Choice of Processing Method for Wrist-Worn Accelerometers Influences Interpretation of Free-Living Physical Activity Data in a Clinical Sample

in Journal for the Measurement of Physical Behaviour
View More View Less
  • 1 Iowa State University
  • | 2 University of Tennessee–Knoxville
  • | 3 St. Ambrose University
  • | 4 University of Iowa
  • | 5 Vanderbilt University
Restricted access

Purchase article

USD  $24.95

Student 1 year online subscription

USD  $39.00

1 year online subscription

USD  $52.00

Student 2 year online subscription

USD  $74.00

2 year online subscription

USD  $99.00

Wrist-worn accelerometers are increasingly used to assess free-living physical activity (PA), but the implications of different processing methods are not well characterized. To advance research in this area it is important to better understand how choice of processing method influences estimates of free-living PA behavior. This study compared PA profiles resulting from processing wrist-worn data collected under free-living conditions using four different methods in a clinical sample of 160 women with chronic pain, a condition for which PA serves as a treatment. Participants wore monitors on their non-dominant wrist for 7 days and completed a self-report PA measure. Processing methods were Hildebrand linear, a modified nonlinear Hildebrand, Staudenmayer linear, and Staudenmayer random forest. Using each method, minutes per day in sedentary, light, and moderate-to-vigorous PA (MVPA) were estimated and individuals were classified as meeting PA guidelines based on their accumulated MVPA. Comparisons of outcomes among processing methods and with self-reported PA were made using repeated measures ANOVA, correlations, and kappa statistics. With few exceptions, estimated time at each intensity differed significantly across processing methods and with self-report (p < .001). Correlations between methods ranged widely (ρrange = 0.09 to 1.00) and showed inconsistent agreement for classifying individuals as meeting PA guidelines (κrange: −0.02 to 0.90). Thus, choice of processing method significantly influenced conclusions regarding free-living PA. Researchers and clinicians should exercise caution when interpreting accelerometer activity data and comparing across existing studies using different processing methods when examining how PA influences clinical conditions.

Ellingson and Welk are with the Iowa State University, Ames, IA. Hibbing is with the University of Tennessee–Knoxville. Dailey is with St. Ambrose University, Davenport, IA. Rakel, Sluka, and Frey-Law are with the University of Iowa, Iowa City, IA. Crofford is with the Vanderbilt University, Nashville, TN.

Ellingson ( is corresponding author.

Supplementary Materials

    • Supplementary Table (PDF 271 KB)