Švihrová, Radoslava (2025). Adaptive Interventions and Causal Inference in Digital Health. (Thesis). Universität Bern, Bern
|
Text
25svihrova_r.pdf - Thesis Available under License Creative Commons: Attribution-Noncommercial (CC-BY-NC 4.0). Download (5MB) | Preview |
Abstract
Non-communicable diseases are the leading causes of morbidity and mortality worldwide, with lifestyle factors such as physical activity, sleep, stress and substance abuse playing a critical role in their prevention and management. Despite evidence-based behavioral change strategies, sustaining long-term adherence in lifestyle change programs remains a major challenge, with high dropout rates and declining engagement over time. The growing ubiquity of wearable devices and mobile health technologies offers new opportunities to address these challenges through continuous, real-world monitoring of physiological and behavioral data. Moreover, their proximity to users enables delivery of timely and context-aware support via digital health interventions. Recent advances in automated decision-making algorithms, particularly reinforcement learning, enable dynamic, real-time adaptation of intervention components to enhance engagement and adherence in lifestyle change programs. When combined with causal inference, these algorithms unify into a causal reinforcement learning framework, enabling decisions that leverage actionable lifestyle insights to optimize adaptive interventions. This thesis investigates the integration of multi-source longitudinal data, causal inference, and multi-armed bandits—tools from reinforcement learning—to develop adaptive digital health interventions that enhance engagement and adherence in lifestyle medicine, presented through five contributions. First, a comprehensive review synthesizes behavioral change theory with methods for designing adaptive interventions using causal inference and multi-armed bandits. Second, a feasibility study evaluates patient engagement with a mobile application enhanced by personalized digital health interventions that nudge adherence to cardiac rehabilitation programs. Third, Bayesian mixed-effects regression is applied to sleep biomarkers for modeling of coping capacity for burnout prevention. Fourth, data-driven causal discovery is employed to infer the structure of longitudinal wearable data, demonstrating alignment with existing literature and feasibility of consumer-grade metrics for monitoring. Finally, fully data-driven causal inference is performed on wearable-derived longitudinal data to quantify effects of lifestyle behaviors on health outcomes, illustrating the potential for causal reasoning in adaptive interventions. Collectively, these studies demonstrate the potential of combining continuous behavioral monitoring, causal reasoning, and adaptive decision-making to enhance engagement and adherence in lifestyle programs. The findings lay the groundwork for the development of intelligent, personalized digital health systems capable of supporting sustained behavior change and the creation of behavioral digital twins.
| Item Type: | Thesis |
|---|---|
| Dissertation Type: | Cumulative |
| Date of Defense: | 28 November 2025 |
| Subjects: | 000 Computer science, knowledge & systems 600 Technology > 610 Medicine & health |
| Institute / Center: | 08 Faculty of Science > Institute of Computer Science (INF) |
| Depositing User: | Sarah Stalder |
| Date Deposited: | 23 Dec 2025 17:11 |
| Last Modified: | 26 Dec 2025 11:16 |
| URI: | https://boristheses.unibe.ch/id/eprint/6999 |
Actions (login required)
 |
View Item |