There is reasonable epidemiological evidence for a causal relationship between sedentary behaviour and all-cause mortality. Within the colour-coding classification used, the eight systematic reviews examined showed clear support for two of the four Bradford Hill factors for assessing causation from population observations: consistency (7 green) and temporality (6 green). There was some support, though less consistent, for strength of association (4 green) and no support for dose–response (5 red). Only three reviews provided data on dose–response relationships. When assessing reviews as a whole, five of the eight reviews were rated green, indicating support for a causal relationship between sedentary behaviour and ACM. Of the 17 primary studies (18 assessments), 67 % were rated green, with evidence for strength and temporality.
Assessing causality using the Bradford Hill framework
Within epidemiological research, judging strength of association is not easy. Complicating our assessments was that the reviews and primary studies we examined differed by how they analysed data, and different measures of sedentary behaviour were often adopted. Rosenthal [
50] has suggested that an odds ratio between 1.5 and 2.5 could be considered of ‘moderate’ strength, but no criteria are provided for hazard ratios. Of the six reviews providing summary data on strength of association, none reached the level of ‘moderate’ strength when comparing highest versus lowest sedentary categories. However, three of the six were close to such a value, showing relative risk ratios between 1.45-1.49. These values are slightly less than what Khaw et al. [
51] reported for mortality for smoking (
RR = 1.77), comparable to five or more daily servings of fruit and vegetables (RR = 1.44), but higher than being physically inactive (RR = 1.24) and consuming more than 14 units of alcohol per week (RR = 1.26). These comparisons are from analyses of 20,244 men and women in England aged 45–79 years in 1993–1997 and followed up to 2006.
In a recent analysis of the Australian ‘45 and Up’ study, Ding and colleagues [
52] calculated hazard ratios for ACM for seven risk factors singly and in combination. The HR for prolonged sitting, when analysed singly, was 1.15. Interestingly, again when considered singly, this was the most prevalent risk factor. But this HR value is small and was less than for physical inactivity alone (HR = 1.61) but higher than poor diet (HR = 1.04).
It is debatable how strong an association we might expect for the behaviour of ‘sitting’ on ACM, but the association that often appears for sitting is sometimes comparable to other risk factors. For example, Bouchard et al. [
4] plot the relative risk for ACM for sitting (from one large study), cardiorespiratory fitness, and MVPA. The trajectories are similar although they reported that while all three have strong associations with premature mortality, the strongest effects are from fitness. It is also possible to argue that sedentary behaviour, even with a slightly smaller effect on ACM that fitness or MVPA, will affect a large proportion of the population, thus making it an important public health issue. This is due to the high volume of total daily sitting for large numbers of people.
Judgments of consistency show that much of the evidence concerns white European and ‘western’ populations, with just one large study from Asia, and none from Africa or South American. There seem to be no obvious differences by sex at the level of reviews, but two primary studies showed contradictory findings with one reporting stronger effects for women [
32] and one for men [
29]. This may be a function of ethnicity with the studies reporting data for US and Japanese adults. That said, a pooled analysis of multiple cohorts from England and Scotland with over 11,000 adults, showed ACM to be associated with occupational sitting for women only [
53]. To strengthen judgements concerning the consistency of evidence, studies are required with more diverse populations using standardised assessments of sedentary behaviour.
The assessment of temporality showed support for the notion of a causal link between sedentary behaviour and ACM. The majority of primary studies had follow-up periods of greater than 7 years, and only three studies had less than 3 years, although the evidence from shorter follow-up studies may not be ready for publication. Most studies we reviewed reported that they did not analyse data for those dying within 2 years of sedentary behaviour assessment thus avoiding the potential confounding influence of occult disease.
Most reviews and primary studies did not report data addressing dose–response effects. From the three studies where data were available, it appears that any effects on ACM are from much higher levels of sedentary behaviour. For example, the review concerning TV viewing by Grontved and Hu [
45] showed that only from around 3 h per day of TV was there an increase in mortality. Similarly, Katzmarzyk and Lee [
46] showed the greatest effects for ACM for the third tertile for both total sitting and TV viewing. Such findings suggest a non-linear or even a threshold effect for sedentary behaviour on ACM, with lower levels of the behaviour having little effect. Indeed, Chau et al’s [
47] analysis of six studies concluded that the risk of ACM increased significantly from about 7 h per day of sitting. Further work on dose–response relationships is required, including analyses of total sedentary time and specific behaviours, such leisure-time TV viewing, workplace sitting and car travel. For now, we cannot conclude with great precision about dose–response effects as most studies (primary and reviews) did not address this (but see [
4]).
Biological plausibility could not be directly assessed in our analyses as we were reviewing only epidemiological studies. However, it can be argued at a general level that higher rates of sitting are plausibly associated with poor health, and this can include an analysis of the factor of ‘analogy’ proposed within the Bradford Hill framework. It has been argued that analogies may contribute to assessments of the weight of evidence [
23]. Moreover, there are at least three ways in which analogies could be drawn when investigating sedentary behaviour and health outcomes. First, there is convincing evidence that low levels of physical activity are associated with ACM [
13]. This has been shown for men and women when comparing multivariate adjusted differences in ACM between those reporting no physical activity and those doing ‘light’ physical activity. While it has been argued elsewhere that ‘sitting’ is not the same as avoiding physical activity (e.g., someone could stand all day but still not move too much) [
2,
5], the argument by analogy is persuasive. Indeed, reductions in sitting time are most likely to transfer into greater amounts of light physical activity, and the meta-analysis by Löllgen et al. [
13] shows that this could be important for ACM. Second, there is evidence that time spent standing is associated in a dose–response manner with ACM [
54]. Standing, and other forms of light movement, are strongly negatively associated with sedentary behaviour [
55].
Third, there is a long history in physiology showing rapid and detrimental effects on health and functional markers from extended bed rest [
56] and from hind limb unloading in rodents [
57,
58]. Finally, evidence on the effects of weightlessness in space, and the associated deleterious health effects, has been suggested as a clear analogy to the low musculo-skeletal loading of prolonged sitting [
59]. Collectively, these types of studies have provided unique insights into the potential causal relationship and possible underlying mechanisms through which physical inactivity may contribute to chronic disease development. Recent reviews have documented evidence on the many physiological responses related to imposed physical inactivity [
56,
60], including a reduced capacity to use fat as a substrate or ATP production, muscle atrophy, a shift in muscle fibres toward fast-twitch glycolytic type, muscle insulin resistance, ectopic fat storage and increased central and peripheral adiposity. In summary, while not all of these fields address ACM, there is a case for ‘analogy’ and ‘biological plausibility’ concerning significant negative health effects of sedentary behaviour from related fields.
While the initial evidence on potential biological mechanisms of poor health markers from sedentary behaviour is largely derived from animal studies [
61], emerging human research is promising, notwithstanding that such studies do not assess mortality outcomes [
62,
63].
Does physical activity matter?
It is important for future studies to control for potential confounding variables in any assessment of links between sedentary behaviour and health outcomes. One such key variable is MVPA. While it may not be too surprising to see the positive effects of physical activity over-power any effects for sitting time, it should be noted that when assessed by accelerometry, large proportions of the adult population have low levels of physical activity [
64], thus sedentary behaviour is still an important behaviour to address for the majority and thus is a significant public health issue.
Nevertheless, there is still some debate if, and by how much, sedentary behaviour is associated with deleterious health outcomes independently of MVPA. For example, while many studies show associations between sedentary behaviour and poor health following adjustment for MVPA levels – indeed many studies did this in their fully adjusted models – the most recent and comprehensive of the systematic reviews [
49] showed that the link between sedentary behaviour and ACM held only for those with low levels of physical activity. In addition, Bjork-Petersen et al. [
65] have reported on a large study of Danish adults and found that being physically inactive during leisure time but also sitting more than 10 h per day was associated with increased risk of ACM, as one might expect. However, for those who were physically active, sitting time had minimal effect on ACM. This is similar to an analysis of cardiovascular disease outcomes in a large study of sitting time and physical activity in women [
66]. Specifically, more time spent sitting increased the risk of CVD except for those classified as the most active.
A recent analysis of the Whitehall II cohort has suggested that sedentary behaviour and ACM are not associated [
67]. However, the sample was a large group of London ‘civil servants’ who were free from CVD and likely to be considered quite healthy. They were not obese and had high levels of physical activity. Moreover, the average age at baseline was in the mid-40s, which may be too young to show any significant mortality effects. In summary, this null finding may be the result of the population studied, but also includes the possible explanation that their physical activity levels partially attenuated the negative health effects of prolonged sitting.
In a novel analysis of over 200,000 mid- and older-aged adults in Australia [
68], analyses investigating the ‘effects’ of substituting certain behaviours for others showed “an independent beneficial effect of standing on mortality (3 % decrease in risk per hour of standing in the whole sample), and this association was present in both those who met and did not meet the physical activity recommendations. These beneficial associations of standing with mortality were even more substantial (5 % decrease in risk for each hour of sitting replaced with standing) when standing time displaced sedentary behaviour” (p. 8). These findings suggest that reductions in sedentary behaviour are important for improved mortality outcomes. Additionally, recent evidence by Healy et al. [
63], also using isotemporal substitution models but with one cross-sectional population data set, showed that replacing sitting with standing was likely to have significant cardiometabolic-related health benefits. Replacing sitting with stepping was more beneficial for adiposity outcomes.
Clearly, more needs to be known about the interaction of sedentary behaviour and physical activity. This may require an analysis of the role of MVPA as a moderator rather than just adjusting for levels of MVPA and taking into account co-exiting behaviours, including sleep [
69,
70]. Accounting for these behaviours across 24 h may require investigation using compositional data analysis [
70,
71]. This is clearly an important future direction.
Measurement of sedentary behaviour
The studies that we examined have several measures of sedentary behaviour exposure – the majority of which were derived from self-report assessment instruments. Television viewing time has consistently emerged as a strong exposure variable, although it is only a subset of overall sedentary behaviour. One of the reasons why this measure has performed so strongly in epidemiological studies is likely to be that recall of television viewing time – typically a consistently repeated habitual behaviour – provides an exposure measure that may be less strongly influenced by measurement error than the other indices that have been used. However, TV time may be seriously confounded by socio-economic status and by unhealthy dietary snacking behaviours concomitant with television time. In addition, self-reported overall sitting time is a consistent predictor of mortality in prospective epidemiological studies, as shown. While the precision of people’s recall of overall daily or weekly sedentary time is likely to be limited, these measures are valuable in large population studies in that they provide a rank of individuals on their total exposure to sedentary behaviour [
72].
The measurement of sedentary behaviour is a moving target – the rapid uptake of cheap, convenient and easy-to use wearable and screen-based devices provides a plethora of new opportunities for sedentary time. Older studies focused primarily on TV viewing time while new studies address ‘screen time’. Ongoing epidemiological studies may soon provide a large body of findings with exposure measures of sedentary behaviour that are derived from small, unobtrusive wearable devices (including accelerometers that capture movement and inclinometers that assess posture directly). Data from these devices will allow not only objectively-assessed overall sedentary time to be employed as an exposure variable more readily, they will also provide the capacity to examine several important and as yet unexplored questions on patterning of sedentary behaviours. Examples include how sitting time is accumulated in different bout lengths, the potential protective role of regularly breaking up sedentary time, and how bouts of physical activity may be protective.
The limitations of our analysis include the usual methodological limitations of observational epidemiological data, and thus excludes a direct analysis of experimental studies that would contribute to biological plausibility. The reliance on self-report exposure measures has potential measurement as well as non-specific and specific unmeasured confounding. Not all studies assessed sedentary behaviour or physical activity in the same way, and some did not assess diet at all. One issue is the extensive use of TV viewing as a proxy measure of sedentary behaviour, when total sitting time may show a direct socio-economic gradient, compared to the inverse gradient seen for TV time alone. [
73]. In addition, our use of the colour coding system, while designed to be helpful to the reader, inevitably relies on subtle judgements and the criteria specified for assessment.