Lessons from a year of heart rate data

And this is a project that is one of the sensors that we were playing around with, where we decided to track heart rate, track it as long as possible and see what we get out with it. And from the system we started to do extra things and probably thought you would class as overboard. So here I am here to present the entire dataset that I collected over a year, and if you want we are actually live streaming this via the Body Blogger Twitter feed so you can watch me having my panic attack in real time.

So I was tracking my heart rate for 24 hours a day, seven days a week and obviously I am collecting this and reflecting on the dataset. I am also streaming it to the Internet via various public channels, so the first one, the original which is the Twitter feed. So this is a numeric format updated every 30 minutes; the average heartbeat rate.

We also have a (pashu?) Blink which is giving people a time series data which is a bit more easy to understand, and we’ve also got my research website, which is color-coded to represent different states. So in the first week of running the system we classified a few basic activities such as sleep, normal work day in the office, high levels of stress, and exercise. And we color-coded that to give a more meaningful representation of data to users. So obviously unless you are a particular expert numeric and time series graphs are a bit difficult to read.

So we ran the program for just over a year and collected about 260 full days of heart beat rate data. So there was lots of daters to play around with and see various and interesting things.

A quick note on how the system was built. Very simple, we have a wireless heart monitor made by a company in Germany and that plugs into my laptop over there, and it collect all my data via a local web service and streams that to the Internet. Because we are using a local web service it’s much easier to develop whatever applications we want to track this and I don’t have to worry too much about that.

So personal experience, what did I learn; well for myself one thing you get to learn is how your body responds to average daily things in your life. So for example, one of the first things and my favorite thing was when I recently moved to Liverpool I found a nice sandwich bar where you get huge subway sandwiches, and then I thought I would check the heart rate data and see what that does, and look at the data going oh dear, I should probably skip these sandwiches, because my heart rate would increase and normally I at 70 beats per minute and I would eat this sandwich and I’m at an elevated of 90 beats per minute for the next two hours. So I probably think that is a bit of an overkill for my system.

Drink, this is something that came up at my last talk is that I may have focused a little bit too much on this, but it’s great to see how your heart responds to different levels of alcohol. So roughly about three or 4 pints worth of alcohol I get an elevated heart rate. The next days’ worth of activity during the sleep cycle doesn’t have much effect. Go over that and your next day’s sleep cycle is completely screwed compare to what it would previously be.

My sleep baseline is roughly about 50 beats per minute, if I drink too much the baseline increases to 10 to 20 beats per minute, and obviously if you are collecting longitude data it becomes evidence of the dataset you collected.

My favorite event that comes in is always is public talks. And part of my line of work I have to do several talks a year and it does get represented in the data. So this is my heart rate during my first talk at Quantified Self and this is actually one of the first talks where I was presenting this system, so I was a bit worried whether it would be accepted.

So you could see my baseline is down at 70 beats per minute, and this is normally what I would be doing if I was sitting down. I was sitting down during most of this thing and I’m running at least 90 beats per minute. As a bit of context, introduce myself and get a huge spike to 140 beats per minute. And then when we were doing the presentation itself I’m at least doing some similar activity represented actually running, my running state is 150, and giving the talk is roughly the same effect, and you can see the recovery period.

Also because you are collecting masses amounts of data you can also see longitudinal effects such as what is your daily sleep cycle and how well do you sleep. So if we look at atypical months in 2010, now this is something we had to learn quite fast is how do you represent longitudinal data. Time series graphs tend to fall apart when you’re tracking more than a single day.

So this is a typical month; the hours of day going across, the day and month going down. Each one of these cells represent an hour. White cells are essentially where no data was collected and you can see the color code in the beats per minute. So if I turn that into hot zones, you can see my sleep cycle right at the beginning a very calm sleep below 60 beats per minute.

You can see something that did surprise me when I was looking at this dataset again is I tended not to get up until 8 o’clock; I actually thought I got up much earlier, which explains why I’m always a little bit late to work.

If we look in the afternoon, so this is an evening and I tend to do a lot of running so you can see all of the recovery periods after I have, off my run. Unfortunately it is very difficult with the system that we set up to track when you have a mobile because I am linked to a laptop.

You can also see how the heart rate response changes to routine, so Christmas a big disruptive event in life, whereas going to see the family and lots of food, so this is my heart rate over Christmas. You can see November as a standard month and all of my daily activities. December, I’m sleeping normally here, but unfortunately because it’s now breaking the cycle and it’s becoming much difficult to reach my normal sleep state. And then when it comes to January too much food, no exercise and my normal sleep state has gone. So it is obviously important for this when you are doing physiological tracking is to be aware that your heart is a dynamic system and its constantly changing in the context around those activities change with it.

The shared experience: obviously, I’m uploading all this data in a public space and how do people react with it. Mainly people use the feeds like friends and colleagues would normally tend to use this in how you are feeling and predominantly use the color codes because they provide the best level of meaning.

Now because we coded different activities, there is a problem with the fact that there are mismatches in what I’m feeling and what the system is telling people. So a good example of this is I was preparing for a conference deadline, and I was they are typing away at a paper. The website is saying I’m highly stressed and it’s all yellow, and my boss sees this and saying yes, he’s working really hard.

A few days later, doing the same thing again, however I’m now green and it doesn’t look like I was stressed and my boss called me up and says, are you working on those papers because you are not as stressed as you should be. I’m going, oh dear. But also I was working on the paper, but it’s the fact that the heart rate is a dynamic system and completely changing, and you have to be aware of these.

Another point is that shared physiology is great again, so when I went to a conference and did another presentation again it was all a race around my physiology.

So in summary what did we learn? A heart rate is a decent monitor for daily activities and changes therein such as the disruptive events which are in December and Christmas. But context is vital through its interpretation. It’s a bit of a no-brainer this but it is very important that you collect some context to your data.

On a long-term basis context emerges from the data itself, so it’s very easy to see sleep patterns and very easy to see when you’re highly active because it is very easy to apply very basic human concepts such as well I sleep in the morning, I do some exercise during the afternoon and I work during the day. But when it comes to very fine grain control you know collect in that context it’s a bit more problematic because it is very data intensive, and one of the things that you’re trying to collect that passively is the problem with that.

Also important is looking at how to tailor the visualization for audience and content. The Twitter feed and the graphical data only really experts are good at reading that stuff. The heart maps and the color coding is much easier for the general audience to look at.

Finally, what does a years worth of heart rate data look like. So this is going through the entire data month by month, and as you can see the standard cycles of sleep, activities in the afternoons. Those red points that you can just about see these are usually indicated of conference deadlines, and the same thing when I have got to get a paper in or a presentation. So the difficult month for the entire year was December when I had non-stop conferences, which is that big yellow section right at the top.