UC Irvine

Well-being is a crucial factor in human lives and society insofar as it is an indicator of satisfaction. Within the pillars of well-being, we favor sleep, physical activity, and mental health because these can represent body health. Furthermore, the entire world has been affected widely by a global virus pandemic which could significantly impact societies with vulnerable factors of well-being. Hence, we have investigated the effect of COVID-19 as one of the representatives of threats to social well-being. Parties interested in illness prevention and health promotion may find it helpful to measure, monitor, and promote well-being. Through the advancement of the Internet of the Things (IoT), it is now possible to monitor health outcomes and biomarkers in everyday free-living conditions without needing to proceed to labs or clinical settings. Taking the above into consideration, we can organize the main contribution of this dissertation into two components. First, we examine the trends and patterns of sleep and mental health disorders at a population level. To do so, we evaluate the sleep parameters of the Oura ring and the Samsung Gear Sport watch compared with a medically approved actigraphy device in a midterm everyday setting, where users engage in their daily routines. We used home-based sleep monitoring to examine the sleep characteristics of 45 healthy people (23 women and 22 men) for 7 days. Then we investigate the sleep trends of 38 pregnant women during the COVID-19 lockdown in Finland. The subjects used the Samsung Gear Sport smartwatch, and their sleep data was recorded. Subjective sleep reports were obtained using a smartphone app designed specifically for this study. Later, we analyze different mental health disorder reports before and during the pandemic and discuss the most vulnerable population. The benefit of such investigations is that capturing real-time information and public attitudes would facilitate policymakers to monitor public health and social wellness.

In the second part, we focus on individual-level analyses. We use Machine Learning and Deep Learning techniques to monitor, reconstruct, evaluate, and forecast various tasks utilizing individuals' data and biomarkers. We begin by reconstructing the blood pressure signal. Continuous blood pressure (BP) monitoring can help individuals manage their chronic diseases such as hypertension, requiring non-invasive measurement methods in free-living conditions. Recent approaches fuse Photoplethysmograph (PPG) and electrocardiographic (ECG) signals using different machine and deep learning approaches to estimate BP non-invasively; however, they fail to reconstruct the complete signal, leading to less accurate models. We propose a cycle generative adversarial network (CycleGAN) based approach to extract a BP signal known as ambulatory blood pressure (ABP) from a clean PPG signal. Our approach uses a cycle generative adversarial network that extends the GAN architecture for domain translation and outperforms state-of-the-art methods by up to 2x in BP estimation. Next, we focus on patients diagnosed with acute respiratory distress syndrome (ARDS) who are in more life-threatening circumstances when it comes to COVID-19, resulting in severe respiratory system failure. We investigate the behavior of COVID19 on ARDS patients by utilizing simple vital signs. We analyze the long-term daily logs of blood pressure (BP), and heart rate (HR) associated with 150 ARDS patients admitted to five University of California academic health centers to distinguish subjects with COVID-19 positive and negative test results. In addition to the statistical analysis, we develop a deep neural network model to extract features from the longitudinal data. Our deep learning model achieved 0.81 area under the curve (AUC) to classify the vital signs of ARDS patients infected with COVID19 versus other ARDS diagnosed patients. Finally, we designed and developed a study to recommend personalized exercises to non-pregnant subjects to increase their physical activity level. We developed smartphone and smartwatch applications to collect, monitor, and suggest exercises using a contextual multi-arm bandit framework. This study includes constructing and developing a personalized model that predicts or recommends different actions depending on individual user biofeedback.