Applying time series analyses on continuous accelerometry data-A clinical example in older adults with and without cognitive impairment

Abstract

Introduction Many clinical studies reporting accelerometry data use sum score measures such as percentage of time spent in moderate to vigorous activity which do not provide insight into differences in activity patterns over 24 hours, and thus do not adequately depict circadian activity patterns. Here, we present an improved functional data analysis approach to model activity patterns and circadian rhythms from accelerometer data. As a use case, we demonstrated its application in patients with mild cognitive impairment (MCI) and age-matched healthy older volunteers (HOV). Methods Data of two studies were pooled for this analysis. Following baseline cognitive assessment participants were provided with accelerometers for seven consecutive days. A function on scalar regression (FoSR) approach was used to analyze 24 hours accelerometer data. Results Information on 48 HOV (mean age 65 SD 6 years) and 18 patients with MCI (mean age 70, SD 8 years) were available for this analysis. MCI patients displayed slightly lower activity in the morning hours (minimum relative activity at 6:05 am: -41.3%, 95% CI -64.7 to -2.5%, p = 0.031) and in the evening (minimum relative activity at 21:40 am: -48.4%, 95% CI -68.5 to 15.4%, p = 0.001) as compared to HOV after adjusting for age and sex. Discussion Using a novel approach of FoSR, we found timeframes with lower activity levels in MCI patients compared to HOV which were not evident if sum scores of amount of activity were used, possibly indicating that changes in circadian rhythmicity in neurodegenerative disease are detectable using easy-to-administer accelerometry.