ESTIMATES AND APPLICATION OF PHYSICAL ACTIVITY DATA ARE NEGATIVELY IMPACTED BY LOW ACCELEROMETER ADHERENCE

PURPOSE: To investigate the influence of low accelerometer adherence on the prediction of physical activity, and the subsequent application of the data. METHODS: One-hundred participants (age=25.5±7.9 yrs; BMI=24.8±3.2 kg/m2; body fat=21.7±8.9%) were asked to wear a triaxial accelerometer 24 hrs/d, 7 consecutive days (raw). Height, weight, and body composition (Air Displacement Plethysmography technology) were also measured during the study. Low accelerometer adherence (10 HR; 10.4 ±0.5 hrs/d) was simulated by imputing zero counts for sleep times, then randomly imputing blocks of 60 min of zero counts during waking hours until each participant had 10 h of data. Sedentary time was defined as \u3c150 cpm, light physical activity (LPA) as 150-2,689 cpm, and moderate-to-vigorous physical activity (MVPA) as ≥2,690 cpm. One-way ANOVA with a Tukey post-hoc test was used to analyze the differences between raw and 10 HR. Linear regression was then used to investigate the relationships between physical activity and body composition (BMI and %body fat) for raw and 10 HR. Absolute percent error (APE) was also calculated to compare raw and 10 HR. RESULTS: Sedentary time (935.0±101.3 min/d) for raw was significantly lower (p\u3c.0001) than 10 HR (1,138.9±56.3 min/d), with an APE of 23.3%. Raw LPA (429.5±88.1 min/d), MVPA (75.4±30.3 min/d), and total physical activity (520.3±139.9 cpm) were all significantly higher (p\u3c.0001) than 10 HR (254.1±49.2 min/d, 47.0±19.2 min/d, and 318.5±84.2 cpm, respectively). APE for the differences in LPA, MVPA, and total physical activity were 40.7, 36.6, and 38.4%, respectively. Time spent in MVPA was significantly associated with %body fat for 10 HR (β=-.092; p\u3c.048), but not for raw (β=-.050; p\u3c.091). CONCLUSIONS: The use of low adherence datasets consistently resulted in profound amounts of measurement error when compared to “true” physical activity, as measured by 24 hr/d. In addition, the subsequent use of low adherence data could potentially impact the prediction of health markers, including the presence of Type 1 errors