Longitudinal Analysis of Real-time Momentary Pain Data in a Cohort of Osteoarthritis Patients
Kelli D. Allen, Center for Health Services Research in Primary Care, Durham VAMC 
*Cynthia J. Coffman, Center for Health Services Research in Primary Care, Durham VAMC 
Robert F. Woolson, Center for Health Services Research in Primary Care, Durham VAMC 

Keywords: longitudinal, pain variability, heterogeneous variance, location-scale mixed effect models

Understanding differences in daily pain levels as well as pain fluctuations and determining factors that are associated with pain and pain variability can greatly aid in the development of pain management strategies for osteoarthritis (OA) patients. We present a study on OA veteran and non-veteran patients where real-time momentary pain data were collected on a visual analog scale using a handheld computer. Measurements were taken on one weekday and one weekend day beginning after waking and then approximately every 2 hours. In this presentation we detail the analytic process and discuss the challenges encountered. Most importantly, we describe what was learned at each step of the process. In the first step, we examined within-day OA-related pain patterns and associated patient characteristics using summary features (e.g. mean daily pain, pain range, area under the curve). Based on the summary features analysis, we found that pain patterns differed substantially across individuals and that joint location (hip or knee vs. hand OA) was associated with greater within day pain range. To exploit more fully the real-time momentary nature of our data, we applied heterogeneous mixed effect models. Using these models we examined the effect of the time-varying covariate of activity level at the time of the pain recording as well as the effect of patient characteristics on pain levels and variability. Finally, we fit a mixed-effect location-scale model to allow for random subject effects for both the subject’s mean pain and the within subject variability. These location-scale models, while challenging to build, have considerable potential for the analysis and interpretation of longitudinal pain OA data.