Background: Questions remain about the effect on mortality of physical activity and sedentary behaviour over time. We summarized the evidence from studies that assessed exposure from multiple time points and critiqued the analytic approaches used.
Methods: A search was performed on MEDLINE, Embase, Emcare, Scopus and Web of Science up to January 2021 for studies of repeatedly assessed physical activity or sedentary behaviour in relation to all-cause or cause-specific mortality. Relative risks from individual studies were extracted. Each study was assessed for risk of bias from multiple domains.
Results: We identified 64 eligible studies (57 on physical activity, 6 on sedentary behaviour, 1 on both). Cox regression with a time-fixed exposure history (n = 45) or time-varying covariates (n = 13) were the most frequently used methods. Only four studies used g-methods, which are designed to adjust for time-varying confounding. Risk of bias arose primarily from inadequate adjustment for time-varying confounders, participant selection, exposure classification and changes from measured exposure. Despite heterogeneity in methods, most studies found that being consistently or increasingly active over adulthood was associated with lower all-cause and cardiovascular-disease mortality compared with being always inactive. Few studies examined physical-activity changes and cancer mortality or effects of sedentary-behaviour changes on mortality outcomes.
Conclusions: Accumulating more evidence using longitudinal data while addressing the methodological challenges would provide greater insight into the health effects of initiating or maintaining a more active and less sedentary lifestyle.
Keywords: time-varying confounding; Physical activity; g-methods; longitudinal studies; mortality; sedentary behaviour; time-varying exposures.
© The Author(s) 2021; all rights reserved. Published by Oxford University Press on behalf of the International Epidemiological Association.