Multiscale entropy of physical activity as an objective measure of activity disorganization in a context of schizophrenia

Abstract

Retardation of physical activity is a criterion used in the evaluation of symptoms and for diagnosis in mental disorders. Although evaluation of physical health status is mostly based on the patient’s report during clinical appointments, long-term measures of physical activity in a given patient at home would be more accurate in informing on the health status of this patient. In this study, Multiscale Entropy, a robust measure of signal structural organization over multiple temporal intervals, was applied to actigraphically recorded, long-term wrist-activity data of 20 patients with schizophrenia and that of 21 gender and age-matched healthy controls. The aim was to test, whether this method allows extracting information from a given set of activity data to differentiate the physical health status of the patients from that of healthy controls. A multivariate Support Vector Machine was trained to provide a binary classification (patient or control) using cross-validation. We found that entropy-based features allow the classification into patients versus controls with 90.24% accuracy when using the two best Multiscale Entropy features. Notably, an entropy calculation on a scale of two minutes was found to be significantly different (p<0.001) between patients and controls and highly predictive for classification. It is suggested that further analysis of Multiscale Entropy derived metrics should be applied to other patient groups for the identification of physiological features related to physical retardation.