In this issue of Diabetologia, van der Berg et al report the results of an interesting observational study [1]. The authors examined the cross-sectional associations of objectively measured sitting with type 2 diabetes and cardiometabolic risk in a sample of nearly 2,500 middle-aged Dutch adults enriched with type 2 diabetes patients. This study is an important addition to the literature aiming to examine the health consequences of sedentary behaviour, recently defined as any waking activity with low energy expenditure and performed in a sitting or reclining posture [2]. The most important asset of the work by van der Berg et al [1] is the implementation, in a large participant sample, of a thigh-worn activity monitor that accurately discriminates between sitting/lying and non-sitting/lying postures. Using this methodology the authors were able to relate total sitting time and indicators of sitting patterns (i.e. the manner in which sitting time is accumulated) to glucose metabolism status and prevalence of criteria of the metabolic syndrome. They found that each additional hour daily of sitting time was significantly associated with increased odds of having type 2 diabetes (1.22, 95% CI 1.13, 1.32) and the metabolic syndrome (1.39, 95% CI 1.27, 1.53). This finding was irrespective of time spent in higher intensity activity and a wide range of socio-demographic, behavioural and health-related confounding factors. The accumulation patterns of this sitting time, whether examined as number of interruptions in sitting, number of prolonged (i.e. ≥30 min) sitting bouts or average sitting bout duration, however, showed much weaker associations.

Clinical and public health relevance of sedentary behaviour

Interest in sitting as a risk factor for type 2 diabetes was sparked 15 years ago by Hu et al [3, 4]. They were the first to associate high levels of TV viewing and other types of sedentary behaviour with increased risk for incident type 2 diabetes. As highlighted in a fairly recent meta-analysis in this journal [5], the link between sedentary behaviour and type 2 diabetes is one of the most consistently observed in this field—a field that has experienced an exponential rise in scientific, media and general community interest. This increased interest has also resulted in the emergence of preliminary public health guidelines focusing on reductions in prolonged sitting (for example, see [6]).

Sedentary behaviour is, however, still a fairly young research area, both in terms of advancements in measurement methodology as well as establishment of robust observational longitudinal and experimental evidence. Current meta-analytical effect estimates are predominantly based on self-reported estimates of sitting time, which are associated with substantial measurement error. Studies with accurate measurement of sitting posture, such as the one by van der Berg et al [1], are therefore most welcome—especially if they can also be extended to a longitudinal design, to minimise the risk of reverse causality (i.e. when individuals are already on the disease pathway and therefore display or report more sitting). To explore whether reverse causality affected the results of their cross-sectional study, the authors ran a sensitivity analysis in which they excluded participants with type 2 diabetes on insulin medication (who were assumed to have more severe diabetes). This did not change the overall results. However, a major limitation of cross-sectional research is the lack of inference on the temporal sequence between exposure (e.g. sedentary time) and outcome (e.g. type 2 diabetes), which will inherently always be associated with substantial risk of reverse causality. Another common limitation in this research area is residual confounding for poorly or unmeasured confounders. This is when a relationship is estimated incorrectly because a confounding factor was not measured (such as dietary intake, as acknowledged by the authors) or not sufficiently accounted for (such as moderate-to-vigorous physical activity, as described further below).

These difficulties hamper our current understanding of the exact impact prolonged sitting may have on the development of chronic disease, such as type 2 diabetes. Excessive sedentary time is, however, ubiquitous in most modern societies, both at an inter- (i.e. in the majority of individuals across all age groups) and intra-individual level (i.e. on average more than half of the waking day) [7]. This means that even if true effect estimates are relatively small, the potential population impact of any behavioural change in terms of prevention of type 2 diabetes and other chronic disease may still be relevant.

Opportunities to improve the evidence base

Patterns of sedentary time: optimising its operationalisation in observational research

Unlike other health behaviours, for example smoking, where even minimal exposure such as passive smoking may increase risk, not all sitting is harmful. Indeed, some sitting is needed for rest and recovery, and prolonged time spent in non-sedentary postures (e.g. static standing) also has inherent risks. Attention has therefore shifted to also considering, and intervening on the patterns of sedentary time, i.e. how sedentary time is accrued.

Definitions of sedentary patterns vary between observational studies. This is testament to the relatively recent interest in this behavioural construct as a potential independent health hazard. Healy et al [8] introduced this concept by relating the number of ‘breaks’ or interruptions in sedentary time to specific cardiometabolic risk factors. This breaks construct, which has been incorporated into public health messages, is now widely used in studies looking into accumulation patterns of sedentary time. However, this frequency measure may not be the most robust estimate of patterns of sitting time, partially because it does not directly assess duration of prolonged sitting bouts per se [9]. Other indicators have therefore also been considered, some of which are also frequency-based (e.g. number of prolonged bouts), whereas others are more duration-based (e.g. average sedentary bout duration). As also reported by van der Berg et al [1], these different constructs show associations of different strengths, which indicates that they may not relate equally strongly to the same latent construct of ‘prolonged sitting’. This diversity may complicate comparisons between study results and general inference about the importance of sitting patterns as regards health. More work is therefore needed in terms of optimising and harmonising pattern definitions for observational research. Constructs that are conceptually more similar to assessing prolonged sitting per se may need to be used more universally, such as time spent in bouts of minimal durations (e.g. ≥30 min). These constructs can also be easily translated into public health guidelines (e.g. ‘Get up at least every 30 min’) if they were to be associated with increased health hazard. Concurrently, more studies should implement posture-discriminating methods that can accurately distinguish sitting from standing, and therefore also identify the transitions between these postures.

Integrating the use of improved measures of sitting time, higher intensity activities and domain-specific information

As mentioned above, the study by van der Berg et al [1] is to be commended for the use of posture-discriminating accelerometry. This is currently rare on such a scale; the vast majority of previous studies have implemented traditional hip- or waist-mounted accelerometers [10]. Despite this, however, time spent in moderate-to-vigorous physical activity, an important co-variate, was measured in a more rudimentary manner. This problem, common in this field of research, could be overcome in future research by combining different objective inference methods, each specialised in accurately capturing specific subcomponents of the intensity or posture-based spectrum of activities. Further important opportunities involve the integrated use of simultaneously measured domain-specific information (e.g. through domain-specific logs [11] or GPS), ensuring participant burden remains low. This would allow an understanding of the context of the behaviour and associated opportunities for intervention.

Considering the 24 h clock

Notably, activities do not occur in isolation, but, rather, are interdependent. That is, given a 24 h day, time spent doing one activity necessarily displaces time spent doing another. The impact of sedentary time is therefore not only due to the time spent in a low energy, low muscle contractile state, but also due to the loss of time spent in higher intensity activities. An individual’s health is impacted by the relative balance between sleep, sedentary behaviour and physical activity. More advanced analytical techniques, such as iso-temporal substitution modelling [12] and compositional data analysis [13], have recently been introduced in sedentary behaviour/physical activity epidemiology. These approaches aim to account for this interdependency of activity behaviours in aetiological analyses with health outcomes. They therefore provide more realistic insights into the relative impact of a behavioural change from one part of the intensity spectrum to another, using observational data. As expected, these models have typically shown that shifting time from sedentary to moderate- or vigorous-intensity activities generally has the greatest benefit on markers of cardiometabolic health [13]. However, given the challenges associated with increasing population levels of moderate- to vigorous-intensity activity, it is encouraging to note that some cardiometabolic benefits may be achieved with shifts from sedentary to light-intensity activities, including standing [14]. An important consideration in the use of these techniques is the assurance that they provide us with insights that are informative and easily translatable into public health guidance.

Establishing robust evidence in a prospective and global setting

The van der Berg et al study [1], and the vast majority of other observational studies that have related objectively measured sedentary time with health outcomes, have been cross-sectional. This is because implementation of these activity monitors in larger aetiological studies is a more recent development, especially those monitors which are able to accurately discriminate between postures [10]. Future longitudinal assessment in cohorts such as that in the Maastricht study [1] will reduce the potential risk of findings being affected by reverse causality. As effect estimates for some of these associations may also be fairly small, future data sharing between multiple cohorts will facilitate an increase in statistical power. This provides additional opportunities in terms of understanding differences in dose–response associations within and between populations. Given the predominance of non-posture-discriminating methodologies in ongoing cohorts, this may also encourage endeavours to harmonise data between different methodologies [10].

In terms of experimental research, several laboratory-based physiological trials have indicated the potential acute benefits of breaking up prolonged sitting, especially in terms of limiting postprandial glucose and insulin excursions (for example, see [15]). Future acute physiological trials, with further control for the frequency, duration and intensity of breaks, will advance our understanding of the physiological underpinnings of observational findings. Alongside these laboratory-based studies, ongoing free-living intervention research is providing evidence for the feasibility of changing prolonged sitting behaviour [16]. These studies now need to be followed by trials that are sufficiently powered to allow assessment of the longer term effects of changes in free-living sitting on cardiometabolic and other health outcomes. These trials should also control for compensatory effects in terms of sitting behaviour and total energy expenditure.

Conclusion

We applaud the study by van der Berg et al [1], which provides important new insights into this area of research. Robust findings from future longitudinal observational and experimental studies, which engage with the wealth of current and future opportunities (as described above), are now needed. These will place us in a better position to (1) make more realistic inferences about the public health impact of changing activity intensity levels in a balanced, efficient and realistic manner; (2) subsequently shape more refined public health guidance in line with this evidence; and, last but not least, (3) translate these insights into individual, micro- and macro-level societal changes, which reduce the public health burden associated with the distorted balance of activity intensity levels, as currently witnessed around the world.