Sedentary behaviours that involve sitting for extended periods are ubiquitous in modern society. Based on self-reports, a recent survey of 20 countries documented a median of 300 min/day spent sitting, ranging from ≤180 min/day in Portugal, Brazil and Colombia to ≥360 min/day in Taiwan, Norway, Hong Kong, Saudi Arabia and Japan.⁵ Several studies have demonstrated positive associations between sedentary behaviours including sitting and television viewing and health outcomes such as type 2 diabetes, cardiovascular disease mortality and all-cause mortality.⁶ Thus, excessive time spent in sedentary behaviour is undoubtedly having an impact on public health. A recent study from Australia estimated that television viewing reduced life expectancy at birth by 1.8 years in men and 1.5 years in women.⁷ The purpose of this study was to determine the impact of sedentary behaviours on life expectancy in the USA.

Three studies were included that studied the association between television viewing and all-cause mortality.^17–19 The levels of television viewing in two of the studies were (1) <2 h, (2) 2–3.9 h and (3) ≥4 h,^{17 18} whereas the levels in one study were (1) <2.5 h, (2) 2.5–3.6 h and (3) >3.6 h.¹⁹ The prevalence of television viewing obtained from NHANES (<2 h, 2–3.9 h and ≥4 h/day) match quite well with the exposure categories from the cohort studies.

Figure 1 presents the results of the meta-analysis for sitting and all-cause mortality. The pooled RRs were 1.18 (95% CI 1.14 to 1.21) and 1.45 (95% CI 1.39 to 1.51) for levels 2 and 3 versus level 1, respectively. Figure 2 presents the results for the meta-analysis for television viewing and all-cause mortality. The pooled RRs were 1.17 (95% CI 1.04 to 1.32) and 1.49 (95% CI 1.22 to 1.82) for levels 2 and 3 versus level 1, respectively.

The prevalences of sitting and television viewing in cases (decedents) and in the source population at baseline in the prospective studies are presented in table 2, along with the average case:source prevalence ratio. These ratios were applied to the population prevalences obtained from the NHANES in order to estimate the prevalences among cases in the population, which are required for inclusion in the PAF calculations. Figure 3 presents the weighted population prevalences of sitting and television viewing in the NHANES, along with the prevalences after adjustment for the average case:source prevalence ratios from table 2.

The PAFs for all-cause mortality associated with sitting and television viewing were 27% and 19%, respectively. The results of the life table analyses indicate a gain in life expectancy from reducing the prevalence of sedentary behaviour from level 2 or 3 to level 1 results in a gain of 2.00 years for reducing sitting prevalence and a gain of 1.38 years from reducing television viewing prevalence. The lower and upper limits from the sensitivity analyses were 1.39 and 2.69 years for sitting and 0.48 and 2.51 years for television viewing, respectively.

This study has several strengths and limitations that warrant discussion. The use of the well-accepted prevalence-based PAF methodology to estimate the public health burden of sedentary behaviour is a marked strength, which allows for comparability with the effects of other established risk factors. However, the PAF provides a theoretical estimate of the effects of a risk factor on a health outcome, and further research is required using an incidence-based approach. Our analysis assumes that there is a causal relationship between sedentary behaviour and mortality. While studies using randomised designs are not possible, further observational studies, which control for confounding, will add to the evidence for causation. Another major strength of this study is the use of representative population data from the NHANES to quantify the exposure of the population to sedentary behaviours. However, our analysis estimated the overall gains in life expectancy at the population level and assumes that the effects of sedentary time on all-cause mortality are consistent across age and demographic subgroups of the population. Each of the cohort studies provided multivariable-adjusted RR estimates for sedentary behaviour and mortality using different combinations of covariates, and we chose to use summary RR estimates based on RR adjusted for age and sex in order to maintain consistency across studies. The degree to which this approach has yielded an overestimation of the independent effect of sedentary behaviour on life expectancy is not known, as important confounders associated with both sedentary behaviour and premature mortality may have been unmeasured or inadequately adjusted for in the primary studies, and we have not made further adjustments for confounding in our analyses. We conducted a sensitivity analysis by reducing the RR from the meta-analysis by the amount observed between age- and sex-adjusted versus multivariable RR estimates from three studies,^{15 17 18} and the estimates of attributable life expectancy were reduced to 1.76 years for sitting and to 0.93 years for television viewing.

This study relied on self-reported engagement in sedentary behaviours, which introduces the possibility for error and recall bias. Future cohort studies should attempt to better quantify sedentary behaviour using objective activity monitors; however, the self-reported estimates of sitting reported in the NHANES are similar to those obtained for the USA in a study of 20 countries,⁵ which indicates some face validity to the results. For television viewing, the categories of exposure reported in two of the cohort studies (<2, 2–3.9 and ≥4 h) matched the categories reported in the NHANES; however, Wijndaele et al¹⁹ used a lower threshold for the upper category (>3.6 h), which may have resulted in an underestimate of the effects of television viewing on life expectancy. The studies on sitting used different exposure categories (ie, total sitting time versus leisure-time sitting only), so assumptions had to be made when estimating the exposure levels in the NHANES. Inaccuracies associated with the assessment of sedentary behaviour using self-report methods in the cohort studies would have led to regression dilution bias and resulted in underestimates of the association with all-cause mortality.

The results of several recent studies have suggested that the effects of sedentary behaviour on health may be independent of the effects of physical activity per se.^{2 20} It has been estimated that a lack of leisure-time physical activity accounts for approximately 0.9 years of life expectancy at birth in Canada.²¹ Given that the studies used to derive the summary RR estimates for the current study in many cases included leisure-time physical activity as a covariate in multivariable-adjusted models, and this did not appreciably change the estimates of RR for sedentary behaviour, the estimates of the effects on life expectancy may also be considered independent. Current life expectancy in the USA (2009) is 78.5 years.¹⁴ The effects of sedentary behaviour on life expectancy reported in this study are on a similar order of magnitude as other chronic disease risk factors. For example, it has been estimated that obesity accounts for between 0.30 and 1.08 years of population life expectancy at birth in the USA, depending on gender, ethnicity and severity of obesity.²² A more recent study has estimated that the current distribution of body mass index, compared with an optimal distribution (a mean of 21 kg/m²), accounts for 1.3 years of current life expectancy at birth in both men and women in the USA.²³ Results from the same study indicate that smoking is also associated with 2.5 and 1.8 years of life expectancy at birth in men and women, respectively.²³

Sitting time is a global measure of sedentary behaviour, whereas television viewing is somewhat more specific. A recent review found that domain-specific behaviours such as television viewing are recalled with more reliability than global measures of sitting and sedentary behaviour.²⁴ The degree to which differences in reliability between the measures used in this study may have affected the estimates is not known. There is some evidence to suggest that using a single global question to measure sitting produces lower estimates than more detailed domain-specific questions.^{24 25} Thus, it is likely that the prevalences of higher levels of sitting reported in this study from the NHANES are likely conservative. Using objective monitoring (accelerometry) in the 2003–2004 NHANES, Matthews et al²⁶ reported that US adults spend approximately 7.7 h/day engaged in sedentary behaviour. There are several potential mechanisms that could explain the association between sedentary behaviour and all-cause mortality rates. Sedentary behaviour is associated with an increased risk of the development of chronic conditions such type 2 diabetes and cardiovascular disease.⁶ Furthermore, human and animal studies indicate that sedentary behaviour is associated with elevated cardiometabolic biomarkers and a poor risk factor profile.^{27 28} For example, hindlimb suspension (unloading) in rats results in marked immediate decreases in lipoprotein lipase activity, triglyceride uptake into red skeletal muscle and reductions in the concentration of high-density lipoprotein cholesterol.²⁹ Future intervention research is required to determine the causal pathways between sedentary behaviour and health outcomes that have the potential to impact mortality rates.