Journal of Physical Activity and Health, 2015, 12, 701 -707 http://dx.doi.org/10.1123/jpah.2013-0212 © 2015 Human Kinetics, Inc.
ORIGINAL RESEARCH
Quantification of Underestimation of Physical Activity During Cycling to School When Using Accelerometry Jakob Tarp, Lars B. Andersen, and Lars Østergaard Background: Cycling to and from school is an important source of physical activity (PA) in youth but it is not captured by the dominant objective method to quantify PA. The aim of this study was to quantify the underestimation of objectively assessed PA caused by cycling when using accelerometry. Methods: Participants were 20 children aged 11 to 14 years from a randomized controlled trial performed in 2011. Physical activity was assessed by accelerometry with the addition of heart rate monitoring during cycling to school. Global positioning system (GPS) was used to identify periods of cycling to school. Results: Mean minutes of moderate-to-vigorous physical activity (MVPA) during round-trip commutes was 10.8 (95% CI: 7.1–16.6). Each kilometer of cycling meant an underestimation of 9314 (95% CI: 7719–11238) counts and 2.7 (95% CI: 2.1–3.5) minutes of MVPA. Adjusting for cycling to school increased estimates of MVPA/day by 6.0 (95% CI: 3.8–9.6) minutes. Conclusions: Cycling to and from school contribute substantially to levels of MVPA and to mean counts/min in children. This was not collected by accelerometers. Using distance to school in conjunction with self-reported cycling to school may be a simple tool to improve the methodology. Keywords: active commuting, children, methodology
A positive association between active transportation to and from school (ATS) and PA levels is well-documented.1 Recent studies confirm this association and highlight the importance of distance from home to school.2–4 This is important as youth PA levels are independently associated with cardiovascular risk factors and the metabolic syndrome.5,6 Furthermore both PA and cardiovascular risk show a tendency of tracking into adulthood7,8 where these are highly associated with type II diabetes and cardiovascular disease.9 Children cycling to school have been reported as being less physically active (as assessed by accelerometry) than those walking to school2,10,11 but more active than passive commuters.2,10 Differences, however, are not significant. This is counterintuitive as transportation to school by cycling, but not walking, is associated with higher cardiorespiratory fitness12,13 and a better cardiovascular risk factor profile14 in youth. Two factors are likely to be explanatory in this regard. A low number of cyclists in some populations1 make it difficult to compare PA levels because of a lack of statistical power. Secondly, as accelerometry is the primary method of objective measurement of PA in children15 and these instruments are known for their inability to measure PA during cycling when positioned at the hip, results could be caused by a systematic bias in PA measurements. In recent years cycling has received attention as a health enhancing behavior16 and the US and the UK have launched initiatives to increase rates of cycling.17,18 This increased focus on cycling means that methods to allow for correction of accelerometer assessments of PA levels of cycling children are warranted. Accelerometry currently offers distinct advantages such as feasibility and ability to differentiate between PA intensity domains but attempts to improve the methodology have been made by combining multiple activity sensors using branched equations and by applying artificial neural networks such as the hidden Markov model.15 Tarp (
[email protected]), Andersen, and Østergaard are with the Institute of Sports Science and Clinical Biomechanics, University of Southern Denmark, Odense, Denmark. Andersen is also with the Dept of Sports Medicine, Norwegian School of Sport Sciences, Oslo, Norway.
Two studies have combined GPS and accelerometry during ATS in walking children19,20 allowing precise estimates of PA during ATS but to date no studies have reported directly measured PA during cycling to school. Employing such an approach, with the addition of a supportive measure of intensity during cycling, PA levels during cycling to school can be assessed. In addition, underestimation during cycling, when using accelerometry, can be quantified and described as the impact on assessments of PA levels is unknown despite the well-known limitation and widespread use of the methodology. As distance is strongly related to the PA performed by active commuters and distance can be easily assessed using geographic information system (GIS) or GPS21 this may offer a simple way of improving assessments of PA when using accelerometry. The aims of this paper are to investigate the extent of underestimation of PA when using accelerometry in children cycling to and from school and quantify implications for assessments of PA
Methods Study Presentation This study uses data from a randomized controlled trial designed to investigate the effect of a change from passive transportation to school (ie, car, bus or train) to active transportation to school by cycling (ATSC) in 4th- to 6th-grade children on cardiorespiratory fitness (CRF) and the metabolic syndrome. The study took place in the municipality of Odense, Denmark, during spring 2011. Only data from the intervention group is used in this article. Criteria for inclusion were owning a bike and not participating in cycling to school for the last 3 months. The intervention group were asked to begin daily cycling to and from school during the 8 weeks the intervention lasted. The intervention group consisted of 23 participants of which 20 had data available for this study. All testing was performed at the University of Southern Denmark. Written informed consent was obtained from participants’ parent or legal guardian. The study was approved by the regional Ethics Committee. For full information on recruitment and procedures see Østergaard et al.22 701
702 Tarp, Andersen, and Østergaard
Anthropometry Before initiation of the intervention anthropometric measurements were taken after a small breakfast on site. Height was measured by a stadiometer (SECA, Hamburg, Germany), participants wearing socks or bare feet. Weight was assessed by a 0.1 kg precision scale (Soehnle Professional, Murrhardt, Germany) wearing only shorts and t-shirts.
V˙O2-HR relationship CRF (V˙ O2peak) was assessed using an incremental step test on an electronically braked cycle ergometer (Monark 839 Ergomedic, Varberg, Sweden). Resistance was increased by 40 W every 2 minutes after a 5 min. warm-up period at 40 W. The test was terminated at volitional fatigue despite verbal encouragement. Cadence was self determined but recommended between 60–80 rpm. Pulmonary gas exchange was measured with a metabolic cart (Amis2001, Innovision, Odense, Denmark) with epoch set at 15 seconds. The equipment was calibrated between each test and was validated against the Douglas bag method before both measurement periods. The test was deemed maximal if participants fulfilled one of the following criteria in addition to being exhausted as judged subjectively by the test leader: heart rate (HR) ≥ 185 beats/min or respiratory exchange ratio ≥ 0.99.23 V˙ O2peak was determined as the mean V˙ O2 (ml O2·kg-1.min-1) of the highest measurements during 30 seconds. HR was measured with a heart rate monitor (Polar RS800CX, Kempele, Finland) with epoch set to 1 sec. Maximal HR (HRmax) was defined as the highest value observed. The incremental test was additionally used to create individual V˙ O2-HR relationships. Individual equations were generated by linear regression of synchronized V˙ O2 and HR data points representing means of 15 seconds. An in-depth description of the establishing of these relationships is included (see Online Supplemental Material).
Assessment of Physical Activity Participants were instructed to report daily mode of school transportation in a custom made transportation diary, which served as registration of compliance. PA levels beyond cycling were assessed using accelerometry (Actigraph GT3X, Fort Walton Beach, FL, USA) during 1 week of the intervention (week 5 of the intervention). These assessments are presented as PA levels (counts/min) and as minutes of MVPA/day. Participants were instructed to wear the device at the right hip for 7 consecutive days and received both verbal and written instructions on how to wear the device. Epoch was set to 2 seconds but data were extracted in 1 min epochs. A minimum of 3 days with minimum 10 hours of recording were required to be included in the analysis. Periods lasting longer than 30 minutes of continuous measurements of 0 counts were discarded as nonwear. Data were extracted using customized software (Propero, Odense, Denmark). Only data from the vertical axis was included. Intensity cut-points for MVPA were based on the age-specific Freedson/Trost equations24 which uses 4 metabolic equivalents (METs) as the threshold for MVPA. During accelerometer assessments of PA participants received a GPS logger (Qstarz BT-Q1300S, Qstarz International Inc., Taiwan) and a heart rate monitor (PolarTeam2, Polar, Kempele, Finland). Participants were instructed to wear the devices during journeys to and from school on 1 school day. Both devices were set at 5 sec epoch. A letter thoroughly explained each step in setting up and how to wear the instruments. All GPS data were transferred to commercial software (Travel recorder v. 5.0) and corrected for drift using manufacturer software. The GPS was the reference for all data from heart rate monitors. Mean HR of school transportation, if verified by the transportation-diary, was calculated as the mean
of the measurements from the first data point when the participants exceeded a speed of 5 km/h when leaving home, to the last data point above 5 km/h when arriving at school. This period was defined as the journey and was the basis of distance and journey minutes. Propero was used to time-match accelerometer data with the GPS and HR data in 1-minute epochs. This was done to compare mean intensity (represented by counts/min) and minutes of MVPA collected by accelerometers during ATSC with that measured by HR during the same period and thus quantifying the extent of underestimation by accelerometers. Accelerometer results during ATSC are presented as counts/min (ACCcount) and minutes of MVPA (ACCMVPA). The mean HR during ATSC was converted to counts/min (HRcount) using the Freedson/Trost equations24 and the V˙ O2-HR relationships. METs were calculated from mean HR during each minute of ATSC using the individual V˙ O2-HR relationships. Thus minutes of MVPA measured by HR (HRMVPA) was determined as each minute of ATSC with a mean HR exceeding 4 METs. METs where found by dividing the V˙ O2 of an activity with the resting oxygen uptake (RMR) calculated using the age-specific Schofield equations including height and weight.25 These equations have been validated against indirect calorimetry and used similarly elsewhere.26 HRMVPA was the reference for minutes of MVPA during ATSC. The underestimation of accelerometer assessed PA during per cycled kilometer to school was found by dividing the total underestimation of counts and minutes of MVPA during journeys by the total distance. Where underestimation was available at only 1 journey (n = 11) this value was doubled. In case of GPS data available for only 1 journey (n = 2) time and distance at the other journey were assumed to be identical to the journey with data. Accelerometer assessments of PA are presented in 2 ways: 1) as conventional estimates (ie, not adjusted for ATSC) and 2) as accelerometry + underestimated minutes of MVPA and counts/min during ATSC (adjusted MVPA/ day and adjusted PA levels). The adjusted values were calculated by multiplying the underestimated minutes of MVPA and counts with the proportion of days the participant cycled to school during assessments of PA (using the transportation diary as reference) and adding this value to the conventional estimate.
Statistics Assumptions of normality were checked by examining residuals within a qq-plot and by using Shapiro Wilk’s test. Most data from accelerometers were positively skewed and were transformed using the natural logarithm. All skewed data justified assumptions of normality after transformation. Parametric tests were used on transformed data and then presented as exponentiated geometric means with 95% CI (95% CI ratios to the geometrics means). The logarithmic transformations makes data more robust to outliers (ie, means and confidence intervals) by using the same data, but at a different scale (as is done with standard deviations). Transformed values are however not easily interpreted and are thus exponentiated (antilogged) so values can be reported at the original scale. This antilogged value is a geometric mean. Results are shown as mean and standard deviations (SD) and geometric mean with 95% confidence intervals (95% CI) if assumptions of normality where justified in untransformed or log-transformed data, respectively. Observations presented as geometric means are forced from 0 to 1 to retain information.27 If assumptions were justified paired t-test was performed to examine whether differences differed significantly from 0. If assumptions were not justified the nonparametric Wilcoxon signedrank test was used. Associations were investigated using Spearman’s rho with confidence intervals (95%) derived after Fischer’s transformation. Effect sizes were calculated using mean difference with
JPAH Vol. 12, No. 5, 2015
Underestimation of Physical Activity During Cycling 703
the unadjusted SD as shared variance as suggested by Dunlop et al28 when dealing with paired designs to avoid overestimating the effect size due to reduced variance in “treatment” groups. We calculated effect sizes on log transformed data. A 2-tailed P-value of 0.05 or less was considered statistically significant. Propero results were exported from Excel to STATA IC V.11.0 (STATA Corp, College Station, Texas, USA) where analyses were performed.
Results Of the 23 participants in the intervention group 3 were excluded. Reasons for excluding were not completing the CRF test (n = 1), invalid GPS measurements (n = 1) and did not attend school during the intervention (n = 1). One participant did not achieve 3 days of valid PA measurements and was not included in analysis of assessments of PA. This report thus includes 20 participants. Baseline characteristics for these participants are shown in Table 1. Accelerometers were worn for a mean (SD) of 6.2 (1.3) days. Mean (SD) wear time was 13.5 (0.8) hours.
Active Transportation to and From School Table 2 shows descriptive results obtained during ATSC. Eighteen participants had GPS data at both journeys. One participant had GPS Table 1 Baseline Characteristics Cycling group
Boys (n = 13) Girls (n = 7) Total (n = 20)
Age (years)
12.2 (0.9)
11.8 (0.5)
12.0 (0.8)
Height (cm)
152.9 (8.5)
152.3 (5.3)
152.7 (7.4)
Weight (kg)
44.5 (8.5)
41.9 (6.8)
43.6 (7.8)
43.9 (7.1)
36.7 (2.5)
41.4 (6.8)
HRmax (bpm)
192.8 (6.9)
195.1 (8.3)
193.6 (7.3)
RMR (ml O2·kg-1.min-1)
4.76 (0.42)
4.37 (0.43)
4.62 (0.46)
CRF (ml
O2·kg-1.min-1)
Note. Baseline characteristics for the 20 participants with available results for this study. Results are shown by gender and total as mean (SD). Abbreviations: CRF, cardiorespiratory fitness; HRmax, maximal heart rate; RMR, resting oxygen uptake.
Table 2 Descriptive Results During Active Transportation to and From School by Cycling
Intensity (% of HRmax)
METs
HRcount (counts/min)
ACCcount (counts/min)
HRMVPA (min)
ACCMVPA (min)
Journey minutes (min)
Distance (km)
To school
From school
72.3 7.6 N = 18 5.0 0.6 N = 18 3249 701 N = 18 966 577 N = 19 5.7 3.9–8.2 N = 18 1.2 0.9–1.4 N = 18 7.2 5.4–9.6 N = 19 1.63 1.17–2.28 N = 19
68.5 7.2 N = 11 4.7 0.7 N = 11 2903 659 N = 11 1088 640 N = 19 5.9 3.3–10.8 N = 11 1.2 0.9–1.6 N = 11 8.0 6.0–10.7 N = 19 1.78 1.32–2.39 N = 19
Mean difference between journeys (95% CI) 5.0 (1.9–8.1) N=9 0.6 (0.3–0.9) N=9 600 (238–962) N=9 –90 (–406 to 225) N = 18 –0.67 (–3.04 to 1.70) N=9
P-value for difference between journeys
