Abstract

The widespread use of wearable devices and smartphones has revolutionized the field of digital medicine, enabling non-invasive, continuous monitoring of various physiological parameters. This technological advancement is crucial for personalized medical and public health interventions. The data collected in free-living conditions are continuous (functional), collected at second or minute level resolution, and are multilevel, reflecting daily, weekly, and monthly patterns. This data complexity necessitates innovative statistical methods to address challenges like uncertainty quantification in predicting future patient trajectories.

This presentation aims to achieve two primary objectives. First, it examines the impact of digital medicine on routine medical tasks, including disease diagnosis, management, and treatment prescription, with a special focus on diabetes care and physical activity interventions. These areas exemplify the transformative effect of digital medicine. Second, the presentation introduces new statistical approaches developed by the author for quantifying uncertainty in outcome prediction. Tailored to the challenges of digital medicine, these algorithms are based on conformal prediction and Bayesian methods. Their practical applications are demonstrated using data from continuous glucose monitors and accelerometers, showcasing their effectiveness in various modeling tasks such as determining diagnosis, characterization of disease phenotypes, creation of new rules for monitoring physical activity profiles, and establishment of new biomarkers.