Concussions, Headaches and the Whole30 Elimination Diet
cognition | diet and weight loss | other
Steven Zhang shares his history of tracking and measuring concussions, headaches and sleep patterns along with an intervention he did to improve his sleep. After constantly suffering from headaches, he decided to take action and track them post a concussion the year prior. He used Taplog to collect his data to analyze if he was indeed improving.
tableau | Taplog
Hey, I’m Steven. I’m going to talk to you today about my history of measuring concussion, headaches, my sleep patterns and then about a specific intervention that I did in how to improve those sleep patterns.
So first, headaches, most of my adult life I have been prone to sleepiness and headaches throughout the every day. But it wasn’t until I got my first concussion last summer, that I started taking logging of that seriously.
I wanted some sort of way to measure whether I was improving from my post-concussion syndrome symptoms, so I started blogging headaches and other body pains and other associated symptoms with concussions.
For the record, when I got my concussion I did not catch the Frisbee, I did it wrong.
I use Taplog to collect my data every time. If I feel a headache, I’ll press the buttons associated with headache. I’ll give it a rating, a comment. I’ll export the data to CSV. I have an automated pipeline for that, just a simple script. And then I imported into Tableau which is visualization software to make charts.
This is a chart that I made of my sleep, so for the past six months, every vertical slice is a day. The negative in the vertical access represents say 10 PM from the night before, would be negative to in this vertical access. So that way, the night before, would carry it over to this current day.
I do manual tracking here as well, because I tend to take a lot of naps and I tend to lie in bed and try to go to sleep, and end up not going to sleep. As far as I know there is no good, consistent automated way of differentiating that.
With the manual logging, I am able to tap out and tag the previous sleep segment was, whether it was a nap, unable to sleep, or actually falling asleep and generate a graph like that.
Another way to look at my sleep data, very widely variable bedtime, means I don’t have very good sleep hygiene, and I also wake up several times a night, usually. If we plot what time I generally go to bed, you can see that I generally go to bed around midnight or one for actual sleeping, and for the naps it’s pretty consistent throughout the afternoon.
We can further tease this apart and generate a 2-D histogram, and this allows you to see that my most common sleep characteristic patterns are, going to bed at midnight or one and sleeping for seven or eight hours.
All those graphs I showed you earlier, I used Tableau again to generate. For the record, I work at Tableau, but I don’t work in sales and marketing. I just use it because I think it’s a useful tool for what I have to do.
Correlation between bedtime and total hours of sleep. So I try to match all these sleep metrics, and try to see which one hand, the strongest correlation. And the one with the strongest correlation was this. It wasn’t immediately obvious to me without looking at the data, because I don’t set alarm clocks.
So even if I go to bed at like 2 AM I’ll try to sleep until 10 AM the following day. But what this is showing me is that if I do that I’ll be less likely to sleep until 10 AM than if I went to bed admit night and tried to sleep until 8 AM.
The thing I’m trying to optimize here, though, isn’t sleep. It’s how I feel the day after, right. So this quality of daily metric, and the best way I thought of measuring this is using a subjective guided scale like this one. I had tried using some sort of algorithm to calculate the quality of day using headaches and how often I felt sleepy, but that got too complicated.
So using that, you can see that even though day-to-day there is a lot of variations, there is some long-term trends. For example, mid July through October was my concussion recovery; I had my concussion mid-July, and then you can see some dips in March and May for sicknesses.
So if I try to seek what sleep metrics correlate with the quality of daily metric, the one with the strongest correlation turns out again to be my bedtime; it has the smallest p value. Though, this only applies in the days that I sleep for a total of eight hours. If I sleep less than eight hours, I suspect I don’t feel the effects of sleep deprivation until the day after. If I sleep more than eight hours, I’m likely recovering from some sort of sleep deprivation from multiple days earlier.
By March, I had fully recovered from my concussion, but I still had lingering headaches and sleepiness. And so I decided to try a new intervention. This is the whole 30 diet. It’s a paleo elimination low carb diet.
One of the things I was trying to reduce is tiredness, but instead of when I tried to measure direct tiredness, like every time I feel tired, logging it, I found that to be inconsistent because I didn’t have a solid threshold, like how tired do I have to be to log it and I would often forget to log it. But I knew my sleep data was very high quality, and I knew that wherever I took a nap, I remember to log it and it was like bright line. I either lay down and take a nap or didn’t.
So using that as a proxy for tiredness, you can see that after the diet things have improved about half, not dramatically, but good enough. I started the diet on April 1, and I’m still doing it right now.
With non-migraine headaches, I had an adjustment period of all of April to adjust to the low carb diet. This is a common thing called the low carb flu for low carb diets, which results in a lot of minor headaches and I suspect that’s the cause of this.
With migraines, though it’s a different story May was better than March and April, but no better than the previous months. So I can’t really draw any solid conclusions from that.
Some overall thoughts. Number one, manual tracking can be really really useful. I manually tracked nap information, and that’s something that currently with technology right now you can’t really track at least consistently.
There’s also some things that will never be able to automatically track, for example, my quality of day scale, which by definition is about how I feel and cannot be measured. So I think in our movement towards more and more automated data collection, and data pipeline, we need to be aware of that, there’s some things that only manual tracking can bring.
And though it is tedious, I found that over time, manual tracking led to mindfulness, especially when I made my quality of day score at the end of every day. I would force myself to look through my logs for that day and see how the day went.
Second point, get the right tools for the job. I see a lot of people trying to use the iPhone to do everything. I mean, if you have infinite time, you could plot everything, and do everything in a programming language. But you don’t have infinite time, so don’t do that. Save the programming for the ETL and data prep, and use a tool like Tableau to generate each of these plots. Each of these plots took me less than a few minutes to make, minus the formatting for this presentation.
And the final point I want to make, I spent three years of trying to gather some insights for my data, and every time I generated a plot, and showed my friends those plots, they would say oh, what you’ve showed are is pretty, but it’s totally obvious, sort of like this XKCD comic right.
But over time, I found that I started getting some insights and what’s more, I had this baseline of data with which to compare future interventions, like the whole 30 intervention that I had. And if you like the visualizations that you saw, you can come to my Tableau office hour tomorrow. There is a free version of Tableau that you can use, and will show you how to use it.
Thanks a lot.