Accelerometer-measured physical activity and sedentary time in a cohort of US adults followed for up to 13 years: the influence of removing early follow-up on associations with mortality

Background Observational studies linking physical activity with mortality are susceptible to reverse causation bias from undiagnosed and prevalent diseases. Researchers often attempt to deal with reverse causation bias by excluding deaths occurring within the first 1 or 2 years from the analysis, but it is unclear if excluding deaths within this time-frame is sufficient to remove bias. Methods We examined associations between total and intensity-specific physical activity and sedentary time with all-cause mortality in a prospective cohort of 3542 individuals from the 2003-2006 NHANES cycles. In order to yield measures of association hypothesized as minimally influenced by reverse causation bias the primary analysis excluded individuals with < 5 years of follow-up. Accelerometer-measured physical activity was linked with recently updated vital status from the National Death Index with a median follow-up of 10.8 years. Results Hazard ratios (95% confidence intervals) were 0.74 (0.53, 1.04), 0.52 (0.37, 0.73), and 0.61 (0.38, 1.01) for ascending quartiles of total physical activity against the least active reference. Hazard ratios for ascending moderate-to-vigorous physical activity quartiles against the reference were 0.67 (0.47, 1.96), 0.67 (0.47, 0.95), and 0.68 (0.39, 1.18). Associations for light intensity physical activity and sedentary time were smaller in magnitude and all confidence intervals included unity. Total activity and moderate-to-vigorous physical activity hazard ratios from analyses only excluding deaths within the first 2 years were inflated by 13 and 26% relative to analysis restricted to >= 5 years of follow-up. Conclusions The pattern of associations suggested total physical activity and moderate-to-vigorous physical activity were associated with lower mortality after more than 10 years of follow-up and excluding the first 5 years of observation time to minimize the impact of reverse causation bias. Excluding deaths within the first 2 years appeared insufficient to minimize the impact of reserve causation bias.