Pediatric Exercise Science, 2015, 27, 21-25 http://dx.doi.org/10.1123/pes.2015-0030 © 2015 Human Kinetics, Inc.
Physical Activity, Inactivity, and Health Alex V. Rowlands University of South Australia
Citation Wolff-Hughes DL, Bassett DR, Fitzhugh EC. Population-Referenced Percentiles for Waist-Worn Accelerometer-Derived Total Activity Counts in U.S. Youth: 2003 – 2006 NHANES. PLoS ONE. 2014; 9(12):e115915. PubMed doi:10.1371/journal.pone.0115915
Background: The total activity volume performed is an overall measure that takes into account the frequency, intensity, and duration of activities performed. The importance of considering total activity volume is shown by recent studies indicating that light physical activity (LPA) and intermittent moderate-to-vigorous physical activity (MVPA) have health benefits. Accelerometer-derived total activity counts (TAC) per day from a waist-worn accelerometer can serve as a proxy for an individual’s total activity volume. The purpose of this study was to develop age- and gender-specific percentiles for daily TAC, minutes of MVPA, and minutes of LPA in U.S. youth ages 6–19 y. Methods: Data from the 2003–2006 NHANES waist-worn accelerometer component were used in this analysis. The sample was composed of youth aged 6–19 years with at least 4 d of ≥10 hr of accelerometer wear time (N = 3698). MVPA was defined using age specific cutpoints as the total number of minutes at ≥4 metabolic equivalents (METs) for youth 6–17 y or minutes with ≥2020 counts for youth 18–19 y. LPA was defined as the total number of minutes between 100 counts and the MVPA threshold. TAC/d, MVPA, and LPA were averaged across all valid days. Results: For males in the 50th percentile, the median activity level was 441,431 TAC/d, with 53 min/d of MVPA and 368 min/d of LPA. The median level of activity for females was 234,322 TAC/d, with 32 min/d of MVPA and 355 min/d of LPA. Conclusion: Population referenced TAC/d percentiles for U.S. youth ages 6–19 y provide a novel means of characterizing the total activity volume performed by children and adolescents.
Commentary Many years ago, I was conversing with Professor Tom Rowland (such occasions are always entertaining but guaranteed to keep you on your toes), when he passed comment that population referenced age- and sex-specific percentiles for physical activity, as have proved valuable for BMI, fitness, height and weight, would be extremely beneficial to clinicians and researchers alike. Fast forward, I don’t know, perhaps a decade and a half, and that is exactly what Wolff-Hughes and colleagues have produced in this paper. In the past few years personal physical activity monitoring has become more and more prevalent in the general population in both adults (e.g., Fitbit, Jawbone UP, Misfit Shine) and children (e.g., Zamzee). While the output from these monitors is not equivalent to the ActiGraph data presented in this paper, the prevalence of personal The author is with the Alliance for Research in Exercise, Nutrition and Activity (ARENA), Sansom Institute for Health Research, University of South Australia, Adelaide, Australia. Address author correspondence to Alex Rowlands at alex. [email protected].
monitoring may mean people would be amenable to and/or interested in wearing monitors if requested to do so, e.g., by a family doctor and/or as part of a regular checkup. In the future, as well as having height and mass checked, children could perhaps be issued with an ActiGraph for a week of monitoring (or another activity monitor for which population data were available). (We will leave potential reactivity to one side at the moment). Use of the population-referenced percentiles published in this paper would allow a child’s centile of activity to be recorded alongside their BMI centile and perhaps fitness centile. Repeated measures during growth would also allow the identification of centile-crossing in one or more measures (although it is noted that the current dataset is cross-sectional), perhaps acting as an early risk indicator and informing decisions on possible implementation of interventions/preventive measures. Significantly, percentile curves are presented for TAC/d as well as for MVPA and LPA. The importance of this measure is outlined very clearly in a key paper also published this year by the same research group (1). In brief: TAC/d incorporates the full continuum of activity intensities; it condenses the frequency, intensity and duration of activity bouts into a single metric; it is close to the parameter measured by 21
Downloaded by Heriot Watt Univ on 09/26/16, Volume 27, Article Number 1
22 Rowlands
the accelerometer (acceleration) and does not rely on assumptions to derive variables, e.g., energy expenditure or time spent at different activity intensities; and finally, but crucially, it is a standardized PA output that can be presented in all studies using the waist-worn ActiGraph. These curves are based on U.S. children (the data are taken from the 2003–2006 National Health and Nutrition Examination Survey (NHANES)), but the curves could presumably also be generated from the International Children’s Accelerometry Database (ICAD) that includes 32,000 young people aged 3–18 years across studies from Europe, the U.S., Brazil and Australia. (http:// www.mrc-epid.cam.ac.uk/research/studies/icad/). The waist-worn ActiGraph was used in all studies in the ICAD database so equivalent population-referenced percentiles for TAC/d, MVPA and LPA could be generated. Where sufficient numbers are available, it would be possible to generate country-specific as well as age- and sex-specific percentiles. As the authors of this paper indicate, the days of these particular percentile curves may be numbered, as there is a move from the waist/hip as the preferred wear-site to the wrist. NHANES changed the accelerometer wear-site to the wrist in the 2008–2011 data collection. In addition, although the ActiGraph is the most widely used accelerometer, other accelerometers are in common use: The Actical is used in the Canadian Health Measures Survey (2) and the GENEActiv, a relatively recent addition to the activity monitor market, is fast gaining in popularity. However, research suggests that, when using an outcome as close as possible to what the accelerometer actually measures, e.g., TAC/d or acceleration (in g or m/s2), this may not be a big problem. Straker and Campbell (7) and Paul et al. (4) have published translation equations for conversion between Actical and ActiGraph minute-byminute counts and average daily counts, respectively. Accelerations measured by the ActiGraph and the GENEActiv, although slightly different in magnitude, are also linearly related (3,5). Further, van Hees and colleagues (8,9) have carried out a substantial amount of work on the generation of transparent standardized output measures from the GENEActiv using the open source software package, R. So moving forward, output based on volume of activity should be translatable within and between monitor brands. A greater problem may be wear-site. As the authors state, the magnitude of accelerometer output is wear-site dependent. They suggest that wear-site specific centiles could be generated or conversions developed between output measured at different wear-sites. For example, new centiles could be generated from the 2008–2011 NHANES data for the wrist-worn ActiGraph. In relation to the potential for conversion, research from our laboratory shows the summary measures of activity (6) are linearly related (r = .86, p < .001). Contemporaneous minute-by-minute accelerations measured at the wrist and the hip also track well (mean r = .89, 25th-75th %ile =
0.86–0.92, p < .001, unpublished data from our laboratory), falling out of synch relatively infrequently and usually only for short periods. These results indicate it may be possible to convert between accelerations measured at the hip and wrist. I see the production of these population-referenced percentiles for accelerometer outcomes as a key step forward in the interpretation of children’s physical activity levels, the tracking of physical activity over time and potentially the identification of children crossing centiles; this is very much akin to the way that anthropometric measures are interpreted. Further, there is now a very large quantity of children’s accelerometer data available that could be used to build on this and contribute to population-referenced percentiles applicable beyond the U.S.
References 1. Bassett BR, Troiano RP, McClain JJ, Wolff DL. Accelerometer-based physical activity: Total volume per day and standardised measures. Med Sci Sports Exer; 2014, 10.1249/MSS.0000000000000468. 2. Colley RC, Garriguet D, Janssen I, Craig CL, Clarke J, Tremblay MS. Physical activity of Canadian children and youth: Accelerometer results from the 2007 to 2009 Canadian Health Measures Survey Statistics Canada, Catalogue no. 82-003-XPE. Health Rep. 2011; 22(1):15–23. PubMed 3. John D, Sasaki J, Staudenmayer J, Mavilia M, Freedson PS. Comparison of Raw Acceleration from the GENEA and ActiGraphTM GT3X+ Activity Monitors. Sensors (Basel Switzerland). 2013; 13(11):14754–14763. PubMed doi:10.3390/s131114754 4. Paul DR, Kramer M, Moshfegh AJ, Baer DJ, Rumpler WV. Comparison of two different physical activity monitors. BMC Med Res Methodol. 2007; 25(7):26. PubMed doi:10.1186/1471-2288-7-26 5. Rowlands AV, Fraysse F, Catt M, et al. Comparison of measured acceleration output from accelerometry-based activity monitors. Med Sci Sports Exerc. 2014; 46: 2308–2316. 6. Rowlands AV, Rennie K, Kozarski R, et al. Children’s physical activity assessed with wrist- and hip-worn accelerometers. Med Sci Sports Exerc. 2014; 2014. PubMed 7. Straker L, Campbell A. Translation equations to compare ActiGraph GT3X and Actical accelerometers activity counts. BMC Med Res Methodol. 2012; 12:54. PubMed doi:10.1186/1471-2288-12-54 8. van Hees VT, Fang Z, Langford J, et al. Autocalibration of accelerometer data for free-living physical activity assessment using local gravity and temperature: an evaluation on four continents. J Appl Physiol. 2014; 117(7):738–744. PubMed doi:10.1152/japplphysiol.00421.2014 9. van Hees VT, Gorzelniak L, Dean León EC, et al. Separating Movement and Gravity Components in an Acceleration Signal and Implications for the Assessment of Human Daily Physical Activity. PLoS ONE. 2013; 8(4):e61691. PubMed doi:10.1371/journal.pone.0061691
Physical Activity and Health 23
Citation
Downloaded by Heriot Watt Univ on 09/26/16, Volume 27, Article Number 1
Montoye AHK, Lanay M, Biswar S, Pfeiffer KA. Energy expenditure prediction using raw accelerometer data in simulated freeliving. Med Sci Sports Exerc. 2014; Epub ahead of print. PubMed doi: 10.1249/MSS.0000000000000597
Purpose: The purpose of this study was to develop, validate, and compare energy expenditure prediction models for accelerometers placed on the hip, thigh, and wrists using simple accelerometer features as input variables in energy expenditure prediction models. Methods: Forty-four healthy adults participated in a 90-min semistructured, simulated free-living activity protocol. During the protocol, participants engaged in a total of 14 different sedentary, ambulatory, lifestyle, and exercise activities for 3–10 min each. Participants chose the order, duration, and intensity of activities. Four accelerometers were worn (right hip, right thigh, and right and left wrists) to predict energy expenditure compared with that measured by the criterion measure (portable metabolic analyzer). Artificial neural networks (ANNs) were created to predict energy expenditure from each accelerometer using a leave-one-out cross-validation approach. Accuracy of the ANNs was evaluated using Pearson correlations, root mean square error, and bias. Several ANNs were developed using different input features to determine those most relevant for use in the models. Results: The ANNs for all 4 accelerometers achieved high measurement accuracy, with correlations of r > .80 for predicting energy expenditure. The thigh accelerometer provided the highest overall accuracy (r = .90) and lowest root mean square error (1.04 METs), and the differences between the thigh and the other monitors was more pronounced when fewer input variables were used in the predictive models. None of the predictive models had an overall bias for prediction of energy expenditure. Conclusion: A single accelerometer placed on the thigh provided the highest accuracy for energy expenditure prediction, although monitors worn on the wrists or hip can also be used with high measurement accuracy.
Commentary People watching advertisements for personal activity trackers, smartwatches and other personal monitoring devices could be forgiven for thinking that it is now straightforward to automatically monitor and classify all human activities with a high degree of accuracy using one body-worn monitor bought from Amazon or in the Apple shop and connected to an app on their smartphone. However, while many of these devices appear to do a pretty good job of tracking total activity and time spent at (unspecified) intensities (4), classifying type of activity from free-living data remains a challenge in both children and adults. This paper, focusing on adults, represents a very different approach from the single standardized metric for accelerometer measured physical activity (TAC/d) used in the previous paper, the importance of which was outlined by Bassett and colleagues (1). I believe both approaches are equally important to our understanding and interpretation of accelerometer data going forward. For example, there is a large body of literature indicating the importance of volume of activity for health; therefore a clear, standardized metric for volume of activity, which can be universally reported in studies using accelerometry to measure activity would be very beneficial. However, as the relationship between accelerometer output and energy expenditure is not linear, classification of type of activity is also important to obtain more accurate and precise estimates of energy expenditure (8). Further, classification of types of behaviors may inform promotion of physical activity and planning of interventions. Several studies have shown pretty accurate classification of activity type using data collected in laboratory
protocols, but when applied to free-living data, typically the classification rate is much poorer. To address this, the use of free-living data to generate the algorithms that classify types of activities is advised. However, this introduces problems. First, there is the need for a criterion measure, a not inconsiderable challenge with free-living data—picture people going about their normal daily activities trying to ignore the mask on their face and gas analyzer on their back (to measure energy expenditure) and the person observing their every move (to record activity type). At a stretch, this might be feasible, until you consider that for the majority of most people’s day they are sedentary; thus, to cover a good range of activities a very long observation period is likely to be needed if relying on free-living data. Long periods with gas analyzers and observation become prohibitively invasive. (Note: More acceptable methods are available, e.g., the IDEEA (Intelligent Device for Energy Expenditure and Activity (9)); alternatively a wearable camera (the SenseCam (2)) or observation can be used as a criterion for activity type, although this would require energy expenditure to be predicted from compendiums (7)). For these reasons, as in this paper, laboratory protocols that simulate a number of free-living activities in a condensed period of time are used, sometimes with some sort of weighting to account for the prevalence of given activities in daily life (8). A neat thing about this study was the degree of flexibility participants had within the protocol. For example, the participants picked the order of their activities and the duration of time, the way and the rate at which they did them. This is a far cry from the way we have done calibration studies in the past with, e.g., metronomes used to standardize the frequency of ‘unregulated’, ‘play’ or ‘lifestyle’ activities.
Downloaded by Heriot Watt Univ on 09/26/16, Volume 27, Article Number 1
24 Rowlands
This study compared different sets of features from the acceleration signal and multiple wear locations simultaneously. All three of the main wear-sites were used (hip, wrist and thigh) with both wrists assessed. The authors built on earlier activity classification work (6) to generate sets of features from the acceleration signal for developing classification algorithms that had the least complexity necessary without compromising accuracy. The goal of the authors was to make the methods as accessible to nonexperts as possible. Perhaps not surprisingly the thigh monitor came out on top for classification accuracy, however, the researchers reported problems securing the monitor to the thigh. As the authors state, this is likely solvable—after all there has been high compliance to the activPAL thigh monitor. The accuracy of the wrist site for activity classification was not far behind that of the hip and thigh and far closer than when using standard linear regression analytic approaches. This supports the findings of others and is encouraging as compliance to wrist-wear has been shown to be very high (3), is optimal for monitoring sleep and has greater potential for classification of type of sedentary behavior than other wear-sites (5). Of note, there was no difference in the accuracy of classification using data from the left and right wrist. It will be interesting to see if this remains the case in true free-living when activities that use a dominant arm may be more prevalent. The big question: How well will the algorithms developed transfer to true free-living better data and will the conclusions on wear-site and degree of algorithm complexity necessary hold up? As the authors state, the big problem with classifying activity type is it all gets much harder when applied in a truly free living setting. If the algorithms transfer well to free-living data then that is a significant step forward. This is what the authors plan to test… watch this space… !
References 1. Bassett BR, Troiano RP, McClain JJ, Wolff DL. Accelerometer-based physical activity: Total volume per day and standardised measures. Med Sci Sports Exerc. 2014; Epub ahead of print. 2. Doherty AR, Kelly P, Kerr J, et al. Using wearable cameras to categorise type and context of accelerometer-identified episodes of physical activity. Int J Behav Nutr Phys Act. 2013; 10:22. PubMed doi:10.1186/1479-5868-10-22 3. Freedson PS, John D. Comment on ‘‘Estimating Activity and Sedentary Behavior from an Accelerometer on the Hip and Wrist’’. Med Sci Sports Exerc. 2013; 45:962–963. PubMed doi:10.1249/MSS.0b013e31827f024d 4. Lee JM, Kim Y, Welk GJ. Validity of consumerbased physical activity monitors. Med Sci Sports Exerc. 2014; 46(9):1840–1848. PubMed doi:10.1249/ MSS.0000000000000287 5. Rowlands AV, Olds TS, Hillsdon M, et al. Assessing sedentary behaviour with the GENEActiv: Introducing the Sedentary Sphere. Med Sci Sports Exerc. 2014; 46:1235–1247. PubMed doi:10.1249/MSS.0000000000000224 6. Staudenmayer J, Pober D, Crouter S, Bassett D, Freedson P. An artificial neural network to estimate physical activity energy expenditure and identify physical activity type from an accelerometer. J Appl Physiol. 2009; 107(4):1300– 1307. PubMed doi:10.1152/japplphysiol.00465.2009 7. van Hees V, Ekelund U. Novel daily energy expenditure estimation by using objective activity type classification: where do we go from here? J Appl Physiol. 2009; 107(3):639–640. PubMed doi:10.1152/japplphysiol.00793.2009 8. van Hees VT, Golubic R, Ekelund U, Brage S. Impact of study design on development and evaluation of an activitytype classifier. J Appl Physiol. 2013; 114(8):1042–1051. PubMed doi:10.1152/japplphysiol.00984.2012 9. Zhang K, Werner P, Sun M, Pi-Sunyer FX, Boozer CN. Measurement of human daily physical activity. Obes Res. 2003; 11(1):33–40. PubMed doi:10.1038/oby.2003.7
Citation Goodman A, Page AS, Cooper AR. Daylight saving time as a potential public health intervention: an observational study of evening daylight and objectively-measured physical activity among 23,000 children from 9 countries. Int J Behav Nutr Phys Act. 2014; 11(1):84. PubMed doi:10.1186/1479-5868-11-84
Background: It has been proposed that introducing daylight saving measures could increase children’s physical activity, but there exists little research on this issue. This study therefore examined associations between time of sunset and activity levels, including using the biannual ‘changing of the clocks’ as a natural experiment. Methods: 23,188 children aged 5–16 years from 15 studies in nine countries were brought together in the International Children’s Accelerometry Database. 439 of these children were of particular interest for our analyses as they contributed data both immediately before and after the clocks changed. All children provided objectively measured physical activity data from Actigraph accelerometers, and we used their average physical activity level (accelerometer counts per minute) as our primary outcome. Date of accelerometer data collection was matched to time of sunset, and to weather characteristics including daily precipitation, humidity, wind speed and temperature. Results: Adjusting for child and weather covariates, we found that longer evening daylight was independently associated with a small increase in daily physical activity. Consistent with a causal interpretation, the magnitude of these associations was largest in the late afternoon and early evening and these associations were also evident when comparing the same child just before and just after the clocks changed.
Physical Activity and Health 25
These associations were, however, only consistently observed in the 5 mainland European, four English and two Australian samples (adjusted, pooled effect sizes 0.03–0.07 standard deviations per hour of additional evening daylight). In some settings there was some evidence of larger associations between day length and physical activity in boys. There was no evidence of interactions with weight status or maternal education, and inconsistent findings for interactions with age. Conclusions: In Europe and Australia, evening daylight seems to play a causal role in increasing children’s activity in a relatively equitable manner. Although the average increase in activity is small in absolute terms, these increases apply across all children in a population. Moreover, these small effect sizes actually compare relatively favorably with the typical effect of intensive, individual-level interventions. We therefore conclude that, by shifting the physical activity mean of the entire population, the introduction of additional daylight saving measures could yield worthwhile public health benefits.
Downloaded by Heriot Watt Univ on 09/26/16, Volume 27, Article Number 1
Commentary There have been many discussions on the possible benefits of additional daylight saving, i.e., increasing the number of waking daylight hours by moving the clocks forward, on health and the environment, but (as far as I am aware) relatively little presented in the way of actual evidence to support claims either side. This very elegant study changes that. This paper addresses a number of hypotheses, providing converging evidence supporting a causal relationship between daylight saving and physical activity in children. Significantly, the extremely large sample size meant that, for over 400 children, the measurement period spanned the period of time both before and after clocks were changed, enabling causality to be inferred between lighter nights and increased activity. These effects were evident in children from England, mainland Europe and Australia, although interestingly not in children living in America, Madeira, or Brazil. The data suggested that the higher temperatures in the latter countries might contribute to the lack of an effect. This did not appear to be the case in Australia, although perhaps helps explains the lower effect size in Australia relative to England and mainland Europe. Of course, the number of daylight hours cannot be manipulated so if daylight hours last later into the evening, then the mornings will be darker. Might this lead to lower activity levels in the morning? It seems not. The authors showed that the morning level of activity was virtually identical regardless of clock changes, whereas there was a clear increase in physical activity in afternoon/evening hours when sunset was later.
The authors report that the evidence suggests that, if fully causal, shifting the clocks forward by one additional hour year round would lead to a small average increase in moderate-to-vigorous physical activity (MVPA) of 1.7 min in English children, fractionally more in children in mainland Europe and slightly less in Australia. Notably average cpm (counts per minute: a measure of average activity intensity—total activity counts per day divided by wear-time) increased during the afternoon and evening hours too, suggesting that, as well as the small increase in MVPA, light activity increased and sedentary behavior decreased. As the effects of MVPA and sedentary behavior on health appear to be independent (1), this is important. Intuitively, the findings of this study make sense. We know that time spent outdoors is related to physical activity (4) and it makes sense that children are more likely to want to, and/or be allowed to, play outside if it is daylight. As noted by the authors, the effects of daylight saving on activity are small. However, this is a population health approach whereby the goal is to increase the health of populations rather than the individual (2). Shifting the population physical activity level, even by a relatively small amount, may have a greater impact on population health than changing the activity level of fewer people by a larger amount (3). Perhaps most encouragingly, from my naïve perspective, this appears to be a population initiative that
would be relatively straightforward and inexpensive to implement. A small shift in the average baseline activity level, applying to every child in the given country on every day, is a nudge in the right direction and, further, the availability of longer daylight hours could facilitate other intervention approaches. Kickabout anyone. . . ?
References 1. Healy GN, Wijndaele K, Dunstan DW, et al. Objectively measured sedentary time, physical activity, and metabolic risk: the Australian Diabetes, Obesity and Lifestyle Study (AusDiab). Diabetes Care. 2008; 31(2):369–371. PubMed doi:10.2337/dc07-1795 2. Kohl, 3rd HW, Craig CL, Lambert EV, et al. The pandemic of physical inactivity: global action for public health. Lancet. 2012; 380:294–305. PubMed doi:10.1016/S01406736(12)60898-8 3. Rose G. Sick individuals and sick populations. Int J Epidemiol. 1985; 14:32–38. PubMed doi:10.1093/ije/14.1.32 4. Schaefer L, Plotnikoff RC, Majumdar SR, et al. Outdoor time is associated with physical activity, sedentary time, and cardiorespiratory fitness in youth. J Pediatr. 2014; 165(3):516–521. PubMed doi:10.1016/j. jpeds.2014.05.029
Physical Activity, Inactivity, and Health Alex V. Rowlands University of South Australia
Citation Wolff-Hughes DL, Bassett DR, Fitzhugh EC. Population-Referenced Percentiles for Waist-Worn Accelerometer-Derived Total Activity Counts in U.S. Youth: 2003 – 2006 NHANES. PLoS ONE. 2014; 9(12):e115915. PubMed doi:10.1371/journal.pone.0115915
Background: The total activity volume performed is an overall measure that takes into account the frequency, intensity, and duration of activities performed. The importance of considering total activity volume is shown by recent studies indicating that light physical activity (LPA) and intermittent moderate-to-vigorous physical activity (MVPA) have health benefits. Accelerometer-derived total activity counts (TAC) per day from a waist-worn accelerometer can serve as a proxy for an individual’s total activity volume. The purpose of this study was to develop age- and gender-specific percentiles for daily TAC, minutes of MVPA, and minutes of LPA in U.S. youth ages 6–19 y. Methods: Data from the 2003–2006 NHANES waist-worn accelerometer component were used in this analysis. The sample was composed of youth aged 6–19 years with at least 4 d of ≥10 hr of accelerometer wear time (N = 3698). MVPA was defined using age specific cutpoints as the total number of minutes at ≥4 metabolic equivalents (METs) for youth 6–17 y or minutes with ≥2020 counts for youth 18–19 y. LPA was defined as the total number of minutes between 100 counts and the MVPA threshold. TAC/d, MVPA, and LPA were averaged across all valid days. Results: For males in the 50th percentile, the median activity level was 441,431 TAC/d, with 53 min/d of MVPA and 368 min/d of LPA. The median level of activity for females was 234,322 TAC/d, with 32 min/d of MVPA and 355 min/d of LPA. Conclusion: Population referenced TAC/d percentiles for U.S. youth ages 6–19 y provide a novel means of characterizing the total activity volume performed by children and adolescents.
Commentary Many years ago, I was conversing with Professor Tom Rowland (such occasions are always entertaining but guaranteed to keep you on your toes), when he passed comment that population referenced age- and sex-specific percentiles for physical activity, as have proved valuable for BMI, fitness, height and weight, would be extremely beneficial to clinicians and researchers alike. Fast forward, I don’t know, perhaps a decade and a half, and that is exactly what Wolff-Hughes and colleagues have produced in this paper. In the past few years personal physical activity monitoring has become more and more prevalent in the general population in both adults (e.g., Fitbit, Jawbone UP, Misfit Shine) and children (e.g., Zamzee). While the output from these monitors is not equivalent to the ActiGraph data presented in this paper, the prevalence of personal The author is with the Alliance for Research in Exercise, Nutrition and Activity (ARENA), Sansom Institute for Health Research, University of South Australia, Adelaide, Australia. Address author correspondence to Alex Rowlands at alex. [email protected].
monitoring may mean people would be amenable to and/or interested in wearing monitors if requested to do so, e.g., by a family doctor and/or as part of a regular checkup. In the future, as well as having height and mass checked, children could perhaps be issued with an ActiGraph for a week of monitoring (or another activity monitor for which population data were available). (We will leave potential reactivity to one side at the moment). Use of the population-referenced percentiles published in this paper would allow a child’s centile of activity to be recorded alongside their BMI centile and perhaps fitness centile. Repeated measures during growth would also allow the identification of centile-crossing in one or more measures (although it is noted that the current dataset is cross-sectional), perhaps acting as an early risk indicator and informing decisions on possible implementation of interventions/preventive measures. Significantly, percentile curves are presented for TAC/d as well as for MVPA and LPA. The importance of this measure is outlined very clearly in a key paper also published this year by the same research group (1). In brief: TAC/d incorporates the full continuum of activity intensities; it condenses the frequency, intensity and duration of activity bouts into a single metric; it is close to the parameter measured by 21
Downloaded by Heriot Watt Univ on 09/26/16, Volume 27, Article Number 1
22 Rowlands
the accelerometer (acceleration) and does not rely on assumptions to derive variables, e.g., energy expenditure or time spent at different activity intensities; and finally, but crucially, it is a standardized PA output that can be presented in all studies using the waist-worn ActiGraph. These curves are based on U.S. children (the data are taken from the 2003–2006 National Health and Nutrition Examination Survey (NHANES)), but the curves could presumably also be generated from the International Children’s Accelerometry Database (ICAD) that includes 32,000 young people aged 3–18 years across studies from Europe, the U.S., Brazil and Australia. (http:// www.mrc-epid.cam.ac.uk/research/studies/icad/). The waist-worn ActiGraph was used in all studies in the ICAD database so equivalent population-referenced percentiles for TAC/d, MVPA and LPA could be generated. Where sufficient numbers are available, it would be possible to generate country-specific as well as age- and sex-specific percentiles. As the authors of this paper indicate, the days of these particular percentile curves may be numbered, as there is a move from the waist/hip as the preferred wear-site to the wrist. NHANES changed the accelerometer wear-site to the wrist in the 2008–2011 data collection. In addition, although the ActiGraph is the most widely used accelerometer, other accelerometers are in common use: The Actical is used in the Canadian Health Measures Survey (2) and the GENEActiv, a relatively recent addition to the activity monitor market, is fast gaining in popularity. However, research suggests that, when using an outcome as close as possible to what the accelerometer actually measures, e.g., TAC/d or acceleration (in g or m/s2), this may not be a big problem. Straker and Campbell (7) and Paul et al. (4) have published translation equations for conversion between Actical and ActiGraph minute-byminute counts and average daily counts, respectively. Accelerations measured by the ActiGraph and the GENEActiv, although slightly different in magnitude, are also linearly related (3,5). Further, van Hees and colleagues (8,9) have carried out a substantial amount of work on the generation of transparent standardized output measures from the GENEActiv using the open source software package, R. So moving forward, output based on volume of activity should be translatable within and between monitor brands. A greater problem may be wear-site. As the authors state, the magnitude of accelerometer output is wear-site dependent. They suggest that wear-site specific centiles could be generated or conversions developed between output measured at different wear-sites. For example, new centiles could be generated from the 2008–2011 NHANES data for the wrist-worn ActiGraph. In relation to the potential for conversion, research from our laboratory shows the summary measures of activity (6) are linearly related (r = .86, p < .001). Contemporaneous minute-by-minute accelerations measured at the wrist and the hip also track well (mean r = .89, 25th-75th %ile =
0.86–0.92, p < .001, unpublished data from our laboratory), falling out of synch relatively infrequently and usually only for short periods. These results indicate it may be possible to convert between accelerations measured at the hip and wrist. I see the production of these population-referenced percentiles for accelerometer outcomes as a key step forward in the interpretation of children’s physical activity levels, the tracking of physical activity over time and potentially the identification of children crossing centiles; this is very much akin to the way that anthropometric measures are interpreted. Further, there is now a very large quantity of children’s accelerometer data available that could be used to build on this and contribute to population-referenced percentiles applicable beyond the U.S.
References 1. Bassett BR, Troiano RP, McClain JJ, Wolff DL. Accelerometer-based physical activity: Total volume per day and standardised measures. Med Sci Sports Exer; 2014, 10.1249/MSS.0000000000000468. 2. Colley RC, Garriguet D, Janssen I, Craig CL, Clarke J, Tremblay MS. Physical activity of Canadian children and youth: Accelerometer results from the 2007 to 2009 Canadian Health Measures Survey Statistics Canada, Catalogue no. 82-003-XPE. Health Rep. 2011; 22(1):15–23. PubMed 3. John D, Sasaki J, Staudenmayer J, Mavilia M, Freedson PS. Comparison of Raw Acceleration from the GENEA and ActiGraphTM GT3X+ Activity Monitors. Sensors (Basel Switzerland). 2013; 13(11):14754–14763. PubMed doi:10.3390/s131114754 4. Paul DR, Kramer M, Moshfegh AJ, Baer DJ, Rumpler WV. Comparison of two different physical activity monitors. BMC Med Res Methodol. 2007; 25(7):26. PubMed doi:10.1186/1471-2288-7-26 5. Rowlands AV, Fraysse F, Catt M, et al. Comparison of measured acceleration output from accelerometry-based activity monitors. Med Sci Sports Exerc. 2014; 46: 2308–2316. 6. Rowlands AV, Rennie K, Kozarski R, et al. Children’s physical activity assessed with wrist- and hip-worn accelerometers. Med Sci Sports Exerc. 2014; 2014. PubMed 7. Straker L, Campbell A. Translation equations to compare ActiGraph GT3X and Actical accelerometers activity counts. BMC Med Res Methodol. 2012; 12:54. PubMed doi:10.1186/1471-2288-12-54 8. van Hees VT, Fang Z, Langford J, et al. Autocalibration of accelerometer data for free-living physical activity assessment using local gravity and temperature: an evaluation on four continents. J Appl Physiol. 2014; 117(7):738–744. PubMed doi:10.1152/japplphysiol.00421.2014 9. van Hees VT, Gorzelniak L, Dean León EC, et al. Separating Movement and Gravity Components in an Acceleration Signal and Implications for the Assessment of Human Daily Physical Activity. PLoS ONE. 2013; 8(4):e61691. PubMed doi:10.1371/journal.pone.0061691
Physical Activity and Health 23
Citation
Downloaded by Heriot Watt Univ on 09/26/16, Volume 27, Article Number 1
Montoye AHK, Lanay M, Biswar S, Pfeiffer KA. Energy expenditure prediction using raw accelerometer data in simulated freeliving. Med Sci Sports Exerc. 2014; Epub ahead of print. PubMed doi: 10.1249/MSS.0000000000000597
Purpose: The purpose of this study was to develop, validate, and compare energy expenditure prediction models for accelerometers placed on the hip, thigh, and wrists using simple accelerometer features as input variables in energy expenditure prediction models. Methods: Forty-four healthy adults participated in a 90-min semistructured, simulated free-living activity protocol. During the protocol, participants engaged in a total of 14 different sedentary, ambulatory, lifestyle, and exercise activities for 3–10 min each. Participants chose the order, duration, and intensity of activities. Four accelerometers were worn (right hip, right thigh, and right and left wrists) to predict energy expenditure compared with that measured by the criterion measure (portable metabolic analyzer). Artificial neural networks (ANNs) were created to predict energy expenditure from each accelerometer using a leave-one-out cross-validation approach. Accuracy of the ANNs was evaluated using Pearson correlations, root mean square error, and bias. Several ANNs were developed using different input features to determine those most relevant for use in the models. Results: The ANNs for all 4 accelerometers achieved high measurement accuracy, with correlations of r > .80 for predicting energy expenditure. The thigh accelerometer provided the highest overall accuracy (r = .90) and lowest root mean square error (1.04 METs), and the differences between the thigh and the other monitors was more pronounced when fewer input variables were used in the predictive models. None of the predictive models had an overall bias for prediction of energy expenditure. Conclusion: A single accelerometer placed on the thigh provided the highest accuracy for energy expenditure prediction, although monitors worn on the wrists or hip can also be used with high measurement accuracy.
Commentary People watching advertisements for personal activity trackers, smartwatches and other personal monitoring devices could be forgiven for thinking that it is now straightforward to automatically monitor and classify all human activities with a high degree of accuracy using one body-worn monitor bought from Amazon or in the Apple shop and connected to an app on their smartphone. However, while many of these devices appear to do a pretty good job of tracking total activity and time spent at (unspecified) intensities (4), classifying type of activity from free-living data remains a challenge in both children and adults. This paper, focusing on adults, represents a very different approach from the single standardized metric for accelerometer measured physical activity (TAC/d) used in the previous paper, the importance of which was outlined by Bassett and colleagues (1). I believe both approaches are equally important to our understanding and interpretation of accelerometer data going forward. For example, there is a large body of literature indicating the importance of volume of activity for health; therefore a clear, standardized metric for volume of activity, which can be universally reported in studies using accelerometry to measure activity would be very beneficial. However, as the relationship between accelerometer output and energy expenditure is not linear, classification of type of activity is also important to obtain more accurate and precise estimates of energy expenditure (8). Further, classification of types of behaviors may inform promotion of physical activity and planning of interventions. Several studies have shown pretty accurate classification of activity type using data collected in laboratory
protocols, but when applied to free-living data, typically the classification rate is much poorer. To address this, the use of free-living data to generate the algorithms that classify types of activities is advised. However, this introduces problems. First, there is the need for a criterion measure, a not inconsiderable challenge with free-living data—picture people going about their normal daily activities trying to ignore the mask on their face and gas analyzer on their back (to measure energy expenditure) and the person observing their every move (to record activity type). At a stretch, this might be feasible, until you consider that for the majority of most people’s day they are sedentary; thus, to cover a good range of activities a very long observation period is likely to be needed if relying on free-living data. Long periods with gas analyzers and observation become prohibitively invasive. (Note: More acceptable methods are available, e.g., the IDEEA (Intelligent Device for Energy Expenditure and Activity (9)); alternatively a wearable camera (the SenseCam (2)) or observation can be used as a criterion for activity type, although this would require energy expenditure to be predicted from compendiums (7)). For these reasons, as in this paper, laboratory protocols that simulate a number of free-living activities in a condensed period of time are used, sometimes with some sort of weighting to account for the prevalence of given activities in daily life (8). A neat thing about this study was the degree of flexibility participants had within the protocol. For example, the participants picked the order of their activities and the duration of time, the way and the rate at which they did them. This is a far cry from the way we have done calibration studies in the past with, e.g., metronomes used to standardize the frequency of ‘unregulated’, ‘play’ or ‘lifestyle’ activities.
Downloaded by Heriot Watt Univ on 09/26/16, Volume 27, Article Number 1
24 Rowlands
This study compared different sets of features from the acceleration signal and multiple wear locations simultaneously. All three of the main wear-sites were used (hip, wrist and thigh) with both wrists assessed. The authors built on earlier activity classification work (6) to generate sets of features from the acceleration signal for developing classification algorithms that had the least complexity necessary without compromising accuracy. The goal of the authors was to make the methods as accessible to nonexperts as possible. Perhaps not surprisingly the thigh monitor came out on top for classification accuracy, however, the researchers reported problems securing the monitor to the thigh. As the authors state, this is likely solvable—after all there has been high compliance to the activPAL thigh monitor. The accuracy of the wrist site for activity classification was not far behind that of the hip and thigh and far closer than when using standard linear regression analytic approaches. This supports the findings of others and is encouraging as compliance to wrist-wear has been shown to be very high (3), is optimal for monitoring sleep and has greater potential for classification of type of sedentary behavior than other wear-sites (5). Of note, there was no difference in the accuracy of classification using data from the left and right wrist. It will be interesting to see if this remains the case in true free-living when activities that use a dominant arm may be more prevalent. The big question: How well will the algorithms developed transfer to true free-living better data and will the conclusions on wear-site and degree of algorithm complexity necessary hold up? As the authors state, the big problem with classifying activity type is it all gets much harder when applied in a truly free living setting. If the algorithms transfer well to free-living data then that is a significant step forward. This is what the authors plan to test… watch this space… !
References 1. Bassett BR, Troiano RP, McClain JJ, Wolff DL. Accelerometer-based physical activity: Total volume per day and standardised measures. Med Sci Sports Exerc. 2014; Epub ahead of print. 2. Doherty AR, Kelly P, Kerr J, et al. Using wearable cameras to categorise type and context of accelerometer-identified episodes of physical activity. Int J Behav Nutr Phys Act. 2013; 10:22. PubMed doi:10.1186/1479-5868-10-22 3. Freedson PS, John D. Comment on ‘‘Estimating Activity and Sedentary Behavior from an Accelerometer on the Hip and Wrist’’. Med Sci Sports Exerc. 2013; 45:962–963. PubMed doi:10.1249/MSS.0b013e31827f024d 4. Lee JM, Kim Y, Welk GJ. Validity of consumerbased physical activity monitors. Med Sci Sports Exerc. 2014; 46(9):1840–1848. PubMed doi:10.1249/ MSS.0000000000000287 5. Rowlands AV, Olds TS, Hillsdon M, et al. Assessing sedentary behaviour with the GENEActiv: Introducing the Sedentary Sphere. Med Sci Sports Exerc. 2014; 46:1235–1247. PubMed doi:10.1249/MSS.0000000000000224 6. Staudenmayer J, Pober D, Crouter S, Bassett D, Freedson P. An artificial neural network to estimate physical activity energy expenditure and identify physical activity type from an accelerometer. J Appl Physiol. 2009; 107(4):1300– 1307. PubMed doi:10.1152/japplphysiol.00465.2009 7. van Hees V, Ekelund U. Novel daily energy expenditure estimation by using objective activity type classification: where do we go from here? J Appl Physiol. 2009; 107(3):639–640. PubMed doi:10.1152/japplphysiol.00793.2009 8. van Hees VT, Golubic R, Ekelund U, Brage S. Impact of study design on development and evaluation of an activitytype classifier. J Appl Physiol. 2013; 114(8):1042–1051. PubMed doi:10.1152/japplphysiol.00984.2012 9. Zhang K, Werner P, Sun M, Pi-Sunyer FX, Boozer CN. Measurement of human daily physical activity. Obes Res. 2003; 11(1):33–40. PubMed doi:10.1038/oby.2003.7
Citation Goodman A, Page AS, Cooper AR. Daylight saving time as a potential public health intervention: an observational study of evening daylight and objectively-measured physical activity among 23,000 children from 9 countries. Int J Behav Nutr Phys Act. 2014; 11(1):84. PubMed doi:10.1186/1479-5868-11-84
Background: It has been proposed that introducing daylight saving measures could increase children’s physical activity, but there exists little research on this issue. This study therefore examined associations between time of sunset and activity levels, including using the biannual ‘changing of the clocks’ as a natural experiment. Methods: 23,188 children aged 5–16 years from 15 studies in nine countries were brought together in the International Children’s Accelerometry Database. 439 of these children were of particular interest for our analyses as they contributed data both immediately before and after the clocks changed. All children provided objectively measured physical activity data from Actigraph accelerometers, and we used their average physical activity level (accelerometer counts per minute) as our primary outcome. Date of accelerometer data collection was matched to time of sunset, and to weather characteristics including daily precipitation, humidity, wind speed and temperature. Results: Adjusting for child and weather covariates, we found that longer evening daylight was independently associated with a small increase in daily physical activity. Consistent with a causal interpretation, the magnitude of these associations was largest in the late afternoon and early evening and these associations were also evident when comparing the same child just before and just after the clocks changed.
Physical Activity and Health 25
These associations were, however, only consistently observed in the 5 mainland European, four English and two Australian samples (adjusted, pooled effect sizes 0.03–0.07 standard deviations per hour of additional evening daylight). In some settings there was some evidence of larger associations between day length and physical activity in boys. There was no evidence of interactions with weight status or maternal education, and inconsistent findings for interactions with age. Conclusions: In Europe and Australia, evening daylight seems to play a causal role in increasing children’s activity in a relatively equitable manner. Although the average increase in activity is small in absolute terms, these increases apply across all children in a population. Moreover, these small effect sizes actually compare relatively favorably with the typical effect of intensive, individual-level interventions. We therefore conclude that, by shifting the physical activity mean of the entire population, the introduction of additional daylight saving measures could yield worthwhile public health benefits.
Downloaded by Heriot Watt Univ on 09/26/16, Volume 27, Article Number 1
Commentary There have been many discussions on the possible benefits of additional daylight saving, i.e., increasing the number of waking daylight hours by moving the clocks forward, on health and the environment, but (as far as I am aware) relatively little presented in the way of actual evidence to support claims either side. This very elegant study changes that. This paper addresses a number of hypotheses, providing converging evidence supporting a causal relationship between daylight saving and physical activity in children. Significantly, the extremely large sample size meant that, for over 400 children, the measurement period spanned the period of time both before and after clocks were changed, enabling causality to be inferred between lighter nights and increased activity. These effects were evident in children from England, mainland Europe and Australia, although interestingly not in children living in America, Madeira, or Brazil. The data suggested that the higher temperatures in the latter countries might contribute to the lack of an effect. This did not appear to be the case in Australia, although perhaps helps explains the lower effect size in Australia relative to England and mainland Europe. Of course, the number of daylight hours cannot be manipulated so if daylight hours last later into the evening, then the mornings will be darker. Might this lead to lower activity levels in the morning? It seems not. The authors showed that the morning level of activity was virtually identical regardless of clock changes, whereas there was a clear increase in physical activity in afternoon/evening hours when sunset was later.
The authors report that the evidence suggests that, if fully causal, shifting the clocks forward by one additional hour year round would lead to a small average increase in moderate-to-vigorous physical activity (MVPA) of 1.7 min in English children, fractionally more in children in mainland Europe and slightly less in Australia. Notably average cpm (counts per minute: a measure of average activity intensity—total activity counts per day divided by wear-time) increased during the afternoon and evening hours too, suggesting that, as well as the small increase in MVPA, light activity increased and sedentary behavior decreased. As the effects of MVPA and sedentary behavior on health appear to be independent (1), this is important. Intuitively, the findings of this study make sense. We know that time spent outdoors is related to physical activity (4) and it makes sense that children are more likely to want to, and/or be allowed to, play outside if it is daylight. As noted by the authors, the effects of daylight saving on activity are small. However, this is a population health approach whereby the goal is to increase the health of populations rather than the individual (2). Shifting the population physical activity level, even by a relatively small amount, may have a greater impact on population health than changing the activity level of fewer people by a larger amount (3). Perhaps most encouragingly, from my naïve perspective, this appears to be a population initiative that
would be relatively straightforward and inexpensive to implement. A small shift in the average baseline activity level, applying to every child in the given country on every day, is a nudge in the right direction and, further, the availability of longer daylight hours could facilitate other intervention approaches. Kickabout anyone. . . ?
References 1. Healy GN, Wijndaele K, Dunstan DW, et al. Objectively measured sedentary time, physical activity, and metabolic risk: the Australian Diabetes, Obesity and Lifestyle Study (AusDiab). Diabetes Care. 2008; 31(2):369–371. PubMed doi:10.2337/dc07-1795 2. Kohl, 3rd HW, Craig CL, Lambert EV, et al. The pandemic of physical inactivity: global action for public health. Lancet. 2012; 380:294–305. PubMed doi:10.1016/S01406736(12)60898-8 3. Rose G. Sick individuals and sick populations. Int J Epidemiol. 1985; 14:32–38. PubMed doi:10.1093/ije/14.1.32 4. Schaefer L, Plotnikoff RC, Majumdar SR, et al. Outdoor time is associated with physical activity, sedentary time, and cardiorespiratory fitness in youth. J Pediatr. 2014; 165(3):516–521. PubMed doi:10.1016/j. jpeds.2014.05.029
