A novel integrative procedure for identifying and integrating three-dimensions of objectively measured free-living sedentary behaviour

Sedentary behaviour (SB) is highly prevalent in the twenty-first century accounting for between 46% - 72% of the waking day [1‐3]. The prevalence of SB is of particular concern as there is evidence to suggest SB is an independent risk factor for deleterious health outcomes including metabolic syndrome, all-cause mortality, cardiovascular disease, type 2 diabetes and obesity [4‐6]. The Sedentary Behaviour Research Network (SBRN), an organization of researchers and health professionals, define SB as “any waking behaviour characterized by an energy expenditure ≤1.5 metabolic equivalents (METs) while in a sitting or reclining posture” [7]. Sedentary time is defined as “the time spent for any duration (e.g., minutes per day) or in any context (e.g., at school or work) in sedentary behaviours” [8].

Free-living SB is notoriously difficult to quantify due to the complexity of human movement. Previous studies have used TV viewing as a proxy measure to reflect total sedentary time [9, 10] however, TV viewing is not representative of overall sedentary time and it is also associated with other health related behaviours such as increased energy intake, particularly from fat [11, 12]. To address the limitations of self-report proxy measures of SB, objective measurement methods are increasingly being used [2, 13, 14]. Devices used to measure free-living movement behaviour reflect different facets of SB and can provide quite different estimates of sedentary time [15]. The SenseWear Armband (SWA) uses information from a triaxial accelerometer (as well as other physiological information) to estimate sleep time and activity intensity and defines SB as activities with an intensity of <1.5 METs (see [16‐18]). Alternatively, the activPAL (AP) uses a triaxial accelerometer to measure posture and defines SB based on a sitting or reclining posture (see [19‐21]). Separately, each activity monitor has been shown to accurately measure the specific dimension of SB they specialise in, however, they are not able to capture multiple components of SB [20, 22, 23]. The multi-component definition of SB makes accurate assessment of sedentary time complex because different components of SB do not always covary; it is possible to be seated whilst expending >1.5 METs and it is also possible to stand whilst expending <1.5 METs [24, 25]. In other words, it is possible to be inactive whilst not in a seated posture; and alternatively to show some activity (>1.5 METs) whilst actually being seated. This issue, reflected by the concepts of passive standing and active sitting, is incorporated into the very recently published terminology consensus from the SBRN [8]. Therefore, there is a need for measurement tools that provide information on activity intensity and posture simultaneously as well as indicating whether the wearer is awake or asleep.

Surprisingly, there is no single objective measurement device that accurately measures sleep, posture (sitting/reclining) and activity intensity (<1.5 METs) simultaneously. Recent research has demonstrated combining information from multiple devices yielded a more accurate measure of SB and the authors encouraged further research using the “multi-method” approach [26]. Furthermore, data from the ActiGraph and AP have been shown to have greater accuracy when classifying activity intensity and estimating energy expenditure (EE) when data from both monitors were integrated [27]. However, these innovative data integration methods were suited to laboratory based research and not free-living conditions in which activities can be measured over multiple days. There is a need to develop a measurement method that can provide information on waking hours as well as posture and activity intensity under free-living conditions to better understand how SB, as defined by the SBRN, impacts on health outcomes.

The aim of this study was to develop a novel integrative procedure to combine data from the SWA and AP to identify and quantify SB based on both posture and activity intensity during waking hours.

Sixty-three female participants, aged 37.1 (13.6) years with a BMI of 29.6 (4.7) kg/m², had ≥5 days (including at least one weekend day) of valid SWA and AP data. Average wear time of the SWA was 23.6 (0.3) hours/day (98.4% of possible wear time) and average wear period was 6.5 (0.7) days. Participants were sedentary (excluding sleep) for an average of 11.7 (1.6) hours/day (70.7% of waking hours), 10.2 (1.8) hours/day (61.2% of waking hours) and 9.1 (1.7) hours/day (54.8% of waking day) determined by the SED^SWA, SED^AP and SED^INT methods, respectively (see figure 2).

The associations between measures of free-living SB are displayed in Table 1. All three measures of sedentary time were significantly positively correlated. The weakest association was between SED^SWA and SED^AP [p = .003], followed by SED^SWA and SED^INT [p < .001] and the strongest association was between SED^AP and SED^INT [p < .001].

There was a significant difference between average daily sedentary time determined by the different measurement methods [F(1.18, 73.15) = 104.70, p < .001]. Post-hoc tests using the Bonferroni correction revealed all three methods were significantly different from each other [p < .001]. SED^SWA recorded the most sedentary time, followed by SED^AP, and the least amount of sedentary time was recorded using the SED^INT method (see figure 2). The three different methods also produced significantly different measures of sedentary time from each other when weekdays [F(1.19, 73.52) = 91.67, p < .001] and weekend days [F(1.24, 76.74) = 100.75, p < .001] were analysed separately. Post-hoc tests using the Bonferroni correction revealed all three methods produced significantly different measures of sedentary time on weekdays and weekend days [p < .001]. SED^SWA recorded the most sedentary time, followed by SED^AP, and the least amount of sedentary time was recorded using the SED^INT method during weekdays and weekend days. Paired sample t-tests revealed the amount of sedentary time accumulated on weekdays compared with weekend days did not differ significantly when measured using any of the measurement methods [p = .04–.25].

Sedentary time accumulation measured using the three measurement methods is displayed in fig. 3. As the duration of the sedentary bout categories increases so too did the amount of sedentary time accumulated in that category. A similar amount of time was accumulated in the shorter sedentary periods for all three methods. SED^SWA gave the highest measure of total sedentary time and this is reflected in the longest sedentary bout category with SED^SWA registering more time in the >40 min category compared with SED^AP an SED^INT. When posture is included in the sedentary time measurement method (SED^AP and SED^INT) less sedentary time is accumulated in the longest bout category.

The difference in SED^SWA and SED^AP determined sedentary time were calculated for each participant to identify whether classification of SB based on either activity intensity or posture alone (during waking hours) resulted in systematically different estimates of sedentary time for the same measurement period. Figure 4 displays the individual variation in the difference in sedentary time estimated based on activity intensity (SWA) or posture (AP). The figure shows that 11 of the participants accumulated more sedentary time when it was measured based on posture, whereas 52 accumulated more sedentary time when it was measured based on activity intensity. Figure 5 displays a series of Bland-Altman plots showing the degree of systematic bias and limits of agreement between all measures of sedentary time. The largest average difference (bias) in measures of sedentary time was between SED^SWA and SED^INT (158.7 min/day) and the smallest difference in sedentary time was between SED^AP and SED^INT (63.8 min/day). Panel B and C of fig. 5 show that SED^SWA and SED^AP consistently provide higher estimates of sedentary time when compared with SED^INT. However, panel A shows that in some individuals SED^AP provides a greater estimate of sedentary time and in others SED^SWA is greater.

The primary aim of the current study was to develop a method to integrate information from two validated activity monitors on both posture and activity intensity during waking hours, the three key dimensions of SB as defined by the SBRN [7].This study demonstrates it is possible to identify free-living sedentary time during waking hours based on posture and activity intensity alone, and in combination using two validated activity monitors and a novel integrative procedure. At present, there is no single field-based measurement device that accurately captures both the activity intensity and postural element of SB during waking hours. The development of a device that can capture these dimensions of SB has been highlighted as a research priority [31]. Until such a device is available, integrating information from multiple activity monitors, such as the method presented in the current paper, may overcome SB measurement limitations. Recently, Ellingson et al. combined postural information from the AP with information on activity intensity from the ActiGraph and found that estimates of activity intensity during sedentary and light activities were more accurate than estimations based on ActiGraph data alone [27]. Although this research demonstrates a methodological advancement in the measurement of SB, its utility in free-living conditions is unknown. The integrative procedure presented in the current paper represents a feasible method for all three dimensions of SB to be measured under free-living conditions.

According to the SBRN SB refers to “any waking behaviour characterized by an energy expenditure ≤1.5 METs while in a sitting or reclining posture” [7]. Although this definition has been widely accepted, other definitions are still being used, for example, Pate et al. define it as “activities that involve energy expenditure at the level of 1.0-1.5 METs” [29]. Whilst both definitions include low activity intensity, there is still debate whether SB should encompass activities at ≤1.5 METs whilst standing [31]. The recently published terminology consensus from the SBRN acknowledges that standing can be either active (>2 METs) or passive (<2 METs), however, it remains unclear whether passive standing should be classified as an active or sedentary behaviour (particularly when the intensity is <1.5 METs) [8]. The method described in the current paper allowed us to investigate the extent to which estimates of SB differed based on the definition and measurement technique used. There was a significant difference between average daily sedentary time determined by the different measurement methods. SED^SWA recorded the most sedentary time, followed by the SED^AP, and the least amount of sedentary time was recorded by the SED^INT method. Furthermore, more sedentary time was accumulated in prolonged bouts when determined by the SED^SWA method. The volume of SB in the current study was large, but not dissimilar to other studies and sedentary time did not differ between weekdays and weekend days [16, 32, 33]. Participants were sedentary for between 9.1 h/day (54.8% of waking hours) and 11.7 h/day (70.7% of waking hours) depending on the measurement criteria. It is understandable that SED^SWA reflects a larger amount of SB since it would include instances of standing (as well as sitting) but with a MET of <1.5. The difference in sedentary time when estimated by different measurement methods carries important implications for the association between SB and health outcomes. Literature reporting the relationship between SB and health outcomes may arrive at different conclusions depending on the measurement device being used and therefore the component of SB being measured. Indeed, previous research has identified differences in associations between SB and cardio metabolic risk when measuring sedentary time subjectively and objectively [34]. However, differences in associations between health outcomes and different measures of objectively determined sedentary time have not been examined. Research to determine the specific properties of SB which relate to diminished health is a key priority [31, 35]. This will inform researcher’s decisions on the most appropriate device to use for their specific research question. Furthermore, existing recommendations to reduce sitting [36] could be updated and refined to provide more specific information about what activity should displace sitting to benefit health.

A strength of the current study is that it shows clearly that the amount of sedentary time identified varies with the particular device used; in turn, this has implications for associations with health and disease endpoints. A limitation of this study is that the sample is all female university employees and therefore, the results of this study may only apply to a similar demographic. Furthermore, the epochs were collapsed from 15 s to 60 s for the AP data but we do not feel that this resulted in a significant change in the degree of resolution.

In this study, we have demonstrated a procedure to obtain a measure of free-living SB based on both activity intensity and posture during waking hours by integrating data from two validated activity monitors. This platform is flexible and can be expanded as new tracking technologies become available. Measures of sedentary time using different objective measurement techniques are not measuring the same components of SB. Indeed, the three measures of SB in the current study differed significantly. Determining whether the postural element of SB contributes to negative health outcomes attributed to SB or whether sitting is a marker for low EE remains a research priority. The novel data integration and processing procedures presented in this study represents an opportunity to investigate whether different components of SB are more strongly related to health outcomes than others.