How Much Coffee is Too Much?
Tracking My Daily Caffeine Consumption
Facebook recently reminded me of a little experiment I ran on myself three years ago.
Back then I was really into coffee. I still love coffee now of course, but at that time the black gold was pretty much at the centre of my life.
My tiny flat in London was full of coffee brewing equipment and other coffee related stuff. Almost my entire first month’s worth of PhD salary went into a nice espresso machine and grinder, which came to dominate my very limited kitchen space.
I gave talks about the science of coffee. I spent more time in London’s many coffee shops than at my office. I even seriously considered starting two coffee related businesses, one hardware related, a smart espresso tamper, and the other software, a kind of LinkedIn centered around meeting up for coffee.
Thanks to the freedom I got as a PhD student I also had ample time to experiment on myself, from unusual “supplementation” to weird sleeping patterns, and I was very interested in the Quantified Self movement.
I kept detailed logs of what food I ate, what supplements I took and in which quantity, and even measured details like my blood sugar levels and blood oxygenation several times a day.
Given these admittedly slightly weird interests and my obsession with coffee it was only natural to dig a bit deeper into my caffeine consumption.
For a little bit over a month I recorded every single caffeinated beverage that I consumed, grouped into several categories such as “single espresso”, “double espresso”, “double espresso (decaf)”, “cold brew”, “V60”, and a few more, including some types of tea.
At the end of it I wrote a simple Python script to model the caffeine concentration in my blood based on the collected data.
The assumptions that went into my model were pretty simple.
I know from analyzing my DNA through 23andMe that I’m a fast caffeine metabolizer, so I assumed a biological half-life time of caffeine of about 5 hours.
I approximated the initial caffeine delivery to the bloodstream to be of Gaussian shape, peaking around 30 minutes after consumption. This also makes the simplifying assumption that I consumed the entire beverage in a single gulp.
Every drink category outlined above was then assigned an average amount of coffee used to brew it, for example 15g of coffee for the average double espresso. The highest caffeine content was in cold brew, which I estimated at 23g of coffee per serving (I used to brew fairly standardized batches and had little glass bottles for each serving, so this was quite consistent).
The final assumption was that all coffee beans have the same biologically available caffeine content by weight of about 7mg/g (except decaf which I generously estimated at 0.5 mg/g), and that all of the caffeine gets extracted during brewing, irrespective of the method used. Caffeine is a very easily soluble substance so this should be a fairly accurate assumption.
Based on my data and this simple model, I produced three plots.
The first was my blood caffeine concentration over the entire period.
It clearly shows a pattern of daily spikes, decaying off into the night until my first coffee in the morning triggers the next spike and repeats the cycle.
The second plot showed my average caffeine concentration over a 24 hour period from midnight to midnight. This revealed some interesting (and slightly concerning) insights.
During my PhD days I had a sleeping schedule that was shifted fairly late into the night. My average sleeping time was somewhere between 2 and 3 am. The plot shows that around that time I had on average still well above 100mg of caffeine coursing through my veins.
In my model that’s roughly the equivalent of a double Espresso taken straight before bed time.
Even by the time I woke up I was still above 50mg. You can actually infer from the graph that I usually got up around 10 or 11 due to the steepening slope. The flatter slope between around 8 and 11 am was mainly due to Wednesdays, the only day of the week where I had to get up early to get to university for my research group’s weekly breakfast meeting (which of course also involved coffee).
The final plot showed the day with the highest peak caffeine during this period.
Waking up around 11 am I made a Flat White (which was usually my first coffee of the day). That was quickly followed by a green tea.
The night before I had made a new experimental batch of cold brew, of which I tried two servings in close succession at 12:43 and 13:02. That’s the steep slope in the graph at around 1 pm.
Adding a few more coffees in the afternoon, I apparently maxed out at almost 800mg of caffeine in my blood at 5pm.
This is still well below the more than 10g that might have been considered a lethal dose, but I don’t think I felt particularly wonderful that evening.
There are certainly days outside this recorded period where I had even higher doses. I very distinctly remember my first London Coffee Festival. The huge number of free tasters left me feeling like I had the heart rate of a humming bird on crack, not to mention the profuse sweating and nausea.An experience I don’t really recommend.
Over the last few years I have definitely cut down massively on my caffeine intake. I haven’t tracked my consumption that rigorously, but I guess I’m now peaking at a bit over 200mg at around 3 or 4 pm on an average day.
Maybe I should repeat this experiment and track my data properly again. I have also been using an Oura ring for the past years to track my sleeping patterns as well as resting heart rate and heart rate variability. Comparing this data could show some very interesting correlations.
Tracking alcohol consumption in a similar way might also be interesting, although the results might be even scarier (and my records might get fairly inaccurate later into a night of drinking).
I’d be curious to hear about your experiences! Have you ever tracked anything like this? What were the results and conclusions you drew from it?
If not, I really encourage you to dig into something that matters to you, track it, look up the science behind it, and build a simple model around it.
Even if the model is wildly incaccurate, you’ll still learn a lot from the experiment. In the words of Peter Drucker, what gets measured gets managed.
My little experiment definitely helped me be more aware of my own habits and their impact, and ultimately improve them for the better.