Algorithms and Digital Health Solutions for Wearable Sensor Data Acquisition, Processing and Interpretation in Chronic Diseases

Wearable devices have revolutionized healthcare by enabling real-time, continuous monitoring of patients in clinical settings and everyday environments. These devices provide valuable data for monitoring chronic conditions, predicting disease onset, and personalizing healthcare interventions. Despite their growing availability, challenges remain in data collection, processing, and exploitation. One main issue in data collection is supporting digital health solutions that seamlessly gather and manage data without increasing the burden on the wearer with manual tasks. The collected data faces challenges in processing and preparation. Ensuring data quality and reliability is crucial, as manual entry errors or improper device use can lead to inaccuracies that affect clinical assessments. Analyzing these data streams requires complex algorithms to filter out noise and extract meaningful insights. While several solutions have been proposed, most focus on non-consumer devices, highlighting the need for robust algorithms tailored to consumer wearables like smartwatches. The challenges of wearable device data collection and processing significantly impact their effectiveness in healthcare. Despite their efficacy, issues such as data interpretability and actionability hinder the clinical implementation of wearable devices. Additionally, while the collected data is often used for calculating statistics and trends, its true potential lies in personalized care through therapy tuning and optimization. This thesis addresses the challenges related to data acquisition, processing, and utilization by developing algorithms and digital health solutions specifically designed for chronic conditions such as Type 1 Diabetes (T1D), post-bariatric surgery hypoglycemia (PBH), amyotrophic lateral sclerosis, and multiple sclerosis. The thesis is divided into three parts, each aimed at describing the solutions developed to address these challenges. Part 1 describes the digital health solutions developed for two clinical studies. Both solutions adapt IMPACT, our core mobile platform for conducting clinical trials, detailed in Chapter 1. Chapters 2 and 3 present the solutions for the first study, which aimed to develop a non-invasive continuous glucose monitoring sensor (CGM), and the second, focusing on individuals with post-bariatric surgery hypoglycemia. Part 2 focuses on developing two solutions for wearable device signal processing. Chapter 5 describes the European Horizon 2020 BRAINTEASER project, during which we developed processing pipelines to prepare signals from a smartwatch for subsequent use. Chapter 6 details the developed pipeline, emphasizing clinical interactions at various development steps. Chapter 7 presents a Bayesian-based algorithm designed to improve the signal-to-noise ratio of the heart rate signal collected from the smartwatch using an adaptive methodology. Part 3 describes three solutions that use wearable device data in a clinical setting. Chapter 10 introduces a novel real-time algorithm to identify mealtimes, augmenting information from glucose CGM sensors with heart rate data for prompt detection. Chapter 11 discusses integrating a real-time glucose prediction algorithm into the mobile application from Chapter 3, allowing patients to receive alerts of imminent hypoglycemic events. Finally, Chapter 12 shifts focus to clinicians, detailing the development of ad-hoc visualizations and statistics to adapt the standardized reporting tool for T1D, the Ambulatory Glucose Profile, for the PBH population. Lastly, Part 4 is dedicated to conclusions and future perspectives. In Chapter 14, we summarize the issues related to implementing wearable devices in clinical settings, highlight the impact of this thesis, and discuss the main outcomes. We also outline possible future work directions, indicating pathways for research to improve the adoption of wearable sensors in healthcare.