Visualizing Diabetes Adherence Data

Creating a tool to help doctors predict diabetes adherence in patients.

A visualization project by Janell Joyner, Simone Pimento, Dan Yang.

Sugar. (Photo by Sharon McCutcheon on Unsplash.)

An estimated 415 million people in the world are living with diabetes. Less than 50% of these people adhere to their medication requirements. Studies performed to better understand what impacts adherence have found education, socioeconomic status and types of recommended therapies, among numerous additional factors to be the key factors. Increasing treatment adherence would lead to decreased hospitalization, diseases, and death caused by complications of diabetes. This information is used to improve glucometer designs, however no tools appear to exist for helping doctors quickly evaluate which glucometer is right for their patient based on their adherence information.

Our goal was to create an interactive data visualization that will allow doctors to easily view factors that impact adherence over time for various patient characteristics and manufacturers. A data visualization tool like this could provide more information to a doctor before they select an adherence treatment. Doctors that regularly collect information from their patients could compare what they see to this visualization or something like it. If they notice changes in adherence that mirror the visualization, then they could generate a list of questions to ask or adjust the person’s adherence. If they see something new, then they could ask questions to gain a better understanding of what is impacting adherence for this patient specifically.

Our Process

We followed a 4 step process to create our final visualization. We began with understanding our data, doing an extensive literature review, followed by sketches and finally creating our visualization.

Our Dataset

The dataset we used is from a diabetes study performed in 2010. The patient information was anonymized and contained a wide variety of information about the 99 subjects that participated in the study. This included age, race, education, non-diabetic medical issues, diabetic related medical issues, as well as the type of glucometer that they used. Data about the subject’s adherence to checking their blood glucose level was collected over 60 days. We attempted to understand and visualize how a person’s adherence changed over time to catch any correlations with performance shaping factors.

Literature Review

We reviewed 22 papers to understand the current medical tools or visualizations that may exist for diabetes adherence. While some companies have been working on visualization tools, much of this has been performed on the patient side. Glucometer manufacturers are making glucometers that can be paired with smartphone applications like Accu-Chek and OneTouch Reveal.

Photo by Kate on Unsplash

These smartphone applications can generate pie charts, line graphs, bar graphs, and various other graph types to show an individual how their blood glucose level has changed over time. The user might be able to input additional information about their self depending on the complexity of the application to help monitor factors that are keeping them from reaching their blood glucose level goals.

These applications are limited in their interactivity and the visualizations are 2-D in nature and do not compare many factors against each other. However, they may also provide a way for people to share their information with their doctor or upload it to a public database for others to analyze.

To compare individually generated information as a group, researchers are generating interactive visualizations using Tableau and R. Using commercially available software such as Tableau or an open source programming platform like R has its pros and cons depending on the type of graph or interactivity desired. For example, Tableau offers more interactivity than R for scatter plots due to the ease of being able to select which data points to examine. While R with ggplot2, offers the ability to customize bar graphs with extra labels which is not as easy to do in Tableau.

A library has been developed in R called AdhereR specifically for computation and visualization of adherence to medication treatments. This library is also designed so that it can be easily integrated into EMRs used by doctors.

Issues that current tools are facing

The biggest issues faced by current tools is the lack of engagement with the data, the quality of the data, and the overwhelming amount of data being generated. There is an increase in the amount of data available for analysis due to the increase in glucometers with mobile applications and integration of other sensors. If the information seems too daunting or is not particularly interesting, people would not want to engage with it. The data being generated by individuals is not guaranteed to be complete data. Chunks of data can be missing for various reasons ranging from forgetfulness to technical malfunctions. The missing data will cause distortions in visualizations if they are not designed to account for this. The size of the data also causes issues of where to start with generating a visualization. A visualization should be able to help make informed decisions about the information quickly, compare relevant pieces of data against one another, and these tools should be compatible across multiple platforms.

For the purpose of our visualization, we took a look into the aforementioned applications. We considered the way that data is presented in these applications to be a starting point for our visualization. We got an idea of what kind of information and features people found useful to them. We got insight into the tools used to create visualizations with similar data which helped us form a path to creating our own.

Sketches

Since our data was multidimensional we wanted to combine visual variables to get the most impact without overwhelming the viewer. During the literature review we saw different groups use box plots, line graphs, and bar charts to display their data. These graphs utilized combinations of visual variables such as using color or shapes to add an extra dimension to the data. Though these data sets were usually done to look at one subject or a small group of subjects and did not focus on comparing many PSFs against each other. Our goal was to create an interactive visualization tool that would allow doctors to select which PSFs they wanted to examine, generate a graph comparing all the data that fitted into the selected PSFs, and then have the ability to compare their own patient against that data if they had data to compare.

This graph displays how many times a certain group of participants checked their blood glucose level over the course of a 60 day experiment. In this design draft, the participants are grouped based on the score they received after completing an anxiety survey. The groupings are denoted by different textures. Each day of the experiment would marked along the x-axis. The higher the points are along the y-axis mark the average number of times each group checked their blood glucose levels. This graph can help with seeing trends in the data amongst the grouped data. This graph set-up could get potentially confusing if the textures are not distinct enough. The groupings might need to be adjusted to make sure no trends are being excluded or artificially created.

This sketch is a scatterplot matrix comparing multiple PSFs over the course of the experiment. The x-axis denotes the adherence time and the y-axis denotes medical adherence. The adherence time could be grouped to reduce the number of graphs that need to be created. Each scatterplot will show how adherence changed over time with each PSF or comparing two PSFs. While different shapes can be used to show off the differences, this could be forcing too much data onto one chart.

These are small graphs of multiple PSFs and how subject subgroups overall average adherence. The textures denote subject subgroups (ex. separation by age, education, depression level, etc.). This graph can also be used to see general trends, but does not show dips in adherence over time.

This pie chart, tries to display a relation between adherence and predicted provider adherence. Each sector can also be further color coded.( based on distance from the center), to show how intended frequency might have an impact on adherence.

This graph shows the relation between various PSFs and doing an unsafe act. Each PSF is color coded and displayed on a graph, using dots. (one for each participant). Each unsafe act has its own graph. The number of PSWs would have to be restricted according to their relevance.

Final Designs

Our final graphs and interactions were generated in Jupyter Notebooks. Users can view a list of performance shaping factors(PSFs) and select the ones they want to visualize. The pie chart shows the daily adherence groups the participants fell under.

Average Daily Adherence Percentage for various age groups

Screenshot of the dashboard, with different performance shaping factors

Demo of our data visualization tool

Findings

• We found strong correlations between age of diagnosis and adherence. Adherence levels were much higher for people diagnosed at an early age while adherence decreased as the age of diagnosis increased.
• People with lower anxieties ie(0–7) did not seem to follow a trend with their adherence. However, people with higher anxiety scores seemed to have a higher daily adherence.
• The depression scores were interesting to view. The graph showed that people with higher depression scores seemed to have a higher average adherence percentage. This may need additional analysis against the anxiety scores.
• Motivations scores did not display any interesting trends in the data.
• People who felt they had a higher support score, had a higher adherence percentage as compared to people who had a lower support score.

Next Steps

Our next steps include adding more interactivity, creating more visualizations with unsafe acts and further analyzing the data for interesting trends.

References

• Bailey, C. J., & Kodack, M. (2011). Patient adherence to medication requirements for therapy of type 2 diabetes. International Journal of Clinical Practice, 65(3), 314–322. https://doi.org/10.1111/j.1742-1241.2010.02544.x
• Browne, S. H., Behzadi, Y., & Littlewort, G. (2015). Let Visuals Tell the Story: Medication Adherence in Patients with Type II Diabetes Captured by a Novel Ingestion Sensor Platform. JMIR MHealth and UHealth, 3(4). https://doi.org/10.2196/mhealth.4292
• Cohen, D. J., Keller, S. R., Hayes, G. R., Dorr, D. A., Ash, J. S., & Sittig, D. F. (2016). Integrating Patient-Generated Health Data Into Clinical Care Settings or Clinical Decision-Making: Lessons Learned From Project HealthDesign. JMIR Human Factors, 3(2), e26. https://doi.org/10.2196/humanfactors.5919
• Delamater, A. M. (2006). Improving Patient Adherence. Clinical Diabetes, 24(2), 71–77. https://doi.org/10.2337/diaclin.24.2.71
• Dima, A. L., & Dediu, D. (2017). Computation of adherence to medication and visualization of medication histories in R with AdhereR: Towards transparent and reproducible use of electronic healthcare data. PLOS ONE, 12(4), e0174426. https://doi.org/10.1371/journal.pone.0174426
• García-Pérez, L.-E., Álvarez, M., Dilla, T., Gil-Guillén, V., & Orozco-Beltrán, D. (2013). Adherence to Therapies in Patients with Type 2 Diabetes. Diabetes Therapy, 4(2), 175–194. https://doi.org/10.1007/s13300-013-0034-y
• Gotz, D., & Borland, D. (2016). Data-Driven Healthcare: Challenges and Opportunities for Interactive Visualization. IEEE Computer Graphics and Applications, 36(3), 90–96. https://doi.org/10.1109/MCG.2016.59
• Hood, M., Wilson, R., Corsica, J., Bradley, L., Chirinos, D., & Vivo, A. (2016). What do we know about mobile applications for diabetes self-management? A review of reviews. Journal of Behavioral Medicine, 39(6), 981–994. https://doi.org/10.1007/s10865-016-9765-3
• Hsu, W. C., Lau, K. H. K., Huang, R., Ghiloni, S., Le, H., Gilroy, S., … Moore, J. (2016). Utilization of a Cloud-Based Diabetes Management Program for Insulin Initiation and Titration Enables Collaborative Decision Making Between Healthcare Providers and Patients. Diabetes Technology & Therapeutics, 18(2), 59–67. https://doi.org/10.1089/dia.2015.0160
• Kebede, M. M., & Pischke, C. R. (2019). Popular Diabetes Apps and the Impact of Diabetes App Use on Self-Care Behaviour: A Survey Among the Digital Community of Persons With Diabetes on Social Media. Frontiers in Endocrinology, 10. https://doi.org/10.3389/fendo.2019.00135
• Mobile Apps | Accu-Chek. (2019). Retrieved October 3, 2019, from Accu-Chek® website: https://www.accu-chek.com/tools-tech/mobile-apps-powerful-data-your-fingertips
• Morrison, C., & Doherty, G. (2014). Analyzing Engagement in a Web-Based Intervention Platform Through Visualizing Log-Data. Journal of Medical Internet Research, 16(11), e252. https://doi.org/10.2196/jmir.3575
• Now Convert Your Smartphone into Smart Blood Sugar Meter. (2019). Retrieved October 3, 2019, from MyDario.com website: https://mydario.com/smart-meter/
• Ogden, J., & Sidhu, S. (2006). Adherence, behavior change, and visualization: A qualitative study of the experiences of taking an obesity medication. Journal of Psychosomatic Research, 61(4), 545–552. https://doi.org/10.1016/j.jpsychores.2006.04.017
• OneTouch Reveal® Mobile and Web Apps. (2019). Retrieved October 3, 2019, from OneTouch® website: https://www.onetouch.com/OneTouchReveal
• Polonsky, W. H., & Henry, R. R. (2016). Poor medication adherence in type 2 diabetes: Recognizing the scope of the problem and its key contributors. Patient Preference and Adherence, 10, 1299–1307. https://doi.org/10.2147/PPA.S106821
• Schneider, J., Börner, D., Van Rosmalen, P., & Specht, M. (2015). Augmenting the Senses: A Review on Sensor-Based Learning Support. Sensors, 15(2), 4097–4133. https://doi.org/10.3390/s150204097
• Shneiderman, B., Plaisant, C., & Hesse, B. W. (2013). Improving Healthcare with Interactive Visualization. Computer, 46(5), 58–66. https://doi.org/10.1109/MC.2013.38
• Smith, B., Frost, J., Albayrak, M., & Sudhakar, R. (2006). Facilitating narrative medical discussions of type 1 diabetes with computer visualizations and photography. Patient Education and Counseling, 64. Retrieved from http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.79.8123&rep=rep1&type=pdf
• Van Wijk, B. L. G., Klungel, O. H., Heerdink, E. R., & de Boer, A. (2006). Refill persistence with chronic medication assessed from a pharmacy database was influenced by method of calculation. Journal of Clinical Epidemiology, 59(1), 11–17. https://doi.org/10.1016/j.jclinepi.2005.05.005
• Wang, J., Wongvibulsin, S., Henry, K., & Fujita, S. (2018). Quantifying and Visualizing Medication Adherence in Patients Following Acute Myocardial Infarction. AMIA Annual Symposium Proceedings, 2017, 2299–2303.
• Wood, E., Yang, Q., Steinberg, D., Barnes, A., Vaughn, J., Vorderstrasse, A., … Shaw, R. J. (2019). Diabetes Mobile Care: Aggregating and Visualizing Data from Multiple Mobile Health Technologies. AMIA Summits on Translational Science Proceedings, 2019, 202–211.