1 Feasibility of measuring sedentary time with thigh worn accelerometry, and sociodemographic correlates: the 1970 British Cohort Study

In large scale cohort studies sedentary behaviour has been routinely measured using self-report or devices that apply count-based threshold. We employed a gold standard postural allocation technique using thigh inclination and acceleration to capture free-living sedentary behaviour . Participants (n=5,346, aged 46.8 ± 0.7 yrs) from the 1970 British Cohort study (U.K.) were fitted with a water-proofed thigh mounted accelerometer device (activPAL3 micro) worn continuously over 7 days, collected 2016 – 2018. Useable data were retrieved in 83.0% of the devices fitted, with 79.6% of the sample recording at least 6 full days of wear (at least 10 waking hours). Total daily sitting time (average 9.5±2.0 hr/d men and 9.0±2.0 hr/d women) accounted for 59.4% and 57.3% of waking hours in men and women respectively; 73.8% of the sample recorded ≥8hr/d of sitting. Sitting in prolonged bouts of more than 60 continuous minutes accounted for 25.3 % and 24.4% of total daily sitting in men and women respectively. In mutually adjusted models, male sex, underweight and obesity, education, poor self-rated health, TV-viewing time and a sitting occupation were associated with higher device measured sitting times. Thigh worn accelerometry was feasibly deployed and should be considered for larger scale national surveys.

Sedentary behaviour has been recognised as a risk factor for health (1)(2)(3)(4)(5). However, to date, the evidence generated from large scale population cohorts has relied on self-report with known biases (6).
Data, for example, from National Health and Nutrition Examination Survey suggested total self-reported hours per day of sitting time increased in adults from 5.5 to 6.4 hr/d during 2007 to 2016 (7) and showed that 25.7% of US adults reported more than 8 hours of total sitting time per day (8). Given the uncertainty regarding the validity of data on self-reported sitting time, it is difficult to estimate true population norms for sitting (9,10).
Wearable devices are being increasingly used to assess free-living sedentary time (11,12), although most of the existing methods have applied criteria based on lack of movement or movement below a certain count threshold (13). The count-based threshold approach that applies cut points to classify movement intensity can lead to misclassification of low-intensity non-sedentary behaviors such as standing (14-16).
Thus, ideally sedentary time should be derived from a combination of both energy expenditure and postural elements (12).
In the present study we sought to use a thigh-mounted accelerometer device (activPAL3 micro; PAL Technologies Ltd., Glasgow, UK) to assess sedentary behavior (12). The device uses derived information about thigh inclination and acceleration to estimate body posture (i.e., sitting/lying and upright) and transition between these postures, stepping, and stepping speed (cadence). Importantly, this technique overcomes concerns raised (14-16) about the face validity of wrist and hip worn monitors to accurately capture postural sitting. ActivPAL was validated for measuring free-living sedentary behaviour against direct observation using an automated camera (17). Although the thigh-mounted accelerometer has been used in relatively small convenience samples thus far (12,18), the present study is the first to use of deploying a potentially more invasive device attached to the skin (compared to devices worn around the wrist or attached to waist belts) so that this methodology could be considered for larger scale national surveys in the future.
The aim of this study was to explore the feasibility of using a gold standard postural allocation technique to capture free-living sedentary behaviour in a large nationally representative cohort study of middleaged British adults. We firstly report on rates of consent, and adherence to the wear protocol. The second aim was to examine sociodemographic and lifestyle correlates of free-living sitting.

Design and participants
The The study used a thigh-mounted accelerometer device (activPAL3 micro; PAL Technologies Ltd., Glasgow, UK) as previously described (12). We utilised a wear protocol previously adopted (18); Devices were programmed to sample at the default frequency of 20 Hz. The device was waterproofed and fitted by a trained nurse on the midline anterior aspect of the upper thigh as recommended by the manufacturer. Participants were requested to wear the device continuously for 7 days, including sleeping, bathing, swimming, and all physical activities. If the device fell off or was removed before the stated end date, participants were requested not to re-attach. Devices were returned via post. Data were processed using freely available software that has been previously validated (21

Statistical analyses
The distribution of activPAL variables were examined for normality and potential outliers. The activity data represents mean hours per day averaged over the number of days the device was worn. Extreme waking hours wear time (n=13; > 20hrs per day) was checked against sleep diaries and in the case of clear discrepancies (± 3 SD) outliers were removed. Acceptability of the device (in terms of consent to participate and number of days worn) was examined in relation to sociodemographic characteristics.
Total sitting time was categorised into tertiles (low: <8.4hr/day; medium: 8.4-10.1 hr/day; and high: >10.1 hr/day) and examined in relation to sociodemographic and lifestyle variables. We also derived data on bouts of uninterrupted sitting time lasting more than 60 min. The sociodemographic and lifestyle variables were selected a priori based on existing literature (23,24). Generalised linear models were used to examine associations between sociodemographic and lifestyle variables, and sitting time as a continuous dependent variable, making adjustment for waking hours wear time.

Results
Useable data were retrieved in 83.0% of the devices fitted (Web Figure 1). Participants declining to wear the device (11.8%) were more likely to be male, smokers, report poorer health, and be obese (Web Table 1). Reasons for declining to wear the device mainly included 'inconvenience' and 'going on holiday' (plans to travel by plane was an exclusion criterion), while relatively few had concerns over attachment of the device to the skin (Web Figure 1).
The final analytic sample comprised 5,346 men and women, (aged 46.8 ± 0.7 yrs). We observed high adherence to the wear protocol, 90.7% of the sample recorded at least 3 full days of device wear, 79.6% recorded 6 full days of wear, and 65.5% wore the device for the full 7 days. Compared to participants with higher wear adherence (> 3days), those with poor adherence (up to 3 days) were more likely to be male, smokers, report poorer health, obese, and non-degree educated (Web Table 2). Interestingly, those with poor adherence were more likely to have worn devices over the summer months.
Nevertheless, no differences were observed between the groups for average sitting time or activity.
Total daily sitting time was normally distributed (Web Figure 2) (average 9.5±2.0 hr/d men and 9.0±2.0 hr/d women) accounted for 59.4% and 57.3% of waking hours in men and women respectively (Table 1).
Overall, 73.8% of the sample recorded ≥8hr/d of sitting. Sitting in prolonged bouts of more than 60 continuous minutes accounted for 25.3 % and 24.4% of total daily sitting in men and women respectively (Web Figure 3).
Participants recording higher sitting times were more likely male, smokers, degree educated, obese, reported higher prevalence of poor health and disability, hypertension and diabetes (Table 2). There was a trend for higher prevalence of all self-reported sedentary behaviours in the highest device measured sitting group ( Total sitting time was socially patterned, higher in degree educated participants. This finding is likely to be partly driven by occupation in that professionals/managers are most likely to be desk bound at work. In contrast, however, previous data suggest lower social status groups report greater sedentary behaviour in leisure time such as TV viewing (23). Thus social patterning of sedentary behaviour is likely to be context specific (not simply volume), although our data suggest it is largely driven by occupational sitting time. The other correlates of device measured sitting time, such as obesity and health indicators, found in this study are also consistent with other studies on self-reported TV viewing (24).
The main strengths of this study are the nationally representative sample, and high adherence to the wear protocol with little data loss. Our wear protocol minimised the problems of non-wear as participants were requested not to re-attach their device if removed prematurely as used previously (18). We employed a novel algorithm to isolate valid waking wear time from sleep/non-wear that enabled large volumes of accelerometry data to be processed more efficiently. Wearable activity monitors are generally designed to be worn for no more than one week to minimise participant burden, which may not adequately reflect habitual behaviour.

Limitations
As is the case in most population studies, respondents that did not consent to wear a device tended to be less educated and report poorer health that may have introduced bias. Participants with greater compliance to wearing the device were also generally healthier although device wear characteristics did not appear to influence the amount of sitting or activity recorded. Since our study was conducted on middle aged adults, before the onset of functional decline, it may not be representative of the wider population. The algorithm was not designed to distinguish physiological sleep periods, and in the absence of a true 'gold standard' we were unable to more fully explore sleep in this study. Data were cross-sectional and we cannot infer directionality of associations presented between sedentary and demographic characteristics. Our measure of sedentary behavior was an average of weekday and weekend activity and participants contributed differential wear time to this average. Nevertheless, average sitting times of participants with full 7 days wear were identical to data from the whole cohort.
In summary, thigh worn accelerometry can be feasibly deployed in large scale population cohorts.
Future studies should be mindful of potential selection biases when using wearable technology, both in terms of consent to participate and wear compliance.