Physical activity assessment – Accelerometery

Accelerometry is a direct measure of acceleration of the body or segments of the body. Accelerometry is the most common objective method used to measure physical activity; it has been used extensively in field settings to monitor activity patterns. Technological advances have resulted in devices that can measure activity accurately, over an extended time period (greater than 7 days), and that are small and discrete for people to wear. The device is enclosed in a case and typically attached to the hip (or lower back, ankle, wrist or thigh) by a strap.

Acceleration is a change in velocity with respect to time (SI unit: m.s^-2). Muscular forces can result in acceleration of body mass. Both the acceleration of the body mass and the amount of body mass being accelerated are in theory related to energy expenditure.

In accelerometry, acceleration is measured by piezoelectric or seismic sensors in one (longitudinal body axis, usually vertical), two (vertical and medio-lateral or vertical and anterior-posterior) and three (vertical, medio-lateral and anterior-posterior) directions (Chen et al, 2005). Accelerometers attached to the waist do not capture upper body movement or cycling, and underestimate walking on an incline or carrying heavy loads. The latter are examples of where the energy cost of physical activity may not necessarily be equivalent to body movement. Accelerometers are usually not waterproof although a few models are splash-proof, and should be removed for water-based activities. Classical accelerometers used in the epidemiological field sampled acceleration at some higher frequency (10-32 Hz) but then reduced the information on-the-fly to store local averages on say a minute-to-minute basis. This is due to memory and battery restriction. It is important to acknowledge that these data processing “decisions”, made by any given monitor makes the stored information fundamentally different to the original acceleration signal. Current monitors store more data, from which it is possible to make more sophisticated inferences, either on-the-fly or at the post-processing stage.

A number of manufacturers produce accelerometers, and studies have shown differences in values both within and between models (Welk et al, 2000; Brage et al, 2003; Welk et al, 2004; Esliger & Trembaly, 2006; Plasqui & Westerterp 2007). Regular mechanical calibration of the sensor is recommended to overcome the former issue. The underlying scientific principles and technical specifications have been comprehensively described (Chen et al, 2005; Pober et al, 2006). Accelerometers used in paediatric studies have also previously been summarised (Reilly et al, 2008).

Outcome measures
The primary outcome measure of accelerometry is body acceleration, often expressed as a count value. A “count” is an arbitrary unit, which varies across devices and even generations of the same device type (Rothney et al, 2008). It is influenced by the amplitude and frequency of acceleration, also to a varying degree between different types of instrument. Secondary outcomes are estimates of bout frequency, duration and intensity of body movement. Output measures may be presented as:

activity counts i.e. counts/min
Units of acceleration (g-force units or m.sec²)
Time spent in physical activity with varying intensity i.e. time spent above an intensity threshold pre-determined by a regression equation
Number of bouts of activity i.e. the number of times there was continuous movement above an intensity threshold (pre-determined by a regression equation) and a duration threshold

Given the error in the estimation of energy expenditure, the arbitrary nature of the count-based cut-offs, and the difference in counts across models, reporting accelerometry data as units of acceleration (m.sec²) has been recommended (Brage et al, 2003; Esliger et al, 2007; Corder et al, 2008).

Prediction equations and cut-points
The aim of energy prediction equations and cut-points to differentiate thresholds of activity intensities is to characterise the relationship of movement counts and other physiological measures of activity. If a threshold reflects a particular intensity level of activity accurately, the time spent above and below this intensity of activity can be determined. In practice, the characterisation of this relationship is complex because the relationship between counts and energy expenditure is not linear across all activities; different relationships exist for different activities (Treuth et al, 2004). The approach used to derive prediction equations varies widely and some studies have used only small numbers of volunteers and sometimes only single sex participants. It has been suggested that establishing the relationship between activity counts and energy expenditure is especially problematic in children due to their growth and development which affects estimates of resting metabolic rate (RMR) and energy expended (i.e. movement economy) during activity. Children’s resting metabolic rate expressed relative to body weight decreases with age and maturation, and similarly the energy expended relative to body mass during for example walking and running also decreases (improved movement economy) with age (Trost, 2007). However, this issue was examined in one study of 3-10 year olds undertaking a wide range of activities; accelerometry output was not found to be systematically different across the ages or activities (Reilly et al, 2008).

Additionally, an accelerometer placed on one body location does not capture all activity of other body sites, although there is usually some cross-correlation (Welk et al, 2004). For example, prediction equations developed during walking-jogging experiments tend to estimate energy expended during walking and jogging reasonably well, but (Crouter et al, 2006a).

Cut-points for defining different intensity levels are somewhat arbitrary and the use of different cut points can have profound impact on the estimate of the physical activity (Freedson et al, 2005; Matthews et al, 2005). A researcher using accelerometry must understand the derivation of prediction equations from calibration studies and the rationales and implications of choosing a particular set of cut points (see Matthews et al, 2005; Welk, 2005; Reilly et al, 2008). For example, published cut points for sedentary from one accelerometer varies from <100 cpm to <800 cpm. Similarly, the range of cut points for moderate intensity activity varies between < 200 cpm to > 3000 cpm.

One study (Schmidt et al, 2003), compared data from a physical activity log against three different sets of cut points and found wide differences (38 to 312 minutes per day) in time spent in moderate vigorous physical activity (MVPA).This difference was explained by a 10-fold difference in cut points (191 cpm to 1953 cpm).

When establishing the relationship between accelerometry output and energy expenditure several points must be considered:

There may be variation between devices so multiple monitors should be used in a calibration study and the coefficient of variation reported
The activities used during calibration should be reflective of the types and intensities of the activities undertaken by the population under investigation
Calibration equations derived from lab-based activities will most likely differ from those derived from free-living activities
Sustained and stop-and-go activity will display different relationships with energy expenditure, the prevalence of which may also differ by population, e.g. children tend to have frequent short bouts of activity compared to adults.
Movement type varies between populations and this should ideally be reflected in the design of a calibration study.
Cut-offs may differ between different groups (e.g. children and elderly) and should be established or chosen for the group under investigation.

Laboratory-derived physical activity energy expenditure equations are not all equally suitable to assess physical activity in free-living populations (Nilsson et al, 2009); equations derived from field-based activities appears to be most suitable to estimate energy expenditure at a group level (Ekelund et al, 2001). Laboratory-derived prediction equations have been found to overestimate free-living energy expenditure by 47% in one study using doubly labelled water (DLW) (Ekelund et al, 2004).

The overall relationship between activity energy expenditure and uniaxial accelerometry during rest, walking, and jogging is fairly linear. However, uniaxial linearity breaks down for high intensity running (Brage et al, 2003) which is better captured with additional measurement of the antero-posterior axis of acceleration (Rowlands et al, 2007). Common for uniaxial and triaxial accelerometry, however, is that those linear relationships derived for rest and ambulation display much poorer validity in biomechanically diverse activities, e.g. cycling or lifting weights. Advanced statistical methods have been proposed to improve prediction equations. For example, a 2-segment regression model has been shown to improve accuracy of energy expenditure estimates, compared with a simple regression (Crouter et al, 2006b). Decision boundaries and receiving operating curves (ROC) have been suggested to more appropriately minimise misclassification error (Jago et al, 2007). An alternative procedure (ArteACC) corrects an individual’s total activity accelerometer count in a time interval by the activity counts for the individually assessed average accelerometer response during two or more standardised activities, e.g. 4 and 6 Km.h^-1 on a treadmill (Ekelund et al, 2003). However, this form of individual calibration is less feasible for use in large-scale studies.