Hannah G Lawman1, M Lee Van Horn2, Dawn K Wilson2, Russell R Pate3. 1. Center for Obesity Research and Education, Temple University, USA. Electronic address: hlawman86@gmail.com. 2. Department of Psychology, University of South Carolina, USA. 3. Department of Exercise Science, Arnold School of Public Health, University of South Carolina, USA.
Abstract
OBJECTIVES: Popular methods for analyzing accelerometer data often use a single physical activity outcome variable such as average-weekly or total physical activity. These approaches limit the types of research questions that can be answered and fail to utilize the detailed, time-specific information available from accelerometers. This study proposes the use of multilevel modeling, which tested intervention effects at specific time periods. DESIGN: The motivating example was the Active by Choice Today trial. Simulations were used to test whether the application of time-specific hypotheses about when physical activity intervention treatment effects were expected to occur (e.g., after-school hours) increased power to detect effects compared to traditional methods. METHODS: Six simulation conditions were tested: (1) no treatment effects (to test the type 1 error rate), (2) time-specific effects, but no traditionally-tested effects, (3) traditionally-tested effects, but no time-specific effects, and (4) combinations of traditional and time-specific effects in 3 proportions. RESULTS: Results showed the proposed multilevel approach demonstrated appropriate type 1 error rates and increased power to detect treatment effects during hypothesized times by 31-38 percentage points compared to traditional approaches. This was consistent across varying proportions of traditional versus time-specific effects, and there was no loss of power using the multilevel approach when only traditional effects were present. CONCLUSIONS: The current study showed potential advantages of testing time-specific hypotheses about intervention effects using a multilevel time-specific approach. This approach may show intervention effects when traditional approaches do not. Future research should explore the application of this additional analytic tool for accelerometer physical activity estimates.
RCT Entities:
OBJECTIVES: Popular methods for analyzing accelerometer data often use a single physical activity outcome variable such as average-weekly or total physical activity. These approaches limit the types of research questions that can be answered and fail to utilize the detailed, time-specific information available from accelerometers. This study proposes the use of multilevel modeling, which tested intervention effects at specific time periods. DESIGN: The motivating example was the Active by Choice Today trial. Simulations were used to test whether the application of time-specific hypotheses about when physical activity intervention treatment effects were expected to occur (e.g., after-school hours) increased power to detect effects compared to traditional methods. METHODS: Six simulation conditions were tested: (1) no treatment effects (to test the type 1 error rate), (2) time-specific effects, but no traditionally-tested effects, (3) traditionally-tested effects, but no time-specific effects, and (4) combinations of traditional and time-specific effects in 3 proportions. RESULTS: Results showed the proposed multilevel approach demonstrated appropriate type 1 error rates and increased power to detect treatment effects during hypothesized times by 31-38 percentage points compared to traditional approaches. This was consistent across varying proportions of traditional versus time-specific effects, and there was no loss of power using the multilevel approach when only traditional effects were present. CONCLUSIONS: The current study showed potential advantages of testing time-specific hypotheses about intervention effects using a multilevel time-specific approach. This approach may show intervention effects when traditional approaches do not. Future research should explore the application of this additional analytic tool for accelerometer physical activity estimates.
Authors: Dawn K Wilson; Heather Kitzman-Ulrich; Joel E Williams; Ruth Saunders; Sarah Griffin; Russell Pate; M Lee Van Horn; Alexandra Evans; Brent Hutto; Cheryl L Addy; Gary Mixon; Susan B Sisson Journal: Contemp Clin Trials Date: 2007-07-17 Impact factor: 2.226
Authors: Richard P Troiano; David Berrigan; Kevin W Dodd; Louise C Mâsse; Timothy Tilert; Margaret McDowell Journal: Med Sci Sports Exerc Date: 2008-01 Impact factor: 5.411
Authors: Thomas L McKenzie; James F Sallis; Judith J Prochaska; Terry L Conway; Simon J Marshall; Paul Rosengard Journal: Med Sci Sports Exerc Date: 2004-08 Impact factor: 5.411
Authors: Kathryn R Hesketh; Alison M McMinn; Ulf Ekelund; Stephen J Sharp; Paul J Collings; Nicholas C Harvey; Keith M Godfrey; Hazel M Inskip; Cyrus Cooper; Esther M F van Sluijs Journal: Int J Behav Nutr Phys Act Date: 2014-01-09 Impact factor: 6.457
Authors: Sarah G Sanders; Elizabeth Yakes Jimenez; Natalie H Cole; Alena Kuhlemeier; Grace L McCauley; M Lee Van Horn; Alberta S Kong Journal: J Phys Act Health Date: 2019-07-17