An evidence-based multi-factorial model to predict the oxygen cost of ventilation during ramp-incremental cycle ergometry exercise

Abstract

Introduction:

During maximal ramp-incremental exercise (RIE), the oxygen uptake–power output relationship (O₂gain) may deviate from linearity near exhaustion. An increased oxygen cost of ventilation (O_2VENT) is a plausible but under-quantified contributor. This study tested a non-linear multi-factorial model using measured O_2VENT and six predictors: resting expired ventilation (_E), weight, height, age, O₂ peak, and maximal heart rate (HRMax) to 1) estimate O_2VENT and its contribution to maximal oxygen uptake (O₂max) in an independent dataset and 2) determine whether correcting O₂ by O_2VENT (O_2VCORR) alters O₂max and O₂gain estimates.

Methods:

Published data from 42 participants (11 women, 31 men; 29 ± 6.5 years; O₂max = 4.02 ± 1.06 L min⁻¹) were used to derive the model. Leave-one-out cross-validation (LOOCV) was used to assess validity, with predictive accuracy and coefficient stability evaluated via bootstrap resampling. The model was applied to an independent RIE dataset to generate O_2VCORR, which was compared with uncorrected O₂ across six %Wpeak intensities using repeated-measures ANOVA and final 30 s slope analysis.

Results:

The model explained 81% of O_2VENT variance (adjusted R² = 0.78). O_2VENT represented 17.43% ± 3.58% of O₂ at O₂max. Across 35%–100% Wpeak, O_2VCORR values (L·min⁻¹) increased with intensity (1.77 ± 0.43, 2.68 ± 0.57, 3.43 ± 0.72, 3.72 ± 0.79, 3.84 ± 0.86, and 3.92 ± 0.82) but remained significantly lower than uncorrected O₂ (p < 0.001), with the final-30 s O₂ slope attenuated following correction (p = 0.002).

Conclusion:

The internally validated model revealed O_2VENT may contribute to a significant fraction of O₂ near exhaustion.

Introduction

There has been growing interest in the presence of a non-linear profile of the oxygen uptake–power output relationship (O₂gain) during maximal ramp-incremental exercise (RIE) (Iannetta et al., 2019; Hansen et al., 1988; Zoładź et al., 1995; Zoładź and Korzeniewski, 2001; Barstow et al., 2000; Scheuermann et al., 2002; Lucía et al., 2002a; Pedersen et al., 2002; Zoładź et al., 1998; Bickham et al., 2004; Lucía et al., 2002b; Boone and Bourgois, 2012). Original interpretations of this relationship described the O₂gain during RIE as constant, meaning that for RIE protocols applying a linear increase in exercise intensity (x-axis), the total oxygen uptake (O₂) response (y-axis) to the protocol was also linear until, for some individuals, a plateau or deviation toward a plateau occurred at volitional exhaustion (Whipp et al., 1981; Davis et al., 1982).

Despite this historical context, it has been well-documented that significant variability exists between participants in the O₂gain of linear RIE protocols, with the O₂gain decreasing (Zoładź et al., 1998; Bickham et al., 2004), increasing (Hansen et al., 1988; Barstow et al., 2000; Scheuermann et al., 2002), displaying mixed non-linear responses (Lucía et al., 2002a; Pedersen et al., 2002; Zoładź et al., 1998), or remaining linear to volitional fatigue (Supplementary Figure S1). The increasing between-subject variability in this relationship and the incidence of a O₂ plateau have raised concerns about the validity of maximal oxygen uptake (O₂max) determination and subsequent constant-work rate (CWR) exercise prescription (Iannetta et al., 2019; Niemeyer et al., 2021). Ultimately, gold-standard criteria for determining O₂max are yet to be established due to differences between researchers and laboratories in data processing strategies, the sensitivity and accuracy of different metabolic carts, large within-subject and between-subject variability in O₂gain profiles toward maximal exhaustion, the absence or presence of a O₂ plateau, and other physiological and data processing concerns (Boone and Bourgois, 2012).

Advancements in understanding O₂gain variability have revealed several possible physiological causes (Boone and Bourgois, 2012). Current insights suggest that increasing O₂gain may be the result of a combination of physiological factors, including increases in lactate, hydrogen ions, or catecholamines, muscle temperature, proton leaks through the inner mitochondrial membrane, decreases in cytosolic phosphorylation potential, increased cost of stabilizing muscles of the upper body, increases in the oxygen cost of ventilation (O_2VENT), altered motor unit recruitment, and/or differences in fitness levels (Boone and Bourgois, 2012; Zoladz et al., 2002; Hug et al., 2004; Vella et al., 2006). The most widely explored, evidence-based physiological factors explaining variable O₂gain are muscle fiber type proportions (slow-twitch (ST) vs. fast-twitch (FT)) (Lucía et al., 2002a; Hug et al., 2004; Jones et al., 2004; Marles et al., 2006). A key limitation of prior interpretations is that the roles of other major contributors to O₂ cost, such as O_2VENT, were not quantified or considered, leading to the prevailing interpretation that increasing O₂gain profiles result from greater recruitment of FT fibers (Lucía et al., 2002a; Zoladz et al., 2002; Hug et al., 2004; Jones et al., 2004; Marles et al., 2006). However, lower mitochondrial-derived adenosine triphosphate (ATP) turnover has been measured in FT fibers, which, hypothetically, should result in slow O₂ kinetics that reduce the ability to increase O₂, especially at the higher power outputs experienced at the end of an RIE test (Wakeling et al., 2006). Such responses would lower the O₂gain. Given these uncertainties and the insufficient research on FT fiber involvement, other contributors beyond skeletal muscle fiber type and percent distribution warrant further inquiry.

One such contributor is the O_2VENT, and the extent to which this measure increases above the gas-exchange threshold (GET) (Vella et al., 2006; Aaron et al., 1992a; Marks et al., 2005; Coast et al., 1993; Harms et al., 1997; Harms et al., 1998). The O_2VENT is defined as the energy requirement of the elevated minute ventilation (_E) by either exercise or voluntary hyperventilation. It is measured by completing multiple trials of breathing in seated (non-exercising) conditions, or other postures of interest, at _E values similar to those at different exercise intensities. It is calculated as the difference in O₂ at rest and during hyperventilation (Vella et al., 2006; Marks et al., 2005). A greater understanding of O_2VENT may provide an important avenue to deepen our understanding of O₂gain dynamics and allow for distinct quantification of its impact relative to skeletal muscle contributions to whole-body O₂ (wbO₂). For example, the disproportionate increase in _E, particularly at higher power outputs during an RIE test, has been proposed to explain increases in O₂gain not accounted for by muscle fiber proportions alone (Vella et al., 2006; Marks et al., 2005).

Bartlett et al. (1958) were the first to reveal an exponential increase in O_2VENT during moderate to high rates of _E. The metabolic cost of this exponential increase was later quantified for untrained humans as being approximately 10% of O₂max (Shephard, 1966; Nielsen, 1936; Aaron et al., 1992b) and, in endurance-trained men, 15% of O₂max (Aaron et al., 1992a). Further research revealed the significant contribution of the O_2VENT to wbO₂, with estimates between 5% and 18% of O₂max (Vella et al., 2006; Marks et al., 2005; Dominelli et al., 2014; Oueslati et al., 2017; Oueslati et al., 2016). Only two prior studies have quantified the impact of the O_2VENT on calculations of O₂max between participants (Vella et al., 2006; Marks et al., 2005). Despite clear evidence that O_2VENT constitutes a significant fraction of wbO₂ and to increasing O₂gain profiles (Vella et al., 2006; Marks et al., 2005), no standardized approach exists to systematically quantify this effect across different individuals and exercise intensities.

Direct measurement of O_2VENT requires a higher percent carbon dioxide (CO₂) gas to be breathed in during testing to prevent hypocapnia in higher _E rate trials. This unique setup and access to medical-grade gas may be inaccessible in athletic testing environments. To address this gap, we developed a computational model based on previously published datasets to estimate O_2VENT during RIE without requiring hyperventilation testing. An evidence-based model would provide a practical, scalable method for estimating O_2VENT across individuals, enabling more accurate interpretation of O₂ and O₂gain during RIE and its underlying physiological components in both research and applied settings.

We aimed to test whether a non-linear multi-factorial model derived from measured O_2VENT and six predictors: resting expired ventilation (_E), weight, height, age, O₂ peak, and maximal heart rate (HRMax) could 1) estimate O_2VENT and its contribution to maximal oxygen uptake (O₂max) in an independent dataset and 2) determine whether correcting O₂ by the O_2VENT (O_2VCORR) alters O₂max and O₂gain estimates. We hypothesized that the model would accurately estimate O_2VENT, representing approximately 15%–20% of O₂max, and that O₂ corrected by the O_2VENT (O_2VCORR) would be significantly lower than uncorrected O₂, resulting in a reduced O₂gain across increasing exercise intensities.

Methods

Data extraction and regression analysis

A multi-factorial non-linear regression model was developed to predict O_2VENT using previously published O_2VENT data from repeated-measures experimental research. However, the analyses and results presented in the manuscript are entirely novel and have not been published elsewhere. The published data (Table 1) were collected by Vella et al. (2006) from 20 healthy, non-smoking, recreational and/or endurance-trained men (n = 18) and women (n = 2) and by Marks et al. (2005) from 22 healthy, non-smoking men (n = 13) and women (n = 9). Permission was granted by the authors to reuse the published work to develop the model, which was also checked and approved against prior ethical guidelines, including the Declaration of Helsinki and the Common Rule. All _E trial data of Vella et al. (2006) were first converted (Equation 1) from BTPS to STPD volume conditions to ensure similar volume conditions to the participants of Marks et al. (2005). All participant data from both studies were then exported and combined into a single MS Excel file.where 273 is 0 °C expressed in Kelvin units (K); Btemp is body temperature ( °C); PB is the barometric pressure (mmHg), and Atm is the standard atmospheric pressure (mmHg).

Model development and statistical procedures

All statistical analyses were performed using IBM SPSS Statistics v29.0.0.0, with results presented as mean ± standard deviation (SD) and significance set at p < 0.05. Model validation procedures, including leave-one-out cross-validation (LOOCV) and bootstrap resampling, were conducted in R (v4.3.1) using the caret and boot packages. Data normality was determined using Shapiro–Wilk tests, and potential outliers were identified by two authors through visual inspection of individual boxplots. Descriptive statistics summarized _E and participant characteristics (Table 1). To determine the fit of O_2VENT data for all participants (n = 42) of Vella et al. (2006) and Marks et al. (2005), the _E rate across multiple _E mimicked trials were plotted against the O_2VENT (Figure 1). Linear and non-linear regression models were then compared to determine the best-fitting relationship, as shown in Figure 1.

FIGURE 1

To avoid collinearity due to repeated measures within the model dataset, only one O_2VENT value per subject (peak O_2VENT) was used. All independent variables were assessed for multicollinearity using pairwise Pearson’s correlations (Supplementary Table S1). Variance inflation factors (VIF) for each predictor were obtained from a multiple regression model (Supplementary Table S2). The model was then developed based on the total individual participant data from Vella et al. (2006) and Marks et al. (2005) for the independent variables of _E, age (years), weight (kg), height (cm), O₂max (L∙min⁻¹), maximum heart rate (HRMax; b∙min⁻¹), and the O_2VENT (L∙min⁻¹) as presented in Figure 2. Each independent variable was plotted (GraphPad Prism, v10) against only the highest O_2VENT (dependent variable) for each participant to adhere to the needed assumption of data independence for multiple regression analyses. Based on lowest standard error for each independent variable, the _E:O_2VENT data (Figure 2a) were fitted with a non-linear third order polynomial (cubic), age:O_2VENT data were fitted with a simple linear regression (Figure 2b); Height:O_2VENT data were fitted with a non-linear exponential growth equation (Figure 2c); Weight:O_2VENT data (Figure 2d) were fitted with a non-linear third order polynomial (cubic); O₂max:O_2VENT data were fitted with a simple linear regression (Figure 2e); and HRMax:O_2VENT data (Figure 2f) were fitted with a simple linear regression. The best-fit coefficient values of each independent variable and equation were then used to create the non-linear regression model. Please see “Full non-linear regression model expression.”

FIGURE 2

Equations 2–8 were developed for each independent variable based on the results of the linear or non-linear function applied to the participant data (Figures 2a–f).where V0, V1, V2, and V3 are the best-fit coefficient values, and VE is the highest _E of the participant’s _E trials.where SLA is the slope of O_2VENT:age, YIA is the y-intercept of O_2VENT:age, and age is the participant’s age in years.where SLH is the slope of height, KH is the constant of height, and height is the height of the participant.where Wt0, Wt1, Wt2, and Wt3 are the best-fit coefficient values, and Weight is the weight of the participant in kg.where O₂SL is the slope of O_2VENT:O₂max, O₂YI is the y-intercept of O_2VENT:O₂max, and O₂max is the measured O₂max of the participant in L·min⁻¹.where SLHR is the slope of O_2VENT:HRMax, YIHR is the y-intercept of O_2VENT:HRMax, and HRMax is the maximum heart rate of the participant expressed as beats·min⁻¹.

The model’s ability to then estimate O_2VENT was assessed by applying the model equation to previously published maximal RIE (ramp function Watts = 36 ± 3) data of 14 participants from the study of O’Malley et al. (2024). All gas-exchange data were measured breath-by-breath and, subsequently, averaged over seven breaths. The parameter estimates from the final iteration of the multi-factor non-linear regression model (Supplementary Table S3) were used to create a custom, functional computational model program in LabVIEW (National Instruments, Austin, TX, United States, v2017). This custom program was then used to predict the O_2VENT of participants from the study of O’Malley et al. (2024) and to generate a graph of O₂max, O_2VCORR, and O_2VENT. The results of the predicted O_2VENT of these participants are reported in Results.

Statistical analysis procedures

Internal validation and predictive performance

Due to the modest sample size (n = 42), internal validation was performed for six predictors (_E, Weight, Height, Age, O₂ peak, and HRMax) using LOOCV. LOOCV systematically holds out each observation in turn, fits the model to the remaining data, and predicts the outcome for the held-out observation. This procedure provides nearly unbiased estimates of predictive performance while mitigating overfitting. Model performance was quantified using adjusted R² to account for the number of predictors relative to the sample size, RMSE, and mean absolute error (MAE), calculated from LOOCV-predicted vs. observed values. Observed vs. predicted plots, as well as residuals vs. predicted plots, were generated to visually assess model fit and calibration. Shaded areas in calibration plots indicate the 95% confidence intervals for predicted values, illustrating uncertainty in individual predictions.

To further characterize uncertainty, bootstrap resampling of the LOOCV framework was conducted with 500 bootstrap iterations. For each iteration, the dataset was sampled with replacement, the model was refitted using the same LOOCV procedure, and both predicted values and model coefficients were recorded. This allowed calculation of 95% bootstrap confidence intervals for model performance metrics (R², RMSE, and MAE) and for each model coefficient to assess that the predictor’s contributions are consistent across resampled datasets.

Model predictions and calibration

Observed values represent measured O_2VENT, while predicted values are the corresponding LOOCV predictions. Observed vs. predicted values were plotted to assess calibration and predictive accuracy. Calibration analysis was performed by fitting a linear regression of predicted on observed values, with the slope and intercept reported to quantify systematic bias and prediction dispersion. The distribution of predicted O_2VENT values across bootstrap iterations was summarized and visualized as histograms to illustrate the stability and variability of model performance.

The model was then applied to an independent sample to predict and compare the predicted O_2VENT for participants from the study by O’Malley et al. (2024) across six different %Wpeak intensities via a one-way repeated-measures analysis of variance (ANOVA). A two-way ANOVA was performed to assess for any significant differences between O₂max and O_2VCORR across these intensities, with follow-up pairwise comparisons for significant interactions. Finally, slope analyses of the final 30 s of O₂max and O_2VCORR were determined using simple linear regression, with differences between slopes evaluated via paired samples t-tests. Descriptive analyses were performed for the 14 participants of O’Malley et al. (2024) for O_{2VENT and}O₂max and the %O_2VENT of O₂max was calculated as the difference between O_{2VENT and}O₂max multiplied by 100.

Results

Model performance

The non-linear regression model converged after 588 major iterations, with the residual sum of squares decreasing significantly from 777.662 to 0.686, meeting the convergence criterion of a relative reduction in residual sum of squares at most 1.000 × 10⁻⁸. The estimated coefficients, standard error, and 95% confidence intervals for each constant are presented in Table 2. The model explained variance in O_2VENT, F (16, 26) = 19.983, p < 0.001, R² value of 0.81, indicating that 81.1% of the variance in O_2VENT was accounted for by the independent variables. Although moderate correlations were observed between some anthropometric and physiological variables (e.g., O₂max and height, r = 0.761), all pairwise correlations (Supplementary Table S1) were below the commonly accepted multicollinearity threshold (r = 0.80). VIFs for all independent variables were also <5 (Supplementary Table S2).

LOOCV indicated good model performance, with RMSE = 0.128, MAE = 0.098, R² = 0.812, and adjusted R² = 0.78, accounting for the six predictors relative to the sample size. Observed vs. predicted O_2VENT values showed good alignment (calibration slope = 0.81, intercept = 0.16), and residuals vs. predicted plots indicated no systematic deviation across the prediction range (Figures 3A,B).

FIGURE 3

Bootstrap resampling of the LOOCV predictions (500 iterations) generated 95% confidence intervals for model metrics: [RMSE = 0.128, 95% CI (0.098, 0.155); MAE = 0.098, 95% CI (0.074, 0.124); R² = 0.78, 95% CI (0.68, 0.90)]. Histograms of the bootstrap distributions (Supplementary Figure S2) illustrate the variability in predictive performance across resamples. Bootstrap-derived 95% confidence intervals for model coefficients are summarized in Supplementary Table S4. Some intervals included zero (e.g., Intercept: −5.36 to 15.17; _E linear term: −0.0616 to 0.0734), reflecting expected variability given the sample size and number of predictors. However, the model shows moderate predictive accuracy and calibration despite these uncertainties.

Application of the model to an independent dataset

Results of the one-way repeated-measures ANOVA with the Greenhouse–Geisser correction revealed a significant main effect for predicted O_2VENT across the different intensity points, F (1.129, 14.681) = 37.680, p < 0.001 (Figure 4).

FIGURE 4

Individual participant data are presented in Table 2 for O_2VENT,O₂max, and the %O_2VENT of O₂max. At O₂max, O_2VENT expressed as a percentage of O₂max, was 17.43% ± 3.48% (Table 2).

Raw data of one participant are presented in Figures 5a–d for O₂max, O_2VCORR, and O_2VENT. The mean slope (final 30 s) of O₂max was significantly greater (0.2866 ± 0.2549 L min⁻¹) than the mean slope of O_2VCORR, (0.0419 ± 0.3954 L min⁻¹; t (13) = 4.003; p = 0.002, d = 1.07).

FIGURE 5

Two-way repeated-measures ANOVA with Greenhouse–Geisser correction revealed a significant main effect for intensity, F (1.31, 17.02) = 115.66, p < 0.001; measure, F (1, 13) = 533.05, p < 0.001; and a significant two-way interaction between intensity and measure, F (1.13, 14.67) = 37.46, p = <0.001 (Figure 6). Follow-up pairwise comparisons revealed a significant difference between O₂ and O_2VCORR across all levels of intensity (p < 0.001; Figure 6). Furthermore, O₂ was significantly different between all levels of intensity (p < 0.001; Figure 6) but O_2VCORR was only significantly different across 35%, 55%, and 75% Wpeak (p < 0.001) but not between 85% and 95% (p = 0.927); 85% and 100% (p = 0.269); and for 95% and 100% (p = 1.00; Figure 6).

FIGURE 6

Discussion

The present study set out to develop and apply an evidence-based computational model for estimating O_2VENT without requiring additional hyperventilation-mimicking trials, and the findings are threefold. First, the model demonstrated close agreement between predicted and measured O_2VENT values (Figures 3A,B), reflecting good calibration and stable performance across participants. Second, when the model was applied to a new dataset of trained participants, the mean predicted O_2VENT, expressed as a percentage of O₂max, was 17.43% ± 3.48% (Figure 4; Table 2; range 11.96%–23.66%), in line with several previous findings. Finally, the mean slope (final 30 s) of uncorrected O₂ was significantly greater (0.2866 ± 0.2549 L min⁻¹) than that of O_2VCORR (Figures 5, 6), highlighting the potentially significant contribution of O_2VENT to total O₂ and, consequently, to the increase in O₂ gain near maximal exhaustion during RIE.

Model performance and O_2VENT contribution to O₂max

The developed model demonstrated good apparent predictive performance and reasonable calibration for estimating O_2VENT (Figures 3A,B; Supplementary Figure S2; Supplementary Table S4). While bootstrap resampling of the LOOCV predictions (500 iterations) indicated moderate variability in performance estimates, reflecting some sample sensitivity inherent to small datasets, the general pattern of predictive behavior remained consistent (Supplementary Figure S2; Supplementary Table S4). Importantly, when the model was applied to an independent dataset, the predicted O_2VENT values aligned closely with previously published observations, supporting its practical applicability and internal validity. These findings suggest that, despite expected uncertainty, the model provides a useful exploratory framework for estimating O_2VENT and warrants further evaluation in larger and more diverse samples to confirm generalizability.

When applied to a separate dataset of trained cyclists, the predicted average contribution of O_2VENT to O₂max was 17.43% ± 3.58%, with individual values ranging from 12% to 24% (Figure 4). These estimates closely align with previous observations of the contribution of O_2VENT to O₂max of approximately 10%–18% in trained populations (Aaron et al., 1992a; Harms et al., 1997).

While the model has not yet undergone full external cross-validation with independent participant data, the LOOCV and bootstrap results, along with the agreement with existing datasets, suggest generally consistent predictive behavior (see Limitations). Importantly, this model is intended as an exploratory framework that opens a new avenue for investigating ventilatory function during exercise. At the same time, the relatively wide bootstrap confidence intervals for some coefficients, the small sample size, and the risk of overfitting suggest caution when applying the model to other datasets, underscoring the value of further validation in larger, independent samples. The observed inter-individual variability, including sex-based differences in O_2VENT and _E, demonstrates the potential of this exploratory framework to enhance the interpretation of complex physiological responses, once further validation is performed (Table 2). Although the predicted O_2VENT was lower in women than in men (Table 2), O_2VENT as a percentage of O₂max was higher, suggesting that women may need to expend a greater percentage of their overall O₂ on breathing to achieve adequate _E. Prior literature (Espinosa-Ramírez et al., 2021; Sheel and Guenette, 2008) also supports men having larger overall metabolic rates, ventilatory capacities, and increased ventilatory efficiency, which is further corroborated by the lower relative, predicted O_2VENT as a percentage of O₂max in the present study. Similar physiological and anatomical differences have also been observed by Dominelli et al. (2014), with higher relative ventilatory values of 13.8% and 9.4% of wbO₂ for women and men, respectively. The model’s consistency in producing O_2VENT estimates aligned with prior observations without requiring additional hyperventilation protocols may support its biological plausibility, at least in trained populations.

The inter-individual variability of predicted O_2VENT is not surprising, as the dataset of O’Malley et al. (2024) consisted of average O₂max values of 4.76 L min⁻¹, with individual values ranging from 2.24 L min⁻¹ to 5.91 L min⁻¹ (Figure 4; Table 2). There is sufficient evidence to support physiological and anatomical differences between sexes that may explain the increasingly higher measured and, in the present study, estimated O_2VENT for men compared to women (Dominelli et al., 2014; Espinosa-Ramírez et al., 2021). The lowest predicted O_2VENT values at O₂max in the present study was 0.53 L min⁻¹ and 0.61 L min⁻¹, which, relative to O₂max, represented 23.66% and 16.05%, respectively (Figure 4; Table 2). Both observations involved two female participants from the O’Malley et al. dataset (O’Malley et al., 2024). Interestingly, the 23.66% value for O_2VENT as a percent of O₂max was for a female participant who had the lowest absolute O₂max (participant 8; Table 2) in the present study. Some of the lowest predicted O_2VENT values relative to O₂max were observed for men who had larger absolute O₂max values (Table 2). During dynamic exercise, women may experience increased resistive respiratory and elastic work, lower forced vital capacity, and high limited expiratory flow, which may explain the predicted increase in their O_2VENT compared to their male counterparts (Sheel and Guenette, 2008).

Non-linear O₂–PO dynamics

During continuous RIE to volitional exhaustion, some participants may demonstrate an increased O₂gain simultaneous with a non-linear increase in O_2VENT above the GET (Vella et al., 2006; Marks et al., 2005). A similar O₂–PO dynamic was observed in the present study (Figures 5a–d), where the mean slope (final 30 s) of O₂ was significantly greater (0.2866 ± 0.2549 L min⁻¹; p = 0.002; Figures 5a,b) than O_2VCORR (0.0419 ± 0.3954 L min⁻¹; Figure 5b), highlighting the significant contribution of O_2VENT near volitional exhaustion. This is further supported by the disproportionate increase in predicted O_2VENT with increasing PO intensities (35%–100% Wpeak; Figure 4), consistent with previous estimates attributing up to 18% of total O₂ during maximal RIE to ventilatory cost (Vella et al., 2006; Harms et al., 2000; Dominelli et al., 2014). The phenomenon reflects the known non-linear increase in ventilatory drive and increased work of breathing (WOB) near volitional exhaustion (Vella et al., 2006; Harms et al., 1997; Dominelli et al., 2014).

While predicted O_2VENT increased simultaneously with increasing PO across the RIE test (Figure 6), the slope of O_2VCORR was substantially attenuated, with no significant differences observed between 85%, 95%, and 100% Wpeak (Figure 6). This plateau in O_2VCORR, despite continued increases in total O₂ with increasing PO, suggests that increased O₂gain is more likely to be the result of increasing O_2VENT rather than by further increases in locomotor muscle oxygen uptake due to increased motor unit (predominantly FT) recruitment.

The increasing O_2VENT observed toward volitional exhaustion (Figures 4, 5a) is consistent with the known ventilatory response during high-intensity exercise, wherein _E increases exponentially, primarily through elevations in breathing frequency (fR) once tidal volume (Vt) plateaus (Vella et al., 2006; Aaron et al., 1992a). However, individual responses in the present study varied. Some participants maintained or further increased Vt alongside fR, whereas others exhibited a leveling response in Vt at intensities above the GET, and others showed a decrease in Vt during the last minutes of the RIE. For these latter individuals, minute ventilation (_I) continued to increase due to a rapidly increasing fR (Figures 5c,d). This shift could be hypothesized to reflect mechanical or neuromuscular limitations on tidal expansion, including dynamic hyperinflation, elevated intrinsic positive end-expiratory pressure, or respiratory muscle fatigue. These factors together may explain the increase in fR rather than depth and therefore, the less efficient _E patterns and higher O_2VENT observed (Harms et al., 1998; Dempsey et al., 2006). Limits to further increases in Vt and the need to increase fR reflect mechanical limitations to _I that result in an increased WOB, particularly in smaller or less trained individuals, where the increased fR causes larger increases in O_2VENT for given increments in _E. In other words, the increased ventilatory demand is met less efficiently, with a higher fR driving a disproportionate increase in O_2VENT. This is further exacerbated in individuals who have a declining Vt near the end of the RIE.

While earlier research has often emphasized increased FT fiber recruitment as the likely explanation for increased O₂gain near and/or at volitional exhaustion (Nielsen, 1936), our findings suggest that ventilatory mechanics and their oxygen cost must at least be considered in parallel. This is especially important for subjects with less efficient breathing strategies, for whom a high fR and reduced Vt may increase WOB and elevate O_2VENT disproportionately.

Perspective

The novel, non-linear regression model developed in this study provides a new method for estimating O_2VENT using commonly derived RIE measures. Such modeling may provide a new exploratory avenue for examining physiological sex-based differences in _E mechanics, which is an area of research in which female participants remain markedly underrepresented across age, health status, and fitness level. Future research should also aim to develop similar models within clinical contexts, and evaluation of this model’s applicability to independent datasets may continue to pave the way for greater reliance on modeling approaches to estimate difficult-to-measure physiological variables in performance testing and physiological data interpretation.

Limitations

Limitations of the present study may include the risk of overfitting despite acceptable LOOCV performance, as the model was developed and validated using the same dataset. External validation using a new dataset of measured O_2VENT may add to the robustness and generalizability of the model for future applications. Bootstrap confidence intervals for some model coefficients, reported in the Supplementary Material, were relatively wide and occasionally included zero. Similarly, some standard errors were larger than anticipated, highlighting uncertainty in specific coefficient estimates. Nevertheless, as this is the first study to model O_2VENT, the results infer high potential for further research inquiry on this topic. Additionally, because the model was developed based on the characteristics of healthy and trained individuals, it may only be applicable to a similar subset of participants. The findings of this study are specific to RIE protocols and may not directly generalize to step-incremental or CWR exercise. Finally, due to the inability to quantify O_2VENT for resting O₂, the model assumes a similar resting O₂ to that measured by Vella et al. (2006) and Marks et al. (2005), even though it is logical that the O_2VENT at rest must be a low-to-moderate percentage of the total resting O₂ measure.

Conclusion

This study developed and internally validated a non-linear regression model to estimate the O_2VENT during RIE without requiring hyperventilation-mimicking trials. The model showed strong agreement with the previously measured O_2VENT data, and when applied to an established dataset, provided estimations of O_2VENT that closely matched previous observations. Further, the model suggests that O_2VENT may contribute to a significant and variable fraction of O₂ near and/or at volitional exhaustion that causes major adjustments to total O₂gain of the RIE in the last minutes of the protocol. The findings, although aligning with previous research, also reveal important individual and sex-based differences in ventilatory patterns during RIE. Notably, increases in O₂ near and/or at volitional exhaustion may be influenced by ventilatory demands, not only by increases in locomotor O₂. By providing a practical, evidence-based method for estimating O_2VENT, this work further reveals the capabilities to adjust whole-body O₂ data for a more valid limb skeletal muscle O₂ response. This has the potential to provide greater understanding of the physiological contributions of limb skeletal muscle to exercise-induced gas exchange, as well as improve the development, detection, and application of criteria used to verify the attainment of O₂max. As such, the results of this study suggest further research is needed to address a significant methodological gap in understanding the changing O₂gain of RIE, and how this could impact the detection of O₂max.

References

• 1

  AaronE. A.SeowK. C.JohnsonB. D.DempseyJ. A. (1992a). Oxygen cost of exercise hyperpnea: implications for performance. J. Appl. Physiol. (1985)72 (5), 1818–1825. 10.1152/jappl.1992.72.5.1818

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 2

  AaronE. A.JohnsonB. D.SeowC. K.DempseyJ. A. (1992b). Oxygen cost of exercise hyperpnea: measurement. J. Appl. Physiol. (1985)72 (5), 1810–1817. 10.1152/jappl.1992.72.5.1810

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 3

  BarstowT. J.JonesA. M.NguyenP. H.CasaburiR. (2000). Influence of muscle fibre type and fitness on the oxygen uptake/power output slope during incremental exercise in humans. Exp. Physiol.85 (1), 109–116. 10.1017/S0958067000019424

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 4

  BartlettR. G.Jr BrubachH. F.SpechtH. (1958). Oxygen cost of breathing. J. Apply. Physiol.12 (3), 413–424. 10.1152/jappl.1958.12.3.413

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 5

  BickhamD. C.GibbonsC.Le RossignolP. F. (2004). O₂ is attenuated above the lactate threshold in endurance-trained runners. Med. Sci. Sports. Exerc.36, 297–301. 10.1249/01.MSS.0000113667.64064.36

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 6

  BooneJ.BourgoisJ. (2012). The oxygen uptake response to incremental ramp exercise: methodogical and physiological issues. Sports. Med.42, 511–526. 10.2165/11599690-000000000-00000

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 7

  CoastJ. R.RasmussenS. A.KrauseK. M.O’KroyJ. A.LoyR. A.RhodesJ. (1993). Ventilatory work and oxygen consumption during exercise and hyperventilation. J. Appl. Physiol.74 (2), 793–798. 10.1152/jappl.1993.74.2.793

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 8

  DavisJ. A.WhippB. J.LamarraN.HuntsmanD. J.FrankM. H.WassermanK. (1982). Effect of ramp slope on determination of aerobic parameters from the ramp exercise test. Med. Sci. Sports. Exerc.14 (5), 339–343. 10.1249/00005768-198205000-00005

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 9

  DempseyJ. A.RomerL.RodmanJ.MillerJ.SmithC. (2006). Consequences of exercise-induced respiratory muscle work. Resp. Physiol. Neurobiol.151 (2-3), 242–250. 10.1016/j.resp.2005.12.015

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 10

  DominelliP. B.RenderJ. N.Molgat-SeonY.FosterG. E.SheelA. W. (2014). Precise mimicking of exercise hyperpnea to investigate the oxygen cost of breathing. Resp. Physiol. Neurobiol.201, 15–23. 10.1016/j.resp.2014.06.010

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 11

  Espinosa-RamírezM.Moya-GallardoE.Araya-RománF.Riquelme-SánchezS.Rodriguez-GarcíaG.ReidW. D.et al (2021). Sex-differences in the oxygenation levels of intercostal and vastus lateralis muscles during incremental exercise. Front. Physiol.12, 738063. 10.3389/fphys.2021.738063

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 12

  HansenJ. E.CasaburiR.CooperD. M.WassermanK. (1988). Oxygen uptake as related to work rate increment during cycle ergometer exercise. Eur. J. Appl. Physiol. Occup. Physiol.57, 140–145. 10.1007/BF00640653

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 13

  HarmsC. A.BabcockM. A.McClaranS. R.PegelowD. F.NickeleG. A.NelsonW. B.et al (1997). Respiratory muscle work compromises leg blood flow during maximal exercise. J. Appl. Physiol. (1985)82 (5), 1573–1583. 10.1152/jappl.1997.82.5.1573

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 14

  HarmsC. A.WetterT. J.McClaranS. R.PegelowD. F.NickeleG. A.NelsonW. B.et al (1998). Effects of respiratory muscle work on cardiac output and its distribution during maximal exercise. J. Appl. Physiol. (1985)85 (2), 609–618. 10.1152/jappl.1998.85.2.609

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 15

  HarmsC. A.WetterT. J.St CroixC. M.PegelowD. F.DempseyJ. A. (2000). Effects of respiratory muscle work on exercise performance. J. Appl. Physiol.89 (1), 131–138. 10.1152/jappl.2000.89.1.131

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 16

  HugF.DecherchiP.MarquesteT.JammesY. (2004). EMG versus oxygen uptake during cycling exercise in trained and untrained subjects. J. Electromyog. Kinesiol.14 (2), 187–195. 10.1016/s1050-6411(03)00081-6

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 17

  IannettaD.de Almeida AzevedoR.KeirD. A.MuriasJ. M. (2019). Establishing the O₂ versus constant-work-rate relationship from ramp-incremental exercise: simple strategies for an unsolved problem. J. Appl. Physiol.127, 1519–1527. 10.1152/japplphysiol.00508.2019

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 18

  JonesA. M.CampbellI. T.PringleJ. S. (2004). Influence of muscle fibre type and pedal rate on the O₂-work rate slope during ramp exercise. Eur. J. Apply. Physiol.91 (2-3), 238–245. 10.1007/s00421-003-0971-7

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 19

  LucíaA.RiveroJ. L.PérezM.SerranoA. L.CalbetJ. A. L.SantallaA.et al (2002a). Determinants of O₂ kinetics at high power outputs during a ramp exercise protocol. Med. Sci. Sports. Exerc.34, 326–331. 10.1097/00005768-200202000-00022

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 20

  LucíaA.HoyosJ.SantallaA.PérezM.ChicharroJ. L. (2002b). Kinetics of O₂ in professional cyclists. Med. Sci. Sports. Exerc.34 (2), 320–325. 10.1097/00005768-200202000-00021

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 21

  MarksD. W.RobergsR. A.NelsonJ.VellaC. A.Bell-WilsonJ.ApkarianM. (2005). Oxygen cost of ventilation and its effect on the O₂ plateau. J. Exerc. Physiol.8 (5), 1–13. Available online at: https://api.semanticscholar.org/CorpusID:99612902.

  • Google Scholar

• 22

  MarlesA.MucciP.LegrandR.BetbederD.PrieurF. (2006). Effect of prior exercise on the O₂/work rate relationship during incremental exercise and constant work rate exercise. Int. J. Sports Med.27 (5), 345–350. 10.1055/s-2005-86566

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 23

  NielsenM. (1936). Die Respirationsarbeit bei Körperruhe und bei Muskel arbeit. Acta Physiol.74, 299–316. 10.1111/j.1748-1716.1936.tb01161.x

  • CrossRef
  • Google Scholar

• 24

  NiemeyerM.KnaierR.BenekeR. (2021). The oxygen uptake Plateau-A critical review of the frequently misunderstood phenomenon. Sports. Med.51, 1815–1834. 10.1007/s40279-021-01471-4

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 25

  O’MalleyB. G. J.RobergsR. A.AstorinoT. A. (2024). Increases in the incremental exercise mean response time across the steady state domain: implications for exercise testing & prescription. Sports. Med. Health. Sci.6 (4), 315–323. 10.1016/j.smhs.2024.02.002

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 26

  OueslatiF.GirardO.TabkaZ.AhmaidiS. (2016). Excess O₂ during ramp exercise is positively correlated to intercostal muscles deoxyhemoglobin levels above the gas exchange threshold in young trained cyclists. Resp. Physiol. Neurobiol.228, 83–90. 10.1016/j.resp.2016.03.010

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 27

  OueslatiF.BooneJ.TabkaZ.AhmaidiS. (2017). Respiratory and locomotor muscle implications on the O₂ slow component and the O₂ excess in young trained cyclists. Resp. Physiol. Neurobiol.239, 1–9. 10.1016/j.resp.2017.01.012

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 28

  PedersenP. K.SørensenJ. B.JensenK.JohansenL.LevinK. (2002). Muscle fiber type distribution and nonlinear. O₂-power output relationship in cycling. Med. Sci. Sports. Exerc.34, 655–661. 10.1097/00005768-200204000-00015

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 29

  ScheuermannB. W.Tripse McConnellJ. H.BarstowT. J. (2002). EMG and oxygen uptake responses during slow and fast ramp exercise in humans. Exp. Physiol.87, 91–100. 10.1113/eph8702246

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 30

  SheelA. W.GuenetteJ. A. (2008). Mechanics of breathing during exercise in men and women: sex versus body size differences?Exerc Sport Sci. Rev.36 (3), 128–134. 10.1097/JES.0b013e31817be7f0

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 31

  ShephardR. J. (1966). The oxygen cost of breathing during vigorous exercise. Q. J. Exp. Physiol.51 (4), 336–350. 10.1113/expphysiol.1966.sp001868

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 32

  VellaC. A.MarksD. W.RobergsR. A. (2006). Oxygen cost of ventilation during incremental exercise to O₂max. Respirology11 (2), 175–181. 10.1111/j.1440-1843.2006.00825.x

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 33

  WakelingJ. M.UehliK.RozitisA. I. (2006). Muscle fibre recruitment can respond to the mechanics of the muscle contraction. J. R. Soc. Interface.3 (9), 533–544. 10.1098/rsif.2006.0113

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 34

  WhippB. J.DavisJ. A.TorresF.WassermanK. (1981). A test to determine parameters of aerobic function during exercise. J. Appl. Physiol. Respir. Environ. Exerc. Physiol.50, 217–221. 10.1152/jappl.1981.50.1.217

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 35

  ZoładźJ. A.KorzeniewskiB. (2001). Physiological background of the change point in O₂ and the slow component of oxygen uptake kinetics. J. Physiol. Pharmacol.52 (2), 167–184.

  • Pubmed Abstract
  • Google Scholar

• 36

  ZoładźJ. A.RademakerA. C.SargeantA. J. (1995). Non-linear relationship between O₂ uptake and power output at high intensities of exercise in humans. J. Physiol.488, 211–217. 10.1113/jphysiol.1995.sp020959

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 37

  ZoładźJ. A.DudaK.MajerczakJ. (1998). Oxygen uptake does not increase linearly at high power outputs during incremental exercise test in humans. Eur. J. Appl. Physiol. Occup. Physiol.77, 445–451. 10.1007/s004210050358

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 38

  ZoladzJ. A.DudaK.KarasinskiJ.MajerczakJ.KolodziejskiL.KorzeniewskiB. (2002). MyHC II content in the vastus lateralis m. quadricipitis femoris is positively correlated with the magnitude of the non-linear increase in the O₂/power output relationship in humans. J. Physiol. Pharmacol.53 (4 Pt 2), 805–821. Available online at: https://pubmed.ncbi.nlm.nih.gov/12510865.

  • Pubmed Abstract
  • Google Scholar