Sleep is important for myriad clinical and personal outcomes such as cognitive function, mood, mental health and heart disease. Sleep patterns vary considerably from individual to individual and from night to night, and are affected by a variety of factors including obesity, respiratory symptoms (coughing, sneezing), age, gender, and alcohol intake. It’s not much of a stretch to imagine that almost all aspects of our daily life - stress, food and drink, exercise, our relationships with our children, partners and friends, the type of work we do and how much, various emotions - could potentially impact our sleep.
Because it’s such a personal experience, sleep provides a perfect line of inquiry for N-of-1 studies, the subject of much recent discussion and research at HD2i.
As a simple example of what is possible with an N-of-1 study, I decided to investigate the impact of alcohol on my sleep. Some people believe a drink before bed helps them to relax. Others, like myself, have the sense that it disrupts their sleep.
This study is simple and fairly qualitative, but it provides a flavor of what we can hope to accomplish with N-of-1 studies. Interestingly, it also corroborates some published sleep results from the 1960s.
The Science of Sleep
Traditional sleep research is highly fragmented: neurologists are interested in neurological symptoms, pulmonologists and cardiologists are interested in breathing and heart function, etc. As a result, clinical sleep studies tend to focus on identifying and treating specific clinical pathologies. Almost none address the broader goal of improving sleep quality and efficiency in the absence of disease.
One of the core difficulties of sleep science is that we have poor insight into our own sleep. If you walk into your doctor’s office to discuss a painful spot on your rib, she might ask you about the history of the problem, how you feel during different movements and activities, and any treatments you’ve tried. On the other hand, if you’re there because you don’t feel your sleep is restorative, all you can really report on is what happens before you fall asleep, yada yada yada, and after you wake up. In the words of Jerry Seinfeld, “You yada yada’d over the best part!”
Sleep studies, currently our best means of diagnosing sleep disorders, are typically conducted in a lab using polysomnography equipment to record physiological data like airflow, respiratory effort, blood oxygenation, snoring, body position, and brain activity. More recently, in-home sleep studies have become a cost effective alternative that allows the patient to sleep in his/her own bed and avoid the first-night effect.
Clinical sleep study compared to consumer sleep device
On the left is a patient in a laboratory sleep study, and on the right is me sleeping at home with our device suite from our ongoing Sleep in the Natural Environment study. Who do you think is having a more realistic night’s sleep?
Monitoring Sleep in the Home Environment
One of the devices in our sleep study is the Withings Aura Smart Sleep System, which consists of an under-mattress pad and a strange-looking lamp. Together these components capture ambient sound, light, temperature, heart rate (yes, through the mattress!) and movement. These data are used to stage sleep and monitor transitions between light, deep and REM sleep over the course of a night.
- Light sleep represents the transitional stage between wakefulness and unconsciousness.
- Deep sleep is characterized by neocortical rest, is needed to consolidate memory, and is necessary to sustain life.
- REM sleep is characterized by rapid eye movements and lack of muscle tone, and is associated with dreams and memory formation.
The Withings Aura Smart Sleep system
Sleep stages are defined by characteristic patterns of brain activity as measured using electroencephalography (EEG). Since most consumer devices used to monitor sleep outside the lab do not measure EEG, these devices can only infer sleep stage indirectly, by measuring other physiological parameters such as heart rate, respiratory rate, and movement.
When compared to gold standard, the best of these devices are only 80-85% accurate. However, we can trade a reduction in device accuracy for an increase in data (more consecutive nights of sleep) in a natural sleep environment. Our hypothesis is that more data will allow us to build a more complete picture of an individual’s sleep than can be assessed in one or two nights in a sleep lab.
Alcohol Sleep Study Design
I designed a small study to test the relationship between sleep and alcohol. I drank myself to a pre-determined blood alcohol level, which I monitored at regular intervals with the BACtrack Mobile Pro1. Once I reached my target blood alcohol level, I drank a glass of water and went to sleep.
Study tools essential for all at-home blood alcohol monitoring in Silicon Valley: Apple products, a portable breathalyzer, and vodka. Missing: shot glass
To measure how different lifestyle choices, like alcohol intake, impact sleep quality, I first needed to establish a baseline for “normal” sleep. This (below) is what a good night sleep looks like for me as measured by the Withings Aura. I know it’s good sleep because I’ve been collecting sleep data on myself for about 3 months.
Percentage of sleep by sleep stage as measured by Withings Aura (n=23 nights)
The Effects of Alcohol by Sleep Stage
This chart compares a normal night’s sleep to a night of sleep after drinking: notice the rather dramatic reduction in REM sleep. While I only completed 2 nights of the drinking protocol, the effect is quite large. Since my baseline is well characterized, I have some confidence in this result. It turns out this result has been observed in the literature as early as 1966! Those studies required EEG sleep patterns to be characterized in a laboratory setting - I could do the same thing from the comfort of my own bed.
Effect of alcohol on percentage of sleep by stage (n=2 nights)
Getting my REM Back… with NAC
Once I knew how, exactly, alcohol disrupted my sleep, I was curious if I could somehow stage an intervention to get my REM sleep back after drinking. In graduate school, the immunology lab where I worked studied an amino acid called N-acetylcysteine (NAC).
NAC is a precursor to the antioxidant glutathione, which is a powerful modulator of neurotransmitter and inflammatory pathways. NAC is primarily used to treat acetaminophen overdose in the ER, but in our lab it was being studied in HIV, autism, allergies and cystic fibrosis. It is safe in very high dosages and readily available over the counter.
In the lab there was a bit of lore that NAC could be used for lots of applications, including treating hangovers. So I figured I would give it a shot and added to my glass of water before bed.
Effect of NAC after alcohol on percentage of sleep by stage (only 1 night!)
Big warning: I only tried this intervention for one night (a man can only get drunk so many times in the service of science), so I don’t feel very confident about the result, but in that single night you can see that my REM sleep recovered to normal levels. Intriguing!
Unfortunately, NAC was not successful at mitigating the hangover.
N-of-1 Studies of Sleep: Some Thoughts
This toy experiment alludes to a much larger idea: that a clinician or individual can formulate and test a hypothesis using unobtrusive data collection and attain a highly personalized result.
This study raised a number of questions for me:
- What is the dose-response relationship between my alcohol consumption and sleep?
- Does NAC reliably recover REM sleep after alcohol consumption?
- How would I design a more robust experiment to test this effect?
Answering these questions would probably require a platform that could help me design scientifically robust experiments, collect and integrate the data into a single database, and analyze the data using appropriate statistical tools.
Fortunately, we’re building it here at HD2i.
Not all quantified-self devices are made equal, but I made some calls and was able to get my breathalyzer calibrated along with a handful of police breathalyzers. To the surprise of myself and the officers, this little gadget is incredibly accurate; within the same error range as the official police field units! ↩