The principal finding of this study was that the use of ICPs to determine the minutes of MVPA/day measured by accelerometry influenced the PA output substantially. Thus, minutes of MVPA determined by ICPs and a GCP varied considerably, indicated by low to moderate correlations and CVs of about 50% between the measures. This means that the use of ICPs will lead to quite different results on an individual level compared to using a common cut point on a group level. However, it is difficult to conclude with respect to the most appropriate procedure for analyzing time in MVPA.
The finding of a strong association between the ICPs and the measured minutes of MVPA/day was unexpected, as this indicates that subjects having lower ICPs were more physically active than subjects having higher ICPs. However, as we showed that two thirds of the variation in ICPs are due to resting VO2, work economy and BMI, most of the variation can be claimed to be true (although measurement error will be present). Since ICPs also attenuated the effects of all other independent variables used to explain the minutes of MVPA, one can argue that we managed to successfully determine ICPs that accounted for sources of variation that influenced work rate and thereby PA level. Thus, we can conclude that the use of ICPs in the present study increased the precision of the PA measurement. However, we will point out two main challenges when drawing this conclusion. First, one third of the variation in the ICPs was unexplained. Differences in gait patterns (beyond work economy), accelerometer-units worn, the attachment and tilt of the instruments are factors likely to explain this variation. The finding of a relatively large SEM/LoA for cut points obtained from the right and left hips, may support this notion (although these differences also could be attributed to biomechanical variations between the dominant and non-dominant side of the body). The attachment of the accelerometer may be a crucial aspect of the precision of the measurement process because tilting of the accelerometer or small differences in hip-placement are known to influence the accelerometer output
[10, 28]. We expected that the cut points derived from two accelerometers would be fairly similar and that typical deviations would be under 8.9%
[10–12]. Because our data were homoscedastic, such a percentage difference is difficult to evaluate. However, the SEM amounted to approximately 27% of the group mean cut point. This result shows that the uncertainty in defining ICPs is substantial and that the percentage variation is greatest at the lower end. Further, because a regression toward the mean effect tended to be supported by the study data (i.e. subjects having high or low cut points from the right accelerometer had less extreme cut points from the left accelerometer), the established ICPs probably shouldn’t be viewed as “true” ICPs. It should be mentioned that the placement of the accelerometers probably would be less standardized in a field setting, compared to our laboratory setting, which would introduce even greater errors in the PA field data
. This unexplained variation will also be carried forward and influence the analyses of PA level. Our results revealed that use of ICPs to explain minutes of MVPA increased the explained variance with 10 – 20%, compared to the use of resting VO2, work economy and age as independent variables. This could imply that a certain degree of measurement error influence the results. Furthermore, measurement error in determining VO2 at rest and treadmill walking will add to this measurement error. Finally, any deviations in gait pattern and work economy between the treadmill walking and field will disturb the findings.
Second, if we believe that individual calibrations increase the measurement precision, we have to accept that use of a GCP may be more or less useless to determine minutes of MVPA on an individual level, as typical deviations is about 50% and shared variance is only 14% for total minutes of MVPA/day derived from ICPs vs. GCP. These differences are caused by great diversity in treadmill walking speed at three METs, which is accounted for by applying the ICPs. However, for describing PA level on a group level, applying ICPs or a GCP may lead to quite similar results, although we recognize that the PA levels derived from the GCP were about 40% higher than the PA level derived from the ICPs (because the estimated cut point from the regression model were somewhat lower than the mean of the ICPs) and that our study may suffer from lack of statistical power to detect a significant difference.
The present results suggest that severely obese individuals achieve a moderate work rate at very different walking speeds (between 1.1 and 3.5 km/h). Thus, because all movement detected above each individual’s walking speed threshold is interpreted as MVPA, it is not surprising that we found a strong association between ICPs and PA level. As stated above, the use of ICPs may significantly alter research findings compared to applying a GCP. Here we will consider three relevant areas. First, an interesting discussion is whether the use of ICPs can increase our ability to detect relationships between PA and diverse health outcomes. However, at this point we have no evidence-based suggestion for how different outcomes would be affected. Nevertheless, the difference between ICPs and GCP would probably be greater in cross-sectional analysis compared to experimental studies were the same cut point is applied repeatedly on an intra-individual level. Second, the use of ICPs vs. a GCP could influence whether subjects are found to achieve PA guidelines or not. In the present study we found a significant degree of re-classification between application of ICPs vs. the GCP (70% agreement, Kappa coefficient = 0.40, p = .299, result not shown) in the analyses of whether subjects achieved ≥ 30 min in bouts of > MVPA/day or not. Third, work rate can be determined absolutely (oxygen consumption or standardized MET-values etc.) or relative to an individual’s maximal work capacity (percentage of VO2max, etc.). This may challenge exercise prescription for subjects having low fitness levels, as “moderate” intensity exercise in absolute terms, can actually be quite demanding
. We could hypothesize that if resting metabolic rate, work economy and/or BMI (which determined the ICPs) was related to VO2max, applying ICPs could possibly relate better to each individual’s maximal capacity than a common GCP. However, we did not find any statistically significant relationship between ICPs and VO2max (ml/kg/min) (r = 0.17, p = .320, result not shown).
Strengths and weaknesses
The main strength of the present study is the use of precise and sophisticated measurements of the metabolic cost of walking and free-living PA. Moreover, the study gives an overview of the measurement properties of accelerometers and the use of individual cut points. Finally, the inclusion of a relatively large sample of subjects ensures the validity of our results.
This study has several limitations. The main limitation is that we do not have a valid criterion measure of minutes of MVPA/day, meaning that a direct comparison of the precision of a GCP vs. the ICPs could not be established. Because there is no “gold standard” for the measurement of time in various PA intensities, the criterion validity for the accelerometer measurements is impossible to establish
[30, 31]. Although PA measurements by accelerometry are found to be moderately correlated with measurements made using doubly labeled water
, this technique is not suitable to measure minutes of PA at different work rates, as doubly labeled water only measure the total energy expenditure over a given time period
[30, 31]. Therefore, it is very difficult to perform a valid comparison of precision between the GCP and the ICPs. However, we established the relative validity using the short-form of the International Physical Activity Questionnaire (IPAQ)
 (results not shown). The minutes spent in bouts of MVPA/day were moderately correlated with the IPAQ for both the GCP (Spearman’s ρ = 0.64, p = .001) and the ICPs (ρ = 0.51, p = .014) (n = 23). Because neither measure showed to be superior in comparison with the other, the finding does not encourage the use of one measure over another. However, it is very well known that objective and subjective PA outcomes are only moderately associated and that large variation are found between studies (mean correlation r = 0.37 (minimum r = −0.71 and maximum r = 0.98) based on 148 studies)
. Therefore, we believe our approach to the evaluation of applying ICPs should be viewed as a meaningful way to answer the research question asked.
Second, our results may not be valid in a normal-weight population. As observed in the laboratory, the attachment of the accelerometers can be more challenging for severely obese subjects than for normal-weight subjects, and tilting of the instrument is known to reduce the level of counts
. In addition, musculoskeletal disorders and other factors that might interfere with walking capacity and work economy is much more common in severely obese subjects, compared to less obese and normal-weight subjects
. These effects may have produced greater variability in this population than would have been found in other populations. Thus, further research should verify or falsify our findings in a sample of less obese subjects.
Third, some issues regarding the performance of the calibration protocol and calculation of the ICPs deserve a comment. First, although we started the calibration procedure at a low speed to account for the high metabolic cost of walking in this group of severely obese subjects, 14 subjects spend more than three METs at two km/h. The extrapolation of the accelerometer counts to three METs may have caused some uncertainty in the count thresholds in these subjects. However, most subjects spent close to three METs at two km/h (n = 7 < 3.20; n = 11 < 3.50 METs). In addition, despite a quadratic fit between counts and metabolic cost was indicated in some individuals, we used linear models to derive ICPs to avoid overfitting of the models based on only five observations. This could have caused some uncertainty. However, applying ICPs derived from quadratic models did not change any findings. Second, we calibrated the accelerometers using a treadmill protocol, while PA was measured in a field setting. Although it doesn’t seem to be any agreement on which setting that causes the highest cut points (treadmill vs. track)
[36, 37], this may have caused variability on an individual level. Third, the participants started a lifestyle treatment program for their obesity in the period between performing the field measurement and the calibration protocol (i.e. one month prior to performing the calibration). This delay could have added variability to the results, as changes in physical fitness, resting metabolic rate or work efficiency could influence the relationship between accelerometer counts and work rate. Forth, the scaling of work rate to each individual subject’s resting metabolic rate to obtain individual MET-values, may have introduced a systematic bias relating to body size, as body composition is the most important determinant of the resting metabolic rate
. However, reanalysis of the data using a standardized MET value (3.5 ml/kg/min) as the reference for the calculation of work rate did not change any main findings. Fifth, as minutes of MVPA could be expected to increase with increased wear time, wear time could influence the findings. However, minutes of MVPA and percentage time in MVPA were very highly correlated (r = 0.98) and the use of percentage time in MVPA did not change any findings.