The N-of-1 that Counts
Last updated: Jun 15, 2022
(This article was reposted with permission from the original Evidation post here.)
A Head for Health
Elvira turned 32 last year. That’s when the migraines started.
They’re episodic, maybe 5-8 a month. She takes medication that helps a bit whenever she has an attack. But she’d really like to figure out what triggers them or makes them worse. It’d be great if she could stop getting them altogether!
Elvira uses a diary app to log various day-to-day events. At any given moment, she can track a number of things, including her stress level, and what she eats and drinks—and whenever she gets a migraine (and how bad it is).
Determined to stop her attacks, Elvira starts poring over her data from the past year.
First, she notices that her stress levels are usually highest at the end of the week. That’s also when she tends to get migraines.
Her stress is largely brought on by her work schedule. She can’t really change this, so she may not be able to prevent her migraines, after all—a disappointing finding.
Unsurprisingly, Elvira’s app also shows that she drinks coffee almost everyday. She logs at least one coffee-drinking instance on most days; sometimes two, and occasionally more. She fondly recalls using her two favorite coffee mugs (which both hold about the same amount of coffee).
But then something interesting shows up. Elvira tends to drink coffee more than twice a day towards the end of the week—when she also seems more likely to suffer an attack!
Could too much caffeine be the true culprit? Unlike her work schedule, she can more easily change her coffee-drinking habits.
Elvira decides to do a little self-experiment. She’s a little apprehensive: Her friend Trisha also gets migraines. But in Trisha’s case, drinking coffee actually helps ease her pain and discomfort. Still, Elvira knows the best way to learn for sure what works for her is to do an experiment.
Each month for two months, she will limit herself to having coffee at most twice a day on half the days of the entire month, which she’ll pick at random. She’ll make sure to drink coffee at least three times on the other days. If too much caffeine is triggering her migraines, then she should get more attacks on days when she drinks coffee more than twice.
In the first month, Elvira suffers six attacks. Of these six, four occurred on high-caffeine days (when she drank coffee more than twice). In the second month, she suffers seven attacks, with five occuring on high-caffeine days. From this, she concludes that drinking coffee more than twice a day does in fact trigger migraine attacks, or at least makes them more likely.
Elvira goes on to limit her coffee consumption over the next few months, and notices that she now only gets 1-3 attacks a month—usually on days when she happens to drink coffee more than twice. Unlike her friend Trisha, drinking more coffee is actually bad for her migraines because it generally leads to more attacks.
Elvira ran what’s called an n-of-1 trial, an individual-focused or “idiographic” experiment where you compare your health-related outcome under normal circumstances—what’s called a “baseline condition”—to your outcome when exposed to a trigger or “intervention condition”. The goal is to see if there’s an overall (average) pattern between the suspected trigger and the health outcome.
Elvira found that drinking coffee more than twice a day (trigger) versus not (normal) generally increased her chances of getting a migraine (health-related outcome).
N-of-1 trials resemble the exercise, workout, or habit-change challenges of various apps. The challenge goals are often to help improve some measure of fitness and health, like lowering weight or body fat, or improving sleep quality or heart health. These are closer to single-case experimental designs (SCEDs) in psychology: they’re like n-of-1 trials with few but long periods of alternating baseline and intervention conditions.
Randomizing the conditions is key in an n-of-1 trial. It’s basically like flipping a coin with “baseline” on one side and “intervention” on the other—and you have to do whichever side comes up. Ideally, randomization helps make sure almost nothing else other than the intervention can cause a change in your health outcome.
Remember how “correlation does not imply causation”? Well, the experimental design of n-of-1 trials and SCEDs actually checks for causation, not just correlation.
This is why randomized controlled trials (RCTs) in clinical research are a gold-standard technique for figuring out if a new intervention or treatment actually works. “Flipping the coin” in a way balances everything else that might confuse or “confound” the way the treatment might impact the health-related outcome.
N-of-1 trials and SCEDs have been used for decades by clinicians to help patients figure out what really works for them in particular. But they haven’t yet been used much with “real world” digital health data from smartphones like Elvira’s diary app, or from wearable sensors like Fitbits and Apple Watches—what we call person-generated health data or PGHD.
Lots of folks may do their own informal self-experiments based on what they find in their PGHD. But these can be unsatisfying, frustrating, or misleading because they aren’t done in a rigorous way that really teases apart causes and effects.
N-of-1 trials and SCEDs add the rigor needed to provide much more definitive answers about what’s really causing what. The hope is that n-of-1 trials and SCEDs can help reduce the unnecessary pain and wasted time of trying out too many ideas.
Elvira’s story shows us the amazing potential for n-of-1 trials and SCEDs to help each of us figure out exactly what’s going on in our own lives. They can help us use our own PGHD to determine what actually influences our own health each and every day, regardless of how those same triggers harm—or may even help—others.