Background Although certain types of sedentary behaviour have been linked to metabolic risk, prospective studies describing the links between sitting with incident diabetes are scarce and often do not account for baseline adiposity. We investigate the associations between context-specific sitting and incident diabetes in a cohort of mid-aged to older British civil servants.
Methods Using data from the Whitehall II study (n=4811), Cox proportional hazards models (adjusted for age, sex, ethnicity, employment grade, smoking, alcohol intake, fruit and vegetable consumption, self-rated health, physical functioning, walking and moderate-to-vigorous physical activity, and body mass index (BMI)) were fitted to examine associations between total sitting and context–specific sitting time (work, television (TV), non-TV leisure time sitting at home) at phase 5 (1997–1999) and fasting glucose-defined incident diabetes up to 2011.
Results Total sitting (HR of the top compared with the bottom group: 1.26; 95% CI 1.00 to 1.62; p=0.01) and TV sitting (1.33; 95% CI 1.03 to 1.88; p=0.05) showed associations with incident diabetes; once BMI was included in the model these associations were attenuated for both total sitting (1.19; 95% CI 0.92 to 1.55; p=0.22) and TV sitting (1.31; 95% CI 0.96 to 1.76; p=0.14).
Conclusion We found limited evidence linking sitting and incident diabetes over 13 years in this occupational cohort of civil servants.
- Physical activity
- Sitting time
Statistics from Altmetric.com
Sedentary behaviour (SB) comprises a set of waking time activities that are characterised by an energy expenditure of ≤1.5 metabolic equivalents (MET) in a sitting or reclining posture.1 Sitting is a ubiquitous behaviour in today’s world and has been linked to broad outcomes such as all-cause mortality.2–4 A growing body of research has examined the cardiometabolic consequences of sitting in both population3 4 and laboratory5 settings. Data on sitting and risk for incident diabetes are scarce. A meta-analysis of 10 cross-sectional and prospective epidemiological studies concluded that the population groups with the greatest levels of SB time had 112% higher risk for type 2 diabetes compared with the lowest SB time groups.6 All included studies used television (TV) viewing as a proxy of total SB7–10, and only one7 adjusted for baseline adiposity. Adiposity is associated with both type 2 diabetes and SB time,11–13 and as such, it may be a confounder that needs to be accounted for. A more recent meta-analysis on SB and cardiometabolic disease events and mortality3 also showed that, across all outcomes, the most consistent associations were seen for risk for type 2 diabetes (>90% increase in risk). All five included studies also had TV viewing as the exposure.3 However, it is unclear whether these findings are driven by the sitting that TV viewing involves per se. TV time is a poor indicator of total SB14 15 and sitting time16 that is confounded by multiple aspects of socioeconomic circumstances,17 dietary factors3 6 and mental health.18 Such a breadth of confounding has not been fully accounted for by studies in the field. Beyond TV time, a recent study of total sitting and incident diabetes in a sample of adults from Denmark found associations only among the physically inactive and the obese groups.19 Sitting can occur in many different contexts (eg, work, leisure time and transportation), and there is a limited number of cohort studies9 20 21 that examined context-specific associations with diabetes risk (all US-based). TV time only,9 20 TV and total leisure-time sitting21 but not work-related sitting9 20 were associated with diabetes; on one occasion, these associations were eliminated once baseline body mass index (BMI) was taken into account.20
The aim of this study was to examine the associations between context-specific sitting with incident diabetes among middle-aged and older British civil servants over a 13-year period. We sought to highlight the role of adiposity by presenting these associations with and without adjustment for baseline BMI.
Participants and study background
The Whitehall II study was established in 1985 to examine the biological mechanisms that account for observed social inequalities in cardiovascular disease (CVD) and diabetes.22 The sample included in this study comprised 4811 individuals (3501 men and 1310 women) from clerical and office support grades, middle-ranking executive grades and senior administrative grades. Baseline examination (phase 1: 1985–1988) involved a questionnaire and a clinical examination, and subsequent measurement phases have alternated between postal questionnaire alone and postal questionnaire accompanied by a clinical examination. Ethical approval was granted by the University College London Medical School Committee. Informed consent was obtained at baseline and renewed at each contact. The detailed measures of physical activity (PA) and sitting included in this report were undertaken during the fifth phase of data collection between 1997 and 1999 with follow-up for incident diabetes until December 2011.
Measurement of sitting behaviours
The questionnaire included items related to both occupational and leisure time sitting behaviours.22 Participants were asked ‘On average, how many hours per week do you spend sitting at work, driving or commuting?’ and ‘sitting at home, for example, watching TV, sewing, at a desk’, and responded by selecting one of eight time categories (none, 1 hour, 2–5, 6–10, 11–20, 21–30, 31–40 and >40 hours). For sitting at home, participants were given an open-text response option to specify two types of sitting and then selected a time category for each. Using the midpoint of each time category (exactly 40 hours was used to represent the >40-hour category), six indicators of sitting expressed as hours per week were computed: 1) work-related sitting time, 2) TV time, 3) non-TV leisure sitting time at home, 4) total leisure time sitting at home (sum of 2 and 3 above), 5) total sitting time (sum of 1 and 3 above) and 6) non-TV total sitting time (sum of 1 and 3 above). Five of these items (1–5) have been used previously,9 11 22 and although there is currently no objective criterion measure of context-specific sitting, these questionnaire items have demonstrated concurrent validity with past weeks’ recall questionnaires (Pearson's r=0.44) and activity diaries (r=0.41).23
Outcomes included incident diabetes up to December 2011. As previously described,24 blood glucose was measured using the glucose oxidase method. Incident cases of diabetes were identified by fasting blood glucose concentration (≥7.0 mmol/L) according to the 2006 WHO classification.25
Height (metres) and weight (kilograms) were measured at the clinical examinations. BMI was computed by dividing the squared height by body weight. Sociodemographic covariates included age, gender, ethnicity (Caucasian vs non-Caucasian ), employment grade (a comprehensive marker of socioeconomic circumstance related to salary, level of responsibility and social status), smoking status (current smoker, previous smoker or never a smoker), alcohol consumption (units per week), frequency of fruit and vegetable consumption, self-rated health (excellent, very good, good, fair or poor) and physical functioning score (continuous) using the 36-item Short-Form health survey scale.26 PA was measured using a 20-item modified version of the previously validated Minnesota Leisure-Time Physical Activity Questionnaire which enquired about occupational, domestic and leisure time physical activities. These questions have been shown to have acceptable criterion validity against accelerometry27 and have demonstrated excellent predictive validity for mortality28 in the Whitehall II study. Physical activities were classified by MET,29 with moderate-intensity activities ranging from 3 to 5.9 MET and vigorous-intensity activities in the range of 6 MET or greater. The energy cost of walking is dependent on walking pace and could not be determined from the phase 5 questionnaire, so walking time did not contribute to the moderate-to-vigorous physical activity (MVPA) and was entered as a separate covariate.
Participants with prevalent diabetes at baseline (based on the 2006 WHO fasting glucose definition25) were excluded from analyses. Due to low numbers in some of the eight original time categories for each sitting exposure, each sitting time variable was regrouped into three categories of near-equal numbers as the data permitted (exact tertiles were not possible due to abnormal distribution): work sitting was grouped as 0 to <15, ≥15 to <35 and ≥35 hours/week; TV sitting as 0 to <11, ≥11 to <16 and ≥16 hours/week; non-TV leisure time sitting at home as 0 to <8, ≥8 to <16 and ≥16 hours/week; leisure time sitting as 0 to <15, ≥15 to<25 and ≥25 hours/week; total sitting as 0 to <33, ≥33 to <50 and ≥50 hours/week and total sitting excluding TV as 0 to <33, ≥33 to <50 and ≥50 hours/week. Participants with missing data in any variables required for this analyses were excluded from analyses.
Cox proportional hazards models for each exposure were fitted to examine the associations between each of the six sitting exposures and incident diabetes up to 2011. HR and 95% CI were estimated for each category of sitting time, by type, with the lowest group being the reference category. Examination of Schoenfeld residuals and Nelson–Aelen cumulative hazards plots provided no evidence for deviations from proportionality in any of the Cox models. Analyses were limited to those who had completed both the survey and clinical examination, who were still working in the civil service or elsewhere and who had no prevalent diabetes (590 cases excluded) or heart diseases (1145 cases excluded) at baseline. Models were first adjusted for age, gender, ethnicity and employment grade (model 1) and then further adjusted for smoking status, weekly alcohol intake, fruit and vegetable consumption, self-rated health and physical functioning (model 2). The final model was also adjusted for PA (model 3). To test for linear trends in individual parameters, the Wald χ2 test was used, and the likelihood-ratio χ2 test was used for non-linear relationships. Similar to the previous study,30 we examined the independence of the observed associations (only for the sitting exposures that were associated with incident diabetes in any of the three models) from adiposity in a separate analysis where, in addition to all covariates specified in model 3, we also adjusted for baseline BMI.
In a sensitivity analysis, we repeated the above Cox analyses examining the associations between sitting behaviours in phase 5 and incident diabetes using a 75 g oral glucose tolerance test (OGTT) at phase 9 (the last phase that such a test was included). OGTT involved determination of 2-hour postload glucose according to the WHO standards25 (2-hour postload glucose ≥11.1 mmol/L). Analyses were conducted in 2016 using Stata V.13.2 (StataCorp, College Station, Texas).
Out of the 10 308 participants at the Whitehall II onset, 7870 took part in phase 5. Among them, 517, 1145 and 1397 were excluded due to existing diabetes, existing CVD and missing data in at least one of the variables needed for the multivariate analyses (see online supplementary table 1), respectively. The characteristics of the phase 5 participants that were included in this analysis are shown in table 1. As previously reported,22 compared with those in the sample of the present study, those lost to follow-up between the study’s inception in 1985 and phase 5 were slightly older at the date of screening, consumed slightly less alcohol and were more likely to be men, obese and in a higher employment grade in 1985. The mean follow-up was 13.0 years, corresponding to 62 463 person years.
In total, 402 cases of fasting glucose defined incident diabetes occurred during the follow-up period. As table 2 shows, leisure time sitting, total sitting and TV time showed associations with incident diabetes in the models with minimal adjustment (model 1), and these associations persisted for TV time and total sitting once the remaining potential confounders were taken into account. Work time sitting, non-TV leisure time sitting and total sitting excluding TV time were not associated with the outcome in any of the three models (table 2). Baseline BMI was associated with incident diabetes (per unit HR: 1.15; 95% CI 1.12 to 1.18, p<0.001) after adjusting for total sitting time, age, sex, employment grade, ethnicity, smoking status, alcohol consumption, PA, general health, physical functioning and frequency of fruit and vegetable consumption.
Independence of the observed associations from baseline BMI
Table 3 presents the results additionally adjusted for BMI for exposures that showed associations with incident diabetes in any of the three models. Once BMI was included in the model (model 4), all associations were attenuated. For example, for TV time, the HR of the top compared with the lowest group was attenuated to 1.31 (95% CI 0.96 to 1.76).
Online supplementary table 2 presents the characteristics of the sample included in the sensitivity analysis with OGTT-defined incident diabetes between phase 5 (1996–1998) and phase 9 (2006–2008) (n=4735; 9.7 average years of follow-up, 439 events and 45 864 person years). Results of the sensitivity analyses were consistent with the main analyses described above, and only TV sitting time and total sitting were associated with OGTT-defined incident diabetes (online supplementary table 3). Once baseline BMI was taken into account, these associations were also attenuated, although TV time maintained a borderline association (online supplementary table 4).
Main findings and comparison with previous literature
Our study addresses several gaps in the SB literature by considering type-specific sitting in relation to incident diabetes over a long follow-up. We found that total sitting time and TV time were both associated with incident diabetes independent of PA, and these associations were attenuated once baseline BMI was taken into account. Our findings are a novel contribution, with previous prospective research being reliant on TV time as the sole marker of sitting and, in most cases, lacking adjustment for baseline adiposity.8–10
A long-term prospective study, broadly comparable to ours, featured a sample of 4554 American women with gestational diabetes and examined the association between different types of sitting (TV, other domestic, non-domestic/work and car driving) and risk of type 2 diabetes over 16 years.20 The results of this study were concordant with ours, as the TV time relative risk prior to adjustment for BMI was 1.41 (1.11–1.79) for 11–20 hours/week of the TV time group20 (vs 1.39, 1.03–1.88 for ≥16 hours/week in our study), and adjustments for baseline BMI attenuated these associations substantially.20 In a multiethnic US cohort, the total leisure time sitting (62% of which was TV time) was associated with incident diabetes over 11 years of follow-up21 in overweight and obese participants but not in participants with a BMI <25. A large Danish study of 72 608 adults with a relatively short follow-up (<5 years) found that, once BMI and PA were taken into account, the total sitting time was associated with HbA1c–defined incident diabetes only among the physically inactive and the obese groups.19 Adjustment for baseline BMI attenuated the associations between weekly TV viewing frequency and clustered cardiometabolic risk in another prospective British study over a 21-year follow-up.30
The role of body mass index
Conceptually, adiposity and sitting may be associated in a bidirectional manner, but not both directions of the association are empirically supported. Albeit limited in volume, existing literature12 13 (including a Whitehall II cohort study11) suggests that previous adiposity determines future SB, and no prospective study, to our knowledge, has indicated that sitting predicts markers of adiposity or obesity. Although some shared variance is likely to exist, it is more likely that BMI is a confounder rather than a mediator of the association between sitting and diabetes—an assertion that is indirectly supported by a recent laboratory study showing that the energy expenditure benefits of simply reducing sitting are negligible.31 While more work is needed in this area, our results and the above literature suggest that studies that examine the links between SB and diabetes without adjusting for adiposity may be compromised. Additional pathways linking SB, adiposity and diabetes include the established relationships between TV viewing and obesogenic diets.32 SB may displace PA time, leading to a decrease in energy expenditure and unfavourable weight changes.33 It is worth noting that the top TV tertile in Whitehall II corresponded to >2.3 hours/day, which is well below the general population in England aged >55, where mean values are 3–4 hours/day34). Despite this relatively low bound of the high-TV group, only 0.007% (7 out of 937) of its members reported less than 3.6 hour of TV per day, and as such, it is unlikely that our analyses underestimated the associations due to a likely threshold effect.
Interpretation of main findings
During prolonged sitting, differences in energy balance have been proposed as a major determinant of the metabolic dysfunction (as indexed by compromised insulin action) observed among non-obese young and fit men and women.35 Acknowledging the generally accepted theorem that increased adiposity (which BMI is thought to reflect reasonably at the population level) is the result of chronic energy imbalance, these findings suggest that future studies examining the links between SB and diabetes will benefit from incorporating more robust assessments of energy intake and expenditure.
One possible explanation for the limited evidence linking sitting and diabetes risk in the current study is the protective effect of the high volumes of total reported MVPA, in particular, daily walking that is reported in the Whitehall cohort. For example, the mean reported daily walking time (42.73±22.70 min/day) is over double the reported UK national average.36 Several recent large prospective studies have showed that the associations between sitting time and incident diabetes19 or CVD37 38 are only observed in the least active participants.
In general, previous literature is consistent in that TV time is prospectively associated with diabetes and other cardiometabolic outcomes3 39 but occupational sitting is not.40 This contradiction suggests that examining total sitting volumes alone may not be sufficiently informative due to the existence of context-specific unmeasured confounding (eg, dietary or socioeconomic). Another important consideration is the pattern of sitting (eg, length of bouts and frequency of interruptions from sitting), which may be relevant to health outcomes but cannot be captured by self-report measures, including the questionnaires used in our study. A study of 164 London office workers41 who wore inclinometers for 7 days found that 69% of sitting bouts are <10 min, with only <10% of all bouts lasting >60 min in duration. Both during work and in the evenings, participants registered approximately two sit-to-stand transitions per hour, a pattern that has been linked to measurable improvements in acute glycaemic responses in laboratory studies testing interruptions of sitting with walking.5 42 Assuming the sitting patterns of the Whitehall II cohort are similar, and the absence of notable effects may be partly attributed to the relatively frequent short PA bouts that confer glycaemic protection in this occupational cohort.
Strengths and limitations
Strengths of our study include the prospective design, the long follow-up, and the six sitting exposures (covering work, recreation and commuting) that allowed us to take into account the sitting context. We were also able to take account of a broad range of important confounding factors, including physical functioning that is linked to acute injury and long-standing illness and may be a contributor to increased sitting. Our study also has limitations. Sitting was measured using self-report that may be subject to recall and social desirability biases. Leisure time questions only captured sitting at home. Occupational cohorts are, by definition, sufficiently healthy at baseline to be in active employment, which may reduce the extent to which our conclusions are generalisable. However, aetiological findings from Whitehall II are broadly consistent with those obtained from representative cohorts.43 Despite threats to the ecological validity of our study, it is reassuring that our results are in agreement with a clinical US cohort.20 We were able to take into account fruit and vegetable consumption, which is an important for diabetes risk44 aspect of diet.
In conclusion, our study makes a unique contribution in the literature by examining prospectively a broad range of type-specific SBs in relation to incident diabetes over a long follow-up period of 13 years in a physically active cohort of British civil servants. We found moderate evidence linking TV time and limited evidence linking total sitting with diabetes, but these links were dependent on baseline BMI. It is important that prospective general population studies using objective measures of sitting patterns (in addition to sitting context-specific measures) and controlled trials replicate our findings.
What are the findings?
The epidemiological literature on sitting and incident diabetes is very scant and rarely acknowledges the confounding role of adiposity.
Occupational, non-television (TV) leisure time at home and total non-TV sitting were not associated with incident diabetes risk over 13 years of follow-up.
TV time and total sitting were associated with diabetes, but once baseline body mass index (BMI) was taken into account, these associations were attenuated.
How might it impact on clinical practice in the future?
Our findings provide little support for developing interventions that specifically target sitting to reduce diabetes risk.
The weak evidence for associations that we found may be partly due to the protective effect of relatively high amounts of daily walking and other moderate-to-vigorous physical activities in this cohort.
Strategies to increase walking and other physical activities and reduce BMI remain the cornerstone of diabetes prevention.
The authors would like to thank Dr Martin Shipley from the Whitehall II study (University College London) for his assistance and advice with reconciling the incident diabetes variables.
Contributors ES: conceived the idea, designed the study, drafted the manuscript, provided initial interpretation of the data and carried out multiple manuscript revisions. RMP: contributed to data acquisition, contributed to study design, did and updated the statistical analysis several times, assisted with data acquisition and drafted parts of the manuscript. EJB: contributed to data acquisition and study design, critical revision of the manuscript for important intellectual content and results interpretation. ARB: contributed to data acquisition, critical revision of the manuscript for important intellectual content and results interpretation. AEB: critical revision of the manuscript for important intellectual content and results interpretation. SJHB: critical revision of the manuscript for important intellectual content and results interpretation. MH: contributed to data acquisition and contributed to the study design, critical revision of the manuscript for important intellectual content and results interpretation.
Funding The Whitehall II study is supported by grants from the Medical Research Council (G0902037), British Heart Foundation (RG/07/008/23674), Stroke Association, National Heart Lung and Blood Institute (5RO1 HL036310) and National Institute on Aging (5RO1AG13196 and 5RO1AG034454). A National Health and Medical Research Council (Australia) Senior Research Fellowship and a National Institute for Health Research Career Development Fellowship (UK) supported the first author of this article during different stages of this work.
Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.