Diabetes affects a growing portion of the population and, like many chronic diseases, it is primarily managed by patients themselves without day-to-day input from doctors. Keeping blood glucose within a healthy range is important for prevent long-term complications of diabetes, but many patients with Type 2 diabetes do not achieve this. As a result, it is important for patients to be invested in and knowledgeable about their treatment goals and plan. One promising approach to address this is shared decision-making (SDM), where a patient and clinician work together to understand the patient's preferences and collaboratively formulate a treatment plan. SDM can potentially increase patient trust and satisfaction, but patients, doctors, and other caregivers begin with different sets of beliefs about disease and treatment. This creates challenges for SDM, as each participant may have a different understanding about what will result from an action, and when a patient's beliefs differ from information provided by a doctor this can lead to communication challenges and reduced trust. Further, treatment guidelines generally focus on one factor at a time, like the role of exercise or nutrition, and are rarely personalized to individuals. Causal models could potentially be used to help people understand the link between their goals and actions, but they can be too complex for people to reason with. This project will lead to methods that can automatically learn personalized causal models that are specific to the decision-making situation and individual's health, and communicated in the context of an individual's knowledge. This work will close the gap from data to decisions by bridging computational methods for causal inference, insight into the cognitive processes underlying decisions, and shared decision-making. The project will also aim to reduce treatment disparities by creating training modules to educate clinicians about patient beliefs and how these influence trust and decision-making.
Motivated by improving outcomes in Type 2 Diabetes (T2D), this work will fundamentally advance computational methods, and our understanding of cognition. While factors affecting blood glucose differ considerably between individuals, prior work has focused on finding population-level models. To address the need for personalized guidance, this work 1) develops novel approaches for finding personalized causal models (e.g. individual factors affecting blood glucose) from limited personal data by leveraging simulation, and 2) develops personalized abstractions of the inferred models, taking into account patient preferences and decision context, to reduce cognitive burden. This allows more relevant information to be delivered during decision-making. Since decisions are made in the context of existing knowledge, the second core focus of the project is linking causal models and mental models. While prior work has examined differences in mental models, it has not shown how to reconcile models across individuals. This work develops new approaches to more efficiently and accurately elicit an individual's mental model, map the elicited model to inferred causal models, and reconcile differences across individuals. The approaches will be deployed in shared decision-making between patient-provider and patient-caregiver pairs for T2D management both online and in local clinics. The methods developed will be applicable to many types of shared healthcare decisions.
This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
