Multistate Hidden Markov Model for High-Frequency Repeated Measures in Applications to Studies of Physical Activities with Accelerometers

• Activity Number: 186
• Type: Contributed
• Date/Time: Monday, August 4, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Jaejoon Song*+ and Karen Basen-Engquist
• Companies: MD Anderson Cancer Center and MD Anderson Cancer Center
• Keywords: Multistate models ; Hidden Markov models ; Accelerometry

Abstract

Accelerometers are non-invasive wearable devices to collect continuous physical activity data over time. Movement frequencies are recorded minute-by-minute in accelerometers to provide rich and objective measurement of physical activity: a typical record for an individual may have more than one thousand data points for a single day, with movement frequencies per minute ranging from zero to thousands. In common approaches to analyze these data, intensity of real time activities are classified based on certain cut points of the movement frequencies (e.g. light, moderate, hard and very hard activities). While the classifications provide convenient translations of the movement frequencies, numerous studies report considerable classification errors arising from population characteristics and environmental factors. Multistate hidden Markov models allow for full consideration of real time transitions of the physical activities, while assuming that the true states of physical activities are unobserved. In this study, multistate Markov models are used to explore physical activity characteristics of endometrial cancer survivors.