Reference values for cut-point-free and traditional accelerometer metrics and associations with cardiorespiratory fitness: a cross-sectional study of healthy adults aged 20 to 89 years

Study design

Data were collected in the population-based, cross-sectional COmPLETE study at the Department of Sport, Exercise and Health at the University of Basel, Switzerland. The study protocol is available elsewhere.[9] All procedures were performed according to the Declaration of Helsinki. The study was registered on clinicaltrials.gov (NCT03986892) and approved by the Ethics Committee of North-western and Central Switzerland (EKNZ 2017– 01451).

Study participants

The COmPLETE study examined healthy adults (HEALTHY) and patients with chronic heart failure (HEART). HEALTHY were ≥20 years of age, non-smoking, and had a BMI <30 kg/m².[9] Exclusion criteria were the presence of chronic exercise-limiting disease, known pregnancy or breastfeeding, hypertonic blood pressure >160/100 mmHg, orthopaedic problems hindering the examinations, inability to follow the study procedures, and contraindications for all-out exercise.[9] Individuals aged ≥20 years and diagnosed with stable chronic heart failure were eligible for HEART.[9] All exclusion criteria and a rationale for the study size are presented elsewhere.[9]

Study procedures

Statistical analyses

Statistical analyses and figures were done in R version 4.2.0.[13] Correlations between accelerometer parameters and VO_2peak were analysed using Spearman’s Rank Correlation. Explorative principal component and partial least squares analyses were performed to examine the relationship between the abovementioned variables.[14, 15] Multiple linear regression models were used to examine between-cohort differences and the association between VO_2peak and accelerometer parameters, i.e. IG, AvAcc, MVPA, VPA and time in PA (TPA=light PA+MVPA). For each cohort, we planned to create three models with cut-point-free metrics only (IG, AvAcc, and both parameters together), two models with cut-point-based metrics only (MVPA and VPA), and two models containing IG or VPA as a measure of PA intensity and TPA as a surrogate of time in PA. All models were adjusted for age and an interaction between BMI and sex. We used restricted cubic splines with four knots placed at appropriate percentiles of the data for age and BMI.[16] Fulfilment of model assumptions was verified using residual diagnostics. The best model fit was selected based on residual standard deviation and adjusted R². A Likelihood-Ratio-Test was used to compare models. Non-nested models were compared using the Vuong Test.[17] Percentile curves were created with Generalised Additive Models for Location, Scale, and Shape with the GAMLSS package (version 5.3-4).[18] Age or VO_2peak as the independent parameter was fitted using p-splines.[19, 20] We adopted the Bayesian information criterion to select the conditional distribution offering the best compromise between model complexity and goodness-of-fit. Model fits were inspected using diagnostic residual plots.[21] We developed the application ‘rawacceleration’ using ‘shiny’[22] to translate our findings into clinical practice. The level of statistical significance was set at 0.05. All tests were two-sided.

Between-cohort differences

HEART had significantly lower IG (p<0.001, 95% CI: -0.185 to -0.079), AvAcc (p<0.001; 95% CI: -8.59 to -3.92 mg), MVPA (p<0.001; 95% CI: -45 to -20 min), VPA (p=0.002; 95% CI: -6.1 to -1.3 min), and TPA (p<0.001; 95% CI: -80 to -33 min) than HEALTHY (see Table 1). HEART were more sedentary than HEALTHY (p=0.006; 95% CI: 10 to 59 min).

Interrelationship between accelerometer metrics and VO_2peak in HEALTHY and HEART

Correlations of VO_2peak, IG, AvAcc, MVPA, VPA, TPA, and sedentary time are available in Figure 1.

• Download figure
• Open in new tab

Figure 1.

Spearman correlations of average acceleration (AvAcc), intensity gradient (IG), moderate-to-vigorous physical activity (MVPA), sedentary time (SED), time in physical activity (TPA), peak oxygen uptake (VO2peak), and vigorous physical activity (VPA) for healthy individuals (HEALTHY) and patients with heart failure (HEART), respectively. Non-significant correlations are crossed.

Results of the explorative principal component and partial least squares analyses are presented and discussed in Supplement 2.

Multiple regression models with key accelerometer parameters as independent variables and VO_2peak as the dependent variable are presented in Table 2. In HEALTHY, model 2 (AvAcc) had a better fit than model 1 (IG) (p=0.022). Model 7 (AvAcc+IG) performed better than model 1 (IG) (p<0.001) and model 2 (AvAcc) (p=0.004). The best cut-point-free model including AvAcc and IG (model 7) and the best cut-point-based model including VPA (model 4) performed similarly.

When combining IG as a metric of PA intensity and TPA as a surrogate of PA time (Model 5), both metrics were independent predictors of VO_2peak. However, when including VPA and TPA (model 6), there was little evidence for the latter being an independent predictor of VO_2peak. In HEART, there was little evidence for a difference in fit between model 1 (IG), 2 (AvAcc), or 7 (IG+AvAcc). Both IG and AvAcc were associated with VO_2peak with only IG being independent of AvAcc but not vice versa. Moreover, the best model with cut-point-based metric (model 3 with MVPA) performed equal to models 1 (IG) and 7 (AvAcc+IG), but was superior to model 2 (AvAcc) (p=0.012). Adding TPA to the model with IG or VPA did not significantly improve model fits (models 5&6).

Reference values for healthy adults

Sex-specific percentile curves were created for IG, AvAcc (see Figure 2), MVPA, and VPA (see Supplementary Figure 3). Empirical data for IG and AvAcc per age decade are presented in Table 3. Data for MVPA and VPA are shown in Supplementary Table 2.

• Download figure
• Open in new tab

Figure 2.

Percentile curves for intensity gradient (A & B) and average acceleration (C & D) in relation to age in healthy adults. The left column shows data for females and the right column for males. Mean values of IG and AvAcc of other cohorts were plotted to highlight the potential of the reference curves for comparability across studies and between devices. Data by Rowlands et al. (1): pre- and post-menopausal women (n_pre=1218, n_post=1316; device: Axivity). Dempsey et al. (23): participants of the UK Biobank study (females, n = 51,509; males, n=36,903, device: Axivity, dominant wrist). Our data of patients with heart failure (HEART; device: GENEActiv) were also plotted.

CRF-specific percentile curves for IG and AvAcc are available in Figure 3.

• Download figure
• Open in new tab

Figure 3.

Percentile curves for intensity gradient (A & B) and average acceleration (C & D) in relation to VO_2peak in healthy adults stratified by sex. The left column shows data for females and the right column for males. Abbreviations: VO_2peak, peak oxygen uptake.

We developed the application ‘rawacceleration’ using these reference data. It is freely available on GitHub (https://github.com/FSchwendinger/rawacceleration). Further insights into ‘rawacceleration’ are given in the Supplement and discussion section.

Inter-relationship between accelerometer metrics and VO_2peak in HEALTHY and HEART

We confirmed previous findings in children and adults with type 2 diabetes suggesting AvAcc and IG provide a complementary picture of PA.[6, 7] Our study demonstrated this in healthy adults and patients with heart failure by examining correlations between accelerometer metrics. These findings are strengthened by our explorative principle component analysis. Lower communalities were seen for IG than for AvAcc or MVPA indicating IG may yield a perspective on PA not covered by other metrics. AvAcc and IG may thus provide a more complete impression of an individual’s PA profile without having to rely on accelerometer cut-points.

AvAcc and IG have been described to reflect PA volume and intensity, respectively.[6] Yet, correlations between TPA, VPA, IG, and AvAcc indicate that AvAcc conveys information on both PA intensity and volume. It should therefore be noted that cut-point-free metrics allow the further investigation of intensity/volume, despite their overlap, as IG captures the distribution as well as quantity of intensity. This is more difficult with cut-point-based metrics because of their categorical and population-dependent nature. Thus AvAcc and IG may be particularly appropriate when attempting to differentiate between the contribution of PA intensity and volume to CRF as well as other health parameters (e.g. body fatness, BMI, physical function).[6, 7]

IG and AvAcc together had the greatest predictive value for VO_2peak in healthy adults and both were independently associated, explaining a unique fraction of the variance. This is partly compliant with a study in children that estimated CRF from a shuttle-run-test.[7] IG or AvAcc as the only independent variable in the adjusted regression models was significantly associated with estimated CRF.[7] Yet, only IG remained significantly associated with CRF when AvAcc was added to the model.[7] AvAcc may consequently be more relevant in adults than in children. This is possibly due to different activity patterns with higher IG and AvAcc together with less sedentary time in children.[7] This highlights the utility of combining these two metrics to gain insight into the relative importance of volume and intensity of the PA profile which differs across populations.[6]

Interestingly, TPA significantly added to model 5 also containing IG. Both intensity distribution (IG) and time spent in PA (TPA) may thus be relevant for VO_2peak. Notably, AvAcc captures both time and intensity of PA, thus reflecting activity volume, while IG captures the distribution of the intensity of activity. Combining IG and AvAcc thus yields complementary perspectives on PA intensity.

In patients with heart failure, cut-point-free metrics had similar predictive value for VO_2peak as the best cut-point-based metric, MVPA. Yet, unlike in healthy adults, only IG was independent of AvAcc but not vice versa. This may reflect relatively little activity of higher intensity in these patients; it is in contrast with findings of Dawkins et al. (25) and suggests that a focus on intensity of activity, i.e. increasing the breadth of the intensity distribution, may be more important than volume if intervening in this population.

This study demonstrated that IG and AvAcc are not only potentially more robust accelerometer metrics than MVPA, VPA, or TPA, as they do not rely on cut-points, but also have similar predictive value for VO_2peak and, in turn, indirectly for the risk of mortality and longevity.[8, 26] Indeed, Dempsey et al. (23) recently demonstrated that AvAcc and IG were independently associated with incident cardiovascular disease in participants from UK Biobank. These results and those of previous studies[6, 7] provide an evidence-based rationale that cut-point-free accelerometer metrics facilitate capture of the volume and intensity distribution of the PA profile across populations, and thus may be a viable alternative to cut-point-based outcomes in the measurement of PA.

Reference values for healthy adults

Our reference values and percentile curves may be pivotal to assessing the appropriateness of PA on population-or individual-level and to determining the level of PA necessary for a healthy lifestyle.[6]

IG and AvAcc at the 50^th percentile are higher than in participants of the UK Biobank Study aged 40 to 69 years, despite them wearing the accelerometer on their dominant hand which typically elicits accelerations approximately 10% higher.[1, 27] This seems reasonable in light of the good health condition (see inclusion/exclusion criteria) and the likely higher CRF[12] of the COmPLETE cohort compared to the aforementioned study. Hence, our reference values might be desirable benchmarks for healthy PA patterns in adults.

Age trajectories of IG and AvAcc were curvilinear and inverse for both sexes (see Figure 2). This is in line with the accelerometer data of another Swiss adult cohort.[28] Interestingly, the decline in IG and AvAcc with age seems to resemble that of CRF and muscular strength in the COmPLETE cohort.[12, 29] With age, less time is spent at higher absolute intensities which might be due to reduced physical functioning as apparent by lower CRF and muscular strength. However, when expressed in relative terms (as % of an individual’s VO_2peak), intensity may still be high. This explanation is underpinned by questionnaire data, which may also reflect the rate of perceived exertion, showing more moderate and vigorous PA in individuals aged > 59 years than in 40 to 59-year-olds.[28] Moreover, the inverse relationship seen between accelerometer-based VPA and age seems not to be apparent in questionnaire data.[28]

The proposed reference values may also be useful when PA is measured with other raw-acceleration, wrist-worn accelerometers. Migueles et al. (24) found high inter-instrument reliability for four common accelerometers (GENEActiv, Movisens Move 4, ActiGraph GT3X+, and Axivity AX3) provided that raw data are processed identically. The between-device difference ranged between 1 (95% CI: -6 to 7) and 8 (95% CI: 1 to 15) min per day.[24] Our reference values may thus be applicable for at least the examined accelerometers (i.e. GENEActiv, Movisens Move 4, ActiGraph GT3X+, and Axivity AX3).[24]

Reference values and percentile curves for cut-point-based metrics (VPA and MVPA, see Supplement: Figure 3 and Table 2) were also created. These are still commonly used and directly interpretable. Comparison to these data is however only possible when using the cut-points by Hildebrand et al. (3) (MVPA=0.1 g and VPA=0.4 g) and processing data identically.[24] Due to the aforementioned disadvantages of cut-point-based accelerometer metrics (also discussed elsewhere[5]), cut-point-free metrics should be reported in future studies examining PA with raw-acceleration accelerometers.

Importantly, the reference values for traditional metrics are specific to accelerometer data and examine the whole activity profile. They are not comparable to e.g. the PA guidelines by the WHO which focus on MVPA and were developed predominantly from self-reported data.[2, 30] This is supported by numerous studies.[31, 32]

Our reference values facilitate interpretation of accelerometer-assessed PA data relative to a ‘healthy norm’, and may thus be an important complement to the current PA guidelines by the WHO.[30]

Limitations

PA levels differ between countries.[33] Nonetheless, it seems to be desirable to promote the present reference values also in countries other than Switzerland considering the good general health of our cohort in combination with their high CRF.[12] Importantly, the present reference values are only applicable if the data are processed similarly (see Methods section and configuration file). The R-package GGIR will facilitate this process.[10] Large differences in sampling frequency of accelerometer data may impact outcomes. In a study comparing 25 Hz with 100 Hz, the lower sampling frequency led to 3.1%–13.9% lower overall acceleration values.[34] The difference may however be smaller at 50 Hz, as used in our study. Yet, sampling frequency should be considered when comparing PA metrics to the present reference values. Available algorithms may help transform PA metrics to match those obtained from higher sampling frequencies.[34] Lastly, no causation between PA and VO_2peak can be implied from the present study.