Validity of harmonised self-reported physical activity behaviours in UK Biobank: a doubly labelled water study.

Abstract

UK Biobank includes multiple instruments reporting various self-reported behaviours which can be used collectively to provide an estimate of physical activity (PA) volume. Our objective was to derive PA volume by calibrating self-report data to wrist accelerometry and examine validity against PAEE assessed using the gold standard doubly labelled water (DLW) method. Our validation study included 53 participants with complete selfreport data and DLW-assessed PAEE. PA volume expressed as average daily wrist acceleration (milli-g) was derived using a mutually-adjusted prediction equation from the self-reported behavioural variables in the accelerometry subsample (n=91653) of the UK Biobank cohort. We further converted PA volume from milli-g to PAEE using a previously published equation derived in 1050 UK adults. We examined the validity of self-predicted vs DLWassessed PAEE using Pearson correlation, mean bias, root mean square error (RMSE) and 95% limits of agreement. The self-reported behaviours explained approximately 4% variance of daily average acceleration in both women (n=54914) and men (n=42874) in the UK Biobank subsample. Following conversion from milli-g, mean(SD) predicted PAEE was 41.4(3.9) kJ/day/kg with range 33.1 to 51.0 kJ/day/kg. Mean(SD) of the criterion DLW-assessed PAEE was 51.8(17.4) kJ/day/kg with range 8.6 to 90.8 kJ/day/kg. Comparing predicted PAEE with DLW-assessed PAEE, mean bias was -10.5 kJ/day/kg (95%CI: -14.8, -6.1), RMSE was 18.8 kJ/day/kg (36% of the mean DLWassessed estimate), and 95% limits of agreement were -42.0 to 21.1 kJ/day/kg. Predicted PAEE and DLW-assessed PAEE were moderately correlated (r=0.51; p=.0001). Harmonisation of multiple self-reported behavioural variables using wrist accelerometry produces a single estimate of PA volume with good relative validity vs PAEE from goldstandard assessment; however absolute validity is weak. The prediction equation can be used to rank PA volume in the full UK Biobank cohort.