Wearable systems for monitoring and analyzing physiological signals in older subjects

Population ageing is reshaping healthcare needs and increasing demand for scalable tools that can monitor health in daily life and support early risk detection. This PhD thesis develops and evaluates wearable sensor methods to quantify physical activity, sleep, and cardiovascular function and to translate measures of these tightly coupled domains into cohort-scale and clinically relevant applications. Section I introduces the state of the art in wearable sensing and outlines the shared methodological foundation used throughout the thesis. It describes the main sensor modalities (accelerometry, wearable electrocardiography (ECG), and photoplethysmography (PPG)), common preprocessing steps for real-world data (data ingestion, calibration, non-wear detection, and synchronization), and the evaluation principles used to compare wearable-derived outcomes against reference standards. Section II focuses on algorithm evaluation and benchmarking. It includes systematic reviews and open-source comparisons of methods for heart rate estimation from motion-corrupted PPG and for PPG artifact detection, and a multicohort evaluation of an accelerometer-based sleep detection algorithm across sensor locations and polysomnography datasets. Section III moves beyond single-modality outcomes by illustrating a multisensor system to characterize nocturnal movement events and their cardiac autonomic correlates, and by evaluating PPG pulse wave features for blood pressure stratification real-world recordings. Finally, Section IV demonstrates translation to clinical utility at scale. It presents an ECG-based cardiovascular risk score learned from foundation-model embeddings and trained using deep phenotyping, then tests its ability to predict future cardiovascular events in an independent clinical cohort. This section also describes scalable wearable data processing pipelines deployed in large observational studies of older adults and caregiver populations (DARE-FALLSPREDICT and iCare.IT). Overall, the thesis provides a practical and generalizable framework for turning real-world wearable data into reliable digital biomarkers to support healthy ageing and risk stratification.