Automating Data Personas for Designing Health Interventions.

In user experience (UX) research, persona development can be particularly beneficial in designing digital health interventions. From our previous work, we identified challenges that diabetic patients face in adhering to guideline-recommended care through persona development. We developed data personas from Electronic Health Records (EHR) using two-step clustering and combined them with proto personas that were generated by a group of medical experts for the same patient population. Not only did our results prove beneficial for intervention design, but also highlighted fairness issues that may result from the underrepresentation of certain populations in EHR datasets. In our current paper, we validate the results of our prior work and we extend our previous work on persona development to build a model that can automatically associate a patient record to its representative data persona. The model was built using the results of the two-step clustering as the inputs. We believe that building such a model will help clinicians quickly trace new patient records to already developed personas, thereby assisting in immediate health interventions based on the health risk characteristics of the patient.