Behavioral Phenotyping Approaches for Diabetes and Pre-Diabetes
Most interventions are deployed as one-size-fits-all which assumes that patients that differ in many ways will each respond similarly to a common intervention approach. However, we know this is not the case. Behavioral phenotyping could improve prediction of response to interventions and thereby allow targeting of specific interventions to specific patients. A significant challenge is the need to collect data beyond demographic information such as information on personality type, risk preferences, or social networks. This data is challenging to collect and the patients who are willing to provide it are likely different than those that do not engage, which hinders the generalizability of these models.
This project makes use of two ongoing clinical trials that leverage Way to Health to collect this information on two adult populations: 360 uncontrolled type 2 diabetics and 185 pre-diabetics. All patients have HbA1c levels at baseline and 6 months. We will create behavioral phenotypes for these patients and test whether this can improve prediction of response to interventions. We will then generate a list of observable data from the electronic health record and test for correlations with responses to the validated assessments to create behavioral phenotypes for 10,000 uncontrolled diabetics and 50,000 pre-diabetics who we identified but did not enroll in the trials. We will then evaluate if behavioral phenotypes can be used to improve prediction of changes in HbA1c in usual care. This model would yield significant methodological insights for behavioral phenotyping and produce relevant clinical applications at Penn Medicine and beyond.