We use cookies on this site to improve our service, perform analytics, and remember website preferences. By using this site, you consent to these cookies. For more information visit our Privacy Policy
A data-informed redesign of a diabetes self-management tool
Managing type 1 diabetes is never straightforward. HHF had the unique opportunity to build on an app we had previously designed following in-depth user research. We expanded the app’s features and helped conceptualize a service model around it that would respond to a more comprehensive picture we had formed about the people who use the app and their needs.
Client
Artificial Intelligence for Medical Systems (AIMS) Lab & the Harold Schnitzer Diabetes Health Center at OHSU
Funded by
The Leona M. and Harry B. Helmsley Charitable Trust
Services
Contextual Interviews, Service Design, UX/UI Design, Usability Testing
Challenge
Beginning in 2018, Researchers at Oregon Health Sciences University (OHSU) collaborated with HHF on an app called DailyDose that uses machine learning to help people living with type 1 diabetes optimize their insulin doses and health management routines. In an earlier case study, we showed how we designed the app to use 24/7 glucose monitoring and information about each person’s diet and activity to make custom tailored recommendations about how they can adjust their behaviour over the following week. The app was then evaluated in a randomized controlled trial which showed that people who followed more of the app’s recommendations saw greater improvements in their glucose management, but that they weren’t following recommendations 100% of the time.
In 2022, the research team reached out to HHF to help build a deep understanding of patient behaviour and answer a key question: if accepting more recommendations was associated with better health outcomes, what strategies could help encourage people to accept recommendations more frequently? By combining quantitative study data and primary research techniques, we were able to: reframe the design problem, make evidence-based iterations of the app’s interface, and work with OHSU to conceptualize a digitally enabled service model for type 1 diabetes care that meets the newly uncovered needs of this diverse population.
If accepting more recommendations was associated with better health outcomes, what strategies could help encourage people to accept recommendations more frequently?
Process
Discovery
To develop a deep understanding of why people were accepting or rejecting the in-app suggestions, we interviewed people who had used DailyDose during the study. Our discussions focused on their experience living with diabetes and the impacts the app had made on their daily management routines.
Because the participants had created months of real app usage data including what they ate, how much insulin they used, and which recommendations they accepted or rejected, we framed the interviews around their real-life data. This presented us with the unique opportunity to help participants remember their frame of mind and what situations influenced their behaviours. It also helped us to get a balanced view of not just what people say they do, but what they actually do. This blending of life history and mental model interviewing with quantitative data produced rich insights.
Synthesis
First and foremost, we found that almost all users loved the DailyDose product and that even people who had not accepted all their recommendations found it useful and even exciting to have new ways of think about their data. However, we did learn that life is always a little bit unpredictable and people who were unsure about the app’s recommendations were often concerned that there was no way for them to adjust the algorithm for unusual events or to override recommendations that seemed at odds with what they knew about their own condition. HHF’s iterative design process always includes continuous refinements to product designs as we learn more about users over time. In this case, our in-depth research allowed us to identify key factors that impact people’s level of confidence and engagement with self-care and how the DailyDose service model could better respond to their needs.
What We Discovered
Advice should feel intuitive
There is a lot of math and interpretation of data involved in calculating insulin doses, and without context users could feel uneasy about following recommendations.
Emphasize the outcomes
Some users felt like some of their recommendations were pushing in opposite directions. Ensuring that the outcome of recommendations was clear would help with this.
Every day is different
Life is full of surprises like late meals, traffic jams, and walking a few blocks when transit is out of service. Ensuring that people can capture surprises will better optimize the app.
The most important insight was that, rather than aiming to have all users accept all recommendations right away, it is more important to build a service around the app that includes the person’s clinical team. Moving forward, our goal became to help clinicians and patients to have better conversations, allowing the clinical team and the patient to better understand what it takes to build a care plan that works for everyone involved.
It is better to create a human service that brings people along for the ride, understands their human needs, and how best to serve them.
Design
Our research findings and user requests informed the redesign of some features in the app so that users would feel more comfortable using it and following the app’s recommendations. We tested these changes with three previous users of DailyDose and five users who were new to the app so that we could validate their effectiveness and iterate as needed. These updates included:
“About this recommendation”
We added more context and clarity to the recommendations to help users understand why the app was providing them with recommendations and why they are important to follow
Flag days
We added the ability for users to flag certain days as atypical so that exceptions could be acknowledged. This made them feel more comfortable using the app at times when their data could show negative trends based on exceptions to their usual routines, or biological events like illnesses or menstrual periods. Flag days will also help researchers understand future UX changes and can guide machine learning for better understanding trends
Timing Controls
It was important to users that they would be able to control the timing of recommendations and snooze them until they are able to address them
Ability to tailor dosages
Being able to adjust the range of insulin was important to users who are comfortable making these decisions based on their schedules and momentary needs. When adjusting the bolus in the app, they would be able to see a real-time projection of the effect this would have, which would guide them in making an informed choice.
Tone of voice
We also refined the tone of voice of the writing, recommendations, and notifications coming from the app, allowing DailyDose to better connect with our relatively young user-base.
Outcomes
As this project progressed, we realized that changes in the app’s interface were only one piece of the larger challenge. The design of a service that is focused on individual rather than collective needs would result in recommendations that felt more relevant and personalized and would allow users to work with their care team on tailored, achievable goals. Our research process taught us that people who live with diabetes are experts in their own care, and rejected suggestions did not necessarily mean the user was not effectively self-managing. A service that understands an individual would encourage more uptake, and our goals became focused on personalizing the experience rather than encouraging 100% acceptance of the app’s recommendations. As we built a clinician dashboard to support this service (read more about that here), we took the same human-centred approach to ensure that the individual’s voice could guide their own care.
Framing the interviews around real-life data helped us to understand deeply why users were accepting or rejecting recommendations.