Home
Search
Collections
Journals
About
Contact us
My IOPscience
An investigation into the minimum accelerometry wear time for reliable estimates of habitual physical activity and definition of a standard measurement day in pre-school children.
This content has been downloaded from IOPscience. Please scroll down to see the full text. 2014 Physiol. Meas. 35 2213 (http://iopscience.iop.org/0967-3334/35/11/2213) View the table of contents for this issue, or go to the journal homepage for more Download details: IP Address: 128.226.136.66 This content was downloaded on 29/05/2017 at 17:19 Please note that terms and conditions apply.
You may also be interested in: Validity of wrist worn accelerometers and comparability between hip and wrist placement sites in estimating physical activity behaviour in preschool children Jane Hislop, Nicole Palmer, Priya Anand et al. Empirically derived cut-points for sedentary behaviour: are we sitting differently? Alexandra M Clarke-Cornwell, Tracey M Farragher, Penny A Cook et al. The minimum number of days required to establish reliable physical activity estimates in children aged 2–15years Minsoo Kang, Kristie Bjornson, Tiago V Barreira et al. Validity of an automated algorithm to identify waking and in-bed wear time in hip-worn accelerometer data collected with a 24h wear protocol in young adults Joanne A McVeigh, Elisabeth A H Winkler, Genevieve N Healy et al. Number of accelerometer monitoring days needed for stable group-level estimates of activity Dana L Wolff-Hughes, James J McClain, Kevin W Dodd et al. Investigating optimal accelerometer placement for energy expenditure prediction in children using a machine learning approach K A Mackintosh, A H K Montoye, K A Pfeiffer et al. Comparison of self-reported measure of sitting time (IPAQ) with objective measurement (activPAL) S F M Chastin, B Culhane and P M Dall Assessing physical activity intensity by video analysis P Silva, C Santiago, L P Reis et al.
Institute of Physics and Engineering in Medicine Physiol. Meas. 35 (2014) 2213–2228
Physiological Measurement doi:10.1088/0967-3334/35/11/2213
An investigation into the minimum accelerometry wear time for reliable estimates of habitual physical activity and definition of a standard measurement day in pre-school children. Jane Hislop1, James Law2, Robert Rush1, Andrew Grainger1, Cathy Bulley1, John J Reilly3 and Tom Mercer1 1
School of Health Sciences, Queen Margaret University Edinburgh, Scotland, UK School of Education, Communication and Language Sciences, Newcastle University, Newcastle, UK 3 Physical Activity and Public Health Science Physical Activity for Health Group, School of Psychological Sciences and Health, University of Strathclyde, Glasgow, Scotland, UK 2
E-mail: [email protected] Received 6 March 2014, revised 14 July 2014 Accepted for publication 29 July 2014 Published 23 October 2014 Abstract
The purpose of this study was to determine the number of hours and days of accelerometry data necessary to provide a reliable estimate of habitual physical activity in pre-school children. The impact of a weekend day on reliability estimates was also determined and standard measurement days were defined for weekend and weekdays. Accelerometry data were collected from 112 children (60 males, 52 females, mean (SD) 3.7 (0.7)yr) over 7 d. The Spearman-Brown Prophecy formula (S-B prophecy formula) was used to predict the number of days and hours of data required to achieve an intraclass correlation coefficient (ICC) of 0.7. The impact of including a weekend day was evaluated by comparing the reliability coefficient (r) for any 4 d of data with data for 4 d including one weekend day. Our observations indicate that 3 d of accelerometry monitoring, regardless of whether it includes a weekend day, for at least 7 h d−1 offers sufficient reliability to characterise total physical activity and sedentary behaviour of pre-school children. These findings offer an approach that addresses the underlying tension in epidemiologic surveillance studies between the need to maintain acceptable measurement rigour and retention of a representatively meaningful sample size. 0967-3334/14/112213+16$33.00 © 2014 Institute of Physics and Engineering in Medicine Printed in the UK
2213
J Hislop et al
Physiol. Meas. 35 (2014) 2213
Keywords: physical activity, pre-school, measurement, reliability, accelerometry (Some figures may appear in colour only in the online journal) 1. Introduction Accelerometry has been used widely as an objective means of measuring physical activity and sedentary behaviour in pre-school children (Pate et al 2010). Despite this there are several outstanding questions on approaches to reduce accelerometry data, whereby accelerometry output is transformed into a meaningful format (Cliff et al 2009). In population-based studies, two of the key decisions in data reduction are to determine how many hours of data constitute a ‘valid’ day, and how many days will provide a reliable estimate of habitual physical activity and time spent in sedentary behaviour (Ojiambo et al 2011). Both have an impact on whether to subsequently exclude participant data with insufficient days or hours of data, and practical implications in terms of the burden on researchers and the study participants. In excluding incomplete participant data, there is a risk of potential inaccuracies, that can result from differences between days that are excluded and days kept in the analysis. Therefore, the decision of whether to exclude incomplete participant data or include data for part of a day can produce biased estimates of physical activity. For example, it is possible that more active children have a greater number of valid days of data than less active children, and exclusion of data from less active children in any sample would result in a biased estimate of a population’s physical activity levels. The decision on whether to include or exclude data could explain the apparent differences in physical activity output between studies (Alhassan et al 2008). In addition, the decision on excluding or including data will have implications for the resulting sample size and hence could impact on the power of associated statistical analysis. To date, there is limited evidence regarding the number of days and hours within days, required to provide a reliable estimate of physical activity in the pre-school population (Hinkley et al 2012). There have been two studies with pre-school children which have explored how many days and hours of data were necessary to provide a reliable estimate of habitual physical activity. Penpraze et al (2006) conducted reliability analysis on 7 d of accelerometry data collected from 76 pre-school children (mean (SD) age: 5.6 (0.4) yr). The authors argued that while 10 h of data per day over 7 d maximised the reliability of estimates of total physical activity (r = 0.80), 3 d of data had sufficient reliability (r ≥ 0.60). They also suggested that the increase in reliability coefficients in hourly increments between 3 and 10 h of data were small (approximately r = 0.02), when 1–7 d of monitoring were used. The authors based their analysis on counts per minute (cpm) data collected in 1 min epochs. This approach has limitations, as more recent evidence suggests that shorter epochs (15 s or less) may be more accurate at capturing the intermittent physical activity behaviour typical of young children (McClain et al 2008, Hislop et al 2012b). In addition, cpm data incorporates all intensities of activity, from sedentary behaviour, light intensity and moderate-to-vigorous intensity activity. As a result it is difficult to determine the reliability of habitual levels of physical activity or sedentary behaviour between days using cpm data. In a more recent study by Hinkley et al (2012), reliability analysis was conducted on data collected from 1004 pre-school children (aged 3 to 5 years), over an 8 d period. The authors concluded that the number of days of data required to achieve a desired intraclass correlation coefficient (ICC) estimate of 0.7 increases as the number of hours of data per day decreases. For example, in their study, 10 h of data per day required 3 d of data collection, while collection of 2214
J Hislop et al
Physiol. Meas. 35 (2014) 2213
7 h of data per day meant that 4 d were required to achieve the same ICC estimate (Hinkley et al 2012). While the Hinkley et al (2012) study included a large sample and based their analysis on percentage time spent in total physical activity from data collected in 15 s epochs, the study was carried out in Australia and there may be cultural, social, and climatic factors influencing the physical activity behaviour of pre-school children. It is therefore important to investigate wear times for different pre-school populations to see if similar wear time recommendations apply. The commonly adopted approach to determine how many hours constitute a valid day and how many days reflect an individual’s usual or ‘habitual’ physical activity, is to first ascertain the intra-individual variability of activity between days (Baranowski and de Moor 2000). ICC calculations can be used to estimate the consistency of activity across days (Baranowski and de Moor 2000). If an individual replicates the same activity pattern every day, then one day of data would be sufficient. However, as variability increases across days an increasing number of days are necessary to reflect habitual activity. Similarly, the number of hours to reflect a ‘typical’ day also contributes to the variability in data between days. Another important consideration in determining habitual activity is whether it is necessary to only include participant data which includes a weekend day, or, conversely, exclude participant data that do not include a weekend day. Both the studies by Hinkley et al (2012) and Penpraze et al (2006) investigated the implications of including a weekend day in data analysis. Penpraze et al (2006) reported significant differences in the mean counts per minute (cpm) between weekdays and weekend days. However, the inclusion of a weekend day had minimal impact on the reliability of the data and the authors suggest that, in their sample, the inclusion of a weekend day in analysis was not necessary. A recent study has reported similar findings of minimal differences in reliability between weekend and weekdays in accelerometry data collected from 7704 seven-year-old children (Rich et al 2013). In contrast, Hinkley et al (2012) reported that total physical activity differed significantly between weekdays and weekend days. Recommending that future studies ensure both weekend and weekdays are included to accurately represent pre-school children’s physical activity across an entire week. Finally, an alternative approach to deciding the minimum wear-time is to define a standard measurement day, whereby at least 70% of the sample are wearing accelerometers for 80% of the time (Catellier et al 2005). While this approach has been implemented in studies of preschool aged children (O’Dwyer et al 2012) and toddlers (Hnatiuk et al 2012), the value of this approach has not been fully described in a pre-school population. In summary, only two studies to date have sought to investigate minimal wear time and whether having a criteria for the inclusion of a weekend day in the analysis with pre-school children is necessary and conflicting findings have been reported. As a result the evidence for the number of hours and days of data to reliably reflect habitual physical activity in the pre-school population is limited and further investigation is warranted. Given that ICC values are constrained to the sample from which they are calculated (Baranowski and de Moor 2000), and that the magnitude of the intra- and inter-individual variances in physical activity are sample-specific (Ridley et al 2009), the current study therefore aims to gather empirical evidence to understand whether or not day to day variation in physical activity and sedentary behaviour are population specific. The aims of this study were: • To determine the recommended wear time required to provide reliable estimates of habitual physical activity and sedentary behaviour of pre-school children. • To establish whether the inclusion of a weekend day is a necessary pre-requisite for reliable estimates of habitual physical activity and sedentary behaviour of pre-school children. • To define a standard measurement day in pre-school children. 2215
J Hislop et al
Physiol. Meas. 35 (2014) 2213
2. Methods This study involved a secondary data analysis of 121 accelerometry data files collected from children aged 42–48 months. The analysis is based on baseline accelerometry data collected at the start (pre-intervention) of a longitudinal intervention study with pre-school children. Ethical approval for the study was obtained from the host institution’s ethics committee. Baseline data were collected between November 2009 and July 2010. Recruitment was undertaken from two demographically similar geographical areas within Scotland. One area was used for recruitment of an intervention group and the other for recruitment of a control group. To determine sample size for this intervention study, it was estimated that 110 participants (55, per group) were necessary to detect changes of 100 cpm d−1 for 80% power. Children were recruited from nurseries in the two geographical areas. To ensure recruitment of participants from a range of socioeconomic backgrounds, postcodes of the nurseries were used to allocate a Scottish Index of Deprivation (SIMD) quintile (with 1 being most deprived and 5 least deprived quintile (Scottish Government 2010)). In addition, children were selected that had postcodes which matched closely with the nursery to try to ensure there was balance across the quintiles. For the purposes of the current reliability study, Hopkins et al (2000) states that at least 50 participants performing three or more trials provides adequate precision in estimating typical error. Data in this analysis had been collected using GT1M (ActiGraph, Walton Beach, FL) and GT3X accelerometers set to uniaxial mode, which were distributed to parents of pre-school children attending nurseries in the identified areas. Good agreement between the GT1M and GT3X models has been reported in studies of adolescents and school aged children (Kaminsky and Ozemek 2012, Robusto and Trost 2012, Vanhelst et al 2012). All accelerometers were set to start recording data in 15 s epochs from 6 am of the first day they were distributed to the parents at their child’s nursery. Parents were asked to attach the accelerometers, which were on an elasticated belt, around their children’s waists during waking hours for seven consecutive days and to remove the accelerometers for any water-based activities (bathing, swimming). Data were transferred to Excel and processed using the MAHUffe software (MRC Epidemiology Unit 2013). Each child had their weight measured using analogue scales to the nearest 0.1 kg and their height measured using a portable stadiometer to the nearest 0.1 cm. Each child’s BMI’s were then calculated as the body mass in kg divided by the square of the height (kg m−2). Good agreement between the GT1M and GT3X models has been reported in studies of adolescents and school aged children (Kaminsky and Ozemek 2012, Robusto and Trost 2012, Vanhelst et al 2012) but to our knowledge no such comparison exists in pre-school children. In addition, the BMI scores were expressed relative to the UK 1990 population reference data (Cole et al 1995) and the proportion of children classified as being ‘healthy weight’, ‘overweight’ and ‘obese’ was determined using the international cut off points for BMI and obesity by sex and age (Cole 2002). Finally the percentiles were reported for males and females (Centres for Disease Control and Prevention 2014). The initial step in data reduction was to clean the 121 accelerometry data files. Each file represented a participant’s accelerometry data collected over 7 d. The first consideration was to identify spurious data (data with very high data counts). The MAHUffe programme presents an individual’s accelerometry data graphically as well as allowing for minute-by-minute inspection of the data. It was therefore possible to undertake minute-by-minute visual inspection of the accelerometry data output files and spurious data were considered to be when the cpm was > 20 000 for a 1 min epoch. This threshold was identified in an earlier study by Colley et al (2010) above which activity is argued to be not biologically possible. In the current study, visual inspection of data revealed that six participants had cpm data which was 2216
J Hislop et al
Physiol. Meas. 35 (2014) 2213
Visual inspection of minute-
No accelerometry data
by-minute accelerometer data
recorded (n=2)
(n=121)
Identification of spurious data (n=119)
Exclusion of data with 20,000 cpm (n= 6 )
Logical sleep wake times identified (n=113) Exclusion of participants with >18 hours of data per day (n=1) Sample included in analysis (n=112)
Figure 1. Stages of data cleaning.
consistently > 20 000 and a particular error pattern was seen in the data with all of the hourly mean cpm for these data files being > 50 000 cpm. These data files, including all data from the six participants were excluded from the final analysis. It is possible that this may relate to an error code indicating monitor malfunction which has been reported in earlier studies of the older generations of ActiGraph (7164 model) as representing voltage saturation (Alhassan et al 2008). To date, this error pattern found within accelerometry data files has not been reported in any other studies using the GT1M or GT3X accelerometers. Visual inspection also revealed two participants who had no accelerometry data (either as a result of device malfunction or due to an error of the technician) and one participant who had worn the accelerometer for over 18 h d−1. This participant’s data were excluded as this did not correspond to logical sleep/wake time (Alhassan et al 2008). This left a total of 112 participants. Figure 1 presents the stages of data cleaning. 3. Data analysis To allow comparison of weekdays with weekend days, the days of the week were identified and separated for analysis. The start and finish wear times for each participant (n = 112) were identified and a frequency plot of all participant wear time data undertaken. Table 1 presents the characteristics of the sample, with 78% classified as ‘healthy’ weight and 22% classified overweight or obese i.e. BMI at or above 85% centile relative to the UK population reference data (Cole 2002). Data with ≥ 20 min of consecutive zeros were identified as non-wear time and excluded from analysis, as recommended by Esliger et al (2005). A ‘standard measurement day’ was identified as the length of time when at least 70% of the sample wore the accelerometer for 80% of the time (Catellier et al 2005).
2217
J Hislop et al
Physiol. Meas. 35 (2014) 2213
Table 1. Characteristics of sample.
Mean (SD)
No. participants Age (years) Height (cm) Body weight (kg) Body mass index (kg m−2) Mean BMI percentile
All
Male
Females
112 3.7 (0.7) 101.3 (3.9) 17.0 (2.3) 16.6 (1.6) —
60 3.7 (0.7) 101.6 (3.7) 17.5 (2.3) 17.0 (1.7) 85th
52 3.7 (0.7) 101.0 (4.1) 16.5 (2.1) 16.2 (1.3) 71st
3.1. Comparison of weekend and weekdays
Data from all 112 participants (60 males, 52 females, mean (SD) age: 3.7 (0.7) yr) were included in the analysis of weekend and weekdays. Using all participant data the mean cpm for participants’ weekdays of data and the mean cpm for their weekend days were extracted to allow comparison between weekend and weekdays. Data were evaluated for normality using the Kolmogorov-Smirnov test. As the data were not normally distributed, descriptive statistics of the medians and inter-quartile ranges (IQR) were calculated, and presented graphically in box plots. The Wilcoxon signed rank test was used to compare differences between cpm for weekend and weekdays. In addition, a comparison was made between the cpm for males and females at weekend and weekdays using a Mann-Whitney U test, the non-parametric equivalent of the independent t-test. The level of significance was set at p
Search
Collections
Journals
About
Contact us
My IOPscience
An investigation into the minimum accelerometry wear time for reliable estimates of habitual physical activity and definition of a standard measurement day in pre-school children.
This content has been downloaded from IOPscience. Please scroll down to see the full text. 2014 Physiol. Meas. 35 2213 (http://iopscience.iop.org/0967-3334/35/11/2213) View the table of contents for this issue, or go to the journal homepage for more Download details: IP Address: 128.226.136.66 This content was downloaded on 29/05/2017 at 17:19 Please note that terms and conditions apply.
You may also be interested in: Validity of wrist worn accelerometers and comparability between hip and wrist placement sites in estimating physical activity behaviour in preschool children Jane Hislop, Nicole Palmer, Priya Anand et al. Empirically derived cut-points for sedentary behaviour: are we sitting differently? Alexandra M Clarke-Cornwell, Tracey M Farragher, Penny A Cook et al. The minimum number of days required to establish reliable physical activity estimates in children aged 2–15years Minsoo Kang, Kristie Bjornson, Tiago V Barreira et al. Validity of an automated algorithm to identify waking and in-bed wear time in hip-worn accelerometer data collected with a 24h wear protocol in young adults Joanne A McVeigh, Elisabeth A H Winkler, Genevieve N Healy et al. Number of accelerometer monitoring days needed for stable group-level estimates of activity Dana L Wolff-Hughes, James J McClain, Kevin W Dodd et al. Investigating optimal accelerometer placement for energy expenditure prediction in children using a machine learning approach K A Mackintosh, A H K Montoye, K A Pfeiffer et al. Comparison of self-reported measure of sitting time (IPAQ) with objective measurement (activPAL) S F M Chastin, B Culhane and P M Dall Assessing physical activity intensity by video analysis P Silva, C Santiago, L P Reis et al.
Institute of Physics and Engineering in Medicine Physiol. Meas. 35 (2014) 2213–2228
Physiological Measurement doi:10.1088/0967-3334/35/11/2213
An investigation into the minimum accelerometry wear time for reliable estimates of habitual physical activity and definition of a standard measurement day in pre-school children. Jane Hislop1, James Law2, Robert Rush1, Andrew Grainger1, Cathy Bulley1, John J Reilly3 and Tom Mercer1 1
School of Health Sciences, Queen Margaret University Edinburgh, Scotland, UK School of Education, Communication and Language Sciences, Newcastle University, Newcastle, UK 3 Physical Activity and Public Health Science Physical Activity for Health Group, School of Psychological Sciences and Health, University of Strathclyde, Glasgow, Scotland, UK 2
E-mail: [email protected] Received 6 March 2014, revised 14 July 2014 Accepted for publication 29 July 2014 Published 23 October 2014 Abstract
The purpose of this study was to determine the number of hours and days of accelerometry data necessary to provide a reliable estimate of habitual physical activity in pre-school children. The impact of a weekend day on reliability estimates was also determined and standard measurement days were defined for weekend and weekdays. Accelerometry data were collected from 112 children (60 males, 52 females, mean (SD) 3.7 (0.7)yr) over 7 d. The Spearman-Brown Prophecy formula (S-B prophecy formula) was used to predict the number of days and hours of data required to achieve an intraclass correlation coefficient (ICC) of 0.7. The impact of including a weekend day was evaluated by comparing the reliability coefficient (r) for any 4 d of data with data for 4 d including one weekend day. Our observations indicate that 3 d of accelerometry monitoring, regardless of whether it includes a weekend day, for at least 7 h d−1 offers sufficient reliability to characterise total physical activity and sedentary behaviour of pre-school children. These findings offer an approach that addresses the underlying tension in epidemiologic surveillance studies between the need to maintain acceptable measurement rigour and retention of a representatively meaningful sample size. 0967-3334/14/112213+16$33.00 © 2014 Institute of Physics and Engineering in Medicine Printed in the UK
2213
J Hislop et al
Physiol. Meas. 35 (2014) 2213
Keywords: physical activity, pre-school, measurement, reliability, accelerometry (Some figures may appear in colour only in the online journal) 1. Introduction Accelerometry has been used widely as an objective means of measuring physical activity and sedentary behaviour in pre-school children (Pate et al 2010). Despite this there are several outstanding questions on approaches to reduce accelerometry data, whereby accelerometry output is transformed into a meaningful format (Cliff et al 2009). In population-based studies, two of the key decisions in data reduction are to determine how many hours of data constitute a ‘valid’ day, and how many days will provide a reliable estimate of habitual physical activity and time spent in sedentary behaviour (Ojiambo et al 2011). Both have an impact on whether to subsequently exclude participant data with insufficient days or hours of data, and practical implications in terms of the burden on researchers and the study participants. In excluding incomplete participant data, there is a risk of potential inaccuracies, that can result from differences between days that are excluded and days kept in the analysis. Therefore, the decision of whether to exclude incomplete participant data or include data for part of a day can produce biased estimates of physical activity. For example, it is possible that more active children have a greater number of valid days of data than less active children, and exclusion of data from less active children in any sample would result in a biased estimate of a population’s physical activity levels. The decision on whether to include or exclude data could explain the apparent differences in physical activity output between studies (Alhassan et al 2008). In addition, the decision on excluding or including data will have implications for the resulting sample size and hence could impact on the power of associated statistical analysis. To date, there is limited evidence regarding the number of days and hours within days, required to provide a reliable estimate of physical activity in the pre-school population (Hinkley et al 2012). There have been two studies with pre-school children which have explored how many days and hours of data were necessary to provide a reliable estimate of habitual physical activity. Penpraze et al (2006) conducted reliability analysis on 7 d of accelerometry data collected from 76 pre-school children (mean (SD) age: 5.6 (0.4) yr). The authors argued that while 10 h of data per day over 7 d maximised the reliability of estimates of total physical activity (r = 0.80), 3 d of data had sufficient reliability (r ≥ 0.60). They also suggested that the increase in reliability coefficients in hourly increments between 3 and 10 h of data were small (approximately r = 0.02), when 1–7 d of monitoring were used. The authors based their analysis on counts per minute (cpm) data collected in 1 min epochs. This approach has limitations, as more recent evidence suggests that shorter epochs (15 s or less) may be more accurate at capturing the intermittent physical activity behaviour typical of young children (McClain et al 2008, Hislop et al 2012b). In addition, cpm data incorporates all intensities of activity, from sedentary behaviour, light intensity and moderate-to-vigorous intensity activity. As a result it is difficult to determine the reliability of habitual levels of physical activity or sedentary behaviour between days using cpm data. In a more recent study by Hinkley et al (2012), reliability analysis was conducted on data collected from 1004 pre-school children (aged 3 to 5 years), over an 8 d period. The authors concluded that the number of days of data required to achieve a desired intraclass correlation coefficient (ICC) estimate of 0.7 increases as the number of hours of data per day decreases. For example, in their study, 10 h of data per day required 3 d of data collection, while collection of 2214
J Hislop et al
Physiol. Meas. 35 (2014) 2213
7 h of data per day meant that 4 d were required to achieve the same ICC estimate (Hinkley et al 2012). While the Hinkley et al (2012) study included a large sample and based their analysis on percentage time spent in total physical activity from data collected in 15 s epochs, the study was carried out in Australia and there may be cultural, social, and climatic factors influencing the physical activity behaviour of pre-school children. It is therefore important to investigate wear times for different pre-school populations to see if similar wear time recommendations apply. The commonly adopted approach to determine how many hours constitute a valid day and how many days reflect an individual’s usual or ‘habitual’ physical activity, is to first ascertain the intra-individual variability of activity between days (Baranowski and de Moor 2000). ICC calculations can be used to estimate the consistency of activity across days (Baranowski and de Moor 2000). If an individual replicates the same activity pattern every day, then one day of data would be sufficient. However, as variability increases across days an increasing number of days are necessary to reflect habitual activity. Similarly, the number of hours to reflect a ‘typical’ day also contributes to the variability in data between days. Another important consideration in determining habitual activity is whether it is necessary to only include participant data which includes a weekend day, or, conversely, exclude participant data that do not include a weekend day. Both the studies by Hinkley et al (2012) and Penpraze et al (2006) investigated the implications of including a weekend day in data analysis. Penpraze et al (2006) reported significant differences in the mean counts per minute (cpm) between weekdays and weekend days. However, the inclusion of a weekend day had minimal impact on the reliability of the data and the authors suggest that, in their sample, the inclusion of a weekend day in analysis was not necessary. A recent study has reported similar findings of minimal differences in reliability between weekend and weekdays in accelerometry data collected from 7704 seven-year-old children (Rich et al 2013). In contrast, Hinkley et al (2012) reported that total physical activity differed significantly between weekdays and weekend days. Recommending that future studies ensure both weekend and weekdays are included to accurately represent pre-school children’s physical activity across an entire week. Finally, an alternative approach to deciding the minimum wear-time is to define a standard measurement day, whereby at least 70% of the sample are wearing accelerometers for 80% of the time (Catellier et al 2005). While this approach has been implemented in studies of preschool aged children (O’Dwyer et al 2012) and toddlers (Hnatiuk et al 2012), the value of this approach has not been fully described in a pre-school population. In summary, only two studies to date have sought to investigate minimal wear time and whether having a criteria for the inclusion of a weekend day in the analysis with pre-school children is necessary and conflicting findings have been reported. As a result the evidence for the number of hours and days of data to reliably reflect habitual physical activity in the pre-school population is limited and further investigation is warranted. Given that ICC values are constrained to the sample from which they are calculated (Baranowski and de Moor 2000), and that the magnitude of the intra- and inter-individual variances in physical activity are sample-specific (Ridley et al 2009), the current study therefore aims to gather empirical evidence to understand whether or not day to day variation in physical activity and sedentary behaviour are population specific. The aims of this study were: • To determine the recommended wear time required to provide reliable estimates of habitual physical activity and sedentary behaviour of pre-school children. • To establish whether the inclusion of a weekend day is a necessary pre-requisite for reliable estimates of habitual physical activity and sedentary behaviour of pre-school children. • To define a standard measurement day in pre-school children. 2215
J Hislop et al
Physiol. Meas. 35 (2014) 2213
2. Methods This study involved a secondary data analysis of 121 accelerometry data files collected from children aged 42–48 months. The analysis is based on baseline accelerometry data collected at the start (pre-intervention) of a longitudinal intervention study with pre-school children. Ethical approval for the study was obtained from the host institution’s ethics committee. Baseline data were collected between November 2009 and July 2010. Recruitment was undertaken from two demographically similar geographical areas within Scotland. One area was used for recruitment of an intervention group and the other for recruitment of a control group. To determine sample size for this intervention study, it was estimated that 110 participants (55, per group) were necessary to detect changes of 100 cpm d−1 for 80% power. Children were recruited from nurseries in the two geographical areas. To ensure recruitment of participants from a range of socioeconomic backgrounds, postcodes of the nurseries were used to allocate a Scottish Index of Deprivation (SIMD) quintile (with 1 being most deprived and 5 least deprived quintile (Scottish Government 2010)). In addition, children were selected that had postcodes which matched closely with the nursery to try to ensure there was balance across the quintiles. For the purposes of the current reliability study, Hopkins et al (2000) states that at least 50 participants performing three or more trials provides adequate precision in estimating typical error. Data in this analysis had been collected using GT1M (ActiGraph, Walton Beach, FL) and GT3X accelerometers set to uniaxial mode, which were distributed to parents of pre-school children attending nurseries in the identified areas. Good agreement between the GT1M and GT3X models has been reported in studies of adolescents and school aged children (Kaminsky and Ozemek 2012, Robusto and Trost 2012, Vanhelst et al 2012). All accelerometers were set to start recording data in 15 s epochs from 6 am of the first day they were distributed to the parents at their child’s nursery. Parents were asked to attach the accelerometers, which were on an elasticated belt, around their children’s waists during waking hours for seven consecutive days and to remove the accelerometers for any water-based activities (bathing, swimming). Data were transferred to Excel and processed using the MAHUffe software (MRC Epidemiology Unit 2013). Each child had their weight measured using analogue scales to the nearest 0.1 kg and their height measured using a portable stadiometer to the nearest 0.1 cm. Each child’s BMI’s were then calculated as the body mass in kg divided by the square of the height (kg m−2). Good agreement between the GT1M and GT3X models has been reported in studies of adolescents and school aged children (Kaminsky and Ozemek 2012, Robusto and Trost 2012, Vanhelst et al 2012) but to our knowledge no such comparison exists in pre-school children. In addition, the BMI scores were expressed relative to the UK 1990 population reference data (Cole et al 1995) and the proportion of children classified as being ‘healthy weight’, ‘overweight’ and ‘obese’ was determined using the international cut off points for BMI and obesity by sex and age (Cole 2002). Finally the percentiles were reported for males and females (Centres for Disease Control and Prevention 2014). The initial step in data reduction was to clean the 121 accelerometry data files. Each file represented a participant’s accelerometry data collected over 7 d. The first consideration was to identify spurious data (data with very high data counts). The MAHUffe programme presents an individual’s accelerometry data graphically as well as allowing for minute-by-minute inspection of the data. It was therefore possible to undertake minute-by-minute visual inspection of the accelerometry data output files and spurious data were considered to be when the cpm was > 20 000 for a 1 min epoch. This threshold was identified in an earlier study by Colley et al (2010) above which activity is argued to be not biologically possible. In the current study, visual inspection of data revealed that six participants had cpm data which was 2216
J Hislop et al
Physiol. Meas. 35 (2014) 2213
Visual inspection of minute-
No accelerometry data
by-minute accelerometer data
recorded (n=2)
(n=121)
Identification of spurious data (n=119)
Exclusion of data with 20,000 cpm (n= 6 )
Logical sleep wake times identified (n=113) Exclusion of participants with >18 hours of data per day (n=1) Sample included in analysis (n=112)
Figure 1. Stages of data cleaning.
consistently > 20 000 and a particular error pattern was seen in the data with all of the hourly mean cpm for these data files being > 50 000 cpm. These data files, including all data from the six participants were excluded from the final analysis. It is possible that this may relate to an error code indicating monitor malfunction which has been reported in earlier studies of the older generations of ActiGraph (7164 model) as representing voltage saturation (Alhassan et al 2008). To date, this error pattern found within accelerometry data files has not been reported in any other studies using the GT1M or GT3X accelerometers. Visual inspection also revealed two participants who had no accelerometry data (either as a result of device malfunction or due to an error of the technician) and one participant who had worn the accelerometer for over 18 h d−1. This participant’s data were excluded as this did not correspond to logical sleep/wake time (Alhassan et al 2008). This left a total of 112 participants. Figure 1 presents the stages of data cleaning. 3. Data analysis To allow comparison of weekdays with weekend days, the days of the week were identified and separated for analysis. The start and finish wear times for each participant (n = 112) were identified and a frequency plot of all participant wear time data undertaken. Table 1 presents the characteristics of the sample, with 78% classified as ‘healthy’ weight and 22% classified overweight or obese i.e. BMI at or above 85% centile relative to the UK population reference data (Cole 2002). Data with ≥ 20 min of consecutive zeros were identified as non-wear time and excluded from analysis, as recommended by Esliger et al (2005). A ‘standard measurement day’ was identified as the length of time when at least 70% of the sample wore the accelerometer for 80% of the time (Catellier et al 2005).
2217
J Hislop et al
Physiol. Meas. 35 (2014) 2213
Table 1. Characteristics of sample.
Mean (SD)
No. participants Age (years) Height (cm) Body weight (kg) Body mass index (kg m−2) Mean BMI percentile
All
Male
Females
112 3.7 (0.7) 101.3 (3.9) 17.0 (2.3) 16.6 (1.6) —
60 3.7 (0.7) 101.6 (3.7) 17.5 (2.3) 17.0 (1.7) 85th
52 3.7 (0.7) 101.0 (4.1) 16.5 (2.1) 16.2 (1.3) 71st
3.1. Comparison of weekend and weekdays
Data from all 112 participants (60 males, 52 females, mean (SD) age: 3.7 (0.7) yr) were included in the analysis of weekend and weekdays. Using all participant data the mean cpm for participants’ weekdays of data and the mean cpm for their weekend days were extracted to allow comparison between weekend and weekdays. Data were evaluated for normality using the Kolmogorov-Smirnov test. As the data were not normally distributed, descriptive statistics of the medians and inter-quartile ranges (IQR) were calculated, and presented graphically in box plots. The Wilcoxon signed rank test was used to compare differences between cpm for weekend and weekdays. In addition, a comparison was made between the cpm for males and females at weekend and weekdays using a Mann-Whitney U test, the non-parametric equivalent of the independent t-test. The level of significance was set at p
