Cardiorespiratory fitness is positively associated with health status, and structured and individually tailored aerobic exercise training programs are recommended to improve cardiorespiratory fitness (^1–5). Structuring an aerobic exercise program involves the manipulation of several parameters (⁶) related to both the overall training regimen (e.g., weekly exercise frequency, volume, progression, etc.) and the single exercise session (e.g., duration, intensity, etc.). The aerobic exercise intensity continuum has long been divided into moderate, heavy, and severe domains, based on clearly identifiable physiological demarcation points (⁷). However, most of the world’s preeminent organizations have adopted a different approach, also used in the present investigation, which uses five intensity zones (e.g., see [¹]). Although the domains and zones do not match perfectly, mainly because of the highly individualized nature of responses to exercise (⁸), heavy and severe domains correspond approximately to the intensity zones vigorous and near-maximal to maximal, respectively, whereas the moderate domain comprises the intensity zones very light to light, light, and moderate.

Intensity is a fundamental consideration when tailoring an aerobic exercise prescription: light intensities are considered safe but may be insufficient to elicit the biological responses necessary to improve cardiorespiratory fitness (⁹), whereas vigorous intensity, although effective in improving cardiorespiratory fitness, may increase the health risks associated with exercise when individuals are not accustomed to it (^6,9).

Aerobic exercise intensity is usually prescribed and monitored with parameters calculated using either oxygen uptake (V˙O₂) or heart rate (HR), both of which increase with increasing aerobic exercise intensity. Indeed, studies investigating the association betweenV˙O₂ and HR during incremental exercise have generally found a linear relationship when values were expressed as percentages of maximalV˙O₂ (V˙O_2max) and maximal HR (HR_max), respectively (^10–12). However, the relationship between %V˙O_2max and %HR_max may be affected by interindividual differences in the maximal (as suggested by Swain and Leutholtz [¹³]) and/or resting values. On the contrary, using the “reserve” values, i.e., the difference between maximal and resting values, allows the correction for nonzero resting values. The concept of reserve, which was introduced by Karvonen et al. for HR (¹⁴), was applied toV˙O₂ by Swain and Leutholtz (¹³) in light of previous findings of Davis and Convertino (¹⁵). These investigations, focusing on young adults, showed that: (a) the percentages of the reserve values ofV˙O₂ (%V˙O₂R) and HR (%HRR) did not differ significantly at four different exercise intensities (¹⁵) and (b) %V˙O₂R and %HRR were strongly correlated and their regression was not distinguishable from the line of identity (¹³), i.e., slope = 1 and intercept = 0. Subsequent studies confirmed that %V˙O₂R and %HRR regressions did not significantly differ from the line of identity in healthy subjects (^16,17), in myocardial infarction (¹⁸), obese (¹⁹), and diabetic (²⁰) patients, or in elite amateur and professional cyclists (²¹).

However, it has been known for years (²²) that the association betweenV˙O₂ and HR, even under controlled conditions, may be affected by several factors such as body temperature, hydration, emotional state, physical activity level, sex, and day-to-day variability in HR response to exercise. Indeed, the actual association between %V˙O₂R and %HRR has been questioned in several reports. Swain et al. (²³) found that regression parameters differed significantly from those of the identity line in healthy adults, and subsequently, the same discrepancies have been found in children and adolescents (²⁴), in overweight and obese pregnant women (²⁵), and in obese (²⁶), CHF (^17,18), CAD (¹⁸), and heart transplant recipient (²⁷) patients. Furthermore, Cunha et al. (²⁸) found that the %V˙O₂R–%HRR relationship was significantly affected by the exercise testing protocol used. Importantly, they also found that %HRR was more closely associated with %V˙O_2max than it was with %V˙O₂R (²⁸), confirming the results of a previous study (²⁴). According to data from a review on this topic (²⁹), the HRR percentage point errors yielded by using the identity line to describe the relationship between %HRR andV˙O₂R of the above-mentioned studies ranged from −8% to +10% at 50% HRR and from −6% to +14% at 70% HRR. This means that, even close to the midpoints of moderate (40%–59% HRR) and vigorous (60%–89% HRR) ranges of the two recommended health-enhancing aerobic exercise intensity categories, the actual exercise intensity may be very close to, or even fall within, other intensity categories. This could lead to major errors in exercise intensity prescription and monitoring because HR is mostly used for this purpose.

Nonetheless, since 1998 (³⁰), the regression between %V˙O₂R and %HRR has been widely accepted as nonsignificantly different from the line of identity, as reported in the latest position stands of the major internationally recognized leading bodies in the field of physical activity and exercise (e.g., see [⁹]). Therefore, the main aim of the present study was to assess the actual relationships between %HRR and %V˙O₂R and between %HRR and %V˙O_2max using the large data set of the HERITAGE Family Study (³¹).

METHODS

Sample

The sample of the present investigation was composed of 737 members of Caucasian and African American families participating in the pretraining assessments of the HERITAGE Family Study (HERITAGE) (for details regarding ethics committee approval, inclusion and exclusion criteria, and study design, see Bouchard et al. [³¹]).

All subjects enrolled in the HERITAGE study (ranging in age from 17 to 65, inV˙O_2max from 15.2 to 54.9 mL·min⁻¹·kg⁻¹, and in HR_max from 136 to 214 bpm) were healthy (i.e., with no significant medical conditions or diseases), not physically active (i.e., they had not engaged in regular physical activity in the previous 6 months), and were not taking any medication that could affect resting and/or exercise HR.

HERITAGE Assessments

The design of the HERITAGE study included several exercise and nonexercise tests performed before and after an aerobic exercise training intervention. In the present study, only baseline (body weight and preexercise HR) and exercise testing (V˙O_2max tests) data of selected pretraining assessments were used (see below).

Body weight and preexercise heart rate

Body mass was measured to the nearest 0.1 kg using a balance beam scale. Resting HR (HR_rest) was measured immediately before the exercise test at the end of a 5-min rest period, with the subject sitting quietly in a chair.

Maximal oxygen uptake

Participants’V˙O_2max was defined based on the results of two cardiorespiratory fitness tests. First, a continuous, step-incremental exercise test to exhaustion (T1) was performed on a cycle ergometer (model 800S; Sensor Medics, Yorba Linda, CA) connected to a mixing-chamber metabolic cart (model 2900, Sensor Medics). In the first 3-min stage, participants pedaled at 50 W, then the resistance of the ergometer was increased by 25 W every 2 min until volitional exhaustion (in older, smaller, or less fit subjects, starting the test with a lower power output (PO) and/or making smaller increases every 2 min was permitted). At least 48 h later, a submaximal, steady-state exercise test, followed by a progressive test to maximum (T2), was performed. After the first phase of the test (which is not relevant to the present investigation and involved having the subjects exercise at a steady-state intensity of about 60% of theV˙O_2max measured in T1), participants pedaled for 3 min at the PO that was intended to correspond approximately to 80% of theV˙O_2max measured in T1. This PO was calculated using a linear interpolation of theV˙O₂ versus PO data recorded in T1. Thereafter, a 2-min stage at the highest PO attained in T1 was performed, and the resistance was then increased, if necessary, by the same increment used in T1, every 2 min until volitional exhaustion. Because the cycle ergometer was able to keep the PO constant regardless of the pedaling frequency, each participant was allowed to choose his/her own “comfortable” cadence (usually around 60 rpm).

In both tests,V˙O₂ (along with other gas exchange variables that were not used in the present investigation) was determined every 20 s and retained for subsequent analysis as the average of the last three 20-s values of each stage, whereas HR was measured continuously by means of ECG (to confirm HR, ECG rhythm strips were taken within the last 15 s of each stage and at maximum).

The criteria used for the attainment ofV˙O_2max were as follows: (a) a plateau inV˙O₂ (i.e., a change <100 mL·min⁻¹ in the last three consecutive 20-s intervals), (b) an HR within 10 bpm of the age-predicted HR_max, and (c) a respiratory exchange ratio >1.1. All participants met at least one of these criteria in one of the two tests (³²), but most subjects met two or more (³³). Hence, when theV˙O₂ peak of only one test met at least one criterion, it was assumed to be theV˙O_2max. When both T1 and T2V˙O₂ peaks met the criteria and the values were within 5% of each other, their average was calculated and assumed to be theV˙O_2max; otherwise, the highest value was assumed to be theV˙O_2max (³²). HR_max was assumed to be the highest value attained during either of the two maximal exercise tests.

Study Data Set Implementation

Before performing the calculations necessary to implement the data set used in the present study, the HERITAGE data were screened and filtered.

HERITAGE data set screening and filtering

Participants whose records had missing data in theV˙O_2max (and/or body weight), HR_rest, or HR_max fields were excluded. Subsequently, each stage of the T1 was inspected and deleted if either theV˙O₂ or the HR fields were missing. Finally, the data integrity of all the above-mentioned variables was assessed by means of range checks: when implausible physiological data were found, the whole participant record and/or the relevant stage(s) of the T1 were excluded (seeFig. 1 for details).

Data preparation and processing

EachV˙O₂ and HR recorded for each stage of the T1 was computed as a percentage of both the reserve and the maximum values using, respectively, the following two formulae: (a) 100 × (recorded value − resting value) / (maximal value − resting value) and (b) 100 × recorded value / maximal value. In the calculation of %V˙O₂R, the restingV˙O₂ was assumed to be 3.5 mL·min⁻¹·kg⁻¹, as suggested by the current American College of Sports Medicine guidelines (⁶).

Once calculated, %V˙O₂R, %V˙O_2max, and %HRR paired data points were used to perform the individual linear regressions (ILR) for the %HRR–%V˙O₂R and the %HRR–%V˙O_2max relationships. As suggested by Swain et al. (^12,13), a regression was performed for each participant, and the %HRR was set as the dependent variable. Data from ILR resulting from fewer than three paired data points were excluded because they were assumed to be potentially not accurate in representing the true underlying physiological relationship.

Statistical Analysis

The analyses were performed using Excel (Microsoft Office, version 2016), SPSS Statistics (IBM, version 20), and R (R Core Team, version 3.2.3; “Robust” package, version 0.4/16) software, with anα level of 0.05.

The study data set was filtered and analyzed twice using a univariate approach and a univariate–bivariate blended approach.

The procedures of the univariate approach were performed according to the methods of analysis routinely used in the literature to facilitate comparability among studies. Such procedures, which have been widely used in previous investigations (^{13,16–19,21,23–26}), were adopted in this study to verify the currently accepted HR–V˙O₂ relationships using the large and heterogeneous HERITAGE study data set, whose quality assurance and control have been supported by several reports (^34–37). Because the results of the univariate approach did not confirm those on which currently available guidelines are based, the data were also reanalyzed using a univariate–bivariate blended approach with more stringent data filtering procedures (to avoid any potential outlier-related bias) and multivariate analyses, providing additional statistically robust interpretations of the data.

In both approaches, data were adjusted for the familial clusters of the original HERITAGE data set (see the specific paragraph below for details). A flowchart illustrating the number of cases resulting from the analyses is presented inFigure 1. SeeTable 1 for details of the characteristics of the participants andTable 2 for the results of the analyses.

Univariate Approach

Data filtering

After excluding the linear regressions whose slopes were lower than zero, the slope of each linear regression was compared with zero using a two-tailed regression slopet-test. The regressions whose slopes were not significantly different from zero were excluded.

Statistics

For each relationship, the mean slope and intercept were compared with the line of identity (i.e., to 1 and 0, respectively) using two two-tailed one-samplet-tests with degrees of freedom corrected for familial clusters.

Univariate–Bivariate Blended Approach

Data filtering

For each relationship, paired data points were filtered using the DFFITS influential statistics and those having an absolute value of DFFITS larger than the size adjusted cutoff (i.e., double the square root of the ratio between the number of the regression’s parameters and the number of paired data points) were excluded, as proposed by Belsley et al. (³⁸). Because the DFFITS procedure requires at least four values to be performed, all the regressions resulting from fewer than four paired data points were also excluded. Subsequently, the ILR was run using the remaining paired data points, and those resulting from fewer than three paired data points and those with a slope lower than zero or not significantly different from zero (two-tailed regression slopet-test) were excluded as well (seeFig. 1).

Thereafter, because the dependent variables for both relationships showed significant correlations, slopes and intercepts were filtered using a bivariate procedure by adapting the ISO 13528:2015 rule (³⁹). Briefly, the 99% confidence ellipse was created using the robust mean and the variance–covariance matrix (calculated using the Huber M-estimator), and all paired data lying outside the ellipse were assumed to be bivariate outliers and excluded (Fig. 2).

Statistics

A test for Pearson’sr significance was performed to evaluate the correlation between intercepts and slopes of both the %HRR–%V˙O₂R and the %HRR–%V˙O_2max relationships.

The slopes and intercepts were used to build a mean vector$\left(\genfrac{}{}{0pt}{}{a}{b}\right)$ that was compared with the expected vector$\left(\genfrac{}{}{0pt}{}{0}{1}\right)$ using the bivariate Mahalanobis distance and the Wishart distribution.Post hoc univariate analyses were then performed using two two-tailed one-samplet-tests to compare the average slopes and intercepts to 1 and 0, respectively. The degrees of freedom used for Mahalanobis distance andpost hoc tests were those obtained from familial-cluster adjusted calculations.

The equations of the ILR retained were also used to calculate the predicted %HRR over theV˙O₂R andV˙O_2max continua (0% to 100%) for each subject. The mean %HRR predicted at 30%, 40%, 50%, 60%, 70%, 80%, and 90% ofV˙O₂R andV˙O_2max were then reported inTable 3 along with the relevant descriptive statistics and the 95% confidence intervals (CI) of the Cohen’sd effect size (ES). To be as conservative as possible, the CI values of the ES were calculated according to Lakens (⁴⁰), using the lower sample size resulting from the correction for familial clusters (i.e., 235; see then^corr row ofTable 2).

Finally, for each relationship, the average root mean square error (RMSE) was calculated as follows. For each participant, the difference between the actual %HRR and the %V˙O₂R or %V˙O_2max of each stage of the T1 was calculated. The sum of the squared differences was then divided by the number of stages completed before calculating the square root of each relationship and their averages. The RMSE values of the two relationships were compared using a two-tailed, paired-samplet-test (for the same reason described above, the sample size was set to 235).

Familial-cluster adjustments

To take into account the familial relatedness effect on each regression variable (seeTable 2), the following procedure was performed: (a) the eta squared (η²) for univariate ANOVA with random effect (family membership) was calculated (the dependent variables were either slope or intercept); (b) theη² was computed in the equation of Shieh (⁴¹) and an intraclass correlation coefficient was obtained; (c) the variance inflation factor was calculated using the intraclass correlation coefficient and the mean size of the grouped data; and (d) the variance inflation factor was used to calculate the corrected sample size (n^corr) for clustered data (⁴²).

RESULTS

The results are presented separately for each approach used.

Univariate approach

The intercepts of both %HRR–%V˙O₂R and %HRR–V˙O_2max regressions were significantly different from 0, whereas only the slope of the %HRR–%V˙O_2max regression was significantly different from 1 (seeTable 2). Additional information regarding the goodness of fit of the ILR can be found in the supplemental content (see Table, Supplemental Digital Content 1, Descriptive statistics of the goodness of fit of the ILR retained after applying the screening and filtering procedures to the original dataset of the HERITAGE Family Study,https://links.lww.com/MSS/C42).

Univariate–bivariate blended approach

Thet-test for the correlation index between the slopes and the intercepts revealed a significant correlation for both the %HRR–%V˙O₂R (r = −0.72,P < 0.0001) and the %HRR–%V˙O_2max (r = −0.79,P < 0.0001) relationships.

The Mahalanobis distance showed a highly significant difference between the mean vector$\left(\genfrac{}{}{0pt}{}{a}{b}\right)$ and the expected vector$\left(\genfrac{}{}{0pt}{}{0}{1}\right)$ for both the %HRR–%V˙O₂R (χ²₍₂₎ = 186,P < 0.0001) and the %HRR–%V˙O_2max (χ²₍₂₎ = 98,P < 0.0001) relationships.Post hoc univariatet-tests (seeTable 2) revealed that the slopes and the intercepts were significantly different from 1 and 0, respectively in both relationships (seeFig. 3 for a graphical representation of the regressions over the expected identity line). Additional information regarding the goodness of fit of the ILR can be found in the Table of the Supplemental Digital Content 1, which reports theirR²,r, and SEE,https://links.lww.com/MSS/C42.

The predicted %HRR values were different from the identity line (i.e., the expected zero ES did not lie within the 95% CI of the ES) for all the percentages calculated for the %HRR–V˙O₂R relationship, whereas for the %HRR–V˙O_2max relationship, they differed significantly from the identity line above 50% ofV˙O_2max (Table 3).

Compared with the %HRR–%V˙O₂R relationship, the average RMSE of the %HRR–%V˙O_2max relationship was significantly lower (7.78% ± 4.49% vs 9.25% ± 5.54%;t = 6.348,P < 0.001), with a mean difference of 1.47% ± 3.55% and an ES of 0.41.

DISCUSSION

The main finding of the present study was that the regression between %HRR and %V˙O₂R differed from the identity line, which conflicts with the currently accepted 1:1 relationship commonly recommended to prescribe and monitor aerobic exercise intensity. Furthermore, the %HRR–%V˙O_2max relationship appeared to be more similar to the identity line than the %HRR–%V˙O₂R relationship. However, the %HRR–%V˙O_2max relationship was not 1:1 either, and the similarity between the two percentages was disrupted at intensities above 50% of theV˙O_2max. In the present study, both the univariate and the univariate–bivariate blended approach provided results oriented in the same direction.

Univariate approach

When straightt-tests were performed on the data retained for the univariate approach, only the slope of the %HRR–%V˙O₂R relationship was nonsignificantly different from the expected result, whereas all the other comparisons showed significant differences.

The results of the univariate approach support those of several studies which found the %HRR–%V˙O₂R relationship to be different from the identity line during incremental exercises (^{17,18,23–27}), yet they conflict with the observations of other investigations that reported no difference between the relationship and the identity line (^13,16–21). Such conflicting results, however, may stem from methodological limitations and differences among the studies. First, several investigators (^17,27) set the %HRR as the independent rather than the dependent variable of the ILR, as suggested by Swain and Leutholtz (¹³). Second, in some of the investigations, the linear regressions were performed also including the resting values of the percentages of the reserve (^{13,16,17,21,23–25}); along with the maximal values, this could induce the slope and intercept to tend to 1 and 0, respectively. Third, in several studies (^{16–19,23,24,26}), the resting HR was not adequately measured, which could affect the extent of the reserve. As suggested by Swain et al. (¹²), a regression was performed for each participant in this study (resting values were excluded), and the %HRR was set as the dependent variable to accurately reflect the variability within the data and not to obscure the individual relationships. This method is theoretically correct from a physiological standpoint because HR does not elicit whole bodyV˙O₂, althoughV˙O₂ is the main factor determining the demand for HR. This method is also statistically correct because the transposition of a linear regression equation does not yield the same values as those that would be obtained if the regression had initially been performed with the dependent and independent variables reversed. In addition, another strength of the data retrieved from the HERITAGE study is that the procedure used to measure resting HR was in line with current guidelines, which recommend that it should be measured after at least 5 min of quiet rest, preferably with the subject in a position similar to the one assumed during the prescribed exercise mode (e.g., see [⁴³]). On the contrary, the restingV˙O₂ was not measured in all subjects of the HERITAGE study; thus, it was assumed to be 3.5 mL·min⁻¹·kg⁻¹ in the present investigation. Although this value is routinely used in the calculations performed for aerobic exercise intensity prescription (⁶), its use may represent a limitation of the present study. Indeed, using the recommended resting value ofV˙O₂ yields results that reflect more directly the daily practice of exercise specialists who prescribe the intensity of aerobic exercise for health-related purposes.

Univariate–bivariate blended approach

The multivariate inferential statistics showed that both the %HRR–%V˙O₂R and the %HRR–%V˙O_2max relationships were significantly different from the identity line, with all slopes and intercepts significantly different from 1 and 0, including the slope of the %HRR–%V˙O₂R relationship, which was not significantly different from 1 in the univariate approach. Compared with the results of the univariate approach, the ES and the mean differences of the slope and the intercept versus the identity line of the %HRR–%V˙O₂R relationship increased in the univariate–bivariate blended approach, whereas those of the %HRR–%V˙O_2max relationship decreased.

When the %HRR values predicted using the identity line were compared at different intensities to the %HRR values predicted using the ILR, the identity line showed an overall low accuracy for prescribing and monitoring different exercise intensities. Indeed, the HRR values predicted using the ILR were different from those predicted using the identity line for anyV˙O₂ ranging from 30% to 90% of the reserve, whereas they differed from the prediction of the identity line for anyV˙O₂ higher than 50% of the maximal. From a practical standpoint, when HR is used to prescribe exercise intensity according to the 1:1 relationship between %HRR and either %V˙O₂R or %V˙O_2max, the exerciseV˙O₂ tends to be overestimated. This means that the actual exercise intensities are lower than those expected close to the midpoints of both the moderate and the vigorous range of exercise intensity. Exercise intensity also affects the prediction error of the 1:1 relationship, which increases in the %HRR–%V˙O_2max relationship and decreases in the %HRR–%V˙O₂R relationship as the exercise intensity rises.

When %HRR–%V˙O₂R and %HRR–%V˙O_2max relationships were compared with the identity line, the difference between predicted and expected HRR, their percentage error, and ES appear to be higher in the %HRR–%V˙O₂R relationship than in the %HRR–%V˙O_2max relationship, suggesting that the %HRR–%V˙O_2max relationship is more similar to the identity line than the %HRR–%V˙O₂R relationship. Likewise, when the predictions of the identity line were compared with the actual values of the %HRR for each subject, the errors in using the identity lines were higher for the %HRR–%V˙O₂R relationship than for the %HRR–%V˙O_2max relationship, which confirms that the %HRR–%V˙O_2max relationship is more similar to the identity line than the %HRR–%V˙O₂R relationship.

The univariate–bivariate blended approach highlights that the relationships are different from the identity line and that %HRR–%V˙O_2max relationship has better agreement with the identity line. The latter result is in line with previous studies (^24,28) and in contrast with the postulated 1:1 relationship between the reserve values. This is particularly evident at lower exercise intensities despite the relatively high error, which is observed in both relationships but is higher for %V˙O₂R–%HRR.

Therefore, assuming a 1:1 relationship for the %HRR–%V˙O₂R and %HRR–%V˙O_2max relationships will yield high average errors in both relationships, although the average errors are lower in the %HRR–%V˙O_2max relationship, calling into question current guidelines.

Future directions

In the present study, we chose not to create subject subgroups (e.g., age, sex, and race) because the current guidelines adopt a 1:1 relationship between %HRR–%V˙O₂R for all subjects. However, the influence of those and other variables (e.g., body composition,V˙O_2max, resting HR, type of ergometer used, and incremental exercise adopted) on both relationships has not been adequately investigated. Therefore, future studies that consider all of these variables may be able to help account for the high variability of the HR–V˙O₂ relationships among different persons found in this study and enhance the accuracy of aerobic exercise intensity prescription.

Finally, the transferability of the relationships between HR andV˙O₂ from incremental to constant-load exercise is a much debated and controversial topic (^44,45) that has implications on the applicability of current aerobic exercise prescription guidelines. Indeed, studies have shown that the actual constant-load exercise intensity is mostly not predictable with accuracy by calculating fixed and standard percentages ofV˙O_2max, HR_max, or HRR measured using an incremental exercise test to exhaustion (^46,47). However, although prediction accuracy may be increased by using incremental exercise protocols with specific characteristics (^48,49), the prolonged duration of constant-load exercise yields physiological responses, such as cardiovascular drift, whose implications on the parameters usually adopted to control for exercise intensity (i.e., HR) need to be carefully considered (⁵⁰). Clearly, this topic is of great interest and still warrants further investigation.

CONCLUSIONS

The %HRR–%V˙O₂R and %HRR–%V˙O_2max relationships are slightly but significantly different from the identity line, and the %HRR is more closely associated with %V˙O_2max than %V˙O₂R. Importantly, the interindividual variability of the ILR (i.e., slopes and intercepts) and of the predicted %HRR at different exercise intensities is high, which suggests that using a standard and unique equation to predict aerobic exercise intensity can yield relatively high error in a single subject. In both relationships, the potential prediction errors of using the 1:1 relationship are relatively high. This shows the inadequacy of the 1:1 relationship in predicting exercise intensity and should raise the question of whether relying on the currently recommended equivalence between HRR andV˙O₂R to prescribe and monitor aerobic exercise intensity is still acceptable.

The authors thank Professor Steven E. Gaskill, who provided them with important information that gave them the initial impetus for this research, and Mr. Timothy C. Bloom for his linguistic revision of the manuscript. The HERITAGE Family Study was supported by multiple grants from the NIH (HL45670, HL47323, HL47317, HL47327, and HL47321). The authors state that the results of the study are presented clearly, honestly, and without fabrication, falsification, or inappropriate data manipulation. The authors report no conflict of interest. The results of the present study do not constitute endorsement by the American College of Sports Medicine.

REFERENCES

Supplemental Digital Content

• MSS_2020_06_22_LUCRETINI_19-01126_SDC1.pdf; [PDF] (141 KB)