we calculated the changes (Δ) in GCoW, NCoW and cadence between the treadmill and overground conditions by first averaging the values across all the speeds per participant (to avoid the inflation of repeated-measures data) and second, subtracting the overground values from the treadmill values. All the equations used are described as follows:

$$\mathrm{RMR} = (15962 + 5155 · {\mathrm{RER}}_{\mathrm{resting}}) · \dot{V}{O}_{2\mathrm{r}\mathrm{e}\mathrm{s}\mathrm{t}\mathrm{i}\mathrm{n}\mathrm{g}}$$

(1)

$$\mathrm{GEE}=(15962+5155 · {\mathrm{RER}}_{\mathrm{walking}}) · \dot{V}{O}_{2\mathrm{w}\mathrm{a}\mathrm{l}\mathrm{k}\mathrm{i}\mathrm{n}\mathrm{g}}$$

(2)

NEE = GEE − RMR

(3)

GCoW = GEE · WS⁻¹

(4)

NCoW = NEE · WS⁻¹

(5)

Cadence = (2 · number of gait cycles) · (Walking time)⁻¹

(6)

Average step length = WS · Cadence⁻¹

(7)

Average stride length = 2 · average step length

(8)

WR = Average step length · Cadence⁻¹

(9)

ΔGCoW = GCoW_TM − GCoW_OG

(10)

ΔNCoW = NCoW_TM − NCoW_OG

(11)

ΔCadence = Cadence_TM − Cadence_OG

(12)

2.6. Statistical Tests

For age, anthropometric parameters, RMR, GEE, NEE, GCoW, NCoW, cadence, average step and stride lengths, and walk ratio, first we checked for normality using the Shapiro–Wilk test. For the RMR, age and anthropometric parameters, we conducted independent samples (between-groups) T test (Student’s or Mann–Whitney) between the males and females in our cohort. We also performed Levene’s test to check for the equality of variances. The effect sizes for the statistically significant effects were reported as either Hedge’s G or Rank-Biserial correlation.

Since, in this study, we wanted to look at and explain the differences in GEE, NEE, GCoW and NCoW between overground and treadmill with differences in the spatiotemporal parameters, we mainly chose to look at the intra-individual (within-group) differences. We then conducted a three-way mixed ANOVA with sex (males versus females) as the between-groups factor, and walking surface (overground versus treadmills) and walking speeds (slow, comfortable, fast) as the within-groups factor. Mauchly’s test was used to do sphericity corrections and Greenhouse–Geisser corrections were applied, when necessary. We performed Levene’s test to further check for the equality of variances for the remaining variables. If there were statistically significant main and interaction effects, only for those effects, we further conducted post hoc tests with Bonferroni corrections. It is probably worthwhile to mention here that if some of the effects were statistically non-significant, they should be interpreted with caution. This is because our total sample size (N = 20) should be well powered to detect significant within-participant main and interaction effects. However, as we had only 10 males and 10 females, this could mean that our statistical power to detect between-participant and interaction effects involving sex would be on the lower side, and so these interaction analyses are primarily exploratory by nature. The effect sizes for the statistically significant effects are reported as partial eta-squared (η_p²).

Furthermore, to explore the relationship between the changes (Δ) in GCoW and NCoW with the changes (Δ) in cadence, we performed correlation analyses between the variables. We checked for the normality distribution of the variables and reported Spearman’s ρ as the correlation coefficient. Additionally, a multiple linear regression analysis was conducted to examine whether the changes (Δ) in GCoW and NCoW are independently predicted by the changes (Δ) in cadence and a few selected anthropometric variables. Since GCoW and NCoW were already body-mass normalized, we included the waist circumference (as a marker of body adiposity) and the LLL up to foot as our anthropometric predictors (see Table 1 in the Section 3 for a list of the various anthropometric parameters). We checked for the normality of the residuals through Q-Q plots and for the absence of multicollinearity of the independent predictors with the Variance Inflation Factor (VIF). Standardized regression coefficients (β) for the predictors and both R² and adjusted R² were reported as key model results. The open-source statistical software JASP (version 0.18.3.0) was used for all the statistical tests and the default value of α = 0.05 was chosen as the level of statistical significance.

3. Results

3.1. Age, Resting Metabolic Rate and Anthropometric Measurements

The results of the RMR, age and anthropometric parameters are given in Table 1, where the mean and the standard deviations (mean ± SD) are mentioned. The effect sizes are calculated as the male group, subtracted from the female group. The statistically significant p-values (p < 0.05) and the effect sizes between the males and females are given for age and only for those anthropometric parameters which have a significant difference between them. The young Japanese males were heavier, taller, with a larger waist circumference and longer lower limb lengths. The females were also significantly younger than the males, but the mean age values indicate that all the participants were young.

Table 1. Age, RMR and anthropometric parameters of study participants.

Variables	Females (N = 10)	Males (N = 10)	Effect Size	p-Values
RMR (J kg−1 s−1)	1.30 ± 0.13	1.36 ± 0.10
Age (years)	21.00 ± 1.94	23.30 ± 1.42	−0.7	p = 0.01
Body mass (kg)	49.58 ± 5.39	63.20 ± 9.06	−1.8	p < 0.001
Height (m)	1.58 ± 0.04	1.73 ± 0.05	−3.3	p < 0.001
Waist circumference (m)	0.72 ± 0.05	0.80 ± 0.07	−1.12	p = 0.02
LLL up to malleolus (m)	0.73 ± 0.03	0.80 ± 0.04	−1	p < 0.001
LLL up to foot (m)	0.79 ± 0.03	0.86 ± 0.04	−1	p < 0.001
BMI (kg m−2)	19.91 ± 1.93	21.1 ± 2.72
Waist to height ratio	0.46 ± 0.03	0.46 ± 0.04

Footnote: All p-values are statistically significant (p < 0.05).

Table 1. Age, RMR and anthropometric parameters of study participants.

Variables	Females (N = 10)	Males (N = 10)	Effect Size	p-Values
RMR (J kg−1 s−1)	1.30 ± 0.13	1.36 ± 0.10
Age (years)	21.00 ± 1.94	23.30 ± 1.42	−0.7	p = 0.01
Body mass (kg)	49.58 ± 5.39	63.20 ± 9.06	−1.8	p < 0.001
Height (m)	1.58 ± 0.04	1.73 ± 0.05	−3.3	p < 0.001
Waist circumference (m)	0.72 ± 0.05	0.80 ± 0.07	−1.12	p = 0.02
LLL up to malleolus (m)	0.73 ± 0.03	0.80 ± 0.04	−1	p < 0.001
LLL up to foot (m)	0.79 ± 0.03	0.86 ± 0.04	−1	p < 0.001
BMI (kg m−2)	19.91 ± 1.93	21.1 ± 2.72
Waist to height ratio	0.46 ± 0.03	0.46 ± 0.04

Footnote: All p-values are statistically significant (p < 0.05).

3.2. Main Statistical Analyses

3.2.1. Gross and Net Energy Expenditures and Metabolic Costs of Walking

From the mixed ANOVA, there were statistically significant main effects ((F(1.4, 25.4) = 70.7 for GEE) and (F(1.4, 25.2) = 78.0 for NEE), p < 0.001, η_p² = 0.8) for walking speeds and walking surface ((F(1, 18) = 23.5 for GEE) and (F(1, 18) = 25.4 for NEE), p < 0.001, η_p² = 0.6) for both gross and net energy expenditures. Post hoc tests revealed significant effects (Bonferroni adjusted) between all the three pairs of speeds (slow–comfortable, slow–fast and comfortable–fast, p < 0.001) and between overground and treadmill (p < 0.001), with the faster speeds and treadmill showing elevated energy expenditures. Due to several statistically significant anthropometric differences between our participants, we are showing the males and females with separate sub-figures in this section representing their GEE, NEE, GCoW, NCoW and the various gait parameters. Figure 3 shows the descriptive plots for Gross EE and Net EE.

For gross metabolic costs, there was a statistically significant main effect for the walking surface (F(1, 18) = 25.3, p < 0.001, η_p² = 0.6), with post hoc tests indicating that treadmills incurred a greater metabolic cost than overground, for all the speeds (p < 0.001). For net metabolic cost of walking, there were statistically significant main effects for walking speed (F(1.5, 27.2) = 9.6, p = 0.002, η_p² = 0.4) and walking surface (F(1, 18) = 24.3, p < 0.001, η_p² = 0.6). Post hoc tests revealed a significant elevation (Bonferroni adjusted) in cost between fast and slow (p < 0.001) and fast and comfortable speeds (p = 0.02) and for treadmill compared to overground (p < 0.001). Figure 4 shows the descriptive plots for GCoW and NCoW.

3.2.2. Spatiotemporal Gait Parameters

For cadence, there were statistically significant main effects for walking speed (F(1.3, 24.0) = 344.7, p < 0.001, η_p² = 0.95), walking surface (F(1, 18) = 122.5, p < 0.001, η_p² = 0.9) and an interaction effect for walking speed*walking surface (F(2, 36) = 28.4, p < 0.001, η_p² = 0.6). Post hoc tests revealed a significant difference (Bonferroni adjusted) in cadence between all the speed combinations (p < 0.001) and a significant elevation on treadmill compared to overground (p < 0.001). For the interaction effect, all the post hoc combinations were statistically significant (p < 0.001), with the only exception being the difference between comfortable speed on overground and slow speed on treadmill (p = 0.8), where the mean difference was around 0.05 (95% CI = −0.03 to 0.12) steps/s. The key result was that at matched speeds, all our participants had a faster cadence on the treadmill compared to overground. Finally, there was a statistically significant main effect of sex (F(1, 18) = 7.4, p = 0.01, η_p² = 0.3), with the post hoc test showing a similar level of significance (p = 0.01). The mean difference revealed that the females walked with a faster cadence than the males. Figure 5 shows the descriptive plots for cadence.

For average step and stride lengths, there were statistically significant main effects for walking speed (F(1.4, 24.4) = 560.0, p < 0.001, η_p² = 0.97), walking surface (F(1, 18) = 122.5, p < 0.001, η_p² = 0.9) and an interaction effect for walking speed*walking surface (F(2, 36) = 31.2, p < 0.001, η_p² = 0.6). Post hoc tests revealed a significant difference (Bonferroni adjusted) in average step and stride lengths between all the speed combinations (p < 0.001) and a significant reduction on treadmill compared to overground (p < 0.001). For the interaction effect, all the post hoc combinations were statistically significant (p < 0.01). The key result was that at matched speeds, all our participants had smaller average step and stride lengths on the treadmill compared to overground. Finally, there was a statistically significant main effect of sex (F(1, 18) = 7.5, p = 0.01, η_p² = 0.3), with the post hoc test showing a similar significance (p = 0.01). The mean differences showed that the average step and stride lengths of females were smaller than males. Figure 6 shows the descriptive plots for average step length. Since, in this study, the average stride length is mathematically just double the average step length, the figure looks like that of average step length and so is shown in the Supplementary Materials.

For walk ratio, there were statistically significant main effects for walking speed (F(1.4, 24.0) = 5.1, p = 0.03, η_p² = 0.22), walking surface (F(1, 18) = 108.7, p < 0.001, η_p² = 0.9) and an interaction effect for walking speed*walking surface (F(2, 36) = 17.2, p < 0.001, η_p² = 0.5). Post hoc tests revealed a significant elevation (Bonferroni adjusted) in walk ratio for the fast speed compared to the slow (p = 0.01) and a significant reduction on treadmill compared to overground (p < 0.001). For the interaction effect, all the post hoc combinations were statistically significant (p < 0.01), with the exceptions being the differences between comfortable and fast speeds on overground (p = 0.55), comfortable and slow speeds on treadmill, fast and slow speeds on treadmill and fast and comfortable speeds on treadmill (all p = 1). The key result was that at matched speeds, all our participants had a lower walk ratio on the treadmill compared to overground. Finally, there was a statistically significant main effect of sex (F(1, 18) = 7.5, p = 0.01, η_p² = 0.3), with the post hoc test showing a similar level of significance (p = 0.01). The mean difference revealed that the females walked with a lower walk ratio than the males. Figure 7 shows the descriptive plots for walk ratio.

3.3. Correlation Analyses

We performed two separate correlation analyses, namely a correlation with ΔGCoW with ΔCadence and then with ΔNCoW with ΔCadence, to examine the association between the metabolic penalty on treadmill with the altered gait parameter (cadence). There was a significant positive association between ΔCadence and ΔGCoW (ρ = 0.6, 95% CI [0.29–1.00], p = 0.003, N = 20), as well as between ΔCadence and ΔNCoW (ρ = 0.6, 95% CI [0.26–1.00], p = 0.003, N = 20). Figure 8 shows the descriptive plots for correlations.

3.4. Multiple Regression Analyses

As can be seen from Table 1, there were several anthropometric parameters which were statistically different between the males and the females. To determine whether the altered gait parameter (cadence) and the selected anthropometric parameters independently predicted the metabolic penalty on treadmill, we performed two multiple linear regression analyses. The model predicting ΔGCoW from ΔCadence, LLL up to foot, and waist circumference was statistically significant (R² = 0.5, adjusted R² = 0.4, F(3, 16) = 5.5, p = 0.009). Within this model, ΔCadence was the only significant predictor (β = 0.5, p = 0.02, unstandardized 95% CI [0.4–3.3]), whereas LLL up to foot (β = 0.1, p = 0.7) and waist circumference (β = 0.4, p = 0.1) were not significant. The VIF for all the three predictors were between 1.1 and 1.6, thus indicating negligible multicollinearity between the predictors.

A similar pattern was observed for the model predicting ΔNCoW from ΔCadence, LLL up to foot, and waist circumference. This model was also statistically significant (R² = 0.5, adjusted R² = 0.4, F(3, 16) = 4.8, p = 0.01). Within this model, ΔCadence was still the only significant predictor (β = 0.5, p = 0.03, unstandardized 95% CI [0.3–3.3]), whereas LLL up to foot (β = 0.1, p = 0.8) and waist circumference (β = 0.4, p = 0.1) were not significant. The VIF for all the three predictors were similarly between 1.1 and 1.6, thus indicating negligible multicollinearity between the predictors.

4. Discussion

Since treadmills are widely used in gait laboratories as a valid proxy for natural, overground walking, in this study we wanted to confirm if healthy YAs incur a metabolic penalty on the treadmill compared to overground at matched speeds and if it is accompanied by an altered gait. Additionally, we also wanted to explore if sex differences influence the metabolic and gait parameters on the treadmill, compared to overground. To answer these questions we recruited healthy, young Japanese males and females and measured their energy expenditure, metabolic cost and spatiotemporal gait parameters at three speeds (slow—1.0 m/s, comfortable—1.3 m/s and fast—1.5 m/s) during walking both overground and on treadmill. Our results indicate that our first hypothesis (H1) was fully accepted. Contrary to our expectations, the second hypothesis (H2) was rejected as, even at the CWS, the treadmill incurred an energetic penalty, marked with an altered gait. Our third hypothesis (H3) was also fully accepted. Additionally, our explorative suggestion was partially confirmed. Our data clearly demonstrate that treadmill walking, even in YAs, imposes a metabolic penalty and elicits an altered gait relative to overground walking across multiple speeds, which is less economical and thus, indicative of a sub-optimal strategy. Below, we will discuss our results in detail and elucidate them by considering previous work in the literature.

The key finding of this study was that GEE, NEE and NCoW increased with walking speed and were higher on the treadmill, whereas GCoW was higher on the treadmill and exhibited a U-shaped relationship with speed. This is expected as walking faster requires greater mechanical power and muscle activations, while treadmill walking imposes additional constraints of enclosed space, different neuromuscular and stabilization requirements and sensory-motor mismatch compared with overground walking ([8]). This results in elevated energy expenditures and metabolic cost, even at matched speeds ([26,32]).

Accompanied by these metabolic penalties, for both overground and treadmill walking, we saw significant elevations in cadence and average step and stride lengths with increasing walking speeds. However, at matched speeds, treadmill walking was characterized by higher cadence and shorter (average) step and stride lengths (and thus lower WR) compared to overground walking. This agrees with many previously published studies ([15,16,18]). Additionally, we showed that, averaged across the three walking speeds, the change in cadence (between overground and treadmill) was also positively correlated with the change in GCoW and NCoW (between overground and treadmill). As overground walking is the ecologically valid mode of human walking, increases in speed are achieved through coordinated increases in both step/stride lengths and cadence, whereas treadmill walking constrains step/stride length modulation and relies more heavily on increasing cadence. This treadmill-specific gait adaptation possibly leads to the increased energetic penalty.

Considering our explorative questions, we saw significant differences in basic anthropometrics between the males and females. The main effect of sex on the gait parameters is consistent with established anthropometric differences between them. As the females had a shorter stature and lower limb lengths, they adopted shorter steps and faster cadence to achieve a given walking speed ([23,24,25]). Importantly, these differences persisted even when speeds were matched. From our multiple regression analyses, we showed that, when adjusted for the selected anthropometric parameters that are different between our males and females, the treadmill metabolic penalty for all the participants is partially explained by their altered gait strategy (increased cadence) on treadmills. All in all, the reduction in walk ratio on the treadmill and the increased cadence leads to a deviation from the energetically optimal gait coordination. This is a key result as it is not due to the anthropometric differences between our males and females.

If we look at some of the previously published literature, refs. [6] and [7] reported higher metabolic cost on treadmill compared to overground in healthy OAs. In this study, by focusing exclusively on healthy YAs, we show that the metabolic penalty is not exclusive to aging and rather may arise from more fundamental demands of treadmill walking. In previous work ([13]), we did not see an elevation in NCoW for our YAs at PWS. Our current results suggest that this apparent equivalence may not generalize beyond PWS and so, in this study, we saw an elevated metabolic cost for all the three tested walking speeds, despite the CWS of 1.3 m/s being very close to the general PWS of humans. A recent systematic review and meta-analysis also reported increased relative oxygen consumption and cadence, and reduced stride/step lengths on treadmill compared to overground walking in healthy adults ([14]), thus showing similarity with our results. The spatiotemporal shift in gait suggests that participants show a cautious strategy and may have prioritized stability over energetic economy. Such strategies likely increase muscle activation and control effort, analogous to what is seen in conditions which are novel or where balance is challenged ([33]). While we did not record Electromyography (EMG), prior work in OAs ([2]) indicates that treadmill walking increases antagonistic muscle co-activation, which can be seen as a compensatory strategy to achieve lower limb joint stability, often at the cost of a metabolic penalty. Even in healthy YAs, like in this study, more stabilizing muscle activity may have driven up the metabolic cost, though this is purely speculative since we neither measured EMG nor analyzed stride-to-stride variability or dynamic stability.

Looking at some of the recently published literature, we see that treadmill and overground walking fundamentally differ in stability-related foot placement control ([34]), implying that the underlying locomotor control problem changes with the walking surface. For our young participants in this study, treadmill walking at matched speeds consistently elicited higher cadence (and thereby shorter average step lengths) and elevated MCoW compared to overground walking. This altered gait strategy is possibly indicative of a more cautious gait on the treadmill, due to tighter foot-placement control ([34]) and the need to perform frequent step-to-step foot placement corrections. Importantly, deviations from the preferred coordination between cadence and step length (WR) are known to increase mechanical power requirements during the swing phase of the gait, particularly when cadence is increased and walk ratio is lowered ([35]). Taken together, these new findings support the interpretation that treadmill walking induces a stability-oriented coordination shift away from energetically optimal gait, thereby leading to increased mechanical and muscle-force production demands ([36]), and a metabolic penalty.

Another interesting finding of this study is the divergent behavior of GCoW and NCoW. GCoW was elevated on the treadmill but did not vary significantly with speed and exhibited the classic U-shaped relationship, whereas NCoW increased with speed and was also elevated on the treadmill. Since GCoW represents the whole-body energy level required to travel a unit distance and has repeatedly been shown to exhibit a minimum near PWS/CWS, it reflects the ecological and energy minimization theories ([37,38,39]) that govern human walking. In contrast, for NCoW, it is expected that we do not see the U-shaped relationship between cost and speed. This is because we measure RMR only at the beginning and subtract the same RMR value from the GEE at every walking speed to calculate the NEE and then divide it by the different walking speeds to get the NCoW. At slow speeds, the RMR is larger and at faster speeds, it is smaller, relative to walking energy expenditure. As such, when we subtract the same RMR from the GEE, at slower speeds, the NEE is disproportionately reduced, thereby suppressing the NCoW. At faster speeds, the similar subtraction does not impact the NEE substantially and so the NCoW increases.

Our study has several limitations. Firstly, we only measured healthy Japanese YAs and so we cannot generalize to older, clinical, or less fit (patient) populations and possibly to other non-Japanese ethnicities. Secondly, though we gave each one of our participants at least ten minutes of treadmill familiarization in the beginning, we only did it at the CWS and this might not eliminate all novelty and adaptation effects. In [13], we gave a similar amount of treadmill familiarization at PWS and saw that it was enough for the healthy YAs in that study. So, we can possibly claim that familiarization was not the main driver for the increased MCoW for YAs on treadmill in this study. Finally, overground walking was paced using visual and auditory cues and participants had to take turns, which may reduce natural gait variability and slightly alter gait parameters. However, we advised the participants to take smooth turns and visually ensured that they were doing so. All the participants were also walking at the desired speed and were in their metabolic steady state after 2 min of overground walking. Nevertheless, it is true that even after taking smooth turns during overground walking, we cannot claim complete equivalence of the overground and treadmill walking protocols. Altogether, we think that these limitations do not substantially affect and change our primary conclusions about YAs and the consistent effect of the treadmill compared to overground is much more dominant.

Looking forward, future studies should aim to combine metabolic measurement with EMG and kinetics and kinematics from motion capture or wearable IMUs to identify muscle activation patterns, joint work, postural stability metrics and their (relative) contributions to increased metabolic cost. This may help establish causal links between the mechanics and energetics of human walking. Investigating short-term (days) and long-term (weeks to months) adaptations to treadmill walking and assessing whether repeated exposure reduces the energetic penalty on treadmills are also fruitful to study. Studies manipulating sensory feedback, such as visual flow via virtual reality, proprioceptive inputs via musculo-tendon vibrations and vestibular control via Electronic Vestibular Stimulation, would furthermore be interesting, to study how sensory context and possible reweighting on treadmills compared to overground modify gait and thereby lead to the metabolic penalty.

5. Conclusions

To the best of our knowledge, this study is one of the few recent investigations that provides clear evidence that even healthy young adults incur a substantial and consistent energetic penalty on treadmills compared to overground. They also exhibit an altered and possibly sub-optimal gait when walking on a treadmill, across a range of matched speeds. Based on these findings, we recommend both researchers and practitioners to review their data from treadmill-based studies against overground data, if they want to understand daily-life energetics and gait behavior. From the results of this study, we can conclude that treadmills may not serve as a direct proxy for the ecologically relevant overground walking. We think this is important for advancing scientific understanding, informing evidence-driven recommendations for future clinical and rehabilitation applications (involving treadmills) and for conducting improved modeling studies trying to connect the mechanics and energetics of human walking, using empirical data.