Differing Context-Specic Sedentary Behaviors in Australian Adults with Higher and Lower Diabetes Risk

Time spent sitting in different settings can pose different risks to health. In Australian adults either with higher and lower diabetes risk, this study examined the differing compositions of self-reported sitting time accumulated in ve contexts (occupational, transport, TV viewing, leisure computer-use and other). Participants (n = 3927; 60 ± 11 years; 45% male) were from the 2011–2012 assessment wave of the AusDiab study. The relative compositions of self-reported context-specic sedentary behaviors to total sitting time were compared between those with and without previously undiagnosed dysglycaemia (impaired fasting glucose, impaired glucose tolerance or newly diagnosed T2D), in working (323 with, 1646 without; 5-part composition) and non-working (433, 1525; 4-part composition) adults. For working adults, compared to those without dysglycaemia, those with undiagnosed dysglycemia spent the same proportion of time sitting at work, 3% more time sitting during transport, 9% more time sitting watching TV, 2% less time sitting using a computer for leisure, and 9% less time sitting during other activities. For non-working adults, compared to those without, those with dysglycemia spent 26% less time sitting during transport, 9% more time sitting while watching TV, 29% less time sitting using a computer for leisure, and 5% more time sitting during other activities. In addition to addressing overall sitting time, those with higher levels of diabetes risk may benet from targeted reductions in context-specic sedentary behaviors, particularly TV viewing time. These ndings also provide a case in point with potential relevance for other health problems associated with sedentary behavior.


Introduction
High volumes of time spent in sedentary behaviors are detrimentally associated with type 2 diabetes (T2D) incidence and with cardiometabolic risk biomarkers (1)(2)(3)(4)(5). Time spent sitting occurs across multiple context-speci c behaviors, including TV viewing, at work, and during transport. Sitting in these different contexts may not be equally detrimental to health outcomes. For example, in Australian adults, accumulating sitting while watching TV or at the computer (during leisure) was found to be adversely associated with cardiometabolic risk, whereas occupational sitting was found to be relatively less harmful (6).
Excessive sitting is a particular concern for those at risk of or living with, T2D. Landmark lifestyle intervention trials have demonstrated the potential for effective prevention of T2D (7,8). Importantly, these trials have been directed at those with previously undiagnosed conditions, particularly re ecting impaired blood glucose control. This has public health relevance, as in developed countries such as Australia, some 24% of the adult population have undiagnosed dysglycemia (impaired fasting glucose, impaired glucose tolerance and undiagnosed diabetes) (9). The main lifestyle issues targeted in the prevention of T2D are healthier bodyweight, improved diet, and achieving su cient leisure time moderateto-vigorous physical activity (MVPA) levels. However, in recognition that sitting time is signi cantly higher (~ 1hr/d) in those with dysglycaemia compared to those with normal glucose metabolism (10), it has become a priority to address sedentary behavior (11). Despite this, little is known about the pro le of sedentary behaviors which constitute overall sitting time in those with underlying metabolic impairment, Page 3/18 and whether these differ from that of healthy individuals. These pro les are likely to differ greatly between workers and non-workers, as occupational sitting can be the largest contributor to daily sitting time on workdays (12).
Treating time spent in context-speci c sedentary behaviors as being bound within a xed composition (as distinct from treating them as absolute or stand-alone attributes) allows information to be expressed relative to each behavior within the set time frame. This accounts for inter-dependence between the different sedentary behaviors -changing time in one behavior necessitates change in one or more other behavior (13)(14)(15)(16)(17). In the context of diabetes risk, compositional methods can identify how variations in the speci c sedentary behaviors that make up overall sitting time might differ between those who are at higher or lower diabetes risk.
We compared the relative contributions to total sitting time of ve separate contexts of sedentary behavior (occupational, transport, TV-viewing, leisure time computer-use and other sitting) in a large sample of middle aged and older, working and non-working, adults with and without previously undiagnosed dysglycemia.

Participants and Procedures
The Australian Diabetes, Obesity and Lifestyle study (AusDiab) is a national longitudinal study, originally designed to examine the prevalence of diabetes and its predictors/precursors in a sample of Australian adults from urban areas and regional cities. Details of the data collection methods and response rates have been described previously (9,18). Brie y, the baseline survey collected data from nationally representative sample of 11,247 adults in 1999-2000. Follow up studies were conducted in 2004-05 (AusDiab2) and 2011-12 (AusDiab3). In the 2011-12 data collection, new survey questions on context speci c sedentary behaviors were added. The present study uses cross-sectional data from participants in AusDiab3 who attended testing sites for a biomedical examination (n = 4,614). After excluding those who: were pregnant (n = 6); had known type 2 diabetes (n = 446); known CVD (n = 189); had implausible total sitting time (greater than 18 hours per day) (n = 10) or had missing self-reported sitting time data (n = 37); the sample size for analysis was 3,927. The sample was strati ed by working status, with working adults being those who reported working either full time or part time, "in a paid position, including selfemployment, or as a volunteer" (n = 1,969). Non-workers included those who reported being retired, not working (but not retired), studying, or other (n = 1,958). The Alfred Hospital Ethics Committee approved the study and written informed consent was obtained from all participants.

Measurement and Data Management
Participants in AusDiab3 attended a biomedical examination which assessed a variety of biomarkers of cardiometabolic health. After an overnight fast (minimum 10 h) they attended a local testing center to complete a standard 75 g 2 h oral glucose tolerance test. Glycemic status was classi ed according to and who had fasting plasma glucose > 7.0 mmol/L or 2 h post-load glucose > 11.1 mmol/L were classi ed as having previously undiagnosed diabetes. For those without known diabetes, fasting plasma glucose > 7.0 mmol/L and 2 h post-load glucose > 7.8 but < 11.1 mmol/L indicated impaired glucose tolerance; impaired fasting glucose was de ned as fasting plasma glucose > 6.1 and < 7.0 mmol/L, with 2 h post-load glucose < 7.8 mmol/L; and normal glucose tolerance was de ned as fasting plasma glucose < 6.1 mmol/L and 2 h post-load glucose < 7.8 mmol/L, without prior-diagnosed diabetes.
Sitting time Participants reported sitting time over the previous seven days, separately for weekdays and weekends, across ve contexts (occupational, transport, television viewing, leisure time computer use and "other" sitting). The measurement properties of the items used have been previously reported; and, the sum of these ve contexts has been validated against total sitting time measured by activPAL (r = 0.46 [95% CI: 0.40, 0.52]) (20). "Other" sitting, while di cult to interpret due to the lack of clarity regarding which behaviors are captured, was included in the analysis as it represented a signi cant portion of total sitting time. Average daily sitting time (h/day) for each of the ve contexts [(weekday/5 + weekend/2)/7)] was then calculated. Total sitting time was calculated as the sum of all ve forms of sitting (including "other") in working adults and for the four relevant forms of sitting (excluding occupational sitting) in nonworking adults.

Social and behavioral attributes:
Attributes used to describe the general health and demographic pro le of the different sub-samples were determined from interviewer-administered questions and include age, gender, parental history of diabetes and educational attainment (high school or less, technical/vocational, bachelor's degree or higher). Leisure-time physical activity (LTPA; h/day), including walking for recreation or transport, and MVPA was assessed for the previous week using the Active Australia Survey Questionnaire (Australian Institute for Health and Welfare, Canberra, Australia). Validation studies of the Active Australia questionnaire have reported good reliability and acceptable validity (21).

Statistical Analyses and Data Handling
Properties of compositional data as they relate to sitting time Previous studies have applied compositional approaches using the 24 h waking day as the composite whole. Another approach of relevance to our study is to apply proportions (i.e. the sum of 1) to the composite of behaviors (22,23). In the context of sitting time, this time-use can be viewed as a combination of mutually exclusive contexts in which sitting is accumulated, such as at work, during transport, while watching TV or using a computer for leisure. When individual data components are treated as compositional, the resulting information is considered to be relative as opposed to absolute; i.e. any associations evident from one component is meaningful only by reference to the other components which make up the composite whole (22).

Calculation of geometric means and isometric log ratios
Appropriate to the analytic method (22) geometric means were calculated. Subsequently, these geometric means were used to calculate the log-ratio differences in sitting times between those with and without undiagnosed dysglycemia (i.e. geometric mean occupational sitting in those with dysglycemia divided by the geometric mean occupational sitting for those with normal glucose metabolism). A positive value indicates, in this case, that those with dysglycemia spent more time sitting in that particular context than those with normal glucose metabolism, and conversely if the value is negative, less time is spent sitting. If the value is zero, the sitting time is equal across the two groups. Isometric log ratios (ILRs) were calculated for each context of sitting for each of the two groups (dysglycemia and normal glucose metabolism). The presence of zeros within compositional data presents issues as log-ratios cannot be applied to zero values (24). Given that in our case, zeros were expected (e.g. if someone reported they spent 0 h watching TV), essential zeros were assigned a very small non-zero value of 1 min/day, which allows for log transformation.

Tests for statistical signi cance
Standard tests (t-tests and chi-square tests) to examine differences in descriptive characteristics of the two groups were used. Hotelling's test (multivariate test) was used to determine whether there was a difference in any of the ILRs of the ve contexts of sitting between those with and without undiagnosed dysglycemia. Bonferroni adjusted p-values were used to determine the signi cance of the test for both working and non-working groups. This test indicated whether the compositions differed overall between groups, but not which individual components differed. To examine this, we developed bootstrap percentile con dence intervals for log ratio differences for each separate component of sitting. If the con dence interval crossed zero, this indicated that there was no statistically signi cant difference between groups with respect to this component (23). All analyses were conducted using Stata (version 14.0 Stata Corporation, College Station,TX, USA) and RStudio (version 1.2.5033 2009-2019 RStudio, Inc.) software using the compositions (25), Hotellings (26), and boot (27) packages. Finally, though the survey used a complex sample design, we did not apply survey weights to the current analysis.

Attributes of Participants
Of the 3,927 study participants; 756 had previously undiagnosed dysglycemia. Those with undiagnosed dysglycemia were older (mean ± SD, 64 ± 11 years) compared to those with normal glucose metabolism (59 ± 11 years). Less were women (48% compared to 59%), and less reporting working either full or part time (44% compared to 53%). Those with undiagnosed dysglycemia spent less time in LTPA (0.68 ± 0.75 h/day) compared to those with normal glucose metabolism (0.90 ± 0.90 h/day, P < 0.001), and less reported meeting at least the recommended 150 minutes of MVPA per week (57% compared to 67%). The majority of those with undiagnosed dysglycemia were categorized as either overweight or obese (82%), compared to 64% of those with normal glucose metabolism. Participant attributes by working status are reported in Table 1. Of the 1969 who were working, 323 had previously undiagnosed dysglycemia, as did 433 of the 1958 non-working adults. Compositional geometric means (calculated using Aitchison geometry) are presented as proportions of the total time sitting. Working adults Figure 1 shows the exponentiated log ratio contrasts (% differences) in each context of sitting time between those without and with dysglycemia, for working adults. Compared to those with normal glucose metabolism, those with undiagnosed dysglycemia spent similar proportions of time sitting while at work, 3% more time sitting during transport (95% CI -10, 16%), 9% more time sitting while watching TV (95% CI -5, 23%), 2% less time sitting while using a computer for leisure (95% CI -21, 20%), and 9% less time sitting for other activities (95% CI -21, 3%). The proportion of time spent in speci c sedentary behaviors was not signi cantly different between those with and without previously undiagnosed dysglycemia (Hotelling's test; P = 0.46).
The variability of the composition of working adults with and without dysglycemia is described in Supplementary Tables 1 and 2 respectively. The variability or proportionality between pairs of components can be understood as indicators of the interchangeability of the speci c components. In general, for working adults, the lowest levels of co-dependency were observed between sitting while at work and leisure-time computer use, meaning these behaviors were related to each other to a lesser extent. The highest level of co-dependency was observed between other sitting and transport sitting, meaning that these two behaviors had the greatest level of interchangeability.

Non-working adults
Non-workers (n = 1958) are described with only four contexts of sitting (i.e., any reported work sitting was ignored). Hotelling's test reported a p-value of P < 0.001, indicating that the proportion of time spent in speci c sedentary behaviors was signi cantly different between those with and without previously undiagnosed dysglycemia. Figure 2 shows the exponentiated log ratio contrasts (% differences) in each context of sitting time between those without and with dysglycemia in non-working adults. Compared to those with normal glucose metabolism, non-working adults with undiagnosed dysglycemia spent 26% less time sitting during transport (95% CI -35, -16%), 9% more time sitting while watching TV (95% CI -0.5, 18%), 29% less time sitting while using a computer for leisure (95% CI -43, -11%), and 5% more time sitting during other activities (95% CI -4, 14%).
For non-working adults with and without undiagnosed dysglycemia, the lowest levels of co-dependence occurred between sitting while watching TV and sitting while using a computer for leisure (Supplementary Tables 3 & 4). For adults without dysglycemia, the highest levels of co-dependence or interchangeability were observed between other sitting and sitting during transport (2.13). For those with dysglycemia, the highest level of co-dependency was observed between other sitting and TV (2.44)

Discussion
In this large sample of Australian adults, we found that compared to those with normal glucose metabolism, working adults with undiagnosed dysglycemia spent the same amount of time sitting at work, more time sitting during transport (3%) and watching TV (9%), and less time sitting using computer for leisure (2%) and during other activities (9%). In non-working adults, those with undiagnosed dysglycemia spent more time sitting while watching TV (9%), and during other activities (5%) and less time sitting during transport (26%) and using a computer for leisure (26%) compared to those with normal glucose metabolism.
To date, no studies have described the differences in time spent in interrelated contexts of sedentary behavior, between those with and without undiagnosed dysglycemia, in both working and non-working adults. Since previous experimental evidence suggests that the impacts of prolonged sitting on postprandial glycemia are greater in those who have a degree of underlying dysglycemia (28,29), an improved understanding of the relative contribution to overall sitting of each respective sedentary behavior is clinically relevant. Irrespective of whether working or not, the observation that those with dysglycemia spent a greater proportion of time sitting watching TV, although not statistically signi cant, provides further support for the recent calls within clinical practice recommendations to target the reduction of speci c sedentary behaviors for the prevention of T2D (11).
Extensive evidence suggests increased TV viewing time is associated with poorer cardiometabolic health (6,(30)(31)(32)(33) and increased risk of CVD events and mortality (34)(35)(36). In comparison to other contexts of sedentary behavior, TV viewing can often occur concurrently with other detrimental health behaviors such as; poorer dietary patterns (37,38); sitting for longer, uninterrupted periods of time; and often occurs in the evening following a large meal (39). Indeed, a recent study reported that overweight/obese adults had a lower postprandial glycemic response when sitting was interrupted during advertisement breaks on TV after an evening meal (40). Some contexts of sitting, such as TV viewing, may be more strongly linked with adverse outcomes due to the nature of the behavior itself -more habitual, less variable and easily recalled -meaning they may be more accurately measured than others (41). Time spent watching TV is also associated with lower socioeconomic status and other factors which are known to strongly in uence health outcomes (30). Regardless, our results indicate that reducing and interrupting sitting while watching TV is an important target area to improve the cardiometabolic health of those with dysglycemia.
The contribution from occupational sitting to total sitting time was similar in both groups. This is not unexpected, since sitting while at work, in most cases, is less discretionary than sitting in other contexts (due to the constraints of the workplace and set working hours). Workplaces have been identi ed as environments which are high risk for excessive prolonged sitting. Occupational sitting contributes a high proportion of total sitting time (41), and intervention evidence supports reducing and interrupting occupational sitting time to improve markers of cardiometabolic health (42,43). Thus, potential future initiatives within occupational health and safety regulations may contribute to improved health outcomes through reductions to overall sitting time, particularly in working adults. The current data does not allow insight into the discretionary nature of occupational sitting (i.e. whether work had to be performed seated) and as such is a limitation.
Those with undiagnosed dysglycemia spent less time sitting while using a computer for leisure than those with normal glucose metabolism (2% in workers, 29% in non-workers). A potential strategy for reducing time spent in what may be deemed, more "metabolically harmful" sedentary behaviors, for example TV viewing (6,32,34,37,(44)(45)(46), aside from promoting movement, could be the reallocation of time from non-active to active sedentary behaviors. A recent study found non-active sedentary behaviors (e.g. sitting on a chair or reclining) were positively correlated with BMI, body fat % and insulin, whereas active sitting (e.g. sitting while typing on a computer) was negatively associated with the same outcomes (45). Additionally, evidence has emerged on the potential impacts of mentally-active compared to passive sedentary behaviors on mental health outcomes. Isotemporal substitution analyses reported reduced odds of depressive symptoms and clinician-diagnosed major depressive disorders when 30 minutes of passive sedentary behaviors (watching TV, listening to music) were substituted with 30 minutes of mentally-active sedentary behaviors (o ce work, sitting in a meeting, knitting) (47). While care should be taken so as not to promote accumulation of large volumes of sitting time in any context, it is possible that the more mentally-active behavior of leisure-time computer use provides an appropriate alternative to passive sedentary behaviors such as TV viewing. Future studies which combine objective measures of contextualized sitting with longitudinal study designs are needed to further explore the temporality of these ndings. Additionally, the current data were collected between 2011-2012; the changing landscape of TV viewing will need to be considered with the rise of web-based streaming services.
Sitting accumulated in 'other' activities contributed greatly to overall sitting time and differed between glycemic groups. In working adults, those with dysglycemia spent 9% less time sitting during other activities than those with normal glucose metabolism. Conversely, in non-working adults, those with dysglycemia spent 5% more time sitting in other activities than those without. The survey question which collected this data was phrased; "…this could include sitting for reading or hobbies, socializing with friends or family including time on the telephone eating meals; or listening to music." Presumably, a combination of both active and non-active sedentary behaviors are captured within this question, which makes interpretation di cult. Future studies could direct attention to expanding the self-reporting of these various behaviors to better understand their unique contributions to overall sitting time.
Categorizing sedentary behaviors into non-active and active sitting (46) is one method of addressing this issue that future studies could employ.
Limitations include the cross-sectional nature of the data; one week of self-reported, recalled data gives limited detail around time of day sitting was accumulated or bout duration, and may not be representative of broader sitting patterns. The large contribution to overall sitting time from 'other' sitting behaviors increases the risk for biases due to residual confounding, meaning that important contexts of sitting are not captured. The level of detail obtained from the data does not provide insight on the underlying biological mechanisms which may contribute to the different sedentary behavior pro les observed. In addition, the sample was not population representative and may be prone to selection bias. Speci cally, the analysis included a sample healthier than the general population, within the third wave of a cohort which was subject to some loss to follow up (48). Since the AusDiab study is limited to middleaged/older adults, the ndings are not generalizable to younger population age groups.
There are several strengths to our study, including the large and diverse sample of Australian adults incorporating measures of both fasting and post-challenge plasma glucose levels to identify those with previously undiagnosed dysglycemia. Furthermore, the context-speci c sitting measurement was a novel aspect; unique to this wave of the AusDiab study. We also employed a compositional data analysis approach, which respects the inherently relative nature of time-use data by expressing the information as a set of log ratios. Most prior analyses of time-use data use statistical methods which focus on the absolute information in the data.

Conclusions And Recommendations For Future Research
In this large and diverse sample of Australian adults, those at higher risk of developing diabetes, or with newly diagnosed diabetes, spent a greater proportion of total sitting time watching TV, a similar proportion of the time sitting while at work (in working adults), and less time sitting to use a computer for leisure, than those at lower risk. Overall, our ndings point to the potential bene t of public health initiatives that speci cally target the contexts of sitting that we have identi ed.  between those with and without previously undiagnosed dysglycemia in working adults. Contrasts are expressed as percentages with error bars denoting bootstrap 95% Con dence Intervals. *A positive gure denotes that those with previously undiagnosed dysglycemia spend more time sitting in that context compared to those with normal glucose metabolism; a negative gure indicates less time spent sitting in those with dysglycemia compared to without. Hotelling's test P = 0.46. Statistical signi cance was set at P < 0.01 for working adults (5-part composition). LTCU: Leisure-time computer use. are expressed as percentages with error bars denoting bootstrap 95% Con dence Intervals. *A positive gure denotes that those with previously undiagnosed dysglycemia spend more time sitting in that context compared to those with normal glucose metabolism; a negative gure indicates less time spent sitting in those with dysglycemia compared to without. Hotelling's test P < 0.001. Statistical signi cance was set at P < 0.0125 for non-working adults (4-part composition). LTCU: Leisure-time computer use.