Making Sense of Multidimensional Health Data to Manage Chronic Conditions: Designing to Support Episode-Driven Data Interaction

Abstract

People with chronic health conditions, such as diabetes, are now able to capture large amounts of health data every day owing to improved medical and consumer sensing technology. These data, known as patient-generated data, have immense potential to inform the care of chronic conditions, both individually by patients and collaboratively by patients and clinicians. Despite the increasing ability to capture personal health data, informatics tools provide limited support to enable routine use of data for disease management. Lack of support for making sense of different types of health data challenges informed decision-making and results in missed opportunities for improving care, leading to suboptimal control and poor health consequences. Motivated by these problems, my dissertation examines the data practices and decisional needs of patients and clinicians to design novel tools for the presentation of multidimensional health data and evaluates these tools in the context of Type 1 diabetes. It employs several qualitative methods that include interviews, observations, focus groups, diary study, think aloud sessions, and user-centered design. By examining how patients and clinicians interpret multiple streams of data from continuous glucose monitors and insulin pumps, I synthesized the episode-driven sensemaking framework, a novel framework that describes the different analytical stages through which multidimensional health data is made actionable. My work describes the four analytical stages of the episode-driven sensemaking framework that include episode detection, episode elaboration, episode classification, and episode-specific recommendation generation. I show that the episode-driven framework provides a promising basis to guide the design of tools for data-based sensemaking and decision-making as the different stages of the framework lend themselves to opportunities for combining computational and user agency in different ways. By examining existing data review platforms, I show that the exploratory nature of these tools makes them underutilized by lay users like patients, in addition to resulting in negative experiences, such as cognitive burden, misinterpretation, and misrepresentation of reality. Given the limitations of exploratory tools, the potential of the episode-driven framework in providing a basis for tool design, and the promise of data-driven narratives in communicating data to the lay users, I designed episode-driven data narratives to help patients review data from continuous glucose monitors and insulin pumps. An exploratory comparison of the episode-driven narratives with the commercially available data review platforms shows that the former improved data comprehension and patients’ ability to make decisions from data; and lowered the cognitive load of engaging with data. Additionally, in nuanced ways, episode-driven narratives enabled user agency in making decisions for self-care. Based on multiple studies to examine practices, and design and evaluate tools, I suggest that to support people in effectively leveraging multidimensional data for managing chronic conditions, tools must do the following - support effective problem-solving with data by creating a shared understanding of data between stakeholders, enable different types of assessments from data and help connect those assessments, and guide analytic focus using a scaffold (e.g., an episode-driven workflow) to organize and present evidence. One promising approach to implement these suggestions in the design of a tool is an episode-driven data narrative, an embodiment of the episode-driven sensemaking framework using narrative visualization techniques. By supporting the generation and presentation of episode-driven narratives from multidimensional data, tools can augment patients’ abilities to effectively inform self-care of chronic conditions with their data.