Exploring and Visualizing Diabetes Data

I’m Jana. I just moved here and I gave a talk a few years ago at the New York Quantified Self about some self-tracking that I did with my diabetes data so this is a bit of a follow up in a way.

I have Type 1 diabetes and for quick and dirty for those of you who aren’t familiar with the different types. This is the type of diabetes where you don’t produce insulin anymore. Your immune system is hacked. Your pancreas basically kills off the cells that produce insulin. So me and other people with Type 1 diabetes are dependent on synthetic insulin to survive to live. So that’s either by infusion with some kind of insulin pump like the one I have here, or by injection but pumps are pretty much becoming dominant these days.

And one of the problems here and the reason why insulin it a treatment not a cure is that it’s really a potent and basically a dangerous drug. You can kill yourself with too much insulin very very easily, even with someone who doesn’t have diabetes.

And it’s difficult to dose properly and part of that is because it’s injected or infused subcutaneously but that’s not an ideal mechanism; it’s takes a long time for it to have an effect, like the curve and it’s not instantaneous at all. So you have to plan in advance and all that kind of stuff, it’s difficult basically. So if you’re a person with Type 1 diabetes you generate a lot of data.

And I also us a Doxcon continuous glucose monitor. I’ve had one since 2011 and the first experience when I got one was that I was basically horrified. I had thought that I was in pretty good control and my doctors were pretty happy with where I was, but when I actually started to see the high-fidelity view of what was going on with my blood glucose I was pretty terrified.

So as Molly said it was a reading every five minutes, and compared to that most insurance companies will only cover 10 test strips a day. So if you’re not using one of these, what you get is 10 readings, but if you are using one of these you get 288 so it’s a big difference.

So this was like the big takeaway from the talk I did in 2012 where I did this experiment to basically make it that I didn’t hate looking at my Doxcon. And so the intervention is that I actually discovered that worked for me in terms of making it a better experience was restricting the carbs in my diet.

But unfortunately this is still rare. So in the diabetes community we all these a ‘no-hitter’, when your Doxcon stays between the two lines that are marking out the low and the high range. But this is still pretty rare even though I just discovered the semi-magic bullet for myself. It’s not easy. Anyone who has tried a carb restricted diet knows it’s not fun or easy sometimes to stick to it. Like I still have issues. So one of the things that I’ve been working on now is like my follow up is to try ways of looking at my data that will help me assess, like okay, am I doing all right this week, this month, this year. Am I getting worse, am I getting better you know, am I happy with where I am right now.

Part of that is motivation for me. Like it’s incredibly motivating for me to see something that, so this is a violin plot, so it’s a box-plot with a curl identity curve indicating the probability of the blood sugar readings in the different ranges. So that’s like a significantly statistical huge difference between 2011 and 2012 is when I started doing the carb restricted diet.

So that graphic came out of R if anyone is familiar with that and my background is that was the tool that I was most familiar with at the time. And now I’ve been trying to make this prettier and interactive and a little bit more accessible.

So this is something I’ve been working on that’s kind of a modified like inspired by a violin plot, where it’s just a histogram that’s a little bit more assessable and the colors you know high-light things that are the areas of interest, the target range and the out of target range. And I’m doing this on different timescales so weeks and years and things like that.

And then I think maps are really great for finding patterns. So this is really useful for me to look at what does my day look like in terms of what my blood glucose looks like over time. This particular graphic comes from when I was still a graduate student. So something that is really common among people with Type 1 diabetes is something called the Dawn phenomenon, and that’s basically the hormonal change that occurs when you wake up, and usually results in higher blood sugar in the morning.

So this actually looks like I don’t have one because my blood sugar goes down around like six, eight, seven in the morning, but this is because I was a grad student and I didn’t actually get up until 11. So I have a dawn phenomenon, it’s just not at dawn.

And then so this is again I’m sort of translating the technology and the week view with the heap maps was the a little bit less useful because it’s not quite enough data. But it does give useful indication around how much variation there is to, like you can see when the squares are more spread out and when they’re most closely packed together and that’s helpful.

And then the final thing that I’ve been experimenting with is that there’s also a big emotional connection to your numbers when you’re have Type 1 diabetes and the disease is pretty relentless in terms of management right. You can’t take a break. You can’t decide you want to stop tracking.

And so I was starting to think about ways to look at your numbers without looking at your numbers. So this is one of the things I read about somewhere and managed to try. So Chernoff faces are a technique where you put in a whole bunch of parameters and they control different aspects of the face. The came out of R again. And so here you can just see what days were different right, and probably worse is what’s happening here. But I think it’s really useful to just say like okay, even if I’m reviewing the last month or something or I’m going to an appointment with my endocrinologist and we want to talk about my data, we might sit down and look at this and say, okay what happened on June 17th and find out I was doing, but let’s say I was travelling. Travelling is one of the things that really make my blood sugar go crazy usually because I can’t eat when I want to eat, so I things become a mess.

So if I look at my calendar and say on the 17th like I took the M-track to Boston. Of course that was a crazy day; we don’t need to talk about right, because it was an outlier. It wasn’t my normal routine so there’s nothing to talk about, there’s nothing to learn from really. The month we’re going to have a big conversation about how we make travel better and a lot that might be relevant.

So basically, my takeaway is where I’m moving on from, my initial experiment is there’s value in all sorts of data presentations, and I actually think the more variety the better. There’s even value in these non-quantitative presentations of data. So just being able to pick out the outliers and decide if you want to talk about them or not without having to be confronted with like my average blood glucose on that day was sky-high. And so right now I’m working on creating like an actual dashboard with a whole bunch of these things all in one display, so that I can easily look at an arbitrary time frame and compare it with another one and all that kind of stuff, and integrate it with things like contextual day about my life from my calendar, things like that.

And I should also mention so here’s all my contact stuff and I do actually have a whole lot of my Doxcon data is online at GitHub repository for anyone to use if you would like to use or if you would like to do something with it. And there’s also like another couple of two weeks data with a whole bunch of other data where I gave away for free, and it’s documented and there’s little galleries of digitalization to.