Intensity and Duration of Physical Activity and Cardiorespiratory Fitness

A cross-sectional study of deidentified data collected during the prescreening process for the Oxfordshire Sedentariness, Obesity, and Cardiometabolic Risk in Adolescents: A Trial of Exercise in Schools study (NCT04118543) was undertaken. The study was approved by The University of Oxford Ethics Committee (Ref. R54302/RE006), was in accordance with the Declaration of Helsinki, and was written in accordance with Strengthening the Reporting of Observational Studies in Epidemiology guidelines.²³ 

Participants aged 13 to 14 years from two secondary schools in Oxfordshire, United Kingdom were studied in the 2018 to 2019 and 2019 to 2020 academic years. The schools informed families about the study and the opportunity was given for their children to be opted-out. Data were collected during physical education lessons run by the school, with support from the study team. Anonymous data on fitness indicators were collected to characterize the year-group. Adolescents who were opted-out participated in the physical education lesson but their data were not collected.

Height to the nearest 0.1 cm and weight to the nearest 0.1 kg were measured using a portable Harpenden Stadiometer (Holtain Ltd, Crymych, UK) and a SECA medical 770 digital floor scale (SECA, Hamberg, Germany), respectively. Body mass index (BMI) was calculated using height and weight and was additionally reported as age- and sex-adjusted z-scores, using WHO 2007 reference data.²⁴ 

CRF was measured using the 20 m shuttle run test (20mSRT; starting at 8 km/h, then 9 km/h, then increasing by 0.5 km/h at each subsequent stage) and recorded as the total number of completed shuttles.²⁵  Sex-specific CRF z-scores were calculated for all participants based on our sample of boys (n = 143) and girls (n = 124) with normal-weight (BMI z-score equal to or greater than −2 and less than +1).

Adolescents were provided with a small triaxial, wrist-worn accelerometer (AX3, Axivity Ltd., Melton Park, United Kingdom) and were asked to wear this for 7 consecutive days. Wrist-worn accelerometery has been shown to accurately predict children’s energy expenditure^26,27  and the AX3 accelerometer has been shown to accurately detect various child postures and PA intensities,^28,29  as validated by its successful use in large pediatric studies.³⁰  A valid recording consisted of at least 3 weekdays and 1 weekend day. Each valid day required >6 hours of awake wear time (>50% of awake time for most participants). We repeated the analyses using a criterion of >10 hours of awake wear time to ensure that this threshold did not alter the results. Sleep periods were not included in analyses. To define the noise floor of stationary AX3 sensors, we placed accelerometers on a flat surface, with no interference for 24 hours. From this, we defined nonwear as periods where the standard deviation (SD) and range of g for 1-second epochs varied by <0.018 g and <0.11 g, respectively, for at least 15 minutes.³¹  Nonwear periods were excluded from analyses. Although an algorithmic process was used initially to identify day and night periods and nonwear time, all data were then screened manually and edited if necessary.

The sampling frequency was set to 100 Hz with a dynamic range of ±16 g. A bespoke MatLab script calibrated the three axes using an established method (Supplemental Information).³²  After calibration, the signal was filtered using a fourth order Butterworth bandpass filter (high 0.2 Hz and low 15 Hz). Euclidean norm minus 1 (ENMO) was then calculated and summed into one second epochs. This metric of PA is referred to as bandpass-filtered followed by Euclidean norm (BFEN). Previously validated thresholds for BFEN were used to determine the average duration of daily PA at each intensity: 0.1 g for light (LPA), 0.314 g for MPA, and 0.998 g for VPA.³³  PA below the LPA threshold was categorized as sedentary time (ST).³⁴ 

Other metrics of PA processing were calculated as described previously and PA bands were defined using published thresholds relevant to each metric.³¹  The metrics were ENMOz (ENMO with negative values rounded to 0; thresholds: LPA, 0.1 g; MPA, 0.192 g; and VPA, 0.696 g),^35,36  HFEN (a fourth order Butterworth high-pass filter [0.2 Hz] applied to each of the three axes followed by EN; thresholds: LPA, 0.1 g; MPA, 0.314 g; and VPA, 0.998 g),³⁴  HFEN+ (HFEN plus ENMO of the three axes after a fourth order Butterworth low-pass filter [0.2 Hz]; thresholds as per HFEN). As there is a significant body of literature summarizing PA as ActiGraph counts, we also translated our signals in ActiGraph counts for comparability, using a validated method.³⁷ 

Statistical analysis was completed using Stata (version 16.1, StataCorp, College Station, Texas). Participants with data on sex, BMI z-score, 20mSRT, and a valid recording of PA were included. Histogram plots were used to assess normality of variables. Differences in the number of boys and girls were assessed using χ² tests. Sex differences were tested using independent t tests or Mann-Whitney U test and represented as mean ± SD or median (interquartile range), depending on distributions. Differences between PA metrics were assessed by analysis of variance and Bonferroni post hoc test or Kruskal-Wallis H tests and Dunn’s post hoc test, depending on distributions.

An individual’s PA at each intensity is strongly correlated with their PA at other intensities (Supplemental Table 3), preventing robust comparison of the effects of different intensities on health outcomes. This can be addressed by constructing variables that contain only the variance at each intensity that is independent of all other intensities (ie, zero-correlated, Supplemental Table 4). We did this using partial multivariable linear regression modeling³⁸  to generate a set of residualized variables for MPA, LPA, and ST, which together with their base variable (unadjusted VPA), are referred to as rPA variables (Supplemental Information). To check that the directionality of this partial modeling did not significantly influence the results, we repeated this process in the opposite direction, starting with ST as the base variable, to generate another set of rPA variables. Standard multivariable regression with all the normal PA intensities was also used to support the findings of the partial models.

We tested the association of each rPA variable with CRF, adjusting for sex, in separate models and, given their mutually independent nature, in a fully adjusted model including all rPA variables. These results were compared with models using normal PA variables.

To capture the underlying relationships between rPA and CRF, we used a filtering mechanism to smooth the raw data, removing any “noise,” and allowing the trends in our data to be revealed. This approach differs from more common statistical methods such as linear regression that fit a function with a predefined shape to the data. By contrast, we used a zero-phase moving-average filter, which makes no assumption about the shape of the relationship and imposes no linearity constraint, to determine relationships between each rPA variable and CRF. The Akaike information criterion (AIC) was used to determine the moving average window size that yielded the optimum goodness-of-fit of each model before overfitting occurred. The threshold at which CRF might no longer improve significantly with increasing PA was determined as the point where any positive increase began to level off. Further details are provided in Supplemental Information. Comparison was made with linear, second order polynomial, and moving median models to demonstrate which approach best fitted the data. The duration of PA needed to achieve median CRF was determined as the point where the moving average model reached a CRF z-score of 0.

To support the results of the moving average model, sex-specific mean differences in VPA between participants ≤50^th percentile (group 1) and participants >50^th percentile (group 2) of CRF were assessed using independent t tests. A range of CRF percentile thresholds were determined from our own 20mSRT data, and from the mean 20mSRT data from European and global reference ranges of 13- and 14-year-olds, for comparison.^39,40 

To assess differences in CRF between low and high amounts of PA, the upper and lower quartiles of each normal PA variable and of each rPA variable were assessed using independent t tests. Statistical significance was set at P < .05.

Of the 480 adolescents who participated in the health assessment, 339 had sufficient, valid data allowing for inclusion (Fig 1). Their key characteristics are reported in Table 1. Height, weight, BMI, and BMI z-score data were missing for 2, 5, 5, and 6 participants, respectively. Boys were taller than girls, but there were no sex differences in weight, BMI, or BMI z-score. Boys performed better in the 20mSRT and did more VPA. Girls had higher MPA, LPA, and ST measures. PA determined by other metrics are reported in Supplemental Table 5. Adolescents that did not have valid PA wear time had similar characteristics, but had lower z-score total lap results, compared with those with valid PA (Supplemental Table 6). Although not directly assessed, schools covered a range of socioeconomic and ethnic groups (Supplemental Information).

Given the sex differences in PA shown in Table 1, all linear models were adjusted for sex. Linear regression of each PA intensity with the CRF measures (total laps and total laps z-score) demonstrated the strongest association with VPA and progressively weaker associations with MPA and LPA (Supplemental Table 7, Model 1). Standard multivariable regression showed that only VPA was independently associated with CRF after adjustment for the other intensities (Supplemental Table 7, Model 2). Further modeling using partial regression was then conducted to examine the independent effects of each intensity. This was done with the rPA variables individually (Supplemental Table 7, Model 3) and in mutually adjusted models with the rPA variables calculated with either VPA (Supplemental Table 7, Model 4) or ST (Supplemental Table 7, Model 5) as the base variable. Only VPA was independently associated with CRF, irrespective of analysis approach. Having thus demonstrated the benefit of the partial regression modeling approach, only rPA variables were used in subsequent analyses.

In the moving average model, greater average daily VPA was associated with better CRF until 19 minutes of VPA, when the relationship plateaued (r² = 0.35; Fig 2). Median CRF was found in adolescents performing 14 (range 12–17) minutes of VPA daily. The range was estimated as the VPA values where the 95% confidence limits of CRF z-scores crossed the median. Results did not change meaningfully when the awake wear time threshold was set at >10 hours (n = 302; plateau = 20 minutes; range 12–17; r² = 0.36). As in the linear models, other rPA measures were unrelated to CRF (Fig 3). Moving average models using other metrics of PA demonstrated similar results to BFEN (Supplemental Figs 4-8). The moving average model provided the best fit of the data when compared with linear, second order polynomial, and moving median models (Supplemental Fig 9). When analyses were repeated by sex, there were steep increases in CRF with increasing VPA up to 20 and 18 minutes in boys and girls, respectively, plateauing after that (Supplemental Fig 10). However, these models were not statistically different from each other (overlapping confidence limits).

To support the findings of the moving average model, independent t tests assessed mean differences in VPA between individuals below (group 1) and above (group 2) a number of different median CRF thresholds (Table 2). Boys and girls in group 2 undertook significantly more VPA than those in group 1, regardless of threshold definition. There were no significant group differences in any of the other rPA intensities. Similarly, CRF z-score differed when comparing the upper and lower quartiles of VPA, but not those of any other rPA measure (1.03 z-scores, 95% CI = 0.75 to 1.30; Figs 2 and 3).

Habitual PA is known to have many health benefits, including higher levels of CRF in all age-groups. We showed that greater average VPA per day was associated with better CRF in our adolescent population up to about 20 minutes of VPA per day. The association plateaued after this, with even greater VPA durations not being associated with significant further improvement of CRF. After intercorrelations between PA intensities were controlled for, only VPA was associated with CRF. As current guidelines set PA activity targets using a combined metric of MVPA that could be satisfied by undertaking only MPA and not VPA, we provide grounds for clearer public health messaging on how to improve CRF in this population. Most adolescents fail to achieve the current recommendation of ≥60 minutes of MVPA per day on average.^18,19  One possible reason is that this duration is quite long, requiring a daily time commitment that some may find difficult to maintain. A shorter target of 20 minutes might be easier to schedule daily and a focus on VPA would simplify messages about the intensity of activity that is likely to improve CRF.

Our findings support those in adults where 15 minutes of VPA was shown to improve both CRF and glycemic control,¹⁶  and are similar to findings in children and adolescents.¹⁵  The current guidelines provide recommendations based on a combined metric of MVPA, which does not specify the duration of either MPA or VPA.¹⁸  Our approach provides such evidence and highlights the importance of VPA over lesser intensity PA for CRF.

Our findings are consistent with other cross-sectional studies in highlighting the benefits of VPA.^15,41  However, to our knowledge, no prior study has attempted to examine the independent effects of different PA intensities, controlling for their intercorrelation. Our approach suggests that, although past reports generally acknowledge the inferiority of MPA or lesser intensities of PA, conclusions about MPA could have been erroneously drawn because of the confounding strong correlation between VPA and MPA.

We showed no additional effect of MPA or lower intensity PA on CRF beyond the influence of VPA. Nevertheless, it is possible that other health-related measures that we did not study may improve with lower intensities of PA.⁴²  However, as we showed, it is important to first control for intercorrelations between PA intensities before examining associations with any health-related measures. Nevertheless, CRF remains an important predictor of cardiometabolic health outcomes.^6–9  It is notable too that earlier studies have generally found that only very high amounts of MPA are associated with improved health outcomes.¹  In studies of exercise interventions, VPA interventions were found to have similar adherence rates to those with lesser intensity, but with reduced time commitments, highlighting no apparent disadvantage of focusing on VPA.⁴¹  Cross-sectional and longitudinal evidence also suggests that VPA improves adiposity and cardiometabolic health to a greater extent than MPA.^15,41,43 

Undertaking 20 minutes of daily VPA to improve CRF may provide the means for adolescents to improve their long-term health outcomes at scale.^6–9,20,21  However, our study design does not address causation in the associations we have shown. Although reverse causation may partially explain our findings, prior knowledge on the relationship between exercise and CRF suggests this is unlikely to be a dominant explanatory factor.¹⁴  Future studies should also aim to evaluate how best to achieve a target of 20 minutes of daily VPA in large populations, as recent studies have shown that it can be difficult to achieve similar VPA targets at scale.⁴⁴ 

We found that, on average, girls undertook less PA and had lower CRF than boys, which is in-line with previous studies.^19,45  However, we found no evidence that our models of the relationship between PA and CRF differed significantly according to sex, supporting the use of the combined model and unified recommendations for all.

This work has a number of strengths and potential limitations. Our findings are based on data from two Oxfordshire, United Kingdom schools and therefore, might not be generalizable to broader populations. However, the sample was well-balanced by sex, covered a range of socioeconomic and ethnic groups, and had a wide distribution of BMI and CRF. Although we acknowledge that our study needs replicating, importantly, there was no significant difference in shuttle run performance between our data and European and Global normative data, suggesting that our population is representative.^39,40  Recently the validity of the 20mSRT as an indirect assessment of CRF has been both questioned⁴⁶  and defended.⁴⁷  Although its utility and validity for ranking fitness in large populations of children is supported in numerous contemporary reviews,^48–51  we acknowledge that replication of our findings with more direct assessment of VO₂max would be desirable, although challenging in practice. There is no universally-accepted metric to analyze PA derived from accelerometry.³¹  We took care in our data processing to calibrate our accelerometery and ensure accurate, manually curated, identification of wear periods. We also used BFEN in our analyses, as this was shown to classify VPA better than other metrics in our target age-group.³³  However, we repeated our analyses using the variety of other PA metrics favored in the literature, which did not meaningfully alter our results.

We provide evidence that adolescents who undertake 20 minutes of daily VPA on average have maximal CRF, with little evidence of additional benefit from undertaking more or from lesser intensities of PA. This modest duration and more specific intensity of PA may provide a better underpinning for future guidelines that currently recommend a longer duration of less specific activity (MVPA) each day. Further work should aim to test whether interventions based on this new target offer significant improvements in adolescent cardiometabolic health.

AIC

Akaike information criterion

BFEN

bandpass-filtered followed by Euclidean norm

BMI

body mass index

CRF

cardiorespiratory fitness

CVD

cardiovascular disease

ENMO

Euclidean norm minus 1

ENMOz –

Euclidean norm minus 1 with negative values rounded to 0

HFEN

highpass filtered Euclidean norm

HFEN+

highpass filtered Euclidean norm plus

LPA

light physical activity

MPA

moderate physical activity

MVPA

moderate to vigorous physical activity

PA

physical activity

rPA

residualised physical activity after partial multivariable regression modelling

ST

sedentary time

VPA

vigorous physical activity

WHO

World Health Organization

20mSRT

20 meter shuttle run test