Accounting For Physical Activity In Sedentary Behaviour Research: A Theoretical Framework

Introduction Given its associations with morbidity and mortality, sedentary behaviour (SB) has emerged as a potentially significant public health issue. Although most studies of relationships between SB and health outcomes acknowledge that some sort of adjustment should be made for physical activity, there is no consensus on the approach for doing this.

Purpose The aim was to clarify the role of moderate to vigorous physical activity (MVPA) in the relationship between SB and mortality, by illustrating models with PA considered as a confounder, mediator and effect modifier and with SB and PA measured both with questionnaires and accelerometry.

Methods Data were from 1906 participants (aged 50+) in the National Health and Nutrition Examination Survey (NHANES) who wore an accelerometer (ActiGraph AM-7164) for 7 days in 2003/04. Mortality data were available until December 31, 2006. The association between SB and mortality was analysed with Cox proportional hazards models. Assumptions for confounding, mediation and effect modification were tested. Analyses were done with separate models for SB and MVPA measured with accelerometry (SB=% time counts<100, MVPA=%time counts<2020) and self-report (SB=h/day tv/computer; MVPA=MET.min/week leisure MVPA).

Results For both accelerometry and self-report, associations between (1) MVPA and SB (p<0.001), (2) MVPA and mortality (p<0.001), and (3) SB and mortality (p≤0.001) were statistically significant, suggesting MVPA was a potential confounder. Additional adjustment for MVPA in the association between SB and mortality resulted in a small attenuation of the hazard ratios, but confidence intervals were wide and the direction of the effect was unclear, so there was insufficient evidence for mediation. The variance inflation factor was low (VIF≤1.2) suggesting no collinearity. The p-value for interaction was 0.26 for accelerometry and 0.48 for self-report. For both measurement methods, stratified analyses showed similar hazard ratios for participants in the low and high MVPA categories, suggesting no effect modification, but confidence intervals were wide.

Conclusions Assumptions for confounding were met and there was no evidence for collinearity. Mediation and effect modification could not be adequately assessed.