Real world multimodal assessment of physical behavior change in health and disease

Sedentary lifestyle is currently considered a global pandemic, associated with major health problems such as cardiovascular disease and premature death. Regular physical activity (PA) is one way to address this problem, as it brings a variety of health benefits, including reducing the risk of non-communicable diseases and improving mental health. Although much research has been conducted on the beneficial effects of PA, there are still unanswered questions regarding the key determinants associated with PA and sedentary behaviors, or the health-related effects of specific PA interventions. Today, advances in wearable technologies offer unique opportunity to monitor and understand the impact of PA interventions in a naturalistic context. However, the ability to relate PA behavior to health depends critically on its accurate measurement and the possibility to extract relevant information. The proposed work aimed to develop multimodal analysis tools to assess the behavioral, physiological, and psychological dimensions of real-world PA in healthy and patient populations. Algorithms were developed and validated to accurately characterize locomotion periods in daily life, including duration and intensity, using foot- or trunk-mounted inertial measurement units. Heart-rate (HR) and heart rate variability (HRV) metrics, generally used as proxies for the autonomous nervous system, were also assessed to evaluate the effects of PA interventions on health status. Furthermore, we evaluated the feasibility of extracting breathing rate from respiratory sinus arrhythmia during moderate to high-intensity exercise. Breathing rate is considered as a marker for physical efforts and recovery, and thus the proposed approach holds promise for yielding further physiological insights, without requiring an additional sensor. The second part of this work consisted of applying the proposed methods to longitudinal studies for evaluating the effect of PA interventions on two populations: (i) patients with multiple sclerosis (MS), and (ii) young and healthy women. In the first intervention, our study provides comprehensive evidence of the beneficial effects of an intensive rehabilitation program on self-reported health questionnaires and walking capacity measured with supervised walking tests. However, these improvements did not translate to home performance in patients with severe MS or only marginally in patients with mild MS. In the second intervention, the effectiveness of an eight-week running program was analyzed during daily activities using a single chest-mounted device, with results showing increased vigorous PA after intervention in half of the participants. Furthermore, participants who adhere to the intervention reported feeling more vital after the running sessions, suggesting that a positive association with running is important for PA behavior change. Our results indicate that motivational and behavioral factors should also be considered and incorporated into intervention strategies to further enhance PA participation.Overall, the signal processing algorithms and the versatile multimodal approach developed in this thesis enable accurate extraction of PA and HR/HRV-derived metrics in real-life contexts. Based on this information, physical therapists and researchers can assess the effects of specific interventions, and identify determinants and barriers of behavior change.