By Joyce Lee, MD

I love interactive data visualization (#dataviz).  It is one of the things that I definitely wanted to explore when I came out to the Bay Area on sabbatical, because I believe that it has great potential for helping both patients and clinicians with diabetes management.  The sheer volume of numbers available for this disease is overwhelming; we need #dataviz tools that can help us achieve greater understanding and make actionable clinical decisions to improve health.

This is what we usually see in clinic: numbers written down on a piece of paper.

Yes there are computer systems that link to blood glucose meters, but there are a number of complexities with the downloading of blood sugar numbers in clinic (which deserves an entire blog post sometime in the future).

You can see there is some visual analysis and annotation that we do perform, albeit primitive.  The circles represent high blood sugars (>150 mg/dl)and the triangles represent low blood sugars (<70 mg/dl). This is almost better than the cave painters don’t you think?

But even the minority of patients who download their BS to the computer, are viewing dashboards like this.

Pie charts, need I say more? I can extract some useful insights from these charts, which improve over the previous one I showed, but a few things strike me: (1) some of the scatter plots overlay weeks of data, which I don’t find helpful because you can’t tell how BS on a given day are responding and relate them to life events; (2) some visualizations show a lot of numbers in many of the sections, and it just becomes onerous to go through them and find trends; (3) many provide statistics (area under the curve, MAD%) which I think only a minority of families and children really understand; (4) although some of the software programs do provide interactivity and let you see the data at different time scales (day, week, month), if you change to a different view, you are stuck trying to remember in your head what you saw on a previous screen because you can’t see the multiple levels at once; (4) finally, I find that the user interface and design could use major improvement.

When I arrived to the Bay Area in the fall, I went to a meetup that I thought was a general #dataviz meetup, but it turns it actually was a d3 meetup.  I couldn’t understand a word of what people were saying (all codese) but I did get some feedback from the group on the pros and cons of different visualization software like tableau, processing, and d3, and after this chance encounter, I began thinking about how I might use d3 for my research projects.

It turns out that in the fall, Jeff Heer was teaching a class at Stanford about #dataviz, and I was fortunate to have the opportunity to have his students (Ben Rudolph and Reno Bowen) work on a diabetes visualization for their class project, and to find a trusted collaborator who was willing to share personal blood sugar data.

This is the prototype, inspired in part by a Trulia heatmap, which shows blood glucose data from a continuous glucose monitoring system (CGM).  You can link to the actual dynamic visualization here.

CGM gives a readout of interstitial blood glucose (BG) every 5 minutes.  The dashboard displays day, week, and month views of BG as a heatmap.  Red represents a high BG and blue represents a low (BG) sugar.  The more intensely red it is, the higher the BG, the more intensely blue it is, the lower the BG.  White represents a normal BG, and grey represents missing data.  Finally, the three small boxes at the top of the day view represent the % of time that individuals are high, low or normal for the month, week, or day.

I really love this prototype for a number of reasons:

1) You can see all of the data, not just pieces of it; you can scroll up and down the month view to see data that spans years.

2) It’s visual.  The colors tell the story   Numbers alone can be totally overwhelming, so the color gives it some visual order.

3) There is a day panel, a week panel, and a month panel that all can be viewed simultaneously, to give a greater sense of perspective.  For the month display, we purposely went with 3 months at a time on the right hand side since that is the interval at which we see our patients.

For the week display, we went with 2-3 weeks as we often will make changes in the insulin regimen using the last 2 weeks of data.

For the day display, I did want to be able to see the actual numbers because I as a clinician might not feel totally comfortable recommending changes in insulin doses based on a color shade, but the nice thing is that it’s a number on demand: it hides, but then only appears if you hover over the line with your mouse.

4) The interactivity is really smooth and all of the displays are connected.  When you click on a given day in the week view, the day view panel changes to match it and the black arrow on the right of the month view also moves to the corresponding week.  Likewise, when you click a day on the month view, the corresponding week is highlighted on the week view, and the day view changes to the corresponding day.

5)  Customizability for the user. You can click the bottom arrow of the week view, and the 3 week view shrinks to a 1 week view, allowing the viewer to simplify the visual space as needed.  Also the user can turn date labels on and off.

6) Finally, it has a great minimalist design and doesn’t have the notorious chartjunk.

Ben and Reno wrote up a summary paper of their project for the class.  They quote my effusive reaction in their paper:

Note to self: Stop using so many exclamation marks in a row.  But seriously, this is how passionate I am about #dataviz + #design + #diabetes.

This is just the first prototype. There is much more that I want to do with this visualization (adding additional datastreams like insulin data, location, physical activity, mood data).  I see this visualization as the starting point for the front end of a mobile technology web-based platform for helping clinicians and patients manage diabetes.  But even more importantly I am interested in talking with patients and families about what they want to see or use, given that they are the real experts with this disease (either through interviews and through crowdsourced initiatives check out medevice-users and this flickr page)  If you have feedback on this first prototype, please share it!  Hopefully there will be more to come soon!

Joyce Lee, MD is a pediatrician, diabetes specialist, and Associate Professor at the University of Michigan. . She is currently on sabbatical at Stanford University through the Center for Health Policy. She blogs about design and healthcare at joycelee.tumblr.com, where this post originally appeared.