Background
Physical activity (PA) plays a key role in health and well-being of an individual. PA is beneficial for all age groups and has both short- and long-term benefits [
1]. There is strong evidence that PA reduces the risk of cancer and cardiovascular mortality [
1,
2]. Also, PA prevents several lifestyle disorders such as diabetes, cardiovascular diseases, arterial stiffness, obesity and metabolic syndrome [
3‐
5]. In addition, higher amounts of PA and/or lesser sedentary behaviour are associated with a lesser risk of depression [
6] and sleep disorders [
7].
In the era of digital development, the use of smartphone applications and other digital devices like wearable PA monitors has become popular for tracking PA in the free-living environment. Recent developments in wearable technology enable regular and long-term tracking of PA, representing a promising measure for the promotion and support of a more active lifestyle [
8]. These devices provide feedback in form of step counts, energy expenditure or total time spent in moderate to vigorous intensity. This feedback can encourage behavioural change towards a more active and less sedentary lifestyle [
9]. Indeed, several studies demonstrated that PA monitors and smartphone applications can be used to promote PA in individuals [
10,
11].
Nowadays, there is an increasing amount of clinical and population-based observational studies evaluating the amount of PA of a person using wearable devices such as smartwatches, smartphones, consumer-grade or research-grade activity monitors. A common characteristic of these devices is that they include an accelerometer (a Micro Electro-Mechanical System or an Inertial Measurement Unit which consists of an accelerometer, a gyroscope and a magnetometer) which measures acceleration relative to the Earth’s gravitational field. Compared to questionnaire-based assessments, wearable devices present the advantage of detailed and objective measurements of the PA behaviour in the free-living environment [
12]. Especially when the PA duration or the intensity is of interest, self-reported results should be interpreted with caution [
12]. Whilst the questionnaires provide information on the purpose of the activity (and sometimes on activity type), the wearable devices quantify the motion performed. Thus, both approaches provide complementary information, which are not interchangeable [
13].
The output of the accelerometers is a three-dimensional time series of accelerations expressed in gravitational units. Developments in the field of computing techniques for the domain of health sciences have made analytical tools more easily available [
14], and enabled researchers from both academic and industrial milieu to decipher the 24-h raw time series acceleration signal. These analytical tools can now generate a number of summary variables. However, until now, most studies are limited to measuring PA in units of time spent in certain intensity levels, which can subsequently be used to compare sub-groups by categorising people as active or inactive [
15,
16]. The latter classification is usually based on PA recommendations of 150 min of moderate PA per week or 75 min of vigorous PA or an equivalent combination of both [
17], although many alternative recommendations exist [
18]. The classification approach is practical and convenient but does not reveal a complete PA profile of a person. Actually, cut-point approaches comes with several limitations such as (1) the complex relationships of acceleration with energy expenditure, activity types and study populations, (2) the many parameters that are somehow arbitrarily determined (e.g. bout length), and (3) the collinearity between PA intensity categories [
19]. The latter implies that the time spent in different PA intensity categories cannot be used in standard regression models because of the compositional nature of the data [
5].
How a person accumulates minutes of sedentary time was demonstrated to be a key determinant regulating different biomarkers of obesity [
20]. Similarly, how a person accumulates active time is of significant importance to provide adequate guidance and appropriate recommendation for each individual [
18]. For example, some people may be moderate to vigorously active for a long bout of exercise whilst spending the rest of the day in sedentary behaviour, whereas some others may be active for frequent short durations by doing household work, gardening, commuting from home to office, etc. Each type of PA behaviour pattern has different health implications [
21]. Indeed, PA is a complex multidimensional human behaviour, which dimensions include amongst others frequency (e.g. number of light activities per week), intensity (e.g. exercise or active sports are usually in moderate to vigorous intensities), time (e.g. time spent sitting), type (e.g. walking) or posture (e.g. sitting), each describing a different aspect of the PA behaviour. The total volume of PA (e.g. total time spent at light intensity) represents an important variable as it combines the intensity and time dimensions of PA. However, it reduces both dimensions to a single summary variable, which comes with a loss of information. In other words, the way PA is accumulated over time does not influence the results. Therefore, advanced approaches taking into account the way PA is accumulated over time and providing multidimensional output may be more suitable to analyse different patterns of activity, investigate their association with health conditions and suggest appropriate recommendations for changes in PA behaviour [
18].
Over the last few years, more advanced analytical approaches to generate new summary variables capturing more appropriately the multidimensional nature of PA behaviour have emerged [
22‐
24]. Irrespective of whether these advanced analytical approaches are cut-point dependent or independent, they allow generating a comprehensive profile of PA pattern (Fig.
2). Accelerometer time-series raw data presents the possibility to evaluate and test whether detailed patterns of activity may be more informative to health outcomes than traditional measures of total activity.
In fact, accelerometry is a field in health sciences that emerged in the 1990s and is still very actively exploring analytical techniques. Currently, there is no universally accepted and standardised approach for measuring PA using wearable accelerometers in health research. Thus, there is considerable heterogeneity in methodology, data acquisition and data processing. Several published systematic reviews aimed to analyse the methods and results of calibration studies relative to energy expenditure (i.e. PA intensity) [
25], data processing criteria applied to the acceleration signal [
26] and the completeness of accelerometer reporting methods [
27]. A preliminary search revealed that none of the previous reviews have mapped existing multidimensional PA summary variables for assessing PA and investigating the association with health outcomes.
The choice of PA summary variables must be based on its ability to reveal clinically relevant features of physical behaviour and to discriminate between experimental groups. Previous authors have supported the need to overcome the limitation of unidimensional PA related-outcomes in defining the activity profile of a person [
21,
28]. Some studies have incorporated a variety of equally important characteristics of PA, such as intensity, duration, distribution and or timing of acceleration intensity over the day, or bouts duration [
19,
23,
29].
Hence, the aim of this scoping review is to map advanced analytical approaches and their multidimensional summary variables used to provide a comprehensive picture of PA behaviour of an individual. Therefore, this scoping review will look at analytical methods that go beyond total PA volume, average daily acceleration and the conventional cut-point approaches, involving tri-axial accelerometer data from any sensor attachment location, and covering both data- and knowledge-driven techniques. We will also review the studies evaluating the association between the identified multi-dimensional summary variables with different health parameters. These methods may be used in future studies investigating the association between PA and health outcomes (e.g. cardio-metabolic health, diabetes, frailty), as well as in future personalised interventions in public health. This work will provide researchers as well as consumer wearable device companies with decisive information on future developments in the data processing, as well as on relevant feedback to the end user.
Discussion
The accuracy of a PA measurement as well as the dimensions covered by the outcomes depend on the analytical approaches and the outcome(s) calculated. It is important to select the appropriate outcomes depending on which dimensions of PA are important for a certain health condition. Through this scoping review, the authors look forward to systematically map the available cut-point independent and multidimensional variables, as well as to identify potential knowledge gaps on analytical methods for the assessment of PA. The results of this review may be used to guide future research related to the assessment of PA and the individualised feedback to each person. The results of this scoping review may be of interest to sports scientists, clinical researchers and smartphone application developers in the field of PA.
This scoping review will have some potential limitations. We will not discuss the accelerometer-based approaches that combine measures from other sensors measuring heart rate variability, temperature changes, oxygen saturation or skin conductance to name but a few. Besides, studies investigating sleep or sedentary behaviour as a single standalone sub-domain of the physical behaviour fall out of the scope of the present review. Studies identifying the type of PA through machine learning and artificial neural network approaches will not be included as this topic was covered by recent systematic reviews.
Ethical approval is not required for scoping reviews as it is based on the analysis of published data and results. Findings of this scoping review will be disseminated through peer-reviewed journals and conferences to contribute towards the development of PA measurement and health promotion in the community.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.