Self-report recall questionnaires

Questionnaires, asking about past activity over a predefined time frame, may be self-completed or administered by an interviewer. Children younger than 10 years do not have the capacity to report their activities accurately⁽ Reference Baranowski, Dworkin and Cieslik ^{20 )}. Parental proxy answers may be no better at accurate reporting: they cannot report school-based activity, and informal play is particularly difficult to assess. In 2006, we undertook a desk-based review of all physical activity questionnaires used in diet and nutrition studies. We concluded that physical activity in children under 11 years could not be measured in sufficient detail by the use of questionnaires to estimate EE. A 2010 review found that no questionnaires for children were both reliable and valid, with worse results in younger children than adolescents⁽ Reference Chinapaw, Mokkink and van Poppel ^{21 )}.

Short questionnaires often enquire about the frequency and average duration of categories of activity (usually light, moderate and vigorous) over a pre-specified time frame, giving examples of typical activities within each group, based on the metabolic equivalent (MET; 1 MET=the rate of energy expenditure while sitting quietly) intensity coding scheme developed by Ainsworth and colleagues⁽ Reference Ainsworth, Haskell and Herrmann ^{22 )}. Despite the advantage of less respondent burden, they provide insufficient detail to estimate EE. Detailed questionnaires ask which activities the survey participant have done in a given time frame, and for ‘each activity’ asks about the frequency and average duration, and questions to ascertain the intensity. For example, whether the activity makes the respondent breathless, hot or sweaty (or not) would categorise cycling as vigorous or moderate, respectively⁽ Reference Ainsworth, Haskell and Herrmann ^{22 )}.

Questionnaires are relatively quick, inexpensive, and easy to administer but have many important limitations⁽ Reference Bailey, Olson and Pepper ^{23 )}. A major problem is recall bias. Questionnaires asking about habitual activity⁽ Reference Corder, Ekelund and Steele ^{24 )} risks responses reporting ‘most active weeks’ instead of ‘usual’ activity or activity in the specified time period, most commonly for children the previous 7 d⁽ Reference Craig, Mindell and Hirani ^{18 )}. Shorter time periods give greater recall accuracy but determine irregular activities less accurately. In some individuals, light activity can be an important component of physical activity energy expenditure (PAEE)⁽ Reference Westerterp and Plasqui ^{25 )} but light activity is extremely difficult to define and capture using self-report or proxy methods. In children, physical activity is sporadic in nature, being characterised by short bursts: during informal play up to 96 % of activity may occur in bouts <10 s⁽ Reference Corder, Ekelund and Steele ^{24 )}. Social desirability bias may play a part among adolescents⁽ Reference Jago, Baranowski and Baranowski ^{26 )}.

Physical activity of children aged 2–15 years was first assessed in the HSE in 1997 (by parental proxy interview for children aged 2–11 years and self-report for those aged 13–15 years). Differing instruments were used subsequently; a new questionnaire was developed de novo for HSE 2008⁽ Reference Craig, Mindell and Hirani ^{18 )}. For the 2012 HSE, additional questions were added, to assess adherence to national recommendations for muscle strengthening exercises as well as sufficient weekly duration moderate or vigorous physical activity (MVPA)⁽ Reference Scholes, Mindell, Craig and Mindell ^{27 )}. In older children, self-report questionnaires overestimated activity levels compared with accelerometry⁽ Reference Craig, Mindell and Hirani ^{18 )}; in younger children, parental proxy questionnaires substantially underestimated MVPA (Fig. 1)⁽ Reference Craig, Mindell and Hirani ^{18 )}.

Source: Reanalysis from Health Survey for England 2008⁽ Reference Craig, Mindell and Hirani ^{18 ) a}MVPA: moderate or vigorous activity.

Fig. 1. Comparison of accelerometry data with reported average daily moderate or vigorous physical activity time in children.

For the rolling programme UK NDNS, a new questionnaire for teenagers was developed to attempt to estimate PAEE accurately at the individual level, by including detailed questions on all forms of activity. It performed well at the population level in categorising participants into activity levels but was not accurate at an individual level when compared with doubly labelled water (DLW; an objective measure of EE used in dietary surveys)⁽ Reference Scholes, Mindell, Bates, Page and Swan ^{19 )}.

For questionnaires there is a trade-off between completeness of the information collected, balancing how comprehensive the questionnaire is with participant burden and the risks of missing data items and/or lack of co-operation on further stages of the dietary survey. In practice, questionnaires have been the most frequently used method for recording physical activity, although there is a trend towards moving to more objective methods for children in population surveys⁽ Reference Trost ^{15 ,} Reference Oliver, Schonfield and Kolt ^{16 )}.

Assessing energy expenditure from physical activity questionnaires

Population-level assessment of EE may be required to compare changes over time with changes in energy intake and obesity prevalence, or to compare population subgroups. Accurate assessment of individuals’ EE is generally used in dietary surveys for comparison with recorded intake, to assess the accuracy of reported diet. Where EE assessment is not sufficiently accurate, underreporting can be assessed by grouping participants into EE categories for comparison of reported intake.

Questionnaires that cover all physical activity domains contribute significantly to EE and in sufficient detail to record frequency, duration and intensity of activity, are very long but fail to provide accurate EE measures at the individual level because such precise recording of activity intensity and duration across all levels of intensity is not feasible. EE assessment using questionnaires is still a common technique to assess free living EE, although seldom valid due to inadequate coverage of all activity domains and inaccurate assignment of MET values (EE) to reported activities, particularly in children. In NDNS, the population means derived from the questionnaire and from the DLW data were close but participants could not be assigned into meaningful EE categories because of the large variance in the results⁽ Reference Mindell, Stamatakis and Brage ^{28 )}.

Validation may be hampered by use of different time periods⁽ Reference Neilson, Robson and Friedenreich ^{29 )}. If the DLW collection period does not overlap fully with the other measurements (other physical activity data, dietary intake), results will contain additional noise among those participants whose activity pattern is unstable with considerable between-day or between-week variation⁽ Reference Scholes, Mindell, Bates, Page and Swan ^{19 ,} Reference Walsh, Hunter and Sirikul ^{30 )}.

Activity diary or log

Use-of-time tools have moderate validity and reliability for assessing young people's physical activity and EE⁽ Reference Foley, Maddison and Olds ^{31 )}. An activity log requires the individual, parent or other responsible adult to record at regular time intervals what activity or activities have been undertaken in that time period. As with questionnaires, young children cannot do this and parents are often not present. Despite the theoretical advantage of better completion, as minimal recall is required, there is a trade-off concerning the time interval selected: the shorter the duration, the more accurately physical activity can be recorded but more frequent recordings add considerably to participant burden and therefore reduce completeness and/or accuracy⁽ Reference Corder, Ekelund and Steele ^{24 )}.

A diary requires the participant to record specific groups of activities. However, diaries usually collect information on MVPA only as a proxy for EE, which is not ideal as light intensity activities, such as standing and non-brisk walking, may contribute substantially to overall PAEE. Activity diaries, used in the first UK NDNS of children⁽ Reference Gregory, Lowe and Bates ^{32 )}, give valid data in adolescents for estimating physical activity category and total EE, assessed by DLW measurement⁽ Reference Bratteby, Sandhagen and Fan ^{33 )} but add to the participant burden. Electronic diaries have been validated against accelerometry in children⁽ Reference Floro, Dunton and Delfino ^{34 )} but are relatively costly and are not commonly used in dietary surveys in children.

Pedometers

Pedometers measure steps taken. They can be a useful motivational tool to encourage activity⁽ Reference Beets, Patton and Edwards ^{37 ,} Reference Duncan, Schofield and Duncan ^{38 )}. Their main advantages are their low cost, compared with other devices, and the simplicity of the data produced: the number of steps taken. They are therefore useful for large population surveys⁽ Reference Eston, Rowlands and Ingledew ^{39 )}, and were used in the Australian National Survey of Children's Diet and Nutrition^{( 40 )}. However, they cannot capture other forms of activity, including cycling and swimming, nor increased EE from walking on an incline or carrying a load. They have been validated in children⁽ Reference McNamara, Hudson and Taylor ^{41 )}, including adolescents⁽ Reference Jago, Watson and Baranowski ^{42 )}, but are not recommended for pre-school children⁽ Reference Oliver, Schofield and Kolt ^{43 )}.

Dietary surveys need research-grade pedometers that have adequate performance criteria⁽ Reference Tudor-Locke, Sisson and Lee ^{44 )}. They also need to be calibrated, although this is simple, as step size depends on age and speed⁽ Reference Trost ^{15 )}. While some authors have found pedometer data can estimate children's EE reasonably accurately⁽ Reference Corder, Ekelund and Steele ^{24 ,} Reference Eston, Rowlands and Ingledew ^{39 )}, most recommend that the data be analysed only as the number of steps taken⁽ Reference Corder, Brage and Ekelund ^{45 )}.

Accelerometers

Accelerometers, introduced in the early 1980s⁽ Reference Matthews ^{46 )}, have been used to assess physical activity and to validate self-reported physical activity. They provide objective, standardised data without recall bias and other subjectivity problems but participants’ co-operation is required to wear the monitor during waking hours⁽ Reference Roth and Mindell ^{47 )} and, if required, to record activities when not worn.

Accelerometers can measure directly the frequency, duration and intensity of most aspects of physical activity. The greater precision of the objective data yields equivalent statistical power with lower sample sizes. Accelerometer counts increase almost proportionally with ambulatory speed⁽ Reference Matthews ^{46 )}, so are excellent for recording time being inactive and time spent on dynamic activity such as walking or running up to a certain speed. However, they are less good at recording static activity, often important for overall EE estimation. Non-wear time can also be an important issue. In relation to estimating EE, the main issue is that the equations for conversion from accelerometers to EE perform poorly in non-controlled free-living situations⁽ Reference Corder, Ekelund and Steele ^{24 )}.

Accelerometers can measure movement in one, two or three dimensions (uni-, bi- or tri-axial measurements). Uni-axial accelerometers produce data that are the simplest to interpret but are limited by the lack of detection of movement in other directions; for example, rowing and cycling are poorly detected. Uni-axial monitors may also not detect inclines; walking more slowly up a 5 % hill may use more energy than walking faster on the level. No devices detect load-carrying, also underestimating EE while carrying a heavy school bag, for example. Most accelerometers are not water-proof, and must therefore be removed for swimming. Some ethics committees may also impose restrictions, e.g. to instruct participants to remove the device when participating in contact sports, including martial arts.

After the accelerometers have been worn, the devices need to be returned or collected, so that data can be downloaded from the devices and processed. Data processing of accelerometer data includes differentiating between periods of non-wear and low counts indicating inactivity (sedentary behaviour); decisions about minimum wear time for inclusion in analysis (minimum number of hours worn per day (‘sufficient wear’) and minimum number of days of sufficient wear)⁽ Reference Roth and Mindell ^{47 )}; which thresholds to use to categorise activity as sedentary, light, moderate or vigorous⁽ Reference Ainsworth, Haskell and Herrmann ^{22 )}; and which equations to use to estimate MET or EE, if that is intended. Data processing has become much more sophisticated over time. For example, probability modelling is replacing a dichotomous decision of whether or not a device was deemed to be worn at any particular moment, and therefore whether that reading should be included.

Accelerometry has been used for population surveillance. The US National Health and Nutrition Examination Survey was the first general population study to use accelerometry in children and adults in a population-wide survey, in 2003/04^{( 48 )}. They used accelerometers with over 1000 participants, aged 6 years and over, demonstrating the feasibility of objective measurement in large population studies. The first surveys to issue accelerometers ‘in the field’ occurred in England⁽ Reference Craig, Mindell and Hirani ^{18 ,} Reference Scholes, Mindell, Bates, Page and Swan ^{19 ,} Reference Mindell, Stamatakis and Brage ^{28 ,} Reference Roth and Mindell ^{47 , 48 )}. Both the HSE and NDNS used the Actigraph model GT1M, a lightweight, digital, accelerometer, smaller than a matchbox. Movements outside normal human motion are filtered out electronically⁽ Reference Rothney, Apker and Song ^{49 )}. The filtered acceleration signal is digitised and summed over a specified time interval to produce a number of ‘counts’ for that time period, which is stored and the summing restarts. We used a 1 min period, to maximise data storage, but shorter periods give more detailed assessment of short bursts of activity, characteristic of children. In the NDNS, a subsample of children provided DLW estimates over 10 d, concurrent with accelerometer wear⁽ Reference Mindell, Stamatakis and Brage ^{28 )}.

The choice of Actigraph was based on cost, ease of use and compliance, reliability, adequacy for physical activity assessment⁽ Reference Trost, Way and Okely ^{50 ,} Reference Pate, Almeida and McIver ^{51 )}, and a wide comparison base. Participants in the HSE and NDNS were asked to wear the accelerometer on their waist, positioned above the right hip, during waking hours for seven consecutive days.

Heart rate monitors

Heart rate monitors are usually worn on a strap around the chest. Measuring changes in heart rate can be a better assessment of physical activity in relation to cardiovascular impact but is not responsive to low-intensity activity and cannot distinguish between different causes of raised heart rates (activity, pain and other emotional stimuli). This is the main issue in free-living studies, where measurement ‘noise’ is more of a problem for heart rate than for motion sensors⁽ Reference Stegle, Fallert and MacKay ^{57 )}. There are also problems of latency, with the heart rate change lagging behind changes in activity, and heart rate elevation persisting after cessation of activity, particularly in less fit individuals. This may conceal intermittent activity, particularly in children.

Combined heart rate and activity monitors

Monitors that detect both movement and heart rate provide more accurate assessment of energy use⁽ Reference Avons, Garthwaite and Davies ^{58 )}. Accelerometry provides accurate data for sedentary and light activity, where heart rate monitors contribute less. Monitoring the heart rate allows the intensity of the activity to be captured more accurately. Such devices can also distinguish well between non-wear and sedentary behaviour, and can help describe the intensity and duration of activity more accurately, although requiring analysis of more complex data. They may provide more accurate assessment of EE⁽ Reference Corder, Ekelund and Steele ^{24 )}, including in children⁽ Reference Eston, Rowlands and Ingledew ^{39 ,} Reference Corder, Brage and Wareham ^{59 )}. However, calibration is required for each individual to optimise data interpretation⁽ Reference Brage, Ekelund and Brage ^{60 )}. Current combined sensor technology can be relatively burdensome for the participant (e.g. certain devices require removal of hair from men's chest) and there is a risk that gains in accuracy may be offset by losses in participant adherence. Furthermore, at the time of preparing this report, combined sensors are still very expensive and beyond the reach of most dietary surveys.