2025 Theses Doctoral

Supporting the Work of Patients and Providers in Complex Chronic Illness

Pichon, Adrienne

Chronic illnesses, particularly poorly understood and complex conditions like endometriosis, present significant challenges for patients and healthcare providers. Endometriosis is a systemic, multifactorial condition affecting approximately 6–10% of women of reproductive age. It is characterized by highly variable and unpredictable symptoms, a lack of biomarkers, no cure, and individualized responses to treatment. This requires significant patient-provider collaboration and ongoing self-management. Despite advances in artificial intelligence (AI) and personal informatics systems for managing chronic illness, limited support exists for complex conditions like endometriosis, where significant uncertainty and variation impede care and management.

This dissertation seeks to understand the needs of individuals faced with the complex chronic illness that is endometriosis and address those needs, particularly through leveraging patient-generated data and personal informatics tools to support care. Throughout this work, we take a human-centered AI (HAI) approach to promote the perspectives of individuals and align their needs and priorities with the capabilities of the technologies we use.

In this research, we first document the work of patients and providers in caring for such a complex chronic illness, and elicit the data and technology needs of patients and providers (Aim 1). There, qualitative research methods, including focus groups and interviews, reveal that patients and providers face barriers in synthesizing complex health data, aligning perspectives, and navigating individualized management pathways.

Next, we develop and evaluate interpretable temporal phenotypes of health status (Aim 2). There, we use a probabilistic modeling approach to generate interpretable, temporal phenotypes of health status from self-tracked data, then validate these health status representations through user feedback and a real-world computational task.

Finally, we identify human and technical specifications for an intelligent system that provides adaptive self-management recommendations using reinforcement learning (RL) (Aim 3). There, we propose and implement a novel HAI framework, which facilitates conducting a mixed-methods study where we map and align human needs and values with technical capabilities and requirements. This research can inform the development of adaptive, explainable, and personalized self-management recommendations. Findings from this dissertation demonstrate the potential of computational approaches and novel intelligent systems to empower individuals with endometriosis by augmenting their understanding and use of their health-related data and self-management efforts with data-driven insights and AI-enabled intelligent systems.

This dissertation has several important contributions. First, this work both leverages and advances human-centered AI. The introduction of the Multi-Perspective Directed Analysis (MPDA) framework provides an approach to bridge human and technical needs in the design of AI-enabled systems. By aligning insights from end-users with the specifications of data science, MPDA operationalizes an HAI approach to design, offering a reproducible approach for other researchers seeking to address similar interdisciplinary challenges. This framework highlights the potential of HAI in translating patient needs into actionable computational design requirements and provides a blueprint for tackling open questions in health and other domains through human-centered AI. We also elaborate on several HAI principles, for example, how to empower patients through control of intelligent systems.

Second, this dissertation contributes to advancing personal informatics technologies. We have documented a range of technology gaps and opportunities to innovate solutions to address these gaps. The development of interpretable, temporal representations of health status and the requirements gathering for an AI-enabled personal informatics tool for individualized recommendations are both novel contributions. These innovations expand the literature on chronic illness support, particularly by demonstrating the potential of AI in addressing a particular real-world, complex health scenario. We also highlight and articulate several sociotechnical gaps where technologies cannot meet the complex needs of users at this time. Despite the barriers in translating these technologies into practical systems, this work explores how AI can enhance chronic disease management through pragmatic, user-centered solutions. Finally, we advance illness-specific endometriosis research.

This dissertation addresses an illness context with significant gaps in research and technologies to support care and management. By identifying the complex work undertaken by patients and providers in care and the associated needs of patients and their care teams, this research provides a foundation for designing tools and systems tailored to this context. While we propose HAI solutions for some of these gaps, we also highlight additional opportunities for computational and interactive systems that could better support individuals and care teams managing endometriosis.

As an ultimate goal, this thesis seeks to understand and address the needs of individuals caring for endometriosis, a complex and poorly understood chronic illness. In particular, we aim to develop HAI technologies to support patients and their care teams. Through the alignment of human needs and values with technological constraints, this research facilitates patient-centered care, empowers individuals with their data, and contributes to the broader fields of biomedical informatics, human-computer interactions, and human-centered AI.