Imputing accelerometer nonwear time in children influences estimates of sedentary time and its associations with cardiometabolic risk

Abstract

Abstract Background A limitation of measuring sedentary time with an accelerometer is device removal. The resulting nonwear time is typically deleted from the data prior to calculating sedentary time. This could impact estimates of sedentary time and its associations with health indicators. We evaluated whether using multiple imputation to replace nonwear accelerometer epochs influences such estimates in children. Methods 452 children (50% male) aged 10–13 were tasked with wearing an accelerometer (15 s epochs) for 7 days. On average, 8% of waking time was classified as nonwear time. Sedentary time was derived from a “nonimputed” dataset using the typical approach of deleting epochs that occurred during nonwear time, as well as from an “imputed” dataset. In the imputed dataset, each nonwear epoch was re-classified as being as sedentary or not using multiple imputation (5 iterations) which was informed by the likelihood of a wear time epoch being classified as sedentary or not using parameter estimates from a logistic regression model. Estimates of sedentary time and associations between sedentary time and health indicators (cardiometabolic risk factor and internalizing mental health symptoms Z-scores) were compared between the nonimputed and imputed datasets. Results On average, sedentary time was 33 min/day higher in the imputed dataset than in the nonimputed dataset (632 vs. 599 min/day). The association between sedentary time and the cardiometabolic risk factor Z-score was stronger in the imputed vs. the nonimputed dataset (β = 0.137 vs. β = 0.092 per 60 min/day change, respectively). These findings were more pronounced among children who had < 7 days with ≥10 h of wear time. Conclusion Researchers should consider using multiple imputation to address accelerometer nonwear time, rather than deleting it, in order to derive more unbiased estimates of sedentary time and its associations with health indicators.

Keywords

• Missing data
• Multiple imputation
• Accelerometer
• Sedentary behaviour
• Child
• Health