Models of Human Activity Behavior in Health and Disease: Bring Together Monitoring Technology and Mathematics

Sunday, February 17, 2013
Auditorium/Exhibit Hall C (Hynes Convention Center)
Anisoara Paraschiv-Ionescu , Ecole Polytechnique Federale de Lausanne (EPFL), Lausanne, Switzerland
Eric Buchser , Hospital of Morges and University Hospital (CHUV), Morges, Switzerland
Christophe Perruchoud , Hospital of Morges, Morges, Switzerland
Blaise Rutschmann , Hospital of Morges, Morges, Switzerland
Kamiar Aminian , Ecole Polytechnique Federale de Lausanne, Lausanne, Switzerland
Background: One important aspect of human behavior consists of activity patterns that may contain a wealth of information about actions, intentions, emotions, personal attributes, physiological responses and the health of the person. In this context, daily-life physical activity (PA) becomes an important proxy measure to assess chronic diseases such as chronic pain, defined as a syndrome of multiple etiologies with consequences for psychological and psychosocial well-being, functionality and quality of life. We postulate that the disease-related behavioral features can be captured better by the temporal dynamics of activity patterns than by global quantitative metrics.

 Methods: PA was monitored in 60 patients suffering from chronic intractable pain and in 20 pain-free healthy subjects. The measurements were performed under free-living conditions during five consecutive days using body-fixed inertial sensors. The PA patterns were modeled, analyzed and quantified in the context of two different mathematical approaches:

(1)   the theory of stochastic bivariate point processes by modeling the PA pattern as a sequence of events, i.e. the sequence of activity-to-rest and rest-to-activity transitions;

(2)   the concept of dynamical structural complexity by mapping the various features of PA, i.e. type, intensity and duration into a ‘barcode’.  The complexity of PA barcode was quantified using the Lempel-Zip complexity end entropic measures.

 Results:The derived PA metrics were compared between age-matched groups of subjects with clinically different pain intensities. The statistical analysis indicated significant differences between groups (p<0.05). As hypothesized, the dynamics of PA patterns revealed clinically relevant information and provided empirical evidence for complex pain behavioral models classified as fear-avoidance, endurance and pacing.

 Conclusions: Ambulatory monitoring devices and advanced mathematical methods provide the opportunity to objectively quantify a number of PA features including dynamical structural complexity of long-term recorded activity patterns. The derived parameters are able to distinguish chronic pain conditions and may have other potential clinical applications in neurological diseases (depression, Parkinson, Alzheimer), geriatrics, cardio-vascular diseases.