CHAPTER THREE
Equation 3.1. Mattocks et al. (2007) energy expenditure regression equation
3.4. Sources of misclassification in the association between physical activity and psychological well-being activity and psychological well-being
3.4.3. Cut-point misclassification
One of the most common sources of misclassification apparent in accelerometer measured physical activity comes from the application of intensity cut-points. One of the most widely used accelerometers; the ActiGraph (ActiGraph, Pensacola, FL) has been the subject of multiple calibrations within similar populations (e.g. children). The availability of multiple thresholds for use with children presents researchers with a conundrum about the most appropriate thresholds to employ, whilst employment of various cut-points may lead to the misclassification of physical activity levels in children.
The impact of employing different thresholds has been examined by relatively few studies, however, those that have, found large differences in time accumulated in PA intensities. For example, McLure et al. (2009) examined the
102 portion of children achieving the recommended 60 minutes MVPA per day (Department of Health, 2011) after the application of two sets of cut-points.
When using the Freedson/Trost (Freedson et al., 2005, Trost, 2002) cut-points for children aged 9-10 (≥ 1100 counts∙min-1), mean time in MVPA was 126.0 min·day-1, with 97% of participants achieving the physical activity guidelines.
However when the Puyau, Adolph, Vohra and Butte (2002) cut-points were applied (≥ 3200 counts∙min-1), mean time in MVPA decreased to 28.9 min·day-1, with the percentage of children meeting physical activity recommendations decreasing to 7%.
This problem was also examined by Stone et al. (2009), who applied three different activity intensity cut-points to ActiGraph data; sample specific (MVPA >
2910 counts·min-1), Mattocks et al. (2007) (> 3581 counts·min-1), and individualised thresholds developed using Ekelund, Aman and Westerterp’s (2003) ArteACC method. Results showed that 8.5%, 48.9% and 100%, of children met the recommendations when sample specific, Mattocks et al. (2007) and ArteAcc, cut-points were applied respectively. Additionally, Richardson, Stewart-Brown, Wilcock, Oldfield and Thorogood (2011) reported differences of 44 minutes of accumulated MVPA between Freedson & Puyau cut-points. More recently, Guinhouya, Samouda & de Beaufort (2013) undertook a literature review of 35 studies of objectively measured physical activity level of children across Europe, reporting results for various cut-points; Using the Freedson age and sex specific equation, between 78 and 100% of children met the recommended level of MVPA, while studies using cut-points of around 2000 counts·min-1 reported 36 – 87% of children met recommendations. For the use of cut-points > 3000 counts·min-1, 3 to 9 % achieved sufficient MVPA while only 1% of children met MVPA recommendations with a cut-point of > 4000. Similarly in adolescents, the use of different cut-points resulted in between 4% and 100%
of adolescents reported as meeting the MVPA guidelines. The findings from these studies highlight the importance of cut-point application upon estimates of physical activity levels in children; however, the number of thresholds examined in each study is relatively small compared to the number of cut-points available within the literature.
103 Recently, researchers have attempted to establish consensus within the literature as to which published thresholds may be most appropriate for accurately determining children’s physical activity levels. Trost, Loprinzi, Moore and Pfeiffer (2011) compared the classification accuracy of a series of regularly used, published cut-points (Evenson, Cattellier, Gill, Ondrak & McMurray, 2008;
Freedson et al., 2005; Trost, 2002; Mattocks et al., 2007; Puyau et al., 2002;
Treuth et al., 2004) to a criterion measure (indirect calorimetery) for sedentary, light, moderate and vigorous physical activity intensities via ROC analysis.
Across all intensities, Evenson et al. (2008) and Freedson / Trost cut-points reported consistently high sensitivity, specificity and area under the curve. The authors therefore, concluded that these cut-points were the most accurate method of assessing children’s time in activity intensities and should be employed in future research in an attempt to establish comparability between studies. Following this, Kim, Beets and Welk (2012) conducted a systematic review of the cut-point literature, they reported that the Freedson and Evenson cut-points gave estimates ranging from 64–124 minutes per day and 47–61 minutes per day respectively, and subsequently noted that the difference between the lower-bound thresholds for MVPA is where possible misclassification of PA occurs.
The possibility of misclassification does not occur solely between the light and moderate boundaries, the discrepancies between cut-points for other intensities may also result in this misclassification. Despite this, researchers are compelled to continue to employ cut-point methods to define accumulated time in physical activity intensities for the immediate future. Thus the potential for misclassification of children’s physical activity levels remains high. The matter becomes yet more confounded as researchers, perplexed as to the most suitable cut-points to use, create their own sample specific cut-points, adding to the plethora already available within the literature (e.g. Mckintosh et al., 2012).
The misclassification of physical activity has recently been examined with regard to the impact it may have upon relationships between physical activity and health related variables. Only a few studies have explored this concept in relation to physical health. For example, both Stone et al. (2009) and Atkin et al.
(2013) have shown that the magnitude and significance of relationships between specific activity intensities and health variables differed depending
104 upon the ActiGraph cut-points employed. Similar results have also been noted with other accelerometer models. Bailey, Boddy, Savory, Denton and Kerr (2013) showed that in children aged 10 to 14 years, the use of different cut-points developed for the RT3 resulted in attenuation of relationships between PA and cardio metabolic risk factors. For example, a relationship between light PA and body fat percentage was apparent in girls with the use of the Rowlands et al., (2004) cut points (r = .303, p = .044), yet this relationship was attenuated with the employment of the Vanhelst et al. (2010) and Chu & McManus (2007) cut-points (r = .055 & r = .283, p’s > 0.05), indicating that misclassification not only alters time reported in activity intensities but also impacts on the observed relationships with health. The impact of misclassification upon health variables will be the focus of chapter four of this thesis.
A potential method of overcoming the misclassification due to cut-point method is to establish a more appropriate way to analyse the data. Newer accelerometers such as the GENEActiv and the ActiGraph GT3X+ allow examination of the raw acceleration data, rather than counts. Furthermore, pattern recognition technologies are continuing to be developed, which remove the need for examination of counts and classification of intensities. Yet, until these methods are available, the cut point method will continue to be used in physical activity research.